A learning automaton methodology for control system design in active vehicle suspensions

Concerns a control system design methodology, applied to the problem of active vehicle suspension system design. Although discussion is limited to a simple chassis system, the methodology is very general, and has the potential to be developed for much more complex industrial systems. The general approach combines concepts from stochastic optimal control with those of learning automata, and extends results obtained previously by the authors (1993). Active suspension system control has been the subject of much research, and many different ideas have been applied, e.g. optimal control, preview control and adaptive control. Two active suspension systems are to be considered. Suspension force actuation is under feedback control; in the first case an ideal full-bandwidth actuator will be assumed, incorporating full-state feedback for both sensor sets. In the second case, a more realistic configuration is considered, with limited bandwidth actuation, and one sensor set consisting of only a single bodymounted accelerometer. The learning automaton selects controller gains, evaluates a performance index, and updates its own internal states, in a way that tends to improve closed-loop system performance. It can be thought somewhat similar to optimization with ´hardware in the loop´, although the automaton is required to work in a stochastic environment. The learning control may also be likened to self-tuning adaptive control; the crucial difference is that for practical application, the automaton does nor require any explicit system model.