Potentials of modern active suspension control strategies – from model predictive control to deep learning approaches

The active suspension system has always been a topic of interest because of its ability to influence the ride quality by exerting independent forces on the suspension by the usage of separate actuators. Various strategies have been proposed over the years in order to estimate the appropriate control action. These strategies are typically feedback oriented and dependent on many factors like the control objective, the frequency of the excitation and system non-linearities that result in the formulation of a complex problem. A simulation-based study using a comprehensive quarter car model with an active suspension is beneficial to summarize the character of each of these approaches by comparing attributes like formulation, performance, robustness, tunability and requirements for implementation on real physical systems.Since active suspensions use an on-board computer and sensor measurements to determine the control action, the enactment of a strategy is limited by the computational requirements and available measurements. This study aims to foresee such challenges when implementing modern control strategies like H∞ control or preview based approaches like Model Predictive Control (MPC) and Deep Learning methods. The latter will focus on both Supervised Learning (SL) and Reinforcement Learning (RL) approaches. These methods are firstly developed in a virtual environment and subsequently implemented on a physical quarter car setup excited by a servo-hydraulic actuator. Finally, a comparison of the performances of the different control approaches is presented.