Click for next page ( 22


The National Academies | 500 Fifth St. N.W. | Washington, D.C. 20001
Copyright © National Academy of Sciences. All rights reserved.
Terms of Use and Privacy Statement



Below are the first 10 and last 10 pages of uncorrected machine-read text (when available) of this chapter, followed by the top 30 algorithmically extracted key phrases from the chapter as a whole.
Intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text on the opening pages of each chapter. Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.

Do not use for reproduction, copying, pasting, or reading; exclusively for search engines.

OCR for page 21
21 1.E-03 1.E-04 Risk of Collision per Operation 1.E-05 1.E-06 1.E-07 1.E-08 1.E-09 7 8 9 10 11 12 13 14 15 16 17 Taxilane/Taxilane Wingtip Separation (ft) Figure 14. Taxilane/taxilane collision probability based on wingtip separation. 1.E-03 1.E-04 Risk of Collision per Operation 1.E-05 1.E-06 1.E-07 1.E-08 1.E-09 1.E-10 4 6 8 10 12 14 16 18 20 Taxilane/Object Wingtip Separation (ft) Figure 15. Taxilane/object collision risk based on wingtip separation. Runway Deviation Modeling Risk if the aircraft loses directional control after touching down and veers off the runway, colliding with fixed or mov- The probability of aircraft collision associated with the sep- able objects. aration between the runway and taxiway or objects depends on whether the movement is a landing or a takeoff operation. For landing there are two types of risk that may be evaluated: These two types of risk may be combined to provide the total risk for landing. The veer-off risk is estimated for every Risk during the final approach phase when the aircraft is air- landing operation, whereas the airborne risk is computed only borne and the combination of large lateral and vertical devi- for missed approaches. ations from the nominal approach path may lead to collision When taking off, the pilot may reject the procedure and with a fixed or movable object in the airfield area (e.g., an intentionally or unintentionally veer-off the runway. If direc- aircraft taxiing in a parallel taxiway). tional control is lost when in high speed, the aircraft may col-

OCR for page 21
22 Figure 16. Plan view of CRM run experiment. lide with fixed or movable objects after departing the runway Several CRM runs were made with obstacles located at var- edge and possibly the RSA. ious separations from the runway centerline and along the run- The methodology used to assess risk during the airborne way length. For Cat I, an obstacle clearance height (OCH) of phase was the CRM. Although the CRM model was developed 200 ft was used, and for Cat II an OCH of 100 ft was used. The in the 1970s and the FAA has been improving these models ranges used were -300, 0, 1,500, 3,000, and 4,500 ft. Figures 16 since the 1990s, the original CRM can serve as a screening tool and 17 illustrate the experiment run for several runway/taxiway to evaluate the feasibility of submitting an MOS. The FAA has separations. In Figure 16, only ranges of -300, 0, and 1,500 ft other tools to evaluate the need for further analysis if the risk are shown. estimated is within a feasible range. A summary of the CRM With the results from the CRM analysis, the maximum risk approach, variables, and models is provided in Appendix B. for each runway/taxiway separation was identified among the five locations along the parallel taxiway. The next step was to develop risk plots for wingtip separation versus risk in terms of Landing--Airborne Phase Model accidents per number of operations. The process was repeated It is recognized that running the CRM demands the avail- for each ADG, and risk plots were prepared for each group and ability of specific software and the expertise to use it. To facili- instrument approach Cat 1 and 2. tate the analysis and the application of the methodology, several The approach to estimating risk using the plots presented CRM runs were made for common situations and different in Appendix A may be viewed as conservative for the follow- ADGs; however, when the analysis involves specific aircraft ing reasons: rather than an ADG, for the airport to obtain more accurate estimates of risk, the assessment should use direct results from Only the highest risk along the runway length was used for the CRM analysis and specific conditions, if possible. the estimate of collision risk. Figure 17. Perspective view of CRM run experiment (not to scale).

