Intelligent Soil Compaction Systems (2010)

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.

 Fundamentals of Roller Measurement Values The reader is referred to Facas & Mooney (2010) for a thor- ough description of roller MV reporting. Each reported MV is often a reflection of vibration data averaged over a time window t MV . Values of t MV vary across manufacturers and are somewhat user programmable (val- ues observed during this study are summarized in Table 3.1). The corresponding spatial window over which an individual roller MV is representative (x MV in Figure 3.2) is a function of t MV and roller speed v (x MV = t MV × v). Values of the spatial window x MV observed using v = 1 m/s (3.3 ft/s) ranged from 0.06 to 1.0 m (0.2 to 3.3 ft) and varied across manufacturers (see Table 3.1). For current roller MVs, the averaging window y MV (see Figure 3.2) is equal to the length of the drum [typi- cally 2.1 m (6.9 ft)]. The spatial reporting resolution of roller MVs in the direc- tion of roller travel (∆x = f report × v in Figure 3.2) also varies across manufacturers. In this study, ∆x varied from 0.2 to 1.0 m (0.66 to 3.3 ft) for v = 1 m/s (3.3 ft/s). The spatial report- ing resolution ∆y is a function of overlap between two adja- cent roller passes. The recommended overlap is 0.1 m (0.3 ft); therefore, ∆y = 2 m (6.6 ft), typically. The combination of ∆x and x MV reflects the continuity and relative coverage of MV data provided by each roller (see Fig- ure 3.3). Records for k s provide complete coverage with no overlap (∆x = x MV ). CMV D and CCV records provide com- plete coverage with considerable overlap (∆x < x MV ). E vib re- cords provide less than complete coverage (∆x > x MV ). It is worth noting that the parameters ∆x and x MV can be modified in the roller software. The ∆x = x MV configuration is optimal since the resulting MVs are reasonably independent while providing complete coverage. For the purpose of consistency in data reporting and sta- tistical analysis and to ensure complete coverage via roller MV records, some standardization of key spatial reporting parameters is recommended. These key parameters include the spatial reporting resolutions (∆x, ∆y) and the spatial av- eraging windows (x MV , y MV ). For all current roller MVs, y MV is c h a p t e r 3 This chapter provides a detailed evaluation of several as- pects of roller measurement values (MVs), including the surface area reflected in individual MVs, spatial resolution in MV records, and uncertainty in roller MVs. Results from in- dependent evaluations of MVs and the relationship between different MVs are presented. The influence of vibration am- plitude and frequency, roller speed, and forward/reverse driv- ing mode on roller MVs is examined. Finally, the influence of soil heterogeneity on roller MVs is examined. 3.1 Roller MV Reporting Characteristics Roller MVs and their position (determined traditionally via wheel encoder and more recently via GPS) are provided as discrete spatial records. Figure 3.1 presents samples of single- lane MV data records from Dynapac (CMV D ), Case/Ammann (k s ), Sakai (CCV), and Bomag (E vib ) smooth drum vibratory rollers. Position data were collected using RTK differential GPS and converted to Cartesian x and y coordinates (x is the driving direction). Figure 3.1 illustrates the discrete nature of MV data as well as the difference in travel direction spatial resolution of each recording system. Roller MVs are computed based on drum sensor data and provided to the onboard PC at a frequency f MV (see Figure 3.2). The onboard PC also receives a stream of position data from the roller-mounted GPS receiver at a frequency f GPS . These two data streams are often provided at different frequencies (i.e., f MV ≠ f GPS ) and must be merged by the manufacturer’s software. The merged MV and position data are reported via PC graphical view or data export at the frequency f report . The resulting spatial resolution of MVs (∆x in the direction of roller travel in Figure 3.2) is a function of f report and roller speed v. Values for f MV , f GPS , f report , and ∆x for the rollers evalu- ated are provided in Table 3.1. The methodology for merging and reporting MVs with GPS-determined positions varies across manufacturers and will clearly evolve in the future.

  Figure 3.1. Sample roller MV data records for (a) Dynapac CMVD , (b) Case/Ammann ks , (c) Sakai CCV, and (d) Bomag Evib over 10 m [32.8 ft; from various test beds (TBs)]. Figure 3.2. Schematic of roller variables.

