Evaluating vehicle dynamic behaviors through an integrated framework of a driver-in-the-loop simulator and artificial neural networks

Given the increasing demand for analyzing large datasets acquired through vehicle instrumentation and simulation, this study aims to create an integrated framework that combines a dynamic vehicle simulator with data science and artificial intelligence techniques. The main goal is to reduce the subjectivity in vehicle behavior analysis and lower simulation time, subsequently reducing vehicle project development costs too. To achieve this aim, is designed a tool capable of assessing the dynamic behavior of virtual models employing parameters that were previously reliant on the subjective assessment of the simulator driver. Making use of software such as MATLAB and VI-CarRealTime, simulations with virtual vehicle models correlated in kinematics and compliance and handling parameters with physical vehicles are conducted in real-time tests involving a human driver, utilizing a dynamic simulator with nine degrees of freedom, model VI-Grade DiM 150. For data preprocessing and artificial neural network development, the Python programming language is applied. Through a multilayer perceptron network, results with satisfactory accuracy are achieved, thereby validating the proposed methodology.