Thermography-Based Electrical Machine Fault Diagnosis Using Explainable Compact Vision in Transformer

The major electrical machines suffer from various mechanical and electrical faults and non-invasive methods are preferred for machine health monitoring. In the realm of sensor technology, infrared thermography (IRT) emerges as a game-changer, offering a non-contact solution to measure temperature distribution-a crucial factor for assessing the health of electrical machinery. The integration of Deep Learning (DL) and IRT has grown in recent years. However, noise and a limited pool of infrared thermal images have hindered DL framework performance. To address these challenges, we present an end-to-end real-time framework that includes image augmentation, enhancement, and fault diagnosis using IRT. The proposed CompViT model achieves a validation accuracy of 99.95% and demonstrates strong overall performance, with a macro F1-score of 97.39% (SD 3.66%, 95% CI: 94.6%–100%) for thermal-image-based fault diagnosis of induction motors. Explainable AI also improves the result interpretability, illuminating fault detection procedures and boosting diagnostic trust and transparency. This novel approach connects cutting-edge technology and sensor-driven insights, bringing precision and reliability to machinery diagnostics.