Misfire Detection Using a Dynamic Neural Network with Output Feedback

This paper presents a crankshaft speed fluctuation model based dynamic neural network misfire detection method to achieve high detection performance and compact network size. In this method, a dynamic neural network with output feedback is utilized to model an inverse system from the engine crankshaft speed signal to the firing event signal. The engine misfire detection is based on the output of the inverse system given the input of engine speed signal. Test results for a 4-cylinder engine show its promising capability of misfire detection even for the low sampling rate data under various engine operating conditions and misfire patterns.