Tayebali, Deacon, Coplantz, Harvey and Monismith 118
Mix and Mode-of-Loading Effects on Fatigue Response
of Asphalt-Aggregate Mixes Akhtarhusein A. Tayebali 1 , John A. Deacon 2 , John S. Coplantz 3 , John T. Harvey 4 and Carl L. Monismith 5 Introduction A series of accelerated performance-related tests for asphalt-aggregate mixes was performed under Project A-003A of the Strategic Highway Research Program's (SHRP) asphalt research activity, and methods were developed for analyzing asphalt-aggregate interactions which significantly affect pavement performance. The fatigue investigation, conducted at the University of California, Berkeley, evaluated a number of promising testing systems both under controlled-stress and controlled-strain modes-of-loading. A frequent question in mix evaluation is whether the mode-of-loading selected for the laboratory testing will influence results of the mix evaluation process. For example, if one mix, Mix A, is judged superior to another mix, Mix B, based on laboratory tests under one mode-of-loading, will it also be superior based on results obtained with another mode-of-loading? Prior speculation suggests that controlled-strain testing might be more appropriate for evaluating mixes for relatively thin pavements on stiff foundations and that controlled-stress testing might be more appropriate for thicker and relatively stiffer structures. Resolution of this issue was an important task of the SHRP A-003A project. The specific objectives of this paper are to explore mix and test variable effects including those of mode-of-loading, on the fatigue response of asphalt aggregate mixes. Following a brief description of the laboratory test programs on which the analyses reported herein are based, the paper examines the effects of various mix properties on fatigue life. Attention is then turned to exploring the effects of mode-of-loading on in-situ mix performance. Laboratory Test Program Data reported and analyzed herein were developed as part of a comprehensive evaluation of fatigue testing equipment under SHRP Project A-003A (Tayebali et al., [1,2,3]; Hicks et al., [4]). This study, referred to as the 22 study because it included two different asphalts and two different aggregates, was designed as a one-half fractional factorial experiment, [1,2,3,4] permitting the estimation of all main effects as well as all two-factor interactions. All testing involved third-point flexure of prismatic beam specimens. Fatigue equipment used in this study applied a pulsed load of 0.1-sec duration at a frequency of 1.67 Hz in both controlled-stress and controlled-strain modes-of-loading. Unconditioned specimens (relative to aging and water conditioning) were used for all testing. Features of the 22 controlled-stress and controlled-strain experiments are summarized in Table 1 and described below: Aggregates: Two MRL (Material Reference Library) aggregates were used - RB, a 100-percent crushed granite characterized by rough surface texture and low stripping potential; and RL, a partially crushed chert, characterized by smooth surface texture, round particle shape and relatively high level of stripping potential. Table 2 shows the gradation used in this study. Asphalt: Two MRL asphalts were used - AAK-1, an AC-30 with relatively lower temperature susceptibility (PI = -0.5); and AAG-1, an AR-4000 with relatively higher temperature susceptibility (PI = -1.5). Asphalt Content: Two asphalt contents were used. For each asphalt-aggregate mix, the lower (optimum) asphalt content was determined using standard Hveem procedures. The upper asphalt content was set 0.6 percent higher (by weight of aggregate) approximately corresponding to the optimum asphalt content found using the Corps of Engineers (Marshall) design procedure 1 . Table 3 shows the asphalt contents used for the Mix and Load Fatigue Effects 119 various mixes considered. Compaction and Air Void Contents: Two levels of compactive effort were used. Low compactive effort was adjusted to provide a target air void content of 8 percent in the specimen. The high level of compactive effort was adjusted to produce a target air void content of 4 percent. A Triaxial Institute Kneading Compactor was used to fabricate the beam specimens. Test conditions: Two temperatures - 0C and 20C, and two stress and strain levels (high and low) were used for the controlled-stress and controlled-strain tests. It should be noted that the target stress/strain levels were adjusted somewhat for the low and high temperatures to ensure failure within a reasonable number of cycles. Table 1. 22 Investigation Asphalts 2 - MRL core asphalts (AAG-1 and AAK-1) Aggregates 2 - MRL aggregates (RB and RL) Asphalt contents 2 - Optimum (Hveem) and optimum plus 0.6 percent Air-void levels 2 - 4 0.5 and 8 0.5 percent Stress/Strain levels 2 - Low and High Replicates at each strain level 2 Mode-of-loading 2 - controlled-stress and controlled-strain Temperature 2 - 0C and 20C Frequency 1 - 1.67 Hz (0.1 sec on, 0.5 sec off) Specimen size 1.5-in. (38mm) height, 1.5-in. (38mm) width, 15-in. (38mm)length Total number of mixes tested 16 Total number of specimens tested 128 (for one-half fractional factorial) Tayebali, Deacon, Coplantz, Harvey and Monismith 120 Table 2. Aggregate Gradation Sieve Size Percent Passing by Weight ASTM Spec. (D 3515) 1 in. (25mm) 100 100 3/4 in.(19mm) 95 90-100 1/2 in. (13mm) 80 - 3/8 in. (9.5mm) 68 56-80 No. 4 (4.75mm) 48 35-65 No. 8 (2.38mm) 35 23-49 No. 16 (1.19mm) 25 - No. 30 (0.60mm) 17 - No. 50 (0.30mm) 12 5-19 No. 100 (0.15mm) 8 - No. 200 (0.074mm) 5.5 2-8 Mix Effects on Fatigue Response Mix and testing effects on laboratory fatigue response of asphalt-aggregate mixes were evaluated primarily using statistical techniques, most importantly a General Linear Model (GLM) formulation. Results are summarized below. Table 3. Asphalt Content Used for Various Mixes Aggregate Stripping Potential a
Temperature Susceptibility a Low (AAK, PI=-0.5) High (AAG, PI=-1.5) Asphalt Content b Asphalt Content b Optimum High Optimum High Low (RB) 5.1 5.7 4.9 5.5 High (RL) 4.3 5.0 4.1 4.8
a Based on the information on the information from the Materials Reference Library (MRL)
