DCGAN Based Data Generation for Process Monitoring

The number of samples under fault conditions is usually much smaller than that of samples under normal conditions in chemical industry fault diagnosis field, which is called unbalanced dataset. Due to the existence of such unbalanced datasets, traditional methods are not easy to detect faults. Deep convolutional generative adversarial networks(DCGAN) is used to solve the fault data generation problem in this paper. The deep learning generation model can generate a large amount of fault data, which can be provided for subsequent fault diagnosis research and analysis. Firstly, DCGAN is proposed to generate more fault data based on the existing data. The statistical characteristics of the original data and the generated data are similar, which implies the generation ability of DCGAN performance well. The generated fault data are added to the original dataset to form a new balanced dataset. Then the convolutional neural network(CNN) is used for fault classification, and the fault classification effect is verified to apply in an actual gas-solid fluidized bed equipment.