OCR for page 21
23 The tallest tail height in the ADG was used to characterize calibrate the veer-off frequency models. Data were gathered the taxiing aircraft as an obstacle. from accidents and incidents and NOD. To avoid the negative The widest wingspan in the ADG was used to characterize effects of multicolinearity on the model, correlations between the dimension of the approaching aircraft. independent variables were tested to eliminate highly correlated The taxiing aircraft was assumed to be a fixed object; how- variables, particularly if they did not contribute significantly to ever, for many airports, a taxiing aircraft will not be present explaining the variation of the probability of an accident. during most of the landings. The selected approach can identify relationships missed by A missed approach rate of 1 percent was kept for the calcu- forward stepwise logistic regression (Hosmer and Lemeshow, lations. Based on the latest FAA data, even a 0.2-percent 2000). The predictor variables were entered by blocks, each con- rate is conservative (FAA, 2008). sisting of related factors, such that the change in the model's substantive significance could be observed as the variables were included. Runway Veer-Off Models for Landing The basic model structure used is logistic, as follows: and Takeoff 1 During the landing, after touchdown, or during the takeoff P { Accident _ Occurrence } = roll, the pilot may lose directional control. Some common 1 + e b0 +b1X1 +b2 X2 +b3X3 +K causes and contributing factors include low runway friction, where mechanical failures, and adverse weather conditions. The basis of the approach used in this study is the probabil- P{Accident_Occurrence} is the probability (0100%) of an ity of aircraft runway excursions and the risk that an aircraft accident type occurring given certain operational conditions, will stop outside the boundaries of the existing or planned RSA. Xi are independent variables (e.g., ceiling, visibility, cross- The approach to model risk of collisions is accomplished by wind, precipitation, and aircraft type), and using a combination of frequency and location models. In a bi are regression coefficients. sense, the modeling considered the bounds of the RSAs rather Several parameters were considered for inclusion in the than the presence of obstacles in the vicinity of the RSAs or the models. The backward stepwise procedure helps identify the aircraft speed when striking obstacles. While the difference most relevant variables for each type of event. One major makes the new models simpler, the approach can be extended improvement relative to models presented in previous stud- to consider risk if this type of analysis is required. The two- ies was the use of tailwind and headwind. These variables were part model approach is represented in Figure 18. not present in the overrun and undershoot models presented in ACRP Report 3 (Hall et al., 2008) because the actual run- way had not been identified in the NOD. The research team Event Probability (Frequency Model) gathered information on the runways used, and the process The likelihood of an aircraft veer-off incident depends on allowed the calculation of the head/tailwind components of operational conditions and human factors. It includes airport the model. characteristics, weather conditions, and aircraft performance, Another major accomplishment that has increased model as well as the relationship between the runway distance required accuracy was the inclusion of a runway criticality factor. The by the aircraft for the given conditions and the runway distance new parameter represents the interaction between the run- available at the airport. way distance required by the aircraft and the runway distance Similar to the approach presented in ACRP Report 3 (Hall available at the airport. The logarithm of the ratio between et al., 2008), backward stepwise logistic regression was used to the distance required and the distance available was used to Two-Part Probability Model Event Location probability probability Probability Operating conditions Conditions (airport, type of operation, RSA configuration, Veer-off and depart runway distances, weather, location of obstacles, the RSA aircraft performance) airplane aircraft wingspan wingspan Figure 18. Modeling approach.

OCR for page 21
24 represent the criticality factor. The greater the value, the more To summarize, the runway distance required was adjusted critical the operation because the safety margin decreases, and for temperature, elevation, runway surface conditions, and in many cases strong braking or the possibility of overruns wind. Table 6 presents the factors applied to the distance may lead to veer-off events. required by the aircraft. A correction for slope was not applied, The distance required is a function of the aircraft per- as this factor had little effect on the total distance required. formance under specific conditions. Therefore, every distance The use of NOD in the accident frequency model was a required under International Organization for Standardiza- major improvement introduced in ACRP Report 3 (Hall et al., tion (ISO) conditions (sea level, 15 deg centigrade) was con- 2008), and it was maintained for this study. The analysis with verted to actual conditions for operations. Moreover, the dis- NOD also adds to the understanding of cause and effect rela- tances were adjusted for the runway surface condition (wet, tionships for veer-off incidents. Table 7 summarizes the model snow, slush, or ice) and for the level of head/tailwind. The coefficients obtained for each veer-off frequency model. adjustment factors for runway surface conditions are those Table 8 summarizes the parameters representing the accu- recommended by the Flight Safety Foundation (2009). racy of each model obtained presenting the R2 and C-values Table 6. Correction factors applied to required runway distance. Local Factor Unit Reference Adjustment Factor Definitions Elevation (E) (i) 1000 ft E = 0 ft (sea Fe = 0.07 x E + 1 Fe is runway level) distance adjustment factor for elevation Temperature (T) (i) deg C T = 15 deg C Ft = 0.01 x (T (15 1.981 Ft is runway E) + 1 distance adjustment factor for temperature Tailwind (TWLDJ) knot TWLDJ = 0 FTWJ = (RD + 22 x FTWJ is runway for Jets (iii) knot TWLDJ)/RD (ii) distance adjustment factor for tailwind (jets) Tailwind (TWLDT) knot TWLDT = 0 FTWP = (RD + 30 x FTWT is runway for Turboprops (iii) knot TWLDT}/RD distance adjustment factor for tailwind (turboprops) Headwind (HWTOJ) knot HWTOJ = 0 FHWJ = (RD + 6 x FHWJ is runway for Jets (iii) knot HWTOJ)/RD distance adjustment factor for headwind (jets) Headwind (HWTOT) knot HWTOJ = 0 FTWP = (RD + 6 x FHWT is runway for Turboprops (iii) knot HWTOT)/RD distance adjustment factor for headwind (turboprops) Runway Surface Yes/No Dry FW = 1.4 FW is runway Condition Wet (W) distance adjustment (iv) factor for wet pavement Runway Surface Yes/No Dry FS = 1.6 FS is runway Condition Snow distance adjustment (S) (iv) factor for snow- covered pavement Runway Surface Yes/No Dry FSl = 2.0 FSl is runway Condition Slush distance adjustment (Sl) (iv) factor for slush- covered pavement Runway Surface Yes/No Dry FI = 3.5 FI is runway Condition Ice (I) (iv) distance adjustment factor for ice- covered pavement i - temperature and elevation corrections used for runway design ii - RD is the runway distance required iii - correction for wind are average values for aircraft type (jet or turboprop) iv runway contamination factors are those suggested by Flight Safety Foundation (2000)

