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.

APPENDIX H MULTI-LA13 VALIDATION STUDY . H-1

APPENDIX H: MULTI-LAB VALIDATION STI)DY INTRODUCTION A multi-laboratory validation study was conducted as a part of the NCHRP I-28 study. The primary purpose of the multi-lab validation program was to evaluate the proposed resilient modulus calibration and test methods, given in NCHRP Interim Report No. 2 [Ho-, for asphalt concrete, aggregate base and subgrade soils. Another benefit from the validation phase of the study was a preliminary examination of the variation in resilient moduli obtained between laboratories for similar materials. Because of tone and budget limitations, a full round-robin multi- laboratory investigation, as described In ASTM E691-92 [H-2], was not conducted. A full round- robin study would have involved 30 or more laboratories. Instead a more modest program was undertaken involving a total of 10 laboratories. ASPHALT CONCRETE Introduction The validation study for asphaltic mixtures was composed of (~) calibration of the testing system using Free synthetic specimens and (2) resilient modulus testing of laboratory compacted specimens provided by the reference laboratory. The two synthetic specimens had resilient moduli of 145,000 psi and 556,000 psi. Both the MTS loading device and the new measurement system developed in this study were used. The following is a summary of the results from the three laboratories participating In the study. These laboratories are designated as Lab I, Lab 2 and Lab 3. Lab ~ is considered as the reference lab since they had considerably more experience performing diametral tests than the other labs. Analysis of Results Figure H-! shows the graphical comparison of the mean values of resilient moduli determined from the three laboratories. MR! and MR2 were obtained using calculated and assumed Poisson's ratios, respectively. The use of the measured Poisson's ratio results in greater mean resilient moduli values than those calculated using an assumed Poisson's ratio. A statistical analysis was performed on the resilient modulus data, and Heir coefficient of variation are plotted in Figure H-2. As shown on these two figures, the moduli determined from the calculated Poisson's ratios show greater differences between the laboratories than those from the assumed Poisson's ratio. Also, when the calculated Poisson's ratios were used in the computation, the coefficients of variation from Lab 2 and Lab 3 are usually higher than those from Lab ~ (the reference laboratory). These observations suggest that although the recommended testing system and protocol yields better estimates of Poisson's ratio, the use of assumed Poisson's ratios gives more consistent moduli. Figures H-1 and H-2 also show that He difference in resilient moduli for calculated versus assumed Poisson's ratios is much smaller for the Lab ~ data than the Lab 2 and Lab 3 data. This trend is probably due to the fact that the Lab ~ operator is much more experienced In the resilient modulus testing of asphalt concrete using the proposed testing system. H-2

5.0E+06 ~4.0E+06 cn ~ _ o' ~ ~ 3.0E+06 Cal o 2.06~06 J In 1.0E+06 O.OE+OG 5.0E~6 - . ~4.0E+06 a, _ , 3.0E+06 o 2.0E+06 u, Al 1.0E+06 O.OE+OO BASED ON MEASURED POISONS RATIO _ _' _] D _ Lab 1 Lab2 Labs ~ 1 104 41 77 TEMPERATURE (F) BASED ON ASSUMED POISONS RATIO 41 Lab 1 Lab 2 Lab 3 h: , _: . 77 TEMPERATURE (F) 104 Figure H- 1. Comparl8On of Resilient ~o~lul1 for Different Labs ~-3

1.0 z 0.8 o - cr: ~S 0.6 IL a `, 0.4 u. 80.2 Q.0 1.0 Z 0.8 - ~ : a: ' ~ 0.6 o <, 0.4 IL O 0.2 0.0 _ . RESILIENT MODULUS BASED ON TESTED POISSONS RATIO Or .. . ~ .~ ~ ,,ss ...~s _~ ~ . ..... . . ... .: ' ''" , ~ Lab 1 _~ O_ Lab 2 Labs 41 77 TEIUPERATURE (F) 104 RESILIENT MODULUS BASED ON ASSUMED POISSONS RATIO Lab 1 rum Lab 2 1 - -] it_ - '1 Labs . I-..,'''-. 1 1 - t:C;Xi~R 104 41 77 TEMPERATURE (F) Figure H - 2. Comparison of Coofflcients of Variation for Different [08 H-4

Mean values of Poisson's ratio obtained from the three laboratories are given in Figure H-3. In addition, their coefficients of variation are also plotted in Figure H-3. Lab ~ has the smallest coefficient of variation among the three laboratories, again ProbablY due to the greater level of experience of the Lab ~ technician. r o Tables H-~(a), H-~(b) and H-~(c) provides a detailed summary of He resilient modulus and Poisson's ratio for each lab at the temperatures of 41°F, 77°F and 104°F, respectively. Their mean values, standard deviations and coefficients of variation are all given. Table H-2 gives a summary of mean resilient modulus, Poisson's ratio and coefficient of variation for each lab at three temperatures. hn general, the data from Lab 2 and Lab 3 show greater variation than those from Lab 1. Regardless of the laboratory and the type of Poisson's ratio used, the coefficient of variation at 104°F was the greatest, suggesting that greater caution has to be exercised when interpreting the MR values determined from the indirect tensile test at high temperatures. It is important to recognize in Table H-2 that the coefficient of variation in the Lab ~ data was not affected much by using the calculated Poisson's ratios instead of the assumed values. However, greater differences can be found with the Lab 2 and Lab 3 data. This finding suggests that an experienced operator can produce equally or even more consistent MR values using Poisson's ratios calculated from measured displacements. All the observations made earlier from the graphical comparison were supported by the statistical analysis. Standard deviation of the MR] values, based on calculated Possion's ratio values, was usually greater Han that of He MR2 values, based on assumed Poisson's ratios. Again suggesting the importance of the operator's experience on obtaining consistent resilient moduli values. Statistical Comparisons. The findings from the graphical comparison were further checked by performing the analysis of variance (ANOVA) tests with generalized linear models. The ANOVA test procedure employs the F-value as the test static to test the null hypothesis that the resilient modulus values at different temperatures are the same. The level of significance (p-value) for this test is the probability of having F-value larger than the calculated F-vatue from a data set for the ~ . . .. . .. . ~ , . . . . . . . factor in question. A smaller value ot tills probablllty unpiles the heavier weight of the sample evidence for rejecting the null hypothesis, and thus, in this investigation, more likely that the resilient moduli are different. A typical criterion of using a critical p-value of 0.05 was employed in this study. Table H-3 summarizes the statistical results on the possible effect of operator-to-operator variation, as well as the effect of fresh and re-tested specimens. Specimens TI, T2,..., T6 were tested first In Lab I. Then, specimens TI, T2 and T3 were tested again at Lab 2, and specimens T4, T5, T6 at Lab 3. Table H-3(a) compares results obtained by Lab ~ with Lab 2 and Lab 3. For the same group of specimens, the p-values of resilient moduli obtained from the assumed Poisson's ratios are usually higher than those from the calculated Poisson's ratios, which indicates that the calculated Poisson's ratio signifies the lab-to-lab variation. Tables H-3(b) and H-3(c) compare the results between fresh specimens (i.e., TI, T2, .... T6 at Lab ~ vs. TI, T2, ..., T6 at Lab 2 and Lab 3, respectively. These tables, indicate that p- values are higher for fresh specimens than revested specimens. This implies possible damage of the re-tested specimens regardless of the variation between operators. Therefore, structural changes during the proposed MR tests are more significant than the lab-to-lab variation. The statistical analysis was performed on a very limited number of samples, making the interpretation of the statistical results less reliable. An extensive validation study is recommended to obtain better evaluations on sample-to-sample and lab-to-lab variations. H-S

