SIGNAL-PROCESSING PATTERN-CLASSIFICATION TECHNIQUES TO IMPROVE KNOCK DETECTION IN SPARK-IGNITION ENGINES

The aim of this paper is to illustrate the contribution of signal proc essing pattern recognition techniques to the resolution of a physical problem: entire knock detection. Knock is an abnormal combustion of sp ark ignition in automotive engines. Classical detectors use the energy of engine vibration in order to detect it. Unfortunately, detection b ecomes erroneous at high speeds when noise level increases.-Pattern re cognition techniques allow the determination of optimal parameters and methods for knock recognition: cepstral coefficients and amplitude hi stograms improve knock recognition at 5500 rpm and for other speeds. A comparison between five speeds using the conventional detector and th e new detector shows that the latter clearly outperforms the former. T he method used here of knock detection can be used in several other ap plications e.g. gear default detection and axle fissure.