Identification of air compressor faults based on ViT and Mel spectrogram

Air compressors are used as a power source in many industries, but they are likely to malfunction during use. However, traditional methods such as fault analysis and judgment on the working sound of air compressors are inefficient and have not fully applied artificial intelligence related technologies in this field. Therefore, we propose a machine working sound recognition method based on Vision Transformer(ViT), which extracts the features of the working sound of the air compressor and converts it into the Mel spectrogram, and use the method of image classification in deep learning to determine whether the air compressor is in a faulty state, thus avoiding more losses and improving work efficiency. The experimental results show that the deep learning method can determine the working status of the air compressor very well, and the ViT shows better classification performance than ResNet, VGG and other models.