U.S. Department of Transportation
Federal Highway Administration
1200 New Jersey Avenue, SE
Washington, DC 20590
2023664000
Federal Highway Administration Research and Technology
Coordinating, Developing, and Delivering Highway Transportation Innovations
REPORT 
This report is an archived publication and may contain dated technical, contact, and link information 

Publication Number: FHWAHRT10035 Date: September 2011 
Publication Number: FHWAHRT10035 Date: September 2011 
To assess differences in the measured moduli determined from the AMPT and TP62 protocols, a joint study was carried out between researchers at the TurnerFairbank Highway Research Center (TFHRC) and NCSU. For this study, TFHRC performed dynamic modulus testing on a mixture following the AMPT TP, and NCSU performed testing on the same mixture using the TP62 protocol.^{(8)} In both cases, three replicates have been tested. To reduce any variability not related to the equipment and protocols, all specimens were fabricated at NCSU and randomly sampled for either AMPT testing or TP62 testing. The details of each testing protocol are summarized in table 43.
Factor  AMPT  TP62 

Temperature (°F)  40, 70, 100, and 130  14, 40, 70, 100, and 130 
Frequency (Hz)  20, 10, 5, 1, 0.5, and 0.1  25, 10, 5, 1, 0.5, and 0.1 
Microstrain target  75–125  50–75 
LVDT gauge length (mm)  70  100 
Load direction  Bottom loading  Top loading 
End treatment  Teflon^{®}  Greased double latex membranes 
Conditioning  External temperature chamber, then equalize in AMPT for 3 min  Equalize for 2.5–3.0 h in test machine 
Rest period between frequencies (s)  0  300 
Calculations  NCHRP 0929 final 10 cycles^{(50)}  NCHRP 0929 final five cycles^{(50)} 
°C = (°F−32)/1.8 
The mixture used for this purpose is a 0.371inch (9.5mm) Superpave™ mixture typically used in North Carolina for surface courses. The gradation of this mixture is given in figure 133, and the relevant volumetric properties are summarized in table 44. All tests were conducted at 5.9 percent ±0.1 percent air void levels.
Figure 133. Graph. Test mixture gradation.
Volumetric Property  Mix Design  Test Samples 

V_{a} (percent)  3.8  5.9 
VMA (percent)  15.6  17.5 
VFA (percent)  75.7  66.2 
Asphalt content (percent)  5.2  5.2 
Percent effective binder content  4.9  4.9 
Dust percentage  1.2  1.2 
G_{mm}  2.616  2.616 
Bulk specific gravity of the aggregate  2.828  2.828 
Effective specific gravity of the aggregate  2.855  2.855 
G_{b}  1.035  1.035 
Results from the experimental study are summarized in figure 134 and figure 135, where the average dynamic moduli from the TP62 protocol are plotted against the average moduli from the AMPT protocol. Error bars in these figures represent a single standard deviation from the mean. From these figures, it is observed that the AMPT test results are systematically lower than those from the TP62 protocol; the difference between the two datasets is approximately 13 percent. Statistical analysis of these values using the stepdown bootstrap method has also been performed. This method is used in lieu of multiple paired ttests due to the effect of experimentwise error rates, which results in statistical errors when making multiple comparisons. Specifically, failing to account for this error rate increases the probability of finding significance when none is present. The statistical analysis results are shown by temperature and frequency in table 45. Note that in this table, the conditions under which the means are statistically similar are bold.
Figure 134. Graph. Comparison of E* measured via AMPT and TP62 protocols in arithmetic scale.
Figure 135. Graph. Comparison of E* measured via AMPT and TP62 protocols in logarithmic scale.
