To address the issues of large model size and long detection time in existing transformer fault diagnosis algorithms, which prevent their deployment on the front end, a transformer winding fault diagnosis algorithm based on Gramian Angular Field (GAF) and Broad Learning System (BLS) is proposed. The GAF is used to convert the original one-dimensional acoustic signal of the transformer into a two-dimensional feature image. Secondly, Convolutional Block Attention Module (CBAM) is added to the Broad Learning System to improve the detection accuracy of the model. The transformed feature image is diagnosed for transformer winding faults through CBAM-BLS. Finally, the proposed algorithm is compared with other existing fault diagnosis algorithms. The experimental results show that the detection accuracy of the proposed algorithm is 96.25%, and the model training time is only 28.56 s.