Abstract: Aiming at the problems of insufficient data and unbalanced sample distribution of oil pumping unit failures, a CDCGAN(Conditional Deep Convolutional Generative Adversarial Networks) model based on self-attention mechanism is proposed. The model adds a regular term to the loss function that constrains the distribution of generated images, improves the quality and diversity of generated images and effectively prevents the occurrence of mode collapse. Using Alexnet, VGG16 and other networks to classify and test the generated pumping unit fault samples, the experimental results show that the improved network generates higher quality data, can effectively balance the pumping unit fault data, and further improves the accuracy of the pumping unit fault diagnosis rate.

Key words: fault diagnosis,  , generative adversarial networks,  , self attention mechanism,  , mode collapse