0 the drum length [2.1 m (6.9 ft) for the rollers used here], and Δy is a function of the overlap between adjacent roller passes. A minimum overlap of 0.1 m (0.3 ft) is recommended. Roller MVs will likely evolve to capture soil heterogeneity along the length of the drum (see Section 3.6) and estimate soil prop- erties within the drum length (i.e., y MV and ∆y less than the drum length). Recommended values of y MV and ∆y should be modified accordingly. Together with the desire to provide complete (full) cov- erage and independent MVs (x MV = ∆x), the recommended reporting parameters x MV and ∆x are influenced by position accuracy provided by GPS. Table 3.2 summarizes the hori- zontal accuracy (x error ) and vertical accuracy (z error ) of avail- able GPS. RTK differential GPS will likely become standard practice for highway-scale projects and was used throughout this study. The recommended x MV should be an order of mag- nitude greater than the error (x MV ≥ 10 x error ). It follows that x MV and ∆x need not be less than 0.25 m (0.82 ft) when using RTK differential GPS with an error of 2.5 cm (0.08 ft). 3.2 Roller MV Position Reporting Error Documenting the correct position of the drum and thus the location of underlying material reflected in each roller MV is important for implementation of QA specifications that re- quire local comparison of roller MV with spot-test measure- ments or pass-to-pass comparisons of MV data (see Facas & table 3.1. Summary of roller MV reporting characteristics. Roller MV f mv (Hz) f GPS (Hz) f report (Hz) t MV (s) x MV a (m) (ft) ∆xa (m) (ft) Case/Ammann k s ~30b 1 1 1.0 1.0 (3.3) 1.0 (3.3) Bomag E vib 3 10 10 0.1 0.06 (0.2) 0.33 (1.1) Dynapac CMV D 1–5 10 1–4 1.0 1.0 (3.3) 0.2–1.0 (0.7–3.3) Sakai CCV 5 10 10 1.0 1.0 (3.3) 0.2 (0.7) aAssuming roller speed = 1 m/s (3.3 ft/s). bBased on operating frequency, here 30 Hz. table 3.2. Levels of GpS position accuracy. GPS Type Horizontal Accuracy,a cm (in) Vertical Accuracy,a cm (in) Real-time kinematic (RTK) 2.5 (0.98) + 2 ppmb 3.7 (1.46) + 2 ppmb Satellite differential GPS (subscription) 10–78 (3.94–30.71) 10–78 (3.94–30.71) Satellite differential GPS 95 (37.4) 95 (37.40) a Two standard deviations; may vary across GPS units. b ppm = parts per million, regarding distance from base station to receiver [e.g., for base station 1 km (0.62 mi) from receiver, 2 ppm = 0.1 cm (0.039 in); for base station 10 km (6.2 mi) from receiver, 2 ppm = 1 cm (0.39 in)]. Figure 3.3. Continuity of roller MV data (from various TBs).

  Mooney 2010). The first of two sources of MV position error results from the physical offset of the GPS receiver from the drum center (see Figure 3.4). The software for most rollers accounts for this GPS offset, and in some cases the offset is zero (e.g., when the GPS receiver is mounted directly above the drum center). The correct reporting position should be verified on project sites by comparing roller GPS-provided position with independently measured drum position using similar accuracy GPS (e.g., RTK handheld unit). This is re- flected in the recommended continuous compaction control (CCC) specifications (Section 7.5.3). The second source of MV position error occurs while the roller is moving and stems from data averaging during the calculation of each roller MV (i.e., t MV and x MV ; see Facas & Mooney 2010). Due to the way in which roller MVs and GPS position data are merged (Section 3.1), the resulting MVs are often associated with a position that reflects the end of the t MV and x MV . This leads to position reporting error. The posi- tion of each roller MV should be reported at the center of the Figure 3.4. Schematic of roller, GPS receiver offset, and spatial window for determination of roller MVs. averaging window. Any latency in the MV calculation prior to merging with GPS position would cause additional position error. This source of position error is roller speed and direc- tion dependent. Roller position error due to physical offset of the GPS receiver from the drum was checked for each roller on each field site and accounted for in the data analysis. To investi- gate the accuracy of MV position reporting during roller travel, wood beams were placed at known locations to create an artificial spike in roller MV data. Two tests beds were con- structed similarly on a 60-m (200-ft)-long by one-lane-wide fully compacted soil. Two wood beams were placed across the lane approximately 10 m (33 ft) apart (see Figure 3.5) to introduce spikes in roller MV data at known locations (measured with GPS). The rollers were driven in the north direction with forward roller travel (north forward, NF) and then in the south direction with forward roller travel (south forward, SF). This was repeated three or four times with each roller. Figure 3.5. Position error observed in CMVD data records (TB FL18): (a) observed MV records with no correction; (b) MV position corrected by -0.70 m (2.3 ft).