1.5 o 1 a: In he o In o ~0.5 o - 1.0 0.8 o - - <0.6 o He ~0.4 - ~L 00.2 0.0 Labs labs Labs 1 _e_ 41 77 TEMPERATURE (F) 104 POISONS RATIO __ - Id:= ... '."'." : ~ . . ~ ,..2.',.:... ~ .'.2.,,:.'...2 ~ it: eel 1 i1 ~ 41 77 TEIYIPERATURE (F) Lab1 Lab 2 Cod Lab 3 104 Figure H - 3. Co~arison of poisson's Ratios for Different Lobe H-6

Table H-~. Comparison of Resilient Moduli and Poisson's Ratio for the Different Labs (a) 4IoF Spec~nen Tl T2 T3 T4 TS T6 Ul ~2 U3 U4 US U6 Avcragc Std. dev. Coef. Var. Speci~ er T1 -T2 T3 T4 T5 T6 U1 U2 U3 U4 U5 U6 Ave~gc Std. dev. Coef. Var. Specimcn T1 T2 T3 T4 TS T6 Ul U2 . U3 . U4 US U6 Averagc Std. dcv. Coef. Var. Poi~n's Ratio (c) 1040 F , Poisson's Ratio Lab 1 ~ Lab 2 ~ Lab 3 Lab 1 0.72 0.18 2.94E+S . . 0.68 0.39 4.69E+5 . 0.56 0.28 2.43E+S 0.70 0.86 2.79E+S 0.56 1.02 392E+5 0.57 ~1.4S 2.28E+5 0.41 0.30 . . 0.37 1.29 1.11 1.15 0.63 1.15 0.32 3.18E+S 0.07 0.19 0.08 8.56E+4 . 0.11 O.t6 0.24 0.27 _ Resilient Modulus (~t') Lab 1 2.02E+6 2.13E+6 2.11E+6 2.00E+6 2.21E+6 2.25E+6 2.12E+6 9.08E+4 0.04 : (b) 770 F Lab2 4.26E+6 4.78E+6 2.61E+6 8.63E+6 2.68E+6 1.25E+6 4.04E+6 2.36E+6 0.58 Lab3 1.62E+6 1.13E+6 1.83E+6 . 1.39E+6 9.32E+S 1.19E+6 1.35E+6 3.03E+S 0.22 Resilicnt Modulus (M, ) Lab 2 . 1.31E+6 2.30E+6 1.26E+6 1.96~+6 1.23E+6 l.l9E+6 1.54E+6 4.30E+S 0.28 Lab3 6.22E+5 5.67E+S 5.64E+5 6.62E+5 5.28E+5 5.29E+S S.79E+S 4.88E+4 0.08 Resilient Modulus (~t') _ Lab 2 3.50E+S 8.07E+S 6.18E+S 5.26E+5 S.4SE+5 4.67E+5 S.52E+S 1.40E+S 0.25 . H-7 . Lab3 1.85E+5 5.03E+5 3.97E+S 2.58E+S 1.71E+S 1.63E+5 2.80E+S 1.28E+5 0.46 Lab I _ 1.44E+6 1.55E+6 1.42E+6 1.57E+6 I.85E+6 1.59E+6 1.57E+6 1.40E+S 0.09 - ~;~ Lab 2 | Lab 3 1 1 l.41E+6 1 1 1.45E+6 T l.25E+6 1 1.94E+6 1 . _- 1 - 1.64E+6 1.S4E+6 | ~3iE~ 11 1.13E+6 ~ 1 1.23E+6 1.93E+6 1.65E+6 1 - 1 1 2.09E+6 1 1 1.BOE+6 1 1.31E+6 1 1 1 2.00E+S 1 1.13E+5 1 011 1 o.og Rcsil~ent Modulus OM`~) | Lab I I Lab2 I Lab3 1 1 1 1 8.17E+5 1 1 6.71 E+S 1 1 1 ~ ~.1 IE+S~ 7.95E+S 1 T 5.8BE+~ 7.15E+5 1 5.20E+S 1 1 1 1 . ~ 8.91E+5 1 9.70E+5 1 1 ~1 -~ 8.47E+S 1 5.21 E+S 1 | 1 6.17E+S I 1 5.48E+5 1 1 5.84E+5 1 ~ - 1 6.08E+S ~1 5.30E+5 T | S.80E+S 1 1 8.23E+5 1 6.21 E+S 1 6.20E+S ~1 1 5~82E+4 1 1.59E+S I 5.53E+4 1 1 1 L 007 1 026 1 o.og Resilient Modulus (M' ) Labl 2.32E+5 3.83E+5 1 2.26E+5 2.21E+5 3.63E+5 2.09E+5 l l 2.72E+S 7.17E+4 0.26 _ Lab 2__~ 1 1 2.40E+5 4.91F,+S 2.75E+5~ 1 1 2.60Eff 3.06E+5 2.54E+S 3.04E+5 8.60E+4 0.28 Lab3 | 3.16E+S 5.87E+S 5.60 1 l 2.61 E+S~ 1 2.33E+5 1 . ~ 1.96E+5 1 1 1 1 _3.64E+S - | 1.53E+S 0.42

