Superelevation Criteria for Sharp Horizontal Curves on Steep Grades (2014)

Below is the uncorrected machine-read text of this chapter, intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text of each book. Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.

40 S E C T I O N 4 This section presents the analytical and simulation modeling work performed to investigate superelevation criteria for sharp horizontal curves on steep grades. Section 4.1 presents the step- by-step analysis approach which integrates both field and simu- lation data and is based upon an increasingly detailed analysis using progressively more sophisticated simulation models. Sections 4.2 through 4.12 present the individual steps of the analysis, first describing the goal and methodology for the step, followed by background information and individual results, concluding with a summary of the key results for the respec- tive step/analysis. The analysis considers a range of horizontal curve and vertical grade combinations and six vehicle types (i.e., three types of passenger vehicles and three types of trucks). The analysis considers situations in which vehicles maintain a constant speed through the curve and situations with progres- sively more aggressive deceleration. The analysis also considers situations where the vehicle’s desired trajectory is to maintain the same lane from the approach tangent through the curve and situations with a lane-change maneuver. The primary perfor- mance measures of interest from the analyses are lateral fric- tion and rollover margins that indicate whether a vehicle can successfully follow its desired trajectory through a geometric condition (i.e., horizontal curve and vertical grade combina- tion) without experiencing a skidding or rollover event. The severity of skidding and rollover events is also described in some situations by considering the duration of the event and the lateral deviation from the desired vehicle trajectory. Most of the analyses/steps focus purely on the dynamic capabilities of the vehicle to traverse the given geometric condition. It is only in the most sophisticated and complex analyses (i.e., multibody models) that the inputs and capabilities of a driver are consid- ered. Section 4.13 summarizes the main, overarching findings from the analytical and simulation modeling. For the analytical and simulation modeling, a sharp horizon- tal curve was defined as a minimum-radius curve as determined from the maximum rate of superelevation and maximum side friction factor for given design speeds. 4.1 Analysis Approach The analysis was designed to use a combination of field data (see Section 3) and simulation results to evaluate geo- metric design criteria specific to sharp horizontal curves on steep grades. The general framework for the analysis is shown in Figure 23. The overall goal of the evaluation framework was to develop recommended modifications to existing AASHTO design pol- icy to improve conditions that may generate concerns at sharp horizontal curves on steep grades. The notion of “substantial error” in the evaluation framework was one where differences were observed in field data versus simulations, and between simulations of different fidelity. Where field data were available to compare with simulation results, the field data were used to verify that simulations were providing reasonable results. In several of the key steps, the primary focus was to deter- mine whether friction demand, f, exceeds supply friction, ftire-pavement. These design conditions should be avoided because they increase the risk of a vehicle skidding and being unable to maintain the desired trajectory on the horizontal alignment. For each analysis, the ftire-pavement values are represented by a friction ellipse that encompasses the maximum friction sup- ply in the longitudinal or x-direction (braking) and lateral or y-direction (side) as shown in Figure 24. Both limits change as a function of speed, tire type, and pavement condition. To determine whether a vehicle can traverse a horizontal curve without skidding or overturning, a minimum require- ment is that the “operating point” representing the friction demand remains within this friction ellipse. Departure of the operating point from within the friction ellipse repre- sents cases where friction demand exceeds friction supply, resulting in skidding of the tire. The operating point changes depending on the curve radius, the superelevation, steering maneuvers, and braking forces used in the horizontal curve. Much of the simulation work focuses on calculating the oper- ating point of a vehicle within the friction ellipse under dif- ferent maneuvers and assumptions. Analytical and Simulation Modeling

41 To ascertain whether the operating point ( fx, fy) lies inside the friction supply ellipse for a given combination of cor- nering and braking demand, the constraint of Equation 15 must be met. While this equation serves as a good check of whether friction supply limits have been exceeded by the vehicle’s demand, it is less useful as a definition of lateral fric- tion margins because it weights braking and cornering mar- gins equally. In practice, however, braking forces should be given priority because, when a vehicle begins to skid, the tire forces are in the opposite direction of the skid, and there- fore the cornering forces are greatly diminished. Thus, hav- ing excess cornering margins but zero braking margins is not very meaningful since the cornering margins will mean little if the vehicle is unable to steer. The goal, therefore, is to define lateral friction margin for the purposes of this study. This definition must be mathemat- ically tractable, must give priority to braking margins first, and should remain consistent with the definition of margin of safety against skidding used in highway design. The defini- tion should also reflect the friction ellipse concept. To develop a definition of lateral friction margin, consider the simple definition in Equation 16: (16)margin y,supply yf f f= − In other words, the lateral friction margin is defined as lat- eral friction supply minus the lateral friction (i.e., cornering friction). Because braking friction demand decreases avail- able lateral supply friction below the nominal value of fy,max, as demonstrated by Equation 15, the following modification is made to fy,max to obtain fy,supply by rearranging the friction ellipse equation: 1 (17)y,supply y,max x x,max 2 f f f f = −     Figure 23. Framework for evaluating analytical and simulation models. Figure 24. Friction ellipse (tire–pavement model).

42 Combining Equations 16 and 17 obtains a usable defini- tion of the lateral friction margin: 1 (18)margin y,max x x,max 2 yf f f f f= −     − This definition of the lateral friction margin therefore depends on the tire’s demanded side force, fy, the demanded braking, fx, and maximum dimensions of the friction ellipse in the braking and lateral directions, fx,max and fy,max. This lat- eral friction margin, where braking forces are assumed to be required first before lateral forces are available, is consistent with tire behavior near skidding. At the onset of a skid, the tire’s force will be applied only opposite the direction of the skid, with little side forces available. This is generally in the braking direction, and thus there are little to no side forces available if braking is maximized. The definition of lat- eral friction margin above appropriately reflects this. With this definition of lateral friction margin, values greater than zero imply that the maneuver will not cause skidding, whereas values less than zero may cause skidding. This defini- tion is used in simulations regardless of the complexity or struc- ture of the simulation. For example, when using the modified point-mass model, the “tire” considered is actually a lumped representation of the sum of forces on all tires possessed by the real vehicle. When considering the per-axle (bicycle) model, each “tire” considered represents two tires lumped together, or even eight tires in the case of the rear tractor and trailer axles. For the per-tire simulations using high-order multibody simu- lation software, the “tire” considered is consistent with a single “tire” on the physical vehicle. This is important because chang- ing normal loads during a simulation due to weight transfer affect the ultimate supply friction available on true tires due to tire load sensitivity, and also change the friction demand on each modeled tire as the model structure complexity increases to approach reality. When evaluating lateral friction margins and rollover margins, the following general qualitative categorization was assumed: • Lateral friction margin ≥ 0.2: Large margin of safety • 0.1 ≤ lateral friction margin < 0.2: Medium margin of safety • 0 ≤ lateral friction margin < 0.1: Low margin of safety • Lateral friction margin < 0: Unacceptable margin of safety For modern roadway designs in nominal conditions, the lateral friction margins are expected to be quite high. The side friction demand in horizontal curve design is usually quite low relative to the side friction that can be supplied by the tire–pavement interface. AASHTO policy for horizontal curve design suggests some maximum friction demand levels, fmax, for use in the design of roadways. These values are par- ticularly conservative because they are based on driver com- fort thresholds rather than skidding or rollover thresholds. Because this study is examining potential modifications to this policy, a research approach was developed to identify situa- tions where the friction demand curves used by AASHTO can be violated due to sharp horizontal curves on steep grades and to investigate these situations further. From this analysis, spe- cific changes in superelevation policy can be recommended to correct for areas of concern. The approach to the analytical and simulation modeling comprises 11 steps as follows: Step 1: Define basic tire–pavement interaction model(s) and estimate lateral friction margins against skidding in AASHTO’s current horizontal curve policy Step 2: Define road geometries and variable ranges for use in subsequent steps Step 3: Develop side friction demand curves and calculate lateral friction margins against skidding considering grade using the modified point-mass model Step 4: Define vehicles and maneuvers to use in non-point- mass models Step 5: Predict wheel lift using quasi-static models Step 6: Predict skidding of individual axles during steady- state behavior on a curve Step 7: Predict skidding of individual axles during braking and lane-change maneuvers on a curve Step 8: Predict skidding of individual axles during transient steering maneuvers and severe braking Step 9: Predict skidding of individual wheels Step 10: Predict wheel lift of individual wheels during tran- sient maneuvers Step 11: Analysis of upgrades The goals, details, and primary results of each step are pre- sented in the corresponding sections. At the start of the research it was generally assumed that vehicle operations on steep downgrades were the more criti- cal situations to investigate compared to steep upgrades. Therefore, much of the analytical and simulation analysis focused on investigating horizontal curves in combination with steep downgrades, but to be thorough, some analyses were performed to investigate vehicle operations on sharp horizontal curves on steep upgrades. Steps 1 through 10 (Sec- tions 4.2 through 4.11) focus on downgrades, while Step 11 (Section 4.12) addresses upgrades. Six classes of vehicles were considered in the analytical and simulation modeling, as appropriate, including three classes of passenger vehicles and three classes of trucks. In presenting results of the first few steps, most of the discussion focuses on

43 the simulation results for passenger vehicles with a brief dis- cussion on the simulation results for trucks. It is not until the last few steps (i.e., beginning with Step 7) that more detailed results for the different truck classes are presented, as the differ- ences between trucks and passenger vehicles become more pro- nounced with these increasingly complex simulation models. 4.2 Step 1: Define Basic Tire– Pavement Interaction Model(s) and Estimate Lateral Friction Margins against Skidding in AASHTO’s Current Horizontal Curve Policy The objective of Step 1 was to develop and refine tire– pavement interaction model(s) that estimate(s) friction supply on typical roads, ftire-pavement, for use in subsequent sim- ulations. The model(s) predict tire forces as a function of tire type, vehicle speed, friction supply measurements, and pave- ment wetness. Friction supply curves from model estimates were then compared to AASHTO’s side friction design curves to estimate lateral friction margins against skidding presently assumed in current AASHTO horizontal curve policy. 4.2.1 Analysis Approach Data from the friction testing (see Section 3.4) were com- bined using the general procedure in Figure 25 to obtain tire force curves for representative passenger vehicle and truck tires on the roads where friction measurements were taken. First, the DF tester measurements (see top portion of Table 17) were fit to a tire force curve for the ASTM tire. This generates the reference skid number measurements of a road. The mea- sured skid numbers are shown in Table 18 for the longitudinal direction (i.e., x-direction) corresponding to tests at 40 mph. Additionally, the CT meter data (see bottom portion of Table 17) and DF tester data can be transformed into lateral forces to generate representative skid numbers for the lateral direction (i.e., y-direction). The corresponding values for pas- senger vehicle tires are shown in Table 19. These lateral skid numbers are not typically reported in the literature. They are reported here for the passenger tire as these values are more appropriate for horizontal curve design than longitudinal skid numbers as they represent the measured values of limiting side force available to a tire before sideways skidding. Com- paring Tables 18 and 19, the lateral skid numbers are generally 9 to 25 lower than the longitudinal skid numbers. Figure 25. Sequence to convert field measurements to representative tire parameters. DF Tester CT Meter ASTM Tire Parameters LuGre Tire Model Passenger Tire Parameters Truck Tire Parameters Site Measurement location Avg Min 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 MD1 77 77 77 74 72 72 78 75 76 73 73 76 74 75 73 77 78 76 71 71 77 74.9 71 MD2 59 61 63 62 60 58 62 62 61 58 60 59 59 67 64 63 59 57 64 62 67 61.3 57 MD3 71 69 70 74 74 70 56 68 68 70 65 68 65 67 65 71 70 70 63 63 62 67.6 56 WV1 74 81 85 87 87 87 89 90 87 87 90 88 78 76 75 78 80 80 90 86 88 84.0 74 WV2 71 75 73 78 78 78 73 74 76 76 76 75 70 70 71 76 74 70 67 63 64 72.8 63 WV3 74 83 83 84 84 83 85 87 82 81 83 84 70 72 68 65 69 67 81 77 78 78.1 65 WV4 69 68 73 73 72 76 76 74 76 81 77 74 73 72 69 70 67 73 71 72 53 71.9 53 WV5 71 73 75 77 77 78 75 74 72 72 71 73 68 70 70 58 69 71 75 77 82 72.8 58 Table 18. Skid numbers in longitudinal direction at skidding (40 mph).

44 Additional information about the road surface is needed to capture the full tire force curves in combined longitudinal and lateral skidding, across a range of skidding values from normal driving to full skids. In particular, the skid numbers only provide the skidding values and therefore do not give a good indication of tire forces transitioning from maximum friction to skidding friction conditions. To describe partial skidding phenomenon, the LuGre tire model was used. The LuGre tire model predicts tire forces by estimating the local deflection, z, of each portion of the tire using a model sim- ilar to a spring/damper system sliding along a surface with a rel- ative velocity, vr. As an analogy, the tire’s deformation is treated like “bristles” on a brush sliding along a contact area moving below; thus, sometimes the LuGre model is referred to as a Bristle tire model. Under these assumptions, the braking force of the tire element, Fxi, can be calculated using the following: (19) 0 1 2 0 1 2 F F z dz dt v F F z dz dt v x n Stiffness Effect Damping Effect r Viscous Effect xi zi Stiffness Effect Damping Effect r Viscous Effect       = σ + σ + σ = σ + σ + σ where Fzi is the normal force on the tire contact patch and s0, s1, s2 are model constants that depend solely on the proper- ties of the tire, and thus are different for passenger vehicle and truck tires. Once the tire properties are determined, the tire models can predict tire friction for pure braking, pure cornering (until skid), and combinations of braking and cornering. The result- ing curves form an ellipse that represents the available tire forces. Figure 26 shows an example friction ellipse for the WV2 site at the second measurement location (see Figure 22). To investigate whether friction changes within a curve in characteristic patterns—for example whether the friction may be lower on the entrance to the curve—the tire mod- els were used to predict the maximum supply tire friction for pure braking and pure cornering across all speeds in the study, at all locations and all sites. The results showed no clear trends to suggest that friction values are different at the beginning, middle, or end of the curve. These results indicated that, for each site, the mean friction and statistical variation in the friction values can be used to model vehicle behavior, rather than detailed location-by-location modeling of friction values. 4.2.2 Analysis Results To determine the range of friction values to consider as rep- resentative of a road surface, the statistical distribution of fric- tion values measured from each site and each measurement location were examined. Figure 27 shows the distributions of the maximum braking and cornering friction values across all sites at 40 mph for passenger vehicle tires. Figure 28 shows the same data for 85 mph. These friction values follow roughly a normal distribution, with a mean friction supply between 0.65 and 0.88 for wet-road conditions. These numbers are in agree- ment with published data for wet roads, at 40 mph test speeds, for well-maintained pavement surfaces and passenger vehicle tires which suggest wet-road friction values of 0.6 or higher. The distribution of the friction data can also be used to determine the minimum values of supply friction to consider when evaluating lateral friction margins against skidding. In this case a Gaussian (normal) probability distribution func- tion was used to fit the data. Taking a conservative approach, the worst-case (i.e., minimum) friction values selected for use in evaluating lateral friction margins against skidding were the 2nd percentile of the distributions, determined by the mean friction minus two standard deviations in the friction data. This suggests minimum supply friction values roughly between 0.5 and 0.7 (as seen in Figures 27 and 28) for evaluat- ing lateral friction margins against skidding. Figures 27 and 28 illustrate the probability distribution functions of the friction data for two speed levels (40 and 85 mph). To cover the full range of speeds considered in this evaluation, friction supply curves for wet-weather conditions were generated for full braking and full cornering for speeds between 25 to 85 mph for both passenger vehicle and truck Table 19. Skid numbers in lateral direction at skidding (40 mph). Site Measurement location Avg Min 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 MD1 61 62 61 58 57 56 63 59 62 59 58 62 58 62 58 62 64 61 56 56 63 59.9 56 MD2 51 52 55 53 51 49 53 54 53 50 49 48 48 58 56 54 50 48 52 52 56 52.0 48 MD3 61 59 60 63 63 60 48 58 59 61 54 58 56 59 57 63 62 62 53 55 54 58.3 48 WV1 47 50 53 55 56 56 58 57 58 59 61 57 55 49 52 59 58 58 62 58 60 56.1 47 WV2 61 63 60 67 68 68 62 62 65 65 66 65 61 60 61 67 65 61 59 55 55 62.7 55 WV3 66 74 75 76 75 75 76 79 75 73 75 76 62 61 56 56 60 58 73 69 70 69.5 56 WV4 59 57 62 62 61 65 64 63 64 69 67 62 61 60 59 58 54 62 59 61 44 60.6 44 WV5 61 64 65 67 69 69 65 64 61 62 60 63 60 61 61 52 61 63 67 69 74 63.7 52

45 tires. Figure 29 illustrates the friction supply curves for the maximum friction measurements, providing both average val- ues and two standard deviations below the average values, for both the full braking and full cornering conditions. Figure 30 presents similar information based on the skidding friction values rather than the maximum friction values. Equivalent curves for truck tires are shown in Figures 31 and 32. For com- parison, Figures 29 through 32 also show the AASHTO maxi- mum side friction factors used in horizontal curve design. A goal of this analysis was to define reasonable estimates of the friction supply, ftire-pavement, as a function of speed, and to represent the values in a manner easily interpreted in terms of lateral friction margins against skidding. The vehicle dynam- ics literature contains a wide array of tire–pavement mod- els, and the choice of the LuGre model is a tradeoff between its comparatively high accuracy and modest computational demands. Because normal driving does not involve signifi- cant skidding, this tire model captures the vast majority of phenomenon of importance in this study. Further, the dif- ference between the AASHTO maximum side friction fac- tors used in horizontal curve design and the field-measured friction curves gives an estimate of the difference between the current geometric design policy based on the point-mass model and the friction levels demanded by more complex models. In Figures 29 through 32, the braking-only and cornering-only curves show a significant lateral friction mar- gin against skidding between these and the AASHTO maxi- mum side friction factors used in horizontal curve design, and thus the main areas of design concern are likely to arise primarily from interaction of braking and cornering forces. In later sections where lateral friction margins are reported, the margins generally represent the difference between friction supply and friction demand. To avoid skidding or departure from a desired trajectory, the lateral friction margin should be positive. To simplify the simulation process, the demanded friction levels are obtained from vehicle dynamic simulations that are run hereafter under “dry-road” assumptions. These dry-road simulations will demand much more tire force than can be achieved in wet-road or icy-road conditions. In contrast, the supply friction will be obtained from the passenger vehicle and truck curves in Figures 29 to 32, which are based on wet-road conditions. This difference in dry-road assumptions for calcu- lating demand versus wet-road conditions for estimating fric- tion supply is not only easier to simulate, but also it produces more conservative results. This conservatism accommodates friction transitions that commonly occur on roads but are hard to consider analytically. For example, a vehicle that is maneu- vering on a dry road may encounter a wet patch of road within that maneuver (e.g., an area of the road that is drying more slowly than the surrounding road segments). In such a case, the tires could be demanding forces on entrance to the maneuver that are from a dry road, but friction availability along other portions of the road may be limited by wet-road conditions. Figure 26. Friction ellipse for friction data collection location 2 (site WV2).

46 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 1.1 1.2 1.3 0 10 20 30 40 50 Peak normalized braking force, F x,max /F z Fr eq ue nc y Data Fit with Mean = 0.878 and StdDev = 0.091 2% friction value (Mean - 2 ) 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 1.1 1.2 1.3 0 10 20 30 40 50 60 70 Peak normalized cornering force, Fy,max/Fz Fr eq ue nc y Data Fit with Mean = 0.711 and StdDev = 0.068 2% friction value (Mean - 2 ) Figure 27. Distribution of maximum friction for longitudinal (braking) and lateral (cornering) directions across all sites for passenger vehicle tires (40 mph).

47 Figure 28. Distribution of maximum friction for longitudinal (braking) and lateral (cornering) directions across all sites for passenger vehicle tires (85 mph). 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 1.1 1.2 1.3 0 10 20 30 40 50 60 Peak normalized braking force, F x,max /F z Fr eq ue nc y Data Fit with Mean = 0.821 and StdDev = 0.080 2% friction value (Mean - 2 ) 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 1.1 1.2 1.3 0 10 20 30 40 50 60 70 80 90 Peak normalized cornering force, Fy,max/Fz Fr eq ue nc y Data Fit with Mean = 0.650 and StdDev = 0.059 2% friction value (Mean - 2 )

48 Figure 29. Passenger vehicle tire measurements of maximum wet-tire friction in longitudinal (braking) and lateral (cornering) directions (mean and two standard deviations below mean of the maximum friction supply). 30 40 50 60 70 80 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 AASHTO AASHTO AASHTO mean braking mean cornering2stddev braking 2stddev cornering Speed (mph) Fr ic tio n va lu e (un itle ss ) Figure 31. Truck tire measurements of maximum wet-tire friction in longitudinal (braking) and lateral (cornering) directions (mean and two standard deviations below mean of the maximum friction supply). 30 40 50 60 70 80 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 AASH TO AASH TO AAS HTO mean braking mean cornering2stddev braking 2stddev cornering Speed (mph) Fr ic tio n va lu e (un itle ss ) Figure 30. Passenger vehicle tire measurements of skidding wet-tire friction in longitudinal (braking) and lateral (cornering) directions (mean and two standard deviations below mean of the skidding friction supply). 30 40 50 60 70 80 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 AASHTO AASHTO AASHTO mean braking mean cornering 2stddev braking 2stddev cornering Speed (mph) Fr ic tio n va lu e (un itle ss ) Figure 32. Truck tire measurements of skidding wet-tire friction in longitudinal (braking) and lateral (cornering) directions (mean and two standard deviations below mean of the skidding friction supply). 30 40 50 60 70 80 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 AASHTO AASHTO AASHTO mean braking mean cornering 2stddev braking 2stddev cornering Speed (mph) Fr ic tio n va lu e (un itle ss )

49 4.2.3 Summary of Key Results from Step 1 The results shown in Figures 29 to 32 allow comparisons between road friction measurements and the maximum side friction, fmax, used in the current AASHTO design policy for horizontal curves. The friction supply curves for both the lateral (cornering) and longitudinal (braking) directions for both passenger vehicles and trucks are higher than the maximum friction demand curves given by AASHTO policy. Thus, current horizontal curve design policy appears to pro- vide reasonable lateral friction margins against skidding. The lowest friction supply curves in Figures 29 to 32 correspond to trucks in skidding conditions on roads whose friction is estimated to be two standard deviations below the mean; but even in these cases, the friction supply curve is at least 0.25 to 0.3 above the AASHTO maximum side friction. These figures already suggest a finding that is supported in later sections of the report: if there is going to be an area of concern based upon AASHTO’s current design policy, it will likely arise pri- marily from the interaction of braking and cornering forces. It is also worth noting that in most cases, the differences between the friction supply curves and demand side friction curves increase with speed, and the friction supply curves are generally the same shape as the maximum side friction curves assumed by AASHTO for horizontal curve design. Finally, there is no indication that friction values vary in a consistent manner based upon location within a curve (e.g., upstream of the curve, at the PC, and within the curve). 4.3 Step 2: Define Road Geometries and Variable Ranges for Use in Subsequent Steps The objective of Step 2 was to define the range of superele- vations, horizontal curve radii, side friction levels, and grades to be considered in the analytical and simulation modeling analyses. Table 3-7 (Minimum Radius Using Limiting Values of e and f ) in the 2011 Green Book provides a range of design values for consideration in this research. For example, design speeds range from 10 to 80 mph in 5 mph increments. Maxi- mum superelevation ranges from 4% to 12%, in increments of 2%; and the maximum side friction factor ranges from 0.08 to 0.38. Current AASHTO policy also indicates some adjustment in superelevation rates should be considered for grades steeper than 5%. At minimum, it was important to investigate the full range of design values to sufficiently address the scope of this research and investigate design val- ues that deviate from the norm to address potential concerns and/or modifications to the existing policy. 4.3.1 Analysis Approach Table 20 illustrates the range of design values considered in the analytical and simulation modeling procedures. Basically, minimum-radius curves on grades of 0% and 4% to 9% in 1% increments were designed for design speeds of 25 to 85 mph, in 5 mph increments; for superelevation rates of 0% and 4% to 16%, in 1% intervals; and for side friction factors from 0.08 to 0.23 (as defined in Table 3-7 in the Green Book). For 85 mph, a side friction factor of 0.07 was assumed. Similarly, horizontal curves designed with curve radii of 0.8 Rmin were analyzed. In addition to analyses of the hypothetical geometrics, the hori- zontal and vertical alignments and cross slopes of the 20 field sites (see Table 5) were fully defined for analysis purposes. For analysis of the hypothetical geometries, speeds/ decelerations of the vehicles also had to be defined. Four speeds/deceleration levels were selected for analyses: • No deceleration (0 ft/s2; i.e., constant speed) • Curve-entry deceleration equivalent to -3 ft/s2 based upon typical deceleration rates when entering a horizontal curve (see Section 3.3.3) • Deceleration rates used in calculating stopping sight dis- tance (i.e., -11.2 ft/s2) • Deceleration rates assumed for emergency braking maneu- vers (i.e., -15 ft/s2; analyzed for select cases) For analyses of the 20 field sites, speed distributions of vehicles collected in the field were used (see Section 3.2). For the variations in the minimum design radius, reduc- ing the design radius from Rmin to 80% of Rmin can either be analyzed as a geometric change, a speed change, or a friction change. This is best understood by considering the point-mass model on which the AASHTO policy is based and considering Table 20. Range of design values for analytical and simulation modeling. Variable input parameter Range R Rmin, 0.8Rmin V 25 to 85 mph (5-mph interval) e 0%, 4% to 16% (1% interval) G 0%, 4% to 9% (1% interval) (downgrades and upgrades) ax Four levels of deceleration (0, −3, −11, and −15 ft/s ) 2

50 a horizontal curve with no superelevation. From Equation 7, the relationship between radius, speed, and friction will be approximately: = + 0.01 (20) 2V gR f e From this equation, if the speed is kept the same while the radius is reduced by 80%, then the acceleration would be bal- anced if the demanded friction and superelevation are both increased by a factor of 1/0.8, or 1.25. Similarly, on the left-hand side of the equation, the effect of decreasing the radius by 0.8 and keeping the speed fixed is equivalent to increasing the speed by 11.8% (the square-root of 1/0.8) and keeping the radius fixed. For horizontal curves without superelevation, the right-hand side of the equation is simply the demanded friction. In this case, a decrease in radius by 0.8 requires an increase in demand friction by 1.25. This, in turn, corresponds to forcing a system- atic downward shift of the friction margins for the Rmin case to the 0.8Rmin. Thus interpretation of the reduced-radius case can be used to additionally understand outcomes for situations of overspeed, low superelevation, or reduced friction margins. Although not explicitly indicated in Table 20, the analyses primarily focus on sharp horizontal curves on steep down- grades; however, consideration is also given to sharp horizon- tal curves on steep upgrades in Section 4.12. 4.3.2 Summary of Key Results from Step 2 The primary purpose of this step was to define the full range of design values for consideration in the analytical and simulation modeling. The range of design values selected for detailed investigation include the following: • Speed: 25 to 85 mph (and actual speeds measured at the study sites) • Superelevation: 0%, 4% to 16% • Grades: 0, 4% to 9% • Curve radius: minimum curve radii (Rmin) based upon cur- rent AASHTO policy (and curves with radii of 0.8Rmin) • Deceleration: 0, -3, -11.2, and -15 ft/s2 4.4 Step 3: Develop Side Friction Demand Curves and Calculate Lateral Friction Margins against Skidding Considering Grade Using the Modified Point-Mass Model The objective of Step 3 was to develop side friction demand curves for hypothetical geometries covering the full range of design values defined in Step 2 using the modified point-mass model and calculate lateral friction margins against skidding considering the friction supply curves ( ftire-pavement) developed in Step 1. Using the modified point-mass model, the calcu- lated side friction factors account for grade and vehicle decel- eration on the curve. The adjusted side friction factors were compared to the friction supply curves from Step 1 to esti- mate the lateral friction margins against skidding. 4.4.1 Analysis Approach The point-mass model (see Section 2.1), which serves as the basis for horizontal curve design, was modified to account for the effects of grade and deceleration. For a given curve radius, superelevation, grade, and design speed, physics is used to cal- culate the tire force utilization for steady driving. This is done via a force balance on the point mass, while using a simple friction ellipse representation of the tire to define skidding events. To develop side friction demand curves, a modified point-mass model was derived for a vehicle traversing a down- grade with superelevation. The assumption of small angle rep- resentation (i.e., cos q = 1 and sin q = q) is made to maintain simplicity within equations. The free body diagrams for the point-mass model are shown in Figure 2 for the lateral direc- tion, and in Figure 33 for the longitudinal direction. In Figures 33 and 2, Fb and Fc represent the braking and cornering forces acting on the vehicle point mass while g and a represent the grade and superelevation angles, respectively. The deceleration, ax, is directed along the vehicle’s longitudi- nal axis. After applying a force balance using Newton’s second law for a body rotating with angular velocity around a curve with constant radius, R, the three governing equations for vehicle motion in the X-, Y-, and Z-directions can be obtained as follows (Varunjikar, 2011). = − 100 (21)xF ma mg G bBraking Equation: = − 100 (22) 2 F m V R mg e cCornering Equation: = (23)N mgWeight Balance Equation: Figure 33. Longitudinal forces acting on a vehicle point-mass model. X Z Fb N W = mg γ

51 These equations can be simplified by substituting Equa- tion 23 into Equations 21 and 22, and then simplifying the result using the friction factors from Equations 13 and 14 to obtain: = − 100 (24)xf a g Gx i = − 100 (25) 2 f V g R e y Here, the terms fx and fy represent the friction demand in the braking (longitudinal) and cornering (lateral) directions. These depend on ax, which is the braking-induced decelera- tion; g, the gravitational constant; G, the road grade (which is negative for downgrades); V, the vehicle forward speed; R, the curve radius; and e, the road superelevation (positive values lean the vehicle to the inside of the curve). Comparing Equations 24 and 25 to Equation 10 used by AASHTO, Equation 25 is equivalent, while Equation 24 adds an additional equation for the longitudinal friction factor. This point-mass section is restrained to constant curves, i.e., curves with a minimum constant-radius design, so the radius, Rmin, is given by Equation 9. If Equation 9 is substituted into Equation 25, fy = fmax, Equation 25 implies that the side fric- tion demand is independent of the superelevation, grade, or braking demand. However, both grade and braking decel- eration influence longitudinal friction demand fx through Equation 24, which in turn reduces the overall lateral friction margin through Equation 18. Thus, in the absence of brak- ing forces, this point-mass vehicle will have the same lateral friction margins for each superelevation and grade. With the addition of braking forces, however, the conditions change slightly as the total friction demand of a point-mass model for a vehicle is represented by fx and fy together. 4.4.2 Analysis Results Plots of friction supply and lateral friction margins are shown in Figures 34 and 35 for passenger vehicles for a range of grades and design speeds, assuming a superelevation of 8% and constant speed. In Figure 34, the effective lateral supply friction values (Equation 17) are plotted versus the AASHTO design friction values. For the point-mass model, the lateral friction demand is equal to the AASHTO design friction for minimum-radius curves. In Figure 34, the deceleration of the vehicle is zero, meaning that braking is applied at a level suffi- cient to prevent the vehicle from accelerating down the grade. Both the mean lateral friction supply and the lower-bound lateral friction supply (mean minus two standard deviations, e.g., the 2-sigma values) are shown to illustrate the statistical range in friction supply. In Figure 35 the lateral friction margins are plotted for the same situations. The friction margin is simply the dif- ference between the lateral friction demand and the effective lateral friction supply. For this case (e.g., the modified point- mass model), the lateral friction margins increase slightly with speed. Throughout nearly all the results that follow, the mean lateral friction margin is roughly 0.12 higher than the 2-sigma lateral friction margin, and so only the 2-sigma lat- eral margin is shown in the plots hereafter. Figure 34. Lateral friction factors from modified point-mass model for passenger vehicle (G  0% to 9%, e  8%) (ax  0 ft/s2). 30 40 50 60 70 80 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 AASHTO AASHTO AASHTO 2-sigma lateral supply mean lateral supply Speed (mph) Fr ic tio n Fa ct or Figure 35. Lateral friction margins from modified point-mass model for passenger vehicle (G  0% to 9%, e  8%) (ax  0 ft/s2). 30 40 50 60 70 80 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 2-sigma lateral margin mean lateral margin Speed (mph) Fr ic tio n M ar gi n

52 For different braking values, the lateral friction margins change because the braking forces utilize some of the reserve lateral friction available. Three decelerations levels (0, -3, and -11.2 ft/s2) are shown in Figure 36 for passenger vehicles and in Figure 37 for trucks. The grades range from 0% to -9% (downgrade) and, to illustrate the effects of superelevation, lateral friction margins are shown for superelevations of 0% and 16%. These figures illustrate that, for the modified point- mass model, lateral friction margins decrease with increased braking and the addition (or lack) of superelevation has no effect on the lateral friction margins. This result may seem counterintuitive, but the primary influence of superelevation for the modified point-mass model is to change the mini- mum radius. Thus, the effect of superelevation is negated by the respective flattening or tightening of the curve radius. 4.4.3 Summary of Key Results from Step 3 Key findings from Step 3 are as follows: 1. Lateral friction margins decrease substantially with increased braking, and also decrease slightly with increasingly steeper downgrades. 2. Current AASHTO policy provides increasing lateral fric- tion margins for increasing speeds for both passenger vehi- cles and trucks. Results presented in later sections show Figure 36. Lateral friction margins from modified point-mass model for passenger vehicles (G  0% to 9%, e  0 and 16%) (ax  0, 3, and 11.2 ft/s2). Figure 37. Lateral friction margins from modified point-mass model for trucks (G  0% to 9%, e  0 and 16%) (ax  0, 3, and 11.2 ft/s2).

53 that this might not apply to more realistic (i.e., complex) vehicle models. 3. In general, trucks have a lateral friction margin about 0.06 less than passenger vehicles, simply due to differences in the tire friction curves. 4. The -3 ft/s2 deceleration case on a level road (0% grade) cor- responds roughly to the zero deceleration case for a grade of -9%; these two curves overlap. In other words, if a driver was trying to maintain a constant speed while approach- ing an unfamiliar downgrade section and was expecting a downgrade of no more than 9%, the expected behavior would be to hit the brakes immediately prior to the down- grade. The amount of lateral friction margin utilized under this situation is consistent with -3 ft/s2 deceleration on a level road. The Bonneson (2000b) study, as well as the measured variation in driver decelerations throughout downgrades that this work measured via the instrumented vehicle, sug- gests that drivers are comfortable with these friction margins. Roadway designs that necessitate deceleration requirements outside the usual variations seen in both this study and by Bonneson (2000b)—e.g., grades outside of 9% magnitudes— may require additional levels of caution and driver warning. 5. For the -11.2 ft/s2 deceleration case, the friction utiliza- tions are all the same, regardless of grade. This is because the stopping sight distance deceleration is assumed in the AASHTO Green Book to vary with grade. This varia- tion nullifies grade’s influence on friction margins. The deceleration value used in simulation was modified as per AASHTO stopping sight distance deceleration policy; spe- cifically, the deceleration used in the actual simulation, ax′, is given by: ′ = − 100 (26)x xa a g G Where G is defined as a positive number representing downgrade, and g is the gravitational constant. This stop- ping sight distance deceleration formula used by AASHTO is based on a simplistic vehicle dynamics model, which ignores potentially important effects like weight transfer, tire load sensitivity due to said weight transfer, and the static weight and friction demand differences between individual axles and tires on a vehicle. 4.5 Step 4: Define Vehicles and Maneuvers to Use in Non-Point-Mass Models The objective of Step 4 was to define the family of vehicles and range of maneuvers (e.g., lane changes, deceleration lev- els) to be considered in subsequent analyses by models other than the modified point-mass model. 4.5.1 Analysis Approach Six classes of vehicles were selected for consideration in subsequent analyses: • Passenger vehicles – E-class sedan (i.e., mid-class sedan) – E-class SUV (i.e., mid-size SUV) – Full-size SUV • Trucks – Single-unit truck – Tractor semi-trailer truck – Tractor semi-trailer/full-trailer truck (double) These vehicle classes were selected because they represent a high proportion of vehicles in the current vehicle fleet, because of their operating characteristics, and in particular because of their propensity for involvement in rollover crashes. In addi- tion, these vehicle classes are commonly incorporated in vehi- cle dynamic simulation packages. Tractor semi-trailer trucks with an attached tanker trailer were not specifically considered in the simulation analyses because existing vehicle dynamics models do not have the capability to simulate the dynamic effects of liquid sloshing in a tank trailer. To “define” a vehicle, each of the models requires a number of vehicle input parameters. A set of vehicle parameters rep- resentative of general vehicle classes were defined through a combination of literature review and default values found in the vehicle dynamics software. The range of input parameters needed for simulation analyses included the following: 1. Inertia properties: mass, z-axis mass moment of inertia about the center of gravity (CG) of the total vehicle, mass of payloads for trucks 2. Dimensions: wheelbase, CG height, distances from CG of sprung/unsprung mass to front/rear axle along x-axis, track width, and location of payloads and hitch points on trucks 3. Suspension: The natural frequency and damping ratio of the vehicle in pitch (Note: results for the bicycle models in Section 4.8 showed that suspension did not have an appre- ciable effect on friction margins.) Appendix B includes the vehicle input parameters selected for use in the simulation modeling. As indicated in Step 2, four deceleration levels were consid- ered to resemble various driving conditions for steady-state and transient behavior for use in non-point-mass models. These maneuvers each provide the braking force required to simulate constant speed (0 ft/s2), natural speed reduction upon curve entry (-3 ft/s2), stopping sight distance deceleration (-11.2 ft/s2), and emergency braking situations (-15 ft/s2). By increasing the amount of brake force, it will decrease the force

54 available in the lateral direction and thus decrease the corner- ing abilities of the vehicle. Speed distributions of vehicles col- lected in the field were also used to confirm that simulations were using a range of speeds similar to those measured from vehicles at the actual field sites. To initiate a braking scenario, the simulated vehicle is initial- ized so that it operates in a steady-state cruising situation. For most analyses that follow, the vehicle enters the horizontal curve from a tangent section. The initial constant speed is the design speed for the curve. The vehicle brakes and a step steering input is applied at various points in the curve. Since the deceleration, ax, is assumed to be constant, the braking inputs are found using a brake-proportioning model that rapidly changes the braking forces to match deceleration. Because suspension dynamics are ignored in the bicycle models that follow, the weight shift due to deceleration is assumed to be rapid compared to the vehicle’s motion through the curve. In the multibody models, the sus- pension dynamics are included and considered. To determine the worst portion of the curve to initiate a brake maneuver, a set of simulations was performed using a transient bicycle model for an E-class SUV cruising at design speed of 60 mph on the tangent section and then entering the curve around t = 2 s with a constant deceleration rate of -3 ft/s2. The steady braking was initiated at different portions of the curve for each simulation: • Case 1: brakes applied after the vehicle enters steady state on the curve • Case 2: brakes applied after the vehicle enters the curve but before it reaches steady state • Case 3: brakes applied at the same time as steering input initiated entering the curve The results for each case were very similar, but the maxi- mum lateral friction demand was obtained when the vehicle brakes after reaching steady state (Case 1). These results were confirmed as well in Step 7 (see Section 4.8). In the sections that follow, when braking maneuvers are applied, they are applied well after the onset of the curve unless otherwise noted. A common lane-change maneuver was also considered for analysis in later sections. Initially traveling at steady state on a curve at the design speed, the vehicle moves from a low- speed lane to a high-speed lane at a constant speed as shown in Figure 38. It was assumed that the curve was to the left, and therefore, the lane change was toward the inside of the curve. A lane width, l, of 12 ft is assumed for analysis purposes. The steering input used for the lane-change simulations is one sine wave with a time period of ts. This sine wave steering input is applied in addition to the nominal steering input, dcurve, required for traveling on a curve as shown in Figure 39. Data on lane-change maneuvers were also collected as part of the speed and vehicle maneuvers studies (see Section 3.2); and in particular, the duration of lane-change maneuvers measured in the field were considered in Steps 7 through 10 of the simulation modeling. 4.5.2 Summary of Key Results from Step 4 In this step the primary vehicle input parameters were selected for use in the simulation modeling. Appendix B provides more detail on the vehicle input parameters. Also it Figure 38. Lane-change maneuver. Figure 39. Steering input for lane-change maneuver.