Temperature (°C)  Frequency (Hz)  E* AMPT (psi)  E* TP62 (psi)  pValue 

4  25.00  2,145,226  2,420,540  0.032 
4  10.00  1,989,606  2,284,746  0.020 
4  5.00  1,838,144  2,111,129  0.030 
4  1.00  1,503,747  1,726,774  0.026 
4  0.50  1,359,729  1,601,117  0.019 
4  0.10  1,050,375  1,234,431  0.023 
21  25.00  1,030,409  1,237,696  0.020 
21  10.00  899,831  1,022,446  0.023 
21  5.00  785,545  881,347  0.025 
21  1.00  550,882  628,569  0.033 
21  0.50  468,842  524,847  0.068 
21  0.10  306,841  358,120  0.057 
37  25.00  385,448  464,233  0.008 
37  10.00  318,540  384,219  0.010 
37  5.00  263,476  330,282  0.002 
37  1.00  160,938  198,110  0.008 
37  0.50  130,346  167,580  0.005 
37  0.10  75,190  96,587  0.011 
54  25.00  153,735  177,050  0.003 
54  10.00  127,039  128,097  0.801 
54  5.00  102,669  101,164  0.672 
54  1.00  58,086  59,737  0.377 
54  0.50  42,997  48,863  0.022 
54  0.10  23,863  33,547  0.005 
°C = (°F−32)/1.8 
To assess the differences observed between the two E* measurement protocols, a more comprehensive analysis was performed using the databases available in this study. The two AMPT and TP62 databases were segregated based on the temperatures at which the E* values were measured. Because these two databases cover different ranges of parameters, it is useful to examine the distribution of the relevant parameters for the two databases. Figure 136 through figure 157 present the distribution and range of each parameter in the two databases. In figure 158 through figure 162, the measured E* data points available for some specific temperatures for each type of database are shown by frequency. Based on observations from these figures and the difference equation shown in equation 100, differences between the databases containing AMPT and TP62 measurements are evident, as can be seen in table 46.
(100) 
Based on this description, the following differences are observed at each temperature:
Figure 136. Graph. Frequency distribution of temperature in AMPT versus TP62 databases.
Figure 137. Graph. Range of temperature in AMPT versus TP62 databases.
Figure 138. Graph. Frequency distribution of frequency in AMPT versus TP62 databases.
Figure 139. Graph. Range of loading frequency in AMPT versus TP62 databases.
Figure 140. Graph. Frequency distribution of percentage retained on ¾inch (19.05mm) sieve (ρ_{34}) in AMPT versus TP62 databases.
Figure 141. Graph. Range of percentage retained on ¾inch (19.05mm) sieve (ρ_{34}) in AMPT versus TP62 databases.
Figure 142. Graph. Frequency distribution of percentage retained on ^{3}/_{8}inch (9.56mm) sieve (ρ_{38}) in AMPT versus TP62 databases.
Figure 143. Graph. Range of percentage retained on ^{3}/_{8}inch (9.56mm) sieve (ρ_{38}) in AMPT versus TP62 databases.
Figure 144. Graph. Frequency distribution of percentage retained on #4 sieve (ρ_{4}) in AMPT versus TP62 databases.
Figure 145. Graph. Range of percentage retained on #4 sieve (ρ_{4}) in AMPT versus TP62 databases.
Figure 146. Graph. Frequency distribution of percentage passing #200 sieve (ρ_{200}) in AMPT versus TP62 databases.
Figure 147. Graph. Range of percentage passing #200 sieve (ρ_{200}) in AMPT versus TP62 databases.
Figure 148. Graph. Frequency distribution of specimen air voids in AMPT versus TP62 databases.
Figure 149. Graph. Range of specimen air voids in AMPT versus TP62 databases.
Figure 150. Graph. Frequency distribution of effective binder volume in AMPT versus TP62 databases.
Figure 151. Graph. Range of effective binder volume in AMPT versus TP62 databases.
Figure 152. Graph. Frequency distribution of VMA in AMPT versus TP62 databases.
Figure 153. Graph. Range of VMA in AMPT versus TP62 databases.
Figure 154. Graph. Frequency distribution of VFA in AMPT versus TP62 databases.
Figure 155. Graph. Range of VFA in AMPT versus TP62 databases.
Figure 156. Graph. Frequency distribution of G* n AMPT versus TP62 databases.
Figure 157. Range of G* in AMPT versus TP62 databases.
Figure 158. Graph. Percentage of difference between AMPT versus TP62 databases based on similar ranges of different variables at 39.9 °F (4.4 °C).
Figure 159. Graph. Percentage of difference between AMPT versus TP62 databases based on similar ranges of different variables at 69.9 °F (21.1 °C).
Figure 160. Graph. Percentage of difference between AMPT versus TP62 databases based on similar ranges of different variables at 100 °F (37.8 °C).
Figure 161. Graph. Percentage of difference between AMPT versus TP62 databases based on similar ranges of different variables at 129.2 °F (54.0 °C).
Figure 162. Graph. Percentage of difference between AMPT versus TP62 databases based on similar ranges of different variables at 129.9 °F (54.4 °C).