b Asphalt content by weight of aggregate Mix and Load Fatigue Effects 121 2 Cumulative dissipated energy to failure is a measure of mix fatigue resistance that was evaluated in the SHRP A-003A project. It is highly correlated with fatigue life. Statistical Analysis A statistical analysis of the 22 fatigue data was conducted to evaluate the effects of mix and testing variables (that is, asphalt type, air-void content, temperature, etc.) on asphalt mix performance as measured by the controlled-stress and controlled-strain flexural beam fatigue tests. Specific tasks for the statistical analysis included the following: Test for correlation among the independent variables (Pearson). Analysis of variance (ANOVA) of full models (all main factors and two-factor interactions) to determine the sensitivity of flexural stiffness, fatigue life, and cumulative dissipated energy 2 to mix and testing variables. General linear modeling (GLM) to develop models for stiffness, fatigue life, and cumulative dissipated energy. These models were then used to adjust for the effects of any off-target air-void contents and off-target stress or strain levels. Summaries of the effects of different mix and testing variables on stiffness, fatigue life, and cumulative dissipated energy based on the results of GLM. Statistical analysis in this study was conducted using the SYSTAT Statistical Software Package (Wilkinson, [6]). For controlled-stress tests, failure is well defined since specimens are cracked through at the end of the test. For controlled-strain tests, the number of cycles to failure was defined to coincide with a 50-percent reduction in flexural stiffness. The flexural stiffness measured at a given number of load repetitions was computed from the corresponding tensile stress and strain. The dissipated energy per cycle was computed as the area within the stress-strain hysteresis loop. Figure 1 illustrates a typical result for the controlled-strain mode-of-loading. The cumulative dissipated energy was computed as the area under the dissipated energy per cycle versus number of load repetitions curve. Figure 2 shows the variation of dissipated energy per cycle with number of load repetition for typical controlled-stress and controlled-strain tests. An assumption necessary for Analysis of Variance (ANOVA) and General Linear Modeling (GLM) is that the variables are normally distributed. For fatigue data, stiffness, fatigue life, and cumulative dissipated energy were found to be approximately log-normally distributed and, therefore, log-transformed (natural logarithms) data were used for the analysis. Some independent variables which could not be precisely controlled but which were carefully measured (such as air voids) were normalized. The purpose of normalizing these variables was to ensure that they could be added or omitted in statistical models without adversely affecting the remaining model coefficients. Air voids, temperature, and the natural logs of applied stress and strain were normalized as follows: X normalized = (X - X mean )/{(X Hi - X Lo )/2} (1) where: X = observed value; X mean = grand mean; X Hi = mean of the high values; and X Lo = mean of the low values. Tayebali, Deacon, Coplantz, Harvey and Monismith 122 Figure 1 Stress-Strain Hysteresis Loop for Controlled-Strain Test Figure 2. Effect of Load Repititions on Dissipated Energy per Cycle The experiment design selected in this study was a one-half fractional factorial, (Tayebali et al., [1,2]; Hicks et al., [3]), which permits the estimation of all main effects of the experimental factors (variables) and their two-factor interactions. Accordingly, for ANOVA and GLM a log-linear model of the following type was utilized: Y i = : + " 1 *Asph + " 2 *Aggr + " 3 *%Asph + " 4 *Voids + " 5 *Temp + " 6 *Stress/Strain + " 7 *Asph*Aggr + " 8 *Asph*%Asph + " 9 *Asph*Voids + " 10 *Asph*Temp + " 11 *Asph*Stress/Strain + " 12 *Aggr*%Asph + " 13 *Aggr*Voids + " 14 *Aggr*Temp + " 15 *Aggr*Stress/Strain + " 16 *%Asph*Voids + " 17 *%Asph*Temp + " 18 *%Asph*Stress/Strain + " 19 *Voids*Temp + " 20 *Voids*Stress/Strain + " 21 *Temp*Stress/Strain (2) Mix and Load Fatigue Effects 123 Table 4. Summary Statistics for 2x2 Fatigue Experiment Table 5. Summary Statistics from GLM of 2x2 Experiment where: Y 1 = log-stiffness; Y 2 = log-cycles-to-failure; Y 3 = log-cumulative-dissipated-energy; : = constant (grand mean); " i = model coefficients; Asph = asphalt type; Aggr = aggregate type; %Asph = asphalt content; Voids = normalized percent air voids; Temp = normalized temperature; and Stress/Strain = log-normalized stress or strain. Table 4 shows the general statistics for the 22 fatigue data based on the log-transformed data. The coefficients of variation (quotient of standard deviation and mean, expressed as a percentage) for flexural stiffness, fatigue life and cumulative dissipated energy, based on natural log-transformed data, are similar for both modes-of-loading and are approximately 7, 18, and 23 percent for controlled-stress tests, and 5, 17, and 20 percent for controlled-strain tests. Table 5 shows the summary statistics from the Generalized Linear Models (GLM) for flexural stiffness, fatigue life, and cumulative dissipated energy. The fit of the models for flexural stiffness is good with coefficients of determination of 0.98 and 0.95, and coefficients of variation of 12 and 16 percent for the controlled-stress and controlled-strain tests, respectively. The models for cycles to failure and cumulative dissipated energy to failure show reasonable fits with coefficients of determination of approximately 0.8 for the former and 0.7 for the latter, and coefficients of variation of approximately 92 to 98 percent for both modes-of-loading. Tayebali, Deacon, Coplantz, Harvey and Monismith 124 3 It should be noted that the difference between optimum and high asphalt content was only 0.6-percent by weight of aggregate as indicated in Table 3. Coefficients of variation presented in Table 5 were computed as follows: CV = 100 * (e MSE - 1) 0.5 (3) where: CV = coefficient of variation in percent; e = base of the natural logarithms; MSE = mean square error. These coefficients of variation are measures of prediction error, testing error, and random error. They are based on the actual (non-log-transformed) data as opposed to the transformed data in Table 4. Because the data, in reality, are log-normally distributed, the coefficients of variation based on non-transformed data are expected to be high. These coefficients of variation are generally in agreement with those observed for trapezoidal cantilever and diametral fatigue