:> 1 o o a] ·= cd o v ~ - - cO - D Ct 1 ._ - D 1;~4 hi. :~, _' .~ ~ _ ~ ~ , ~ O ~ ~ ~ of, O O O _ ~ _ oo _ O _ O by, _ ~" ~ O O 00 ~ _ , ~ ~ ~ ~ O ~ m. O 00 00 0\ 1` . ~ ~ ~ ~ ~ O O O O _ _ -, O O O O ~ ~ [_ ._ '5 ~ ~ ~ _ _ ~ ~1 _ _ _ _ _ o .~ I: o U. ¢ o I: - ._ ~o., .o o - ~: ._ ~ o C~ 5 w C~ Ct H-S .o ·e o ·" o v ~ o, _ 0 0 _ ~ oo ~ ~ oo oo ~ ~ ~ - " o ~ ~ o ~ o ~ ~ ~ ~ o o o o o o o o o ~ o o ~ oo ~ - ~ ~ ~ - ~ ~ o - ~ - - ~ o o o o o o o o o - - - - - - - - - - - + + + + + + + + + . ~ LO ~ -. ~ - ~ ~ o -. o ~ ~ - - oo ~ ~ ~ ~ ~ -- - - - - - So k + + + + + + + + + -" ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ o ~ ~ x ~ o - o ~ o ~ ~ - ~ x ~ ~ - - - ~ ~ ~ ~ s o ~ o o o ~ ~ o =. - o - -- - - o --~ ~ -~ ~ ~ ~ -= ~ ~ ~ ~ ~ ~ -t - ~ -- - 5ex5 ~ - a~ so oo o ~ ~ o o o o~ - - - - ~ ~ - 2 E ~ . o~ o -

UNSTABILIZED AGGREGATE BASE EXPERIMENT Introduction The aggregate base multi-laboratory resilient modulus validation testing program involved 4 different laboratories. A fifth laboratory was to have participated in the experiment, but their testing equipment could not be set up in time. The validation study consisted of two parts: (1) system calibration, including using aluminum and polymer specimens, and (2) the testing of 3 different aggregate base materials. The specimen preparation and testing procedures used in the experiment are given in NCHRP 1-28 Interim Report No. 2. The tentative test procedure required the use of two EVDTs mounted outside He biaxial cell to measure axial deformation. The testing procedures, except as noted, were reasonably similar to those given in the SHRP P46 (November, 1989 procedure). The resilient modulus mode} used in the multi-laboratory study is as follows: M =KI ~K2 oK3 R d where MR _ _ _ K1,K2,K3 = resilient modulus = bulk stress (~ 02 03) = deviator stress = model constants (H-1) Mode! constants were evaluated by multiple regression analysis techniques. The statistical analyses employed in this study were performed using the general techniques described in Chapter 2 for the aggregate base study. Materials and Notation Base Materials. The 3 different nonplastic aggregate materials tested in the multi-laboratory study are described as follow: Base Notation Description Base ~ B! Base 2 B2 Base 3 C4 Well graded, nonplastic fines, I.5 in. maximum size, subangular to angular crushed stone. Four percent fines. Specimen compacted to 100 percent T-~80 density. Well graded, nonplastic, 1.25 in. maximum size, subangular to angular crushed stone. Ten percent fines. Specimen compacted to 100 percent T-~80 density. Well graded, nonplastic Minnesota DOT Class 4 base material with partial crushing of faces. H-9

Representative samples of aggregate for specimen preparation were thoroughly mixed to a homogenous condition, and then split into the required quantities following accepted sampling practices. Sufficient material for two extra labs were prepared at the same time and held in reserve. The samples sent to each laboratory were selected using a random number procedures. The samples were placed in 5 gal. buckets, labeled, sealed and shipped to each laboratory. System Calibration. Preliminary system calibration consisted of evaluating the testing system alignment and extraneous deformation in the testing system. Extraneous system deformation was evaluated using an aluminum cylinder 12 in. high and 6 in. diameter which simulated a specimen of essentially rigid material. Preliminary calibration procedure used is given in Appendix D. Final calibration was achieved using a stiff and a soft synthetic polymer specimen. These reference specimens were shipped to each lab. The two synthetic specimens used have a known, nominal resilient modulus of 8,000 psi and 50,000 psi. These specimens are the same synthetic specimens as used in the laboratory base study described in Chapter 2 of this study. The specimen having a stiffness of 50,000 psi is representative of an unbound aggregate base subjected to a relatively high bulk stress (~) of about 45 to 55 psi. The soft specimen with a resilient modulus of 8,000 psi is representative of a typical unbound aggregate base subjected to a low bulk stress of about ~ to 10 psi. Steel top and bottom plates were epoxy glued to the ends of the 50,000 psi synthetic specimen to avoid cupping of the ends with use of the specimen. Cupping of the end of the synthetic Type ~ base specimens used in the SHRP program was a serious problem. Unfortunatelv. One steel end cap was knocked off the specimen as it was moved between labs. _ _ ,, Notation. The notation described here is used on the tables and figures given in subsequent sections. The four laboratories included in the study are indicated as Ll, L2, L3 and L4. Lab ~ was used as the reference laboratory since good agreement In resilient moduli values was obtained from axial displacement measurements made using external EVDTs and inside clamp supported EVDTs. The three bases tested, which are described above, are designated as BI, B2 and C4. The two replicate specimens of the same material prepared and tested by the same lab are indicated by A and B. Therefore, the notation "DEBRA" on a figure or table indicates (Base 2 (i.e., B2) which was tested by Lab 2 (i.e., L2) and the results are for the "A" specimen. The notation "~3BlAB" indicates lab 3 (i.e., L3) tested Base ~ (i.e., Bl) and the data for both the A and B specimens have been combined and the results presented. Resilient Modulus Test Results Within Lab Reproducibility of Test Results. To perform reliable resilient modulus tests, the test results for a specific laboratory (and a single test apparatus and operator) must be reproducible for replicate specimens of the same material and the same method of specimen preparation. Figures H-4 and H-5 illustrate the typical variability observed between replicate specimens of the same material prepared and tested by the same lab. As shown on the figures, 3 different values of deviator stress were applied in the resilient modulus test for each of the 5 values of confining pressure used. The 95 % upper (UL`) and lower (Lid) confidence intervals for the replicate pair of tests are shown on the figure in addition to the actual test points for each of the two tests. The consistency comparisons for replicate specimens given in Table HA are for a 90% confidence level and are based on a comparison of the individual model constants KI, K2 and K3. H-10

