Bearing cross-domain fault diagnosis based on domain adversarial network

Abstract

A new method has been suggested to tackle the issue of irregularly dispersed bearing vibration data when subjected to varying operational conditions, along with the deficiency of data in the intended domain. This method combines two techniques: the Wasserstein Generative Adversarial Network with Gradient Penalty (WGAN-GP) and Domain Adversarial NeuralNetwork (DANN).To begin with, the limited target domain data is fed into the WGAN-GP. This network consists of a generator and a discriminator engaged in a competitive process. The generator's objective is to create synthetic data that bears a strong resemblance to the data in the target domain. In contrast, the discriminator's goal is to differentiate between authentic data from the target domain and the data generated by the generator. Through this adversarial process, the target domain dataset is augmented. The augmented target domain data and source domain data (derived from distinct yet related domain data) are then fed into the DANN. The DANN comprises two components: a label classifier and a domain classifier. The primary role of the label classifier is to detect faults in bearings, whereas the domain classifier endeavors to ascertain whether the data comes from the source domain or the expanded target domain created through augmentation. Through adversarial training between these two classifiers, the network learns to align the data distributions of the two domains. By following this process, the method effectively bridges the distribution gap between the source and target domains. This approach proves capable of accurately diagnosing bearing faults in scenarios where target domain data is scarce. In summary, the proposed methodology employs a combination of WGAN-GP and DANN to enhance the target domain dataset using synthetic data generation and adversarial domain alignment. This approach effectively addresses data distribution disparities between different domains and facilitates accurate bearing fault diagnosis in situations with limited target domain data.