A deep domain adaption model with multi-task networks for planetary gearbox fault diagnosis

Abstract

The planetary gearbox plays an important role in many advanced electromechanical mechanisms. The mechanical fault is a major factor that threatens the service performance of a planetary gearbox. Deep learning (DL) algorithms have been widely used to identify faults and health status of industrial equipment. However, owing to diversified equipment structures, variable working conditions and disparate data acquisition, the service performance of DL-based methods may degrade significantly when applied to different industrial sites. Domain adaptation emerges as a promising idea that aims to transfer knowledge from a source domain to a different but related target domain. In this paper, we introduce a novel deep domain-adaptive multi-task learning model Y-Net, which is exploited to enable domain-adaptive diagnosis of faults in planetary gearboxes. The SE-Res modules are utilized to reduce the redundancy of the model and improving the separability of deep features. Furthermore, a soft joint maximum mean discrepancy (SJMMD) is introduced to link the two pipeline in order to reduce both the marginal and conditional distribution discrepancy of the learned features, with the enhancement of auxiliary soft labels. The domain adaption between different planetary gearbox under variant operating condition is realized by the Y-Net. Experiments demonstrate the superiority of the proposed SJMMD over conventional maximum mean discrepancy, especially when the datasets of different domains suffer different imbalances.