OCR for page 21
25 Table 7. Independent variables for veer-off frequency models. Variable LDVO TOVO1 Variable LDVO TOVO1 Adjusted Constant 3.088 15.612 Temp from 5 to 15C 0.453 0.420 User Class G 1.682 2.094 Temp more than 25C 0.291 0.921 Aircraft Class A/B 0.770 0.852 Icing Conditions 2.67 3 Aircraft Class D/E/F 0.252 0.091 Rain 0.126 1.541 Visibility less than 2 SM2 2.143 2.042 Snow 0.548 0.963 Visibility from 2 to 4 SM 3 0.808 Frozen Precipitation 0.103 3 Visibility from 4 to 8 SM 3 1.500 Gusts 0.036 3 Xwind from 5 to 12 kt 0.653 0.102 Fog 1.740 3 Xwind from 2 to 5 kt 0.091 3 Turboprop 2.517 1.522 Xwind more than 12 kt 2.192 0.706 Foreign Origin/Destination 0.334 0.236 Tailwind from 5 to 12 kt 0.066 3 Hub/Non-Hub Airport 3 0.692 Tailwind more than 12 kt 0.98 3 Log Criticality Factor 4.318 1.707 Temp less than 5C 0.558 0.988 Night Conditions 1.360 3 1 LDVO = landing veer-off; TOVO = takeoff veer-off. 2 SM = statute miles. 3 Blank cells indicate that there are no coefficients associated with these parameters. Where Equipment Class Ref: C Large jet of MTOW 41k-255k lb (B737, A320 etc.) Heavy Acft AB Heavy jets of MTOW 255k lb+ Large commuter of MTOW 41k-255k lb (small RJs, Commuter Acft D ATR42 etc.) Medium aircraft of MTOW 12.5k-41k lb (biz jets, Medium Acft E Embraer 120 Learjet 35 etc.) Small aircraft of MTOW 12.5k or less (small, single or Small Acft F twin engine Beech90, Cessna Caravan etc.) User Class Ref: C = Commercial or F = Cargo or T/C = Taxi/Commuter User Class G G = GA Turboprop Turboprop engine(yes/no) Ref: Turbojet Ceiling Height feet Visibility statute miles Crosswind knots Tailwind knots Gusts Yes/No Ref: No Icing Conditions Yes/No Ref: No Snow Yes/No Ref: No Rain Yes/No Ref: No Frozen Precipitation Yes/No Ref: No Fog Yes/No Ref: No Air Temperature deg C Turboprop Aircraft Yes/No Ref: No NonhubApt Yes/No Ref: Yes for hub airport If Log(CF) > 0, available runway distance is smaller than required Log Criticality Factor distance Night Conditions Night, Dawn or Dusk Ref: Daylight Notes: Ref: indicates the reference category against which the odds ratios should be interpreted. Non-hub airport: airport having less than 0.05% of annual passenger boardings for each model. Relatively low R2 values are the norm in logis- Table 8. Summary statistics for tic regression (Ash and Schwartz, 1999), and they should not veer-off frequency models. be compared with the R2 of linear regressions (Hosmer and Model R2 C Lemeshow, 2000). A better parameter to assess the predictive LDVO 0.32 0.88 capability of a logistic model is the C-value. This parameter TOVO 0.14 0.82 represents the area under the sensitivity/specificity curve for

