Travail de fin d'études et stage[BR]- Travail de fin d'études : Ride and Comfort Optimization of Autonomous Vehicles[BR]- Stage d'insertion professionnelle

Abstract

Autonomous vehicles require improved comfort without loss in handling. To accomplish this, a controller for active suspension is necessary. This thesis focuses on controller design and answers the following question: which controllers perform best for comfort and handling in the context of road preview and active suspension? Active suspension requires three components: force elements, sensors, and controllers. Controllers exploit road preview data from sensors then command force elements like actuators to counter disturbances. Four controller groups are defined: feedforward, feedback, robust and objective function control. A quarter-car model and Gaussian-curve road obstacle are established, along with a single, hard constraint on tire load to ensure road contact. Controller performance is measured in terms of comfort, handling, suspension deflection, and actuation. A global index considers all in one term. Pertinent controllers for the formulated problem are feedforward, skyhook, PID, fuzzy logic, H2/H∞, LQR and MPC. A pre-selection process yields feedforward, MPC and skyhook for implementation. Certain and uncertain simulations are carried out on the selected controllers, plus a combined feedforward-skyhook. MPC performs the best globally but has robustness issues. Despite worse comfort and handling than feedforward, it wins by greatly conserving on actuation. Feedforward performs globally better than skyhook but adding skyhook to feedforward does not improve performance. Indeed, MPC performs best, but this conclusion should not be overvalued, as it is subjective. The proposed framework on comparing and testing controllers is more important since it can be used by anyone with their own performance priorities.