Data Privacy Protection Diagnostic Algorithm for Industrial Robot Joint Harmonic Reducers Based on Swarm Learning

Abstract

Harmonic reducers play a crucial role in industrial robots. Their high load capacity and low friction performance make them highly favored. However, obtaining a large amount of high-quality data on all factory faults is not easy in actual industrial applications. At the same time, data sharing between factories is limited due to privacy concerns. To address this challenge, this article proposes an innovative solution by integrating convolutional neural networks (CNNs) into a swarm learning (SL) framework. In this framework, multiple factories act as edge computing nodes, sharing data features through the fusion of network parameters without directly sharing the data itself. First, we use CNNs to train each node and select a decision-maker before training to merge the model parameters. Secondly, the decision-maker chosen by SL collects the models from other nodes. Finally, the decision-maker disseminates the integrated model to the other nodes. We validated the proposed method using a harmonic reducer dataset and confirmed its reliability. The experimental results show that the proposed framework can improve computational efficiency without relying on a central server, and the shared model can also improve the fault diagnosis accuracy of each edge node.