tests (Tayebali et al., [3]) and the approximately 80 percent coefficient of variation reported in the literature for fatigue life based on fitted data. Table 6 identifies those main effects and two-factor interactions which are statistically significant at the 95-percent probability level for each of the three response variables. Tables 7 and 8 show the average values of stiffness, cycles to failure, and cumulative dissipated energy for the controlled-stress and controlled-strain tests. For these tables, averages were computed by first estimating the values for the complete data set using the GLM models. The use of these regression models also facilitated corrections for the air voids and stress/strain levels to the targeted low and high values for individual test results. The data set was then partitioned according to the test temperature and the independent variable, and the logarithmic averages obtained. The effects of mix and testing variables on the response variables are discussed in the following sections. Flexural Stiffness Statistical analysis for flexural stiffness shows that both controlled-stress and controlled-strain tests are sensitive to asphalt type (Table 6), with mixes containing asphalt AAG-1 showing a higher stiffness than those containing asphalt AAK-1 (Tables 7 and 8). The percentage difference (expressed as a percentage of greater value) between the mixes containing asphalts AAG-1 and AAK-1 ranges from 21 to 44 percent at 0C, and 54 to 58 percent at 20C. Both tests show a significant interaction between asphalt type and temperature, which was expected due to the high temperature susceptibility of the AAG-1 asphalt. Aggregate type shows a significant effect on flexural stiffness for both test types. The average stiffness of mixes containing RL aggregate is approximately equal to that of mixes containing RB aggregate at the low temperature (0C), but the average RL mix stiffness is significantly lower at the higher temperature (20C). This fact is reflected by the interaction between aggregate type and temperature which was significant for both test types. The effect of asphalt content on flexural stiffness was not significant either directly or through the two-factor interaction for the controlled-stress test, but was significant for the controlled-strain test, with mixes containing the optimum asphalt content showing approximately 11 percent higher stiffness compared to mixes with the higher asphalt content 3 . As expected, both air voids and temperature significantly influence flexural stiffness with mixes containing lower voids and lower temperatures exhibiting higher stiffness as compared to mixes with higher voids and higher temperatures. Stress level was observed to have a small (less than 10 percent) but statistically significant effect on flexural stiffness in controlled-stress tests. For controlled-strain tests, strain level did not have a significant influence on flexural stiffness. Mix and Load Fatigue Effects 125 4 Note that low and high stress levels for 0C and 20C are different (Table 7). Table 6. Statistically Significant Effects in GLM of 2x2 Experiment Fatigue Life (Cycles-to-Failure) Based on the results of the statistical analysis, fatigue life as a response variable is sensitive to all mix and test variables either directly or through two-factor interactions for both test types (Table 6). For the controlled-stress test, mixes containing asphalt AAG-1 show higher fatigue life as compared to those mixes with asphalt AAK-1 for both temperatures. The difference is larger at 0C (86 percent) than at 20C (26 percent) 4 . For controlled-strain tests, as anticipated, the effect of asphalt type is reversed with mixes containing asphalt AAG-1 showing lower fatigue life compared to mixes containing asphalt AAK-1, the difference being approximately 62 percent for both temperatures. Both tests exhibited sensitivity to aggregate type with mixes containing aggregate RL having lower fatigue life than mixes containing aggregate RB. The difference at both temperatures was approximately 68 percent. Asphalt content was noted to have a significant effect on fatigue life only through interactions. For controlled-stress tests, interaction between asphalt content and temperature was significant, with fatigue life being higher for mixes with the optimum asphalt content at 0C, and lower at 20C when compared to the mixes containing the high asphalt content. This result is unexpected since asphalt content did not have a significant influence on stiffness for either temperature in the controlled-stress test. For controlled-strain tests, average fatigue lives are approximately the same for both asphalt contents, although there are significant interactions between asphalt content and asphalt type, and asphalt content and air void content. Air void content has a significant influence on fatigue life for both test types with fatigue life for low voids mixes being higher than for high voids mixes. Note that, for controlled-strain tests, the effect of asphalt type (Table 8) was such that those specimens exhibiting higher stiffness (mixes containing asphalt AAG-1) had lower fatigue life as compared to specimens having lower stiffness (mixes containing asphalt AAK-1). The effect of air void content is such that specimens with higher stiffness (lower air voids) exhibit higher fatigue life as compared to Tayebali, Deacon, Coplantz, Harvey and Monismith 126 T a b l e
7 .
A v e r a g e
R e s u l t s
f r o m
G L M
M o d e l ,
F l e x u r a l
B e a m
F a t i g u e
C o n t r o l l e d - S t r e s s
T e s t s E f f e c t L e v e l T e m p e r a t u r e
o f
0 C T e m p e r a t u r e
o f
2 0 C S t i f f n e s s
( p s i ) F a t i g u e
L i f e ( c y c l e s ) C u m u l .
D i s s . E n e r g y
( p s i ) S t i f f n e s s
( p s i ) F a t i g u e
L i f e ( c y c l e s ) C u m u l .
D i s s . E n e r g y
( p s i ) A v e r a g e 2 , 4 5 4 , 7 0 0 5 , 8 3 4 , 0 0 0 2 8 , 8 0 0 4 2 5 . 1 0 0 3 4 , 5 0 0 1 , 1 0 0 T a b l e
8 .
A v e r a g e
R e s u l t s
f r o m
G L M
M o d e l ,
F l e x u r a l
B e a m
F a t i g u e
C o n t r o l l e d - S t r a i n
T e s t s E f f e c t L e v e l T e m p e r a t u r e
o f
0 C T e m p e r a t u r e
o f
2 0 C S t i f f n e s s
( p s i ) F a t i g u e
L i f e ( c y c l e s ) C u m u l .
D i s s . E n e r g y
( p s i ) S t i f f n e s s
( p s i ) F a t i g u e
L i f e ( c y c l e s ) C u m u l .
