Engine misfire fault diagnosis based on sound deep learning
A lightweight convolutional neural network based on deep learning was constructed to realize misfire fault detection of the engine sound signals.The computer's microphone array was used to record engine sound signals in different states including normal,one-cylinder misfire and two-cylinder misfire with different engine speeds.These sound signals were then converted into time-frequency images.These images were applied to the training,verification and testing of a convolutional neural network model.The sound time-frequency image feature extraction network is mainly composed of separable convolution modules.The feature extraction network connects the image classifier.Comparative analysis of training,verification,test experiments are conducted on the network model with different numbers of feature channel grouping convolution modules.The designed convolutional neural network has 99.60%accuracy applying in engine misfire fault detection.The calculation amount of the network is small and the detection time is short.A deep learning neural network based on feature channel packet convolution can quickly complete the detection and diagnosis of the sound signal of engine misfire fault.The method provides intelligent decision support for the online real-time detection of the engine misfire fault.
NETL
NSTL
万方数据
