Desenvolvimento de um sensor virtual para a velocidade longitudinal de um automóvel

Abstract

The development of inteligent vehicles has increased over the years, but to have access to the magnitudes of the environment and its dynamics, the vehicle needs reliable sensors to carry out these measurements. However, physical sensors are susceptible to problems, such as measurement errors, availability, and high-cost. An alternative, without high economic costs, to overcome these problems is the use of virtual sensors. This work describes the process of identifying a virtual sensor, capable of estimating the longitudinal speed of an intelligent vehicle by means of low-cost physical sensors, such as the accelerometer of a smartphone. For this, NARX (Non-linear AutoRegressive with eXogenous inputs) polynomial models were indentified, with parameters obtained by the least squares estimator and regressors chosen by the error reduction rate. Finally, robust models to mass variations of the vehicle were implemented and analyzed. The results obtained were satisfactory in terms of robustness to vehicle mass variations. One way to increase the applicability of these virtual sensors would be to obtain robustness in relation to variations in track sloping, which is desired for future work.