Classificação de descontinuidades em vias asfaltadas usando redes neurais treinadas a partir de modelos físicos da dinâmica vertical de veículos

Abstract

With the growth of cities, there has been a noticeable increase in the amount of time people spend within transportation means on paved roads. Consequently, various ergonomic-related issues arise. In this context, this undergraduate thesis proposes the development of a methodology for training artificial neural networks to classify discontinuities in asphalt roads, using physical mod- els of vehicle vertical dynamics and acceleration signals measured by mobile devices. Initially, a discrete full-car physical model with 7 degrees of freedom was developed in a computational environment, aiming to represent the vertical vehicle dynamics. From this model, estimates of the vehicle’s vertical acceleration were obtained, simulating a smartphone placed on the car’s dashboard. Subsequently, a classification algorithm based on artificial neural networks was imple- mented, aiming to analyze the vehicle’s vibration signals and identify different discontinuities in roads, such as speed bumps, potholes, and normal roads. The classification algorithm was tested with simulated data, considering various discontinuities and vehicle speed conditions, and it was parameterized to allow for changes in the model’s speed. Finally, the neural network was trained and validated, demonstrating the effectiveness of the proposed methodology in creating systems capable of identifying discontinuities in asphalt roads. This contribution aids in improving road quality and, consequently, the safety and comfort of users.