Reconstruction of Road Defects from Dynamic Vehicle Accelerations by Using the Artificial Neural Networks

Abstract

Monitoring of roads is considered the first step in establishing a successful road maintenance program, which includes scheduling adequate maintenance to a certain road section at the right time. Road monitoring assesses the road's condition for later analysis, which helps the prioritization of maintenance activities. Optimizing the process of scheduling and prioritizing maintenance activities is driven by cost-efficiency and is critical for preserving the overall road networks' health and smooth operation. However, the implementation of road monitoring, using traditional equipment such as profilometers, total stations, and automated road meters, could be exceedingly time-consuming. This research proposes the usage of Artificial Neural Networks (ANNs) and dynamic vehicle accelerations to reconstruct road defects in a time/cost-effective manner. A multi-degree of freedom numerical model is used to simulate the dynamics of a vehicle passing on various road defects such as potholes and speed bumps. These road defects are reconstructed by using two different methods. The first method employs a frequency domain approach to inversely reconstruct the road defects numerically. The second method uses ANNs to reconstruct the road defects with time histories of the vehicle's acceleration as the input to the network. The ANN model was trained by dynamic vehicle accelerations as inputs and various road defect profiles as outputs. Both methods were later compared on the basis of their accuracy and efficiency, and the ANN method was found to be more promising for implementation on experimental data.