MODAL CHARACTERIZATION AND ROAD ROUGHNESS RECONSTRUCTION USING DYNAMIC VEHICLE ACCELERATIONS AND ANNS

Abstract

Road networks are considered huge and critical infrastructures that support the development and growth of societies. These infrastructures deteriorate over time due to regular usage and/or external environmental factors. Deteriorating road networks eventually cause ride discomfort for their users and the production of ground-borne noise and vibrations. Thus, maintaining these infrastructures is essential, and monitoring the condition of the roads is one of the most important steps in maintaining these road networks. Road roughness could be considered as one of the main indicators of the road's overall health. The International Roughness Index (IRI) is used to describe the road roughness profile numerically in a single value. Traditionally the IRI is obtained through manual or automated profilometers, profilographs, or dipstick profilers, which could be time/money consuming. Therefore, this study investigates the ability of Artificial Neural Networks (ANNs) in reconstructing road roughness profiles from dynamic vehicle accelerations. This study also investigates the ANNs ability to predict the model characteristics of a 7-DOF Full Car (FC) model, which is constructed to extract the dynamic vehicle accelerations of a vehicle moving on various roughness profiles. First, the FC model will be moving over a certain obstacle so that the developed ANN could take the dynamic vehicle accelerations as inputs and predict the FC model characteristics. Once the model characteristics are obtained, another ANN will be trained using the dynamic vehicle acceleration of an FC model with the same characteristics to reconstruct the road roughness profile.