55 was determined that the maximum lateral friction demand is required when the vehicle brakes after reaching steady state (Case 1). Therefore, in subsequent analyses investigating braking maneuvers, the brakes are applied well after the onset of the curve unless otherwise noted. 4.6 Step 5: Predict Wheel Lift Using Quasi-static Models The objective of Step 5 was to find the static rollover thresh- olds for the six vehicle classes included in this study to check if vehicle maneuvers at design speeds on downgrades with curves could induce wheel-lift events for a given road geom- etry considering horizontal curvature, grade, and supereleva- tion. Because roadway design is focused on providing low levels of side friction demand for vehicles relative to the maxi- mum side friction supply at the tire–pavement interface, it is possible that a vehicle could experience wheel lift prior to a skid event occurring. This step is aimed at predicting wheel lift for a vehicle traveling on a curve using quasi-static models. 4.6.1 Analysis Approach In this step, a rollover model to predict wheel lift was devel- oped to account for the effect of superelevation. The predic- tion of wheel lift involves expressing the rollover threshold of the vehicle using laws of mechanics. For roads without super- elevation or grade, the rollover threshold for a rigid-vehicle model using quasi-static analysis is = 2 (27)f T h rollover where T is the track width and h is the CG height (Gillespie, 1992). This is a well-known and classic result, but it does not include superelevation effects or suspension effects. To include superelevation and suspension within the clas- sical analysis, this step involved the following tasks: 1. Derive the quasi-static rollover model for a rigid and/or suspended vehicle accounting for superelevation. 2. Find rollover threshold for each representative vehicle and compare it with the lateral accelerations obtained from the modified point-mass analysis in previous steps. 3. Identify those roadway conditions, for further investiga- tion, where the lateral accelerations generated are higher than the rollover threshold. The static rollover/wheel-lift predictions do not directly depend on the tire–pavement friction. However, this method will indicate whether a wheel-lift event or a skidding event will occur first as vehicle speed increases. For example, if the wheel- lift threshold for lateral acceleration is higher than the friction limit, then skidding will take place before wheel lift. Further, if the wheel-lift threshold is significantly higher than the actual lateral acceleration necessary to negotiate the curve, then again wheel lift is unlikely during normal maneuvers on a curve. The quasi-static rollover model for use on superelevated roads is based on a static force balance on a simplified rep- resentation of a vehicle, which includes only a rudimentary representation of suspension effects. The approach is nearly identical to the point-mass, rigid-vehicle model analysis that produces Equation 27, except superelevation is considered and the roll axis of the vehicle is added. The setup of the model is shown in Figure 40 which illustrates the rear view of a suspended vehicle traversing a curve to the right. Figure 40 shows the forces acting on the suspended vehicle. Due to lat- eral load transfer, the normal load on the outside wheel, Fzo, increases. This can be associated with the sprung mass rolling with a lateral shift in the CG toward the outside of the curve. The sprung-mass CG rotates about a point called the roll cen- ter, whose position depends on the suspension geometry. For the analysis, it is assumed that the roll-center position: • Does not change, • Is aligned with the center of the vehicle, and • Is a fixed height above the road surface. The parameters in the static rollover model shown in Fig- ure 40 are defined as follows: h = Height of sprung-mass CG hr = Height of roll center T = Track width f = Roll angle Fzi = Normal load on inner tires Fzo = Normal load on outer tires Fyi = Lateral force on inner tires Fyo = Lateral force on outer tires Figure 40. Static rollover model modified to include superelevation.

56 This model associates the rollover event with the onset of wheel lift, characterized by the normal load on the inside wheels going to zero (Fzi = 0). It can be assumed that Fzi ≈ 0 just before wheel lift occurs. Balancing the moments about the outer tire contact point, i i i i i ∑ ( ) ( ) ( ) ( ) = − − α  + − − φ α = sin 2 cos 0 (28) 2 M h m V R mg T h h mgr Substituting ay = V 2/R into Equation 28, and using the small angle approximation yields: i i( )( )− − α  + − − φ =2 0 (29)h ag T h hy r and therefore: i ( ) + − φ = + 2 100 (30) a g h h h T h ey r For steady-state analysis, the roll angle of the vehicle body can be written as a roll gain, in rad/g, multiplied by the lateral acceleration in g’s (e.g., iφ = φR a g y ). Substituting this expression, Equation 30 can be rewritten in the final form used in this study: i i i( ) ( ) + − = + = + + − φ φ 2 100 , or 2 100 1 1 (31) a g h h h R a g T h e a g T h e h h R y r y y r Equation 32 gives the rollover margin based on lateral acceleration, which represents the difference between the maximum lateral acceleration allowable before wheel lift and the curve-induced lateral acceleration, ay.    i( )= + + − − φ 2 100 1 1 (32)RM T h e h h R a g ay r maximum steady acceleration prior to wheel lift y normalized acceleration within curve This rollover margin is for a rigid vehicle considering a simple suspension model and the superelevation of the roadway. From Equation 32, a few observations can be made imme- diately. First, since Equation 32 only depends on lateral forces, the grade of the road has no effect on wheel lift, nor does speed influence the rollover threshold. For vehicles with suspension, the worst-case conditions are those vehicles with a high roll gain (Rf) and low roll axis compared to CG height (e.g., hr/h is close to zero or even negative). This agrees with intuition, as these assumptions represent top-heavy vehicles with “soft” suspensions. Further, if one assumes a rigid vehicle without suspension (e.g., Rf = 0), then Equation 32 becomes: ( )= + −2 100 (33)RM Th e agay y For this rigid-vehicle model, increasing superelevation directly shifts the rollover threshold upward. This agrees with intuition, as well as the current AASHTO design policy which allows tighter curve radii in the presence of higher superel- evation. For a vehicle without any suspension roll and on a road without any superelevation, the rollover threshold por- tion of Equation 33 reduces to T/2h, which agrees exactly with Equation 27. To develop approximate estimates of how a suspension- vehicle model will differ from a rigid-vehicle model, an approx- imate value of Rf = 0.17 rad/g was assumed given the fact that most vehicles exhibit approximately 1° of roll per 0.1 g of lat- eral acceleration (10°/g corresponds to 0.17 rad/g). The ratio of hr/h is generally between 0.25 and 0.75 for most vehicles, but a worst-case value would be to set this ratio to zero. Similarly, the worst-case road is one without superelevation. Therefore, the worst-case rollover margin would be approximately: ( ) ( )≈ ++ − ≈ − ≈ −φ2 1001 11.17 2 0.85 2 (34) RM T h e R a g T h a g T h a g ay y y y 4.6.2 Analysis Results Sample rollover thresholds (T/2h values) for the vehicles used in this study are given in Table 21 considering a super- elevation of 4%. For trucks, these margins are given for their trailers, as the trailer margins are far lower than the tractor; however, the trailer can be loaded in an infinite number of configurations, resulting in a wide range of margins that could potentially be achieved. The maximum side friction ( fmax) rec- ommended by AASHTO policy, for the speeds considered in this study, ranges from 0.07 for a design speed of 85 mph to 0.23 for a design speed of 25 mph. Comparing the rollover thresholds to AASHTO’s maximum side friction values for

57 design speeds of 25 mph or greater, it can be deduced that wheel lift will not occur for passenger vehicles or trucks driving at the design speed on a curve designed according to current AASHTO policy. Note, though, that at design speeds below the scope of this research (i.e., design speeds of 10 and 15 mph), the maximum side frictions according to current AASHTO policy are 0.32 and 0.38, respectively. The rollover margin at these low design speeds is still positive, but the margin is decreasing with speed because the maximum side friction ( fmax) recommended by AASHTO policy increases with speed. Note also that Harwood et al. (2003) reported conserva- tive (worst-case) rollover thresholds for trucks to be approxi- mately 0.35. Assuming 85% of this value gives approximately 0.30 for the estimated rollover threshold, when accounting for superelevation and suspension. Thus, at lower design speeds rollover becomes more of a concern. For tractor semi-trailers, the truck configuration and type of cargo influences the vehicle’s roll stability. The effects of liquids in cargo tank trucks are of particular concern. While detailed simulations of fluid–vehicle interaction is beyond the scope of this research, previous work provides good approximations of rollover thresholds suitable to estimate situations that may lead to the onset of a rollover. Notable work includes that of Gillespie and Verma (1978) who found that lateral accelera- tion at wheel lift was 0.36 for liquid-cargo tank trucks (due to their higher CG and different suspension) versus 0.54 for the typical tractor semi-trailer—a value similar to the 0.56 value found in modern simulations and studies (Table 21). Their work also noted that liquid-cargo tank trucks were much more susceptible to rollover due to rearward amplification effects. A comprehensive study of slosh dynamic effects was conducted by Ervin et al. (1985) to assist in federal rule making for liquid- cargo transport. They found that, of 30 reported crashes from California data, 22 crashes occurred during transport of under- filled cargo containers; definitive cause/effect relationships between liquid-cargo motion and vehicle rollover, however, could not be established. Subsequent analyses revealed that some rollover cases would have occurred even for rigid-cargo motion. Ervin et al. note that peak liquid force amplitudes were 2 to 3 times larger for liquid cargo than for the same mass of rigid cargo. These amplification factors closely agree with the amplification factor of 2 numerically computed by Modaressi- Tehrani et al. (2007). To quantify the effect of liquid-cargo influence, Evrin et al. examined the difference in lateral accelerations resulting in overturn. These results indicate that liquid-cargo tank trucks may have rollover thresholds that are half of comparable rigid- cargo rollover thresholds. However, the minimum lateral roll- over thresholds for liquid-cargo tank trucks are nearly always the same as that of an empty tanker. All lateral rollover thresholds were 0.25 to 0.30, which are similar to the rollover thresholds assumed by Harwood et al. (2003) for truck rollover stability when accounting for superelevation and suspension effects. Ervin et al. (1985) also note that the worst-case lateral slosh frequencies are between 0.5 and 0.8 Hz, with lower frequencies corresponding to less-full cases. The effect of sight distance on a vehicle’s excitation at various frequencies is also consid- ered, with results showing that roads with more limited sight distance will tend to cause more excitation at frequencies of liquid-cargo resonance (between 0.2 and 0.4 Hz) versus typical steering input excitations for roads with unrestricted sight dis- tance, which tend to contain frequencies around 0.15 Hz. For these oscillation frequencies, Ervin et al. note that lateral accel- erations of 0.25 or less will generally not cause overturn based on a harmonic analysis. Again, these results are in agreement with the experimental results presented earlier and assump- tions by Harwood et al. Further, both results suggest that the 50% full-loading condition is likely the “worst-case” loading situation for harmonic fluid motion. For braking, recent work by Biglarbegian and Zu (2006) showed that liquid-cargo tank trucks require approximately 30% more distance than rigidly loaded trucks due to weight-transfer effects of the fluid. Thus, the most conservative interpretation of the litera- ture on liquid-cargo tank trucks is to assume a lateral rollover threshold value of 0.30 or half a rigid vehicle’s nominal value, whichever is less. The lowest rollover threshold for trucks in Vehicle class Rollover threshold in ’s Adjusted rollover threshold (~0.85 T/2h)a E-class sedan 1.36 1.16 E-class SUV 1.10 0.94 Full-size SUV 1.22 1.04 Single-unit truck 0.87 0.74 Tractor semi-trailer truck 0.56 (trailer) 0.48 Tractor semi-trailer/full-trailer truck 0.56 (trailer) 0.48 a Rollover threshold (T/2h) = ~ 0.85 × rollover threshold in g’s. Table 21. Rollover thresholds (T/2h) for vehicles used in this research.

58 Table 21 is 0.56, and half this value is 0.28, which for practical purposes is the same as a rollover threshold value of 0.30. This suggests that the lateral accelerations in curves, and hence the maximum side friction values used for design, should be lim- ited to values less than 0.30. In general practice, a rollover threshold of 0.28 to 0.30 is particularly conservative since the default loading of most trucks is expected to have nominal T/2h values of approximately 0.56 as noted in Table 21. Fur- ther, liquid-cargo tank trucks in modern practice are gener- ally discouraged from carrying half-filled tanks, and thus the completely filled or empty tanks produce rigid-load behaviors that are generally more predictable and more in agreement with the 0.56 value than 0.30. The difference between the 0.56 rollover threshold value expected in practice, versus the 0.30 rollover threshold value based on the summary of previous research, suggests that there is conservatism added to the low- order model analysis that likely includes extreme cases (i.e., partially filled liquid-cargo tanks) as well as expected errors inherent in such a simple rollover vehicle model. It should also be noted that in the Comprehensive Truck Size and Weight Study (FHWA, 1995), an appendix states that crash data show so few fatalities with rollover thresholds less than 0.35 that rates cannot be calculated, suggesting that few vehicles on the road have rollover thresholds less than 0.35. 4.6.3 Summary of Key Results from Step 5 The following findings were obtained from the analysis in Step 5 focusing on roll margins for steady-state driving, e.g., driving without abrupt steering inputs that might excite tran- sient lateral accelerations: 1. For passenger vehicles, the rollover thresholds are far higher than the available friction on the road. There seems to be no concern of a passenger vehicle rolling over while traveling at the design speed on a curve designed accord- ing to current AASHTO policy. This is simply because the tires will skid before the rollover threshold is reached. 2. For trucks, the rollover thresholds are much lower. For design speeds greater than 30 mph, trucks are not expected to exhibit wheel lift under current AASHTO design pol- icy. At design speeds of 25 or 30 mph, AASHTO policy allows maximum side friction values that are nearer to the rollover thresholds for trucks, but not sufficient to cause rollover. For design speeds below the scope of this research (e.g., 10 and 15 mph), the rollover margins are still positive but are decreasing with speed. Thus, rollover for trucks is of more concern at lower design speeds than at higher design speeds. 3. Based upon a review of the literature, the lowest rollover thresholds for tanker trucks (i.e., liquid-cargo tank trucks) are in the range of 0.28 to 0.30. In later sections, multiaxle and multibody models are used to check individual axle and individual tire normal forces on passenger vehicles and trucks. These latter analyses supple- ment the steady-state analysis presented here to verify the results and to determine if transient maneuvers are sufficient to excite momentary wheel lift. 4.7 Step 6: Predict Skidding of Individual Axles during Steady-State Behavior on a Curve The objective of Step 6 was to identify whether steady-state axle forces obtained based on the steady-state bicycle model violate the available friction supply. Using a bicycle model with the vehicles classes chosen for study, an analysis was per- formed based upon steady-state behavior to determine force requirements on each axle. From the force requirements, fric- tion demand was inferred and compared to available friction supply (i.e., from Step 1). 4.7.1 Analysis Approach In this step, a steady-state bicycle model was developed to predict skidding of individual axles accounting for the effects of vehicle type, grade, superelevation, and deceleration. A primary criticism of the point-mass model is that it does not account for the per-axle force generation capabilities of a vehicle. The point- mass model used currently by AASHTO to determine expected friction demand adds the front- and rear-axle lateral forces to determine if a vehicle can maneuver through a curve. It does not check if one of the axles requires more or less friction rela- tive to the other. While the average of forces on each axle might not express skidding, one axle might be beyond the friction sup- ply limit, while another is well below the limit. Nearly all vehicles have different tire loads on the front and rear axles caused by the center of gravity of the vehicle not being located midway between the axles. For example, a typical passenger vehicle has an approximately 60/40 weight split from front to rear. When the vehicle is in a curve, this weight differ- ence means that the lateral forces required on the front axle are usually much different than those on the rear axle. Indeed, the lateral forces required on each axle are proportional to the mass distributed over each axle; thus, on a flat road (i.e., one with no superelevation or grade), the weight distribution on the tires is exactly the same proportion as the lateral forces required from each axle. This is beneficial to curves on level roads: the verti- cal forces pushing down on each axle are pushing hardest on the axles that most need cornering forces. The net effect is that, for level roads, the weight differences are generally ignored for friction analysis without much error. However, on grades and in cases where there is deceleration, the weight shift from the

59 rear to the front of the vehicle may significantly change the relative amounts of vertical tire force on each axle. If there is a curve on a grade, the cornering forces required from each axle remain proportional to the mass above each axle, not the weight. This difference between the mass-related cornering forces and the weight-related friction supply illustrates why curves on grades may be problematic for ensuring sufficient lateral friction margins. To calculate the effect of per-axle friction utilization, a common simplification in vehicle dynamics was assumed for this analysis: the vehicle is idealized as a rigid beam, and each axle is represented as a single tire situated at the midline of the vehicle. The resulting model is termed a “bicycle model” because of its appearance (see Figure 41). This classical bicycle model is typically used to study vehicle maneuvers on a flat road. One goal of this analysis was to expand this model to evaluate a steady turning maneuver taking into consideration the horizontal alignment, grade, and superelevation. The effects of constant braking were also included. This model was used to check the friction demand for each axle and to check if the friction supply generated by the tire–pavement is sufficient for cornering and/or braking. A number of assumptions were made for the steady-state bicycle model as follows: 1. The velocity changes slowly relative to the forward and turning motions, such that the speed is approximately constant over the maneuver analysis window (generally a few seconds). 2. The vehicle is assumed to be steered only by the front tires. 3. There is no lateral load transfer. 4. The vehicle is right/left symmetric. 5. The roll and pitch of the vehicle and tires are ignored, other than the steady contributions due to grade and superelevation. 6. Aerodynamics and rolling resistance of the tires are ignored. 7. The vehicle’s suspension is assumed to be stiff and non- moving throughout the curve. 8. The deceleration (if any) is assumed to be constant. 9. The vehicle is assumed to be driving forward down the road at a slip angle to the road that is small enough to ignore the sideways skidding of the vehicle. 10. The grade and superelevation angles are assumed to be small enough that small angle approximations can be used. In this and in later sections, the vehicle may be braking with a specified deceleration. This deceleration is specified in the vehicle’s coordinate system, and hence the equations of motion are most conveniently written in this frame of refer- ence. In previous sections, the equations of motion were writ- ten in a global frame of reference, and so to distinguish one reference frame from the other, lower-case x, y, and z are used hereafter to denote the vehicle’s coordinate system, while the upper-case X, Y, and Z denote the earth-referenced coordi- nate system. Both designations are shown in Figure 41. For a vehicle traveling steadily on a curve, the force bal- ances can be conducted in the local longitudinal (x-axis), lateral (y-axis), and vertical (z-axis) directions separately, as the motions for each will be orthogonal. The forces acting along each axis are shown in Figure 41. Using this figure’s force direction conventions, and small angle approximations where appropriate, in the longitudinal direction (braking), the governing equation is: = + = − 100 (35)F F F ma mg G b bf br x In the lateral direction (cornering), the governing equa- tion is: + = − 100 (36) 2 F F m V R mg e cf cr Finally, the weight balance equation gives: + ≈ (37)N N mgf r Equations 35 to 37 are similar to Equations 21 to 23 derived earlier for the modified point-mass model. The only differ- ence is that the steady-state bicycle model is derived from per- axle forces whereas the modified point-mass model uses only body-aggregated forces. From Equations 35 to 37 some preliminary observations can be formulated. First, the longitudinal friction demand depends on the grade and deceleration levels as shown in the braking equation. Thus, the lateral friction margins should change with both grade and braking effort. While it would appear that the cornering equation depends on supereleva- tion, in the case of steady-state driving on curves with the X Z Fbr W=mg γ Nr Fbf Nf Y X x y Fbf Fbr Fcr Fcf Figure 41. Forces acting on a vehicle in a steady turn on a superelevated curve with a downgrade.

60 AASHTO minimum curve radii, this is not the case. Equa- tion 36 for the steady-state bicycle model can be rewritten as: + = − 100 (38) 2 min F F m V R mg e cf cr If Equation 38 is compared to the AASHTO design equa- tion for minimum-radius curves, Equation 9, the two equa- tions can be combined to obtain: i+ = (39)maxF F mg fcf cr This result shows that the side forces on the vehicle follow- ing a minimum-radius curve depend only on the maximum side friction, fmax. This factor, according to AASHTO design policy, depends only on the design speed, not on supereleva- tion. Therefore, the only geometric design variable affecting cornering forces is the design speed. This makes the lateral friction demand independent of the superelevation for the steady-state analysis, e.g., the superelevation of the curve will not affect friction demand at all. To calculate the friction supply available to each axle, the normal forces on each axle must be known. The individual axle forces are obtained by the moment balance about the y-axis and z-axis. Shown in Figure 42, a moment balance about the y-axis direction (at front and rear tire contact point) yields the normal force on the front and rear axles on a downgrade while the vehicle is braking: ( )= + − 100 (40)N mg bL m a g G hLf x ( )= − − 100 (41)N mg aL m a g G hLr x The moment balance about the z-axis, shown in Figure 43, gives the ratio of front- and rear-axle cornering forces: = ⇒ + = + = (42) F F b a F F F b b a b L cf cr cf cf cr Therefore, the lateral (cornering) forces at the front and rear are given by: = −   = −   100 100 (43) 2 2 F b L mV R mg e F a L mV R mg e cf cr Using the formulas for the cornering forces and weights on each axle, the lateral friction factor expressions for each axle are: = =, (44)f F N f F N yf cf f yr cr r Substituting the expressions for side forces and weights on each axle, and noting the weight of the vehicle, W = mg, the closed-form expressions for the side friction factors per axle are: ( ) ( ) ( ) ( ) ( ) ( ) = = − + − = = − − − 100 100 100 100 (45) 2 2 f F N b L mV R W e W b L m a g G h L f F N a L mV R W e W a L m a g G h L yf cf f x yf cr r x These represent the quasi-static friction demands on the front and rear axles. To determine whether these friction demands exceed the friction supply, the friction ellipse of the tire is used to modify the friction supply by the amount of friction used for braking. To complete the analysis for the steady-state bicycle model, the prediction for the braking forces on each axle is required. A simple braking model is introduced to illustrate how brake forces are split between each axle. It is important to under- Figure 42. Moment balance about the y-axis for a vehicle braking on a downgrade. X Z Fbr W = mg γ Nr Fbf Nf My Figure 43. Moment balance about z-axis. Y X x y Fbf Fbr Fcr Fcf Mz a b L

61 stand that braking is sometimes (and intentionally) not dis- tributed equally between axles. Passenger vehicles typically use hydraulic brakes which transfer braking pressure from the controlling unit to the actual brake mechanism. Balancing the brake outputs on the front and rear axles is achieved by “pro- portioning” the brake pressure appropriately for the brakes installed on a vehicle (Limpart, 1999). The proportioning valve is a critical component in the brake system which acts to prevent rear tire skidding prior to the front tire skidding to avoid vehicle spin-out; at higher braking levels, it switches to cause more braking force to the front axle. To maintain consistency in notation and presentation, the brake system is presented here. In the simple model used in this analysis (see Figure 44), the braking torque is the product of brake pressure and brake gain for each axle. The brake force can be obtained by dividing the brake torque by the tire’s rolling radius, Rtire. i i = = 1 1 (46) F R G P F R G P bf tire f f br tire r r To avoid rear-axle lock-up that causes spin-out of a vehi- cle, the brake outputs are reduced at higher braking efforts by appropriately adjusting the braking pressures at the front and rear axles. In typical passenger vehicles, the brake pres- sure output for the rear axle is reduced to approximately 30% after a certain application pressure, Pa′. This reduction in brake pressure can be represented by the following equations: = = ≤ ′for (47)f r a a aP P P P P and for 0.3 f a a a r a a a( ) = > ′ = ′ + − ′ P P P P P P P P The values of the parameters involved in this brake- proportioning model are listed for the passenger vehicle classes in Table 22. The truck models simulated in this study do not have brake-proportioning valves. The above relationships relate to brake pressures, but not to brake forces. To be useful in the model, a relation- ship between brake force and brake pressure is needed. To derive this, first note that the net braking force, Fb, required for a decelerating vehicle is given by Equation 35, and the net braking force is Fb = Fbf + Fbr. The braking force distribution for the front versus rear axle depends on whether the appli- cation pressure, Pa, is greater or less than Pa′, the pressure at which the brake-proportioning valve begins to prevent rear wheel lock. The corresponding braking force, Fb′, when Pa = P a′ is given by: i( )′ = + ′1 (48)F R G G Pb tire f r a and the corresponding deceleration that initiates the brake- proportioning valve is given by: = ′ + 100 (49),a F m g G x p b Figure 44. Brake-proportioning flowchart. Vehicle class (ft-lbf/psi) (ft-lbf/psi) ′ (psi) (ft) E-class sedan 4.07 3.05 363 1.19 E-class SUV 4.07 3.05 290 1.26 Full-size SUV 5.09 3.56 290 1.32 Table 22. Per-axle brake-proportioning parameters for passenger vehicles.

62 Table 23 shows the decelerations at which each vehicle’s proportioning valve would initiate a reduction in rear tire braking force, for a level grade situation and for a 9% down- grade. Thus, of the four decelerations levels (0, -3, -11.2, and -15 ft/s2) considered throughout these analyses, Table 23 indicates that the two highest decelerations may cause brake- force redistribution to the rear tires through activation of the brake-proportioning valve. In general the stopping sight distance (-11.2 ft/s2) and emergency braking (-15 ft/s2) decelerations would not be considered “steady-state” driving situations, as the vehicles’ speed is changing too abruptly to satisfy the model assump- tions. However, the equations in this analysis are “steady” in that they assume constant terms in the equations, including decelerations, and thus they will give good estimates of nec- essary tire forces at the onset of the maneuver before speed changes significantly. These results are therefore included here despite the fact that they are not steady-state or constant- speed maneuvers. Additional discussion of emergency, tran- sient maneuvers is presented in Section 4.9. To relate brake pressure to brake-force distribution per axle, two cases have to be considered: 1. Fb ≤ Fb′ 2. Fb > Fb′ In the case of Fb ≤ Fb′, the brakes are lightly used and the brake- proportioning valve is not reducing the rear brakes to prevent lock-up. In this case the braking forces per axle are simply: i= 1 (50)bf tire f aF R G P and 1 .br tire a=F R G Pr i And hence, the braking forces are distributed according to the brake gain on each axle: F G G G Fbf f f r bi= + and i= + (51)F G G G Fbr r f r b In the second case, when the brake-proportioning valve is acting to reduce rear lock-up, the brake force is Fb > Fb′. In this case, the braking forces per axle must be determined by two different brake pressures, e.g., i= 1 F R G Pbf tire f f and i= 1 .F R G Pbr tire r r The values of Pf and Pr are different and can be found by first obtaining the value of the application brake pressure, Pa. The net braking force for this case is given by: i i( )= + = +1 (52)F F F R G P G Pb bf br tire f f r r Substituting the equation for the brake-proportioning valve: i i( )( )( )= + = + ′ + − ′1 0.3 (53)F F F R G P G P P Pb bf br tire f a r a a a Rearranging: i i( )( )= + + ′1 0.3 0.7 (54)F R G G P G Pb tire f r a r a which can be rearranged to solve for the brake pressure: i i = − ′ + 0.7 0.3 (55)P R F G P G G a tire b r a f r Once Pa is known, the per-axle braking forces can be found by using an equation for the brake-proportioning valve. Using the per-axle braking forces, the longitudinal friction factors can be found using their basic definitions: = =, (56)f F N f F N xf bf f xr br r Using the equations for the brake forces, the reduction in friction supply can be determined. The lateral friction supply factors are defined per axle in the same manner as described previously for the point-mass model in Equation 17: Front Axle: = −    1 (57)yf,supply y,max xf x,max 2 f f f f Rear Axle: = −    1 (58)yr,supply y,max xr x,max 2 f f f f Vehicle class , 0% grade , −9% grade E-class sedan −17.21 ft/s2 −14.31 ft/s2 E-class SUV −12.82 ft/s2 −9.92 ft/s2 Full-size SUV −10.92 ft/s2 −8.02 ft/s2 pxa , pxa , Table 23. Decelerations for passenger vehicles at which brake-proportioning valve activates.

63 When the longitudinal friction factor exceeds the lon- gitudinal friction supply, fx,max, the lateral friction supply is assumed to be zero. Using the above equations for the steady-state bicycle model, the friction demand and friction supply analysis is performed for each individual axle. If the lateral friction supply for the rear axle, fyr,supply, is less than the lateral friction demand, fyr, then the rear axle is likely to skid. This individual axle skidding may not be noticed in the point-mass model, and is the advantage of using the bicycle model over the point-mass model. 4.7.2 Analysis Results Figure 45 shows a comparison of the per-axle friction demand for a steady-state E-class sedan assuming -11.2 ft/s2 deceleration on the curve, for a road with no superelevation and a 9% downgrade. The three lines at the top of the fig- ure represent friction supply, and the four lines at the bot- tom of the figure represent friction demand. As expected, the point-mass friction demand agrees exactly with the AASHTO design friction curves, which agrees with intuition because they both utilize the same vehicle model. In Figure 45 two effects are occurring simultaneously that cause the steady-state model to have lower friction margins than the point-mass model: the rear demand is increasing while the rear supply is decreasing. Both are caused by brak- ing which causes a rear-to-front weight shift as predicted by Equations 40 and 41. The change in fx as predicted by Equa- tion 44 explains the reduction in the friction supply on the rear axle and increase in the supply on the front axle. The same weight shift changes the normal forces in the fy calcula- tion in Equation 44, with the result that the lateral demand is increasing. Thus, on the rear axle, braking and downgrades cause the friction supply to go down, while simultaneously increasing friction demand. Because load transfer depends on the mass properties of the vehicle, different vehicle setups will result in different per- axle friction demand. In the case of an E-class SUV, for exam- ple, the load-transfer effect is more pronounced for exactly the same conditions (Figure 46) due to higher CG height, h. Also, these figures are identical across different supereleva- tions; like the modified point-mass model, the steady-state bicycle model results are independent of the grade when stopping sight deceleration is considered. While Figures 45 and 46 illustrate the simultaneous change in demand and supply, the most important information is the difference between lateral friction supply per axle and lateral friction demand per axle which provides the lateral friction margin. Consistent with how the lateral friction margins are calculated for the point-mass model in Equation 16, the lat- eral friction margins for the bicycle model can be defined per axle as follows, for the front tire: = − (59), ,f f fmargin f yf supply yf And for the rear tire: = − (60), ,f f fmargin r yr supply yr In practice, however, the lateral friction demand can have both positive and negative values; hence, the absolute value 30 40 50 60 70 80 -0.1 0 0.1 0.2 0.3 0.4 Speed (mph) Fr ic tio n Fa ct or AASHTO Design Friction Point-Mass Supply Point-Mass Demand Front Axle Supply Front Axle Demand Rear Axle Supply Rear Axle Demand Figure 45. Friction factors for E-class sedan (G  9%, e  0%) (ax  11.2 ft/s2).

64 of the demand is more appropriate. Thus, the lateral friction margins are formulated as: = − = − (61) , , , , f f f f f f margin f yf supply yf margin r yr supply yr To illustrate how both grade and braking effort change the friction margins, Figure 47 shows the lateral friction margins for an E-class sedan and an E-class SUV. For both vehicles, the effect of grade is to decrease lateral friction margins at each braking level, except for the stopping sight distance decelerations (-11.2 ft/s2) as these decelerations reduce with increasing grade per AASHTO guidelines. The largest factor, however, is the level of braking effort applied. As the brak- ing effort increases, the friction margins drop to where, for emergency braking levels (decelerations of -15 ft/s2), they can become negative. Figure 47 shows that, for the E-class sedan, the use of brak- ing increases the detrimental effects of grade. For example, with no braking (ax = 0 ft/s2), each percent change in grade Figure 46. Friction factors for E-class SUV (G  9%, e  0%) (ax  11.2 ft/s2). 30 40 50 60 70 80 -0.1 0 0.1 0.2 0.3 0.4 Speed (mph) Fr ic tio n Fa ct or AASHTO Design Friction Point-Mass Supply Point-Mass Demand Front Axle Supply Front Axle Demand Rear Axle Supply Rear Axle Demand Figure 47. Lateral friction margins for E-class sedan and E-class SUV (G  0% to 9%, e  0%) (ax  0, 3, 11.2, and 15 ft/s2). 30 40 50 60 70 80 −0.1 0 0.1 0.2 0.3 0.4 Speed (mph) Fr ic tio n M ar gi n −0.1 0 0.1 0.2 0.3 0.4 Fr ic tio n M ar gi n Grade = 0% Intermediate grades Grade = −9% 30 40 50 60 70 80 Speed (mph) Grade = 0% Intermediate grades Grade = −9% Constant Velocity, ax = 0 ft/s2 Curve Entry Deceleration, ax = −3ft/s2 SSD Deceleration,ax = −11.2 ft/s2 Emergency Decleration, ax = −15 ft/s2 Vehicle: EclassSedan, e = 0% Vehicle: EclassSUV, e = 0% SSD Deceleration,ax = −11.2 ft/s2 Curve Entry Deceleration,ax = −3 ft/s2 Constant Velocity, ax = 0 ft/s2

65 decreases the lateral friction margin by approximately 0.001. For curve-entry deceleration (ax = -3 ft/s2), the effect of grade is to decrease lateral friction margin by 0.002 per percent of grade, approximately. For emergency braking decelera- tion (ax = -15 ft/s2), the lateral friction margin decreases by 0.02 per percent of grade decrease. Note that stopping sight distance (SSD) decelerations are not affected by grade, but this is because the actual decelerations vary by grade per AASHTO policy. Figure 47 also illustrates that individual vehicles experience different lateral friction margins. For example, the E-class sedan is able to maintain much higher lateral friction margins, even positive margins for much of the grade situations, whereas the E-class SUV has emergency braking friction margins that are all negative (below -0.1). In the sections that follow, the steady-state bicycle model results are compared to results from more complex models. These comparisons include additional vehicles not shown here, for example the full-size SUV and trucks. 4.7.3 Summary of Key Results from Step 6 In summary, the following findings were obtained from the analysis in Step 6 that examined the steady-state bicycle model predictions of friction margins: 1. If AASHTO design policy is used for curvature design, and the vehicle is following the curve at the design speed, the equations of motion predict per-axle tire forces will change only with design speed, not with changes in supereleva- tion or grade. Thus, superelevation- and grade-induced changes in lateral friction margin will occur only due to changes in the tire’s normal force and braking inputs. 2. The effects of the brake-proportioning valve built into most passenger vehicles do not activate at the curve-entry decel- eration rates considered in this study. However, the valve does activate at much lower levels on downgrades than on level roads and may in fact activate during stopping sight distance decelerations as well as emergency braking. 3. The steady-state bicycle model predicts friction supply and demand that are very similar to the point-mass model in that the lateral friction margins increase with design speed, namely because the demand at higher speeds drops faster than the supply at higher speeds. 4. Due to weight shift on downgrades and decelerations, the steady-state bicycle model predicts that the front-axle supply is always higher than the point-mass model and the demand is lower. The reverse is seen on the rear axle. Thus, the front-axle margins are nearly always better than predicted by the point-mass model, and the rear tire is nearly always less. Thus, the rear axle of passenger vehicles nearly always has the lowest lateral friction margin. 5. Different vehicles have different lateral friction margins from the steady-state bicycle model. 6. The use of braking increases the detrimental effects of grade. For example, with no braking (ax = 0 ft/s2), each per- cent change in grade decreases the lateral friction margin by approximately 0.001 for the E-class sedan. For curve- entry deceleration (ax = -3 ft/s2), the effect of grade is to decrease the lateral friction margin by 0.002 per percent of grade, approximately. For emergency braking deceleration (ax = -15 ft/s2), the lateral friction margin decreases by 0.02 per percent of grade decrease. Stopping sight distance decelerations are not affected by grade because the actual decelerations vary by grade per AASHTO policy. 7. The steady-state bicycle model predicts that high brak- ing situations are likely to cause negative friction margins resulting in vehicles skidding while traversing horizontal curves on downgrades. 4.8 Step 7: Predict Skidding of Individual Axles during Braking and Lane-Change Maneuvers on a Curve The objective of Step 7 was to identify whether braking, lane changes, and other non-steady steering maneuvers affect the ability of a vehicle to traverse a sharp horizontal curve without skidding, taking into consideration horizontal cur- vature, grade, and superelevation. Using the bicycle model inclusive of non-steady effects, simulations were run modi- fying the transient steering inputs for each vehicle class of interest in this study to determine cornering forces and fric- tion factors. The results of these simulations are compared to results from previous steps. Data from the speed and vehicle maneuver studies (Section 3.2) and instrumented vehicle studies (Section 3.3) were used as inputs for this analysis. 4.8.1 Analysis Approach The basis of this transient analysis is to determine whether the driver’s change in braking or steering inputs to the vehicle might introduce temporarily changes in the vehicle motion (transient behavior) that could affect the friction demand of each axle. For this analysis, a bicycle model suitable for tran- sient maneuver analysis is developed taking into consideration the horizontal curvature, grade, and superelevation. Like the model in Section 4.7, this formulation of the classical bicycle model assumes a two-wheel vehicle whose behavior is similar to a beam; but unlike in Section 4.7, the equations of motion are not solved in static force balance but in differential equa- tion form by finding numerical solutions. One of the simplest differential equations for vehicle motion inclusive of per-axle tire forces is a 2 degree-of-freedom differential equation model

66 with yaw rate and lateral velocity as the motion variables. The input variables in this model—the steering input, d, and veloc- ity of the vehicle—are assumed to be under the driver’s control. This same model is commonly employed in vehicle stability systems, such as electronic stability control, to confirm that the measured vehicle behavior agrees with model predictions. The derivation of this modified transient bicycle model is much more involved than previous models, so only the salient points are presented here. Additional details on the model for- mulation can be found in Varunjikar (2011). Several assump- tions are used in the derivation of the transient bicycle model, and most of the assumptions are similar to the steady-state bicycle model assumptions. The additional assumptions are as follows: 1. The moments acting on the vehicle about the vertical z-axis are not always balanced and hence give a differential equation for the spin motion of the vehicle. 2. The road grade and superelevation angles are constant within the curve. 3. The tire forces are linear (i.e., the angle between the tire and the roadway is small enough that a doubling of the relative angle doubles the tire forces). 4. The steer angle is small enough that coordinate transforms from the tire angle to the vehicle’s body angle can be sim- plified using small angle approximations. 5. Braking forces per axle are obtained from the steady-state results. The resulting equations of motion are obtained by a force and moment balance on the vehicle. The results are similar to the previous bicycle model except that the lateral and spin motions of the vehicle are governed by a dynamic force balance, instead of a static force balance. On a typical vehicle, the lateral and spin motions most affect the vehicle side forces; hence, the added detail of the differential equation solution is intended to yield more insight into the lateral forces acting on the vehicle in a curve on grade. From Section 4.7, Equation 37 remains the same, describing the normal force on the tires. Equation 35, for longitudinal dynamics (braking), must be modified to include the rotation of the coordinate system attached to the vehicle: ( )− − = − − − 100 (62)m a rV F F mg G x bf br Here the variables are defined as in Section 4.7, except that a new variable, the spin rate of the vehicle, r, is introduced. This is the rotational rate of the vehicle about the z-axis of the vehicle (through the vehicle’s CG). Equation 36, for the lateral dynamics of the vehicle (cornering), becomes: ( )+ = + + 100 (63)m dVdt rV F F mg ey cf cr Here again the variables are defined as before, except that the lateral sliding velocity of the vehicle is introduced, Vy. This is the sideways speed of the vehicle as it moves across the road surface, as measured at the CG of the vehicle (for trucks, it is measured at the tractor’s CG). This velocity is usually very small, but it is non-zero and generates appreciable errors if ignored for high-speed dynamic motion. Finally, the yaw dynamics equation is introduced, which does not appear in Section 4.7: i i= − (64)I dr dt a F b Fzz cf cr This equation predicts how the vehicle’s spin rate, r, will speed up or slow down depending on the unbalanced moments produced at the front and rear axles. Izz is the moment of iner- tia of the vehicle (or tractor in the case of articulated trucks) about the z-axis of the vehicle. In the models in this section and Section 4.9, it is assumed that the brakes, when applied, are done so using a constant value of net braking force, Fb. The brake-proportioning model described in Section 4.7 is again used to find the per-axle brak- ing forces, Fbf and Fbr, per Equation 35. The values of normal loads acting on the front (Wf) and rear axles (Wr) are found also using the formulations given earlier in Equations 40 and 41. A major difference between the transient and steady-state formulation of the bicycle model is that with the transient model in this step, the tire forces change with the dynamic angle of the vehicle to the road surface. For this reason, a simple explanation of tire modeling is provided that focuses on topics that may affect lateral friction margins. Tires experience very small amounts of sideways skidding, called lateral slip, as they roll under cornering conditions for normal driving. This well-known phenomenon is used to accurately predict how a tire will develop a lateral force, Fc ( Gillespie, 1992). The slip angle of the tire is measured from the tire’s orientation (x′-axis) to the tire’s direction of travel (i.e., tire’s velocity vector relative to the road directly underneath). A diagram of the tire’s slip angle, ai, is shown in Figure 48. In contrast, the tire’s steer angle, d, is the angle measured from Figure 48. Transient bicycle model.