 table 3.3. observed position reporting errors for roller MVs. MV Position Reporting Error (m) CMV D +0.70 k s -0.40a CCV +0.55 E vib +0.80 a ± 0.5 m uncertainty. Figure 3.5 illustrates the CMV D response from all NF and SF passes. The solid black lines represent the location of the wood beams. Four passes in each direction demonstrate re- peatable behavior within NF and SF passes. The difference in roller MVs between NF and SF passes is due to transverse heterogeneity and is discussed in Section 3.6. As shown in Figure 3.5a, the location of each dip in the recorded MV data is direction dependent and is different from the locations of the wood posts by +0.7 m (2.3 ft) in the direction of roller travel. By applying a position correction of –0.7 m (–2.3 ft) to the MV data with respect to the direction of roller travel, the resulting NF and SF records align (Figure 3.5b). The –0.7- m position correction is consistent with the spatial shift re- quired to report CMV D at the center of its averaging window x MV . Recall that t MV = 1 s for CMV D . With the observed v = 1.2 m/s (3.9 ft/s) and a required averaging window correction of -0.5 s, the position correction due to averaging should be -0.60 m (-2.0 ft). The 0.10-m (0.33-ft) discrepancy may be attributed to GPS error and/or potential latency in the MV calculation. This method was applied to the other rollers (see Table 3.3 and Appendix B). In practice these offsets should be determined in the field and accounted for as appropriate. 3.3 Repeatability of Roller Measurement Values Repeatability tests were performed to quantify the pass- to-pass uncertainty in roller MV records and to verify the proper working condition of the roller measurement systems. An example of smooth drum vibratory roller MV data from consecutive roller passes with similar operating parameters (direction, speed, amplitude, etc.) is presented in Figure 3.6 (other data are presented in Appendix B). The pairs of MV data from each pass are visually repeatable (e.g., the various peaks and valleys in MV are consistently captured in both passes). Mild increases or decreases in roller MVs are attrib- uted to minor compaction or loosening and to minor devia- tions in driving path. To quantify the repeatability, the percent difference in pass- to-pass roller MV (%∆MV i ) at each spatial location was deter- mined via Equation 3.1 and is plotted in Figure 3.6b. In Equa- tion 3.1, MV i represents the MV data for pass i, and %∆MV i represents the percent difference between pass i and i – 1. Each pass of MV data was placed on a similar x-direction grid using linear interpolation to enable the spatial quantification. % – – – ∆MV MV MV MVi i i i = ×1 1 100 (3.1) The mean and standard deviations of the %∆MV values (µ %∆MV and σ%∆MV) are shown as a solid line and dashed line, respectively. The µ %∆MV reflects the effect of minor compac- tion, loosening, and roller track deviation. Observed µ %∆MV ranged from 0.1% to 3.6% for the testing performed here. Minor changes in pass-to-pass MV are unavoidable in prac- tice. The σ %∆MV provides a quantifiable measure of repeatabil- ity and pass-to-pass uncertainty. This quantity may be used to Figure 3.6. Percent difference in CMVD (smooth drum) determined via repeatability testing (TB FL18).

  verify the proper working condition of a roller measurement system and should be considered when interpreting pass-to- pass changes in MVs commonly used in CCC specifications (see Sections 7.5.3 and 7.7.2). Smooth drum vibratory roller MV records were generally found to be repeatable, and val- ues of σ %∆MV typically ranged from 5% to 7%. Data records for pad foot vibratory roller MVs collected during this study exhibited a high degree of scatter and general lack of repeat- ability. Values of σ %∆MV for pad foot vibratory roller MVs were found to exceed 25%. Repeatability testing plays an important role in verifying the proper working condition of a vibratory roller and/or roller measurement system. Figures 3.7 and 3.8 provide two examples where repeatability testing revealed a problem. In Figure 3.7, sporadic shifts in the roller MV data are visually evident and result in σ %∆MV = 14%. In Figure 3.8, roller MV data from a vibratory pad foot roller revealed a lack of repeat- ability (σ %∆MV = 27%). In both cases the roller MV system should not be used in roller-integrated CCC quality assur- ance (QA). To this end, repeatability testing is recommended and detailed in Section 7.5.3. A suggested criterion is σ %∆MV ≤ 10%, though the engineer of record may approve the roller measurement system based on a qualitative evaluation of the repeatability data. Figure 3.7. Percentage difference in MV (smooth drum) due to faulty roller measurement system. Figure 3.8. Percentage difference in MV (vibratory pad foot) determined via repeatability testing.

