Data-Driven Probabilistic Model of Magneto-Rheological Damper for Intelligent Vehicles using Gaussian Processes

In this paper, we focus on Magneto-rheological damper (MRD) which is essential to intelligent vehicle suspension systems. We attempt to build a data-driven model of MRD with complex dynamics in a probabilistic view in order to address the issues of sample efficiency and the robustness against environmental disturbances in the current non-parametric approaches. Compared with the conventional approaches like neural networks that are sensitive to the noisy data and require a massively training samples, we employ Gaussian processes to model the target system in a full Bayesian way while considering both mean and variance of the prediction. Evaluated by a MRD simulation platform in different working conditions and a noisy external environment, the model learned via GP clearly demonstrated not only good performances in prediction accuracy and uncertainties representation, but also a good balance between the model quality and computational complexity under different sparse scales. These results indicate the great potential of GP as an emerging direction of modeling MRD in intelligent vehicles.