D i s s . E n e r g y
( p s i ) A v e r a g e 2 , 0 4 8 , 5 0 0 4 5 , 5 0 0 7 0 0 7 8 9 , 1 0 0 8 7 , 4 0 0 1 , 0 0 0 Mix and Load Fatigue Effects 127 specimens with lower stiffness (higher air voids). This negative effect of stiffness on fatigue life due to asphalt type and air void content suggests that for a given asphalt type, reducing the stiffness (increasing the air void content) will decrease the fatigue life; whereas, for a given air void content, decreasing the stiffness (changing the asphalt type) will increase the fatigue life for controlled-strain mode-of-loading. For controlled-stress an increase in stiffness due to a different asphalt type and/or a reduction in air void content will result in increased fatigue life. Both temperature and stress/strain have a significant influence on fatigue life. For controlled-stress tests, fatigue life is longer at 0C than at 20C. This result is expected as the stiffness is higher at 0C. In the controlled-strain test, fatigue life is shorter at 0C compared to 20C, which is also expected. Fatigue life for both test types decreases with increase in stress or strain level. For both test types, significant interaction between temperature and stress/strain level was noted. This was expected since during testing the low and high stress/strain levels were varied depending on the temperature in order to assure that the specimens failed at a reasonable number of cycles. Cumulative Dissipated Energy In general, the results for cumulative dissipated energy are similar to those observed for fatigue life. Asphalt type was significant only for controlled-strain testing. However, there is a significant asphalt type and temperature interaction for controlled-stress testing, indicating that the ranking between mixes containing asphalt AAG-1 and AAK-1 actually reverses at 20C, with mixes containing asphalt AAG-1 having lower cumulative dissipated energy although the fatigue life for the mixes containing this asphalt is higher at this temperature. Both tests are sensitive to aggregate type, air void content, temperature and stress/strain levels. Mixes containing aggregate RB show higher cumulative dissipated energy as compared to mixes containing aggregate RL. Increasing air void content and increasing stress/strain level decreases the cumulative dissipated energy. Cumulative dissipated energy decreases with increasing temperature for controlled-stress tests, and increases with increase in temperature for controlled-strain tests. Summary Based on the results of the controlled-stress and controlled-strain flexural beam fatigue tests, the response variables (flexural stiffness, fatigue life, and cumulative dissipated energy) were found to be sensitive to asphalt type, aggregate type, air-void content, and temperature. Stiffness is insensitive to strain level for controlled-strain tests, but sensitive to stress level in controlled-stress tests. Fatigue life and cumulative dissipated energy are sensitive to strain level. The ranking of mixes for the factor levels of each response variable for the two modes-of-loading are similar except for the influence of asphalt type and temperature which have opposite effects for the controlled-stress and controlled-strain loading conditions. The effects of the significant variables based on the GLM are summarized in Table 9. Mode-of-Loading Effects on Fatigue Response Although the 22 fatigue experiment was primarily designed to evaluate the different laboratory fatigue testing procedures, it also provided information useful for assessing possible effects of mode-of-loading on fatigue behavior. Mode-of-loading may be an important consideration in mix analysis because, for similar initial stress/strain conditions, fatigue life is typically greater in controlled-strain loading than in controlled-stress loading and, more importantly as noted previously, mixes of greater stiffness (due to asphalt type and temperature) tend to perform better in controlled-stress loading but worse in controlled-strain loading. The mode-of-loading analysis was done using least squares calibrations of models of the following type: N f = a exp b MF exp c Vo (, o or F o ) d (S o ) e (4) where: N f = cycles to failure; Tayebali, Deacon, Coplantz, Harvey and Monismith 128 MF = mode factor assuming values of 1 and -1 for controlled-strain and controlled-stress loading, respectively; V o = initial air-void content in percent; , o = initial flexural strain in in/in; F o = initial flexural stress in psi; S o = initial mix stiffness in psi; and a,b,c,d,e = regression constants. A separate model was calibrated for each mode-of-loading and each test temperature. Models were also calibrated for data sets aggregated over both temperatures and both modes-of-loading. The Table 10 summary of the regression analyses is based on calibrations in which outliers, defined as cases where the absolute values of the residuals (ln) Table 9. Summary Effects of Mix and Testing Variables Variable Flexural Stiffness Fatigue Life Cumulative Dissipated Energy Controlled-Stress Tests Asphalt Type AAG-1 > AAK-1 AAG-1 > AAK-1 Temp. dependent a Asphalt Content Negligible Temp. dependent b Temp. dependent b Aggregate Type RB > RL c RB > RL RB > RL Air Voids Low > High Low > High Low > High Temperature Low > High Low > High Low > High Stress Negligible Low > High Low > High Controlled-Strain Tests Asphalt Type AAG-1 > AAK-1 AAG-1 < AAK-1 AAG-1 < AAK-1 Asphalt Content Opt > High Negligible Negligible Aggregate Type RB > RL RB > RL RB > RL Air Voids Low > High Low > High Low > High Temperature Low > High Low < High Low < High Strain Negligible Low > High Low > High a AAG-1 > AAK-1 at 0C and AAG-1 < AAK-1 at 20C b Optimum > High at 0C and Optimum < High at 20C c Negligible difference at 0C Mix and Load Fatigue Effects 129 Table 10. Comparative Regression Models for Controlled-Stress and Controlled-Strain Testing Model Tem p. (C) R 2 CV (%) Controlled-Stress Models Regressed on Stress N f = 7.6527*10 11 exp -0.4975 Vo (F o ) -4.7425 (S o ) 0.6816 20 0.8 7 73 N f = 8.6720*10 -7 (F o ) -5.7260 (S o ) 3.9524 0 0.9 0 69 N f = 8.7309 exp -0.2130 Vo (F o ) -3.7040 (S o ) 2.119 Both 0.8 2 82 Controlled-Strain Models Regressed on Strain N f = 4.2554 exp -0.4742 Vo (, o ) -3.9200 (S o ) -1.4792 20 0.8 7 73 N f = 9.5223*10 -11 (, o ) -4.0117 0 0.9 0 55 N f = 3.2513*10 -4 exp -0.2542 Vo (, o ) -3.9211 (S o ) -0.8824 Both 0.8 5 66 Controlled-Stress Models Regressed on Strain N f = 7.8984*10 11 exp -0.4966 Vo (, o ) -4.7620 (S o ) -4.0757 20 0.8 7 73 N f = 2.7755*10 -12 (, o ) -4.2203 0 0.8 9 73 N f = 8.5292 exp -0.2126 Vo (, o ) -3.7148 (S o ) -1.5900 Both 0.8 2 82 Combined Models Regressed on Strain N f = 2.2934*10 4 exp 0.5222 MF exp -0.4040 Vo (, o ) -3.0091 (S o ) -1.6475 20 0.8 6 69 N f = 3.3596*10 -11 exp 0.8611 MF (, o ) -4.0336 0 0.9 0 55 N f = 0.9500 exp 0.4472 MF exp -0.2566 Vo (, o ) -3.3669 (S o ) - 1.1633 Both 0.8 4 75 Tayebali, Deacon, Coplantz, Harvey and Monismith 130 Figure 3 Effect of Testing Temperature on M j -0 0 exceeded 1.1, were removed. Removal of outliers not only enhanced the accuracy of the models but, more importantly, improved their consistency and reasonableness. Predictor variables in each of the models shown in Table 10 were significant at a probability level of 0.05 or less. A frequent question in the interpretation of fatigue test data is whether the effect of temperature on fatigue behavior is fully accounted for by its related effect on mix stiffness. Those models in which strain was used as a predictor variable instead of stress certainly demonstrate differences between models calibrated at 20C and those calibrated at 0C. Most notably, effects