A non-stationary system diagnosis using fuzzy pattern recognition. Application to engine knock diagnosis

摘要

Knock is a well-known problem for spark-ignition engine manufacturers . Knock detection helps achieve the best compromiseudbetween increasing engine efficiency, fuel consumption and present requirements with regard to exhaust emission legislation .udThe ignition timing is usually controlled so that knock never occurs even with fuel quality changes . The advantage of a knockuddetection method is to work within close range of knocking conditions but avoid its occurrence . The purpose of our study consists inudhighlighting several knock intensities from block vibration signals provided by an accelerometer. Our aim is to differentiate threeudkinds of engine cycles : absence of knock, increasing knock and heavy knock . The developped diagnostic approach deals withudfuzzy pattern recognition . After describing the diagnostic architecture line modules and their functions, experimental results fromuddata acquired from an engine test bench are reported . The method, experimented on a learning set, leads to several diagnosesudthat cooperate .