67 the vehicle’s longitudinal orientation (vehicle’s x-axis) to the tire’s direction of heading (i.e., tire’s x′-axis). The tire’s slip angle is rarely the same as the steering angle; however, the pur- pose of the steering angle is to influence the tire’s skid angle to obtain the desired vehicle trajectory. For small tire slip angles (5° or less) that are typical of nor- mal driving, the cornering force for an ordinary tire under a fixed normal load increases linearly with the tire slip angle (i.e., if the slip angle of the tire is doubled, the lateral force from the tire doubles). This proportionality constant for the cornering force to a is called the “cornering stiffness,” Ca. This linear tire model is used in this section and Section 4.9 to find cornering forces, Fcf and Fcr, in Equations 63 and 64. The cornering forces on the front and rear axles are: i i = α = α α α (65) F C F C cf f f cr r r where Caf and Car are the cornering stiffness values for the front and rear axles, respectively. Like a friction force, the cornering stiffness strongly depends on normal load and is assumed to change proportionally to normal load, as a first approximation (Gillespie, 1992). The cornering coefficient, CC, is defined as the ratio of the cornering stiffness to the normal load (Fz), such that one can calculate the cornering stiffness given a normal load on the tire: i= +α (66)C CC F CCz offset Figure 49 shows the cornering stiffness at four different loads for a passenger vehicle tire, and a linear curve-fit using the least-square method. These cornering stiffness values were obtained from tire curves, which are taken from data sets for passenger vehicle and truck tires (see CarSim and TruckSim documentation for example data sets). The slope of the linear curve-fit is the cornering coefficient of a tire. For most tires this value is in the range of 10 to 25 [1/rad]. Table 24 shows the cornering coefficients assumed for vehicles in this analysis. To use the cornering stiffnesses for tire force calcula- tions, the tire’s slip angle (i.e., the angle of the tire with respect to the road) must be known. The slip angle of the tire can be found using geometry as described by Bundorf (1968) and Pacejka (2006). For the front axle, the tire’s slip angle is: i( )α = + − δ−tan (67)1 V a rVf y And for the rear tire: i( )α = −−tan (68)1 V b rVr y Using small angle approximations, Equations 67 and 68 can be rewritten as: i i α = + − δ α = − (69) V a r V V b r V f y r y Figure 49. Tire cornering stiffness for normal loads on passenger vehicle tire. 200 400 600 800 1000 1200 1400 1600 1800 2000 2200 1 1.5 2 2.5 3 3.5 4 4.5 5 x 10 4 Normal Load (lbf) Co rn er in g St iff ne ss (lb f/ra d) C f = 21. 38 * Load + 4785 Vehicle CC (1/rad) CC offset (lbf/rad) E-class sedan 21.38 4,785 E-class SUV 10.55 6,848 Full-size SUV 10.55 6,848 Single-unit truck 7.08 7,336 Tractor semi-trailer truck 7.08 7,336 Tractor semi-trailer/full-trailer truck 7.08 7,336 Table 24. Cornering coefficients for vehicles used in this research.

68 Substituting these expressions into the equations of motion given earlier, Equations 63 to 69, the lateral dynamics (cornering) equation becomes: i i i i ( ) ( ) ( ) ( ) ( ) + = + + − δ + + − + . . . 100 (70) m dV dt rV CC N CC V a r V CC N CC V b r V mg e y f offset y r offset y And the yaw dynamics equation, Equation 64, can be found through similar substitutions: i i i i i i ( ) ( ) ( ) ( ) = + + − δ − + − (71) I dr dt a CC N CC V a r V b CC N CC V b r V zz f offset y r offset y These two coupled differential equations are solved for each vehicle trajectory using a built-in numerical ordinary dif- ferential equation (ODE) solver, the fixed-step Runge-Kutta method, using a time step of 0.01 s. The Simulink® software within MATLAB was used to solve these differential equations. 4.8.1.1 Definition of Maneuvers This analysis considers a vehicle entering a simple horizon- tal curve with constant radius, varying the horizontal curva- ture, grade, and superelevation. This allows for the analysis of both the transient dynamics of the vehicle due to a sudden change in steering input, and also the steady-state tire forces for a given scenario as described in Step 6. The vehicle is assumed to be traveling at a constant speed up until the point that brakes are applied. The curve approach and pre-braking speed is assumed to be the same as the design speed for the curve. Depending on the situation, the vehicle applies brakes, steers into the curve, steers into the curve and performs a lane change, or combinations thereof. To calculate the steady steering for the vehicle entering the curve, the level-road steering equation was first used. This equation, shown below, predicts the steer angle, d, necessary for a vehicle of length L between the front and rear axles to tra- verse a curve of radius R′ in steady state. If the front and rear tire slip angles, af and ar, are known, the equation is given by: δ = ′ + α − α (72) L R f r The rotation radius, R′, in Equation 72 represents the effec- tive radius of the vehicle maneuver path. For a superelevated curve, the rotation radius is greater than the curve’s radius, R, as seen in Figure 50. The rotation radius can be found using geometry, which results in the following equation: ′ = θ = θ cos sec (73)R R R The final steering angle in a curve on a superelevated roadway, d, can be found by combining the steady turning equation, Equation 72, with the equation for the cornering stiffness, Equation 66, to obtain: iδ = ′ − −     − α α 100 (74) 2L R b C a C m L V R g e f r In the simulation of the transient bicycle model, the steer- ing inputs are done “open-loop,” where the steering values are fed into the simulation as inputs with no corrections if the vehicle does not follow the correct trajectory. On the curves and on tangents, the steering inputs required are readily calculated using Equation 74—for tangents, the radius is set to infinity. However, it is especially difficult to predict the steering inputs required for the transitions on the tangent approach to a curve. This is because the super elevation is changing from a normal crown to full superelevation. To simplify the analysis and to produce “worst-case” results, it is assumed in the simulations that the tangent approach is fully developed prior to entry into the curve. This gives the worst-case friction margins for the entire trajectory, because the steering change from tangent to curve keeping is the most abrupt with fully developed super elevation on the tangent. The steering input is assumed to transition quickly from the tangent steering value to the curve value, to provide worst-case responses. The worst-case situation would be to model the steering change as a step steering input; however, this is equivalent to instantaneously turning the front tires on the road and such an unrealistic steering change will automatically induce front tire skidding. To represent a fast but reasonable transition from tangent to curve steering, the transition from one to the other was assumed to take at least 2 s. Comparisons are presented later between the multi- body simulations in Sections 4.10 and 4.11 (which include Figure 50. Rotation radius and curve radius for superelevated curve.

69 a feedback-driver model), and it is seen that the transient bicycle model is noticeably more conservative in predict- ing lateral friction margins on entry to the curve because of these assumptions of fully developed superelevation on the tangent approach. Similar to the steering inputs, the braking inputs are defined as step inputs. The magnitudes of the deceleration values are calculated prior to the simulation to ensure that rates corre- spond with the braking situations appropriate for this analysis, per braking Equations 50 to 55. Representative inputs into the simulation are plotted versus time in Figure 51 for a vehicle that is first steering into a curve at t = 3 s and then braking at t = 6.75 s. The top plot shows the application of the brake input, the middle plot shows the steady decrease in vehicle speed dur- ing the maneuver, and the bottom plot shows the change in steering input applied over a 1 s duration from driving the tan- gent to driving the curve. Note vehicle speeds are limited to a minimum of 5 mph. Since the deceleration, ax, is assumed to be constant, the braking inputs are found using the brake-proportioning model described in Step 6 (Section 4.7). This analysis assumes that the weight shift due to deceleration is instantaneous since suspension dynamics are being ignored. A comparison of lat- eral friction margins for a suspension-less vehicle and for a simulation inclusive of suspension was performed, and both models gave nearly identical results for the predicted maxi- mum lateral forces and minimum lateral friction margins. Thus, for purposes of simplicity and clarity, the suspension- free model is used in this analysis. With the simulation equations and parameters now defined, a trajectory can be simulated. At each time step, the numeri- cal solver calculates a solution to the differential equation, and then moves incrementally to the next time instant using the previous solution as an initial condition for the cur- rent step. This is repeated until an entire time trajectory is produced. Each trajectory is simulated for at least 10 s and more as necessary to ensure a sufficient duration to capture both the transient and the steady responses to the curves. The forces on the tires during each simulated trajectory are saved and used to calculate the friction margins throughout the maneuver, and the worst-case friction margins are saved for plotting purposes. For many of the plots of friction margin that follow, each data point in each margin curve represents one simulation. When multiple curves are presented for various situations, some trends become evident. 4.8.2 Analysis Results 4.8.2.1 Effects of Curve Keeping at Constant Speed The first set of simulations performed using the transient bicycle model were used to study the differences between the point-mass model, the steady-state bicycle model from Sec- tion 4.7, and the transient bicycle model in Section 4.8. To simplify the analysis and to choose a situation where all mod- els should nominally agree, these first sets of analyses consider vehicles traveling at constant speed on the curves. There are Figure 51. Simulation inputs for E-class sedan (V  85 mph, G  9%, e  12%) (ax  11.2 ft /s2). 0 1 2 3 4 5 6 7 8 9 10 0 5 10 Br ak e De m an d(f t/s 2 ) 0 1 2 3 4 5 6 7 8 9 10 60 80 100 Sp ee d (m ph ) 0 1 2 3 4 5 6 7 8 9 10 0 2 4 x 10-3 Time (s) St ee rin g in pu t (r ad )

70 no steering inputs other than those to maintain the vehicle within the lane, and no braking inputs other than those to prevent the vehicle from accelerating on a downgrade. When analyzing the simulation results, there were effects at high and low speeds that caused disagreement between the transient model and the other models. One of these effects only occurs at larger superelevations, while the other only occurs on the front tires. Figure 52 shows the lateral fric- tion margins for the front and rear tires versus speed for two different superelevations, where the differences between the models, particularly the transient bicycle model and the steady-state bicycle model, can be observed. To understand the high-speed model disagreements, these situations are plotted showing the normalized forces, the friction supply, and the resulting friction margins for both the front and rear tires in Figure 53. In the margin plots for high superelevations (i.e., 12%), the minimum margins are seen to occur immediately prior to entry to the curve, not on the curve itself. This behavior is not seen in the low super- elevation case. This indicates that the superelevated road on entry to the curve is requiring more friction utilization than the curve itself. However, recall that for the simulations, the superelevation is assumed to be fully developed prior to entry into the curve. This is not typical design practice. AASHTO policy indicates that the proportion of the superelevation runoff length [i.e., the length of roadway needed to accom- plish a change in outside-lane cross-slope from zero (flat) to full superelevation] on the tangent should be in the range of 0.6 to 0.9 (60% to 90%) for all speeds and rotated widths. Therefore, these results should be interpreted carefully. To assist with design of superelevation, it is possible to derive conditions for which the superelevation on the tangent approach will give worse margins than the curve, at least for steady driving. In Section 4.7, the steady-state front and rear friction factors are given by Equation 45. For the situation with no braking, ax = 0, this set of equations simplifies to: ( ) ( ) = − − = − + 100 100 100 100 (75) 2 2 f b m W V R e b h G f a m W V R e a h G yf yr On the approach tangent, the radius is infinite, and so the radius term can be simplified. Additionally, the supereleva- tion may only be developed by some fraction, ptangent, which is the proportion of the design or maximum superelevation that is attained at the point of curvature for a simple curve. = − − = − + 100 100 100 100 (76) , , f p be b h G f p ae a h G yf tangent tangent yr tangent tangent For the front friction margin on the approach to be less than within the curve, the friction factor on the tangent must be less than the friction factor in the curve: ( )− − < − − 100 100 100 100 (77) 2 p be b h G b m W V R e b h Gtangent 30 40 50 60 70 80 -0.1 0 0.1 0.2 0.3 0.4 Fr on t F y M ar gin Transient Model Steady State Model Point-Mass Model 30 40 50 60 70 80 -0.1 0 0.1 0.2 0.3 0.4 Speed (mph) Re ar Fy M ar gin E-Class Sedan, e = 0% 30 40 50 60 70 80 -0.1 0 0.1 0.2 0.3 0.4 Fr o nt F y M ar gin Transient Model Steady State Model Point-Mass Model 30 40 50 60 70 80 -0.1 0 0.1 0.2 0.3 0.4 Speed (mph) Re ar Fy M ar gin E-Class Sedan, e = 12% Figure 52. Front (top plots) and rear (bottom plots) lateral friction margins from point-mass, steady-state bicycle, and transient bicycle models for E-class sedan (G  9%, e  0% and 12%) (ax  0 ft/s2).

71 < − 100 100 (78) 2 p e V gR e tangent < +100 1 1 (79) 2e p V gRtangent The same result is obtained for the rear friction margin. This result indicates that, if the superelevation at the PC entry is larger than this value, then the lateral friction margin on entry to the curve is likely to be less than the lateral friction margin on the curve. Further, Equation 79 does not depend on the road grade, the road friction levels, or the vehicle type. The example shown in Figure 53 was simulated at 85 mph for e = 12%, with a design radius of 2,542 ft. For the simulation, the road is fully superelevated at curve entry, and so ptangent = 1. For these values, Equation 79 predicts that the largest super- elevation that should be used is 9.5%. Above this value, the superelevation on the approach tangent reduces friction values prior to the beginning of the curve (i.e., PC) more than it helps within the curve. Similarly, at 55 mph for e = 12%, with an AASHTO design radius of 807 ft, and ptangent = 1, the maximum superelevation is 12.5%. As shown in Figure 52, the lateral fric- tion margin for the transient model begins to diverge from the steady-state model at 55 mph, as predicted. Thus, the disagree- ment between the transient model and the steady-state model at high speeds and superelevations is due to the transient model identifying lateral friction margin reductions on the tangent approach, whereas the steady-state model ignores this. To understand the low-speed model disagreements, again plots are made of these specific situations. The top plots in Figure 54 show the friction forces and normalized margins for a 25 mph curve with 0% superelevation. The steering Figure 53. Friction supply and normalized forces (left plots) and resulting lateral friction margins (right plots) from transient bicycle model for E-class sedan (V  85 mph, G  9%, e  0% and 12%) (ax  0 ft/s2). 0 1 2 3 4 5 6 7 8 9 10 -0.2 0 0.2 0.4 0.6 0.8 Time (s) Fr ic tio n an d No rm al ize d Fo rc es E-Class Sedan, e = 0% Front Supply Rear Supply Front Transient Demand Rear Transient Demand 0 1 2 3 4 5 6 7 8 9 10 -0.2 0 0.2 0.4 0.6 0.8 Time (s) La te ra l F ric tio n M ar gi ns E-Class Sedan, e = 0% Front Transient Rear Transient Front Transient Minimum Rear Transient Minimum 0 1 2 3 4 5 6 7 8 9 10 -0.2 0 0.2 0.4 0.6 0.8 Time (s) Fr ic tio n an d No rm al ize d Fo rc es E-Class Sedan, e = 12% Front Supply Rear Supply Front Transient Demand Rear Transient Demand 0 1 2 3 4 5 6 7 8 9 10 -0.2 0 0.2 0.4 0.6 0.8 Time (s) La te ra l F ric tio n M ar gi ns E-Class Sedan, e = 12% Front Transient Rear Transient Front Transient Minimum Rear Transient Minimum Front Steady Rear Steady Front Steady Rear Steady

72 input in this case is a 1 s transition from the tangent steering value to the curve-keeping steering value, but the resulting tire forces show a small peak at the end of the steering transi- tion. This peak is due to the additional forces necessary to accelerate the vehicle in rotation versus the steady forces nec- essary for maintaining the vehicle’s spin and tire forces within the curve. Because the acceleration depends on how quickly the vehicle transitions from driving the tangent (i.e., straight- line driving) to driving the curve, this peak should decrease if the transition is spread out over a longer interval. The bottom plots in Figure 54 show the friction forces and normalized margins for a 25 mph curve following a 2 s transition. These plots show a reduced overshoot of tire forces versus the 1 s transition case, and thus hereafter the 2 s transition is used. Once the curve entry friction margins were understood, simulations were repeated to study the differences between the point-mass model, the steady-state model from Sec- tion 4.7, and the transient bicycle model within the curve for simple curve-keeping steering inputs (i.e., the intended trajectory of the vehicle is within the same lane on the approach tangent and through the curve). The results are shown in Figure 55. The first observation is that all of the models predict more lateral friction margin at high speeds than low speeds, except for the case with high superelevation (i.e., e = 16) where, due to the entry approach issues men- tioned previously, the friction margins are similar at very low and very high speeds. The next observation is that at all speed ranges for lower superelevations (i.e., 4% and 8%), all three models largely agree. The point-mass model and steady-state bicycle model also largely agree at all speed ranges for higher supereleva- tions (i.e., 12% and 16%) as well. However, at higher speeds Figure 54. Friction supply and normalized forces ( left plots) and resulting lateral friction margins (right plots) from transient bicycle model with a 1 s transition (top plot) and a 2 s transition (bottom plots) for E-class sedan (V  25 mph, G  9%, e  0%) (ax  0 ft/s2). 0 1 2 3 4 5 6 7 8 9 10 -0.2 0 0.2 0.4 0.6 0.8 Time (s) Fr ic tio n an d No rm al ize d Fo rc es Front Supply Rear Supply Front Transient Demand Rear Transient Demand 0 1 2 3 4 5 6 7 8 9 10 -0.2 0 0.2 0.4 0.6 0.8 Time (s) La te ra l F ric tio n M ar gi ns Front Transient Front Steady Rear Transient Rear Steady Front Transient Minimum Rear Transient Minimum 0 1 2 3 4 5 6 7 8 9 10 -0.2 0 0.2 0.4 0.6 0.8 Time (s) Fr ic tio n an d No rm al ize d Fo rc es Front Supply Rear Supply Front Transient Demand Rear Transient Demand 0 1 2 3 4 5 6 7 8 9 10 -0.2 0 0.2 0.4 0.6 0.8 Time (s) La te ra l F ric tio n M ar gi ns Front Transient Front Steady Rear Transient Rear Steady Front Transient Minimum Rear Transient Minimum

73 and higher superelevations, the transient bicycle model esti- mates lower friction margins than the other models. Comparing the steady-state bicycle model and the point- mass model, the point-mass model predicts a rear-axle friction margin that is 0.006 higher than predicted by the steady-state model. The front lateral friction margin is similarly under- predicted by the point-mass model. This very minor differ- ence is observed over all speeds, and across all grades, and is due to the weight shift caused by grade. While there remains some disagreement between the mod- els for the front-axle friction margins, the rear tire predictions for the steady-state and transient bicycle models are in agree- ment for the lower speeds for the rear tire. This is important because the rear tire margins appear to be the limiting case, e.g., the rear tires appear to be the first to lose friction at the higher speeds expected of high-speed downgrades. In the plots that follow hereafter, only the minimum lateral friction margins between the front and rear axles are presented. This minimum is calculated at each speed by taking the minimum of the front- and rear-axle friction values. A dividing line is shown in the plot where the minimum margins occur at the front tires versus rear tires; in general this line is at 30 mph. Figure 56 shows the minimum lateral friction margins for cornering plotted versus speed for grades from 0% to -9%, for four superelevations (4% to 16% in 4% increments), for an E-class sedan. There is a minor but consistent influence of grade seen across superelevations: increasing grade decreases the friction margins available. Specifically, the approximate 10% of grade change in each plot (from 0% to -9%) spans a margin of 0.01, so each percentage increase of grade (i.e., a steeper downgrade) reduces the friction margins by about 0.001 at speeds higher than 40 mph. This consistent effect is due to the rear tire saturation and is relatively minor com- pared to the lateral friction margin variations due to speed. Below 40 mph, the change in behavior was analyzed by examining the simulation trajectories one-by-one. It was Figure 55. lateral friction margins from point-mass, steady-state bicycle, and transient bicycle models for E-class sedan (G  9%, e  4% to 16%) (ax  0 ft/s2). 30 40 50 60 70 80 -0.1 0 0.1 0.2 0.3 0.4 Fr on t F y M ar gi n Transient Model Steady-State Model Point-Mass Model 30 40 50 60 70 80 -0.1 0 0.1 0.2 0.3 0.4 Speed (mph) R ea r F y M ar gi n E-Class Sedan, e = 4% 30 40 50 60 70 80 -0.1 0 0.1 0.2 0.3 0.4 Fr on t F y M ar gi n Transient Model Steady-State Model Point-Mass Model 30 40 50 60 70 80 -0.1 0 0.1 0.2 0.3 0.4 Speed (mph) R ea r F y M ar gi n E-Class Sedan, e = 8% 30 40 50 60 70 80 -0.1 0 0.1 0.2 0.3 0.4 Fr on t F y M ar gi n Transient Model Steady-State Model Point-Mass Model 30 40 50 60 70 80 -0.1 0 0.1 0.2 0.3 0.4 Speed (mph) R ea r F y M ar gi n E-Class Sedan, e = 12% 30 40 50 60 70 80 -0.1 0 0.1 0.2 0.3 0.4 Fr on t F y M ar gi n Transient Model Steady-State Model Point-Mass Model 30 40 50 60 70 80 -0.1 0 0.1 0.2 0.3 0.4 Speed (mph) R ea r F y M ar gi n E-Class Sedan, e = 16%

74 found that the minimum friction margins occur at the front axle for all these cases and is caused by the front tires requiring additional friction during the transition from the tangent to the curve steering levels. This transition becomes increasingly abrupt with increasing superelevation. For a very high super- elevation (e.g., 16%), the vehicle has to steer significantly to the outside of the curve immediately before the curve. In the curve, the steering effort must reverse to produce force to the inside of the curve. To confirm that this curve entry effect occurs consistently across different vehicles, Figure 57 shows the minimum lateral friction margins for a fixed grade of -9%, for super elevations ranging from 0% to 16%, for four vehicles. At lower speeds, all vehicles have lower lateral friction margins across the range of superelevations. At higher speeds, the higher super- elevation curves have lower lateral friction margins because the superelevation on the tangent approach is actually requir- ing more friction utilization than within the curve. To summarize the above constant-speed plots, they illus- trate that grade and superelevation have very little effect on the friction margins for these maneuvers. The biggest insight offered is that the point-mass model slightly over-predicts available margin on the rear tires, and slightly under-predicts margin on the front tires. Without braking, however, the dif- ference between the models is minimal. Note that vehicle designs vary widely, and the effects of weight distribution on individual axle friction margins could be more significant for some vehicles. This can be seen in Figure 57 where the differ- ent vehicles yield slightly different margin predictions. These vehicle-to-vehicle differences are minor among passenger vehicles. However, the figure shows that the lateral friction margin for the truck is approximately 0.1 lower than the pas- senger vehicles across all speeds. This is largely due to the lower friction available to truck tires versus passenger tires. The results in Figure 57 suggest that, at low speeds, the steering adjustment from tangent to curve keeping can cause 30 40 50 60 70 80 -0.1 0 0.1 0.2 0.3 0.4 Speed (mph) M in im um F y M ar gi n E-Class Sedan, e = 4% Rear Axle Limits Margin Grade = 0% Intermediate grades Grade = -9% 30 40 50 60 70 80 -0.1 0 0.1 0.2 0.3 0.4 Speed (mph) M in im um F y M ar gi n E-Class Sedan, e = 8% Tangent Approach Limits Margin Rear Axle Limits Margin Grade = 0% Intermediate grades Grade = -9% 30 40 50 60 70 80 -0.1 0 0.1 0.2 0.3 0.4 Speed (mph) M in im um F y M ar gi n E-Class Sedan, e = 12% Tangent Approach Limits Margin Rear Axle Limits Margin Grade = 0% Intermediate grades Grade = -9% 30 40 50 60 70 80 -0.1 0 0.1 0.2 0.3 0.4 Speed (mph) M in im um F y M ar gi n E-Class Sedan, e = 16% Tangent Approach Limits Margin Rear Axle Limits Margin Grade = 0% Intermediate grades Grade = -9% Figure 56. Lateral friction margins from transient bicycle model for E-class sedan (G  0% to 9%, e  4% to 16%) (ax  0 ft/s2).

75 reduction in lateral friction margins if transitioned too quickly. 4.8.2.2 Effect of Curve-Entry Deceleration Another set of simulations were conducted to represent a mild deceleration within a curve. Specifically, a constant deceleration value of -3 ft/s2 was initiated 3.75 s after curve entry. (This choice of timing is discussed in later sections.) This deceleration value was not adjusted for grade, so to maintain the same deceleration, the net braking friction demand increases slightly as grade becomes steeper. Figure 58 compares the results of the modified point- mass model, the steady-state bicycle model, and the tran- sient bicycle model for this curve-entry deceleration case. As before, all models predict increasing lateral friction margin with increasing design speed. And again, the transient model agrees closely with the steady-state model for speeds above 35 mph and as long as the superelevation on approach is not higher than the thresholds given by Equation 79. Figure 59 shows the same situation as Figure 58 to illustrate the effects of grade. Only the transient model is presented for grades from 0% to -9%. Again, there is a distinct transition in margins at around 35 mph, representing the transition from front-axle skidding-dominated behavior at low speeds to rear-axle skidding at higher speeds. The minimum friction margins of the constant-speed case (ax = 0 ft/s2) in Figure 56 and the curve-entry deceleration case (ax = -3 ft/s2) in Figure 59 have very similar minimum margins at lower speeds (around 0.34 to 0.35), but at higher speeds, the (ax = -3 ft/s2) braking case has larger changes in the margin with increasing grade. Specifically, each of the “bands” of 10 grades in each deceleration plot of Figure 59 spans a friction margin of approximately 0.02. This mean that each 1% increase in the downgrade slope results in a 0.002 decrease in friction margin on that downgrade during Figure 57. Lateral friction margins from transient bicycle model for E-class sedan, E-class SUV, full-size SUV, and single-unit truck (G  9%, e  4% to 16%) (ax  0 ft/s2). 30 40 50 60 70 80 -0.1 0 0.1 0.2 0.3 0.4 Speed (mph) M in im um F y M ar gi n Vehicle: E-Class Sedan, Grade = -9% e = 0% Intermediate superelevations (4% increments) e = 16% 30 40 50 60 70 80 -0.1 0 0.1 0.2 0.3 0.4 Speed (mph) M in im um F y M ar gi n Vehicle: E-Class SUV, Grade = -9% e = 0% Intermediate superelevations (4% increments) e = 16% 30 40 50 60 70 80 -0.1 0 0.1 0.2 0.3 0.4 Speed (mph) M in im um F y M ar gi n Vehicle: Full-Sized SUV, Grade = -9% e = 0% Intermediate superelevations (4% increments) e = 16% 30 40 50 60 70 80 -0.1 0 0.1 0.2 0.3 0.4 Speed (mph) M in im um F y M ar gi n Vehicle: Single Unit Truck, Grade = -9% e = 0% Intermediate superelevations (4% increments) e = 16%

76 the same levels of deceleration. As before, this level is par- ticularly small and dwarfed by the change in friction mar- gins versus speed. However, this value is 1.5 to 2 times that of the case when there is no braking present (Figure 56). Thus, the results suggest that braking inputs to the vehicle magnify grade-related reductions in friction margin. In Figure 60, the minimum lateral friction margins are plotted for different vehicles across the entire range of super- elevations. Finally, and most importantly, the figure shows that there is still some amount of friction in reserve with curve-entry deceleration (ax = -3 ft/s2) while the vehicle is following the curve. One can see that increasing supereleva- tion has a very slight effect on the lateral friction margin, and more specifically the 16% superelevation improves the fric- tion margin by 0.01 versus the 0% case. Thus, this effect is very minor compared to the changes in margin with respect to speed, or with respect to differences between vehicles. For passenger vehicles above 55 mph for the 16% super- elevation case, or above 75 mph for the 12% superelevation case, Figure 60 shows that the additional superelevation does not benefit the passenger vehicles (due to the thresholds given by Equation 79). But for trucks, the curve-entry deceleration results in Figure 60 are quite different than the constant- speed case plotted in Figure 57. Even this small difference in deceleration drops the margin by 0.1, enough that the margin is lowest in the curve rather than on the approach, a result that of course would be different if braking were applied on the entry to the curve. Comparing the truck to the passenger vehicles, the truck has a margin 0.2 lower than the passenger vehicles. Previously, for the constant-speed case, the truck’s margin was 0.1 lower than the passenger vehicles. Thus, brak- ing inputs tend to reduce lateral friction margins for trucks much more severely than for passenger vehicles. 4.8.2.3 Analysis of Friction Margins When Curve Radius Is 80% of AASHTO Minimum Design Radius One goal of this study was to understand how modifications to the existing AASHTO roadway design policy might affect Figure 58. Lateral friction margins from point-mass, steady-state bicycle, and transient bicycle models for E-class sedan (G  9%, e  4% to 16%) (ax  3 ft/s2). 30 40 50 60 70 80 -0.1 0 0.1 0.2 0.3 0.4 Fr on t F y M ar gi n Transient Model Steady-State Model Point-Mass Model 30 40 50 60 70 80 -0.1 0 0.1 0.2 0.3 0.4 Speed (mph) R ea r F y M ar gi n E-Class Sedan, e = 4% 30 40 50 60 70 80 -0.1 0 0.1 0.2 0.3 0.4 Fr on t F y M ar gi n Transient Model Steady-State Model Point-Mass Model 30 40 50 60 70 80 -0.1 0 0.1 0.2 0.3 0.4 Speed (mph) R ea r F y M ar gi n E-Class Sedan, e = 8% 30 40 50 60 70 80 -0.1 0 0.1 0.2 0.3 0.4 Fr on t F y M ar gi n Transient Model Steady-State Model Point-Mass Model 30 40 50 60 70 80 -0.1 0 0.1 0.2 0.3 0.4 Speed (mph) R ea r F y M ar gi n E-Class Sedan, e = 12% 30 40 50 60 70 80 -0.1 0 0.1 0.2 0.3 0.4 Fr on t F y M ar gi n Transient Model Steady-State Model Point-Mass Model 30 40 50 60 70 80 -0.1 0 0.1 0.2 0.3 0.4 Speed (mph) R ea r F y M ar gi n E-Class Sedan, e = 16%

77 vehicle behavior on curves with steep grades. To investigate how tighter curve geometries might affect friction margin, the fric- tion margins were evaluated for curves with radii that were 80% of the AASHTO minimum-radius curves. To keep the analysis simple, no braking inputs were added for this analysis. For purposes of comparing the effects of the reduced design radius, Figure 61 shows lateral friction margins for a -9% grade, for superelevations ranging from 0% to 16%, for an E-class sedan, E-class SUV, full-size SUV, and single-unit truck. The figure shows the nominal radius case side-by-side with the low-radius case. The reduced design radius situ- ations reduce the lateral friction margins at low speeds by about 0.1 to 0.14, and by about 0.02 at high speeds. Thus, the effect of radius reduction appears to be more significant at lower design speeds than for higher design speeds. In the low-radius case, the addition of superelevation appears to reduce the margin across all speeds. Further, for the lower- radius design, the margins change much more with changes in superelevation, i.e., the sensitivity of the design to changes in superelevation is much higher. 4.8.2.4 Effect of Lane-Change Maneuver at Constant Speed The effects of a lane-change maneuver within the curve were also studied. This subsection is organized to first intro- duce how the lane changes were modeled. Next, the worst- case timing for lane-change inputs is investigated. These worst-case lane changes are then simulated for a variety of geometric and vehicle situations to understand the influence of lane changes on lateral friction margins at constant speed. To begin, it was assumed that for a lane-change maneuver, the vehicle travels from a low-speed lane to a high-speed lane at a constant speed. This assumption was made as lane changes are often made to avoid slower-moving traffic in the right lane. For the analysis, it was assumed that the curve was to the left, and therefore the lane change was toward the inside of the curve. This was chosen to require higher tire forces, since this type of lane change effectively tightens the turning radius of the vehicle. For most driving, the steering input used for the lane- change simulations can be approximated by one period of Figure 59. Lateral friction margins from transient bicycle model for E-class sedan (G  0% to 9%, e  4% to 16%) (ax  3 ft/s2). 30 40 50 60 70 80 -0.1 0 0.1 0.2 0.3 0.4 Speed (mph) M in im um F y M ar gi n E-Class Sedan, e = 4% Rear Axle Limits Margin Grade = 0% Intermediate grades Grade = -9% 30 40 50 60 70 80 -0.1 0 0.1 0.2 0.3 0.4 Speed (mph) M in im um F y M ar gi n E-Class Sedan, e = 8% Tangent Approach Limits Margin Rear Axle Limits Margin Grade = 0% Intermediate grades Grade = -9% 30 40 50 60 70 80 -0.1 0 0.1 0.2 0.3 0.4 Speed (mph) M in im um F y M ar gi n E-Class Sedan, e = 12% Tangent Approach Limits Margin Rear Axle Limits Margin Grade = 0% Intermediate grades Grade = -9% 30 40 50 60 70 80 -0.1 0 0.1 0.2 0.3 0.4 Speed (mph) M in im um F y M ar gi n E-Class Sedan, e = 16% Tangent Approach Limits Margin Rear Axle Limits Margin Grade = 0% Intermediate grades Grade = -9%

78 a sine wave, and so this steering waveform was used as an idealization of the driver’s input. To determine the appro- priate duration of the lane-change portion of the steering input, field data were used for guidance. Based upon lane- change duration data collected in the field (see Section 3.2.3), most lane changes are completed within approximately 3 s for passenger vehicles and 4 s for trucks. Thus, the period of the sine wave steering input was limited to 3 s for passenger vehicles and 4 s for trucks. This sine wave steering input is applied in addition to the nominal steering required for trav- eling on a curve. A challenge in simulating lane-change maneuvers is that the steering amplitude of the single sine wave required for a lane change depends on the vehicle and on speed. At low speeds, larger steering inputs are required to obtain the same lateral motion of small-amplitude, high-speed steering inputs. To calculate how the lane-change steering amplitude changes with speed, the vehicles were simulated on a tan- gent section of roadway and given increasingly larger steer- ing amplitudes. The assumption is that the steering inputs are additive: e.g., one can add a curve-keeping steering input to a straight-road lane-change steering input to obtain the steering input for changing lanes on a curve. This addition of steering inputs assumes that the superposition principle is valid for the vehicle system, which is generally true as long as the steering inputs are small and the vehicle behavior is linear (e.g., it is not near skidding). In the cases where the vehicle is actually near skidding, this will be discerned in the friction analysis and in the comparison of steering inputs between this transient model and in the multibody simulations in later sections. The simulated road was made infinitely wide and uniform in friction to avoid roadway departure effects. Also the steer- ing inputs specified a lane-change maneuver of 12 ft. 30 40 50 60 70 80 -0.1 0 0.1 0.2 0.3 0.4 Speed (mph) M in im um F y M ar gi n Vehicle: E-Class Sedan, Grade = -9% e = 0% Intermediate superelevations (4% increments) e = 16% 30 40 50 60 70 80 -0.1 0 0.1 0.2 0.3 0.4 Speed (mph) M in im um F y M ar gi n Vehicle: E-Class SUV, Grade = -9% e= 0% Intermediate superelevations (4% increments) e = 16% 30 40 50 60 70 80 -0.1 0 0.1 0.2 0.3 0.4 Speed (mph) M in im um F y M ar gi n Vehicle: Full-Sized SUV, Grade = -9% e = 0% Intermediate superelevations (4% increments) e = 16% 30 40 50 60 70 80 -0.1 0 0.1 0.2 0.3 0.4 Speed (mph) M in im um F y M ar gi n Vehicle: Single Unit Truck, Grade = -9% e = 0% Intermediate superelevations (4% increments) e = 16% Figure 60. Lateral friction margins from transient bicycle model for E-class sedan, E-class SUV, full-size SUV, and single-unit truck (G  9%, e  4% to 16%) (ax  3 ft/s2).

30 40 50 60 70 80 -0.1 0 0.1 0.2 0.3 0.4 Speed (mph) M in im um F y M ar gi n Vehicle: E-Class Sedan, Grade = -9% R = Rmin e = 0% Intermediate superelevations (4% increments) e = 16% 30 40 50 60 70 80 -0.1 0 0.1 0.2 0.3 0.4 Speed (mph) M in im um F y M ar gi n Vehicle: E-Class Sedan, Grade = -9% R = 0.8*Rmin e = 0% Intermediate superelevations (4% increments) e = 16% 30 40 50 60 70 80 -0.1 0 0.1 0.2 0.3 0.4 Speed (mph) M in im um F y M ar gi n Vehicle: E-Class SUV, Grade = -9% R = Rmin e = 0% Intermediate superelevations (4% increments) e = 16% 30 40 50 60 70 80 -0.1 0 0.1 0.2 0.3 0.4 Speed (mph) M in im um F y M ar gi n Vehicle: E-Class SUV, Grade = -9% R = 0.8*Rmin e = 0% Intermediate superelevations (4% increments) e = 16% 30 40 50 60 70 80 -0.1 0 0.1 0.2 0.3 0.4 Speed (mph) M in im um F y M ar gi n Vehicle: Full-Sized SUV, Grade = -9% R = Rmin e = 0% Intermediate superelevations (4% increments) e = 16% 30 40 50 60 70 80 -0.1 0 0.1 0.2 0.3 0.4 Speed (mph) M in im um F y M ar gi n Vehicle: Full-Sized SUV, Grade = -9% R = 0.8*Rmin e = 0% Intermediate superelevations (4% increments) e = 16% 30 40 50 60 70 80 -0.1 0 0.1 0.2 0.3 0.4 Speed (mph) M in im um F y M ar gi n Vehicle: Single Unit Truck, Grade = -9% R = Rmin e = 0% Intermediate superelevations (4% increments) e = 16% 30 40 50 60 70 80 -0.1 0 0.1 0.2 0.3 0.4 Speed (mph) M in im um F y M ar gi n Vehicle: Single Unit Truck, Grade = -9% R = 0.8*Rmin e = 0% Intermediate superelevations (4% increments) e = 16% Figure 61. Lateral friction margins for AASHTO minimum-radius curves and 80% of minimum radius for E-class sedan, E-class SUV, full-size SUV, and single-unit truck (G  9%, e  0% to 16%) (ax  0 ft/s2).

80 Sensitivity analyses of the steering input assumptions revealed several important points. Lane-change-steering amplitudes strongly depend on speed. However, the effect of grade is quite small, so small that it can be ignored in calculat- ing the lane-change-steering inputs. Additionally all vehicles reach steady curve-keeping tire forces within several seconds of curve entry. This allows the simulations to be simplified in several ways. First, they do not have to be simulated for long durations: in this study, 10 s was seen to be more than sufficient to understand the resulting vehicle motions and friction margins. Second, the lane-change maneuvers can be simulated within several seconds before and after entry of the curve, and the resulting analyses allow understanding of the entire curve behavior as long as the speed conditions are similar. For this reason, most of the analysis that follows focuses on curve entry conditions and maneuvers near curve entry, yet the insights apply throughout the curve. It was unclear at the start of the study whether the “worst- case” tire forces and friction margins would occur if the lane change happened immediately before curve entry, during curve entry, or after the curve entry when curve-keeping tire forces were fully developed. The reason for this uncertainty is because, unlike curve-keeping steering inputs, the shape of the lane-change steering input changes directions in time. A sample steering profile is summarized in Figure 62. To understand the lane-change effects in more detail, sim- ulations were conducted to find the worst time to initiate a lane-change maneuver within a curve. The vehicle simulations were set up to stagger the lane-change initiation time relative to the curve entry time. For each simulation, the minimum lateral friction margin was recorded across both front and rear axles. The results of these simulations indicated that, except for low-speed turns, the worst location to perform a lane change is well within a curve. For low speeds (i.e., speeds below 35 mph), the lowest lateral friction margins occur when the lane change occurs only a second or so after curve entry. This is because the tire forces overshoot in low-speed cases because the vehicle is much more responsive to steering changes and steering amplitudes must be much larger at these lower speeds for the same maneuver. However, these represent very aggres- sive curve entry conditions at low speed. Therefore, hereafter, the worst-case lane-change inputs are simulated well after the curve-entry point, at least 2 s or more after entering the curve. To analyze and quantify lateral friction margins further, a series of simulations were conducted studying whether lane- change maneuvers affect the agreement between the differ- ent models. Note that neither the point-mass model nor the steady-state bicycle model can predict friction margins for lane changes because these maneuvers violate the assumptions of steady behavior in both of the models. For the transient model, the lane-change event was initiated 3 s after curve entry, and thus the lateral tire forces for steady curve keeping are fully developed prior to the start of the lane-change maneuver. Figure 63 shows a comparison of the lateral friction mar- gins for the transient model, the steady-state model, and the point-mass model for the lane-change maneuver for an E-class sedan on -9% grade, and for 4%, 8%, 12%, and 16% superel- evations. In earlier sections examining steady maneuvers, the point-mass and steady-state models agreed very well with the transient model. In contrast, for the lane-change situations, the transient model predicts friction margins that are lower than the other models by 0.25. This is because the steady-state and point-mass models are unable to predict tire forces for situations where the steering inputs are changing, such as dur- ing a lane change. Additionally, this decrease in friction mar- gin does not appear to occur at one particular speed range, but rather appears to be a uniform decrease in margin across all speeds. Figure 63 shows some variation in the minimum friction margins with changing superelevation. A comparison specifically focusing on the effects of grade and speed is shown in Figure 64. The plots indicate that sev- eral effects are consistent across vehicles and superelevations. First, the margins increase by 0.062 from 40 to 85 mph, or about 0.0015 margin increase with each 1 mph increase in design speed; this is due to the AASHTO design policy, which decreases the design friction with increasing speeds. Further, as grades change from 0% to -9%, the margins reduce by approximately 0.015, or a margin reduction of 0.0015 per degree of grade. Figure 65 shows the effect of speed, superelevation, and vehicle type on friction margins. For all vehicles and all speeds, as superelevation increases, there is a very slight increase in friction margins across all speeds (i.e., a 0.02 increase in mar- gin across 16% of superelevation change, or about 0.001 in margin increase per degree of superelevation added). The Figure 62. Steering inputs for lane-change simulations.