of air-void content and stiffness are statistically significant only at 20C. However, extrapolations of the 20C controlled-strain calibrations to stiffnesses more characteristic of the 0C temperature compare nicely with the 0C calibrations (Figure 3). Whether the differences of Figure 3 are of practical significance probably depends upon how the information is to be used. Overall, results of the regression analysis can be summarized as follows: At a given stress level in controlled-stress testing, stiffer mixes have greater fatigue resistance (positive stiffness power in controlled-stress models regressed on stress). At a given strain level in controlled-strain testing, stiffer mixes have less fatigue resistance (negative stiffness power in controlled-strain models regressed on strain). The effects of mix stiffness and air-void content are much less significant at 0C than at 20C. Nevertheless, the effect of temperature on fatigue life can be accounted for by its related effect on stiffness, but with some loss in accuracy. In general, fatigue lives under controlled-strain loading are approximately 2.4 times greater than those under controlled-stress loading (from the combined model, exp 0.4472 MF equals 0.639 for controlled-stress loading and 1.56 for controlled-strain loading). The critical unknown in mix evaluation is whether the mode-of-loading selected for laboratory testing will influence results of the evaluation process. For example, if one mix, Mix A, is judged superior to another mix, Mix B, based on laboratory tests under one mode-of-loading, will it also be superior based on results obtained with the opposite mode-of-loading? Prior speculation has suggested that controlled-strain testing might be necessary to evaluate mixes for relatively thin pavements on stiff foundations and that controlled-stress testing might be more appropriate for thicker and relatively stiffer structures. Fatigue behavior of in-situ mixes is determined by two key factors: 1) their resistance to the destructive effects of repetitive stresses/strains, and 2) the levels of stress/strain to which they are subjected under traffic loading. Laboratory testing is necessary to establish the fundamental fatigue behavior (N f -, o relationship), and mechanistic analysis is necessary to establish critical levels of stress/strain. A combination of fatigue testing and mechanistic analysis is required in order to evaluate likely in-situ behavior. Mix and Load Fatigue Effects 131 To address the mode-of-loading issue thus requires that mix performance be simulated for a suitable range of in-situ pavement conditions. For convenience, the mechanistic analysis (ELSYM5) for this study assumed linear elastic behavior for all pavement layers. The surface thickness of the two-layered structures ranged from 2 to 12 in. (51 to 305mm) and subgrade moduli were varied, as indicated in Table 11, to exaggerate the relative stiffness of the surface layer to that of its support. Poisson's ratios of 0.35 and 0.30 were used for surface and subgrade layers, respectively. Loading consisted of 10,000 lb (4536kn) on 100 psi (690kPa) dual tires, spaced 12 in. (305mm) center-to-center. The first step in the simulations was to estimate the maximum principal tensile strains in these six pavement structures for mixes that had been tested in the laboratory. These strains were then used with the previously calibrated N f vs. , o relationships to determine the simulated cycles to failure in situ. Finally, the simulated cycles to failure were examined to determine if mixes were ranked identically for both the controlled-stress and controlled-strain modes. The one complexity in the analysis stemmed from the fact that, at 20C, stiffnesses measured under controlled-stress loading were considerably smaller than those measured under controlled-strain loading. These differences may be due in part to the fact that step loads (0.1 sec on followed by 0.5 sec off) were used in the controlled-stress testing while haversine loads (also 0.1 sec on followed by 0.5 sec off) were used in the controlled-strain testing. Table 11. Surface Thickness and Subgrade Moduli for Two-Layered Structure Surface Thickness in. (mm) Subgrade Modulus psi (kPa) 2 (51) 30,000 (206,850) 4 (102) 25,000 (172,395) 6 (152) 20,000 (137,900) 8 (203) 15,000 (103,425) 10 (254) 10,000 (68,950) 12 (305) 5,000 (34,475) Another contributing factor may have been the rather lengthy delay (of up to eight months) between completion of the controlled-stress testing and completion of the controlled-strain testing. Limited evidence of stiffening due to steric hardening, particularly with storage at room temperatures, has been observed by A-003A investigators at Oregon State University. In any case, in-situ simulations are comparable only when identical stiffnesses are employed for each of the two laboratory modes-of-loading. This was rather easily accomplished by factoring controlled-stress Tayebali, Deacon, Coplantz, Harvey and Monismith 132 5 For this purpose a calibration was considered acceptable if, for both controlled-stress and controlled-strain loading, the coefficients of determination of the ln N f -ln , o relationships exceeded 0.90 and the ln , o term was statistically significant at a probability level of 0.05 or smaller. Table 12. Effect of Laboratory Mode of Loading on Simulated Fatigue Life of Three Mixes Mode of Loading Surface Thickness in. (mm) Simulated Fatigue Life Mixture 1 575,000 psi (3.96 MPa) Mixture 2 715,000 psi (4.93 MPa) Mixture 3 1,013,000 psi (6.98 MPa) Controlled- Stress Testing 2 (51) 2,000 10,000 40,000 4 (102) 6,000 35,000 213,000 6 (152) 17,000 107,000 1,275,000 8 (203) 39,000 257,000 5,524,000 10 (254) 77,000 528,000 17,209,000 12 (305) 129,000 957,000 40,394,000 Controlled- Strain Testing 2 (51) 5,000 11,000 27,000 4 (102) 16,000 35,000 65,000 6 (152) 50,000 96,000 167,000 8 (203) 126,000 212,000 363,000 10 (254) 269,000 408,000 660,000 12 (305) 476,000 688,000 1,035,000 stiffnesses such that the average stiffness for each mix tested under controlled-stress loading was the same as the average under controlled-strain loading. Only three A-003A mixes, all tested at 20C, were found which yielded acceptable calibrations 5 of the N f -, o relationships. Over the range of pavement structures evaluated and using either controlled-stress or controlled-strain data, the ranking of the three mixes was unchanged (Table 12). Mix 3 was always superior, and Mix 1 was always inferior. This limited analysis suggests, therefore, that the evaluation of mix performance may well be independent of laboratory mode-of-loading. This analysis was extended to include four hypothetical mixes having stiffnesses of 400,000 and 600,000 psi (2.76 and 4.14 MPa) and air-void contents of 4 and 7 percent. Because of the hypothetical nature of these mixes, their fatigue response was estimated by regression models calibrated from testing at 20C. For controlled-strain testing, the applicable model is (Table 10): N f = 4.2554 exp -0.4742 Vo (, o ) -3.9200 (S o ) -1.4792 (5) For controlled-stress testing, a recalibration using adjusted stiffnesses yielded the following model: Mix and Load Fatigue Effects 133 N f = 2.5263*10 5 exp -0.2007 Vo (, o ) -3.4134 (S o ) -2.1239 (6) Once again, results of the simulation, summarized in Table 13, generally seem to be independent of mode-of-loading. For the same mix stiffness, low air-void mixes were always superior to high air-void mixes. As anticipated, for thin pavements stiffer mixes demonstrated inferior fatigue resistance while, for thick pavements, stiffer mixes were preferred. The only difference between mode-of-loading is in identifying