Fault Diagnosis Scheme for the Rotary Machine Group: A Deep Mutual Learning-Based Approach With Cloud-Edge-End Collaboration
Zhichen He, Dingguo Liang, Ying Yang
Abstract
In this brief, a deep mutual learning (DML)-based model construction method with the cloud-edge-side collaboration implemented is proposed to develop the fault diagnosis scheme for the gearbox of the rotary machine group. To be specific, two networks equipped with different layers are trained mutually in the cloud server, where the powerful large network is trained to improve the fault diagnosis accuracy and generalization capability of the small network. Then, the small network is transferred to all edge nodes and retrained by using the local data set, and its robustness performance and accuracy can be increased afterwards. Simulation on the Drivetrain Prognostics Simulator (DPS) platform is conducted to demonstrate the effectiveness of the proposed method.