OCR for page 21
26 the model, which is known as the receiver operating charac- number of accidents and incidents, and the total traffic of teristic (ROC) curve. relevant operations from 1982 to 2009. Sensitivity and specificity are statistical measures of the From Table 9, one can see that landing veer-offs are performance of a binary classification test. Sensitivity mea- approximately four times more likely to occur than takeoff sures the proportion of true positives that are correctly iden- veer-offs. tified as such (the percentage of accidents and incidents that are identified when using the model). Specificity measures the Event Location proportion of true negatives that are correctly identified (the percentage of normal operations that the model can identify The farthest location of the veer-off path from the runway as non-incident). These two measures are closely related to edge was used to develop the location models. The probabil- the concepts of Type I and Type II errors. A theoretical, opti- ity of this distance during an incident is not equal for all loca- mal prediction can achieve 100-percent sensitivity (i.e., pre- tions measured from the runway centerline or runway edge. dict all incidents) and 100-percent specificity (i.e., not predict The probability of veer-off with lateral deviation in the prox- a normal operation as an incident). imity of the runway edge is higher than at larger distances To assess how successful the models are in classifying flights from that boundary. This dependence is represented by the correctly as "incident" or "normal," and to find the appropri- accident location model, which is the second main element of ate cut-off points for the logistic regression model, the ROC the risk assessment approach. The accident location models curves were defined for each model to calculate the C-value as are based on historical accident and incident data for aircraft shown in Table 8. The values achieved for the veer-off model veer-offs. are considered very good, with the area under the curve rep- Worldwide data on accidents and incidents were used to resenting a C-value higher than 80 percent. develop the location models. The model structure is an expo- The frequency models developed under this study will nential decay function similar to that used in the research require the use of historical information on operations and reported in ACRP Report 3 (Hall et al., 2008). Based on the weather for the specific airport. The necessary information on accident/incident location data, two cumulative probabil- operations includes the time of the flight, runway used, type ity distribution models were developed. With the functions of aircraft, type of flight, and whether the operation was an obtained, the fraction of accidents involving locations exceed- arrival or departure. In addition, it is necessary to collect the ing a given distance from the runway edge can be estimated. weather information for the same period that operational When the probability estimated with the location model data are available, usually for 1 year. is multiplied by the frequency of accident occurrence, it Weather information can be acquired directly from the is possible to quantify the overall frequency of incidents National Oceanic and Atmospheric Administration (NOAA) involving locations exceeding a given distance from the database for the weather station located at the airport. How- runway edge. ever, the information on operations, particularly for non- Figure 19 shows the runway edge origin location used to rep- towered airports, may be harder to obtain, particularly the resent veer-off incidents. The reference location of the aircraft identification of the runway used. For towered airports oper- is its nose wheel. The y-axis origin is the edge of the runway, not ational data can be requested from the FAA. Another challenge necessarily the edge of the paved area when the runway has is running the analysis because computations can be made only shoulders. with the help of a computer and specific software that incorpo- The model structure for the location models is the following: rates these models. To facilitate the analysis, a series of plots were developed P { Location > y } = e - bym based on average veer-off incident rates for the United States. where The rates are presented in Table 9 and were combined with the location models to build the risk plots presented in P{Location > y} is the probability that the veer-off distance Appendix A. The average incident rates are based on the from the runway edge is greater than y, Table 9. Average veer-off incident rates (19822009). Type of Number of Incident Rate per Incident Rate in Incident Incidents Operation Operations per Incident LDVO 512 1.195E-06 837,000 TOVO 111 2.590E-07 3,861,000

OCR for page 21
27 Prob=exp((-.02568)*Y**(.803946)) R2=99.5% 1.0 Probability of Stopping Beyond Y y 0.8 0.6 Figure 19. Y origin for aircraft veer-offs. 0.4 y is a given location or distance from the runway edge, and b and m are regression coefficients. 0.2 A typical transverse location distribution is presented in Figure 20. 0.0 0 200 400 600 800 1000 The actual model parameters are presented in Table 10 and Distance Y from Runway Edge (ft) illustrated in Figures 21 and 22. Figure 21. Location model for landing veer-off. Probability location Prob=exp((-.01639)*Y**(.863461)) R2=94.2% Exceeds y 1.0 m P{ Location y} e by Probability of Stopping Beyond Y 0.8 P{Loc > y1} 0.6 rwy border y1 Distance y from runway border y 0.4 Figure 20. Typical model for aircraft veer-offs. 0.2 Table 10. Summary of veer-off location models. 0.0 0 200 400 600 800 1000 Distance Y from Runway Edge (ft) Type of Type of Model R2 # of Accident Data Points Figure 22. Location model for takeoff veer-off. LDVO Y P{d y} e 0.02568 y 0.803946 99.5% 126 TOVO Y P{d y} e 0.01639 y 0.863461 94.2% 39