Temp (°F)  0 ≤ ρ_{34} ≤ 15  5 ≤ ρ_{38} ≤ 50  30 ≤ ρ_{4} ≤ 70  3 ≤ ρ_{200} ≤ 7  5 ≤ V_{a} ≤ 9  8 ≤ V_{beff} ≤ 14  12 ≤ VMA ≤ 20  50 ≤ VFA ≤ 80  1e2 ≤ G* ≤ 1e5 

40  46.08  39.80  41.14  43.54  42.75  45.65  44.81  44.29  43.60 
70  59.39  47.54  51.74  57.66  57.67  60.23  59.91  58.32  57.84 
100  62.63  49.53  51.35  61.61  63.38  64.49  64.36  62.66  61.95 
129  45.60  51.02  49.65  46.26  N.A  63.01  46.16  52.09  46.26 
130  57.55  40.76  44.14  57.46  60.50  59.99  60.54  57.93  57.83 
°C = (°F−32)/1.8 
Similar ranges of each variable have been considered for each temperature, and the percentage of error has been calculated based on the difference of average TP62 versus AMPT E* measurements for the corresponding temperature.
A preliminary study was conducted to determine the feasibility and predictability of the ANN modeling technique relative to the existing models. This feasibility study was first conducted based on G* because more existing closedform models use this parameter as their primary input parameter. The ANN models used in this preliminary study are not the final models suggested by the research team, but they are similar in form and validation. To ensure full coverage of the expected conditions, the most recent Witczak database with available measured G* data and a portion of the dataset obtained at NCSU with support from the NCDOT were utilized as the TP62 training database. Also, appropriate portions of the FHWA mobile trailer database and the WRI database (from Kansas and Nevada sites) were considered as the AMPT training database (see table 47).^{(51,52)} New parameters were not identified through this study. Instead, only those that have been used in the modified Witczak model are incorporated. For verification purposes, three different sets of independent databases were used (see table 48). As a corollary to this study, an additional ANN model was trained that uses the Hirsch model input parameters. The results from this model are given in this section, as well.
Type of Database  AMPT  TP62  Total  

FHWA I  WRI  Witczak  NCDOT I  
Number of mixtures  409  24  106  24  563 
Number of data points  7,827  500  3,180  644  12,151 
Number of binders  13  8  17  5  43 
Number of gradation variations  13  12  13  19  57 
Number of volumetric variations  256  13  98  24  391 
Note: FHWA I consists of the mixtures from 12 States 
Type of Database  AMPT  TP62  Total  

FHWA II  Citgo  NCDOT II  
Number of mixtures  84  8  12  104 
Number of data points  1,652  168  338  2,158 
Number of binders  3  2  3  8 
Number of gradation variations  3  1  12  16 
Number of volumetric variations  75  1  12  88 
Note: FHWA II consists of the mixtures from three States in the FHWA mobile trailer database with the following site IDs: 1IA0358, 2WA0463, and 3KS464. 
It should be noted that the two TPs, AMPT and TP62, were used to measure the E* values in the various databases. To illustrate any possible differences between the two protocols, three different ANNs were developed using the Witczakbased input parameters, as shown in table 49. GGR pANN was trained using data from both the AMPT and TP62 protocols, whereas AMPT pANN and TP62 pANN models were trained using the data from AMPT only and TP62 only. Table 49 summarizes the databases used to train and verify the ANNs.