81 vehicle-to-vehicle differences amount to approximately 0.06 in margins. The full-size SUV had the worst margins among the two-axle vehicles simulated here (articulated vehicles are studied in later sections). 4.8.2.5 Effect of Lane-Change Maneuver at Curve-Entry Deceleration In addition to analyzing the effects of lane changes at con- stant speed, the effect of minor decelerations during lane changes was also studied. As before, it was unclear when the worst time would be to apply brakes within a curve, particu- larly if a lane change was also occurring in the curve. To investigate the worst time for braking in combination with a lane change, a range of braking inputs were applied, with the braking application time measured relative to the start of the lane change. The lane-change maneuver was defined using the worst-case situation found earlier: occurring 3 s after curve entry. As the stagger time between brake application and lane-change initiation was changed, the margins for each simu- lation were noted. The brake times were varied substantially, from 2 s before the lane change to 4 s after the lane-change maneuver was completed. It was determined that the worst time to initiate a brake input was approximately 0.75 s after the lane change starts (e.g., when the vehicle is just beginning to spin toward the target lane). The simulations hereafter for combined lane-changing, braking, and curve-keeping inputs are set up so that the lane change initiates 3 s after curve entry and the brake inputs occur 0.75 s after the lane change starts. Both situations correspond to the worst-case conditions for each situation. Figure 66 compares the lateral friction margins for the point- mass model, steady-state bicycle model, and the transient bi- cycle model for an E-class sedan. The resulting margins from this non-steady situation are significantly lower than the steady- state model and the point-mass model predicted margins. To study the effect of grade, superelevation, and vehicle type in lane-change and braking situations, another series of simulations was conducted. The results are shown in Figure 63. Lateral friction margins from point-mass, steady-state bicycle, and transient bicycle models for E-class sedan (G  9%, e  4% to 16%) (ax  0 ft/s2 and lane change). 30 40 50 60 70 80 -0.1 0 0.1 0.2 0.3 0.4 Fr on t F y M ar gi n Transient Model Steady-State Model Point-Mass Model 30 40 50 60 70 80 -0.1 0 0.1 0.2 0.3 0.4 Speed (mph) R ea r F y M ar gi n E-Class Sedan, e = 4% 30 40 50 60 70 80 -0.1 0 0.1 0.2 0.3 0.4 Fr on t F y M ar gi n Transient Model Steady-State Model Point-Mass Model 30 40 50 60 70 80 -0.1 0 0.1 0.2 0.3 0.4 Speed (mph) R ea r F y M ar gi n E-Class Sedan, e = 8% 30 40 50 60 70 80 -0.1 0 0.1 0.2 0.3 0.4 Fr on t F y M ar gi n Transient Model Steady-State Model Point-Mass Model 30 40 50 60 70 80 -0.1 0 0.1 0.2 0.3 0.4 Speed (mph) R ea r F y M ar gi n E-Class Sedan, e = 12% 30 40 50 60 70 80 -0.1 0 0.1 0.2 0.3 0.4 Fr on t F y M ar gi n Transient Model Steady-State Model Point-Mass Model 30 40 50 60 70 80 -0.1 0 0.1 0.2 0.3 0.4 Speed (mph) R ea r F y M ar gi n E-Class Sedan, e = 16%

82 Figure 67 for an E-class sedan, in Figure 68 for an E-class SUV, in Figure 69 for a full-size SUV, and in Figure 70 for a single- unit truck. As observed before, for passenger vehicles, each per- centage decrease in grade appears to reduce friction margin by approximately 0.001. However, for the truck, the lateral friction margin decreases by 0.002. This is due to many factors: a higher center of gravity that causes more weight shift, different tire type, and greater tire force magnitudes required for braking. Across all vehicle types, the margins are worse at low speeds. At low speeds and on high grades, the margins are particularly low. For the single-unit truck, the combination of low speeds on high grades results in negative lateral friction margins. Con- sidering the effect of superelevation on lateral friction margins, the addition of superelevation does increase the lateral friction margins slightly, but this effect is consistently very small across all the vehicles considered in this study. When lane changes combined with braking (-3 ft/s2; see Fig- ures 67 to 70) are compared to the no-braking (0 ft/s2) lane- change case (see Figures 64 and 65), the addition of braking reduced the lateral friction margins for constant-speed lane changes by an additional 0.05 for passenger vehicles and by 0.15 for the single-unit truck. For grades of -8% and -9%, the single- unit truck has negative friction margins in this case for design speeds less than 40 mph. Thus, the effects of braking and lane changes on lateral friction margins can accumulate to ultimately give very low or negative lateral friction margin situations. With the negative friction margins observed in Figure 70, it is worthwhile to review how these margins are physically obtained and what they signify. As noted in Section 3.4, the minimum friction supplies are obtained from the statistical distribution of friction values obtained from field measurements for wet pave- ment conditions. To calculate the friction supply, the statisti- cal distribution of supply friction is calculated at two standard deviations below the mean; as reference, these 2nd percentile friction values are generally 0.1 to 0.15 below the mean values. To calculate the friction demand, the simulations are conducted on dry roads as these will generally not excite skidding (which is difficult to simulate) and also generally demand the most tire forces. Further, the simulation procedure above is intention- ally seeking out the lowest friction margins (i.e., the worst-case maneuver combinations). Thus, the negative friction margins do not imply that the roadway design will cause skidding for a particular vehicle; rather, it indicates that if the road conditions are wet, if the road condition is at the 2nd-percentile friction 30 40 50 60 70 80 -0.1 0 0.1 0.2 0.3 0.4 Speed (mph) M in im um F y M ar gi n E-Class Sedan, e = 4% Rear Axle Limits Margin Grade = 0% Intermediate grades Grade = -9% 30 40 50 60 70 80 -0.1 0 0.1 0.2 0.3 0.4 Speed (mph) M in im um F y M ar gi n E-Class Sedan, e = 8% Tangent Approach Limits Margin Rear Axle Limits Margin Grade = 0% Intermediate grades Grade = -9% 30 40 50 60 70 80 -0.1 0 0.1 0.2 0.3 0.4 Speed (mph) M in im um F y M ar gi n E-Class Sedan, e = 12% Tangent Approach Limits Margin Rear Axle Limits Margin Grade = 0% Intermediate grades Grade = -9% 30 40 50 60 70 80 -0.1 0 0.1 0.2 0.3 0.4 Speed (mph) M in im um F y M ar gi n E-Class Sedan, e = 16% Tangent Approach Limits Margin Rear Axle Limits Margin Grade = 0% Intermediate grades Grade = -9% Figure 64. Lateral friction margins from transient bicycle model for E-class sedan (G  0 to 9%, e  4% to 16%) (ax  0 ft/s2 and lane change).

83 values of all roads, and if the maneuver combination is at the worst timing and location within the curve, then skidding may occur. Thus, the results do not provide absolute pass/fail criteria for a road design; instead they serve as indicators of trends and identify combinations of designs and operational conditions that might cause concern. Because several of the simulation results above show very low margins, the more complex simulation models’ accountings of horizontal curvature, grade, and superelevation are used to simulate these maneuvers in later steps (see Section 4.11) to confirm whether this transient model is accurately predict- ing the possibility of skidding during a lane-change maneuver combined with braking inputs. 4.8.2.6 Tractor Semi-Trailers Tractor semi-trailer behavior was also considered for the same geometry and the same maneuver types. The model used in the following tractor semi-trailer simulations was developed with the same assumptions as the passenger vehicle bicycle model. One key difference, however, is that since tractor Figure 65. Lateral friction margins from transient bicycle model for E-class sedan, E-class SUV, full-size SUV, and single-unit truck (G  9%, e  4% to 16%) (ax  0 ft/s2 and lane change). 30 40 50 60 70 80 -0.1 0 0.1 0.2 0.3 0.4 Speed (mph) M in im um F y M ar gi n Vehicle: E-Class Sedan, Grade = -9% e = 0% Intermediate superelevations (4% increments) e = 16% 30 40 50 60 70 80 -0.1 0 0.1 0.2 0.3 0.4 Speed (mph) M in im um F y M ar gi n Vehicle: E-Class SUV, Grade = -9% e = 0% Intermediate superelevations (4% increments) e = 16% 30 40 50 60 70 80 -0.1 0 0.1 0.2 0.3 0.4 Speed (mph) M in im um F y M ar gi n Vehicle: Full-Sized SUV, Grade = -9% e = 0% Intermediate superelevations (4% increments) e = 16% 30 40 50 60 70 80 -0.1 0 0.1 0.2 0.3 0.4 Speed (mph) M in im um F y M ar gi n Vehicle: Single Unit Truck, Grade = -9% e = 0% Intermediate superelevations (4% increments) e = 16% Figure 66. Lateral friction margins from point-mass, steady-state bicycle, and transient bicycle models for E-class sedan (G  9%, e  4%) (ax  3 ft/s2 and lane change). 30 40 50 60 70 80 -0.1 0 0.1 0.2 0.3 0.4 Fr on t F y M ar gi n Transient Model Steady-State Model Point-Mass Model 30 40 50 60 70 80 -0.1 0 0.1 0.2 0.3 0.4 Speed (mph) R ea r F y M ar gi n E-Class Sedan, e = 4%

84 semi-trailers typically have multiple axles spaced close together longitudinally at the back of the cab and at the back of the trailer unit, these axles were each lumped into single represen- tative axles for the simulations. A diagram outlining the model structure for the low-order tractor semi-trailer dynamic equa- tions is presented in Figure 71. To explain the equations that follow, Table 25 defines the symbols used in the equations in addition to those symbols previously defined. Key measurements are labeled in Figure 71. The tractor semi-trailer has three wheel clusters: the front of the tractor, the back of the tractor, and the back of the trailer. These are referred to as the “front,” “rear,” and “trailer” axles in the plots and discussion that follow. To calculate the tire forces, the analysis of the tractor semi- trailer is more complex than a passenger vehicle because the hitch point transmits braking forces between the tractor and trailer. To solve for each axle’s normal force, the braking forces on each axle must be known. In this analysis they are assumed to be distributed according to their normal loads. Since the total brak- ing force on the combined tractor semi-trailer is given by the requested deceleration, max, the braking forces for the trailer are: i= = (80)F N mg a m N a g bt t x t x Similarly, = = (81) F N a g F N a g bf f x br r x For the trailer, the normal force on the trailer axle is found by moment balance around the hitch point: i ( )= + − −   − +100 (82)2 2 2 N m g f g h a g m a g G h h g h a g t t t h x x h t h x 30 40 50 60 70 80 -0.1 0 0.1 0.2 0.3 0.4 Speed (mph) M in im um F y M ar gi n E-Class Sedan, e = 4% Grade = 0% Intermediate grades Grade = -9% 30 40 50 60 70 80 -0.1 0 0.1 0.2 0.3 0.4 Speed (mph) M in im um F y M ar gi n E-Class Sedan, e = 8% Tangent Approach Limits Margin Grade = 0% Intermediate grades Grade = -9% 30 40 50 60 70 80 -0.1 0 0.1 0.2 0.3 0.4 Speed (mph) M in im um F y M ar gi n E-Class Sedan, e = 12% Tangent Approach Limits Margin Grade = 0% Intermediate grades Grade = -9% 30 40 50 60 70 80 -0.1 0 0.1 0.2 0.3 0.4 Speed (mph) M in im um F y M ar gi n E-Class Sedan, e = 16% Tangent Approach Limits Margin Grade = 0% Intermediate grades Grade = -9% Figure 67. Lateral friction margins from transient bicycle model for E-class sedan (G  0% to 9%, e  4% to 16%) (ax  3 ft/s2 and lane change).

85 Figure 68. Lateral friction margins from transient bicycle model for E-class SUV (G  0% to 9%, e  4% to 16%) (ax  3 ft/s2 and lane change). 30 40 50 60 70 80 -0.1 0 0.1 0.2 0.3 0.4 Speed (mph) M in im um F y M ar gi n E-Class SUV, e = 4% Grade = 0% Intermediate grades Grade = -9% 30 40 50 60 70 80 -0.1 0 0.1 0.2 0.3 0.4 Speed (mph) M in im um F y M ar gi n E-Class SUV, e = 8% Tangent Approach Limits Margin Grade = 0% Intermediate grades Grade = -9% 30 40 50 60 70 80 -0.1 0 0.1 0.2 0.3 0.4 Speed (mph) M in im um F y M ar gi n E-Class SUV, e = 12% Tangent Approach Limits Margin Grade = 0% Intermediate grades Grade = -9% 30 40 50 60 70 80 -0.1 0 0.1 0.2 0.3 0.4 Speed (mph) M in im um F y M ar gi n E-Class SUV, e = 16% Tangent Approach Limits Margin Grade = 0% Intermediate grades Grade = -9% And the sum of vertical forces on the trailer gives the verti- cal load at the hitch point to be: i ( ) = − = − + − −   − + 1 100 (83) 2 2 2 2 N m g N m g f g h a g m a g G h h g h a g h t t t h x x h t h x The braking force passed through the hitch is given by the amount of deceleration force necessary for the trailer that is not compensated by the trailer axle: i( )= − (84)2F m g N a g bh t x The moment balance about the rear axle of the tractor gives: ( )= + − + + − 100 (85)1 1 1N m g bL b dL N hL N m g hL ag Gf h h h h x And the front gives: ( ) ( )= + − − − + − 1 100 (86) 1 1 1 N m g a L b d L N h L N m g h L a g G r h h h h x The above equations often have to be used to calcu- late axle forces on flat surfaces (G = 0%, e = 0%) with no acceleration. These equations simplify in this case to the following: i ( ) ( ) = = + − −     = + + −     1 1 (87) 2 1 2 1 2 N m g f g N m g b L m g b d L f g N m g a L m g a d L f g to t t f h t t r h t t

86 30 40 50 60 70 80 -0.1 0 0.1 0.2 0.3 0.4 Speed (mph) M in im um F y M ar gi n Full-Sized SUV, e = 4% Grade = 0% Intermediate grades Grade = -9% 30 40 50 60 70 80 -0.1 0 0.1 0.2 0.3 0.4 Speed (mph) M in im um F y M ar gi n Full-Sized SUV, e = 8% Tangent Approach Limits Margin Grade = 0% Intermediate grades Grade = -9% 30 40 50 60 70 80 -0.1 0 0.1 0.2 0.3 0.4 Speed (mph) M in im um F y M ar gi n Full-Sized SUV, e = 12% Tangent Approach Limits Margin Grade = 0% Intermediate grades Grade = -9% 30 40 50 60 70 80 -0.1 0 0.1 0.2 0.3 0.4 Speed (mph) M in im um F y M ar gi n Full-Sized SUV, e = 16% Tangent Approach Limits Margin Grade = 0% Intermediate grades Grade = -9% Figure 69. Lateral friction margins from transient bicycle model for full-size SUV (G  0% to 9%, e  4% to 16%) (ax  3 ft/s2 and lane change). The full equations of motion for the tractor semi-trailer are extensive and significantly more complicated than the bicycle model presented here (Pacejka, 2006). Effect of Curve Keeping at Constant Speed for Tractor Semi-Trailers. A set of simulations for tractor semi-trailers was run to investigate situations where the vehicle transitions from a straight tangent to a constant-radius curve at a con- stant speed. As before, it was assumed that the superelevation is fully developed on the approach to the curve and thus is constant throughout the maneuver. A comparison of the point-mass, steady-state bicycle, and transient bicycle models for the tractor semi-trailer for a constant-speed curve entry did not show appreciable differ- ences between the three model predictions. This is expected, as the constant-speed, curve-keeping driving situation is the least likely to excite transient motions that might favor one axle or another. As with the passenger vehicle situations dis- cussed earlier, the lateral friction margin increases with speed from 0.29 at 25 mph up to 0.41 at 85 mph. Additionally, the transient model predicts slightly less margin at low speeds. Again, this is due to the change in steering input necessary at the beginning of the curve. Because the agreement between the models was quite good across all grades, superelevations, and speeds for this driving situation, only one example plot comparing the models is shown in Figure 72. Shown in Figure 73 are the minimum lateral friction mar- gins for grades from 0% to -9%, and for four superelevations (4% to 16% in 4% increments), for the tractor semi-trailer. As noted with passenger vehicles, there is a minor but consistent influence of grade seen across all superelevations. For speeds above 35 mph, the 10% change in grade reduces the friction margin by about 0.01 to 0.02, or about 0.001 to 0.002 margin reduction per each 1% change in grade. These numbers are consistent with those observed for two-axle vehicles noted earlier. Like passenger vehicles, tractor semi-trailers are lim- ited at low speeds (35 mph and lower) by the margins available on the front tires, and thus the steering change at the onset of the curve causes the lowest friction margins. At high speeds and high superelevations, the friction margins are limited by

87 Figure 70. Lateral friction margins from transient bicycle model for single-unit truck (G  0% to 9%, e  4% to 16%) (ax  3 ft/s2 and lane change). 30 40 50 60 70 80 -0.1 0 0.1 0.2 0.3 0.4 Speed (mph) M in im um F y M ar gi n Single Unit Truck, e = 4% Grade = 0% Intermediate grades Grade = -9% 30 40 50 60 70 80 -0.1 0 0.1 0.2 0.3 0.4 Speed (mph) M in im um F y M ar gi n Single Unit Truck, e = 8% Tangent Approach Limits Margin Grade = 0% Intermediate grades Grade = -9% 30 40 50 60 70 80 -0.1 0 0.1 0.2 0.3 0.4 Speed (mph) M in im um F y M ar gi n Single Unit Truck, e = 12% Tangent Approach Limits Margin Grade = 0% Intermediate grades Grade = -9% 30 40 50 60 70 80 -0.1 0 0.1 0.2 0.3 0.4 Speed (mph) M in im um F y M ar gi n Single Unit Truck, e = 16% Tangent Approach Limits Margin Grade = 0% Intermediate grades Grade = -9% Figure 71. Bicycle model representation of tractor semi-trailer, each set of tires is represented by a single axle.

Symbol Meaning Mass of tractor Mass of loaded trailer Mass of tractor and trailer together Distance from hitch to trailer axle Distance from hitch to trailer CG Deceleration along -axis , , Braking force (front, rear, trailer axle) , , Cornering force (front, rear, trailer axle) , , , Normal loads (front, rear, trailer axle, hitch) Vehicle weight ( ) , Tractor CG to front- and rear-axle distance Tractor CG to hitch distance Wheelbase of tractor (i.e., a + b) Height of tractor CG, trailer CG, and hitch point Table 25. Variables used to extend bicycle model to tractor semi-trailer. 30 40 50 60 70 80 0 0.2 0.4 Fr on t F y M ar gi n Transient Model Steady-State Model Point-Mass Model 30 40 50 60 70 80 −0.1 0 0.1 0.2 0.3 0.4 R ea r F y M ar gi n 30 40 50 60 70 80 0 0.2 0.4 Speed (mph) Tr ai le r F y M ar gi n e = 8% Figure 72. Lateral friction margins from point-mass, steady-state bicycle, and transient bicycle models for tractor semi- trailer (G  9%, e  8%) (ax  0 ft/s2). Figure 73. Lateral friction margins from transient bicycle model for tractor semi-trailer (G  0% to 9%, e  4% to 16%) (ax  0 ft/s2). 30 40 50 60 70 80 −0.1 0 0.1 0.2 0.3 0.4 Speed (mph) M in im um F y M ar gi n −0.1 0 0.1 0.2 0.3 0.4 M in im um F y M ar gi n −0.1 0 0.1 0.2 0.3 0.4 M in im um F y M ar gi n −0.1 0 0.1 0.2 0.3 0.4 M in im um F y M ar gi n Tractor Trailer, e = 4% Rear Axle Limits Margin Grade = 0% Intermediate grades Grade = −9% 30 40 50 60 70 80 Speed (mph) Tractor Trailer, e = 8% Tangent Approach Limits Margin Rear Axle Limits Margin 30 40 50 60 70 80 Speed (mph) Tractor Trailer, e = 12% Tangent Approach Limits Margin Rear Axle Limits Margin 30 40 50 60 70 80 Speed (mph) Tractor Trailer, e = 16% Tangent Approach Limits Margin Rear Axle Limits Margin Grade = 0% Intermediate grades Grade = −9% Grade = 0% Intermediate grades Grade = −9% Grade = 0% Intermediate grades Grade = −9%

89 the approach tangent rather than the curve itself. Again, this is consistent with observations made from passenger vehicles. To further illustrate the similarities and differences between passenger vehicles and tractor semi-trailers, Figure 74 shows the minimum lateral friction margins for cornering for a fixed grade of -9%, and for five superelevations (0% to 16% in 4% increments), for both the tractor semi-trailer and an E-class sedan. As noted before, while grade and superelevation have some visible influence on the friction margins, overall these effects are minor compared to the influence of speed and maneuvers on the road. For this configuration of a tractor semi-trailer, the superelevation influenced the friction mar- gin by a maximum of approximately 0.10 when comparing 0% superelevation margins to 16% superelevation margins at 85 mph, but the effect of superelevation decreases with speed. Thus, as observed with passenger vehicles, the primary benefit of superelevation appears to be to allow designers to decrease the radius of curvature. And like passenger vehicles, the tractor semi-trailer also exhibits lower friction margins on the curve approach for high superelevations, at higher design speeds. Thus, as noted before, superelevations above 12% cause decreasing friction margins at high speeds compared to roads with lower superelevations. Effect of Curve-Entry Deceleration for Tractor Semi- Trailers. To consider the curve-entry deceleration case, another set of simulations were conducted to represent a mild deceleration on the curve. Specifically, a constant deceleration value of -3 ft/s2 was initiated 6.75 s after curve entry. (This choice of timing is discussed in later sections.) As with pas- senger vehicles, this deceleration value was not adjusted for grade, so the net braking friction demand increases slightly as grade becomes steeper and steeper. Figure 75 compares the results of the transient model to the steady-state model and the point-mass model for the curve-entry deceleration case. As with passenger vehicles, all models predict increasing friction margin with increasing design speed. Again, the transient model agrees closely with the steady-state model and point-mass models, with a very slightly lower predicted margin on the rear and trailer axles. These predictions agree as long as the superelevation on the approach is not too high, as mentioned in previous sections. Because the agreements between the models are so close for this maneuver, only one example is shown. Figure 76 shows the effect of grade and superelevation on the tractor semi-trailer’s lateral friction margins for curve-entry Figure 74. Lateral friction margins from transient bicycle model for E-class sedan and tractor semi-trailer (G  9%, e  0% to 16%) (ax  0 ft/s2). 30 40 50 60 70 80 −0.1 0 0.1 0.2 0.3 0.4 Speed (mph) M in im um F y M ar gi n −0.1 0 0.1 0.2 0.3 0.4 M in im um F y M ar gi n Vehicle: E-Class Sedan, Grade = −9% e = 0% Intermediate superelevations (4% increments) e = 16% 30 40 50 60 70 80 Speed (mph) Vehicle:Tractor Trailer, Grade = −9% e = 0% Intermediate superelevations (4% increments) e = 16% 30 40 50 60 70 80 -0.1 0 0.1 0.2 0.3 0.4 -0.1 0 0.1 0.2 0.3 0.4 -0.1 0 0.1 0.2 0.3 0.4 Fr on t F y M ar gi n Transient Model Steady-State Model Point-Mass Model 30 40 50 60 70 80 R ea r F y M ar gi n 30 40 50 60 70 80 Speed (mph) Tr ai le r F y M ar gi n e = 8% Figure 75. Lateral friction margins from point-mass, steady-state bicycle, and transient bicycle models for tractor semi-trailer (G  9%, e  8%) (ax  3 ft/s2).

90 Figure 76. Lateral friction margins from transient bicycle model for tractor semi-trailer (G  0% to 9%, e  4% to 16%) (ax  3 ft/s2). 30 40 50 60 70 80 −0.1 0 0.1 0.2 0.3 0.4 Speed (mph) M in im um F y M ar gi n −0.1 0 0.1 0.2 0.3 0.4 M in im um F y M ar gi n −0.1 0 0.1 0.2 0.3 0.4 M in im um F y M ar gi n −0.1 0 0.1 0.2 0.3 0.4 M in im um F y M ar gi n Tractor Trailer, e = 4% Rear Axle Limits Margin Rear Axle Limits Margin Grade = 0% Intermediate grades Grade = −9% Grade = 0% Intermediate grades Grade = −9% Grade = 0% Intermediate grades Grade = −9% Grade = 0% Intermediate grades Grade = −9% 30 40 50 60 70 80 Speed (mph) Tractor Trailer, e = 8% Tangent Approach Limits Margin 30 40 50 60 70 80 Speed (mph) Tractor Trailer, e = 12% Tangent Approach Limits Margin Rear Axle Limits Margin 30 40 50 60 70 80 Speed (mph) Tractor Trailer, e = 16% Tangent Approach Limits Margin Rear Axle Limits Margin deceleration. Again, there is a distinct transition in margins at around 35 mph, representing the transition from front- axle skidding-dominated behavior at low speeds, to rear-axle skidding at higher speeds. This same behavior is observed in passenger vehicles, as seen in Figure 59; however, the friction margin for a tractor semi-trailer is approximately 0.08 to 0.1 lower. This is most likely due to differences in the tires between passenger vehicles and trucks. The constant-speed case (ax = 0 ft/s2) in Figure 73 and the curve-entry deceleration case (ax = -3 ft/s2) in Figure 76 have very similar minimum friction margins at lower speeds. In other words, the minor decelerations observed on the approach and within the curve do not appear to significantly affect the lateral friction margins. In Figure 77 the effects of superelevation are plotted for the tractor semi-trailer case and compared to the closest passen- ger vehicle case. Specifically, the minimum friction margins are plotted for the tractor semi-trailer across the entire range of superelevations and compared to the E-class sedan. Com- paring the tractor semi-trailer to the E-class sedan, the truck has a friction margin that is 0.08 to 0.1 lower across all speeds. While this could be significant, it reveals that the tractor semi- trailer is not the most sensitive to braking situations. In com- parison, the single-unit truck had a friction margin in the same situations that was 0.2 lower than the passenger vehicles. But even considering supply frictions two standard deviations below the mean measured friction across all sites, there is still some amount of friction in reserve for normal driving maneu- vers with curve-entry deceleration (ax = -3 ft/s2). Effect of Lane-Change Maneuver at Constant Speed for Tractor Semi-Trailers. Like the studies done for passenger vehicles, the effects of lane-change maneuvers on a tractor semi- trailer were also studied. As before, it was assumed that for a lane-change maneuver, the vehicle travels from a low-speed lane to a high-speed lane at a constant speed, the curve was to the left, and therefore the lane change was toward the inside of the curve. The sine wave steering input was used for the tractor semi-trailer with a 4 s period as noted for the single-unit truck. One large difference in the simulation of a tractor semi-trailer is that the lane-change maneuver was performed further in the curve than with other vehicles, at 5 s into the curve rather than the 3 s for other vehicles. Again, this represents the worst-case time to initiate a lane-change maneuver. This time difference is

91 because tractor semi-trailers were seen to take longer to reach steady-state after the curve entry than other vehicles, due to the trailer dynamics and the vehicle’s larger mass. Figure 78 shows a comparison of the lateral friction mar- gins for the transient model, the steady-state model, and the point-mass model for the lane-change maneuver for the trac- tor semi-trailer for 8% superelevation. Other superelevations were also simulated (0% to 16% in 4% intervals), and the results were nearly identical. As expected, the lane-change event reduces friction margin noticeably versus the steady- state bicycle model and the point-mass model. This is seen across all three axles. This reduction in friction values ranges from 0.1 to 0.15. However, comparing the tractor semi-trailer results to the passenger vehicle results in Figure 63, the fric- tion margin reductions for tractor trailers are actually much less than the passenger vehicle case. In other words, passenger vehicles are far more sensitive to lane-change maneuvers than are tractor semi-trailers. This is primarily due to the slower response of the tractor semi-trailers relative to passenger vehicles; the lane changes for the larger vehicles are not only initiated over a longer interval, but it takes longer to complete even when the intervals are kept the same. To determine the effects of grade, superelevation, and speed on friction margin, a series of plots are shown in Figure 79. Several effects are consistently observed across superelevations. First, the lateral friction margins increased from approximately 0.15 to 0.30 as speeds increase from 25 to 85 mph. Second, the effects of superelevation on lateral friction margins appear to be small, as the plots are nearly indistinguishable from each other between the 4%, 8%, 12%, and 16% superelevation cases. As grade changes from 0% to -9%, the margin changes by approx- imately 0.02, and thus the grade’s influence on margin is about 0.002 per each 1% change in grade. Interestingly, for a tractor semi-trailer, the lateral friction margins improve for increasing grade at speeds less than 40 mph, and above 60 mph the mar- gins are worse for increasing grade. Between 40 and 60 mph, the effect of grade has mixed impacts on the resulting margin. Overall, the effects of grade and superelevation are small com- pared to the effects of speed, vehicle type, and maneuvers. Figure 80 shows the effect of superelevation and vehicle type on friction margins, comparing a tractor semi-trailer to the worst-performing passenger vehicle, the full-size SUV. The plots show a slight increase in lateral friction margins across all speeds. However, the influence of superelevation is Figure 77. Lateral friction margins from transient bicycle model for E-class sedan and tractor semi-trailer (G  9%, e  0% to 16%) (ax  3 ft/s2). 30 40 50 60 70 80 −0.1 0 0.1 0.2 0.3 0.4 Speed (mph) M in im u m F y M ar gi n −0.1 0 0.1 0.2 0.3 0.4 M in im u m F y M ar gi n Vehicle: E-Class Sedan, Grade = −9% e = 0% Intermediate superelevations (4% increments) e = 16% 30 40 50 60 70 80 Speed (mph) Vehicle:Tractor Trailer, Grade = −9% e = 0% Intermediate superelevations (4% increments) e = 16% Figure 78. Lateral friction margins from point-mass, steady-state bicycle, and transient bicycle models for tractor semi- trailer (G  9%, e  8%) (ax  0 ft/s2 and lane change). 30 40 50 60 70 80 -0.1 0 0.1 0.2 0.3 0.4 -0.1 0 0.1 0.2 0.3 0.4 -0.1 0 0.1 0.2 0.3 0.4 Fr on t F y M ar gi n Transient Model Steady-State Model Point-Mass Model 30 40 50 60 70 80 R ea r F y M ar gi n 30 40 50 60 70 80 Speed (mph) Tr ai le r F y M ar gi n e = 8%

92 30 40 50 60 70 80 −0.1 0 0.1 0.2 0.3 0.4 Speed (mph) M in im um F y M ar gi n Tractor Trailer, e = 4% Rear Axle Limits Margin Grade = 0% Intermediate grades Grade = −9% 30 40 50 60 70 80 -0.1 0 0.1 0.2 0.3 0.4 Speed (mph) M in im um F y M ar gi n Tractor Trailer, e = 8% Tangent Approach Limits Margin Rear Axle Limits Margin Grade = 0% Intermediate grades Grade = −9% 30 40 50 60 70 80 -0.1 0 0.1 0.2 0.3 0.4 Speed (mph) M in im um F y M ar gi n Tractor Trailer, e = 12% Tangent Approach Limits Margin Rear Axle Limits Margin Grade = 0% Intermediate grades Grade = −9% 30 40 50 60 70 80 -0.1 0 0.1 0.2 0.3 0.4 Speed (mph) M in im um F y M ar gi n Tractor Trailer, e = 16% Tangent Approach Limits Margin Rear Axle Limits Margin Grade = 0% Intermediate grades Grade = −9% Figure 79. Lateral friction margins from transient bicycle model for tractor semi-trailer (G  0% to 9%, e  4% to 16%) (ax  0 ft/s2 and lane change). Figure 80. Lateral friction margins from transient bicycle model for full-size SUV and tractor semi-trailer (G  9%, e  0% to 16%) (ax  0 ft/s2 and lane change). 30 40 50 60 70 80 −0.1 0 0.1 0.2 0.3 0.4 Speed (mph) M in im um F y M ar gi n Vehicle: Full-Sized SUV, Grade = −9% e = 0% Intermediate superelevations (4% increments) e = 16% 30 40 50 60 70 80 −0.1 0 0.1 0.2 0.3 0.4 Speed (mph) M in im um F y M ar gi n Vehicle: Tractor Trailer, Grade = −9% e = 0% Intermediate superelevations (4% increments) e = 16%

93 small: a 0.02 increase in margin occurs across a 16% super- elevation change, or about 0.001 increase in margin increase per 1% of superelevation added. This, as observed earlier, is almost negligible. Thus, the effect of superelevation appears mainly to allow designers to reduce road curvature. Compared to passenger vehicles, the tractor semi-trailer has higher lateral friction margins across all speeds by a factor of about 0.08 to 0.1. Again, this is due to the more gradual lane changes that these vehicles perform versus passenger vehicles. Effect of Lane-Change Maneuver at Curve-Entry Decel- eration for Tractor Semi-Trailers. The effect of minor decelerations during lane changes was also studied for the tractor semi-trailer. As in the passenger vehicle case, it was unclear when the worst time would be to apply brakes within a curve, particularly if a lane change was also occurring on the curve. To investigate the worst time for braking, a sensitivity analysis was performed varying braking inputs for the tractor semi-trailer. Through this analysis it was determined that the worst time to initiate a brake input for a tractor semi-trailer is approximately 0.75 s after the lane change starts (e.g., when the vehicle is just beginning to spin toward the target lane). This is nearly identical to the passenger vehicle case as shown with the sedan. From the results of the worst-case braking time, the simulations hereafter have the brake inputs occur 0.75 s after the lane change starts. From the same sensitivity analysis, it was also determined that the limiting axle on tractor semi-trailers changes with speed. At low speeds, the front axle has the lowest margin. At intermediate speeds, the tractor’s rear axle has the lowest mar- gin, and at high speeds, the trailer’s axle has the lowest margin. To analyze model-specific effects for a tractor semi-trailer, Figure 81 provides a comparison of the point-mass model, steady-state bicycle model, and transient bicycle model with the additional effects of brake inputs included with the lane change. Note that only the transient model includes the lane-change effects, and the margins from this model are significantly lower than the others, as expected. The results shown here are con- sistent with the results discussed earlier for passenger vehicles. To study the effect of grade, superelevation, and speed in lane-change and braking situations, another series of simula- tions was conducted, the results of which are plotted in Fig- ure 82. Comparing these results to the two-axle vehicle cases shown in Figures 67 through 70, the plots show that a trac- tor semi-trailer has much higher margins than other vehicles for combined braking and lane-change situations. While all other vehicles had margins below 0.1 (and sometimes below zero), the tractor semi-trailer margins were all above 0.15. As observed for passenger vehicles, each percentage decrease in grade appears to reduce the lateral friction margin by approx- imately 0.001. When compared to other vehicles, tractor semi-trailers are relatively insensitive to lane-change inputs. Finally, Figure 83 presents the effect of superelevation on lateral friction margins, comparing a tractor semi-trailer to the worst-case two-axle vehicle for this situation (i.e., the single-unit truck). As noted earlier for the combined curve- entry deceleration and simple lane-change scenarios, the tractor semi-trailer has much higher lateral friction margins than other vehicles. This is due to the very long length, which results in much lower rear-to-front weight shift. Effect of Loading for Tractor Semi-Trailers. A series of simulations was conducted to vary the loading conditions to understand these effects. In the previous simulations, the trailer load was set to 22,050 lb, situated 19.7 ft to the rear of the hitch. This results in 14,053 lb on the front axle, 24,778 lb on the rear (tractor) axle, and 18,378 lb on the trailer axle. The total load is 57,209 lb. This is considered the “standard” load in TruckSim, a common commercial truck simulation software tool. However, this is not a worst-case load. To simulate a truck near the overload condition, the pay- load of the trailer was increased to 44,841 lb. At this load, if the position of the load is kept at the default (19.7 ft from the hitch), the rear axle of the tractor carries a weight in excess of 37,000 lb, which exceeds the per-axle limit of most Figure 81. Lateral friction margins from point-mass, steady-state bicycle, and transient bicycle models for tractor semi-trailer (G  9%, e  8%) (ax  3 ft/s2 and lane change). 30 40 50 60 70 80 −0.1 0 0.1 0.2 0.3 0.4 Fr on t F y M ar gi n Transient Model Steady-State Model Point-Mass Model 30 40 50 60 70 80 −0.1 0 0.1 0.2 0.3 0.4 R ea r F y M ar gi n 30 40 50 60 70 80 −0.1 0 0.1 0.2 0.3 0.4 Speed (mph) Tr ai le r F y M ar gi n e = 8%

94 30 40 50 60 70 80 -0.1 0 0.1 0.2 0.3 0.4 Speed (mph) M in im um F y M ar gi n Tractor Trailer, e = 4% Rear Axle Limits Margin Grade = 0% Intermediate grades Grade = -9% 30 40 50 60 70 80 -0.1 0 0.1 0.2 0.3 0.4 Speed (mph) M in im um F y M ar gi n Tractor Trailer, e = 8% Tangent Approach Limits Margin Rear Axle Limits Margin Grade = 0% Intermediate grades Grade = -9% 30 40 50 60 70 80 -0.1 0 0.1 0.2 0.3 0.4 Speed (mph) M in im um F y M ar gi n Tractor Trailer, e = 12% Tangent Approach Limits Margin Rear Axle Limits Margin Grade = 0% Intermediate grades Grade = -9% 30 40 50 60 70 80 -0.1 0 0.1 0.2 0.3 0.4 Speed (mph) M in im um F y M ar gi n Tractor Trailer, e = 16% Tangent Approach Limits Margin Rear Axle Limits Margin Grade = 0% Intermediate grades Grade = -9% Figure 82. Lateral friction margins from transient bicycle model for tractor semi-trailer (G  0 to 9%, e  4 to 16%) (ax  3 ft/s2 and lane change). Figure 83. Lateral friction margins from transient bicycle model for single-unit truck and tractor semi-trailer (G  9%, e  0% to 16%) (ax  3 ft/s2 and lane change). 30 40 50 60 70 80 -0.1 0 0.1 0.2 0.3 0.4 Speed (mph) M in im um F y M ar gi n Vehicle: Single Unit Truck, Grade = -9% e = 0% Intermediate superelevations (4% increments) e = 16% 30 40 50 60 70 80 -0.1 0 0.1 0.2 0.3 0.4 Speed (mph) M in im um F y M ar gi n Vehicle: Tractor Trailer, Grade = -9% e = 0% Intermediate superelevations (4% increments) e = 16%

95 departments of transportation (DOTs) of 34,000 lb. However, if the weight is shifted rearward to 24.0 ft behind the hitch, this results in the per-axle weights becoming 14,053 lb on the front axle, 33,393 lb on the rear (tractor) axle, and 32,554 lb on the trailer axle. Note that the front tire load does not change; this is because the hitch for tractor semi-trailers is generally designed to lie exactly on the rear axle of the vehicle. This pro- vides much greater steering repeatability since the front tire loads are not changing. The total load in this case is 80,000 lb (i.e., the maximum weight limit for this vehicle per most DOT specifications). Additionally, the per-axle loads do not violate DOT limits of 34,000 lb. This vehicle hereafter is considered the fully loaded tractor semi-trailer case. Figure 84 shows the difference between the normally loaded and fully loaded tractor semi-trailer situations. The additional loading has a small influence on margins, and indeed the lateral friction margins for the fully loaded case are slightly higher (by 0.03) than the normally loaded case. The limiting axle for each case, however, is notably different. As the trailer becomes more fully loaded, it appears that the trailer axle margins decrease relative to the other axles. Above 55 to 60 mph, it is the trailer’s rear axles that generally have the lowest margins for the fully loaded trailer situation. Effect of Brake Variation for Tractor Semi-Trailers. In the previous tractor semi-trailer simulations, brake forces were assumed to be distributed proportional to the static load on each axle; this results in the most repeatable behavior of the vehicle. However this also assumes that the braking forces are correctly distributed between the tractor and the trailer. In the case of design differences between the two different brake sys- tems, several cases were considered where the braking force on the tractor was 25% higher than the nominal values and 25% lower. To calculate the trailer-tire forces, these are increased or decreased to maintain the requested deceleration. Results of the brake variation simulation analysis are shown in Figure 85. The variation in braking in both cases resulted in a slight decrease in lateral friction margins around 0.01, but the Figure 84. Comparison of normally loaded (top plots) and fully loaded (bottom plots) tractor semi-trailers (G  9%, e  8%) (ax  3 ft/s2 and lane change). 30 40 50 60 70 80 -0.1 0 0.1 0.2 0.3 0.4 Speed (mph) M in im um F y M ar gi n Tractor Trailer, e = 8% Tangent Approach Limits Margin Rear Axle Limits Margin Grade = 0% Intermediate grades Grade = -9% F to R F to R F t o R R to T R to T R to T Speed (mph) Lateral Friction Margins Axle with Lowest Margins G ra de Lane Change, e = 8%, ax= -3 ft/s2 30 40 50 60 70 80 -9 -8 -7 -6 -5 -4 -3 -2 -1 0 30 40 50 60 70 80 -0.1 0 0.1 0.2 0.3 0.4 Speed (mph) M in im um F y M ar gi n Fully Loaded Tractor Trailer, e = 8% Tangent Approach Limits Margin Rear Axle Limits Margin Grade = 0% Intermediate grades Grade = -9% F t o R F to R F to R R to T R to T R to T Speed (mph) G ra de Lane Change, e = 8%, ax= -3 ft/s2 30 40 50 60 70 80 -9 -8 -7 -6 -5 -4 -3 -2 -1 0

96 30 40 50 60 70 80 -0.1 0 0.1 0.2 0.3 0.4 Speed (mph) M in im um F y M ar gi n Tractor Trailer, e = 8% Tangent Approach Limits Margin Rear Axle Limits Margin Grade = 0% Intermediate grades Grade = -9% F to R F to R F t o R R to T R to T R to T Speed (mph) Lateral Friction Margins Axle with Lowest Margins G ra de Lane Change, e = 8%, ax= -3 ft/s2 30 40 50 60 70 80 -9 -8 -7 -6 -5 -4 -3 -2 -1 0 30 40 50 60 70 80 -0.1 0 0.1 0.2 0.3 0.4 Speed (mph) M in im um F y M ar gi n Tractor Trailer (+25% to Tractor Brakes ), e = 8% Tangent Approach Limits Margin Rear Axle Limits Margin Grade = 0% Intermediate grades Grade = -9% F to R F to R F t o R Speed (mph) G ra de Lane Change, e = 8%, ax= -3 ft/s2 30 40 50 60 70 80 -9 -8 -7 -6 -5 -4 -3 -2 -1 0 30 40 50 60 70 80 -0.1 0 0.1 0.2 0.3 0.4 Speed (mph) M in im um F y M ar gi n Tractor Trailer (-25% from Tractor Brakes ), e = 8% Tangent Approach Limits Margin Rear Axle Limits Margin Grade = 0% Intermediate grades Grade = -9% F to R F t o R R t o T R to T R to T Speed (mph) G ra de Lane Change, e = 8%, ax= -3 ft/s2 30 40 50 60 70 80 -9 -8 -7 -6 -5 -4 -3 -2 -1 0 Figure 85. Comparison of ordinary braking (top plots), 25% higher braking on tractor (middle plots), and 25% less braking on tractor (bottom plots) (G  9%, e  8%) (ax  3 ft/s2 and lane change).

