Abstract: In order to deeply understand the mechanism of misfire fault of diesel engine and increase the correct rate of misfire fault diagnosis, a misfire fault diagnosis method of diesel engine is proposed based on wavelet and deep belief network (DBN) in this paper. Firstly, equalangle sampling method is used to sample the cylinder head vibration signals of diesel engine, with smooth angulardomain signals obtained and the disturbance of cyclic fluctuation eliminated. Then, by means of continuous wavelet transform, an angularfrequency analysis is conducted on angulardomain signals, which are then reconstructed after the frequency band near ignition frequency is extracted. Next, in accordance with the working cycle of diesel engine, 12 commonlyused feature parameters such as variance, kurtosis and peak value and so on are extracted section by section from reconstructed signals with the diagnostic parameter matrix constructed. Finally, by using DBN, the dimension of the diagnostic parameter matrix is reduced and the features are extracted second time, based on which misfire faults are diagnosed. The method is applied to a diesel engine, with a result showing that the method can accurately extract fire failure information and effectively diagnose misfire faults.

Key words: Diesel engine, Fault diagnosis, Ignition frequency, Continuous wavelet transform, Deep belief network, diesel engine, misfire fault diagnosis, ignition frequency, continuous wavelet transform, deep belief network