1oo go 80 ~0 60 ~0 ~0 30 20 10 o = _ LEGEb D A: L1 B2A + L1 B2B o LID ~ UL : = _ _ _ , +x ~ _+~. . LL = Lamer 95 °~6 Confidence Umits UL = Upper 95°h Confidence Um~ts 0 20 40 60 80 100 120 Bulk Stress, ~ (psi) Figure H4. Companson of Resilient Moduli for Replicate Specimens- Lab do 3S ._ tax ~ 26 - o c - 30 20 16 10 o LEGEND S L2B1A ~ L2BiB -O- LL - ~ UL B ~' 4'. I' : o ALL 1 1 ~ C pit $. I. O l iaA~ . ~ \,rO~.~ ,0-'~ ,#0' ' O' U = Lower 95 % Confidence Umits UL = Upper 95% Confidenoo Umits _ 0 20 40 60 Bulk Stress, 8. (psi) 80 100 Figure H-S. Compar~son of Resilient Moduti for Replicate Specimens- Lab H-11 A: a3 - 3 PSI B.: CT3 ~ S PSI C: ~3 2 10 PSI D: CS3 - IS PSI E: O3 ~ 20 PSI A: CS3 = 3 PS] B: CT3 3 S PSI C: O3 = 10 PSI D:~3= lSPSI E: a3 - 20 PSI

The mode! constants between the two replicate specimens were found to be statistically the same for Labs I, 2 and 3 at both the 90 and 95 % confidence levels. Lab 4 had, for both confidence levels, statistically different intercepts (i.e., K! values) but the same slopes (i.e., K2 and K3 values) between replicates for both base B! and B2. A comparison of constants KI, K2 and K3 between replicates is statistically correct but quite rigoruos. If the resilient moduli for the replicate specimens from Lab 4 calculated using equation (H-~) are plotted, the results from the two tests are sufficiently close to the same for practical purposes. These findings indicate He resilient modulus test results using the same material, lab and test equipment are statistically reproducible following the proposed test procedure. Lab 4 did have some unexplained variation in equipment behavior or test procedure that influenced He test results. The square of the coefficient of variation (R2) typically varied for Labs I, 2 and 4 from 94 to 98 % for the combined replicate data as shown in Table H-4. These results indicate the Uzan resilient mode! used is valid which was also shown to be true in Chapter 2 for not only this mode! but also the UT-Austin and UTEP models. The results also indicate the testing technique and specimen preparation procedures of the labs are generally good. The resilient modulus mode! fit of the results for specimen B2 from Lab 3, however, exhibited a correlation coefficient squared of only 45%. This poor mode! correlation was apparently due to the use of grouted specimen ends which was not specified in the test procedure. The grouted ends probably resulted in the development of a variable pressure within the specimen greater than the required atmospheric pressure. The net effect of this excess pressure would be to reduce the beneficial effect of the applied confining pressure on resilient modulus. Indeed, the Lab 3 test results showed that the resilient modulus did not increase nearly as much with increasing conjuring pressure as it did for the other labs. Remember that the use of grouted specimen ends by Lab 3 violated the proposed specimen test procedure. Coefficient of Variation. Determining the typical range for different labs of the coefficient of variation (CV) between replicate specimens can serve as a useful too! in evaluating if a laboratory needs to consider further refinements in testing procedures. These refinements could include additional equipment calibration or operator training. The coefficients of variation calculated for the replicate specimen pairs for each lab are shown in Table H-5. These coefficients are based on He mode! constants K,, K2 and K3. The average coefficient of variation (CV) of these mode! constants is 19% with the range being from 10 to 22%. These values of CV omit the Lab 3 results, which used grouted specimen ends. Also Lab I/Lab 2 C4A data (CV = 28%) was omitted since it was for a replicate set involving two labs rather than for a single lab. With experience and careful equipment calibration if should be possible to reduce the average coefficient of variation from 19% down to about 10 to 14% which is about what was observed in the base course study described In Chapter 2. Experience developed during this study clearly shows much less variation exists in the resilient modulus calculated from a realistic mode! such as equation (~) than obtained considering the variations of the individual mode! constants. Table H-6 gives coefficients of variations based on the resilient module calculated using the resilient modulus mode! given in equation (Hal) for the three stress levels summarized in the footnote at the bottom of the table. The mode! constants (Kit, K2 and K3) used to calculate the CVs based on resilient modulus given in Table H-6 were the same as employed in developing the CV values presented in Table H-5. H-12

Table H4. Resilient Modulus Mode} Constants and Statistical Comparison of Replicate Specimens at 90% Confidence Level r ~ _ _ ~ C4mpatIson LAB SPECIMEN Rot __ K2 ~R2 MSE1 p1, p2, p3 ., . . L1B1A 2.511 ~.256 .918 1 98.10 0.02721 0.150>0-01 L1StB 3.107 ~.317 0.935 0.377~0.01 0.821~0.01 L1 . ~ L1B2A 2.971 4.421 1.006 98.30 0.0248 0.860>0.01 L1 B28 3.041 -0.298 0.903 0.059>0.01 0.30>0.01 T L2B1A 1 4.552 0110 0.361 1 9430 0.0326 1 0.689>.01 U81B 4.269 0.042 0.417 0.333>0.01 0.443>0.01 rUB2A 1 3.555 0.005 0.469 1 94.60 0.0319 1 0.662>0.01 L2B28 3,810 0.011 0.454 0.821>0.01 0.829>0.01 1 1 - 1 1 t3B2A T 11290 -0.101 0.330 1 45.00 0.0640 10.820~0.01 L382B 10.238 4.367 0.478 0.088~0.01 0.458~0.01 L481A 0.868 ~.523 1.505 98.50 0.0260 0.000<0.01 L4131B 1.815 ~.530 1.293 0.924~0.01 0.027~0.01 L4B2A 1.109 -0.714 1.514 98.40 0.0239 0.000~.01 L4B2B 2.300 4.660 1.301 0.469>0.01 0.016~0.01 . ~4 1 UC4A 1 2.590 -0.573 1.179 1 96.io 0.03901 0.020>0.01 L404A 1.422 ~.343 1.073 0.040>0.01 1 1 1 1°.395~0.01 Result - Simitar Regress. Lines simdaf Regress. Lines Similar Regress. Lines Similar Regress. Unes . Similar Regress. Lines Different Intercept Same Slopes Different Intercept Same Slopes Similar Regress. ines . Note 1: Notation: R2= Correlation Coeffiaent Squared MSE= Mean Square Error p1, p2, p3 Levei of Significance for Constants K1, K2 and K3 H-13