97 overall shape of the margin curves and trends are unchanged. The biggest effect of the brake variations is to change the axle with the minimum force in the speeds from 45 to 60 mph to be the trailer axle instead of the rear axle of the tractor. 4.8.3 Summary of Key Results from Step 7 In summary, the following findings were obtained from the analysis in Step 7: 1. In transitioning from the tangent approach to a curve, if the steering transition is faster than 2 s, the front tires will often skid, particularly at lower design speed curves. 2. The vehicle’s suspension has a negligible impact on lat- eral friction margins. 3. The worst-case lateral friction margins for situations with braking and steering changes appear to occur when lane changes or other steering inputs occur well into the curve, after the tire forces have built up after curve entry. The worst-case braking situation was found to be when brakes are activated shortly after the steering maneuver has started, by 0.75 s. This is because the lateral tire forces take some duration to build up, whereas the braking forces act nearly instantly in comparison. 4. The transient bicycle model agrees very closely with the steady-state bicycle model and the point-mass model except in the situations that only the transient model can study: for example, curve-entry steering transitions and lane-change maneuvers. 5. On curve entry at speeds lower than 35 mph, the lateral friction margins predicted by the transient bicycle model are 0.05 lower than predicted by the point-mass model or the steady-state bicycle model. 6. For roads with superelevations higher than 12%, vehi- cles often exhibited the lowest friction margins on the approach to the curve rather than within the curve. This is due to the vehicle steering up the superelevated roadway on the approach, then reversing steering inputs toward the inside of the curve. For roadways with 12% superelevation, the tangent began to reduce lateral fric- tion margins for design speeds above 65 mph. For roads with 16% superelevation, the tangent began to reduce margins for design speeds above 45 mph. In general, the tangents will reduce margins when the superelevation is larger than: < +100 1 1 2e p V gRtangent This result is independent of grade. In practical terms, this suggests that superelevations larger than 12% should be avoided. Also, the above condition should be used to check that the superelevation achieved at the PC of a simple horizontal curve is less than the threshold value computed based on the given design speed–curve radius combina- tion. Based upon further analyses, the condition above is satisfied for maximum-superelevation/minimum-radius curves for all design speeds. However, the condition above may be violated when using greater than minimum hori- zontal curve radii. 7. The effect of grade is to decrease the lateral friction margin by 0.001 for each percent grade decrease for situations with no braking (ax = 0 ft/s2) and, like the observations earlier for the steady-state model, the effect of grade increases with increased deceleration, to about 0.002 / grade percent when ax = -3 ft/s2. 8. Compared to the influence of the design speed, the effects of grade and superelevation on the resulting friction margins are relatively minor, as long as vehicles travel at the design speed. 9. The worst-case vehicle in terms of lateral friction mar- gins was the single-unit truck, which has friction margins 0.2 less than other vehicles for the same maneuvers for curve-entry deceleration cases. This is due to the much larger weight shift of this particular vehicle relative to other vehicles. The difference of this vehicle to other pas- senger vehicles (which also have two axles) is only 0.1 when maintaining the same speed from the approach through the curve; thus, the addition of braking inputs increases the relative differences between vehicles. 10. If curves are designed with tighter radii than present AASHTO design policy, this will reduce the friction mar- gin but this reduction depends on design speed. Specifi- cally, for curves that are 80% of the design radius, yet used at the same design speed, vehicles will undergo a 0.1 to 0.15 reduction in friction margin at low speeds (25 mph) and 0.02 margin reduction at high speeds (85 mph). 11. Lane-change maneuvers within a curve reduce the lateral friction margins by 0.2 to 0.25 for two-axle vehicles. This reduction appears consistent across all speeds, vehicles, grades, and superelevations. The tractor semi-trailer was less sensitive to lane changes, with the margins reduced by approximately 0.15 across all speeds. 12. In lane-change situations, the full-size SUV had the most substantial reduction in margins among all vehicles. 13. When lane changes were combined with braking, the addition of braking reduced the margins for constant- speed lane changes by an additional 0.05 for passenger vehicles and by 0.15 for the single-unit truck. The single- unit truck will have negative friction margins in this case for design speeds less than 45 mph. 14. A tractor semi-trailer is much less sensitive to braking inputs and lane changes than passenger vehicles, and thus the lateral friction margins for a tractor semi-trailer do

98 not change as significantly as do passenger vehicles for combined lane changes and braking. This is primarily due to the longer length, slower response, and tires that are less sensitive to changes in loading conditions. 15. There were minimal differences in friction margins between a normally loaded tractor semi-trailer and fully loaded tractor semi-trailer. 16. Unlike passenger vehicles, it is difficult to predict which axle on the tractor semi-trailer will experience the mini- mum lateral friction margins. The lowest-margin axle changes depending on the maneuver, the loading condi- tion, and braking situation. 4.9 Step 8: Predict Skidding of Individual Axles during Transient Steering Maneuvers and Severe Braking The objective of Step 8 was to identify whether severe braking while traversing a sharp horizontal curve affects the ability of a vehicle to traverse the curve without skidding, taking into consideration the horizontal curvature, grade, and superelevation. Using the transient bicycle model from Section 4.8, additional braking inputs were simulated to determine cornering forces and friction factors for the same vehicle/maneuver sets as used earlier. These simulations were used to check whether the acceptable road geometries in pre- vious steps are still suitable for decelerations rates assumed in calculating stopping sight distance design criteria and emer- gency braking maneuvers. Further, the results of the transient and steady-state bicycle models were compared to determine whether the steady-state models of Section 4.7 agree with the transient models of Section 4.8 for severe braking events. Finally, for situations where the margins become zero or neg- ative, the lateral skid distances were calculated, assuming the vehicle is skidding during the duration of negative margins. These lateral distances represent approximations of how far a skidding vehicle will deviate out of the lane, and thus give some means of comparing severity of skid events that may occur in extreme situations. 4.9.1 Analysis Approach The aim of this step was to further utilize the transient bicycle model described in Step 7 to consider severe brak- ing conditions. The equations of motion from the previous model still apply; however, they are used under some assump- tions that must be clear to understand the results that follow. First, the vehicles are simulated under high-friction con- ditions, and the lateral friction margins are calculated by comparing the resulting tire forces with the friction sup- ply assuming low-friction situations (i.e., wet-road, 2nd- percentile road conditions). This means that the high-friction road simulations will usually not exhibit skidding-related effects that would otherwise occur in low-friction roads, such as spin-out, skid-reduced steering, and longer deceleration times. As an example, consider a vehicle that is braking in two scenarios: for 5 s at 10 ft/s2 and for 10 s at 5 ft/s2. Assume also that the high-friction road can maintain these deceleration braking levels, but that the low-friction road will cause skid- ding in both cases. If the durations of skidding are calculated from the two maneuvers using a high-friction road simula- tion, then the lower deceleration braking situation would appear to skid longer, despite a lower applied braking level. This result would not occur in practice, and thus the results are erroneous. One way to prevent these types of errors is to simulate the skidding event on low-friction roads; however, this is quite difficult because the model complexity vastly increases for skidding cases due to many factors including the additional fidelity necessary in the tire model, the presence of ABS, and the need for a well-defined driver model. Further, the transition between road types becomes a critical factor. An actual vehicle can drive from a high-friction surface onto a low-friction surface, and thus the friction demand gener- ated at the onset of a maneuver (e.g., on dry pavement) might not be met by the roadway further within the maneuver (e.g., on wet pavement). For this analysis, most of the maneuvers of interest should have no skidding, or often very short- duration skidding when it does occur. For simulation results presented in Section 4.9.2 where there are long-duration skidding events (i.e., more than a few seconds), the results should be interpreted with caution. The lateral friction margins alone can also be deceptive, as a negative margin during a maneuver does not indicate the severity of the low-friction behavior. For example, consider two vehicles that have the same minimum margins of -0.1 for a particular maneuver. However, one vehicle may have a 10 s skidding event, while the other a skid duration of 0.01 s. Thus, the margin alone does not indicate how long a vehicle is oper- ating within that margin, and thus it is sometimes not a suf- ficient indicator of road condition. This is particularly true for transient situations like lane changes where only short dura- tions of high friction supply are needed, and thus only very short durations of skidding would be expected. To differentiate between skidding events, the duration of the skidding event can be noted in each simulation of the transient vehicle model. Indeed, this is a key benefit of this model versus the point-mass and steady-state models. If the skid duration is known, then one can calculate how much the vehicle is expected to deviate laterally from its lane (i.e., intended path) during the skid. This can be done for each skidding event to classify the severity of the skid. For short- duration skids (i.e., fractions of a second), these estimates should be fairly accurate. For long-duration skidding events,

99 these lateral deviation estimates will be less accurate due to the assumptions mentioned previously as well as due to approximations used in the derivation of the lateral devia- tion distance. To calculate the lateral deviation distance, some basic assumptions must be made about the vehicle within the skid. First, the lateral deviation is measured from the center of the original lane, and the distance is obtained by simple integra- tion of the lateral acceleration to obtain lateral distance. For this integration, the velocity of the vehicle is assumed to be constant during the skid event, at a value equal to the speed at the onset of the skid (usually the design speed of the road). It is also assumed that the driver does have the capability of steering back into the lane, which implies that there is no ABS or other stability systems on the vehicle. Under these assump- tions, the lateral deviation distance is given by: i = −  12 100 (88)Lat Dev 2 skid 2y V R g e t Where yLat Dev is the lateral deviation distance, V is the forward velocity of the vehicle at the onset of the skid, R is the curve radius, g is the gravitational constant, e is the superelevation, and tskid is the time duration of the skid event. While the constant-speed assumption is quite good for short-duration skid events, it will be a very poor assumption if the skid lasts long enough for speed to change appreciably, generally more than a second or two, or if the vehicle departs the high- friction driving lane into a low-friction shoulder, for example. Thus, for lateral deviation distances of more than half a lane width, the lateral deviation estimates rapidly become errone- ous. Further, vehicles with ABS or other stability systems will generally be able to steer in a manner to maintain position on the road; however, their braking forces will be limited to the peak friction values of the road and thus the actual decelera- tion and lateral motion of the vehicle are likely to not match driver expectations. Figure 86 provides a sample illustration of lateral deviation distances experienced by a vehicle during a skidding event. In presenting the results of this analysis, comparisons of lateral friction margins are provided for four deceleration levels (ax = 0, -3, -11.2, and -15 ft/s2) at 10 different grades (0% to -9%) at each speed. Each deceleration level, simulated across the 10 grades, tends to produce a “ribbon” of margins, and thus there are four different “ribbons” of margins in each plot. Each ribbon is plotted so that the 0% grade case is the thick black line, the -9% case is a darker grey line, and inter- mediate grades are light grey lines between these high/low levels. For some situations, the plots of lateral friction mar- gins are so far below zero that it is not practical to extend the axes lower without making them much more different than the other plots and thus making comparisons quite difficult. In cases where the margins do not appear on the plot, the highest and lowest margins are noted via text near the bottom of the figure. 4.9.2 Analysis Results The first set of simulation results compare the predictions of the steady-state bicycle model to the transient bicycle model, for situations where there was only braking on the curve (i.e., no lane-change maneuvers). The results for the E-class sedan, E-class SUV, full-size SUV, single-unit truck, and tractor semi-trailer are shown in Figures 87 to 91. These figures show that the overall agreement between the two models is quite good. Both models show similar trends across all vehicles and braking conditions, and the numerical values for the lateral friction margins are within acceptable error levels, generally with ± 0.04 differences in margins. In comparing the results from the steady-state bicycle models with the results from the transient bicycle models in Figures 87 to 91, the areas of disagreement between the two models are important to mention. First, for the higher brak- ing levels, for ax = -15 ft/s2 in particular, the lateral friction margins appear to increase slightly with increasing speed for the steady-state models, but they are often flat or decrease with increasing speed for the transient models. This is because the transient models, unlike the steady-state models, include the additional forces necessary to initiate rotation into the curve. These “turn-in” forces are apparently small; hence, the general agreement between the models. However, the forces appear to increase with speed; hence, the reason that the tran- sient models have slightly different trends than the steady- state models. Figure 86. Diagram showing lateral deviation distance for skidding vehicle.

100 Figure 87. Lateral friction margins from steady-state bicycle ( left plots) and transient bicycle (right plots) models for E-class sedan (G  0% to 9%, e  0% and 16%) (ax  0, 3, 11.2, and 15 ft/s2). 30 40 50 60 70 80 -0.1 0 0.1 0.2 0.3 0.4 Speed (mph) M in im um F y M ar gi n SSD Decel, ax = -11.2 ft/s2 E-Class Sedan, e = 0% Emergency Decel, ax = -15 ft/s2 Grade = 0% Intermediate grades Grade = -9% 30 40 50 60 70 80 -0.1 0 0.1 0.2 0.3 0.4 Speed (mph) M in im um F y M ar gi n Constant Speed, ax = 0 ft/s2 Curve Entry Deceleration, ax = -3 ft/s2 SSD Decel, ax = -11.2 ft/s2 E-Class Sedan, e = 0% Emergency Decel, ax = -15 ft/s2 Grade = 0% Intermediate grades Grade = -9% 30 40 50 60 70 80 -0.1 0 0.1 0.2 0.3 0.4 Speed (mph) M in im um F y M ar gi n Constant Speed, ax = 0 ft/s2 Curve Entry Decel, ax = -3 ft/s2 SSD Decel, ax = -11.2 ft/s2 E-Class Sedan, e = 16% Emergency Decel, ax = -15 ft/s2 Grade = 0% Intermediate grades Grade = -9% 30 40 50 60 70 80 -0.1 0 0.1 0.2 0.3 0.4 Speed (mph) M in im um F y M ar gi n Constant Speed, ax = 0 ft/s2 Curve Entry Decel, ax = -3 ft/s2 SSD Decel, ax = -11.2 ft/s2 E-Class Sedan, e = 16% Emergency Decel, ax = -15 ft/s2 Grade = 0% Intermediate grades Grade = -9% Curve Entry Decel, ax = -3 ft/s2 Constant Speed, ax = 0 ft/s2 The second area of disagreement between the model results is seen by comparing the 0% superelevation case (top plots) to the 16% superelevation case (bottom plots). In the transient models, for the high superelevation cases (the bottom right plots for each vehicle), the lateral friction margins under mild braking conditions appear to drop, whereas the margins for the steady-state model appear to rise, with increasing speeds. This is due to the situation noted in Section 4.8 where the vehicles in very high superelevation cases are actually expe- riencing their lowest margins on the tangent approach to the curve, rather than within the curve itself. This situation, explained earlier, manifests itself as decreasing margins with increasing speeds for normal driving, but disappears when brakes are being applied because the braking events happen within the curve. And, this situation is only of concern on curves with high superelevation (greater than 12%). The situations considered in Figures 87 to 91 are aggres- sive enough that lateral friction margins approach zero or become negative. In both models, this generally occurs at the 0% grade situation, for the emergency braking situation. Further, for the transient models, the margin curves at this level are generally “flat” across all passenger vehicles (e.g., this boundary, averaged over all passenger vehicles, neither rises nor drops with respect to speed). The flatness and location of this specific lateral friction margin is important. If a friction demand curve is zero and flat, this means that the AASHTO maximum side friction curve exactly matches the difference between supply and demand for this situation. Thus, the plots indicate that the present AASHTO policy, in general, supplies enough friction for all non-emergency maneuvers on wet roads, as long as the friction levels on those roads are no less than two standard deviations below the mean. Thus, the pres- ent AASHTO policy curves appear to form a good estimate of the curve-keeping margins necessary for all non-emergency maneuvers for passenger vehicles. The limiting vehicle, as noted in Section 4.8, appears to be the single-unit truck, shown in Figure 90. This vehicle has par- ticularly low lateral friction margins, indeed so low that the fric-

101 Figure 88. Lateral friction margins from steady-state bicycle ( left plots) and transient bicycle (right plots) models for E-class SUV (G  0% to 9%, e  0% and 16%) (ax  0, 3, 11.2, and 15 ft/s2). 30 40 50 60 70 80 -0.1 0 0.1 0.2 0.3 0.4 Speed (mph) M in im um F y M ar gi n Curve Entry Decel, ax = -3 ft/s2 SSD Decel, ax = -11.2 ft/s2 E-Class SUV, e = 0% Emergency Decel, ax = -15 ft/s2 Grade = 0% Intermediate grades Grade = -9% 30 40 50 60 70 80 -0.1 0 0.1 0.2 0.3 0.4 Speed (mph) M in im um F y M ar gi n Constant Speed, ax = 0 ft/s2 Curve Entry Decel, ax = -3 ft/s2 SSD Decel, ax = -11.2 ft/s2 Emergency Decel, ax = -15 ft/s2 Grade = 0% Intermediate grades Grade = -9% 30 40 50 60 70 80 -0.1 0 0.1 0.2 0.3 0.4 Speed (mph) M in im um F y M ar gi n Curve Entry Decel, ax = -3 ft/s2 SSD Decel, ax = -11.2 ft/s2 E-Class SUV, e = 16% Emergency Decel, ax = -15 ft/s2 Grade = 0% Intermediate grades Grade = -9% 30 40 50 60 70 80 -0.1 0 0.1 0.2 0.3 0.4 Speed (mph) M in im um F y M ar gi n Constant Speed, ax = 0 ft/s2 Curve Entry Decel, ax = -3 ft/s2 SSD Decel, ax = -11.2 ft/s2 Emergency Decel, ax = -15 ft/s2 Grade = 0% Intermediate grades Grade = -9% E-Class SUV, e = 0% E-Class SUV, e = 16% Constant Speed, ax = 0 ft/s2 Constant Speed, ax = 0 ft/s2 tion margins for the stopping sight deceleration and emergency braking deceleration situations are all below zero. This is true even for the 0% grade case. Additionally, the sensitivity of this vehicle to grade is quite high, as evidenced by the very “thick” bands for each situation, as compared to the other vehicles. The reason this vehicle is such an anomaly versus the others is best understood by noting that it has a center of gravity, roughly twice as high (3.85 ft) as the E-class sedan’s (1.93 ft), yet its dis- tance from the front to rear axles is only 64% longer (16.4 ft for the truck versus 10.0 ft for the sedan). Thus, there is a much larger rear-to-front weight shift on this vehicle than for other vehicles. This results in the rear tires having much lower normal force, which means that the friction ellipse, the size of which is roughly proportional to normal load on the tire, shrinks con- siderably. However, the braking and cornering forces necessary for a maneuver are governed by the vehicle’s mass, and thus do not change significantly. Thus, the friction ellipse is shrinking on the rear tires for this vehicle precisely when the demanded forces are growing. The result is very low lateral friction margins. In contrast to the single-unit truck, the tractor semi-trailer (Figure 91) exhibits very little change in lateral friction mar- gin relative to different braking conditions, at least compared to other vehicles. This is likely because the situation for the tractor semi-trailer is nearly opposite that of the single-unit truck. Its CG height (5.45 ft) is 2.8 times that of the E-class sedan, but the length of the semi-trailer alone is 4.5 times lon- ger. If the semi-trailer and tractor are included together, the tractor semi-trailer is 6.4 times longer than the sedan. Thus, the braking and grade sensitivity of the tractor semi-trailer is expected to be 1⁄3 that of the sedan, whereas the sensitivity of the single-unit truck would be roughly 25% higher. The next set of simulations compared the lateral friction margins for normal curve-following maneuvers (i.e., the intended trajectory of the vehicle is within the same lane on the approach tangent and through the curve) to the margins that are observed during lane-change maneuvers within the curve. In both cases, the transient bicycle model was used. The results for the E-class sedan, E-class SUV, full-size SUV,

102 Figure 89. Lateral friction margins from steady-state bicycle (left plots) and transient bicycle (right plots) models for full-size SUV (G  0% to 9%, e  0% and 16%) (ax  0, 3, 11.2, and 15 ft/s2). 30 40 50 60 70 80 -0.1 0 0.1 0.2 0.3 0.4 Speed (mph) M in im um F y M ar gi n Constant Speed, ax = 0 ft/s2 Curve Entry Decel, ax = -3 ft/s2 SSD Decel, ax = -11.2 ft/s2 Full-Sized SUV, e = 0% Emergency Decel, ax = -15 ft/s2Grade = 0% Intermediate grades Grade = -9% 30 40 50 60 70 80 -0.1 0 0.1 0.2 0.3 0.4 Speed (mph) M in im um F y M ar gi n Constant Speed, ax = 0 ft/s2 Curve Entry Decel, ax = -3 ft/s2 SSD Decel, ax = -11.2 ft/s2 Emergency Decel, ax = -15 ft/s2 Grade = 0% Intermediate grades Grade = -9% 30 40 50 60 70 80 -0.1 0 0.1 0.2 0.3 0.4 Speed (mph) M in im um F y M ar gi n Curve Entry Decel, ax = -3 ft/s2 SSD Decel, ax = -11.2 ft/s2 Full-Sized SUV, e = 16% Emergency Decel, ax = -15 ft/s2 Grade = 0% Intermediate grades Grade = -9% 30 40 50 60 70 80 -0.1 0 0.1 0.2 0.3 0.4 Speed (mph) M in im um F y M ar gi n Constant Speed, ax = 0 ft/s2 Curve Entry Decel, ax = -3 ft/s2 SSD Decel, ax = -11.2 ft/s2 Emergency Decel, ax = -15 ft/s2 Grade = 0% Intermediate grades Grade = -9% Full-Sized SUV, e = 16% Full-Sized SUV, e = 0% Constant Speed, ax = 0 ft/s2 single-unit truck, and tractor semi-trailer are shown in Fig- ures 92 to 96, respectively. For passenger vehicles, the lane- change maneuver reduces the margins by approximately 0.25 across all speeds. Interestingly, the single-unit truck and tractor semi-trailer margins are reduced by only 0.1 to 0.15; as noted before, this is due to the larger mass of these vehicles and their slower lane-change durations. In the pas- senger vehicles and in the single-unit truck, the presence of a lane-change maneuver magnifies the effect of grades on margins, as seen by the thicker “ribbons” associated with each situation. Additionally, with lane changes, the margins for the stopping sight distance deceleration situations gener- ate wider “ribbons” with increasing speed. Most notably, with lane-change maneuvers combined with stopping sight distance decelerations or emergency brak- ing decelerations, all vehicles except the tractor semi-trailer exhibit negative margins. For the E-class sedan (Figure 92), the stopping sight distance deceleration is only slightly nega- tive (with margins around -0.05) and relatively “flat” with little change with speed. In contrast, the SUVs (Figures 93 and 94) and single-unit truck (Figure 95) exhibit stopping sight distance and emergency braking deceleration lateral friction margins that are well below zero when lane changes are required during these maneuvers. Because these negative margins occur during a lane change, it is appropriate to con- sider how “severe” these events are by analyzing the corre- sponding lateral deviation distance. Shown in Figures 97 to 101 are the lateral deviation dis- tances for all cases of grades 0% to -9% and zero supereleva- tion where negative lateral friction margins were observed in the combined lane-change and deceleration cases. Note, for the large superelevation case (i.e., e = 16%), the lateral deviation distances were roughly within 5% of the 0% super- elevation case. For the passenger vehicles (Figures 97 to 99), the stopping sight distance deceleration cases (ax = -11.2 ft/s2) show very little lateral deviation in general, as all values are less

103 Figure 90. Lateral friction margins from steady-state bicycle ( left plots) and transient bicycle (right plots) models for single-unit truck (G  0% to 9%, e  0% and 16%) (ax  0, 3, 11.2, and 15 ft/s2). 30 40 50 60 70 80 -0.1 0 0.1 0.2 0.3 0.4 Speed (mph) M in im um F y M ar gi n Constant Speed, ax = 0 ft/s2 Curve Entry Decel, ax = -3 ft/s2 Single Unit Truck, e = 0% (not shown) Emergency Decel, ax = -15 ft/s2 (from -0.14 to -0.55) Grade = 0% Intermediate grades Grade = -9% 30 40 50 60 70 80 -0.1 0 0.1 0.2 0.3 0.4 Speed (mph) M in im um F y M ar gi n Constant Speed, ax = 0 ft/s2 Curve Entry Decel, ax = -3 ft/s2 (not shown) SSD Decel, ax = -11.2 ft/s2 (from -0.15 to -0.36) Single Unit Truck, e = 0% (not shown) Emergency Decel, ax = -15 ft/s2 (from -0.24 to -0.54) Grade = 0% Intermediate grades Grade = -9% 30 40 50 60 70 80 -0.1 0 0.1 0.2 0.3 0.4 Speed (mph) M in im um F y M ar gi n Constant Speed, ax = 0 ft/s2 Curve Entry Decel, ax = -3 ft/s2 Single Unit Truck, e = 16% (not shown) Emergency Decel, ax = -15 ft/s2 (from -0.14 to -0.55) Grade = 0% Intermediate grades Grade = -9% 30 40 50 60 70 80 -0.1 0 0.1 0.2 0.3 0.4 Speed (mph) M in im um F y M ar gi n Constant Speed, ax = 0 ft/s2 Curve Entry Decel, ax = -3 ft/s2 (not shown) SSD Decel, ax = -11.2 ft/s2 (from -0.24 to -0.34) Single Unit Truck, e = 16% (not shown) Emergency Decel, ax = -15 ft/s2 (from -0.30 to -0.50) Grade = 0% Intermediate grades Grade = -9% (not shown) SSD Decel, ax = -11.2 ft/s2 (from -0.11 to -0.36) (not shown) SSD Decel, ax = -11.2 ft/s2 (from -0.11 to -0.36) than 1.5 ft. The worst-case situations are for 0% grades, and for design speeds between 55 and 75 mph. For most of these situations, the durations of the potential skid are so small that it is questionable whether it would affect the driver’s ability to maintain the vehicle on the road. In contrast, the situa- tions with emergency braking decelerations (ax = -15 ft/s2) show much larger lateral deviation distances. Particularly for the SUV cases in Figures 98 and 99, the lateral deviation distances become particularly severe at -4% grade for high speeds (85 mph) and at all speeds for grades larger than -7%. The 12 ft contour that extends from 85 mph/-4% grade to 25 mph/-7% grade is an important dividing line, as this lat- eral deviation distance represents one full lane width. For the single-unit truck lane-change situations shown in Figure 100, the lateral deviation distances are particularly severe. However, some anomalies are evident in that the lower deceleration appears to give larger lateral deviation distances; as noted earlier, this is due to the methodology to calcu- late lateral deviation distance from simulations that do not include skidding dynamics. Because the deceleration is lower, the vehicle will be operating for a longer duration in simula- tion. In reality, the skidding vehicle will likely be unable to achieve the higher emergency braking deceleration levels of ax = -15 ft/s2, and thus the vehicle will actually skid longer than predicted by simulations. Therefore, the very large lateral deviation distances in this plot are likely low estimates due to the over-estimation of available deceleration, but they are also likely to be high estimates due to the assumption that the speed is constant during the skid and equal to the skid-onset speed. In any case, the magnitude of the lateral deviation dis- tance indicates that the single-unit truck experiences lateral friction margins low enough to be of concern. Just as in the lane-change situations, lateral deviation dis- tances can be calculated for the negative lateral friction margin situations where there are no lane changes, just simple curve keeping (i.e., the intended trajectory of the vehicle is within

104 30 40 50 60 70 80 -0.1 0 0.1 0.2 0.3 0.4 Speed (mph) M in im um F y M ar gi n Constant Speed, ax = 0 ft/s2 SSD Decel, ax = -11.2 ft/s2 Tractor Trailer, e = 0% Emergency Decel, ax = -15 ft/s2 Grade = 0% Intermediate grades Grade = -9% 30 40 50 60 70 80 -0.1 0 0.1 0.2 0.3 0.4 Speed (mph) M in im um F y M ar gi n Constant Speed, ax = 0 ft/s2 Curve Entry Decel, ax = -3 ft/s2 Tractor Trailer, e = 0% Emergency Decel, ax = -15 ft/s2 Grade = 0% Intermediate grades Grade = -9% 30 40 50 60 70 80 -0.1 0 0.1 0.2 0.3 0.4 Speed (mph) M in im um F y M ar gi n Curve Entry Decel, ax = -3 ft/s2 Tractor Trailer, e = 16% Emergency Decel, ax = -15 ft/s2 Grade = 0% Intermediate grades Grade = -9% 30 40 50 60 70 80 -0.1 0 0.1 0.2 0.3 0.4 Speed (mph) M in im um F y M ar gi n Constant Speed, ax = 0 ft/s2 Curve Entry Decel, ax = -3 ft/s2 Tractor Trailer, e = 16% Emergency Decel, ax = -15 ft/s2 Grade = 0% Intermediate grades Grade = -9% Constant Speed, ax = 0 ft/s2 SSD Decel, ax = -11.2 ft/s2 Curve Entry Decel, ax = -3 ft/s2 SSD Decel, ax = -11.2 ft/s2 SSD Decel, ax = -11.2 ft/s2 Figure 91. Lateral friction margins from steady-state bicycle ( left plots) and transient bicycle (right plots) models for tractor semi-trailer (G  0% to 9%, e  0% and 16%) (ax  0, 3, 11.2, and 15 ft/s2). Figure 92. Lateral friction margins while maintaining the same lane ( left plots) and with a lane change (right plots) for E-class sedan (G  0% to 9%, e  0%) (ax  0, 3, 11.2, and 15 ft/s2). 30 40 50 60 70 80 -0.1 0 0.1 0.2 0.3 0.4 Speed (mph) M in im um F y M ar gi n Curve Entry Deceleration, ax = -3 ft/s2 SSD Decel, ax = -11.2 ft/s2 E-Class Sedan, e = 0% Emergency Decel, ax = -15 ft/s2 Grade = 0% Intermediate grades Grade = -9% 30 40 50 60 70 80 -0.1 0 0.1 0.2 0.3 0.4 Speed (mph) M in im um F y M ar gi n Constant Speed, ax = 0 ft/s2 Curve Entry Decel, ax = -3 ft/s2 SSD Decel, ax = -11.2 ft/s2 E-Class Sedan, e = 0% (not shown) Emergency Decel, ax = -15 ft/s2 (from -0.16 to -0.51) Grade = 0% Intermediate grades Grade = -9% Constant Speed, ax = 0 ft/s2

105 Figure 93. Lateral friction margins while maintaining the same lane (left plots) and with a lane change (right plots) for E-class SUV (G  0% to 9%, e  0%) (ax  0, 3, 11.2, and 15 ft/s2). 30 40 50 60 70 80 -0.1 0 0.1 0.2 0.3 0.4 Speed (mph) M in im um F y M ar gi n Constant Speed, ax = 0 ft/s2 Curve Entry Decel, ax = -3 ft/s2 SSD Decel, ax = -11.2 ft/s2 Emergency Decel, ax = -15 ft/s2 Grade = 0% Intermediate grades Grade = -9% 30 40 50 60 70 80 -0.1 0 0.1 0.2 0.3 0.4 Speed (mph) M in im um F y M ar gi n Constant Speed, ax = 0 ft/s2 Curve Entry Decel, ax = -3 ft/s2 SSD Decel, ax = -11.2 ft/s2 E-Class SUV, e = 0% (not shown) Emergency Decel, ax = -15 ft/s2 (from -0.35 to -0.72) Grade = 0% Intermediate grades Grade = -9% E-Class SUV, e = 0% Figure 94. Lateral friction margins while maintaining the same lane ( left plots) and with a lane change (right plots) for full-size SUV (G  0% to 9%, e  0%) (ax  0, 3, 11.2, and 15 ft/s2). 30 40 50 60 70 80 -0.1 0 0.1 0.2 0.3 0.4 Speed (mph) M in im um F y M ar gi n Constant Speed, ax = 0 ft/s2 Curve Entry Decel, ax = -3 ft/s2 SSD Decel, ax = -11.2 ft/s2 Emergency Decel, ax = -15 ft/s2 Grade = 0% Intermediate grades Grade = -9% 30 40 50 60 70 80 -0.1 0 0.1 0.2 0.3 0.4 Speed (mph) M in im um F y M ar gi n Constant Speed, ax = 0 ft/s2 Curve Entry Decel, ax = -3 ft/s2 (not shown) SSD Decel, ax = -11.2 ft/s2 (from -0.16 to -0.25) Full-Size SUV, e = 0% (not shown) Emergency Decel, ax = -15 ft/s2 (from -0.39 to -1.00) Grade = 0% Intermediate grades Grade = -9% Full-Size SUV, e = 0% Figure 95. Lateral friction margins while maintaining the same lane ( left plots) and with a lane change (right plots) for single-unit truck (G  0% to 9%, e  0%) (ax  0, 3, 11.2, and 15 ft/s2). 30 40 50 60 70 80 -0.1 0 0.1 0.2 0.3 0.4 Speed (mph) M in im um F y M ar gi n Constant Speed, ax = 0 ft/s2 Curve Entry Decel, ax = -3 ft/s2 (not shown) SSD Decel, ax = -11.2 ft/s2 (from -0.15 to -0.36) Single Unit Truck, e = 0% (not shown) Emergency Decel, ax = -15 ft/s2 (from -0.24 to -0.54) Grade = 0% Intermediate grades Grade = -9% 30 40 50 60 70 80 -0.1 0 0.1 0.2 0.3 0.4 Speed (mph) M in im um F y M ar gi n Constant Speed, ax = 0 ft/s2 Curve Entry Decel, ax = -3 ft/s2 (not shown) SSD Decel, ax = -11.2 ft/s2 (from -0.41 to -0.56) Single Unit Truck, e = 0% (not shown) Emergency Decel, ax = -15 ft/s2 (from -0.60 to -0.83) Grade = 0% Intermediate grades Grade = -9%

106 30 40 50 60 70 80 -0.1 0 0.1 0.2 0.3 0.4 Speed (mph) M in im um F y M ar gi n Constant Speed, ax = 0 ft/s2 Curve Entry Decel, ax = -3 ft/s2 Tractor Trailer, e = 0% Emergency Decel, ax = -15 ft/s2 Grade = 0% Intermediate grades Grade = -9% 30 40 50 60 70 80 -0.1 0 0.1 0.2 0.3 0.4 Speed (mph) M in im um F y M ar gi n Constant Speed, ax = 0 ft/s2 Curve Entry Decel, ax = -3 ft/s2 SSD Decel, ax = -11.2 ft/s2 Tractor Trailer, e = 0% Emergency Decel, ax = -15 ft/s2 Grade = 0% Intermediate grades Grade = -9% SSD Decel, ax = -11.2 ft/s2 Figure 96. Lateral friction margins while maintaining the same lane (left plots) and with a lane change (right plots) for tractor semi-trailer (G  0% to 9%, e  0%) (ax  0, 3, 11.2, and 15 ft/s2). Figure 97. Lateral deviation distances (ft) for all situations with negative margins for E-class sedan (G  0% to 9%, e  0%) (ax  11.2 and 15 ft/s2 and lane change). 0.0 30. 06 0.06 0.08 0 . 08 0.08 0 .08 0.08 0.0 8 0.11 0 .11 0. 11 0.110 .14 0 .14 0. 14 0 .17 0.17 Speed (mph) G ra de (% ) EclassSedan, e = 0%, ax=11.2 ft/s2 30 40 50 60 70 80 -9 -8 -7 -6 -5 -4 -3 -2 -1 0 0.6 5 0.65 0.8 2 0.8 2 0 .82 0.98 0.98 0.9 8 0 .98 1.1 5 1.15 1.15 1.3 1 1.31 1 .31 1.48 1.48 Speed (mph) G ra de (% ) EclassSedan, e = 0%, ax=15 ft/s2 30 40 50 60 70 80 -9 -8 -7 -6 -5 -4 -3 -2 -1 0 Figure 98. Lateral deviation distances (ft) for all situations with negative margins for E-class SUV (G  0% to 9%, e  0%) (ax  11.2 and 15 ft/s2 and lane change). 0. 23 0. 3 0. 3 0. 37 0. 37 0. 37 0. 44 0. 44 0. 44 0. 51 0. 51 0. 51 0.59 0 59 0. 59 0. 59 0. 59 0. 59 0. 59 0. 59 Speed (mph) G ra de (% ) EclassSUV, e = 0%, ax=11.2 ft/s2 30 40 50 60 70 80 -9 -8 -7 -6 -5 -4 -3 -2 -1 0 11.45 11.45 11.45 21.58 21.58 21.58 31 . 71 31.71 31.71 41 . 84 41.84 41.84 51.97 51.97 62 .1 62.1 72.23 Speed (mph) G ra de (% ) EclassSUV, e = 0%, ax=15 ft/s2 30 40 50 60 70 80 -9 -8 -7 -6 -5 -4 -3 -2 -1 0

107 Figure 99. Lateral deviation distances (ft) for all situations with negative margins for full-size SUV (G  0% to 9%, e  0%) (ax  11.2 and 15 ft/s2 and lane change). 0. 83 0 .83 0. 89 0. 89 0. 89 0. 9 5 0. 95 0. 95 0. 951. 01 1. 01 1. 01 1. 01 1. 01 1. 01 1 .07 1. 07 1. 07 1. 07 1. 07 1. 07 1 .13 1 .13 1. 13 1. 13 1. 13 1. 13 1. 19 Speed (mph) G ra de (% ) FullSUV, e = 0%, ax=11.2 ft/s2 30 40 50 60 70 80 -9 -8 -7 -6 -5 -4 -3 -2 -1 0 11.5 11.5 11.5 21.62 21.62 21.62 31.74 31.74 31.74 41.87 41.87 51.99 51.99 62 . 11 72 .23 Speed (mph) G ra de (% ) FullSUV, e = 0%, ax=15 ft/s2 30 40 50 60 70 80 -9 -8 -7 -6 -5 -4 -3 -2 -1 0 Figure 100. Lateral deviation distances (ft) for all situations with negative margins for single-unit truck (G  0% to 9%, e  0%) (ax  11.2 and 15 ft/s2 and lane change). 55 . 34 55 . 34 55 . 34 83 . 9 83 . 9 83 . 9 11 2. 46 11 2.4 6 11 2.4 6 14 1.0 2 14 1.0 2 141.02 16 9.5 8 169.58 198 . 14 198 .14 Speed (mph) G ra de (% ) SingleUnitTruck, e = 0%, ax=1 .2 ft/s2 30 40 50 60 70 80 -9 -8 -7 -6 -5 -4 -3 -2 -1 0 23 .3 9 23 .3 9 31 .5 3 31 .5 3 39 . 67 39 .6 7 47 .8 1 47 .8 1 5 5 . 95 55 . 95 64 .0 9 64 .0 9 7 2 . 2 3 72 . 23 Speed (mph) G ra de (% ) SingleUnitTruck, e = 0%, ax=15 ft/s2 30 40 50 60 70 80 -9 -8 -7 -6 -5 -4 -3 -2 -1 0

108 the same lane on the approach tangent and through the curve). These results are shown in Figure 102 for all vehicles, for the zero superelevation case. Again, there was very little difference in lateral deviation distances versus changes in superelevation, and hence these plots are not repeated here. The plots indicate that, again, the single-unit truck is of particular concern. For passenger vehicles, the E-class sedan undergoes large lateral deviation for grades steeper than -7% to -8%, and for speeds faster than 50 to 60 mph. The SUV cases again start to exhibit large lateral deviations during emergency braking situations defined by a line connecting the 25 mph/-7% grade situation to the 85 mph/-4% grade situation. 4.9.3 Summary of Key Results from Step 8 In summary, the following findings were obtained from the analysis in Step 8: 1. For situations without lane-change maneuvers, the lat- eral friction margins from the steady-state bicycle models agreed quite closely with those from the transient bicycle models, except in situations with high superelevation and high speeds. 2. The disagreement between the models becomes more pro- nounced with increasing braking levels, and with increas- ing CG height of the vehicle. For the most severe braking levels for the single-unit truck, the transient bicycle model estimates lateral friction margins 0.15 lower than the steady-state bicycle model. 3. For passenger vehicles, lateral friction margins generally are above 0.15 for stopping sight distance decelerations when a vehicle stays within the same lane from the tangent approach through the curve. Lateral friction margins are near zero or become negative when passenger vehicles undergo stopping sight distance deceleration combined with a lane change, and when they undergo emergency braking maneuvers in the curve. 4. Even when the intended trajectory of the vehicle is within the same lane on the approach tangent and through the curve (i.e., no lane change), most vehicles will exhibit near-zero lateral friction margins when experiencing emergency braking decelerations. 5. For all decelerations except stopping sight distance decel- eration, the addition of grade reduces the lateral friction margins. This effect is relatively minor except for emer- gency braking where each percentage drop in grade cor- responds to a large drop in margin, by about 0.03 per each 1% change in grade for full-size SUVs. 6. For transient models with severe braking, the lateral fric- tion margins do not necessarily increase with speed and may often drop slightly with increasing speed. 7. The worst-case vehicle for the transient bicycle model with severe braking is the single-unit truck with lateral friction margins as low as -0.50 for emergency braking and -0.34 for stopping sight distance decelerations. All margins for both braking types, for all speeds, are negative, even for the 0% grade. 8. The tractor semi-trailer is predicted to have relatively high lateral friction margins and, for all the maneuvers evaluated, the lateral friction margins were positive. The tractor semi-trailer was also less sensitive to the effects of grade compared to other vehicles. 9. The presence of a lane-change maneuver within a curve reduces lateral friction margins by approximately 0.25 across all speeds for passenger vehicles and by approxi- mately 0.1 to 0.15 for single-unit trucks and tractor semi- trailers. For passenger vehicles and single-unit trucks, steeper downgrades cause more decrease in margins dur- ing lane changes. 10. When lane-change maneuvers are combined with stop- ping sight distance or emergency braking decelerations, all vehicles except the tractor semi-trailer exhibit negative lateral friction margins. 11. Examining the lateral motion of vehicles during a poten- tial skid, in many situations the vehicles are skidding only a short duration (and distance) when lateral friction margins are potentially negative, for example less than a foot for stop- ping sight decelerations with lane changes for the E-class vehicles. The duration and level of lateral motion did not change noticeably with increasing superelevation. 12. For the worst-case skidding situations, the lateral motion of vehicles—particularly the single-unit truck—is poten- tially quite severe (more than two lanes of lateral motion). Figure 101. Lateral deviation distances (ft) for all situations with negative margins for tractor semi-trailer (G  0% to 9%, e  0%) (ax  15 ft/s2 and lane change). 7 .0497e -007 7.0497e -007 7 .0497e -007 1 .01 1 .01 1 .01 1 .01 2 .01 2.01 2 .01 2 .01 3 .02 3.02 3 .02 3 02 4.03 4 .03 5 .04 6 .04 Speed (mph) G ra de (% ) TractorTrailer, e = 0%, ax=-15 ft/s2 30 40 50 60 70 80 -9 -8 -7 -6 -5 -4 -3 -2 -1 0

109 2.8125e-007 2.8125e-007 2.8125e-007 0.4 0.4 0.8 0.8 1.21 1.612.012.41 Speed (mph) G ra de (% ) EclassSedan, e = 0%, ax=15 ft/s2 30 40 50 60 70 80 -9 -8 -7 -6 -5 -4 -3 -2 -1 0 10.32 10.3 2 10.32 20 .6 4 20.64 20.64 30.9 6 30.96 30.96 41 . 28 41.28 41.28 51.6 51.6 61 .9 1 61.91 72.23 Speed (mph) G ra de (% ) EclassSUV, e = 0%, ax=15 ft/s2 30 40 50 60 70 80 -9 -8 -7 -6 -5 -4 -3 -2 -1 0 10.32 10.32 10.32 20.64 20.6 4 20.64 30.96 30.96 30.96 41.28 41.28 51.6 51.6 61 . 91 72 .23 Speed (mph) G ra de (% ) FullSUV, e = 0%, ax=15 ft/s2 30 40 50 60 70 80 -9 -8 -7 -6 -5 -4 -3 -2 -1 0 55 . 34 55 . 34 5. 34 83 . 9 83 . 9 83 . 9 11 2. 46 11 2.4 6 2.4 6 14 1.0 2 14 1.0 2 141.02 16 9.5 8 169.58 198 . 14 198 .14 Speed (mph) G ra de (% ) SingleUnitTruck, e = 0%, ax=11.2 ft/s2 30 40 50 60 70 80 -9 -8 -7 -6 -5 -4 -3 -2 -1 0 23 .3 9 23 .3 9 31 .5 3 31 .5 3 3 9 . 67 39 . 67 47 .8 1 47 .8 1 55 . 95 55 . 95 64 .0 9 64 .0 9 7 2 . 2 3 72 . 23 Speed (mph) G ra de (% ) SingleUnitTruck, e = 0%, ax=15 ft/s2 30 40 50 60 70 80 -9 -8 -7 -6 -5 -4 -3 -2 -1 0 4 .2382e -007 4 .2382e -007 0 .61 1 .21 1 .822 .423 03 Speed (mph) G ra de (% ) TractorTrailer, e = 0%, ax=-15 ft/s2 30 40 50 60 70 80 -9 -8 -7 -6 -5 -4 -3 -2 -1 0 Figure 102. Lateral deviation distances (ft) for all situations with negative margins for vehicles of interest (G  0% to 9%, e  0%) (ax  11.2 and 15 ft/s2).

