A lightweight three-phase induction motor fault detection method based on thermal imaging

With the wide application of power electronic drive technology in industrial automation and smart grid, the fault monitoring of inverter-driven three-phase induction motors has become a key link to ensure the stable operation of the system and power quality. To this end, a lightweight three-phase induction motor fault detection method based on thermal imaging is proposed in this paper. For the model's ability to adapt to irregular hot spots and complex edges, we optimize and introduce a dynamic adaptive group convolution (AKGConv) in the P3 layer of YOLOv11 to maintain the stability of the convolution structure. Meanwhile, to solve the problem of redundant parameters of multi-scale detection head, we design a lightweight shared convolutional detection head (LSCD-Detect) to achieve feature sharing and efficient prediction, which effectively improves the discrimination ability of the model. In addition, in order to improve the convergence speed and robustness of model training, generalized IoU (GIoU) is introduced to optimize the loss. The experimental results show that, compared with the baseline model YOLOv11 on our constructed induction motor fault thermography dataset, the method in this paper reduces the number of parameters and the computational volume by 11.15% and 13.6%, respectively, while maintaining the high accuracy, which fully verifies the superiority of the method in the field of industrial automation and intelligent manufacturing.