the borderline between "thin" and "thick" pavements. Based on this analysis, this difference becomes important for surface thicknesses in the range of 3 to 5 in. (75 to 125mm). The borderline thickness, however, is expected to vary depending on such factors as temperature, mix properties, and the stiffness of the pavement surface layer relative to that of its support. In summary this analysis has demonstrated the importance of mode-of-loading in the proper interpretation of laboratory fatigue data. It has confirmed that fatigue lives under controlled-strain loading generally exceed those under controlled-stress loading and that, upon Table 13. Effects of Laboratory Mode of Loading on Simulated Fatigue Life of Hypothetical Mixes of Varying Air Voids and Surface Stiffness Mode-of- Loading Surface Thickness in. (mm) Simulated Fatigue Life 4 Percent Air Voids 7 Percent Air Voids 400,000 psi (2.76 MPa) 600,000 psi (4.14 (MPa) 400,000 psi (2.76 MPa) 600,000 psi (4.14 MPa) Controlled Stress Testing 2 (51) 46,000 30,000 25,000 16,000 4 (102) 103,000 91,000 56,000 50,000 6 (152) 290,000 294,000 159,000 161,000 8 (203) 695,000 766,000 381,000 420,000 10 (254) 1,324,000 1,641,000 725,000 899,000 casual inspection, effects of mix stiffness on fatigue life are generally reversed for the two modes-of-loading. However, when test results are interpreted in terms of the performance expected of the pavements in which they are placed, it appears that controlled-stress and controlled-strain testing may yield similar mix rankings especially for the substantial pavement structures characteristic of the nation's primary trucking highways. Summary and Conclusions Both the controlled-stress and controlled-strain flexural beam fatigue tests evaluated in this study were found to be sensitive to mix and test variables either directly or through interactions for the three response variables stiffness, fatigue life and cumulative dissipated energy. The variation of asphalt content in the tested mixes was small and, as a result, the observed asphalt content effects on stiffness and fatigue life were small and inconsistent. The effects of mode-of-loading on the ranking of mixes were evaluated by simulating in-situ pavement performance based on the flexural beam controlled-stress and controlled-strain fatigue data. Results of the analysis indicate that fatigue lives under controlled-strain loading generally exceed those under controlled-stress loading and that the effects of mix stiffness on fatigue life are generally reversed for the two modes-of-loading. However, when test results are interpreted in terms of the performance expected of the pavements in which they are placed, it appears that controlled-stress and controlled-strain testing may yield similar mix rankings, especially for more substantial pavement structures. Tayebali, Deacon, Coplantz, Harvey and Monismith 134 Major findings and conclusions of this investigation include the following: At a given stress level in controlled-stress testing, stiffer mixes generally have greater fatigue resistance (positive stiffness power in controlled-stress models regressed on stress). At a given strain level in controlled-strain testing, stiffer mixes generally have lesser fatigue resistance (negative stiffness power in controlled-strain models regressed on strain). Specific mix variables may, however, produce results that are contrary to general expectations. For example, lower air void content and crushed, rough-textured aggregate (RB) meant increased stiffness and increased fatigue life not only in controlled-stress tests but also in controlled-strain tests as well. The effects of mix stiffness and air-void content are much less significant at 0C than at 20C. Nevertheless, the effect of temperature on fatigue life can be accounted for by its related effect on stiffness but with some loss in accuracy. In general, fatigue lives under controlled-strain loading are approximately 2.4 times greater than those under controlled-stress loading. When test results are interpreted in terms of the performance expected of the pavements in which they are placed, it appears that controlled-stress and controlled-strain testing may yield similar mix rankings provided they yield comparable estimates of flexural stiffness. Having laboratory test data on mixes is a necessary condition for characterizing fatigue behavior. However, laboratory testing must be interpreted using mechanistic analyses to determine how mixes are likely to perform in the pavement structure under anticipated traffic loads and environmental conditions. Acknowledgments The research reported herein was conducted as part of the Project A-003A of the Strategic Highway Research Program (SHRP). This project, entitled "Performance Related Testing and Measuring of Asphalt-Aggregate Interactions and Mixtures," is being conducted by the Institute of Transportation Studies of the University of California at Berkeley. SHRP is a unit of the National Research Council that was authorized by Section 128 of the Surface Transportation and Uniform Relocation Assistance Act of 1987. The laboratory testing was conducted at the University of California at Berkeley. Thanks are due to John Harvey under whose direction all the specimens were manufactured for these tests. Thanks are also due to Bor-Wen Tsai, Elie Abijaoude, K. S. Yapa, Thomas Mills, and all other laboratory technicians and helpers for their efforts in specimen preparation, testing, and analysis. Messrs. Clarence Chan, Ed Nicks, and Todd Merport provided the necessary support for the manufacture/redesign and repairs of the test equipment and the electronics. The statistical design of the laboratory testing program was prepared by the SHRP Project A-003A statistician, Mr. Lou Painter. Disclaimer This paper represents the views of the authors only and is not necessarily reflective of the views of the National Research Council, the views of SHRP, or SHRP's sponsor. The results reported here are not necessarily in agreement with the results of other SHRP research activities. They are reported to stimulate review and discussion within the research community. References 1. R.G. Hicks, J. A. Deacon, L. Painter, and C.L. Monismith, "Laboratory Study Plan for SHRP Project A-003A," Technical Memorandum 89-8, Prepared for Strategic Highway Research Program, Project A-003A, Institute of Transportation Studies, University of California, Berkeley, August 1989. 2. A. Tayebali, J. S. Coplantz, J. T. Harvey, and C. L. Monismith, "Interim Report on Fatigue Response of Asphalt-Aggregate Mixtures," Report Number: TM-UCB-A-003A-92-1, SHRP Project A-003A, Asphalt Research Program, Institute of Transportation Studies, University of California, Berkeley, February 1992. 3. A. Tayebali, J. A. Deacon, J. S. Coplantz, J. T. Harvey, and C. L. Monismith, "Fatigue Response of Asphalt-Aggregate Mixtures, Part I - Test Method Selection" SHRP Project A-003A, Asphalt Research Program, Institute of Transportation Studies, University of California, Berkeley, March 1993. Mix and Load Fatigue Effects 135 Figure A. Tensile Strain versus Thickness of Asphaltic Layer 4. A. Tayebali, J. A. Deacon, J. Coplantz, and C. L. Monismith, "Modeling Fatigue Response of Asphalt-Aggregate Mixtures," Journal of the Association of Asphalt Paving Technologists, Vol. 62, 1993. 5. United States Army Corp of Engineers, Engineering and Design Manual, Flexible Pavements, Airforce, HM1110-45-302, 1958. 