Predição da força lateral em pneus inteligentes a partir de técnicas de aprendizagem de máquina

Abstract

Intelligent tires are tires that feature sensors within their structure capable of communicating with advanced driver assistance systems for the purpose of estimating parameters and states of vehicles. This project aims to develop a model to estimate the lateral force in a intelligent tire using Machine Learning techniques, based on measured acceleration signals, which exhibit high levels of noise that impact the estimation process. To achieve this, the reconstruction of the signal is proposed through the use of the adaptive heuristic EMD (Empirical Mode Decomposition) for each tire rotation, with distinct configurations that optionally consider the application of a low-pass Butterworth filter at 400Hz. The reconstruction of the signals was applied to the acceleration signals along the x, y, and z axes corresponding to approximately one complete tire rotation (0° to 359.5°), as well as to a specific rotation interval of the tire (145.5° to 215.0°) that aligns more closely with the contact region of the tire with the ground. To evaluate each proposed configuration, Machine Learning models such as Random Forest (RF) and Gaussian Process Regression (GPR) were used to estimate the lateral force, with NRMSE (Normalized Root Mean Square Error) and the coefficient of determination r² metrics to assess the performance of each proposed learning model. The two best models obtained using RF and GPR, particularly those incorporating the signal reconstruction step and information from outside the tire-ground contact region, exhibited NRMSEs of 16.33% and 16.57%, respectively, with r² values of 0.91 and 0.90, respectively. These results highlight the models’ ability to explain the lateral force based on acceleration signals, emphasizing the significance of data collected outside the tire-ground contact region. Compared to the results of Barbosa et al. (2022), the models demonstrated robust performance, especially those integrating signal reconstruction via EMD, underscoring its potential in improving the prediction of lateral force.