110 Without ABS, the transient bicycle model predicts that this vehicle may not be able to maintain its position in the lane on a curve while undergoing stopping sight distance or emergency braking decelerations. 13. For SUVs, there appears to be a boundary for skidding during lane-change maneuvers and stopping sight dis- tance decelerations that extends from grades steeper than -4% at design speeds of 85 mph to grades steeper than -7% at design speeds of 25 mph. 4.10 Step 9: Predict Skidding of Individual Wheels The objective of Step 9 was to use high-order multibody models to predict skidding of individual wheels as a vehicle traverses a sharp horizontal curve on a steep grade. Using com- mercially available vehicle dynamic simulation software (i.e., CarSim and TruckSim), high-order multibody models were used to predict skidding of individual wheels as a vehicle tra- verses a sharp horizontal curve, taking into consideration a range of conditions such as the horizontal curvature, grade, and superelevation. Rather than simulating the full range of hypothetical geometries considered throughout this research, this analysis focused on those situations identified in previous steps as areas of concern. 4.10.1 Analysis Approach In this step, the commercial vehicle simulation package CarSim and the similar truck-oriented software package TruckSim were used to perform nearly full-fidelity simulations of vehicle behavior during traversals of simulated curves. The focus of the simulations is on situations identified in previous sections as areas of concern. These software packages were chosen primarily because they are the most widely used in industry for similar studies. They also allow direct import of known road geometries, or relatively easy specification of hypo- thetical geometries. Further, there is a comprehensive library of vehicles to choose from that cover all the vehicle types in this research. To simulate a vehicle driving down a road with a particu- lar geometric profile in CarSim and TruckSim, the software requires a three-dimensional model of the road. CarSim and TruckSim represent the three-dimensional road on which the virtual vehicle is to be driven based on the following para- metric specifications: 1. Plan-view (XY) geometry of the lane centerline 2. Global road centerline height 3. Local height offset of each lane edge In short, for each of the roadway geometries simulated within CarSim and TruckSim (i.e., either hypothetical curves or the actual curves included in the speed and vehicle maneu- ver studies), the research team created global X,Y,Z coor- dinates describing the roadway geometry for import into CarSim and TruckSim. After constructing the lists of X,Y,Z points describing the road geometry, these lists were com- piled into CarSim/TruckSim for use in simulation. Portions of these procedures were performed using MATLAB. A ren- dering of a hypothetical roadway geometry used within the CarSim and TruckSim simulations is shown in Figure 103. Note, unlike the analyses in Sections 4.9 and earlier, in the multibody models the superelevation transition is simulated (i.e., designed) according to the Green Book policy. For these simulations, the roadway possesses a very high coefficient of friction, higher than any friction supply value used in the preceding sections. This allows for calculations of lateral friction margins according to the method used throughout the analyses by subtracting demand from supply, and taking the tire forces provided by the simulation software as demand only. Again, this is primarily beneficial in that it decouples the simulation results from a specific tire–pavement interaction model used in the simulation that might change for the range of friction values (which change with speed) as measured from field data, as noted in Section 3.4. The implications of the high-friction assumption in simula- tion are important. If the vehicle was simulated on a road with a friction margin as low as the supply values calculated in Sec- tion 4.2, the Force vs. Slip curves would differ drastically from those in Section 4.8, because of the difference in tire model type between CarSim and the transient bicycle model. CarSim uses a Pacejka-type tire model, which has slightly different behavior than the modified linear tire model used in Sections 4.8 and 4.9, and different yet from the LuGre tire model used to derive friction supply values in Section 4.2. To minimize discrepan- cies between models as much as possible, the high-friction road was used to help ensure that the friction demand reported by the multibody model matched the friction demand reported by the transient bicycle model as much as possible. CarSim requires that certain driver and vehicle parameters be specified. Vehicle inertial, tire, and suspension properties were input according to the representative vehicle families dis- cussed earlier. The CarSim driver model is based on a human- like preview controller developed by MacAdam (1981). It requires little user interaction and follows the road centerline by default. When specifying a lane-change maneuver during the traversal of a curve, the plan-view geometry of the lane- change maneuver with respect to the road centerline is speci- fied. If the vehicle should depart from the trajectory, the driver model changes the steering input to correct accordingly. This differs from the methodology employed in Section 4.8, where the vehicle inputs (steering and braking) were specified for a particular lane change directly. This allows the software to more realistically simulate human behavior and driver response.

111 CarSim also requests a desired speed profile for each simula- tion. Speed profiles for vehicles were generated using MATLAB for input into CarSim. Where there are instances of braking, the speed profiles were determined by piece-wise integration of the decelerations starting at the time and distance locations where braking is first applied. Comparison of vehicle inputs such as steering angle, vehi- cle speed, and deceleration assumed for the transient bicycle models in Section 4.8 and for the multibody models in this step showed close agreement. Agreement between the models indi- cated that the input assumptions used for specifying the CarSim simulations were reasonable; and thus, results of the CarSim simulations could be viewed with a high level of confidence. 4.10.2 Analysis Results 4.10.2.1 Validation of Transient Bicycle Models in Step 7 at Constant Speed and Curve-Entry Deceleration To check the fidelity of the transient bicycle models developed in Section 4.8 with a minimum of confounding variables, the most modest situation was first analyzed: a constant-speed traversal of a horizontal curve while main- taining the same lane. This very simple situation was selected as it is important to compare lateral friction margins obtained from CarSim/TruckSim to the transient bicycle model. Once agreement between the transient bicycle model and the multi- body model is confirmed for mild maneuvers, boundary or questionable cases can be evaluated using CarSim/TruckSim to gain a better understanding of which highway design and maneuver scenarios are most concerning. Section 4.8 showed that the overall effects of grade and superelevation, for minimum-radius curves, were far less sig- nificant than vehicle speed, vehicle maneuver type, and vehicle type when determining the lateral friction margins. Thus, most of the simulations that follow are conducted using a moderate combination of superelevation and grade, while the focus is on the three factors that showed the most effects in previous sec- tions: vehicle speed, vehicle maneuver type, and vehicle type. Because the full-size SUV was determined to be the worst- case passenger vehicle for friction margins in Section 4.8 and for rollover propensity in Section 4.6, this vehicle is a focus for many of the following analyses. Figure 104 shows inputs and outputs for both the transient bicycle model and multibody model for a full-size SUV assuming a 55 mph design speed, Figure 103. Three-dimensional rendering of simulated road within CarSim/TruckSim; situation shows an E-class SUV skidding while changing lanes and emergency braking.

112 grade of -6%, and superelevation of 4% and assuming a con­ stant speed throughout the maneuver and the intended tra­ jectory of the vehicle is within the same lane on the approach tangent and through the curve. The resulting lateral friction margins match fairly well between these two models for this mild, steady maneuver. When the multibody model was run for an entire range of speeds, there remained good agreement between the transient bicycle model and multibody model, as shown in Figure 105. In Figure 105, the lateral friction margins for the front and rear axles for the CarSim simulations were determined by tak­ ing the minimum margin experienced by the inside and outside wheels on each axle, respectively. The agreement between multi­ body and transient bicycle models is excellent for this scenario. For trucks, the single­unit truck was determined in previ­ ous sections to be the worst­case vehicle for many scenarios, so this vehicle type is given attention here. Lateral friction margins from the transient bicycle model and multibody 0 5 10 15 20 25 -1 0 1 2 D ec el er at io n (ft/ s2 ) Transient Bicycle Model Multibody Model 0 5 10 15 20 25 54.8 54.9 55 55.1 Sp ee d (m ph ) Transient Bicycle Model Multibody Model 0 5 10 15 20 25 -5 0 5 10 15 x 10-3 Time (s) St ee rin g In pu t (r ad ) Transient Bicycle Model Multibody Model Outside Tire Multibody Model Inside Tire 0 5 10 15 20 25 0.35 0.4 0.45 0.5 0.55 0.6 Fr on t T ire M ar gi n Transient Bicycle Model Multibody Model Outside Tire Multibody Model Inside Tire 0 5 10 15 20 25 0.35 0.4 0.45 0.5 0.55 0.6 Time (s) R ea r T ire M ar gi n Transient Bicycle Model Multibody Model Outside Tire Multibody Model Inside Tire Figure 104. Inputs and outputs from transient bicycle and multibody models for full-size SUV (V  55 mph, G  6%, e  4%) (ax  0 ft/s2). Figure 105. Lateral friction margins from transient bicycle and multibody models for full-size SUV (G  6%, e  4%) (ax  0 ft/s2). e = 4%, G = -6%, FullSUV Speed (mph) R ea r F y M ar gi n Fr o n t F y M ar gi n

113 model are shown in Figure 106 for a -6% grade and a 4% superelevation, assuming constant speed. The rear­margin minimums of the transient model agree with the front­margin minimums from the multibody model. The rear margins are most important, as these must prevent vehicle spin­out; the transient model gives lower margins and thus is slightly more conservative than the multibody model for this vehicle. This has many potential causes, not the least significant of which is likely air resistance. This is an interesting result, as it suggests that the transient bicycle model of the single­unit truck may be the worst­case vehicle choice for highway design. The rec­ ognition of the worst­case vehicle for the most conservative low­order model, in general, is useful; a roadway design that works with this vehicle in this simulation model may ensure a roadway design that is suited for all vehicles. In Figure 107, the lateral friction margins of the transient bicycle model and the multibody model for the tractor semi­ trailer are compared. This comparison is important as the tran­ sient bicycle model for the tractor semi­trailer has an added level of complexity versus the same type of model for a two­ axle vehicle, and thus there are additional potential sources of error. Again, the match between the transient bicycle model and the multibody model is quite good, but the picture is a little less clear with the tractor semi­trailer than with the full­ size SUV or the single­unit truck. There are larger discrepan­ cies visible on all of the axles, particularly on the rear axle of the tractor and the trailer axle. One reason for these discrep­ ancies for the tractor semi­trailer, as mentioned in Section 4.8, is that the transient bicycle model for the tractor semi­trailer “lumps” the tires in each axle group and represents them by one, effective tire. Thus, in the transient bicycle model, eight tires in an axle group might be represented by only one tire. The multibody model for the tractor semi­trailer models five axles and 18 wheels: axle 1 has one tire on each side, while axles 2, 3, 4, and 5 (working toward the back of the loaded combination) have four tires per axle. In summary, Figure 107 shows the minimum lateral friction margins experienced by the inside and outside tires on each axle for the multibody model. It is clear that there is some disparity between the mar­ gins predicted for axles 2 and 3, and again between axles 4 and 5. Still, the “average” margins between the two axles in the “rear” axle group and the “trailer” axle group are in agreement with the transient bicycle model, which is really all one can expect from a model that, by definition, averages tire forces from the two axles in the group. Interestingly, the disparity between the two axles at the rear of the tractor and the two axles at the rear of the trailer can be quite large. They appear to handicap the tractor semi­trailer even at constant speed when speeds are low. For low­speed curves, the trailer articulation angle is high, the steering angle required to navigate the curve is relatively high, and the lat­ eral acceleration required to negotiate the curve is relatively high. Still, care must be taken when jumping to the conclu­ sion that this apparent handicap will cause loss of control of a tractor semi­trailer. Axles 2 and 3 at the rear of the tractor are closely spaced and fixed, as are axles 4 and 5 at the rear of the trailer. These closely spaced axles cannot steer around the same turn center; therefore, turning requires that at least Figure 106. Lateral friction margins from transient bicycle and multibody models for single-unit truck (G  6%, e  4%) (ax  0 ft/s2). e = 4%, G = -6%, SingleUnitTruck Speed (mph) R ea r F y M ar gi n Fr o n t F y M ar gi n

114 one axle exhibits slip. This can result in low margins for one axle of a closely spaced pair. This behavior will be particu­ larly pronounced for low­speed turns, since the turn radii are generally much smaller for lower speeds. Thus, at low speeds, alternating low and high margins for adjacent, closely spaced axles can be expected and do not necessarily constitute a safety concern. To illustrate this phenomenon, consider Figure 108, which shows the margin trajectories for a tractor semi­trailer navigating a 500 ft radius curve at 25 mph. In this low­speed, large­radius turn, the lateral acceleration is so low as to be neg­ ligible, but even so, the rear axle on the tractor unit and the rear axle on the trailer unit exhibit different, “split” margins than their “partner axles,” located just forward of each, respectively. The split is in the opposite sense for this low­speed turn (i.e., the rearward axle in each group has a higher lateral friction margin than the forward axle in each group). This occurrence happens because the axle that “skids” depends heavily on tire slip and weight shift, and both of these factors are affected by the specifics of each driving maneuver. To investigate this “splitting” phenomenon among the tires in each tractor semi­trailer axle group further, and examine its pervasiveness across maneuver types, consider the com­ parison between the transient bicycle model and multibody model for the curve­entry deceleration traversal shown in Figure 109. The “splitting” of margins between the axle groups still occurs, but the transient bicycle model seems to over­predict lateral friction margins for the trailer axle group. This is a potential problem when using the transient bicycle model to evaluate the dynamics of a tractor semi­trailer for geometric design purposes. Assuming that the multibody model is more representative of real­world situations, this disagreement shows that the transient bicycle model is pre­ dicting higher lateral friction margins than they likely are in reality and therefore may miss the occurrence of negative fric­ tion margins. Thus, for tractor semi­trailers or other vehicles with multiple adjacent axles, a multibody simulation should be run to confirm whether negative margins result or not. To investigate why the transient bicycle model does not pre­ dict the trailer margins correctly, consider Figure 110, in which the 30 mph input and margin trajectories for a tractor semi­ trailer, assuming curve­entry deceleration and curve­keeping steering, are shown. Comparing the time trajectory of inputs and margins for this scenario, the reason for the discrepancy between the low­order model and the multibody model is Speed (mph) Tr a ile r F y M ar gi n R e a r Fy M ar gi n Fr o n t F y M ar gi n Figure 107. Lateral friction margins from transient bicycle and multibody models for tractor semi-trailer (G  6%, e  4%) (ax  0 ft/s2).

115 Figure 108. Lateral friction margins from multibody model for all five axles of a tractor semi-trailer (V  25 mph, G  0%, e  0%, R  500 ft). 0 5 10 15 20 25 30 0 0.2 0.4 e = 0%, G = 0%, v = 25 mph, R = 500 ft, TractorTrailer Fr on t T ire M ar gi n Multibody Model Outside Tire Multibody Model Inside Tire 0 5 10 15 20 25 30 0 0.2 0.4 R ea r T ire M ar gi n Multibody Model Outside Tire Axle 2 Multibody Model Inside Tire Axle 2 Multibody Model Outside Tire Axle 3 Multibody Model Inside Tire Axle 3 0 5 10 15 20 25 30 0 0.2 0.4 Time (s) Tr ai le r T ire M ar gi n Multibody Model Outside Tire Axle 4 Multibody Model Inside Tire Axle 4 Multibody Model Outside Tire Axle 5 Multibody Model Inside Tire Axle 5 apparent. At around 14.5 s, the deceleration rate overshoots the desired curve-entry deceleration value of -3 ft/s2. This is due to the simulated driver’s braking model employed by the multibody software, which represents the TruckSim simulated driver’s efforts to maintain a particular deceleration rate by modulating the brakes. The spike in deceleration (resulting from a spike in brake force) at the trailer axle group corre- sponds exactly with the downward spike in trailer lateral fric- tion margin at 14.5 s. This suggests that, while the mean lateral friction margin of each axle group matches well between the transient bicycle model and the multibody model, the driver- influenced braking dynamics of the tractor semi-trailer are especially significant for the trailer axle (e.g., there are oscil- lations in the braking systems of trailers that cause spikes in braking to occur when brakes are suddenly applied). The spike at 14.5 s in Figure 110 corresponds with the minimum reported margin in Figure 109 for 30 mph but should not really be seen as cause for concern, as the spike to a lateral friction margin of nearly zero on the trailer axle is momentary, and an artifact of the simulated driver control of the braking system. 4.10.2.2 Investigation of Weight-Transfer Effects on Lateral Friction Margins In examining the results in the preceding subsections, it becomes apparent that the multibody model, in its predic- tion of the forces present at each tire on the vehicle, predicts slightly different lateral friction margins for the inside and outside tires. This phenomenon has a few contributing fac- tors, like steering geometry (for the front tires), as well as longitudinal and lateral weight transfer during the maneuver (especially on the rear tractor and trailer axles). Tire behav- ior changes as a function of vertical loading, so the forces contributed by each tire are coupled not only to the tire’s slip angle, but also to the way load shifts from the inside to the outside tire during cornering. This behavior is assumed to be nearly negligible for most driving scenarios and is ignored in the assumptions used to derive the transient bicycle model of Section 4.8. To confirm that lateral weight shift has a relatively small influence on the overall lateral friction margin predictions for

Figure 109. Lateral friction margins from transient bicycle and multibody models for tractor semi-trailer (G  6%, e  4%) (ax  3 ft/s2). Speed (mph) Tr a ile r F y M ar gi n R e a r Fy M ar gi n Fr o n t F y M ar gi n 0 2 4 6 8 10 12 14 16 18 20 -6 -4 -2 0 2 D ec el er at io n (ft/ s2 ) Transient Bicycle Model Multibody Model 0 2 4 6 8 10 12 14 16 18 20 15 20 25 30 35 Sp ee d (m ph ) Transient Bicycle Model Multibody Model 0 2 4 6 8 10 12 14 16 18 20 -0.05 0 0.05 0.1 0.15 Time (s) St ee rin g In pu t (r ad ) Transient Bicycle Model Multibody Model Outside Tire Multibody Model Inside Tire 0 2 4 6 8 10 12 14 16 18 20 -0.2 0 0.2 0.4 0.6 Fr on t T ire M ar gi n Transient Bicycle Model Multibody Model Outside Tire Multibody Model Inside Tire 0 2 4 6 8 10 12 14 16 18 20 -0.2 0 0.2 0.4 0.6 R ea r T ire M ar gi n Transient Bicycle Model Multibody Model Outside Tire Axle 2 Multibody Model Inside Tire Axle 2 Multibody Model Outside Tire Axle 3 Multibody Model Inside Tire Axle 3 0 2 4 6 8 10 12 14 16 18 20 -0.2 0 0.2 0.4 0.6 Time (s) Tr ai le r T ire M ar gi n Transient Bicycle Model Multibody Model Outside Tire Axle 4 Multibody Model Inside Tire Axle 4 Multibody Model Outside Tire Axle 5 Multibody Model Inside Tire Axle 5 Figure 110. Inputs and outputs from transient bicycle and multibody models for tractor semi-trailer (V  30 mph, G  6%, e  4%) (ax  3 ft/s2).

117 normal driving, Figure 111 compares the inside versus out- side tire lateral friction margins for a full-size SUV. The over- all differences between the inside and outside tire margins are only a maximum of about 0.06. The front tires experi- ence more disparity than the rear tires mainly due to steering geometry issues (i.e., the vehicle does not possess so-called “Pure Ackermann” steering geometry, which means that the act of steering at moderate to high speeds will, in itself, induce a small relative amount of slip in either the inside or outside tires). This effect may seem detrimental to performance, but it does have design advantages as it can be used to design and tune chassis behavior (e.g., to ensure that vehicles skid prior to rollover). This effect is often present in passenger vehicles. The Ackermann disparity is magnified at low speeds, since the smaller radii of curves with lower design speeds will, by definition, require higher steering amplitudes. Figure 112 shows that the effects of lateral weight shift on friction margins remains small for normal maneuvers, even for a tractor semi-trailer. The TruckSim model predicts lat- eral, longitudinal, and vertical forces for all five axles (and all 18 tires) on a tractor semi-trailer, while the transient bicycle model used in Section 4.8 provides estimates for three rep- resentative “lumped” axles. Again, the differences between inside and outside lateral friction margins are small across the range of speeds for normal driving cases. In summary, the lateral friction margin plots illustrate that Ackermann geometry errors, combined with weight-transfer effects, are together probably small contributors to overall lateral friction margins. Both likely only play a role in the lateral friction margin estimates in extreme cases where the margins predicted by the transient bicycle model are already close to zero due to other factors. While this is certainly sup- portive of the modeling assumptions made in Section 4.8, there is the possibility that under some of the more extreme driving conditions considered, lateral weight-transfer effects could have a more severe effect. Some of these cases are dis- cussed in the sections that follow. 4.10.2.3 Effect of Lane-Change Maneuver at Curve-Entry Deceleration One of the situations of concern identified in Section 4.8 dealt with combined lane-change and braking maneuvers while traversing a horizontal curve. When assuming a lane- change maneuver combined with curve-entry deceleration, the transient bicycle models for the E-class sedan, E-class SUV, and tractor semi-trailer estimated relatively high lateral friction margins, but for the full-size SUV and the single-unit truck, the transient bicycle model estimated relatively low lateral friction margins for these cases. To determine whether this situation is actually of concern for the full-size SUV and the single-unit truck, the worst-case horizontal curves (i.e., curves on -9% Figure 111. Effect of lateral weight shift on lateral friction margins for full-size SUV (G  6%, e  4%) (ax  0 ft/s2). 30 40 50 60 70 80 0.32 0.34 0.36 0.38 0.4 0.42 0.44 Fr on t F y M ar gi n e = 4%, G = -6%, FullSUV Inside Tire Outside Tire 30 40 50 60 70 80 0.34 0.36 0.38 0.4 0.42 Speed (mph) R ea r F y M ar gi n Figure 112. Effect of lateral weight shift on lateral friction margins for tractor semi-trailer (G  6%, e  4%) (ax  0 ft/s2). 30 40 50 60 70 80 0.25 0.3 0.35 0.4 Fr on t F y M ar gi n e = 4%, G = −6%, TractorTrailer Inside Tire Outside Tire 30 40 50 60 70 80 0.38 0.4 0.42 Ax le 2 F y M ar gi n 30 40 50 60 70 80 0.2 0.3 0.4 Ax le 3 F y M ar gi n 30 40 50 60 70 80 0.38 0.4 0.42 Ax le 4 F y M ar gi n 30 40 50 60 70 80 0.1 0.2 0.3 Speed (mph) Ax le 5 F y M ar gi n

118 grade with 0% superelevation) as determined by the simula- tions in Section 4.8 were further evaluated. Some matching scenarios with curves on -9% grade with 8% superelevation are also shown for comparison. Recall, previous results from the transient bicycle model and the multibody model show that the effect of superelevation is quite small overall. A comparison of inputs and outputs is shown in Figure 113 for the transient bicycle model and the multibody model for the single-unit truck and for a 55 mph constant-speed curve traversal on a -9% grade and 0% superelevation. Figure 114 shows the same situation, except for 8% superelevation. The results are nearly identical. Both figures show that the multi- body model predicts higher lateral friction margins than the transient bicycle model. This may be due to extra factors con- sidered in the multibody simulation that are neglected in the transient model, including air resistance, tire lag, roll dynamics, and chassis stiffness. Each of these factors has the potential to “soften” the response of the vehicle to control inputs. Addition- ally, there are slight differences in the simulation inputs (brake, steering). Notably, the steering amplitude is smaller in the multi body model, and the peak of the sinusoid no longer aligns perfectly with the onset of braking. Both effects are due to the multibody simulation’s use of a more human-like driver model. To further examine the differences between the transient bicycle model and the multibody model across speeds, Fig- ure 115 shows lateral friction margins for the front and rear axles for a single-unit truck assuming a lane-change maneu- ver combined with curve-entry deceleration. Both 0% and 8% superelevation cases are shown, and again these cases are quite similar to each other. The multibody model predicts 0 2 4 6 8 10 12 14 16 18 20 -6 -4 -2 0 2 e = 0%, G = −9%, v = 55 mph, SingleUnitTruck D ec el er at io n (ft/ s2 ) Transient Bicycle Model Multibody Model 0 2 4 6 8 10 12 14 16 18 20 30 40 50 60 Sp ee d (m ph ) Transient Bicycle Model Multibody Model 0 2 4 6 8 10 12 14 16 18 20 -0.02 0 0.02 0.04 Time (s) St ee rin g In pu t(r ad ) Transient Bicycle Model Multibody Model Outside Tire Multibody Model Inside Tire 0 2 4 6 8 10 12 14 16 18 20 0 0.2 0.4 Fr on t T ire M ar gi n Transient Bicycle Model Multibody Model Outside Tire Multibody Model Inside Tire 0 2 4 6 8 10 12 14 16 18 20 0 0.2 0.4 Time (s) R ea r T ire M ar gi n Transient Bicycle Model Multibody Outside Tire Multibody Inside Tire Figure 113. Inputs and outputs from transient bicycle and multibody models for single-unit truck (V  55 mph, G  9%, e  0%) (ax  3 ft/s2 and lane change). 0 2 4 6 8 10 12 14 16 18 20 −6 −4 −2 0 2 e = 8%, G = −9%, v = 55 mph, SingleUnitTruck D ec el er at io n (ft/ s2 ) Transient Bicycle Model Multibody Model 0 2 4 6 8 10 12 14 16 18 20 30 40 50 60 Sp ee d (m ph ) Transient Bicycle Model Multibody Model 0 2 4 6 8 10 12 14 16 18 20 −0.02 0 0.02 0.04 Time (s) St ee rin g In pu t (r ad ) Transient Bicycle Model Multibody Model Outside Tire Multibody Model Inside Tire 0 2 4 6 8 10 12 14 16 18 20 0 0.2 0.4 Fr on t T ire M ar gi n Transient Bicycle Model Multibody Model Outside Tire Multibody Model Inside Tire 0 2 4 6 8 10 12 14 16 18 20 0 0.2 0.4 Time (s) R ea r T ire M ar gi n Transient Bicycle Model Multibody Outside Tire Multibody Inside Tire Figure 114. Inputs and outputs from transient bicycle and multibody models for single-unit truck (V  55 mph, G  9%, e  8%) (ax  3 ft/s2 and lane change).

119 low margins but predicts that there will be no skidding across the range of speeds. In contrast, the transient bicycle model predicts that the rear axle could skid or be close to skidding for nearly all speeds. This suggests some conservatism on the part of the transient bicycle model for two-axle vehicles. To determine whether the difference in lateral friction margins is an effect strictly seen on the single-unit truck, Fig- ures 116 and 117 compare the transient bicycle model to the multibody model for the full-size SUV assuming a lane-change maneuver combined with curve-entry deceleration. Again, the multibody model predicts much higher rear-axle margins than the transient bicycle model. And again, this is likely due to the differences in inputs. This should not necessarily be taken as an indication that all lane-change maneuvers combined with curve-entry deceleration result in lateral friction margins greater than zero; rather, it suggests that using the transient bicycle model to evaluate the dynamics of two-axle vehicles for geometric design purposes is a conservative approach. Addi- tionally, Figures 116 and 117 indicate that certain higher-order braking effects have the potential to obscure the fundamental trends in margin calculations for certain cases. While this may be the case for two-axle vehicles, the possibility of disagree- ment between the transient bicycle model and the multibody model for a tractor semi-trailer has already been discussed for curve-entry deceleration without the addition of a lane change. When a lane-change maneuver is superimposed onto the curve-entry deceleration for a tractor semi-trailer, the results of a comparison between models is even more revealing. Con- sider the comparison between models for a 55 mph horizon- tal curve (Figure 118) and across design speeds (Figure 119). Figure 115. Lateral friction margins from transient bicycle and multibody models for single-unit truck (G  9%, e  0% and 8%) (ax  3 ft/s2 and lane change). Speed (mph)Speed (mph) R ea r F y M ar gi n Fr on t F y M ar gi n R ea r F y M ar gi n Fr on t F y M ar gi n Figure 116. Inputs and outputs from transient bicycle and multibody models for full-size SUV (V  55 mph, G  9%, e  0%) (ax  3 ft/s2 and lane change). 0 5 10 15 20 25 30 35 −5 0 5 D ec el er at io n (ft/ s2 ) Transient Bicycle Model Multibody Model 0 5 10 15 20 25 30 35 0 20 40 60 Sp ee d (m ph ) Transient Bicycle Model Multibody Model 0 5 10 15 20 25 30 35 −0.01 0 0.01 0.02 0.03 Time (s) St ee rin g In pu t (r ad ) Transient Bicycle Model Multibody Model Outside Tire Multibody Model Inside Tire 0 5 10 15 20 25 30 35 0.1 0.2 0.3 0.4 0.5 0.6 Fr on t T ire M ar gi n Transient Bicycle Model Multibody Model Outside Tire Multibody Model Inside Tire 0 5 10 15 20 25 30 35 0 0.2 0.4 0.6 0.8 Time (s) R ea r T ire M ar gi n Transient Bicycle Model Multibody Model Outside Tire Multibody Model Inside Tire

120 Figure 117. Lateral friction margins from transient bicycle and multibody models for full-size SUV (G  9%, e  0% and 8%) (ax  3 ft/s2 and lane change). Figure 118. Lateral friction margins from transient bicycle and multibody models for tractor semi-trailer (V  55 mph, G  9%, e  0%) (ax  3 ft/s2 and lane change). 0 5 10 15 20 25 0.2 0.4 0.6 0.8 Fr on t T ire M ar gi n Transient Bicycle Model Multibody Model Outside Tire Multibody Model Inside Tire 0 5 10 15 20 25 0 0.2 0.4 0.6 0.8 R ea r T ire M ar gi n Transient Bicycle Model Multibody Model Outside Tire Axle 2 Multibody Model Inside Tire Axle 2 Multibody Model Outside Tire Axle 3 Multibody Model Inside Tire Axle 3 0 5 10 15 20 25 −0.5 0 0.5 1 Time (s) Tr ai le r T ire M ar gi n Transient Bicycle Model Multibody Model Outside Tire Axle 4 Multibody Model Inside Tire Axle 4 Multibody Model Outside Tire Axle 5 Multibody Model Inside Tire Axle 5

121 Figure 119. Lateral friction margins from transient bicycle and multibody models for tractor semi-trailer (G  9%, e  8%) (ax  3 ft/s2 and lane change). As Figure 119 shows, the transient bicycle model for a tractor semi-trailer does not capture an apparently significant effect that leads to trailer-tire saturation between 30 to 65 mph design speeds. Part of this phenomenon can be explained by looking at the simulation inputs, specifically braking, shown in Fig- ure 120. These inputs correspond to the same 55 mph truck simulation shown in Figure 118. The braking deceleration is oscillatory, because the simulation attempts to achieve curve-entry deceleration by mimicking a human driver applying and releasing the brakes in an unsteady manner. For this same situation, the brake’s master cylinder “control” pressure is shown in Figure 121. This master cylinder pres- sure corresponds directly with foot pressure on the brake pedal by the driver model, and it exhibits the oscillatory behavior seen in the deceleration plot of Figure 120. Each axle’s individual brakes lag behind the master cylinder pres- sure due to their own dynamic properties, which are related in part to the distance of the axle from the master cylinder. To illustrate this, Figure 121 also shows the brake pressure in axle 5 (i.e., the last axle on the trailer of the tractor semi- trailer). The peaks in this axle’s pressure oscillations lag behind the master cylinder pressure significantly, which is the likely cause of the excessive spiking specifically seen in the trailer margins as predicted by the multibody model. As a result of the oscillations resulting from simulated driver control of deceleration in the multibody model, Figure 120 shows the deceleration value predicted by the multibody model momentarily spike higher than -3 ft/s2, which con- tributes to the low margins. Additionally, as mentioned in the previous section, while weight-transfer effects on lateral friction margins are small for most driving, the tractor semi- trailer has a high center of gravity, which amplifies weight- transfer effects. The high CG of the tractor semi-trailer, combined with the relative severity of the lane-change maneuver, indicates the possibility of weight transfer playing a role in the negative margins. To examine this possibility, the individual lateral

122 0 5 10 15 20 25 -5 0 5 D ec el er at io n (ft/ s2 ) Transient Bicycle Model Multibody Model 0 5 10 15 20 25 20 40 60 Sp ee d (m ph ) Transient Bicycle Model Multibody Model 0 5 10 15 20 25 -0.02 0 0.02 0.04 Time (s) St ee rin g In pu t (r ad ) Transient Bicycle Model Multibody Model Outside Tire Multibody Model Inside Tire Figure 120. Inputs for multibody model for tractor semi-trailer (V  55 mph, G  9%, e  0%) (ax  3 ft/s2 and lane change). Figure 121. Braking pressures for multibody model for tractor semi-trailer (V  55, G  9%, e  0%) (ax  3 ft/s2 and lane change). 0 5 10 15 20 25 0 10 20 30 40 50 60 70 Time (s) Br ak e Pr es su re (p si) Control (Master Cylinder) Axle 5 Outside Wheel Axle 5 Inside Wheel

123 friction margins are shown for the inside and outside tires in Figure 122 for the 0% superelevation case and in Figure 123 for the 8% superelevation case. As anticipated, the figures indicate that the weight transfer effects on lateral friction margins are higher for this maneuver than for a steady-curve traversal. Both axles 4 and 5 have positive friction margins on the inside tire, but negative friction margins are estimated for the outside tire. This further supports the conclusion from the previous section that the transient bicycle model for a tractor semi-trailer should be used with extreme caution when predicting behavior for aggressive maneuvers. Although the multibody model predicts negative lateral friction margins for a tractor semi-trailer for the case of a lane-change maneuver combined with curve-entry decel- eration, the question of severity remains (i.e., how severe is the tire saturation and/or how far from its intended path does the tractor semi-trailer skid?). These questions were answered in Section 4.8 using the transient bicycle model by examining time-of-skidding for various speeds. Fig- ure 118 shows the time trajectory of a tractor semi-trailer for a 55 mph traversal with a lane change and curve-entry deceleration. And although the lateral friction margins are Figure 122. Effect of lateral weight shift for tractor semi-trailer (G  9%, e  0%) (ax  3 ft/s2 and lane change).

124 zero or negative, the vehicle only skids for a short period of time (i.e., less than 1 s). Simulations were next conducted to determine whether a short period of skidding affects the ability of the vehicle to continue to navigate the curve. Figure 124 shows the lateral position of the vehicle on the road for the 55 mph traversal, and the desired and actual positions agree quite closely. This suggests that, while the margins on certain individual tires become negative for a short period of time while travers- ing the curve, the vehicle is still able to negotiate the curve and perform the lane change as desired. Recall that the road surface in the multibody software has a higher coefficient of friction than the assumed friction supply for the margin calculation—this is to help ensure that the friction demand reported by the multibody software does not reflect prema- ture tire saturation to maintain consistency with the transient bicycle model simulations. While this means that vehicles simulated in the multibody software have a slightly better ability to maintain tracking with “negative margins,” some of the more aggressive maneuvers simulated still produced tire saturation and activation of ABS. These cases are discussed in the subsections that follow. Figure 123. Effect of lateral weight shift for tractor semi-trailer (G  9%, e  8%) (ax  3 ft/s2 and lane change).

