Intelligent fault diagnosis of rolling bearings under varying operating conditions based on domain-adversarial neural network and attention mechanism.

As a domain adaptation method, the domain-adversarial neural network (DANN) can utilize the adversarial learning of the feature extractor and domain discriminator to extract the domain-invariant features, thus realizing fault identification of rolling bearings. In the cross-domain diagnosis of rolling bearing faults, how to obtain fault-related discriminative domain-invariant features from the noisy signals is a key to improving the diagnostic result. In response to this, this paper proposes an intelligent diagnosis model based on the DANN and attention mechanism to identify rolling bearing faults. In order to relieve the influence of noisy data on feature extraction and improve the quality of the learned features, the ensemble empirical mode decomposition (EEMD) is first adopted to denoise the raw sample data to weaken the influence of noise on feature extraction. Secondly, a feature extractor composed of three feature extraction modules in series is designed, and each feature extraction module is composed of a convolution layer, an attention mechanism module and a pooling layer. The feature extractor with attention mechanism enables the model to learn and retain key features related to the faults during training process. Meanwhile, the global average pooling layer is used to replace some fully connected layers in the fault classifier and domain discriminator to reduce model parameters and avoid model overfitting. Finally, the analysis using two sets of rolling bearing experimental about the performance of the presented method show that the proposed method has the potential to become a promising tool for the fault diagnosis of rolling bearings.