Table H 5. Indiv~<iUal Test MR Model Constants and Statistics With Coefficient of Variation Analysis Based on Mode} Constants LAB SPECIMEN . . 2.511 -0.256 O.918 3.107 -0.3t7 0.935 4~7 0.~7 0.422 0.043 0.012 _ 15.016 __ tS.12t tom 2871 ~.42t 1. 3.041 4.298 0.903 3.006 -0.360 0.955 0.049 0.087 0.073 1.647 24.210 7.~1 4.552 0.110 0.361 4. - 0.~2 0.417 4.410 0.076 0. - 0.200 0.048 0.040 4. ~.7" 10.= . 3.565 0.0 ~0.44;9 3.810 0.011 0.^ 3.682 0. ~0- - 0.180 0.005 0.011 4.897 60.2t ~2~7 t 1290 -0.101 0.330 10238 -0.367 0.478 10.7" -0. ~0. 0.744 0.188 0.105 8.815 80.381 25.8~;3 . _ . 0.868 4.523 1.505 1.915 no.= tom 1.392 4.527 1.399 0.740 0.005 0.150 53.186 1.037 10.736 1.109 -0.714 1.514 2300 4.860 1.301 1.704 -0.687 1.407 0.842 0.038 0.151 49.434 5.464 10.733 2.590 "o.m 1.17g 1.422 As. - 1.073 2.006 -0.458 1.126 0.826 0.163 0.075 41.1Bt 35.5= 6. - 22.06 ~35.716 8. - r A~.CY en) Stan (1) . R2 o _ 97.70 0.0310 98.60 0.0~8 98.15 0.0269 0.638 0.0058 0.648 - .oo o.om 98.70 0.0218 98.35 0.0247 0.495 0.~41 . 0.543 93.50 0.0359 95.50 0.0290 94.50 0.0324 1.41 ~0.~ 1.497 ~ 93.60 0.0036 96.00 o.m7' 94.80 0.0157 1.697 0.01 n 1.790 47.50 0.0513 38.30 0.0746. 41.90 0.~30 7.920 0.0164 1 8.901 - 99.20 0.0215 97.50 0.0= 98.35 0.0~6 1.202 O.OC" 1.m 89.50 0.0146 96.60 0.0005 98.15 0.0~6 1.909 0.01 13 1.945 _ 9420 0.0425 _ - .10 0.0Os1 85.15 O.OB88 1.344 0.0052 1.412 _ . _ 3.490 _ L1 L1B1A L1B18 AV - + cv~, _ 10.468 L1B2A L1B28 Av"+ CV(%) 11.1n U L2B1A ~B1B Aver cv~) 2B.187 L28" UB2B Aver~e cv{%) 22.483 L' LSB2A L3B2B Average CY(%J 37.71 9 u L481A L481B Average e' CY~) 21.6S3 - L482A L4828 AV6r~ cvt%) 21.sn L2U LX" L~A AYBr&96 (%) - 27.784 AV~ CV% ?2.118 N~e 1: N ~ ~ - ~ ~" MSEa hban Sqwe p1,p ~S-~sK1,~ o. S~d D~ .CV ~C - md~ of V~on H-14

Table H-6. Coefficient of Variation Analysis Of MR Results Based on Calculated MR Values-- Three Stress Levels LAB SPECIMEN Model Constants R7 ~K3 - 2.511 ~ 256 0.818 3.107 ~.317 0.935 Average a cv(%) 2.871 ~.421 1.006 3.041 -0.298 0.903 Average a CV(%) 4.552 0.110 0.361 4.269 0.042 0.417 Average cv(%) 3.555 0.005 0.469 3.810 0 011 0.454 Average cv(%) 11.290 ~.101 0.330 10.238 -0.367 0.478 Average Cow 0.868 -0.523 1.505 1.915 ~.530 1.283 Average CV(%) 1.109 -0.714 1.514 2.300 4.660 1.301 Average _ 2.590 4573 1.179 1.422 ~.343 1.073 Average a CV(%) _ . AV9. CV(% Resilient Modulus (1) ~. 11 111 26.770 30.534 48.327 30.348 33.361 51.190 28.559 31.948 48.758 2.530 1.998 Z025 8.858 6.257 4.069 28.728 30.098 45.147 27.890 30.846 46.gO7 28.359 30.472 46.027 0.522 0.529 1.244 1.840 1.736 2.703 . 18.716 22.803 31.597 17.969 21.279 29.236 18.342 22.041 30.417 0.528 1.078 1.669 2.879 4.8g2 5.487 16.272 19.116 26.549 16.862 19.801 27.339 16.567 19.458 2~.944 0.418 0.485 0.559 2.522 2.4g2 2.075 - 25.900 26.987 31.633 20.484 18.655 20.146 23.192 22.821 25.890 3.830 5.892 8.122 ~3 ~ 51T ~ w 33.110 38.211 75.551 36.213 38.733 65.694 34.661 38.487 70.623 2.194 0.347 6.970 6.330 0.903 9.869 28.050 28.470 49.581 33.064 32.403 so.4ss 30.557 30.437 50.045 3.546 2.781 0642 11.605 9.138 1.282 30.78e 30.769 46.821 20.388 23.059 38.235 25.587 26.~14 42.531 7.353 5.452 6.075 2B. 736 20.255 14.284 - 9.910 8.83B 8.893 AV9. CV(%) l. ll' ~nd III L1 B1A L1B18 L1 6.395 L1B2A L1 82B 2.093 ~B1A L2B1B L2 4.419 L28" UB2B 2.363 L3 UB2AL1 UB2BL1 24.568 L4B1A L4B1B L. 5.700 L482A L482B 7.342 L2/L. L2C4A L4C4A 21.092 9.247 Notes: 1. S~ L~ 1: s~ t~ Il: Stress Levet Ill: ~s' ~ · pel od-10-p ~"26 pd a3Jlpd ad~20pd "3Cp~ ~s3~10 pal e~d-O p ~~70 p~ 2 Notation: CV (96) = Coeffiaent of Vanation ~s = Standard DeYiation H-15

