The existing fault diagnosis methods can achieve good results when various status fault data are available. However, the construction of the diagnosis model is often unachievable in the actual application because only normal data are available, which is actually a few-shot fault diagnosis problem. Therefore, a novel intelligent few-shot fault diagnosis method of rotating machinery based on the convolutional neural network (CNN) using virtual samples generated by the mechanism character generative model (MCGM) integrating the generative adversarial network (GAN) is proposed. The distribution pattern of common parameters that reflect the fault category is learned using the GAN and source domain fault data. Then, the normal state data of the target domain is combined with the distribution common parameters to generate virtual samples in target domain based on the MCGM. Moreover, the fault diagnosis model is trained by virtual samples based on the CNN. Finally, the proposed fault diagnosis method is validated using the laboratory bearing data, the industrial data and the public data of the rotating machinery, respectively. The results show that the proposed method achieves an average accuracy of 93.38% in the diagnostic task, exhibiting at least 4.56% better performance than other comparison methods.

The development of artificial intelligence has brought new opportunities and challenges in the field of mechanical fault diagnosis. Especially, data-driven intelligent fault diagnosis (DIFD) methods are favored by researchers. However, commonly used DIFD methods can only recognize the fault of a known class instead of an unknown class, which results in the DIFD's inability to self-learning. To handle this problem, novel probability confidence convolutional neural network (PCCNN) is proposed in this article. The probability of belonging to each known class calculated by the PCCNN and the confidence of each known class is used to recognize the known and unknown classes. At the same time, the ability to recognize a new class is achieved by updating the architecture and parameters of the PCCNN model with a self-learning method. The bearing experiment dataset, the gearbox experiment dataset, and the rotating machinery failure data are used in the case studies. The study results show that the proposed method achieves an average accuracy of 97.42% and 96.87% for recognizing unknown and known classes, respectively, exhibiting at least 5.18% better performance than the 12 comparison methods. All the code and experimental data are available on the website.

Any failure of the turbofan engine, as one of the key components of space shuttle, can lead to serious accidents. Therefore, it is necessary to predict the remaining useful life (RUL) to guarantee its reliability and safety. This article proposes a multitask learning (MTL)-based self-attention encoding atrous convolutional neural network called MSA-CNN to effectively realize RUL prediction. Specifically, in order to extract fault feature information, an atrous convolutional neural network (ACNN) is used as the auxiliary task network, which is more efficient than the traditional convolutional neural network (CNN) in the process of down sampling. Moreover, a model with ACNN and self-attention encoder (SAE) is used as the main task network to capture short-long term dependencies in a time sequence and thus realize RUL prediction. Compared with other recurrent neural networks (RNNs), SAE proposed in this article has the advantage of parallel computation. Besides, a novel multitasking loss function is also proposed to realize the interaction among multiple tasks. After MSA-CNN experiments on four subsets of commercial modular aero propulsion system simulation (C-MAPSS) dataset, the root mean square error (RMSE) average between the predicted RUL and the real value is about 14.66, which is better than the existing methods. Several other comparative experiments were conducted to verify the benefits of each submodule.