Simultaneous fault type and severity identification using a two-branch domain adaptation network

Abstract

Simultaneous fault type and severity identification is critical for timely maintenance actions to prevent accidents from industrial machinery. The former can indicate occurrences of specific faults, and the latter can track early fault evolutions. Existing methods generally assume training and testing data are drawn from the same data distribution. However, in real industries, due to the change of working conditions, domain shift phenomenon can be triggered. The existing intelligent diagnosis methods are less effective in such scenarios for lack of domain adaptation ability. To address such problems, a novel two-branch domain adaptation network is developed. A deep convolutional neural network with two branches, as the main hierarchical architecture, is designed to handle two recognition tasks. The maximum mean discrepancy based multi-kernel learning is embedded to reduce the distribution discrepancy between the source domain and target domain. As such, the domain-invariant features with a hierarchical structure can be effectively extracted and the fault types and fault severities can be recognized at the same time. Experiments on a bearing dataset, a gear dataset and a motor bearing dataset are carried out to validate the effectiveness of the proposed approach. The results demonstrate that the proposed method can effectively perform fault type and severity identification simultaneously and obviously outperforms other state-of-the-art methods.