The average coefficient of variation (CV) for all three stress levels is equal to 5 % based on resilient modulus calculated using the resilient modulus model. This value of CV is only 26 % of the average value calculated considering the variation of the individual mode! constants (CV = 19%~. Stress levels ~ and IT (8 = 25 psi and 35 psi, respectively) used in Table H-6 are approximately representative of the average stress conditions within the base of typical moderate and strong pavement sections. For these two stress levels the coefficient of variation is still 5 % . The general finding that the CV variation of resilient modulus is much less than that for the CV based on He individual mode! constants is in agreement with the findings presented in Chapter 2 for unstabilized aggregate base. Between Lab Variation. Table H-7 shows the observed variation in resilient moduli between labs. Column ~ in this table gives the two labs being compared. For example "~/~2" in column ~ indicates the comparison is between Lab ~ and Lab 2. For the comparisons given in Table H-7, for 5 out of the 6 comparisons the individual mode] constants obtained by a given lab are statistically different than for the comparison reference lab testing the same material. Also, very large differences exist In the values of similar mode] constants for one lab compared with another (Table Hub. Figures H-6 through H-8 compare the resulting resilient moduli, which are the desired product of the test, for Lab ~ win test results for Labs 2, 3 and 4 for the 5 different values of confining pressure (03) used in the tests. The very important implication of this finding is that the use of axial displacement measuring devices outside the biaxial cell is not a practical technique for most laboratories. The main problem in using external measurement devices is the somewhat tedious and time consuming equipment calibration required to obtain reproducible results. Agreement between resilient moduli obtained using externally mounted EVDTs and clamps mounted on the specimen were obtained in the experunental investigation of base material reported in Chapter 2. This finding formed the basis for using externally mounted EVDTs in the round-robin study. General Findings The important findings from the multi-laboratory validation study, which in general reinforce the general findings of He overall aggregate base study, can be summarized as follows: Synthetic specimens of known resilient moduli are extremely useful and should be routinely used by any laboratory performing the resilient modulus test. Synthetic reference specimens that are shipped from one laboratory to another, however, take a tremendous beating and will very likely become an unreliable reference standard with repeated use. This finding was true for both the present round-robin study as well as the SHRP round-robin 6 in. diameter specimen base study. Therefore, synthetic specimenEs) should be purchased by each laboratory. The reference resilient modulus of the synthetic specimen should be verified periodically by making independent axial displacement measurement directly on the specimen. Accurate, independent measurements can, for example, be made using a pair of large gage length SR4 type strain gages or a pair of EVDTs attached to plugs glued into the synthetic specimen. The resilient modulus of synthetic specimens are temperature dependent. Hence, temperature effects must be considered in using reference specimens. H-16

Table H-7. Resilient Modulus Constants and Statistical Comparison at 90% Confidence Level Between Labs of Combined Results of 4 Tests - - . . , Model Constants Statistics (1 ) Comparison ResuKs LAB SPECIMEN K1 K2 Kim R2MSE p1, p2, p3 , L1Q2 1 GTB1A 1 2.783 4.288 0.9271 96.400 03801 0 001c0.01 |Different MNB1A8 4.405 0.076 0.388 94.600.0317 0.000C0.01 Regress. ~ 0.000<0.01 Runes L1lL4 1 GTB1AB 1 2.793 -0.288 0927T 86.400-03801 0000~001 Tbifferent VMBtAB 1.289 4.530 1.400 94.600.0317 0.002C0.01 Regress. 0.000<0.01 Lines ~ 7 L1/L2 GTB2AB 3.004 ~.360 0.955 97.90 0.0272 0.061>0.01 Same Interc. MNB2A8 3.685 0.005 0.461 94.50 0.0323 0.000c0.01 Diff. Slopes. . O.OOOC0.01 -. .. . . L1/L4 T GTB2A 1 3.004 -0.360 0.9551 97.90 0.0272| O.OOOcO.01 |Different VMB2AB 1.597 ~.687 1.407 95.20 0.0410 0.000C0.01 R~ress. O.000C0.01 Lines . ~. ~. ~_ L1/L3 1 GTB2A 1 3.004 -0.360 0.9551 97.90 0.0272| 0.000~0.01 |Different UTS2AB 1.063 ~.206 0.366 24.30 0.075t 0.111~0.01 Regress. ~0.000<0.01 ~Lines L2/L4 | MNC4A r 2.594 -0.573 1.179| 94.22 0.0425| 0.02~0.01 Tsams VMNCdA 1.422 ~.343 1.073 96.10 0.0351 0.04>0.01 Regress. 1 ~ ~0.395>0.01 ~Lines Note 1: Notation: R2 = Correlation Coeffiasnt Squared MSE= Mean Square Error p1, p2, p3 Level of Significance for Constants K1, K2 and K3 = Standard Deviation CV = Coefficient of Variation H-17

80 70 60 , 60 g 40 E 30 - o 20 10 0 20 lo Q LEGEb D ~ L282 O LL(L1 B2) - +U4L1Bz} D E ~: IAF LL - LaNot 95 * Confidenos Um~ UL = Upper 85% Confidence Limits - 1- ~ 1- 1 60 80 t 00 120 Bulk Stress, 8, (psi) A: CI3 ~ 3 PSI B.: O3 ~ S PSI C: O3 ~ 10 PSI D:~3~1SPSI E: O3 ~ 20 PSI Figure H-6. Compar~son of 95% Conf~dence L~mit Mode! Fit of Lab ~ MR Results with Lab 2: Base B2 120 ~o ~oo ^ 90 E 80 , 70 60 0 - ._ ~0 40 30 20 10 O 0 20 . . _ _ LEGEND · ~ - - L4B2 O LL(L1 B2) O UL(L1 B2) l je: 1' 1 1 .11 . 1 1 _ r.,. ..,~ 1 ~ Lab L4 1 u Rcferencc Lab L1 LL = Lower BS % Confidence Limits 1 1 1 1 1 40 60 80 Bulk Stress, 8, (psI) 100 120 |A CI3 = 3 PSI B: O3=SPSI ;C:CS3 5IO PSI D: CT3 = IS PSI E: O3=20PS! Figure H-7. Companson of 95% Conf~dence L~mit Mode! Fit of Lab ~ MR Results with Lab 4: Base B2 H-18