Model  Data Used in ANN Training  Description  Reference Scale  Statistical Parameters for Training Data  Statistical Parameters for Verification Data 


AMPT  TP62  FHWA II  NCDOT II  Citgo  
GGR pANN  FHWA I  Witczak  ANNs trained with modified Witczak parameters  Arithmetic  Se/Sy = 0.29 R^{2} = 0.92  Se/Sy = 0.38 R^{2 }= 0.86  Se/Sy = 0.33 R^{2} = 0.97  Se/Sy = 0.52 R^{2} = 0.94  
WRI  NCDOT I  Log  Se/Sy = 0.15 R^{2} = 0.98  Se/Sy = 0.35 R^{2} = 0.91  Se/Sy = 0.27 R^{2} = 0.96  Se/Sy = 0.59 R^{2} = 0.96  
AMPT pANN  FHWA I  Arithmetic  Se/Sy = 0.24 R^{2} = 0.94  Se/Sy = 0.36 R^{2} = 0.91  Se/Sy = 0.63 R^{2} = 0.87  Se/Sy = 0.37 R^{2} = 0.88  
WRI  Log  Se/Sy = 0.16 R^{2} = 0.97  Se/Sy = 0.38 R^{2} = 0.90  Se/Sy = 0.60 R^{2} = 0.89  Se/Sy = 0.48 R^{2} = 0.91  
TP62 pANN  Witczak  Arithmetic  Se/Sy = 0.34 R^{2} = 0.88  Se/Sy = 2.08 R^{2} = 0.77  Se/Sy = 0.24 R^{2} = 0.95  Se/Sy = 1.20 R^{2} = 0.97  
NCDOT I  Log  Se/Sy = 0.18 R^{2} = 0.97  Se/Sy = 0.99 R^{2} = 0.82  Se/Sy = 0.27 R^{2} = 0.93  Se/Sy = 0.53 R^{2} = 0.99  
Modified Witczak Model 
Arithmetic  Se/Sy = 0.92 R^{2} = 0.91  Se/Sy = 0.71 R^{2} = 0.91  Se/Sy = 0.64 R^{2} = 0.98  
Log  Se/Sy = 0.58 R^{2} = 0.92  Se/Sy = 0.19 R^{2} = 0.98  Se/Sy = 0.26 R^{2} = 0.99  
Hirsch Model  Arithmetic  Se/Sy = 0.30 R^{2} = 0.92  Se/Sy = 0.47 R^{2} = 0.97  Se/Sy = 0.11 R^{2} = 0.99  
Log  Se/Sy = 0.39 R^{2} = 0.92  Se/Sy = 0.26 R^{2} = 0.97  Se/Sy = 0.09 R^{2} = 0.99  
AlKhateeb Model  Arithmetic  Se/Sy = 0.48 R^{2 }= 0.89  Se/Sy = 0.55 R^{2} = 0.93  Se/Sy = 0.36 R^{2} = 0.93  
Log  Se/Sy = 0.43 R^{2 }= 0.92  Se/Sy = 0.40 R^{2} = 0.93  Se/Sy = 0.17 R^{2} = 0.97  
Note: Blank cells indicate information is not applicable. 
The ANN models perform well, as shown in figure 163 to figure 180, which display the prediction accuracies of the different models for the combined AMPT and TP62 data (figure 163 to figure 168), TP62 data only (figure 169 to figure 174), and AMPT data only (figure 175 to figure 180). Also, these three groups of figures show the prediction accuracies of the ANNs separately. In these three figures, the type of data (i.e., AMPT versus TP62) used in the ANN training matches the type of data used in the verification (e.g., figure 163 shows the prediction accuracy of the GGR pANN model trained with the combined AMPT and TP62 data on the combined AMPT and TP62 data, etc.). It is noted that the data used in these figures were not included in the ANN training.
Figure 181 through figure 204 further demonstrate the differences between the AMPT and the TP62 data and their effect on the prediction accuracies of the different ANNs. FHWA II data used in figure 181 through figure 188 are obtained using the AMPT protocol. The TP62 pANN model trained with the TP62 data and the modified Witczak model overpredict the measured E* values. Figure 189 through figure 196 present the prediction results for the NCDOT II data, which were measured using the TP62 protocol. These figures illustrate the opposite effect on the prediction bias, that is, the effect of using the TP62 data in the ANN training and predicting the AMPT data. In this case, the AMPT pANN model, trained using the AMPT data, underpredicts the E* values. The GGR pANN model provides a promising ANNbased E* model, and the TP62 pANN model shows good predictions without any significant bias. With the exception of the Citgo dataset, the GGR pANN model provides high goodness of fit and correlation, as seen in table 49. The promising feature of the GGR pANN model is that it improves the bias of E* predictions, particularly at high and low temperatures. This new ANN model is more sensitive to, and thus more likely to capture, the changes in volumetric parameters than all the other existing predictive models.
The findings from figure 163 to figure 204 are summarized as follows:
Figure 163. Graph. Prediction of the combination of AMPT and TP62 data using the modified Witczak and GGR pANN models in arithmetic scale.
Figure 164. Graph. Prediction of the combination of AMPT and TP62 data using the modified Witczak and GGR pANN models in logarithmic scale.
Figure 165. Graph. Prediction of the combination of AMPT and TP62 data using the Hirsch model in arithmetic scale.
