It entails the formation of thermal images from battery cells under different conditions, capturing crucial thermal patterns such as hotspots, insulation degradation, and overheating. For robust model training, data preprocessing and augmentation techniques are applied. The U-Net model, known for its expertise in semantic segmentation tasks, is applied to evaluate thermal images and to detect fault-related features. The results demonstrate the U-Net's unique precision, sensitivity, and specificity in detecting thermal anomalies. This research adds to the improvement of the safety and dependability of EV battery systems, with applications in the electric mobility and automotive industries.