80 70 ~ 60 x - ~ 50 - E 4o - ~ 30 - 20 10 o LEGEND - L3B2 LL(L1 El2) U4L1 B2) Reference L1 Lab B lAB L3 Ma LL = L DW" 95 % Confidence Limil s UL = Upper 85% Confidence Limits .! ~ ~ 0 20 40 60 Bulk Stress, 8, (psi) A: tS3 = 3 PSI B: C,3 - S PSI C: O3 = 10 PSI D:~3= Is PSI E: c53 = 20 PSI 80 100 120 Figure H-8. Comparison of 95% Confidence Limit Model Fit of Lab 1 MR Results with Lab3:BaseB2 H-19

Not all laboratories carefully follow calibration and/or testing procedures. This error in several instances In the multi-lab study lead to inaccurate resilient modulus test results. For the same operator and test equipment, reproducible resilient moduli for replicate specimens can be obtained using the aggregate base specimen preparation and testing procedures. When axial deformations are measured outside of the biaxial cell, equipment calibration becomes a dominating factor in obtaining reliable resilient modulus results. The required {eve! of equipment calibration is complicated and time consuming. The results of this multi-lab study, although admittedly of limited scope, cast serious doubt as to whether most production type labs (or even research labs) have the required expertise and are able to expend the necessary time and effort to achieve accurate resilient modulus results using externally mounted EVDTs. 5. REFERENCES The multi-laboratory study shows the need for measuring axial deformation, which is used to calculate the resilient modulus, directly on the specimen rather than outside of the biaxial cell. Measurement of axial deformation on the specimen makes the measured resilient modulus considerably less sensitive to sysem calibration than when externally mounted EVDTs are used. Hal. Barksdale, R. D., Alba, I., Khosla, P. N., Kim, R., Lambe, P. C., and Rahman, M. S., (1993), Laboratory Determination of Resilient Modulus for Flexible Pavement Design, Interim Report No. 2, Prepared for NCHRP Project I-28, Georgia Tech Project E20-634, Atlanta, GA, 445p. H-2. ASTM, (1992), "Standard Practice for Conducting an InterIaboratory Study to Determine the Precision for Test", Designation E691-92, American Society for Testing and Materials. H-20

SUBGRADE COHESIVE SOIL EXPERIMENT Introduction The subgrade cohesive soil multi-laboratory validation resilient modulus testing program involves! 3 different laboratories. A fourth laboratory was to have participates] but the laboratory director resigned before the tests could be completed; without him the lab could not perform the tests. The participating laboratories measured cyclic deformations using LVDTs located inside the biaxial cell ant! therefore did not complete the detailed calibration program described in Appendix D. Each of the labs used the loading schedule proposed in Interim Report No. 1 for NCHRP I-28; they used their own specimen preparation procedure and equipment setup. Since eventually resilient modulus tests will be performed on unclisturbed samples, it was expected that experimenters could obtain He most consistent specimens by using their own well developed laboratory compaction procedures. Materials and Notation Cohesive Soils. Samples of two air dried North Carolina subgrade soils were supplied to the 3 laboratories in 5 gallon white plastic buckets at the end of summer, 1994. Figure H-9 shows the T-99 compaction curves along with the Atterberg limits for each of these soil samples. Sample No. 13 classifies as an A-5 and sample No. 14 classifies as an A-7-5. Based upon the T-99~curves the laboratories were directed to prepare specimens at the optimum water contents ant! maximum dry densities as shown in Table H-8; reported results indicate that the No.13 specimens dried by up to 1.3% and that the No. 14 specimens ciriecl by up to 4.3% cluring compaction. Table H-8. Specimens Tested Lab Specimen w SO ad, pcf Specified 13 18 ~96.9 L1 ~13A 17.4 98.2 13B 17.4 98.2 13C 17.4 98.2 L2 13 17.5 92.8 L3 13A 16.9 13B 16.7 Specified - 14 25 93.8 L1 14A 25.4 94.3 14B 25.5 95.6 14C 25.5 95.6 L2 14 24.2 92.3 L3 14A 20.7 _ 14B 23.5 H-21

Notation. The three laboratories are called Ll, L2, and L3 and the specimens are referenced by soil sample number and a following letter. For example Ll-13A represents the first test performed on soil sample No. 13 by laboratory L1. Laboratory Ll ran three tests on both soil samples No.13 and No. 14, laboratory L2 ran only one test on each soil sample and laboratory L3 ran two tests on each soil sample. In contrast to the aggregate base experiment, He laboratories participating in the subgrade soil experiment did not perform a calibration testing program. While requested to perform such testing He technicians were unprepared and when testing fell well behind schedule they completed just He requested tests. Resilient Modulus Test Results Figures M-IO, H-~! and H-12 portray the test results measured by the three laboratories plotted as resilient modulus in ksi versus deviator stress in psi. Each test involved 15 load sequences of 100 cycles at all the combinations of five different deviator stresses and three different confining stresses. Figure H-IOa shows that with the exception of a confining stress equal to 4 psi on specimen 13B confining stress had little influence on the measured resilient moduli for No. 13. Figure H-IOb also shows no dependence of resilient modulus on confining stress for No. 14 but has an unexpected rising resilient modulus with confining stress at ~ and 2 psi confining stress. The No. 14 specimens had much higher resilient moduli than the No. 13 specimens so the vertical scales had to be increaser! by a factor of S. Figure H-} ~ shows the test results for laboratory L2 with similar qualitative patterns but resilient moduli for No. 13 about 2 times larger than L1 and for No. 14 about 1/2 as large as Ll. Figure H-12 shows that L3 measured resilient moduli 10 to 20 times smaller than either Ll or L2. It appears L3 has some systematic error which should have been discovered during a calibration test program. Table H-9 summarizes the statistics for the subgracle testing program baser! upon the model: MR = K ~ by,] O These results suggest poor agreement Motif within each lab and for lab versus lab. Variability in tested specimens, and not just the testing procedure, contributed to the observed variability. An effective calibration program using synthetic specimens would help identify tl~e source of errors. General Findings Poor resilient modulus results were obtained while specifying soil compaction at the optimum water content which should produce the most consistent specimens. Problems getting the tests performed indicated that each of the three laboratories were experiencing significant trouble just getting the test equipment operational. The sophistication of the electronics and the difficulty to visually observe tile behavior mean tint a greater level of care is required to obtain meaningful values. Because the subgrade soil specimens are softer than tile other pavement materials, they can be damaged while setting up the specimen for testing. Before actually testing subgrade specimens, tile participating laboratories need to subunit to a well planned and supervised calibration program using synthetic specimens. In addition some of the equipment used by these laboratories does not uncork as advertised, an] their technicians have too little experience to identify the source of problems. H-22