 The repeatability tests described here provide useful in- sight into the limitations of roller MV data in QA specifi- cations, particularly the spatial comparison of pass-to-pass roller MV. As described in Section 2.3.2, one Austrian and ISSMGE specification approach for roller-integrated CCC ex- amines the pass-to-pass change in mean MV (averaged over the evaluation section). A natural extension of this approach afforded by GPS is to introduce a spatial percentage change criterion (i.e., the percentage change in local roller MV—sin- gle MV) must be less than a specified percent. Such a criterion would address localized weak zones that may be overlooked in the averaging approach. The results of repeatability testing illustrate that the uncertainty (one standard deviation) alone in point-based percentage difference is 5% to 7%. To this end, the criterion for pass-to-pass spatial percent difference could not reasonably be less than 10% to 15% (two standard devia- tions) if based on single-point measurements. It follows that spatial averaging (e.g., via kriging) might be used to reduce this uncertainty and allow a smaller percentage difference cri- terion (e.g., 5%). 3.4 Comparison of Roller Measurement Values As described in Chapter 2, a number of roller measurement values are used in practice. Though different in methodol- ogy, each roller MV is based on the principle that measurable changes in roller vibration are due to changes in soil stiff- ness (and damping). This section examines roller MV trends via side-by-side comparison. The independent calculation of roller MVs using research team instrumentation also enables verification of roller MVs and how they work (i.e., dispelling the “black box” approach that tends to inhibit adoption by the engineering community). 3.4.1 Independent evaluation of mvs Sakai and Caterpillar smooth drum vibratory rollers were equipped with instrumentation to independently validate and assess roller MVs. Published algorithms to compute CMV, CCV, and k s (summarized in Chapter 2) were used. E vib was not calculated independently due to the significant differences in eccentric excitation technology and the differ- ences that vectoring the excitation force would have on roller MVs. Independently determined MVs are denoted with the subscript CSM (Colorado School of Mines) to distinguish them from the roller manufacturer MVs. The independently determined CCV CSM is compared with Sakai CCV in Figure 3.9, and the independently determined CMV CSM is compared with Caterpillar CMV C in Figure 3.10. The data match closely; subtle differences may be due to dif- ferences in how the systems average data and small differ- ences in sensor locations. Dynapac CMV D was compared with CMV CSM (determined with Sakai vibration data) by consecutively driving the two rollers in the same lane (see Figure 3.11). While minor dif- ferences will result from different roller properties, dis- crepancies in the values here are attributed to deviations Figure 3.9. Comparison of Sakai CCV and CCVCSM determined via independent instrumentation (TB FL12). Figure 3.10. Comparison of Caterpillar CMVC and CMVCSM determined via independent instrumentation (TB MN 29).

  in the tracks of the two rollers and differences in reporting methods. The Case/Ammann k s value was compared with the k s-CSM determined from the Sakai roller vibration data (see Fig- ure 3.12a). k s is calculated differently based on whether the drum is in contact or partial loss of contact operational mode (Anderegg & Kaufmann 2004). The two different algorithms were used for the independent assessment and are shown in Figure 3.12. The k s-CSM trends well with k s but does not match. The Ammann k s is scaled by a fixed factor and might be the reason for the discrepancy, though the occurrence of k s = 0 suggests possible errors within the measurement system. The k s-CSM was transformed using a linear equation (see Fig- ure 3.12b) to provide a good match with the manufacturer’s k s . Given this need to apply a scale factor, the black box sur- rounding k s partially remains. 3.4.2 relationship Between roller measurement values Figure 3.13 presents a comparison of E vib , k s , and CMV C data collected from a test bed with considerable stiffness variability (TB MN42). The soft areas reflect clay subgrade, and the stiff areas reflect >1 m (3.3 ft) thickness of crushed rock. Roller MVs were normalized by their peak values for direct comparison. In general, the roller MVs matched each other over the range of underlying stiffness present. Figure 3.14 shows the relation- ship between CMV D , CCV, and k s collected during operation on granular soil (TB FL18). Each data set was normalized by its peak value. These MVs also trend similarly. Figure 3.15 shows the relationship between k s-CSM , CMV CSM , and CCV CSM computed with data from independent instru- mentation on the Sakai roller. The k s-CSM values were com- Figure 3.11. Comparison of Dynapac CMVD and CMVCSM determined via independent instrumentation on Sakai roller (TB FL18). Figure 3.12. Comparison of Case/Ammann ks and ks-CSM determined via independent instrumentation: (a) raw data; (b) scaled data: k*s-CSM contact = 0.94 *ks-CSM contact-19, k*s-CSM = 0.5*ks-CSM-7.2 (TB NC27).

