Identification of time-varying non-linear systems with application to knock detection in combustion engines

Knock is a serious problem affecting internal combustion engines, and not only causes unwanted noise from the motor vehicle, but may also cause engine damage. In this paper, the reconstruction of the pressure signal in the combustion chamber from vibration sensors on the engine block is examined. A non-linear model, the Volterra series, is used as the basis for this purpose. A linear model is also utilized and a comparison made. A method of Volterra series kernel estimation based upon the singular value decomposition of the observation matrix is discussed. This method is validated using simulated data