Research on Visual Fault Diagnosis Technology for Power Transformers Based on Diffusion Model Sample Augmentation

The issue of insufficient sample size is a key factor restricting the effectiveness of deep learning techniques in the field of visual fault diagnosis for power transformers. To address this, we propose a visual fault diagnosis technology for power transformers based on diffusion model sample augmentation. This method first fine-tunes the diffusion model using LoRA to generate visual fault samples for transformers, thereby expanding the dataset. Subsequently, the diagnostic model is improved to better adapt to the complexity of transformer fault samples, aiming to enhance diagnostic accuracy. The enhanced diagnostic performance of the model is evaluated using an assessment system composed of accuracy, F1 score, and a confusion matrix. The experimental results demonstrate that our proposed method has good diagnostic effectiveness for visible faults in transformers.