 Figure 3.14. Comparison of Case/Ammann ks , Dynapac CMVD , and Sakai CCV data on granular soil (TB FL18). Figure 3.13. Observed relationship between Case/Ammann ks , Bomag Evib , and Caterpillar CMVC (all smooth drum) on a mixed material test bed (TB MN42). puted at ~30 Hz, whereas the CMV CSM and CCV CSM were computed at 1 Hz; therefore, differences within a meter are not considered. The biggest difference in trends can be ob- served from x = 18 to 25 m. The CMV and CCV MVs trend well with each other and with k s when the values are greater than approximately 8 to 10. The correlations between various roller MVs are shown in Figure 3.16. The diagonals reveal the individual roller MV histograms. As anticipated, CMV CSM and CCV CSM exhibit the strongest correlation. Assuming that k s is an effective measure of soil stiffness, the correlation plots reveal lack of sensitivity of both CCV CSM and CMV CSM for values less than approxi- Figure 3.15. Relationship between independently determined ks-CSM , CMVCSM , and CCVCSM (TB NC27).

  mately 10, underscoring the findings in the literature (see Chapter 2). For values above 10, CMV CSM and CCV CSM are approximately linearly related to k s and therefore provide ef- fective measures of soil stiffness. 3.5 Measurement Value Dependence on Machine Parameters Measurement systems are typically standardized; that is, the magnitude, rate, and procedure are established to be con- stant from test to test. With CCC and IC rollers, however, measurement occurs during roller operation, and roller op- eration parameters can vary considerably. The fact that roller MVs are influenced by roller operating parameters is reason- ably well known (as was described in Chapter 2). What is not clear is whether the influence of operational parameters on roller MVs is predictable and therefore can be allowed during roller-integrated CCC specifications. This section character- izes the influence that eccentric force amplitude, vibration frequency, roller speed, and direction have on roller MVs. 3.5.1 dependence of roller measurement values on eccentric excitation amplitude The Ammann smooth drum IC roller was operated on the previously discussed TB MN42 to evaluate the influence of theoretical amplitude A (per Equation 2.2). Figure 3.17a presents roller MV records for low, medium, and high am- plitudes. The nature of the roller k s dependence on A varies along the 50 m (164 ft) test bed. The dependence of k s on A was considerably more significant for x = 20 to 30 m than for x = 37 to 48 m [see percentage difference in mean roller MV Figure 3.16. Observed correlations between ks-CSM , CMVCSM , and CCVCSM (TB NC27). Figure 3.17. Influence of A on ks (smooth drum) on granular and fine-grained material (TB MN42). Percentage difference in mean roller MV shown for two regions: x = 20 to 30 m (left); x = 37 to 48 m (right).