Table H - 9. Summary of Statistics for Subgrade Soil Testing Lab Specimen K1 K2 K3 R2 s Avg. CV% . . L1 1 L1-13 1 1.737 ~.114 0.177 1 0.877 0.078 L1 -13B 0.725 -0.586 0.769 0.89S 0.134 L1 -13C 1.946 -0.293 0.162 0.874 0.082 Average 1.469 4.331 0.369 ~ 0.653 0.238 0.346 CV~ 44.442 72.054 93.666 70.0S4 L1 -14A 30.399 0.091 -0.009 0.378 0.09 L1-14B 39.615 -0.084 0.011 0.241 0.09 L1 -1 4C 30.122 0~085 ~OeO01 0.289 0.089 Average 33.379 0.031 0.000 ~ 5~403 0.099 0.010 CV% 16e 186 322~868 5076~662 1 80S.239 LZ 4 1 1~448 -0~241 0~483 T 0~928 0~058 1 ~ 1 30~517 -0~269 0~002 1 0.791 ~ 0.11 1 ~ . L3 L3-13A 0~307 -0~456 -0.112 0~911 0.1 31 L3-t 3B 0~288 -0~446 -0~089 0~866 0.159 Average 0.297 ~.451 -0.100 ~ 0~013 0~006 0~016 CV% 4~474 1~427 15~870 7~257 0~790 -0~017 0~105 1 0~133 0~118 | L3-1 4B 0~576 -0~249 0~459 0~612 0.163 Average 0.683 -0. 133 0.282 ~ 0.151 0.164 0.250 CV% 22e 147 123~239 88~611 77~999 H-23

Sam le LL -#40, % -#200, % G #13 45 5 87 44 2.77 #14 58 25 82 63 2.67 '°°T~ At- ~ ------ ~ - ~ ' 1 - 1 " 1 \\ 1 G = 2 72 ;=Sample#~] ~ p 90 1 ~ ~ | ~ it Sample#14 ~ ~ l 1 ''-''-'''''''---1-'''-''''-':'--- ''- - ''''1''''- '''-''' /' ·1'''''---'' -'-'-'-' ''--1'''''--'-' ' ''-' -'' 1 ''''\\'~"''-'- ' 1 '- '-'''''"'''''---''''''-- -'1 '--'''''-''-''-'-' '-"'''-''''''-"1 "--'--''''/ i''''-''' '' '1'-'' ''-''' -:---'2'' '"'"'''' ' '1"' ' ' ' '''' ' -'''' '''1' '''' '\'''' '' -'-'' 1 ' ' ' -'''' - ----' - '-1 '"' " '''--I ''--- ''"'-'-'I 1 1 ~ - 1 ~' ' '1' ' ' - '"' -~''i' ""----'''' '- 1 . . I ., : 85 5 1 0 1 5 20 25 30 35 Water Content, oh Figure H - 9. Cohesive Soil Samples used for Subgrade Tests H-24

6- ~ ~ - cO 4- in o3 - 2 _ a) a: 1 o 50 40 . co x in ~5 30 _ O ID 20- e_ co ~ 10 - ~0 ~ 0~ - - O o , . . . . . * · ·, O 4 psi (A) ~ 2 psi (A) 0 1 PA (A) 4 psi (B) 2 pal (B) 1 psi (B) x 4 psi (C) + 2 psi (C) * 1 psi (C) _ ~I l I ~ I ~ r I T I I 1 0 5 10 15 Deviator Stress (psi) (a) Results for L1-13 0 4 psi (A) ~ 2 pSi (A) 0 1 PA (A) 4 psi (B) 2 pSi (B) 1 psi (B) 4 psi (C) 2 pSi (C) * l 5 . 10 15 Deviator Stress (psi) (b) Results for L1-14 Figure H - 10. Subgrade Soil Test Results for Lab 1 H-25

6 5- .^ ye _, AL _ an ~ ~3 a) ._ ~ _ ~ _ ·_ _ a) a: 1 O 50 40 ._ . _ an ~5 30 0 20 t a) ·_ ._ tar: 10 O O ~O O 4 psi ~ 2 psi 0 1 psi O O 0 A O ~ r I I I T I - 1 1-- 1 - 0 5 10 15 Deviator Stress (psi) (a) Results for L2-13 - O 4 psi ~ 2 psi 0 1 pad 1~ 10 15 - O- o 5 Deviator Stress (psi) (b) Results for L2-14 Figure H - 1 1. Subgrade Soil Test Results for Lab 2 H-26

0.6 - l 0.5 _` in . ,,, 0.4 0.3 :-0.2 ~n A: 0.1 O- , o to 18 I\ A IF ~- O 4 psi (A) 2 psi (A) 1 psi (A) 4 psi (B) 2 psi (B) 1 psi (B) . ~ 5 10 15 Deviator Stress (psi) (a) Results for L3-13 4 · _ _' in :,3 O ~ ~ 2 a) ._ ._ In tr: 1- . o _ _ ~ ~ -e - ,- l l l o r I I 1 O 4 psi (A) ~ 2 psi (A) 0 1 psi (A) 4 psi (B) 2 psi (B) 1 psi (B) 1 1 1 () 5 10 15 Deviator Stress (psi) (b) Results for L3-14 Figure H - i2. Subgrade Soil Test Results for Lab 3 H-27