A Bayesian Knock Event Controller

A new Bayesian knock event control system is proposed in which the controller maintains the probabilities of different theories regarding the unknown spark angle of BorderLine audible knock, using them to select an optimal control move. Unlike previous controllers that reset every time a control move is made, the new controller incorporates all prior information, thereby giving improved control as more data arrives. Uncertainty can also be injected into the model to maintain transient performance and adaptability to real changes (as opposed to disturbances) that might occur as a result of changes in operating condition. This tradeoff and several variants of the new controller design are investigated using Monte Carlo simulations to obtain rigorous and repeatable statistical assessments of closed-loop behavior. The results show that the controller can deliver fast transient response with reduced cyclic variability and improved steady-state performance with respect to a classical control strategy.