125 For the sake of completeness in the analysis of trucks, this same maneuver on the same geometry of Figure 124 was simulated with the standard STAA Double twin-trailer truck (i.e., tractor semi-trailer/full-trailer). The standard model of the STAA Double in TruckSim comprises a two-axle lead unit (tractor), a loaded semi-trailer with one rear axle, and a sec- ond full-trailer with two total axles (one in front, one in rear). This vehicle configuration has five axles, like the tractor semi- trailer considered in this research, but the axles are in differ- ent locations and perform different functions. Therefore, the equations of motion derived for the transient bicycle model for the tractor semi-trailer do not apply. Figure 125 shows the margins on each axle of the STAA Double for the curve-entry deceleration with lane-change maneuver. The plots show that the STAA Double seems to have negative margins at speeds less than 55 mph for this configuration. To investigate the severity of these low margin predictions for low design speeds, a plot of the margin trajectories at 25 mph is shown in Figure 126. This 25 mph situation represents the worst-case speed for the STAA Double, and the time traces show that margins on any one axle are negative for a short time duration. Additionally, the outside tire on each axle, the one car- rying more load and generating more cornering force on each axle, maintains positive lateral friction margins throughout the maneuver. The inside tire, which is lighter, exhibits temporary negative margin spikes. Though the inputs are not shown, the lateral friction margin spikes are due to brake activation. Like the tractor semi-trailer, the pneumatic braking system dynam- ics combined with the multibody model’s simulated driver model are responsible for the negative margins here. Comparing the tractor semi-trailer in Figure 119 with the STAA Double in Figure 126, the trends exhibited by the STAA Double are similar to those exhibited by the tractor semi- trailer. The STAA Double, in fact, shows margins that are, in general, either comparable or higher than those exhibited by the tractor semi-trailer. In both situations, the rearmost axles on each vehicle exhibit the lowest margins, but the tractor semi-trailer exhibits lower margins on this axle. With this in mind, the results from the tractor semi-trailer simulations will be considered the worst-case articulated vehicle for the primary analysis of roadway design. 4.10.2.4 Effect of Stopping Sight Distance Deceleration In Section 4.9 stopping sight distance deceleration (ax = -11.2 ft/s2) scenarios were considered for all of the vehicles. Only the single-unit truck produced negative lateral friction margins for curve-keeping steering inputs alone. However, care must be taken when considering the tractor semi-trailer dynamics for this type of traversal as well. Previously, the transient bicycle model for the tractor semi-trailer was shown to be overly optimistic in predicting lateral friction margins when any but gentle steering and braking inputs were consid- ered. Therefore, the multibody models for a single-unit truck and a tractor semi-trailer were used to analyze the stopping Figure 124. Lateral position of tractor semi-trailer with respect to road centerline (V  55 mph, G  9%, e  0%) (ax  3 ft/s2 and lane change). 0 5 10 15 20 0 2 4 6 8 10 12 Time (s) La te ra l D ev ia tio n (ft) e = 0%, G = -9%, v = 55 mph, TractorTrailer Desired Deviation Actual Deviation

126 e = 0%, G = -9%, STAADouble Figure 125. Lateral friction margins from multibody model for tractor semi-trailer/full-trailer truck (Double) (G  9%, e  0%) (ax  3 ft/s2 and lane change). sight distance deceleration case without the lane-change maneuver. Figure 127 compares the results of the transient bicycle and multibody models for the single-unit truck tra- versing a curve while undergoing stopping sight distance deceleration while maintaining the same lane. The transient bicycle and multibody models agree well, with the multibody model offering slightly less-negative margin predictions. For the tractor semi-trailer, Figure 128 shows that at 55 mph both the trailer and rear tractor axles exhibit negative lateral friction margins on all tires for the stopping sight distance deceleration case. This is, again, a relatively small spike and the lateral friction margins are negative for a relatively short period of time, but these two axle groups exhibit skidding for this maneuver across all speeds (Figure 129). This provides further support for the previous claim that the transient bicycle model for a tractor semi-trailer should not be relied upon to estimate lateral friction margins when combined cornering and brak- ing inputs are involved. Again, the low-frequency oscillatory behavior of the simulated driver’s braking inputs also contrib- utes to lower margins for the multi body model, as does the high-frequency oscillation caused by activation of the multi- body simulation’s ABS system. This effect has to be expected for real driving scenarios as well, since human drivers will also exhibit a range in variation in the applied brake pressure. To emphasize the significance of the braking system in determining lateral friction margins, and the significance of the driver model in determining whether a tire will skid, consider the following curve-keeping, stopping sight dis- tance deceleration traversal of an E-class sedan at 65 mph. This vehicle, which consistently produced the highest lateral friction margins in Sections 4.8 and 4.9, exhibits negative margins for several seconds during stopping sight distance

Figure 126. Margin trajectories for tractor semi-trailer/ full-trailer truck (Double) (V  25 mph, G  9%, e  0%) (ax  3 ft/s2 and lane change). 0 5 10 15 20 25 30 0 0.5 1 e = 0%, G = -9%, v = 25 mph, STAADouble Fy M ar gi n Outside Tire Inside Tire 0 5 10 15 20 25 30 -1 0 1 Fy M ar gi n Outside Tire Axle 2 Inside Tire Axle 2 0 5 10 15 20 25 30 0 0.5 1 Time (s) Fy M ar gi n Outside Tire Axle 3 Inside Tire Axle 3 0 5 10 15 20 25 30 -1 0 1 Time (s) Fy M ar gi n Outside Tire Axle 4 Inside Tire Axle 4 0 5 10 15 20 25 30 -1 0 1 Time (s) Fy M ar gi n Outside Tire Axle 5 Inside Tire Axle 5 Figure 127. Lateral friction margins from transient bicycle and multibody models for single-unit truck (G  6%, e  4%) (ax  11.2 ft/s2). 30 40 50 60 70 80 -0.1 0 0.1 0.2 0.3 0.4 Fr on t F y M ar gi n Transient Bicycle Model Multibody Model 30 40 50 60 70 80 -0.1 0 0.1 0.2 0.3 0.4 R ea r F y M ar gi n Speed (mph) e = 4%, G = -6%, SingleUnitTruck (not shown) Transient Bicycle Model Axle 2 (from -0.14 to -0.35) (not shown) Multibody Model Axle 2 (from -0.08 to -0.17)

128 0 2 4 6 8 10 12 14 16 18 20 -20 -10 0 10 D ec el er at io n (ft/ s2 ) Transient Bicycle Model Multibody Model 0 2 4 6 8 10 12 14 16 18 20 0 20 40 60 Sp ee d (m ph ) Transient Bicycle Model Multibody Model 0 2 4 6 8 10 12 14 16 18 20 -0.01 0 0.01 0.02 0.03 Time (s) St ee rin g In pu t (r ad ) Transient Bicycle Model Multibody Model Outside Tire Multibody Model Inside Tire 0 2 4 6 8 10 12 14 16 18 20 0.2 0.4 0.6 0.8 Fr on t T ire M ar gi n Transient Bicycle Model Multibody Model Outside Tire Multibody Model Inside Tire 0 2 4 6 8 10 12 14 16 18 20 -0.2 0 0.2 0.4 0.6 R ea r T ire M ar gi n Transient Bicycle Model Multibody Outside Tire Axle 2 Multibody Inside Tire Axle 2 Multibody Outside Tire Axle 3 Multibody Inside Tire Axle 3 0 2 4 6 8 10 12 14 16 18 20 -0.2 0 0.2 0.4 0.6 Time (s) Tr ai le r T ire M ar gi n Transient Bicycle Model Multibody Outside Tire Axle 4 Multibody Inside Tire Axle 4 Multibody Outside Tire Axle 5 Multibody Inside Tire Axle 5 Figure 128. Inputs and outputs from transient bicycle and multibody models for tractor semi-trailer (V  55 mph, G  6%, e  4%) (ax  11.2 ft/s2). deceleration based upon a multibody simulation. The results of this traversal are shown in Figure 130. It is unclear from the plots of the lateral friction margins in Figure 130 why the margins are estimated to be lower for the multibody model than for the transient bicycle model. However, upon close inspection of the simulation inputs (Figure 131) to simulation outputs (Figure 130), the minimum lateral friction margins occur precisely when the most aggressive braking is activated. This braking maximum is due to an oscillation caused when the multibody simulation software’s driver model attempts to maintain not only the desired deceleration of -11.2 ft/s2, but also the vehicle’s position in the center of the lane. The result is that both the brake pressure (and thus deceleration) and the steering input oscillate together. This causes the two peaks in deceleration observed at approximately 7.5 and 11 s during the trajectory. These spikes correspond with the downward spikes in lateral friction margin in Figure 130. This behavior occurs again later with higher deceleration values. In Section 4.9 it was assumed that the required deceleration value was reached immediately and without error, which is not realistic for a human driver (or a computer approximation of a human driver) that is simply trying to match a deceleration profile. Thus, the transient bicycle model is better than the multibody model in that it more readily illustrates the effects of severe braking, but the transient model is deficient in that it neglects variations expected of human driving. Variations are simulated within the multibody model. Thus, the outputs of each model have to be compared judiciously. The transient model provides a simplified predic- tion of worst-case lateral friction margins, and the multibody model modifies these margins due to human variability. At the very least, the consistent behavior of the transient bi cycle model provides a more reasonable and predictable approach to assess lateral friction margins under these cases than a multi body simulation making use of a driver model. The transient bicycle model has no driver dynamics, and so only tests the vehicle dynamics themselves to determine lateral friction margins. The same model ignores the potential for variability on the part of a driver or an ABS system, and these results show that this variability can actually lead to negative friction margins. Although negative lateral friction margins are estimated from multibody models for single-unit trucks and tractor semi-trailers for the stopping sight distance deceleration case while the desired trajectory is to maintain position in the same lane, in both vehicles the ABS system is initiated, and both vehicles are able to main their desired trajectory. The ability for a single-unit truck and tractor semi-trailer as predicted from multibody models to maintain their desired trajectory through a curve for stopping sight distance deceleration while maintaining position in the same lane is made more evident in the following section, which shows that both vehicles are able to maintain their desired trajectory through a curve while undergoing stopping sight distance deceleration with

Figure 129. Lateral friction margins from transient bicycle and multibody models for tractor semi-trailer (G  6%, e  4%) (ax  11.2 ft/s2). Speed (mph) e = 4%, G = -6%, TractorTrailer Tr ai le r F y M ar gi n R ea r F y M ar gi n Fr on t F y M ar gi n Figure 130. Trajectory of lateral friction margins for E-class sedan (V  65 mph, G  6%, e  4%) (ax  11.2 ft/s2). 0 2 4 6 8 10 12 14 16 18 20 0.1 0.2 0.3 0.4 0.5 0.6 Fr on t T ire M ar gi n Transient Bicycle Model Multibody Outside Tire Multibody Inside Tire 0 2 4 6 8 10 12 14 16 18 20 -0.2 0 0.2 0.4 0.6 Time (s) R ea r T ire M ar gi n Transient Bicycle Model Multibody Outside Tire Multibody Inside Tire

130 a lane change (i.e., an even more aggressive maneuver). In both cases, the ABS system enables both vehicles to traverse the curve without experiencing a skidding event. 4.10.2.5 Effect of Lane-Change Maneuver at Stopping Sight Distance Deceleration While several of the vehicle types considered maintained positive lateral friction margins for stopping sight distance deceleration while maintaining position in the same lane, all of the vehicles considered in Section 4.9 using the transient bicycle models exhibited negative lateral friction margins when a lane change was combined with a stopping sight distance deceleration event. To check whether the same is true using the multibody model, consider the lateral friction margin plots for the E-class sedan (Figures 132 to 134), the full-size SUV (Figures 135 to 137), single-unit truck (Figures 138 to 140), and tractor semi-trailer (Figures 141 to 143). These figures show the margins during traversals at three speeds (25, 55, and 85 mph), at a grade of -9% and superelevations of 0% and 8%. Comparing the effects between 0% and 8% superelevations in Figures 132 to 143, the minimum lateral margins are found 0 2 4 6 8 10 12 14 16 18 20 −20 −10 0 10 D ec el er at io n (ft/ s2 ) Transient Bicycle Model Multibody Model 0 2 4 6 8 10 12 14 16 18 20 0 50 100 Sp ee d (m ph ) Transient Bicycle Model Multibody Model 0 2 4 6 8 10 12 14 16 18 20 −0.02 0 0.02 0.04 Time (s) St ee rin g In pu t (r ad ) Transient Bicycle ModelMultibody Outside Tire Multibody Inside Tire Figure 131. Trajectory of simulation inputs for transient bicycle and multibody models for E-class sedan (V  65 mph, G  6%, e  4%) (ax  11.2 ft/s2). Figure 132. Lateral friction margin trajectories from transient bicycle and multibody models for E-class sedan [V  25 mph, G  9%, e  0% ( left plots) and 8% (right plots)] (ax  11.2 ft/s2 and lane change). 0 2 4 6 8 10 12 14 16 18 20 −0.5 0 0.5 1 Fr on t T ire M ar gi n Transient Bicycle Model Multibody Model Outside Tire Multibody Model Inside Tire 0 2 4 6 8 10 12 14 16 18 20 −0.5 0 0.5 1 Time (s) R ea r T ire M ar gi n Transient Bicycle Model Multibody Model Outside Tire Multibody Model Inside Tire 0 2 4 6 8 10 12 14 16 18 20 −0.5 0 0.5 1 Fr on t T ire M ar gi n Transient Bicycle Model Multibody Model Outside Tire Multibody Model Inside Tire 0 2 4 6 8 10 12 14 16 18 20 −0.5 0 0.5 1 Time (s) R ea r T ire M ar gi n Transient Bicycle Model Multibody Model Outside Tire Multibody Model Inside Tire

Figure 133. Lateral friction margin trajectories from transient bicycle and multibody models for E-class sedan [V  55 mph, G  9%, e  0% ( left plots) and 8% (right plots)] (ax  11.2 ft/s2 and lane change). 0 2 4 6 8 10 12 14 16 18 20 −0.5 0 0.5 1 Fr on t T ire M ar gi n 0 2 4 6 8 10 12 14 16 18 20 −0.5 0 0.5 1 Time (s) R ea r T ire M ar gi n Transient Bicycle Model Multibody Model Outside Tire Multibody Model Inside Tire Transient Bicycle Model Multibody Model Outside Tire Multibody Model Inside Tire Transient Bicycle Model Multibody Model Outside Tire Multibody Model Inside Tire Transient Bicycle Model Multibody Model Outside Tire Multibody Model Inside Tire 0 2 4 6 8 10 12 14 16 18 20 −0.5 0 0.5 1 Fr on t T ire M ar gi n 0 2 4 6 8 10 12 14 16 18 20 −0.5 0 0.5 1 Time (s) R ea r T ire M ar gi n Figure 134. Lateral friction margin trajectories from transient bicycle and multibody models for E-class sedan [V  85 mph, G  9%, e  0% ( left plots) and 8% (right plots)] (ax  11.2 ft/s2 and lane change). 0 2 4 6 8 10 12 14 16 18 20 −0.5 0 0.5 1 Fr on t T ire M ar gi n Transient Bicycle Model Multibody Model Outside Tire Multibody Model Inside Tire 0 2 4 6 8 10 12 14 16 18 20 −0.5 0 0.5 1 Time (s) R ea r T ire M ar gi n Transient Bicycle Model Multibody Model Outside Tire Multibody Model Inside Tire 0 2 4 6 8 10 12 14 16 18 20 −0.5 0 0.5 1 Fr on t T ire M ar gi n Transient Bicycle Model Multibody Model Outside Tire Multibody Model Inside Tire 0 2 4 6 8 10 12 14 16 18 20 −0.5 0 0.5 1 Time (s) R ea r T ire M ar gi n Transient Bicycle Model Multibody Model Outside Tire Multibody Model Inside Tire Figure 135. Lateral friction margin trajectories from transient bicycle and multibody models for full-size SUV [V  25 mph, G  9%, e  0% ( left plots) and 8% (right plots)] (ax  11.2 ft/s2 and lane change). 0 2 4 6 8 10 12 14 16 18 20 −0.5 0 0.5 1 Fr on t T ire M ar gi n Transient Bicycle Model Multibody Model Outside Tire Multibody Model Inside Tire 0 2 4 6 8 10 12 14 16 18 20 −0.5 0 0.5 1 Time (s) R ea r T ire M ar gi n Transient Bicycle Model Multibody Model Outside Tire Multibody Model Inside Tire 0 2 4 6 8 10 12 14 16 18 20 −0.5 0 0.5 1 Fr on t T ire M ar gi n Transient Bicycle Model Multibody Model Outside Tire Multibody Model Inside Tire 0 2 4 6 8 10 12 14 16 18 20 −0.5 0 0.5 1 Time (s) R ea r T ire M ar gi n Transient Bicycle Model Multibody Model Outside Tire Multibody Model Inside Tire

0 2 4 6 8 10 12 14 16 18 20 −0.5 0 0.5 1 Fr on t T ire M ar gi n Transient Bicycle Model Multibody Model Outside Tire Multibody Model Inside Tire 0 2 4 6 8 10 12 14 16 18 20 −0.5 0 0.5 1 Time (s) R ea r T ire M ar gi n Transient Bicycle Model Multibody Model Outside Tire Multibody Model Inside Tire 0 2 4 6 8 10 12 14 16 18 20 −0.5 0 0.5 1 Fr on t T ire M ar gi n Transient Bicycle Model Multibody Model Outside Tire Multibody Model Inside Tire 0 2 4 6 8 10 12 14 16 18 20 −0.5 0 0.5 1 Time (s) R ea r T ire M ar gi n Transient Bicycle Model Multibody Model Outside Tire Multibody Model Inside Tire Figure 136. Lateral friction margin trajectories from transient bicycle and multibody models for full-size SUV [V  55 mph, G  9%, e  0% ( left plots) and 8% (right plots)] (ax  11.2 ft/s2 and lane change). 0 2 4 6 8 10 12 14 16 18 20 −0.5 −0.5 0 0.5 1 Fr on t T ire M ar gi n Transient Bicycle Model Multibody Model Outside Tire Multibody Model Inside Tire 0 2 4 6 8 10 12 14 16 18 20 −0.5 −0.5 0 0.5 1 Time (s) R ea r T ire M ar gi n Transient Bicycle Model Multibody Model Outside Tire Multibody Model Inside Tire 0 2 4 6 8 10 12 14 16 18 20 0 0.5 1 Fr on t T ire M ar gi n Transient Bicycle Model Multibody Model Outside Tire Multibody Model Inside Tire 0 2 4 6 8 10 12 14 16 18 20 0 0.5 1 Time (s) R ea r T ire M ar gi n Transient Bicycle Model Multibody Model Outside Tire Multibody Model Inside Tire Figure 137. Lateral friction margin trajectories from transient bicycle and multibody models for full-size SUV [V  85 mph, G  9%, e  0% ( left plots) and 8% (right plots)] (ax  11.2 ft/s2 and lane change). 0 5 10 15 20 25 −0.5 0 0.5 1 Fr on t T ire M ar gi n Transient Bicycle Model Multibody Model Outside Tire Multibody Model Inside Tire 0 5 10 15 20 25 −0.5 −0.5 −0.5 0 0.5 1 Time (s) R ea r T ire M ar gi n Transient Bicycle Model Multibody Model Outside Tire Multibody Model Inside Tire 0 5 10 15 20 25 0 0.5 1 Fr on t T ire M ar gi n Transient Bicycle Model Multibody Model Outside Tire Multibody Model Inside Tire 0 5 10 15 20 25 0 0.5 1 Time (s) R ea r T ire M ar gi n Transient Bicycle Model Multibody Model Outside Tire Multibody Model Inside Tire Figure 138. Lateral friction margin trajectories from transient bicycle and multibody models for single-unit truck [V  25 mph, G  9%, e  0% ( left plots) and 8% (right plots)] (ax  11.2 ft/s2 and lane change).

133 0 5 10 15 20 25 30 −0.5 0 0.5 1 Fr on t T ire M ar gi n Transient Bicycle Model Multibody Model Outside Tire Multibody Model Inside Tire 0 5 10 15 20 25 30 −0.5 0 0.5 1 Time (s) R ea r T ire M ar gi n Transient Bicycle Model Multibody Model Outside Tire Multibody Model Inside Tire 0 5 10 15 20 25 −0.5 0 0.5 1 Fr on t T ire M ar gi n Transient Bicycle Model Multibody Model Outside Tire Multibody Model Inside Tire 0 5 10 15 20 25 −0.5 0 0.5 1 Time (s) R ea r T ire M ar gi n Transient Bicycle Model Multibody Model Outside Tire Multibody Model Inside Tire Figure 139. Lateral friction margin trajectories from transient bicycle and multibody models for single-unit truck [V  55 mph, G  9%, e  0% (left plots) and 8% (right plots)] (ax  11.2 ft/s2 and lane change). Figure 140. Lateral friction margin trajectories from transient bicycle and multibody models for single-unit truck [V  85 mph, G  9%, e  0% (left plots) and 8% (right plots)] (ax  11.2 ft/s2 and lane change). 0 2 4 6 8 10 12 14 16 18 20 -0.5 0 0.5 1 Fr on t T ire M ar gi n Transient Bicycle Model Multibody Model Outside Tire Multibody Model Inside Tire 0 2 4 6 8 10 12 14 16 18 20 -0.5 0 0.5 1 Time (s) R ea r T ire M ar gi n Transient Bicycle Model Multibody Model Outside Tire Multibody Model Inside Tire 0 2 4 6 8 10 12 14 16 18 20 -0.5 0 0.5 1 Fr on t T ire M ar gi n Transient Bicycle Model Multibody Model Outside Tire Multibody Model Inside Tire 0 2 4 6 8 10 12 14 16 18 20 -0.5 0 0.5 1 Time (s) R ea r T ire M ar gi n Transient Bicycle Model Multibody Model Outside Tire Multibody Model Inside Tire to be nearly identical. The 0% superelevation cases of these figures represent the worst-case scenarios as identified in Sec- tion 4.9 for each vehicle. These situations had friction mar- gins close to zero for most speeds with the transient model, and these low margins are confirmed in general by the multi- body simulation results. One effect of larger superelevations appears to be that, where there are intervals of low margins in the maneuver, these intervals seem to last longer in the 8% superelevation case than in the 0% superelevation case. An example of this can be seen in Figure 133. In many of the cases, Figure 133 as an example, one can see the bias in lateral weight caused by the superelevated curve. This is caused by the multi- body model’s ability to account precisely for a vehicle’s weight transfer and suspension dynamics during and after skidding occurs. The main message, however, remains that the effects of reasonable superelevation values on margins are small as long as appropriate guidelines for developing the superelevation in the tangent are followed. Figures 132 to 143 also show that the multibody model matches the transient bicycle model fairly well for most of the traversal, with some caveats. First, the time traces between the transient model and the multibody model do not always align, but this is generally due to mismatched inputs between the two models, not necessarily due to model differences. Also, they exhibit the same sort of brake oscillations present in the pre- ceding section on curve-keeping steering with stopping sight distance deceleration. This causes the simulations to some- times over-predict or under-predict the lateral friction margins

134 0 5 10 15 20 25 0 0.2 0.4 0.6 0.8 Fr on t T ire M ar gi n Transient Bicycle Model Multibody Outside Tire Multibody Inside Tire 0 5 10 15 20 25 -0.5 0 0.5 1 R ea r T ire M ar gi n Transient Bicycle Model Multibody Outside Tire Axle 2 Multibody Inside Tire Axle 2 Multibody Outside Tire Axle 3 Multibody Inside Tire Axle 3 0 5 10 15 20 25 -0.5 0 0.5 1 Time (s) Tr ai le r T ire M ar gi n Transient Bicycle Model Multibody Outside Tire Axle 4 Multibody Inside Tire Axle 4 Multibody Outside Tire Axle 5 Multibody Inside Tire Axle 5 0 5 10 15 20 25 0 0.2 0.4 0.6 0.8 Fr on t T ire M ar gi n Transient Bicycle Model Multibody Outside Tire Multibody Inside Tire 0 5 10 15 20 25 -0.5 0 0.5 1 R ea r T ire M ar gi n Transient Bicycle Model Multibody Outside Tire Axle 2 Multibody Inside Tire Axle 2 Multibody Outside Tire Axle 3 Multibody Inside Tire Axle 3 0 5 10 15 20 25 -0.5 0 0.5 1 Time (s) Tr ai le r T ire M ar gi n Transient Bicycle Model Multibody Outside Tire Axle 4 Multibody Inside Tire Axle 4 Multibody Outside Tire Axle 5 Multibody Inside Tire Axle 5 Figure 141. Lateral friction margin trajectories from transient bicycle and multibody models for tractor semi-trailer [V  25 mph, G  9%, e  0% ( left plots) and 8% (right plots)] (ax  11.2 ft/s2 and lane change). Figure 142. Lateral friction margin trajectories from transient bicycle and multibody models for tractor semi-trailer [V  55 mph, G  9%, e  0% ( left plots) and 8% (right plots)] (ax  11.2 ft/s2 and lane change). 0 5 10 15 20 25 0 0.2 0.4 0.6 0.8 Fr on t T ire M ar gi n Transient Bicycle Model Multibody Outside Tire Multibody Inside Tire 0 5 10 15 20 25 -0.5 0 0.5 1 R ea r T ire M ar gi n Transient Bicycle Model Multibody Outside Tire Axle 2 Multibody Inside Tire Axle 2 Multibody Outside Tire Axle 3 Multibody Inside Tire Axle 3 0 5 10 15 20 25 -0.5 0 0.5 1 Time (s) Tr ai le r T ire M ar gi n Transient Bicycle Model Multibody Outside Tire Axle 4 Multibody Inside Tire Axle 4 Multibody Outside Tire Axle 5 Multibody Inside Tire Axle 5 0 5 10 15 20 25 0 0.2 0.4 0.6 0.8 Fr on t T ire M ar gi n Transient Bicycle Model Multibody Model Outside Tire Multibody Model Inside Tire 0 5 10 15 20 25 -0.5 0 0.5 1 R ea r T ire M ar gi n Transient Bicycle Model Multibody Model Outside Tire Axle 2 Multibody Model Inside Tire Axle 2 Multibody Model Outside Tire Axle 3 Multibody Model Inside Tire Axle 3 0 5 10 15 20 25 -0.2 0 0.2 0.4 0.6 Time (s) Tr ai le r T ire M ar gi n Transient Bicycle Model Multibody Model Outside Tire Axle 4 Multibody Model Inside Tire Axle 4 Multibody Model Outside Tire Axle 5 Multibody Model Inside Tire Axle 5

135 compared to the transient bicycle model, primarily because of the driver behavior used by the multibody simulation software. Most significantly, the multibody model shows that the ABS may be activated for some of these maneuvers even when the simulations are conducted on high-friction roads. The effects of the input variations are illustrated in particular within Fig- ures 138 to 143. The single-unit truck and the tractor semi- trailer models used by the multibody simulation software are both equipped with ABS, which activates during the travers- als to help the vehicles avoid losing control. For these trucks, the ABS tends to activate in the multibody model in situations where the transient bicycle model predicted severely negative margins; so while ABS prevented the grossly negative margins, the multibody model does confirm that these situations would likely have had very negative margins if ABS were not available. In summary even though the simulations using the tran- sient bicycle model are capable of showing very negative lat- eral friction margins, and even though the road surface used in the multibody model was simulated with a high coefficient of friction to ensure that the demand predicted would be suf- ficient for all cases simulated, the ABS still activates during the stopping sight distance deceleration traversal with lane change for the two trucks considered. This prevented the lateral friction margins from going too far below zero and assisted the vehicle in staying under control. However, it also indicates quite clearly that lane changes combined with stop- ping sight distance decelerations will cause ABS activation even on high-friction roads. 4.10.2.6 Effect of Lane-Change Maneuver at Emergency Braking Deceleration Given the variability of the multibody simulation results caused by the driver model, and given that nearly all scenarios predict negative margins, only a few scenarios are evaluated using emergency braking deceleration rates. In particular, Section 4.9 predicted that the E-class SUV, full-size SUV, and single-unit truck would skid considerable distances when a combined lane-change/emergency braking deceleration maneuver occurred. In previous sections, the lateral deviation distance was predicted assuming, among other things, that the driver made no corrective steering inputs and that ABS is not present—this situation is similar to a worst-case vehicle situation. However, in the multibody model, the vehicles are simulated with ABS, and there is a driver model present that steers the vehicle toward the desired path even in the presence of deviations. This is analogous to a best-case scenario. Thus, comparison of the multibody results with the results of Sec- tion 4.9 illustrates the range of tracking behavior that might occur in the presence of negative lateral friction margins. Figure 144 shows the path tracking performance of four vehicles under emergency braking deceleration with a lane Figure 143. Lateral friction margin trajectories from transient bicycle and multibody models for tractor semi-trailer [V  85 mph, G  9%, e  0% ( left plots) and 8% (right plots)] (ax  11.2 ft/s2 and lane change). 0 2 4 6 8 10 12 14 16 18 20 0.2 0.3 0.4 0.5 Fr on t T ire M ar gi n Transient Bicycle Model Multibody Model Outside Tire Multibody Model Inside Tire 0 2 4 6 8 10 12 14 16 18 20 -0.2 0 0.2 0.4 0.6 R ea r T ire M ar gi n Transient Bicycle Model Multibody Outside Tire Axle 2 Multibody Inside Tire Axle 2 Multibody Outside Tire Axle 3 Multibody Inside Tire Axle 3 0 2 4 6 8 10 12 14 16 18 20 -0.2 0 0.2 0.4 0.6 Time (s) Tr ai le r T ire M ar gi n Transient Bicycle Model Multibody Outside Tire Axle 4 Multibody Inside Tire Axle 4 Multibody Outside Tire Axle 5 Multibody Inside Tire Axle 5 0 2 4 6 8 10 12 14 16 18 20 0.2 0.3 0.4 0.5 Fr on t T ire M ar gi n Transient Bicycle Model Multibody Model Outside Tire Multibody Model Inside Tire 0 2 4 6 8 10 12 14 16 18 20 -0.2 0 0.2 0.4 0.6 R ea r T ire M ar gi n Transient Bicycle Model Multibody Outside Tire Axle 2 Multibody Inside Tire Axle 2 Multibody Outside Tire Axle 3 Multibody Inside Tire Axle 3 0 2 4 6 8 10 12 14 16 18 20 -0.2 0 0.2 0.4 0.6 Time (s) Tr ai le r T ire M ar gi n Transient Bicycle Model Multibody Outside Tire Axle 4 Multibody Inside Tire Axle 4 Multibody Outside Tire Axle 5 Multibody Inside Tire Axle 5

136 change at 70 mph, predicted for the multibody simulation in situations where there are negative lateral friction margins. Note that the ABS was activating during the simulation, even for high-friction roads, which again indicates that the fric- tion margins are extremely low. All four vehicles were able to maintain control through the curve through the use of ABS and the multibody software’s driver model. Figure 145 shows the time trajectories of the lateral friction margin for the same vehicles for the same traversal as shown in Figure 144. The margin trajectories between the transient bicy- cle model and the multibody model match in shape, although in the case of the full-size SUV and the single-unit truck—the two worst vehicles for lateral deviation distance during this type of traversal (see Section 4.9)—the effects of ABS are evident in the oscillations of tire force when the margins approach zero. The fast, pulsing action of the ABS on the simulated vehicles allows the multibody software’s driver model to continue to navigate the prescribed maneuver without any significant skidding and subsequent lateral deviation. The fact that none of these vehi- cles departed significantly from the intended trajectory while traversing the curve suggests that the lane change on the curve is not so severe that an ABS-equipped vehicle will lose control. Recall that the roads used in the multibody simulation possessed high coefficients of friction (corresponding to fric- tion supply) so as not to distort the computation of friction demand as defined in Section 4.2. This was done to make the lateral friction margins comparable for as many driving sce- narios as possible between Section 4.8 and this current analy- sis. But even high-friction roads did not stop the ABS from activating due to excessive wheel slip for large braking values 0 5 10 15 20 0 2 4 6 8 10 12 Time (s) La te ra l D ev ia tio n (ft) e = 8%, G = -9%, V = 70 mph, EclassSedan Desired Deviation Actual Deviation 0 5 10 15 20 0 2 4 6 8 10 12 Time (s) La te ra l D ev ia tio n (ft) e = 8%, G = -9%, V = 70 mph, EclassSUV Desired Deviation Actual Deviation 0 5 10 15 20 0 2 4 6 8 10 12 Time (s) La te ra l D ev ia tio n (ft) e = 8%, G = -9%, V = 70 mph, FullSUV Desired Deviation Actual Deviation 0 5 10 15 20 25 0 2 4 6 8 10 12 Time (s) La te ra l D ev ia tio n (ft) e = 8%, G = -9%, V = 70 mph, Single Unit Truck Desired Deviation Actual Deviation Figure 144. Lateral deviation distance from multibody models for E-class sedan, E-class SUV, full-size SUV, and single-unit truck (V  70 mph, G  9%, e  8%) (ax  15 ft/s2 and lane change).

137 or for the most aggressive maneuvers. Additionally, using a high-friction road to generate friction demand values could inadvertently allow the vehicles in the multibody simulation to maintain path tracking artificially well. To analyze whether the high-friction simulations are artifi- cially improving skid performance, another series of simula- tions was performed. This set of simulations used conditions identical to those run in Figures 144 and 145, but the road surface was given a low coefficient of friction of 0.50 for trucks and 0.55 for passenger vehicles, which represent the actual two-sigma low friction supply values in the lateral (cornering) direction for 70 mph as calculated in Section 4.2. The results of the low-friction simulations, shown in Fig- ure 146, are nearly identical to the high-friction road simula- tions shown in Figure 144. This suggests that even with low friction in the multibody model, the simulated driver is able to maintain control of the vehicle during the lane change. This result is clearly not a sweeping generalization about all human drivers and all low-friction environments. Drivers often panic when tires lose traction, even with ABS and/or stability con- trol assisting them, and often do not make the correct control decisions necessary to keep a vehicle on a desired trajectory at the onset of a skid. However, the fact that the simulated driver is able to maintain lane following and a lane-change maneuver while the ABS system activates on a low-friction road suggests that the lane changes are physically possible on curves, even under emergency braking circumstances. In the simulations presented so far, friction demand is sub- tracted from friction supply to obtain lateral friction mar- gin; thus, in cases of negative margin, it is possible that the vehicle is not actually decelerating at the assumed rate. In the multibody simulation, this can cause the vehicle to achieve a 0 5 10 15 20 25 -0.5 0 0.5 1 e = 8%, G = -9%, v = 70 mph, EclassSedan Fr on t T ire M ar gi n Transient Bicycle Model Multibody Model Outside Tire Multibody Model Inside Tire 0 5 10 15 20 25 -0.5 0 0.5 1 Time (s) R ea r T ire M ar gi n Transient Bicycle Model Multibody Model Outside Tire Multibody Model Inside Tire 0 5 10 15 20 25 -0.5 0 0.5 1 e = 8%, G = -9%, v = 70 mph, EclassSUV Fr on t T ire M ar gi n Transient Bicycle Model Multibody Model Outside Tire Multibody Model Inside Tire 0 5 10 15 20 25 -0.5 0 0.5 1 Time (s) R ea r T ire M ar gi n Transient Bicycle Model Multibody Model Outside Tire Multibody Model Inside Tire 0 5 10 15 20 25 -0.5 0 0.5 1 e = 8%, G = -9%, v = 70 mph, FullSUV Fr on t T ire M ar gi n Transient Bicycle Model Multibody Model Outside Tire Multibody Model Inside Tire 0 5 10 15 20 25 -0.5 0 0.5 1 Time (s) R ea r T ire M ar gi n Transient Bicycle Model Multibody Model Outside Tire Multibody Model Inside Tire 0 5 10 15 20 25 30 -0.5 0 0.5 1 e = 8%, G = -9%, v = 70 mph, SingleUnitTruck Fr on t T ire M ar gi n Transient Bicycle Model Multibody Model Outside Tire Multibody Model Inside Tire 0 5 10 15 20 25 30 -0.5 0 0.5 1 Time (s) R ea r T ire M ar gi n Transient Bicycle Model Multibody Model Outside Tire Multibody Model Inside Tire Figure 145. Lateral friction margin trajectories from transient bicycle and multibody models for E-class sedan, E-class SUV, full-size SUV, and single-unit truck (V  70 mph, G  9%, e  8%) (ax  15 ft/s2 and lane change).

138 lane change and/or lane follow, but not actually decelerate at the intended amount. Further investigations confirmed that the deceleration rates (as measured at the vehicle’s sprung- mass center of gravity) for all vehicles were comparable to the desired deceleration rate of -15 ft/s2. 4.10.3 Discussion This section addresses a couple of key issues associated with the use of complex vehicle dynamics models for estimating lateral friction margins. The first issue deals with the accu- racy of the vehicle simulations themselves. For two-axle vehi- cles, the transient bicycle model is a fairly good predictor of vehicle behavior with a worst-case driver, e.g., one who does not attempt to correct for any lane-keeping errors. In circum- stances where the ABS inputs and driver corrections are both negligible, the results show that the transient bicycle model for two-axle vehicles produces margins that match quite well with the multibody models. For multiaxle vehicles, it was shown that if adjacent tractor or semi-trailer axles are averaged, then the multibody model largely agrees with the transient bicycle model. However, it was found that this averaging may hide some axle-specific variation in lateral friction margin that could result in low or even negative friction margins. The second insight is that, as maneuvers become more aggressive, the agreement between different models becomes less exact, and more dependent on the driver model and on the presence or absence of ABS. Thus the outputs become harder to compare. It is expected that the multibody simu- lations’ simulated driver behavior is more representative of 0 5 10 15 20 0 2 4 6 8 10 12 Time (s) La te ra l D ev ia tio n (ft) e = 8%, G = -9%, V = 70 mph, E-class Sedan Desired Deviation Actual Deviation 0 2 4 6 8 10 12 0 2 4 6 8 10 12 Time (s) La te ra l D ev ia tio n (ft) e = 8%, G = -9%, V = 70 mph, E-class SUV Desired Deviation Actual Deviation 0 2 4 6 8 10 12 0 2 4 6 8 10 12 Time (s) La te ra l D ev ia tio n (ft) e = 8%, G = -9%, V = 70 mph, Full-Sized SUV Desired Deviation Actual Deviation 0 2 4 6 8 10 12 14 16 18 0 2 4 6 8 10 12 Time (s) La te ra l D ev ia tio n (ft) e = 8%, G = -9%, V = 70 mph, Single Unit Truck Desired Deviation Actual Deviation Figure 146. Lateral deviation distance for low-friction simulation from multibody models for E-class sedan, E-class SUV, full-size SUV, and single-unit truck (V  70 mph, G  9%, e  8%) (ax  15 ft/s2 and lane change).

139 what a human driver will do during a traversal of a curve, but it is also acknowledged that these inputs may no longer be the worst case because a worst-case driver response cannot be defined. When the deceleration values exceed curve-entry deceleration values, oscillations in the braking input caused by the simulation trying to maintain the desired deceleration become significant and can lead to negative margin predic- tions. On the other hand, the ABS activates in the multibody software to prevent margins from becoming too negative and allows the vehicle simulated by the multibody software to maintain the desired path, when driven by the software’s optimal preview steering controller (i.e., driver model). 4.10.4 Summary of Key Results from Step 9 In summary, the following findings were obtained from the analysis in Step 9: 1. The multibody model confirms the results of the transient bicycle models for two-axle vehicles for all but the most aggressive maneuvers, which makes them suitable for roadway design. 2. The transient bicycle models for tractor semi-trailers tend to predict vehicle behavior well for moderate maneuvers but should not be used to predict lateral friction margins for combined cornering, lane-change, and braking maneu- vers. Disagreement between the transient bicycle model and the multibody model was found for these cases. The most disagreement was a function of either averaging of axle forces, the presence of ABS, or the specific responses of the simulated driver’s braking inputs. 3. Lateral friction margins for the STAA Double are, for the cases that have been studied, slightly higher than those for the tractor semi-trailer. 4. The simulated driver’s braking and steering control used by the multibody simulation software, which attempts to approximate human behavior, can cause any of the vehi- cles considered to temporarily skid. This was seen during a curve-keeping event with or without a lane change when decelerating at stopping sight distance levels on grades. 5. The single-unit truck and the full-size SUV both have positive lateral friction margins during combined curve- entry deceleration and lane changes. But both have nega- tive lateral friction margins during stopping sight distance and emergency deceleration with or without lane-change events for all speeds. 6. Due to the driver model within CarSim and TruckSim, as well as the use of ABS in the multibody simulations, all vehi- cles were able to maintain the desired trajectory around the curve and through a lane change for all braking scenarios considered. This is in contrast to the results of Section 4.9, which did not consider ABS or driver corrections. 7. The path tracking performance of the simulated vehicle in the multibody model was not degraded when a low- friction road was simulated. 4.11 Step 10: Predict Wheel Lift of Individual Wheels during Transient Maneuvers The objective of Step 10 was to use high-order multibody models to predict wheel lift of individual wheels. Using com- mercially available vehicle dynamic simulation software (i.e., CarSim and TruckSim), high-order multibody models were used to predict wheel lift of individual wheels as a vehicle traverses a sharp horizontal curve, taking into consideration a range of conditions such as the horizontal curvature, grade, and superelevation. Rather than simulating the full range of hypothetical geometries and vehicles considered throughout this research, this analysis focused on those situations identi- fied in previous steps as areas of concern. Note, unlike the analyses in Sections 4.9 and earlier, in the multibody mod- els the superelevation transition is simulated (i.e., designed) according to the Green Book policy. 4.11.1 Analysis Approach To extend the results of Section 4.6, a more sophisticated method of identifying rollover margins was considered in this step using multibody models. The analysis consid- ers the cases where the vehicle traverses a curve without a lane change, with a lane change, and with deceleration. The analysis focuses on vehicles identified in Section 4.6 as having low rollover thresholds as defined by Equation 27. This con- sists especially of the single-unit truck and the tractor semi- trailer truck. Note that tractor semi-trailer/full-trailer trucks are assumed to have the same rollover thresholds as tractor semi-trailer trucks. This assumption is made because in the static case, both vehicles are limited by the rollover threshold of their trailers. The CG height and track width of the trail- ers for both vehicles are assumed to be quite similar. Thus, the tractor semi-trailer results are assumed to represent the tractor semi-trailer/full-trailer as well. This assumption of similarity between the rollover margins for the tractor semi- trailer and the tractor semi-trailer/full-trailer is confirmed in the analysis that follows. In Section 4.6 the rollover threshold, and hence the roll- over margin, was defined according to the static configu- ration of the vehicle, and thus may not be appropriate to predict dynamic wheel-lift events. Thus, a new definition of proximity to wheel lift is needed. In this analysis, the load transfer ratio (LTR), a metric commonly used in the vehicle dynamics community to predict wheel lift and examine the

140 relative severity of a maneuver with respect to wheel lift, is utilized. The metric is defined for each axle as: = − + (89)LTR N N N N i o i o Where Ni and No are the normal (vertical) loads on an axle’s inside and outside tires while cornering, respectively. Because the normal loads on a tire cannot be less than zero, this metric can only vary from -1 to +1. As defined, the LTR for an axle is -1 when the outside tire bears all of the axle’s load (i.e., the inside wheel lifts) and +1 when the inside tire bears all of the axle’s load. For a symmetrically loaded vehicle on a tangent roadway with 0% superelevation, the LTR for any axle would be 0. As a vehicle traverses a horizontal curve, the LTR will tend toward -1 immediately prior to wheel lift. From a qualitative perspective, the LTR can be thought of as the portion of the total axle load carried by the outside tire. This constitutes a sort of “roll demand” that, when subtracted from unity, gives a per-axle dynamic rollover margin defined by the proximity of the LTR to a value that causes wheel lift, e.g., = − − + 1 (90)RM N N N N LTR i o i o This “dynamic” rollover margin, in contrast to the “static” rollover margin of Section 4.6 defined in Equation 27, repre- sents the proximity of the axle to wheel lift. 4.11.2 Analysis Results In Figure 147, the rollover margins are plotted for a single- unit truck and a tractor semi-trailer traversing a curve with- out making a lane change and keeping a constant speed, for a curve with a grade of -6% and 4% superelevation. These plots show the differences between the rollover margins of individual axles for speeds from 25 to 85 mph. Because the current AASHTO policy provides for higher levels of lateral acceleration at low design speeds, there is more weight shift at lower speeds than with higher speeds. This weight shift causes lower rollover margins for this steady traversal at lower speeds. The rollover margins increase with increasing design speed. For these test cases, the minimum rollover margins are approximately 0.4 to 0.48 for a speed of 25 mph. To test the boundaries of the rollover margin envelope in the scope of the maneuvers considered in this research, the more aggressive maneuvers are of greatest interest. Figure 148 shows the rollover margins for a single-unit truck and a trac- tor semi-trailer considering curve-entry deceleration com- bined with a lane change for a -9% grade and superelevations of 0% and 8%. The plots indicate that both vehicles have a large amount of load remaining on the inside tire. In Fig- ure 148, the -9% grade with 0% superelevation represents the worst case identified in Section 4.8 for lateral friction margins for both of these vehicles. It was assumed that this worst-case friction situation will likely generate the worst- case rollover margin, since in general lateral friction margins and rollover margins are both worst when lateral accelera- tions on a vehicle are highest. In Figure 148 the worst-case margins occur at low speed for the tractor semi-trailer. The 8% superelevation case was simulated for the tractor semi-trailer/full-trailer at 25 mph to compare to the tractor semi-trailer. The results are shown in Figure 149, and the minimum rollover margins are nearly identical to the tractor-trailer (0.32 versus 0.36). 30 40 50 60 70 80 -0.1 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 R ol lo ve r M ar gi n Speed (mph) e = 4%, G = -6%, SingleUnitTruck Axle 1 Axle 2 30 40 50 60 70 80 -0.1 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 R ol lo ve r M ar gi n Speed (mph) e = 4%, G = -6%, TractorTrailer Axle 1 Axle 2 Axle 3 Axle 4 Axle 5 Figure 147. Rollover margins for individual axles of single-unit truck and tractor semi-trailer (G  6%, e  4%) (ax  0 ft/s2).

