Motor Bearing Fault Diagnosis Based on Industrial Internet of Things and Transfer Learning

Daiqiao Guo^*

• Hebi Polytechnic, Hebi, China

To address the challenges of traditional motor bearing fault diagnosis methods, such as poor adaptability across operating conditions and a strong reliance on labeled data, this paper proposes an innovative fault diagnosis model that integrates the Industrial Internet of Things with transfer learning. By incorporating a convolutional neural network with a bidirectional gated recurrent unit structure, an adaptive multi-source feature fusion mechanism, and a dual-source domain transfer module based on joint maximum mean difference, this model improves fault identification accuracy, cross-domain generalization, and deployment practicality. Experimental findings demonstrated that the proposed model achieved a fault identification accuracy of 94.7% and maintained a false alarm rate of just 6.8% following model convergence. In the state matching test, the highest state recognition match reached 97.3%. Meanwhile, in practical application tests, the model delivered fast diagnosis with a response time of 19.6 seconds and a GPU usage rate of only 8.9%. It consistently maintained a diagnosis accuracy above 93% across various operating conditions. The proposed fault diagnosis model for motor bearings effectively achieves accurate fault detection and robust adaptation across varying operating conditions, thereby offering reliable technical support for the predictive maintenance of industrial systems. This helps enterprises reduce maintenance costs while improving efficiency and equipment stability.