Gradient Descent Based Classification of Road Induced Disturbances for Active Suspension Systems

Abstract

Active suspensions are designed to meet conflicting performance requirements such as ride comfort and safety. Achievable ride comfort performance without reaching the limits of road holding and suspension bottoming, is limited by the road disturbance roughness level. In order to obtain best ride comfort performance against different road induced disturbances, it is essential to switch among different controllers according to road roughness level. In this study, a classification algorithm based on logistic regression trained by gradient descent was presented to switch the controller with respect to road disturbance values. The classification algorithm with logistic regression model is trained by the road disturbance data provided by standards. A disturbance observer to estimate the road induced disturbance is designed, then a sigmoid activation function was proposed to change the controller by using only the road disturbance data. The suggested algorithm was tested on the road induced disturbance produced by observer. It was proved that the algorithm without complexity classified the road induced disturbance with the one hyperplane reducing the overfitting condition in training process. As a result, the proposed algorithm can be efficiently used to detect the controller switching instants in real time application.