PREDICTING ROAD ROUGHNESS PROFILE USING DYNAMIC VEHICLE ACCELERATIONS AND ARTIFICIAL NEURAL NETWORKS

Abstract

Road roughness can cause ride discomfort and contribute to the emission of ground-borne noise and vibration. A robust monitoring regime of road roughness is essential for the efficient maintenance of a country's road network. The International Organization for Standardization (ISO) presents an International Roughness Index (IRI) in Standard ISO 8606 to classify road roughness into eight different categories ranging from class A to class H. ISO 8606 summarizes the level of road roughness into a single number that could be used to describe the state of the road. The IRI can be measured using well-established techniques such as profilometers, profilographs, and automated road meters. However, these traditional techniques are time-consuming and/or not cost-effective. We propose using vehicle acceleration and artificial neural networks (ANNs) as an efficient and cost-effective alternative to the aforementioned techniques. We use a multi-degree of freedom model to simulate the dynamics of a vehicle traveling on a road with various road roughness classes. The road profile will be predicted using two methods. The first method uses an ANN which was trained using a library of inputs (dynamic vehicle accelerations) and outputs (road profiles), to predict the road profile. In the second method, we formulate the inverse problem in the frequency domain to obtain the road profile from the dynamic vehicle accelerations. The second method will serve as our benchmark solution. We compare the accuracy and computational efficiency of both methods and demonstrate that the trained ANN has a noticeable advantage over the inverse problem in the frequency domain.