Advanced Computing Techniques in Structural Evaluation of Flexible Pavements Using the Falling Weight Deflectometer
Type of Degreedissertation
MetadataShow full item record
The use of nondestructive deflection testing has become an integral part of the structural evaluation and rehabilitation process of pavements in recent years. The falling weight deflectometer (FWD) is commonly used to obtain material properties that can be used in mechanistic-empirical pavement design. These properties are currently obtained through an iterative process called backcalculation which has several limitations with one of the most notorious: the non-uniqueness of the results. The use of artificial neural networks (ANN) is currently being studied as a more reliable methodology and an advanced alternative. In addition, the loading frequency of the FWD impact loading can be considered similar to that of vehicle loading at a high speed. Hence, significant error could result between calculated from FWD and measured strain responses from traffic loads at operational speeds. The objectives of this study were to develop neural networks capable of predicting pavement layer moduli rapidly and reliably; and to determine correction factors for the high frequency/high speed FWD pavement responses to typical operating speed responses. The deflection basin database from the 2009 structural sections at the NCAT Test Track and the FWD test results from a section of the low volume route Lee 159 were used for verification of ANN models. The software 3D-Move was used to determine pavement theoretical response correction factors that were applied to actual values. The results indicated that the backcalculation process tended to overestimate the moduli of the asphalt concrete layer while the moduli of the subgrade were little or not affected. Besides the significant reduction in computed errors, the use of ANNs showed a clear advantage over conventional backcalculation: a couple of seconds to obtained ANN outputs versus minutes to hours from backcalculation. The capability for ANNs to predict pavement layer moduli was validated using multiple load levels and full slip condition as a layer interaction. This presented a clear advantage over previous studies that have been focused on one load level and full bond conditions. The analysis of measured versus predicted pavement responses indicated that significant errors can be obtained from using high speed/high frequency FWD backcalculated moduli to predict highway speed pavement responses. Therefore, correction factors should be applied on pavement responses from backcalculated moduli to represent highway speed loads. For the conditions and scenarios evaluated in this study, correction factors helped close the gap between measured and predicted pavement responses.