Development of an Artificial Neural Network-Based Procedure for the Verification of Traffic Speed Deflectometer Measurements

Abstract

Since the 1980s, the falling weight deflectometer (FWD) has been the primary deflection-measuring device in the United States to evaluate the structural conditions of in-service pavements. However, the stop and go nature of the FWD limits its application at the network level. In the early 2000s, the traffic speed deflectometer (TSD) was introduced as an alternate deflection-measuring device for network-level applications. TSD collects deflection measurements while traveling at traffic speed, which provides improved spatial coverage and no traffic disturbance. The verification of TSD measurements is of great interest as many agencies move toward widespread implementation. This study aims at developing a reliable and straightforward procedure for the verification of TSD measurements using limited FWD measured deflection measurements. The verification procedure employs a trained artificial neural network (ANN) model to shift TSD deflections to their corresponding FWD deflections. The ANN model was trained and verified based on FWD and TSD measurements from two deflection-testing programs. The developed model accurately predicted FWD measurements with a coefficient of determination (R2) of 0.994. The suitability of the proposed verification procedure was evaluated using statistical and engineering-based measures and showed acceptable accuracy. Results also validated that the proposed method could be used to verify TSD measurements before its use for conducting deflection measurements at the network level.

Performance Evaluation of Traffic Speed Deflectometer Based on Virtual Standard Test Road

“…Research by Li et al [7] showed that the TSD had a high correlation with BBD measurements, satisfying the requirements of road network-level deflection measurement. Liao et al [1] and Abohamer et al [2] declared that the correlation coefficients of TSD and FWD measurements reached 0.945 and 0.972, respectively. Nielsen et al [10] concluded that the FWD and TSD measurements, although similar, were not equivalent, and their details differed.…”

“…Earlier, instruments such as the Benkelman Beam Deflectometer (BBD) and the Falling Weight Deflectometer (FWD) were used to measure it. However, such instruments need to stop for 1-2 minutes during each test and suffer from poor safety and low efficiency [2], [3]. Currently, the focus of researchers has gradually shifted to high-speed, continuous, and non-destructive measurements [4], [5], [6].…”

“…In particular, the load of traditional instruments such as FWD is relatively constant when working, and its values are independent of speed, road roughness, and other factors. In contrast, the fluctuation amplitude of TSD is as high as 33% of that under the static load [2], [19]. Although the values may vary, there is a strong correlation between FWD and TSD measurements.…”

“…For example, Manoharan et al [20] even used a linear model to create the functional relationship between TSD and FWD measurements. In addition, some scholars have also built artificial neural network (ANN) models to train and predict the measured values from instruments with different measuring principles [2], [5], [21].…”

“…All these current situations pose challenges to the efficient evaluation of TSD. Fortunately, many in-service roads have been used for comparative test analysis of TSD [2], [15], [25]. Therefore, the core of this research is to construct the ideal ASTRs through the actual in-service roads and propose scientific evaluation methods and indexes.…”