سينا منشي زاده

This study investigates the performance of a Model Predictive Controller (MPC) based on the CVXPY library for vibration control in a half-car suspension system equipped with a Magnetorheological (MR) damper. The half-car model, designed with four degrees of freedom, was analyzed using standard road profiles from ISO 8608 and bumpy road conditions. The primary goal was to eva‎luate the impact of the MPC controller on improving the system’s dynamic responses, such as the sprung mass position, vertical acceleration, and suspension movement. The dynamic model of the system, including the MR damper force equations, was initially formulated, and the state variables were defined. Various road profiles were simulated based on ISO 8608 standards. The MPC controller was designed to optimize system responses, and its parameters were tuned accordingly. Simulations showed that, on smoother roads, vertical acceleration and sprung mass displacement decreased by 13.16% and 12.12%, respectively. On rougher roads, these reductions increased to 26.93% and 30.22%, respectively. Under bumpy road excitations, significant reductions were observed in RMS metrics, including 69.36% in sprung mass displacement, 46.72% in damper force, and 40.23% in vertical acceleration. It was also revealed that extending the prediction horizon improved performance but reached saturation beyond a certain limit. Machine learning models were subsequently explored to predict the control signal, with Random Forest identified as the most effective model. This model accurately predicted the control signal and enhanced the controller’s performance. The findings demonstrate that combining MPC with MR dampers and leveraging machine learning significantly improves vehicle comfort and safety under various road conditions. Future recommendations include exploring more complex models and conducting experimental tests.

ميراگر مگنتورئولوژيكال , درخت‌هاي تصادفي , يادگيري ماشين , كنترل‌كننده ‌پيش‌بين مدل

Magnetorheological Damper , Random Forest , Machine Learning , Model Predictive Controller

Sina Monshizade

Javad Marzban Rad