 (%ΔMV) in Figures 3.17b and c]. Similar results were found for the other rollers and soils (see Appendix B). The depen- dence of roller MV on A is not simply MV dependent but also (as discussed later in Chapter 4) depends on the soil and layering present. The dependence of roller MVs on excitation frequency was examined with the Sakai and Ammann/Case rollers (see Fig- ures 3.18 and 3.19). The strong frequency dependence of CCV is related to partial loss of contact, similar to that described above. At f = 20 Hz the Sakai roller is close to resonance and exhibits a considerably higher degree of loss of contact than at f = 25 Hz. The influence of frequency on k s is more subtle yet still evident. 3.5.2 Influence of roller Speed on measurement values The influence of forward travel speed on roller MVs was investigated with numerous CCC and IC rollers. Shown below are the results of testing on fully compacted material using various rollers. In all cases, roller MV records from two forward travel speeds are compared. The E vib data shown in Figure 3.20 exhibit a mean decrease of 6.3% from 0.8 m/s (2.6 ft/s) to 1.5 m/s (4.9 ft/s). However, some areas reveal similar E vib for the two speeds; therefore, the results are inconclu- sive. Both Sakai CCV (see Figure 3.21) and Dynapac CMV D (see Figure 3.22) exhibit a decrease in value with increasing speed. At higher speed the vibration energy is spread over more soil, and therefore the degree of partial loss of contact is reduced. These results agree with the literature presented in Section 2.1.2. 3.5.3 Influence of forward and reverse driving directions on roller measurement values Most production compaction efforts occur in a forward/ reverse driving pattern. It follows that measurement would be convenient during both forward and reverse operation. The influence of driving mode was investigated on the Am- mann, Bomag, Dynapac, and Sakai rollers. Typical results are presented in Figure 3.23. The figure indicates that Case/Am- mann k s decreases by approximately 10% when driving in the reverse direction. The significant discrepancy for x = 15 to 30 m may be due to misalignment of the roller in the presence of heterogeneous soil (see Section 3.6). Results from other roller investigations are provided in Appendix B. The data pre- sented here and in Appendix B, though limited, suggest there are only subtle differences in forward versus reverse mode driving. Site-specific evaluation of the influence of driving mode on roller MV should be performed if roller MVs are to be considered in both directions. Figure 3.18. Influence of excitation frequency on Sakai CCV (TB FL6). Figure 3.19. Influence of excitation frequency on Case/Ammann ks (TB NC27).

  Figure 3.21. Influence of roller speed on Sakai CCV (f = 20 Hz) (TB F6). Figure 3.20. Influence of roller speed on Bomag Evib (TB CO42). Figure 3.23. Influence of forward and reverse driving modes on ks (smooth drum; TB MN17). Percentage differ- ence (forward/reverse) is shown in plot b. Figure 3.22. Influence of roller speed on Dynapac CMVD (TB CO33).

0 3.6 Influence of Transverse Soil Heterogeneity on Roller Measurement Values Soil density, moisture, and stiffness vary locally. A review of the roller MV records presented in this chapter illustrates that soil properties vary, sometimes considerably, within a length scale of tenths of a meter (one-third of a foot). With a report- ing resolution as low as 0.1 m (0.3 ft), roller MVs are capable of capturing the variability in the x direction. In the trans- verse direction, however, the 2.1-m (6.9-ft)-long rigid drum coupled with current instrumentation and MV algorithms are unable to reflect y-direction (transverse) heterogeneity. This section addresses the influence that transverse soil het- erogeneity can have on roller MVs and, in turn, the limitation of roller MVs as a result of transverse heterogeneity. A 100-m (330-ft)-long fully compacted lane [2.1 m (6.9 ft)] of clayey silt was traversed in the forward north and for- ward south travel directions. Figure 3.24 shows the result- ing E vib and CMV D data records. The accelerometers on the Bomag roller from which the MV is derived are housed on the drum end, as depicted in Figure 3.24 (the driver’s left). The accelerometer location for the Dynapac roller is opposite (the driver’s right). This 100-m (330-ft)-long stretch of subgrade is clearly heterogeneous in the x direction. Figure 3.24 shows that the bidirectional roller MV records are different for al- most 40% of the 100-m (330-ft) length. The directional dependence of roller MVs was investi- gated by performing light weight deflectometer (LWD) tests across the drum lane at the locations in Figure 3.24 identified with arrows. The LWD results are presented in Figure 3.25, along with the roller MVs reported at the posi- tion of the accelerometer. At each location eight LWD tests were performed across the lane. The LWD results explain the directional dependence observed in the roller MVs. In Figure 3.25a, LWD data clearly show the lane is stiffer on the west side. This is corroborated by the higher roller MVs when the sensor is in the west (W) position. In Figure 3.25b the roller MVs are relatively equal in both directions, yet the LWD data are quite heterogeneous. The heterogeneity is symmetrical, however, and therefore leads to symmetry in the roller MV data. The directional dependence of roller MVs in the presence of local heterogeneity has implications for the use of roller- integrated CCC and the implementation of specifications. To best relate roller MVs to spot-test measurements, multiple spot tests should be performed across the drum lane. The implication of directional dependence suggests that if pass- to-pass analysis is to be performed, consecutive passes must follow similar paths. Figure 3.24. Observed differences in directional roller MV (TB MD20) due to soil heterogeneity: (a) Bomag roller Evib , (b) Dynapac roller CMVD .

