102 / 2025-04-17 15:29:22

Effective Fault Diagnosis Method for Induction Motors via Knowledge-Enhanced Graph Convolution Networks

Fault Diagnosis,graph convolution network,prior knowledge,induction motor

Graph Convolution Network (GCN) has been widely applied in industrial fault diagnosis due to their advantages in small-sample learning and interoperability. However, most existing GCN-based methods are trained using measured data, with limited integration of prior knowledge. Moreover, prior knowledge-based methods are often trained under a single operating condition, limiting their applicability in highly dynamic industrial environments such as metallurgical and rolling mills. In such cases, a single prior knowledge approach struggles to cover all operating conditions. These challenges hinder the application of GCN in industrial settings.To address this issue, this paper proposes a novel method for induction motors fault diagnosis based on an improved GCN. Specifically, the method first utilizes prior knowledge to perform a a preliminary fault diagnosis and construct an association graph based on the diagnostic results. Then, two sub-networks are trained using fault data under two representative motor load conditions (high and low). A dynamic observer is designed to estimate the real-time operating load of the motor. Finally, the two sub-networks are integrated to form a complete GCN model that covers the full load range.This diagnostic framework effectively integrates prior knowledge and GCN, achieving superior performance compared to existing fault diagnosis methods in experiments. In particular, under varying operating conditions, the proposed method demonstrates better diagnostic performance than other deep learning approaches and unmodified GCN models.