6. L. Wilkinson, "SYSTAT: The System for Statistics," Evanston, IL, SYSTAT, Inc. 1990. Discussions MR. GEOFF ROWE: (Prepared Discussion) I would like to make some comments, which you could respond to, on the conclusion that the rankings from controlled stress and controlled strain testing are the same at the end of the evaluation using a mechanistic design method which incorporates a linear elastic model for the behavior of the pavement structure and on the use of an elastic analysis method. In order to do this I would like to show a few slides. Using the soil foundation properties for the 50-mm (2-in.) pavement structure quoted in the paper the tensile strain is calculated for a range of thicknesses. The results obtained are presented in Figure A. It can be observed that beyond the thickness of 50-mm (2-in.) the value of strain starts to reduce. What we see happening is that the tensile strain reduces as the pavement gets thicker and as you move from a thick to thin pavement we see that tensile strain increasing. However, in the range of thickness between 75-mm (3-in.) to 25-mm (1-in.), the tensile strain flattens off and then tends to reduce again. This is a function of the neutral axis moving higher up within the pavement structure. There is a need, when you have very thin pavements to be careful, about using elastic layered models because the calculated strain at the underside of the wheel may not necessarily be an adequate indicator of pavement performance. In addition there is often a switch in the position of the maximum tensile strain between the underside of a single wheel versus the centroid of a multi-wheel loading problem. Tayebali, Deacon, Coplantz, Harvey and Monismith 136 Figure B. Dissipated Energy Contour for a Typical Pavement Structure The second figure that I wish to show is basically a contour damage plot (Figure B). In this instance the damage is considered as a function of dissipated energy for the pavement structure. The analysis was conducted using a visco-elastic material model to compute the dissipated energy in the pavement structure. Dissipated energy, as you are aware, is considered as one the main indicators of fatigue performance from the results of the work which was performed during the SHRP A-003A contract. This figure shows for the conditions evaluated that fatigue cracking will start at the underside of the pavement structure where the dissipated energy is the greatest. Now let us consider what happens at the sides of wheel loadings and ask the question - where is the right place to look at the stresses and the strains? If we look at the strain at the surface (see Figure C) related to the distance from the wheel loading (in this case I have taken just one of your wheel loads rather than a dual wheel load) we see at the center of the loaded area the material is in compression and then as we move away from the loaded wheel area it goes into tension. When you are evaluating materials for overlays, this surface stress starts to become important. If we now consider a case (in terms of dissipated energy) where a stiff upper layer is placed over a less stiff material we can observe that the dissipated energy contours plotted (which are contours of damage) indicate that the highest fatigue is occurring at either side of the wheel load, see Figure D. This location is consistent with the high stresses presented in Figure C. The effect of the shear stresses that occur at the side of the wheel load is also considered in the dissipated energy calculations. Thus, if consideration is made of the dissipated energy and visco-elastic modeling of the entire pavement structure, rather than the tensile strain at the underside of the structure computed using an elastic analysis, the rankings obtained may actually change. What I am suggesting is that, if we analyzed the materials and pavement structure using a visco-elastic model, we would find that sometimes you need to have a soft elastic material as an overlay rather than a stiff material. The rankings will actually change depending on the type of model (elastic versus viscoelastic) that is used. In terms of thin overlays and thin pavement structures, I am suggesting that it is probably better to use soft elastic materials whereas for thick overlays and thick pavement structures the use of a stiffer material (which give you a lower tensile strain at the underside of the structure) is desirable. This whole aspect needs to be considered in terms of visco-elasticity. This concludes my discussion. Mix and Load Fatigue Effects 137 Figure C. Strain at Surface of Pavement due to a Single wheel Loading (Radius 101 mm) Figure D. Dissipated Strain Energy Contour for a Pavement with a Stiff Top Layer PROFESSOR A. TAYEBALI: Thank you, Geoff, for your prepared presentation. I would like to point out that this was the first part of our SHRP study and it was in a pulse loading. We did not have any viscoelastic layered programs. Even if we had, we did not have the correct input for that program. So it was not possible for us to look at the energy dissipation model. Also, if you see the results of this paper, it clearly indicates that stiffer layers are preferred for thick pavements and less stiff layers are preferred for thin pavements. Both modes of loading show the same thing. MR. ROWE: What I was trying to suggest is that in the paper the analysis is being regressed on strain for the control strain testing and stress for the controlled stress testing. You might find that soft elastic layers are going to give you better performance when you consider overlays and the rankings will sometimes change because of the problems with elastic analysis. How we normally look at the tensile strain at the bottom of the pavement structure might not necessarily be valid in all cases. Tayebali, Deacon, Coplantz, Harvey and Monismith 138 PROF. TAYEBALI: I agree. If you change the program in how you compute your stresses and strain things might change. If you also change your pavement structure, things might change too. We only looked at two layer structures. MR. RAJ DONGRE: I have a question out of curiosity. Your paper mainly dealt with mode of loading effects. You are loading to factor that if you are not careful you may be looking at the same thing but just because you tested it differently you are getting different results. This has been seen for other materials. One of the explanations given in the differences you see because of mode of loading is that when you keep the stress constant, your cracks increase and that stress that is being kept constant becomes more critical as the length of the crack increases. However, when you keep strain constant that doesn't happen. My question is when you say fatigue life, what is your criteria for determining that fatigue life? I have always wondered about this and no one has answered. It seems to me that if we changed our criteria for stress control, and then changed our criteria for strain control, we may get the fatigue life. Do you agree with that? PROF. TAYEBALI: Your first part of the question was correct. In control strain you also account for the crack propagation. That is why your life is actually higher than the control stress. MR. DONGRE: How do you define life? PROF. TAYEBALI: For control stress it is actually when the specimen fails. It is clearly defined. For control strain it is a 50 percent reduction in the modulus. MR. DONGRE: What I am trying to point out is that the differences you see may be due to the differences in the criteria you use. If you were to come up with a criteria that is independent of the mode of loading, then you may not see the differences. I am trying to point out that it is a mechanism related problem. If you study the mechanism of how the crack propagates in fatigue -- What Geoff pointed out in viscoelastic analysis, I don't know if a material is