Diesel engine fault diagnosis method based on WACGAN and IRCNN

In order to solve the problems of over-fitting and low accuracy of the vibration data-driven diesel engine fault diagnosis method caused by the lack of data，a diesel engine fault diagnosis method based on Wasserstein auxiliary classifier generative adversarial network(WACGAN) and Inception residual convolutional neural network(IRCNN) was established from two perspectives of fault data enhancement and diagnosis model optimization. Firstly，Wasserstein distance and gradient penalty were introduced into the auxiliary classifier generative adversarial network，and the WACGAN was established to augment the small-scale training set. Then，the inception structure was introduced into the convolutional neural network(CNN) and continuously differentiable exponential linear units(CELU) were used to improve the feature extraction ability of the model，while residual structure was added to avoid the loss of feature information，and IRCNN model was built. Finally，the IRCNN model was trained with the augmented training set to realize the diesel engine fault diagnosis under small samples. The method was validated by diesel engine fault simulation experiments，which can achieve fault recognition rate of 95% using only 10 operating cycles of vibration data in each fault state. Compared with the traditional oversampling algorithm and the CNN algorithm before and after optimization，the method works best when fault samples are scarce and can achieve fault diagnosis accuracy of 95.47%. © 2025 Chinese Society for Internal Combustion Engines. All rights reserved.