Figure 166. Graph. Prediction of the combination of AMPT and TP62 data using the Hirsch model in logarithmic scale.
Figure 167. Graph. Prediction of the combination of AMPT and TP62 data using the AlKhateeb model in arithmetic scale.
Figure 168. Graph. Prediction of the combination of AMPT and TP62 data using the AlKhateeb model in logarithmic scale.
Figure 169. Graph. Prediction of the AMPT data using the modified Witczak and AMPT pANN models in arithmetic scale.
Figure 170. Graph. Prediction of the AMPT data using the modified Witczak and AMPT pANN models in logarithmic scale.
Figure 171. Graph. Prediction of the AMPT data using the Hirsch model in arithmetic scale.
Figure 172. Graph. Prediction of the AMPT data using the Hirsch model in logarithmic scale.
Figure 173. Graph. Prediction of the AMPT data using the AlKhateeb model in arithmetic scale.
Figure 174. Graph. Prediction of the AMPT data using the AlKhateeb model in logarithmic scale.
Figure 175. Graph. Prediction of the TP62 data using the modified Witczak and TP62 pANN models in arithmetic scale.
Figure 176. Graph. Prediction of the TP62 data using the modified Witczak and TP62 pANN models in logarithmic scale.
Figure 177. Graph. Prediction of the TP62 data using the Hirsch model in arithmetic scale.
Figure 178. Graph. Prediction of the TP62 data using the Hirsch model in logarithmic scale.
Figure 179. Graph. Prediction of the TP62 data using the AlKhateeb model in arithmetic scale.
Figure 180. Graph. Prediction of the TP62 data using the AlKhateeb model in logarithmic scale.
Figure 181. Graph. Prediction of the FHWA II data using the modified Witczak and GGR pANN models in arithmetic scale.
Figure 182. Graph. Prediction of the FHWA II data using the modified Witczak and GGR pANN models in logarithmic scale.
Figure 183. Graph. Prediction of the FHWA II data using the AMPT pANN and TP62 pANN models in arithmetic scale.
Figure 184. Graph. Prediction of the FHWA II data using the AMPT pANN and TP62 pANN models in logarithmic scale.
Figure 185. Graph. Prediction of the FHWA II data using the Hirsch model in arithmetic scale.
Figure 186. Graph. Prediction of the FHWA II data using the Hirsch model in logarithmic scale.
Figure 187. Graph. Prediction of the FHWA II data using the AlKhateeb model in arithmetic scale.
Figure 188. Graph. Prediction of the FHWA II data using the AlKhateeb model in logarithmic scale.
Figure 189. Graph. Prediction of the NCDOT II data using the modified Witczak and GGR pANN models in arithmetic scale.
Figure 190. Graph. Prediction of the NCDOT II data using the modified Witczak and GGR pANN models in logarithmic scale.
Figure 191. Graph. Prediction of the NCDOT II data using the AMPT pANN and TP62 pANN models in arithmetic scale.
Figure 192. Graph. Prediction of the NCDOT II data using the AMPT pANN and TP62 pANN models in logarithmic scale.
Figure 193. Graph. Prediction of the NCDOT II data using the Hirsch model in arithmetic scale.
Figure 194. Graph. Prediction of the NCDOT II data using the Hirsch model in logarithmic scale.
Figure 195. Graph. Prediction of the NCDOT II data using the AlKhateeb model in arithmetic scale.
Figure 196. Graph. Prediction of the NCDOT II data using the AlKhateeb model in logarithmic scale.
Figure 197. Graph. Prediction of the Citgo data using the modified Witczak and GGR pANN models in arithmetic scale.
Figure 198. Graph. Prediction of the Citgo data using the modified Witczak and GGR pANN models in logarithmic scale.
Figure 199. Graph. Prediction of the Citgo data using the AMPT pANN and TP62 pANN models in arithmetic scale.
Figure 200. Graph. Prediction of the Citgo data using the AMPT pANN and TP62 pANN models in logarithmic scale.
Figure 201. Graph. Prediction of the Citgo data using the Hirsch model in arithmetic scale.
Figure 202. Graph. Prediction of the Citgo data using the Hirsch model in logarithmic scale.
Figure 203. Graph. Prediction of the Citgo data using the AlKhateeb model in arithmetic scale.
Figure 204. Graph. Prediction of the Citgo data using the AlKhateeb model in logarithmic scale.