141 Figure 148. Rollover margins of individual axles for single-unit truck and tractor semi-trailer (G  9%, e  0% and 8%) (ax  3 ft/s2 and lane change). 30 40 50 60 70 80 -0.1 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 R ol lo ve r M ar gi n Speed (mph) e = 0%, G = -9%, SingleUnitTruck Axle 1 Axle 2 30 40 50 60 70 80 -0.1 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 R ol lo ve r M ar gi n Speed (mph) e = 0%, G = -9%, TractorTrailer Axle 1 Axle 2 Axle 3 Axle 4 Axle 5 30 40 50 60 70 80 -0.1 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 R ol lo ve r M ar gi n Speed (mph) e = 8%, G = -9%, Single Unit Truck Axle 1 Axle 2 30 40 50 60 70 80 -0.1 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 R ol lo ve r M ar gi n Speed (mph) e = 8%, G = -9%, TractorTrailer Axle 1 Axle 2 Axle 3 Axle 4 Axle 5 Considering an even more severe scenario, Figure 150 shows the rollover margins for a single-unit truck and a tractor semi- trailer considering stopping sight distance deceleration com- bined with a lane change for a -9% grade and superelevations of 0% and 8%. The results indicate that rollover margins for both vehicles are at least 0.4 or greater for this maneuver and roadway design. As it happens, the rollover margins in this case are actually higher than they are for the mild deceleration case, primarily due to the effect of the friction ellipse. At the higher deceleration level, more of the tire force is partitioned to braking and thus less is available for lateral acceleration. The previous sections showed that this situation begins to produce very low lateral friction margins, and thus the roll- over margins are actually higher than expected. When emergency braking deceleration is considered, the results of the rollover margin analysis for the single-unit truck are quite different than previous results. These rollover mar- gins are shown in Figure 151. While the tractor semi-trailer still does not come close to lifting a wheel, the same is not true for the single-unit truck. The single-unit truck appears to lift a wheel from 50 to 65 mph under emergency braking deceleration with a lane change. To understand the wheel-lift situations for the single-unit truck, the specific rollover margin trajectories are plotted for all wheel-lift speeds (i.e., 50 to 65 mph) in Figure 152. It is clear that the rollover margins are generally very positive through almost the entire trajectory, and only spike down to zero for an instant. This spike is most likely due to resonance in the suspension and/or braking systems associated with actions of the ABS controller on the simulated vehicle. The rollover mar- gins near zero represent only a very momentary “wheel lift” on the inside tire, and do not present a condition where the

142 vehicle will likely rollover. This is evidenced by the fact that the multibody simulation itself did not predict a rollover event, which the software is fully capable of simulating. Thus, even the vehicle with the highest center of gravity did not run the risk of rolling over during any of the worst-case maneuvers considered in this research. This confirms the assertion of Sec- tion 4.6 that skidding is a much more pressing issue for nor- mal maneuvers on highways than is rollover. Finally, the above situations were simulated for the E-class sedan, E-class SUV, and full-size SUV. The lowest margins observed were for the SUVs, and particularly the E-class SUV, but no margins were low enough to elicit any concern. The lowest rollover margin detected was 0.63. This occurred at 70 mph for an E-class SUV conducting an emergency braking maneuver with a lane change. Clearly, wheel lift is not a con- cern for passenger vehicles conducting the various maneu- vers considered in this research. 4.11.3 Summary of Key Results from Step 10 In summary, the following findings were obtained from the analysis in Step 10: 1. For the vehicles considered in this research, rollover is not a direct concern for a vehicle traversing a sharp horizon- tal curve on a steep downgrade, not even when a vehicle performs a lane change, with or without braking while traversing the curve. 2. The single-unit truck exhibits momentary wheel lift at speeds of 50 to 65 mph for the case of emergency braking deceleration combined with a lane change on a steep grade of -9% and a superelevation of 0%. This momentary inside wheel lift is an artifact of the ABS’s actuation and suspension behavior but does not rep- resent a condition where the vehicle will likely roll- 0 5 10 15 20 25 0 0.5 1 R ol lo ve r M ar gi n Axle 1 0 5 10 15 20 25 0 0.5 1 R ol lo ve r M ar gi n Axle 2 0 5 10 15 20 25 0 0.5 1 R ol lo ve r M ar gi n Axle 3 0 5 10 15 20 25 0 0.5 1 Time (s) R ol lo ve r M ar gi n Axle 4 0 5 10 15 20 25 0 0.5 1 Time (s) R ol lo ve r M ar gi n Axle 5 Minimum Margin: 0.32 Figure 149. Rollover margins of individual axles for tractor semi- trailer/full-trailer truck (Double) (G  9%, e  8%) (ax  3 ft/s2 and lane change).

143 Figure 150. Rollover margins of individual axles for single-unit truck and tractor semi-trailer (G  9%, e  0% and 8%) (ax  11.2 ft/s2 and lane change). 30 40 50 60 70 80 -0.1 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 R ol lo ve r M ar gi n Speed (mph) e = 0%, G = -9%, SingleUnitTruck Axle 1 Axle 2 30 40 50 60 70 80 -0.1 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 R ol lo ve r M ar gi n Speed (mph) e = 0%, G = -9%, Tractor Trailer Axle 1 Axle 2 Axle 3 Axle 4 Axle 5 30 40 50 60 70 80 -0.1 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 R ol lo ve r M ar gi n Speed (mph) e = 8%, G = -9%, Single Unit Truck Axle 1 Axle 2 30 40 50 60 70 80 -0.1 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 R ol lo ve r M ar gi n Speed (mph) e = 8%, G = -9%, Tractor Trailer Axle 1 Axle 2 Axle 3 Axle 4 Axle 5 over for this worst-case roadway design and maneuver combination. 3. Assuming the CG height and track width of the trailers for both vehicles are the same, tractor semi-trailer/full- trailer trucks have very similar rollover margins compared to tractor semi-trailer trucks. This was evident based upon the static rollover margins estimated in Step 5 (see Section 4.6), and analyses using the “dynamic” rollover margin based upon the LTR confirmed the rollover simi- larities between both vehicles. 4. For tractor semi-trailer or tractor semi-trailer/full-trailer trucks with unusual loading, or those loaded to capac- ity, rollover may be more of a concern, especially for very aggressive avoidance maneuvers. But for the loading situ- ations simulated here—considered quite typical for these vehicles—there appears to be adequate wheel-lift margins. 4.12 Step 11: Analysis of Upgrades The objective of Step 11 was to analyze the effects of up- grades on lateral friction and rollover margins. Using the tran- sient bicycle and multibody models, this analysis estimated lateral friction margins for passenger vehicles traversing hori- zontal curves on upgrades, assuming that passenger vehicles maintain their desired speed on the upgrade and curve. For tractor semi-trailers, the analysis accounted for reduced speeds on upgrades (i.e., crawl speeds) and for traversing upgrades at the design speed. Vehicle rollover issues were also considered.

144 4.12.1 Analysis Approach On upgrades, the direction of the grade requires traction forces instead of braking forces to be applied on vehicles. While this generally makes braking efforts easier, for situa- tions without braking it means that more of the friction mar- gin may be used. The primary difference between braking and traction is that, for braking, the brake forces are distributed among all tires, but for traction, the drive forces are distrib- uted only to the drive axles. For trucks, it is unclear whether the traction forces required may cause the drive axles to skid during maneuvers on upgrades. If the vehicles require signifi- cant traction on the upgrade, skidding may occur for front- wheel-drive and rear-wheel-drive passenger vehicles as well. This analysis extends the results of Section 4.9, using a modified version of the bicycle transient models. These modifications are specifically to add the effects of grade and to distribute traction forces only to the drive axles of the vehicle. To calculate the theoretical crawl speeds of tractor semi- trailers on upgrades, Figure 153 is used to determine the force balance on the vehicle. In general, the traction force at the wheels, after losses due to transmission, rolling resistance, and other losses, requires a power output equal to the forces acting on the body, multiplied by the vehicle’s speed: i= (91)P F Vx Here P represents the wheel-horsepower of the tractor semi-trailer, V represents the speed of the tractor semi-trailer, and Fx represents the forces acting in the longitudinal direc- tion on the body of the vehicle. Fx is also equal to the final 30 40 50 60 70 80 -0.1 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 R ol lo ve r M ar gi n Speed (mph) e = 0%, G = -9%, Single Unit Truck Axle 1 Axle 2 30 40 50 60 70 80 -0.1 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 R ol lo ve r M ar gi n Speed (mph) e = 0%, G = -9%, TractorTrailer Axle 1 Axle 2 Axle 3 Axle 4 Axle 5 30 40 50 60 70 80 -0.1 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 R ol lo ve r M ar gi n Speed (mph) e = 8%, G = -9%, Single Unit Truck Axle 1 Axle 2 30 40 50 60 70 80 -0.1 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 R ol lo ve r M ar gi n Speed (mph) e = 8%, G = -9%, Tractor Trailer Axle 1 Axle 2 Axle 3 Axle 4 Axle 5 Figure 151. Rollover margins of individual axles for single-unit truck and tractor semi-trailer (G  9%, e  0% and 8%) (ax  15 ft/s2 and lane change).

145 Figure 152. Rollover margin time trajectories for individual axles for single-unit truck (V  50 to 65 mph, G  9%, e  0%) (ax  15 ft/s2 and lane change). 0 2 4 6 8 10 12 14 0 0.2 0.4 0.6 0.8 1 Fr on t R ol lo ve r M ar gi n e = 0%, G = -9%, V=50mph, Single-Unit Truck 0 2 4 6 8 10 12 14 0 0.2 0.4 0.6 0.8 1 R ea r R ol lo ve r M ar gi n Time (s) 0 2 4 6 8 10 12 14 0 0.2 0.4 0.6 0.8 1 Fr on t R ol lo ve r M ar gi n e = 0%, G = -9%, V=55mph, Single-Unit Truck 0 2 4 6 8 10 12 14 0 0.2 0.4 0.6 0.8 1 R ea r R ol lo ve r M ar gi n Time (s) 0 2 4 6 8 10 12 14 0 0.2 0.4 0.6 0.8 1 Fr on t R ol lo ve r M ar gi n e = 0%, G = -9%, V=60mph, Single-Unit Truck 0 2 4 6 8 10 12 14 0 0.2 0.4 0.6 0.8 1 R ea r R ol lo ve r M ar gi n Time (s) 0 2 4 6 8 10 12 14 16 18 0 0.2 0.4 0.6 0.8 1 Fr on t R ol lo ve r M ar gi n e = 0%, G = -9%, V=65mph, Single-Unit Truck 0 2 4 6 8 10 12 14 16 18 0 0.2 0.4 0.6 0.8 1 R ea r R ol lo ve r M ar gi n Time (s) tractive force at the driving wheels when no braking is being applied. Summing forces on the whole tractor semi-trailer system yields: i i i∑ = = − − ρ 100 1 2 (92)2F m dV dt P V mg G C A Vd Here, m is the total mass of the vehicle, g is the gravitational constant, Cd the drag coefficient, r is the density of air, and A the frontal area of the vehicle. When the tractor semi-trailer is no longer able to accelerate, it has reached its crawl speed and Equation 92 reduces to: i i= − − ρ0 100 1 2 (93)2 P V mg G C A VdFigure 153. Force balance on a tractor semi-trailer to determine crawl speeds.

146 which can be rearranged by multiplying by V to obtain: i i i= − − ρ0 100 1 2 (94)3P V mg G C A Vcrawl d The wheel-horsepower, P, is related to the rated engine horsepower, Peng; the static power load on the engine, Pstatic; and the rolling power coefficient, Croll, by the following: i= − − (95)P P P C Veng static roll So that one can write: i i i( )( )= − − + − ρ0 100 12 (96) 3P P V C mg G C A Veng static crawl roll d Equation 96 can be solved many ways. For this equation, two of the roots form a complex conjugate pair, and the third real root represents the crawl speed, Vcrawl, of the tractor semi-trailer. Representative values (for a tractor semi-trailer) for terms in Equation 96 were assumed as follows (McCallen et al., 2006; TruckSim): Cd = 0.79 r = 0.0739 lb/ft3 (at 77°F) A = 84.0 ft3 Peng = 35 Hp Pstatic = 413 Hp Croll = 1.15 Hp/mph Note, it was assumed that the driver shifts into a gear that allows maximum usage of the engine power at the crawl speed. These numerical values were used with Equation 96 to calculate crawl speeds across a range of grades. This solution was obtained for a tractor semi-trailer loaded with 22,000 lb. Comparisons of the theoretical crawl speeds with crawl speeds calculated within TruckSim for the same conditions/ assumptions confirmed the accuracy of the theoretical calcu- lations of crawl speeds for use in this analysis. On upgrades, the direction of the grade requires traction forces instead of braking forces to be applied on vehicles. While this generally makes braking efforts easier, the trac- tion force is concentrated on the drive axle. The traction force required for the upgrade is: i i= 100 (97)F m g G x truck While the normal force on the rear (drive) axle is: i i= (98)F m g pz truck r Here pr is the proportion of the truck weight on the rear axle, and it is usually about 44% of the total mass of the truck. Thus, the normalized friction required is the ratio of these two values: i( )= 100 (99),f G px traction r The equation predicts that the worst-case longitudi- nal traction force will occur for the steepest upgrades. For example, for an upgrade of 9%, the demand traction fric- tion is roughly 0.2. The supply friction for crawl speeds typical of trucks is approximately 0.62, and thus the trac- tion forces require 32% of the available longitudinal force on the tire. The friction ellipse modifies the available lateral force by the longitudinal force used, or for this example by the numerical value of √1-0.32 = 0.95. Thus, the maximum traction forces for 9% upgrades reduce the available lateral force by 95% of the non-traction values. Thus, the traction forces are not expected to significantly affect the lateral fric- tion margins. 4.12.2 Analysis Results 4.12.2.1 Passenger Vehicles Traveling at Design Speed For passenger vehicles, simulations were conducted at the design speed of the roadway as the power-to-weight ratios of these vehicles can be high enough to traverse steep grades at the design speed. Lateral friction margins were calculated for upgrades. For comparison, Figure 154 shows the lateral friction margins for upgrades from 0% to 9% and for down- grades from 0% to -9%. The lateral friction margins overall are slightly higher for upgrades due to the effect of grade. The exceptions are the constant-speed and the stopping sight distance deceleration cases. For the constant-speed case (ax = 0 ft/s2), upgrades have lower margins because the drive axle must utilize some additional friction to maintain the vehicle at speed. This pushes the margins for the constant-speed situations down to a level where they overlap with the curve- entry deceleration margins (ax = -3 ft/s2). For the stopping sight distance deceleration margins (ax = -11.2 ft/s2), these lateral friction margins are nearly identical between down- grades and upgrades. 4.12.2.2 Tractor Semi-Trailer Traveling at Crawl Speed on Upgrades Figure 155 shows the minimum lateral friction margins across all axles for tractor semi-trailer simulations from the transient bicycle model for 0% to 9% grades, for curve- keeping situations with a very large superelevation (16%).

147 30 40 50 60 70 80 -0.1 0 0.1 0.2 0.3 0.4 Speed (mph) M in im um F y M ar gi n Constant Speed, ax = 0 ft/s2 Curve Entry Decel, ax = -3 ft/s2 E-Class Sedan, e = 0% Emergency Decel, ax = -15 ft/s2 Grade = 0% Intermediate grades Grade = 9% 30 40 50 60 70 80 -0.1 0 0.1 0.2 0.3 0.4 Speed (mph) M in im um F y M ar gi n Constant Speed, ax = 0 ft/s2 Curve Entry Decel, ax = -3 ft/s 2 E-Class Sedan, e = 0% Emergency Decel, ax = -15 ft/s2 Grade = 0% Intermediate grades Grade = -9% 30 40 50 60 70 80 -0.1 0 0.1 0.2 0.3 0.4 Speed (mph) M in im um F y M ar gi n Constant Speed, ax = 0 ft/s2 Curve Entry Decel, ax = -3 ft/s2 E-Class SUV, e = 0% Emergency Decel, ax = -15 ft/s2 Grade = 0% Intermediate grades Grade = 9% 30 40 50 60 70 80 -0.1 0 0.1 0.2 0.3 0.4 Speed (mph) M in im um F y M ar gi n Constant Speed, ax = 0 ft/s2 Curve Entry Decel, ax = -3 ft/s2 E-Class SUV, e = 0% Emergency Decel, ax = -15 ft/s2 Grade = 0% Intermediate grades Grade = -9% 30 40 50 60 70 80 -0.1 0 0.1 0.2 0.3 0.4 Speed (mph) M in im um F y M ar gi n Constant Speed, ax = 0 ft/s2 Curve Entry Decel, ax = -3 ft/s2 Full-Size SUV, e = 0% Emergency Decel, ax = -15 ft/s2 Grade = 0% Intermediate grades Grade = 9% 30 40 50 60 70 80 -0.1 0 0.1 0.2 0.3 0.4 Speed (mph) M in im um F y M ar gi n Constant Speed, ax = 0 ft/s2 Curve Entry Decel, ax = -3 ft/s2 Full-Size SUV, e = 0% Emergency Decel, ax = -15 ft/s2Grade = 0% Intermediate grades Grade = -9% SSD Decel, ax = -11.2 ft/s2 SSD Decel, ax = -11.2 ft/s2 SSD Decel, ax = -11.2 ft/s2 SSD Decel, ax = -11.2 ft/s2 SSD Decel, ax = -11.2 ft/s2 SSD Decel, ax = -11.2 ft/s2 Figure 154. Lateral friction margins from transient bicycle model for passenger vehicles comparing upgrades (left plots) (G  0% to 9%) to downgrades (right plots) (G  0% to 9%), (e  0%) (ax  0, 3, 11.2, and 15 ft/s2).

148 Figure 156 shows the same situation with lane changes on the upgrade. The simulations were conducted assuming the vehicle was initially traveling at the crawl speed on the grade. The presence of upgrades generally did not cause worse margins than the 0% grade case. In fact, for nearly all con- ditions, the upgrades caused the tractor semi-trailer to slow so significantly that the lateral accelerations were greatly diminished, resulting in much higher lateral friction margins because the vehicle traverses the curves at much slower speeds. The single exception to this trend, that increasing grade improves the lateral friction margins, is the stopping sight distance deceleration case (ax = -11.2 ft/s2) where margins become worse for increasing upgrades. This is because the decelerations become more aggressive for increasing grades, due to how the stopping sight distance decelerations are cal- culated. In all the cases where the grades are low enough that the vehicle can operate at the design speed, the addition of grade causes worse margins for the stopping sight distance deceleration case. This is particularly notable for low speeds (i.e., 35 mph or less), where the addition of grade can cause negative friction margins. 4.12.2.3 Tractor Semi-Trailer Traveling at Design Speed on Upgrades There may be situations where a truck that has a slow crawl speed can traverse a steep upgrade at a high design speed. One example is when the upgrade is short and fol- lows a long stretch of roadway with a downgrade or level grade. In these situations, the truck’s momentum can main- tain the vehicle’s speed through much of the curve, resulting in much higher speeds in the curve. To study these situa- tions, tractor semi-trailer simulations were conducted using the transient bicycle model at the design speed, rather than the crawl speed, for curve-keeping situations (i.e., no lane changes). The results, shown in Figure 157, are compared to the downgrades and show that the margins are generally higher for upgrades than for downgrades, particularly for 30 40 50 60 70 80 -0.1 0 0.1 0.2 0.3 0.4 Speed (mph) M in im um F y M ar gi n ax = 0 ft/s2, e = 16% Grade = 0% Intermediate grades Grade = 9% 30 40 50 60 70 80 -0.1 0 0.1 0.2 0.3 0.4 Speed (mph) M in im um F y M ar gi n ax = -3 ft/s2, e = 16% Grade = 0% Intermediate grades Grade = 9% 30 40 50 60 70 80 -0.1 0 0.1 0.2 0.3 0.4 Speed (mph) M in im um F y M ar gi n ax = -11.2 ft/s2, e = 16% Grade = 0% Intermediate grades Grade = 9% 30 40 50 60 70 80 -0.1 0 0.1 0.2 0.3 0.4 Speed (mph) M in im um F y M ar gi n ax = -15 ft/s2, e = 16% Grade = 0% Intermediate grades Grade = 9% Figure 155. Lateral friction margins from transient bicycle model for tractor semi-trailer initially traveling at crawl speed (G  0% to 9%, e  16%) (ax  0, 3, 11.2, and 15 ft/s2).

149 the emergency deceleration cases. Like the situation observed for passenger vehicles, there is a reduction of margin for the constant-speed cases due to the traction required to main- tain speeds. For the stopping sight distance decelerations, there is a much larger sensitivity to grade: higher grades cause much larger reductions in margins, again due to the more aggressive deceleration demanded on upgrades versus downgrades. The worst margins are for low speeds, and the margins, while positive for all situations, are only marginally positive for the 25 mph case at both stopping sight distance and emergency braking decelerations. 4.12.2.4 Checking Transient Bicycle Model Results for Tractor Semi-Trailers Using the Multibody Model Because the lateral friction margins for the tractor semi- trailer are low for the stopping sight distance situations at low design speeds, the stopping sight distance deceleration scenarios were examined more closely using the multibody model for the tractor semi-trailer traveling initially at the design speed of the roadway. The first set of simulations considered the apparent worst-case upgrade (G = 9%) and superelevation (e = 16%). Figure 158 shows the individual inside/outside tire margins for the tractor semi-trailer for the multibody model. In Figure 158, the outside tire lateral friction margin disap- pears from the plot above 65 mph for axle 5, the very rearmost axle on the tractor semi-trailer. The margin is not shown here because it is actually infinitely negative as defined originally in Section 4.4, because the tire is lifted off the ground. This, along with the downward trend in friction margins with increasing speed, is in conflict with the predictions of the transient bicycle model. To gain more insight into why the margins could disagree so drastically, and even become infinitely negative, consider Figure 159, which shows the rollover margins for the same roadway design and maneuver. The rollover margin for high Figure 156. Lateral friction margins from transient bicycle model for tractor semi-trailer initially traveling at crawl speed (G  0% to 9%, e  16%) (ax  0, 3, 11.2, 15 ft/s2 and lane change). 30 40 50 60 70 80 -0.1 0 0.1 0.2 0.3 0.4 Speed (mph) M in im um F y M ar gi n ax = 0 ft/s2, e = 16% Grade = 0% Intermediate grades Grade = 9% 30 40 50 60 70 80 -0.1 0 0.1 0.2 0.3 0.4 Speed (mph) M in im um F y M ar gi n ax = -3 ft/s2, e = 16% Grade = 0% Intermediate grades Grade = 9% 30 40 50 60 70 80 -0.1 0 0.1 0.2 0.3 0.4 Speed (mph) M in im um F y M ar gi n ax = -11.2 ft/s2, e = 16% Grade = 0% Intermediate grades Grade = 9% 30 40 50 60 70 80 -0.1 0 0.1 0.2 0.3 0.4 Speed (mph) M in im um F y M ar gi n ax = -15 ft/s2, e = 16% Grade = 0% Intermediate grades Grade = 9%

150 speeds is low and hits a minimum of zero for speeds from 70 to 80 mph. As defined in Section 4.11, a rollover margin of zero implies that a wheel has lifted on the axle, which implies a zero normal force on that tire. With a zero normal force, the friction demand approaches infinity, since friction demand is defined as the cornering force required on the axle divided by the normal (vertical) force on the axle. With an infinite friction demand, infinitely negative lateral friction margins are inevitable. The low minimum rollover margins shown in Figure 159 are likely the reason for the decreasing trend in friction mar- gins with speed as shown in Figure 158. With very little nor- mal load in reserve, the “lighter” tires on an axle will often artificially decrease the minimum friction margin for a given maneuver, while the vehicle may still be controllable and maintain its intended path. Figure 160 shows both the normal force on each tire for the maneuver considered and the lateral deviation to offer more insight on the severity of the low rollover and friction margins seen for higher speeds. The plots show a pronounced oscillation in the vertical loads on the rear tractor and the trailer axle groups, yet a reason- able path tracking performance. The vertical load oscillation appears to coincide with the spike in lateral deviation shown, which is under 2 ft at its maximum. This indicates that the vehicle was able to maintain its path on the road, but that the trailer began to rock back and forth while the vehicle was decelerating, leading to a brief (under 1 s) wheel-lift event for axles 4 and 5 for the outside tire at approximately 17 s into the maneuver. The most interesting point to note for this case is that it is the outside tire which lifts momentarily, leading to the predicted infinitely negative lateral friction margin and zero rollover margin for the 70 mph design speed maneuver. What this means, in the face of 16% superelevation, is that the high wheel is lifting off the ground, implying that the vehicle is leaning heavily toward the low edge of the road. 30 40 50 60 70 80 -0.1 0 0.1 0.2 0.3 0.4 Speed (mph) M in im um F y M ar gi n Constant Speed, ax = 0 ft/s2 Curve Entry Decel, ax = -3 ft/s2 Tractor Trailer, e = 0% Emergency Decel, ax = -15 ft/s2 Grade = 0% Intermediate grades Grade = 9% 30 40 50 60 70 80 -0.1 0 0.1 0.2 0.3 0.4 Speed (mph) M in im um F y M ar gi n Constant Speed, ax = 0 ft/s2 2 Emergency Decel, ax = -15 ft/s2 Grade = 0% Intermediate grades Grade = -9% 30 40 50 60 70 80 -0.1 0 0.1 0.2 0.3 0.4 Speed (mph) M in im um F y M ar gi n Constant Speed, ax = 0 ft/s2 /s2 Tractor Trailer, e = 16% Emergency Decel, ax = -15 ft/s2 Grade = 0% Intermediate grades Grade = 9% 30 40 50 60 70 80 -0.1 0 0.1 0.2 0.3 0.4 Speed (mph) M in im um F y M ar gi n Constant Speed, ax = 0 ft/s2 Curve Entry Decel, ax = -3 ft/s2 Tractor Trailer, e = 16% Emergency Decel, ax = -15 ft/s2 Grade = 0% Intermediate grades Grade = -9% Tractor Trailer, e = 0% SSD Decel, ax = -11.2 ft/s2 SSD Decel, ax = -11.2 ft/s2 SSD Decel, ax = -11.2 ft/s2 SSD Decel, ax = -11.2 ft/s2 Curve Entry Decel, ax = -3 ft/s Curve Entry Decel, ax = -3 ft Figure 157. Lateral friction margin from transient bicycle model for tractor semi-trailers comparing upgrades ( left plots; G  0% to 9%) to downgrades (right plots; G  0% to 9%) (e  0% and 16%) (ax  0, 3, 11.2, and 15 ft/s2).

151 Figure 158. Lateral friction margins for inside and outside tires from multibody models for tractor semi-trailer (G  9%, e  16%) (ax  11.2 ft/s2).

152 30 40 50 60 70 80 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 Ro llo ve r M ar gi n Speed (mph) e = 16%, G = 9%, TractorTrailer Axle 1 Axle 2 Axle 3 Axle 4 Axle 5 Figure 159. Rollover margins of individual axles for tractor semi-trailer (G  9%, e  16%) (ax  11.2 ft/s2). Figure 160. Vertical tire loads of individual axles ( left plots) and deviation from intended path (right plots) for tractor semi-trailer (V  70 mph, G  9%, e  16%) (ax  11.2 ft/s2). 0 2 4 6 8 10 12 14 16 18 20 0 5000 10000 15000 N or m al L oa d (lb f) Outside Tire Axle 1 Inside Tire Axle 1 0 2 4 6 8 10 12 14 16 18 20 0 5000 10000 15000 N or m al L oa d (lb f) Outside Tire Axle 2 Inside Tire Axle 2 Outside Tire Axle 3 Inside Tire Axle 3 0 2 4 6 8 10 12 14 16 18 20 0 5000 10000 15000 Time (s) N or m al L oa d (lb f) Outside Tire Axle 4 Inside Tire Axle 4 Outside Tire Axle 5 Inside Tire Axle 5 0 2 4 6 8 10 12 14 16 18 -1.5 -1 -0.5 0 0.5 1 1.5 Time (s) La te ra l D ev ia tio n (ft) Desired Deviation Actual Deviation Figure 161 depicts why this happens and gives some impor- tant context to the wheel-lift event. It shows the inputs as a function of time for the 70 mph curve-keeping maneuver. Figure 161 brings several very important facts to light. First, for the multibody model, the vehicle speed decreases from 70 mph gradually as soon as the simulation starts. This is because, while the transient bicycle model simulates the vehicle at either the design speed or the crawl speed, the multi body model includes all engine effects and will natu- rally result in a decrease in speed toward the crawl speed from the moment the simulation begins. This also results in the time shift of the deceleration event between the transient model and the multibody model. The former initiates the deceleration at a particular time, and the latter initiates the deceleration at a particular distance on the curve. Second, notice the sine-wave-like steering input occurring in the multibody model just after 15 s. This is caused by the simu- lated driver attempting to maintain the desired trajectory within the lane. This corrective steering is caused for trucks because, at speeds below the design speed, the down-slope side of the vehicle is experiencing more braking force than the up-slope side of the vehicle, causing the vehicle to steer

153 Figure 161. Trajectory of simulation inputs for transient bicycle and multibody models for tractor semi-trailer (V  70 mph, G  9%, e  16%) (ax  11.2 ft/s2). 0 5 10 15 20 25 -30 -20 -10 0 D ec el er at io n (ft/ s2 ) Transient Bicycle Model Multibody Model 0 5 10 15 20 25 0 50 100 Sp ee d (m ph ) Transient Bicycle Model Multibody Model 0 5 10 15 20 25 -0.15 -0.1 -0.05 0 0.05 Time (sec) St ee rin g In pu t (r ad ) Transient Bicycle Model Multibody Model Outside Tire Multibody Model Inside Tire down-slope slightly. The simulated driver attempts to correct for this by steering up-slope slightly. This steering change creates a special situation unique to low-speed traversal of upgrades: this situation is potentially more demanding than the curve keeping with lane-change maneuver simulated in Section 4.9. For maneuvers near the design speed, this would not be a problem, but as the speed decreases during deceler- ation, the lateral acceleration required to maintain the turn also decreases. Thus, steering toward the uphill end of the cross-slope constitutes a maneuver that allows the superele- vation to effectively decrease the vehicle’s rollover threshold. This effect, chiefly due to the efforts of the simulated driver to stay on the road, combined with the inside/outside load differences on each axle, cannot be captured by the transient bicycle model. These results suggest that special consideration may be needed in the choice of the design superelevation on upgrades to prevent tractor semi-trailers from lifting a wheel as their drivers attempt to decelerate and maintain their position on the road during a stopping sight distance deceleration event. To find a superelevation that avoids this issue for a 9% upgrade, Figure 162 shows the rollover margins for the tractor semi-trailer during the same curve-keeping maneuver, but for more moderate design superelevations of 8% and 12%. Figure 162. Rollover margins of individual axles for tractor semi-trailer (G  9%, e  8% and 12%) (ax  11.2 ft/s2). 30 40 50 60 70 80 0.2 0.3 0.4 0.5 0.6 0.7 0.8 Ro llo v e r M ar gi n Speed (mph) e = 8%, G = 9%, TractorTrailer Axle 1 Axle 2 Axle 3 Axle 4 Axle 5 30 40 50 60 70 80 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 Ro llo v e r M ar gi n Speed (mph) e = 12%, G = 9%, TractorTrailer Axle 1 Axle 2 Axle 3 Axle 4 Axle 5

154 The predicted rollover margins for a superelevation of 8% are positive for all speeds, although a minimum of 0.1 for the 80 mph traversal is still rather low. It appears that the 12% superelevation will still result in wheel lift at a 70 mph design speed. To examine whether the same wheel-lift behavior observed at 9% upgrades and 12% superelevation is observed at dif- ferent grades and/or intermediate superelevations, a battery of simulations was run for superelevations of 8%, 9%, 10%, 11%, and 12% for upgrades from 4% to 9%. The rollover margins for simulations of 8% and 12% superelevations are summarized in Figure 163 for a 4% upgrade, in Figure 164 for a 5% upgrade, and in Figure 165 for a 7% upgrade. The sen- sitivity analysis revealed that on upgrades of 4% and super- elevations between 8% and 12%, the lowest speed at which wheel lift occurred for a tractor semi-trailer undergoing stop- ping sight distance deceleration was 75 mph at 12% super- elevation. For upgrades of 5%, low rollover margins began to occur near speeds of 60 mph at 8% superelevation; and as superelevation increased, wheel lift occurred at speeds as low as 55 mph at 12% superelevation. For upgrades of 7%, low rollover margins occurred near speeds of 55 and 60 mph for all superelevations evaluated, but wheel lift did not occur until an initial speed of 70 mph at 12% superelevation. As Figure 163. Rollover margins of individual axles for tractor semi-trailer (G  4%, e  8% and 12%) (ax  11.2 ft/s2). Figure 164. Rollover margins of individual axles for tractor semi-trailer (G  5%, e  8% and 12%) (ax  11.2 ft/s2). 30 40 50 60 70 80 0.2 0.3 0.4 0.5 0.6 0.7 0.8 R ol lo ve r M ar gi n Speed (mph) e = 8%, G = 5%, TractorTrailer Axle 1 Axle 2 Axle 3 Axle 4 Axle 5 30 40 50 60 70 80 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 R ol lo ve r M ar gi n Speed (mph) e = 12%, G = 5%, TractorTrailer Axle 1 Axle 2 Axle 3 Axle 4 Axle 5

155 a general rule, these results suggest that on upgrades of 4% and greater, the maximum superelevation should be limited to 9% for curves with design speeds of 55 mph and higher. Alternatively, if it can be verified that the available sight dis- tance is such that deceleration at -11.2 ft/s2 is unlikely to be required on upgrades of 4% and greater, minimum-radius curves could be designed using a maximum superelevation up to 12% on these steep upgrades. 4.12.3 Summary of Key Results from Step 11 In summary, the following findings were obtained from the analysis in Step 11: 1. For passenger vehicles on upgrades, the lateral friction mar- gins predicted by the transient bicycle model were similar and/or generally higher than those observed on down- grades. The exceptions are the margins for the constant- speed and stopping sight distance deceleration cases. On upgrades, for the constant-speed case, the lateral friction margins are slightly reduced due to the traction necessary to maintain speed on upgrades. This reduction in margin is fairly minor and is similar in magnitude to the margin reductions caused by curve-entry deceleration. For stop- ping sight distance deceleration, lateral friction margins are nearly identical between upgrades and downgrades. 2. Upgrades in general require more traction forces, but for tractor semi-trailers the slower crawl speeds on upgrades significantly reduce the lateral forces. The result is that lateral friction margins are generally better for upgrades than downgrades for tractor semi-trailers. 3. As vehicles undergo stopping sight distance deceleration on upgrades, the net effect on lateral friction margins is to actually reduce friction margins in these braking situations. 4. Design superelevations on upgrades of 4% and greater should be limited to a maximum of 9% to avoid the pos- sibility of wheel lift on tractor semi-trailers as predicted by the multibody model for speeds above 55 mph when undergoing stopping sight distance deceleration on the curve. Alternatively, if it can be verified that the available sight distance is such that deceleration at -11.2 ft/s2 is unlikely to be required on upgrades of 4% and greater, maximum superelevation values up to 12% may be used for minimum-radius curves. 4.13 Summary of Analytical and Simulation Modeling Results from Step 1 of the analysis provide comparisons between road friction measurements and the maximum side friction, fmax, used in the current AASHTO design policy for horizontal curves. The friction supply curves for both the lateral (cornering) and longitudinal (braking) directions for both passenger vehicles and trucks are higher than the maxi- mum friction demand curves given by AASHTO policy. Thus, current horizontal curve design policy appears to provide reasonable lateral friction margins against skidding. These results suggest that if there is going to be an area of concern based upon AASHTO’s current design policy, it will likely arise from the interaction of braking and cornering forces. A series of analyses was undertaken incorporating more complex vehicle dynamics simulation models within the pro- cedures to investigate margins of safety against skidding and Figure 165. Rollover margins of individual axles for tractor semi-trailer (G  7%, e  8% and 12%) (ax  11.2 ft/s2). 30 40 50 60 70 80 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 Ro llo ve r M ar gi n Speed (mph) e = 8%, G = 7%, TractorTrailer Axle 1 Axle 2 Axle 3 Axle 4 Axle 5 30 40 50 60 70 80 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 Ro llo ve r M ar gi n Speed (mph) e = 12%, G = 7%, TractorTrailer Axle 1 Axle 2 Axle 3 Axle 4 Axle 5

156 rollover for a variety of vehicles types when traversing sharp horizontal curves on steep grades. The point-mass model was the simplest model considered, while the transient bicycle and multibody models are more complex and simulate vehicles using multiple axles and multiple tires, respectively. Incor- porating more complex vehicle dynamics simulation models to investigate margins of safety against skidding and rollover of vehicles traversing sharp horizontal curves on steep grades revealed several significant findings: • When maintaining a vehicle operating speed at or near the design speed on a horizontal curve, grade and supereleva- tion appear to have little effect on the margins of safety against skidding and rollover. • When vehicles change lanes on a horizontal curve, the margins of safety against skidding decrease considerably for all vehicle types. When lane changing occurs in combi- nation with severe braking (i.e., stopping sight distance or emergency braking deceleration levels), significant reduc- tions in margins of safety against skidding can occur. • The superelevation attained at the point of curve entry should be checked and compared to a lateral friction mar- gin condition to ensure that the lateral friction margin on the curve entry is not less than the margin within the curve. • The more complex models (i.e., the transient bicycle and multibody models) indicate that the point-mass model generally overestimates the margins of safety against skid- ding and rollover across all vehicle types.