viscoelastic, non-linear, or linear the mode of loading will have the same effect if your criteria are not changed to fatigue resistance. Do you agree with that? MR. ROWE: Could I make a comment? Going back to paper that Aktar and I put together in 1990 at AAPT and some follow up work with SHRP. If you actually look at the way these materials are damaging, the actual reduction in modulus and the actual definitions of failure in a control stress test and control strain test are somewhat similar. I believe that these are both associated with initiation of a crack. You don't get much lifetime due to crack propagation in a control stress test because at the time that crack starts to propagate you get very high stress intensities factors and that crack is zooming through the material very quickly. The opposite thing occurs with the control strain test. You stop the test when a crack starts to propagate. So the actual definitions of failure are somewhat similar in the two tests. That might not be an issue. I think the mode of loading is an issue when you compare the test types. The factor is about right in terms of the work Aktar has done. In terms of an actual definition of failure in a test, I think they are probably fairly similar. MR. TAYEBALI: I agree with that, Geoff. MR. DONGRE: I don't agree with that but we'll just let it go. MR. RICHARD DAVIS: Do I correctly understand that your conclusions are entirely drawn on laboratory specimens at 0C and 20C? PROF. TAYEBALI: Yes. MR. DAVIS: I would like to point out that the pavement universe is much broader than that. What happens in the upper temperatures is much different from what happens in the lower temperatures. You may have healing in those Mix and Load Fatigue Effects 139 higher temperatures. The general idea of fatigue cracking is that it starts out as micro-cracks and they grow. If they are reversed at the higher temperatures, then, of course, you have to start over again. I particularly want to say that my experience in the field shows that softer asphalts are less likely to crack over time than are harder asphalts. I don't doubt the results of your tests that you ran at 0C and 20C, but I don't think we should feel that applies to the full pavement universe. PROF. TAYEBALI: Jack Deacon and I wrote a paper for TRB which showed that most of the cracking would occur at about 20C. MR. DAVIS: Pavements 50 years old with no fatigue cracking have existed and they had soft asphalt in them. I have looked at pavements for years and I have looked for why pavements crack. I nearly always found hard asphalt cracks rather than soft asphalt. Pell said just what you said and what everybody says that tests at the lower temperatures the harder asphalt gives longer fatigue life. I don't know about the experience of other people, but I have looked at pavements for many years. Those pavements that don't crack are likely to have softer asphalt. Those pavements that are all cracked up nearly always have harder asphalt in them. At the time Pell made his first talk at which I was present, I mentioned this fact. I have not seen anything to change my mind. I did not doubt what Dr. Pell's results were, I don't doubt yours, but I don't think that they necessarily apply to pavements over the full temperature regime. PROFESSOR CARL MONISMITH: Dick, if you were talking about 2 or 3 in. (51 or 75mm) thick asphalt bound layers on granular material, I would agree with you. There are Warrenite Bitulithic pavements constructed using 40-pen. asphalt or even a little harder; these pavements were at least 8 in. (203mm) in thickness. No cracking was observed. In addition, when evaluating fatigue we must look at where cracks start; there is substantive evidence that cracks start on the bottom of the pavement. If you have a very thin pavement, it is possible that those cracks might heal. However, dust is going to intrude into these cracks and preclude healing. Quite often we see what appears to be healing in the summer but the cracks open in the same spot where they occur when it becomes cooler on the surface of the pavement. As you get to thicker pavements, you have to realize that the temperature is not going to change as much a few inches down in the pavement as it does at the surface. Therefore, you are not going to have the temperatures at a level where you can expect healing. PROF. TAYEBALI: Let me answer a question also. You are saying that in thin pavements less stiff asphalts are preferable. Do I understand you correctly? MR. DAVIS: Carl may be right to a certain extent. I did not dig down 12 in. (305mm). PROF. TAYEBALI: Your original question was that for thin pavements you would prefer softer asphalts. MR. DAVIS: My experience indicates that if you look at older pavements which I have done for years you will find some pavements that have done very well and show no signs of distress. No cracking. I have dug into many of them and found softer asphalt. I have also dug into pavements, some of which may have as much as 40 penetration, and they were in good condition also. The real problem depends on the temperature regime when you get down below 30 pen. One of the things you have to watch out for with early pavements, some of the contained lake asphalt which had dust in it. The penetrations do not reflect the true softness of the binder. So you can find materials which had penetrations of even 10 or 15 even though the binder was quite soft. The filler in it reduced the penetration a lot. I have run across cases in older pavements where people think that the penetration was low when the actual binder there was quite soft. I do know, after talking to a lot of people, that Warren Brothers wanted as soft asphalt as they could get in their pavements. They made a strong point of that. It may have been after 40 or 50 years it was 40 pen. which is not so hard in the warm California climate. I don't know whether it had filler in it or not. I am not saying that my experience precludes anybody else's. I welcome other people's experience in this area. But ever since I first heard about fatigue I have heard that mixtures with harder binder have higher fatigue resistance as you mentioned. My actual experience in the field has not indicated that. Tayebali, Deacon, Coplantz, Harvey and Monismith 140 PROF. TAYEBALI: What I said was that for thick pavements stiffer mixes were superior and for thin pavements stiffer mixes were inferior. Conversely, for thin pavements, less stiff mixes were superior. That is about what you are concluding also. MR. ROWE: I just wanted to comment on the use of the Pell nomograph. Professor Peter Pell has been very much involved in the recent work. He has been one of my colleagues and supervisors in the work on dissipated energy. It is a function of pavement geometry and pavement thickness. In the UK where that work was done, the approach to pavement design, was to use a controlled stress test because the UK has very thick pavements. In a thick pavement situation when the cracking is principally controlled by the level of tensile strain at the underside of the pavement, stiff materials do give better performance. However, I do recognize what you say. If you get a situation where you have a thin pavement structure, aging and surface cracking can be a problem. There is a need to go to more soft and elastic binders. But in terms of the context of that nomograph and design method, that was correct for the situation as it was being applied. In the UK we are still using these very stiff bases. We are using 50 pens. and harder as the base material. They are giving very good performance compared to some of the softer materials we were using 20 years ago.