  3.7 Conclusions and Recommendations The following conclusions can be drawn from the results presented in this chapter: • Each vibration-based roller MV investigated—Ammann and Case/Ammann k s , Bomag E vib , Dynapac CMV D , and Sakai CCV—is a reflection of soil stiffness over a spatial distance that varies across MVs [0.06 to 1.0 m (0.2 to 3.3 ft) observed]. The reporting resolution of roller MVs in the direction of roller travel varied from 0.2 to 1.0 m (0.66 to 3.3 ft). To ensure full coverage with roller-integrated CCC, the spatial distance over which a single roller MV is re- ported should equal the reporting resolution. The report- ing spatial resolution should be no less than 10 times the GPS accuracy; for example, if RTK differential GPS is used (recommended), the spatial resolution should be no less than 0.25 m [0.82 ft; with GPS accuracy = 2.5 cm (0.082 ft)]. • The GPS-based position reporting of roller MVs exhibited errors of 0.4 to 1.5 m (1.3 to 4.9 ft) for the rollers and mea- surement systems investigated. A position reporting verifi- cation procedure was developed and is recommended for roller-integrated CCC specifications (see Chapter 7). • Repeatability testing of properly working CCC/intelligent compaction rollers and roller measurement systems re- vealed an uncertainty of ±10% (one standard deviation); that is, a repeated pass over the same area will yield in- dividual roller MVs within ±10%. A repeatability testing procedure was successful in identifying when a roller or roller measurement system was faulty. Repeatability test- ing of pad foot measurement systems revealed single MV uncertainties in excess of ±25%. The repeatability testing procedure outlined in this chapter is recommended for CCC specifications and is described further in Chapter 7. • Roller MVs are influenced by the magnitude of eccentric force (or theoretical drum vibration amplitude A). From low to high A vibration on the same material, roller MVs were found to vary by as much as 100%. The amplitude dependence of roller MVs was not determinate and not predictable. Roller MVs were found to increase, decrease, or remain the same with increasing A depending on the soil and layering conditions. The nature of A dependence of roller MVs is discussed from a perspective of soil me- chanics in Chapter 4. Owing to the unpredictability in A dependence, a fixed amplitude is recommended for roller- integrated CCC. • CCV and CMV D were found to decrease considerably with increased roller speed. The influence of roller speed on E vib was inconclusive within the uncertainty of the measure- ment approach and the limited data collected. These find- ings suggest that roller speed should be fixed when using at least CCV and CMV D during QA. • Roller MVs were found to be mildly dependent on forward versus reverse driving modes. Roller MVs differed by 2% to 13% in forward versus reverse driving mode. Given that typical compaction work involves forward/reverse driving Figure 3.25. Comparison of transverse ELWD profiles with bidirectional roller MVs (TB MD20); roller MVs are de- picted at the sensor offset locations.

 sequences, there is considerable benefit to using roller MVs in both forward and reverse modes. Forward and reverse mode measurement should be considered; however, site- specific calibration is required to characterize and verify the relationship between forward-measuring and reverse- measuring roller MVs. • The vibration-based roller MVs investigated—Ammann and Case/Ammann k s , Bomag E vib , Dynapac CMV D , and Sakai CCV—correlate well with each other over a range of soft to stiff soil conditions. CCV and CMV were found to be insensitive to soil stiffness below values of approxi- mately 8 to 10, as is consistent with the literature. • Many of the roller MVs employed by manufacturers were validated using independent instrumentation and imple- mentation of published roller MV algorithms. This dispels the “black box” mentality that would inhibit implementa- tion within the engineering community. • Local soil heterogeneity transverse to the direction of roller travel has a significant influence on roller MVs. Owing to the nature of drum instrumentation, roller MVs are direction- ally dependent on heterogeneous soil. Roller MVs collected by traversing in opposite directions were found to vary by 100% due to transverse soil stiffness variability. This was confirmed by LWD testing across the drum lanes. As a re- sult, spot testing should be conducted across the drum lane when correlating to roller MVs, and great care should be used when performing spatial statistical analysis of pass-to- pass data maps in the presence of heterogeneity.