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.
150 Analysis Method-3 applied an artificial neural network (ANN) analysis process to a broad array of climate, traffic, and pavement input values against the recorded LTPP performance over time. The input and output dataset used to achieve Analysis-2 and Analysis-3 results were the same. The final ANN solution was software driven using the most significant variables that influence pavement performance. As-constructed AV were always included as an input variable regardless of its degree of significance. With input values for a specific pavement, Analysis-3 generates a family of curves showing the incremental influence of as-constructed AV on the selected performance characteristic (y-axis) over time (x-axis). Each analysis method involved some interpretation and generalized assumptions of the data from LTPP to maintain a reasonable dataset to accomplish the analysis. Too much refinement of the required data diminishes the size of each group of data. Below is a list of interpretations and assumptions made regarding the data. ⢠Rutting is predominantly influenced by the surface layer of asphalt mixture. The research team recognizes that rutting could be related to pavement structure (base/subgrade rutting) but did not examine the rutting data to the extent needed to separate rutting type. ⢠All wheel path cracking is related to fatigue cracking on new construction sections. The research team acknowledges that there are other causes of cracking in the wheel path, such as asphalt layer delamination and construction segregation, but did not examine the cracking data to the extent needed to identify the cause of wheel path cracking. ⢠All transverse cracking is related to low-temperature thermal cracking on new construction sections. The research team recognizes that there are other causes of transverse cracking but did not examine the cracking data to the extent needed to determine if the cracking was not related to low-temperature thermal cracking. ⢠All cracking on pavement rehabilitation sections is reflective cracking. The research team recognizes that fatigue and thermal cracking could occur on pavement rehabilitation sections but did not examine the pre- and post- pavement condition to determine the cause of the cracking. Analysis Method-2 included a model parameter for pre- rehabilitation cumulative traffic but it is only a placeholder for future research. Study Conclusions Based on the efforts to complete the NCHRP 20-50(18) study, the following conclusions were made. These conclusions are solely based on the analysis of LTPP sections, which were not constructed for the purpose of examining the influence of as-constructed AV. The research team anticipated that the effect of as-constructed AV would improve pavement performance as the value reduced from 9% to 4% and may have a negative effect at values below 4%. A number of laboratory performance tests require the asphalt mixture specimens to be compacted to 7% AV to standardize the response to the test with the knowledge that the test results would show improvement if the specimens were compacted to 4%. Using this laboratory perspective, the research team applied the following general terms to the results of the analysis. ⢠Meets Expectation: The performance improved as as-constructed AV decreased and performance may decline at air void levels below 4%.
151 ⢠No Influence: The performance would be similar across the entire range of AV. ⢠Contradicts Expectation: The performance declined as AV decreased. On a broad basis, lower as-constructed AV do have a positive effect (meet expectation) on the performance of an asphalt pavement, but the effect was not consistent between pavement demographics (climate, traffic, and pavement structure), types of pavements (new construction and rehabilitation), and performance characteristics (rutting, fatigue cracking, thermal cracking, and ride). Further, the models were predominantly developed with low to moderate traffic and caution is needed when applying the models to parameters outside the range of the LTPP dataset. The results of Analysis Method 1 were based on a subjective examination of 27 subgroups of LTPP sections. Analysis Method 1 concluded that the influence of as-constructed AV was mixed. ⢠For rutting, 62% of the subgroups met expectation and 12% contradicted expectation. ⢠For fatigue cracking, 62% met expectation and 19% contradicted expectation. ⢠For thermal cracking, 46% met expectation and 50% contradicted expectation. ⢠For ride, 40% met expectation and 40% contradicted expectation. The results of Analysis Method 2 were based on separate regression models developed for eight combinations of pavement type (new construction and rehabilitation) and performance characteristic (rutting, fatigue cracking, thermal cracking, and ride). A set of average input variables and increments of as-constructed AV were entered into the models to create a series of predicted performance curves to assess the influence of AV. Analysis Method 2 concluded that the influence of as-constructed AV was mixed. ⢠For rutting of new construction, the regression model prediction nominally contradicted the expectation. ⢠For rutting of rehabilitation, the regression model prediction minimally contradicted the expectation. ⢠For fatigue of new construction, the regression model prediction significantly met the expectation. ⢠For fatigue of rehabilitation, the regression model prediction nominally met the expectation. ⢠For thermal cracking of new construction, the regression model prediction nominally met the expectation. ⢠For thermal cracking of rehabilitation, the regression model prediction nominally met the expectation. ⢠For ride of new construction, the regression model prediction nominally met the expectation. ⢠For ride of rehabilitation, the regression model prediction nominally met the expectation. The results of Analysis Method 3 were based on separate ANN models developed for eight combinations of pavement type and performance characteristic. Increments of as- constructed AV were applied with the climate, traffic, and other material inputs for each LTPP
152 section to create a series of predicted performance curves to assess the global influence of as- constructed AV. Analysis Method 3 concluded that the influence of as-constructed AV was mixed. ⢠For rutting of new construction, the ANN model prediction curves nominally met the expectation. ⢠For rutting of rehabilitation, the ANN model prediction showed no practical influence. ⢠For fatigue of new construction, the ANN model prediction only met the expectation at high as-constructed AV. ⢠For fatigue of rehabilitation, the ANN model prediction significantly met the expectation. ⢠For thermal cracking of new construction, the ANN model prediction showed no practical influence. ⢠For thermal cracking of rehabilitation, the ANN model prediction nominally met the expectation. ⢠For ride of new construction, the ANN model prediction nominally met the expectation. ⢠For ride of rehabilitation, the ANN model prediction nominally met the expectation. A validation of the three analysis methods using external datasets also had mixed results. One validation approach found that rutting and fatigue trends agreed (met expectation) with Analysis Method 1 for the climate 2 subgroup but differed for the climate 4 subgroup. The other validation approach concluded that Analysis Method 2 regression models were a fair fit for the LTPP dataset and were applicable to the external validation dataset for several cases. On the other hand, Analysis Method 3 artificial neural network models were a significantly better fit for the LTPP dataset but were not applicable to the external validation dataset. The validation used datasets that were predominantly on the upper end of the LTPP dataset range and could be outside the reasonable application of the models.
