Chemical processes are usually toxic, corrosive, flammable and explosive. If the process fails, the danger is extremely high. Traditional model-based fault diagnosis methods need to establish an accurate mathematical model of the system, while modern engineering processes are usually large in scale and complex, and it is difficult to establish an accurate mathematical model. Artificial neural network has been widely used in chemical process because of its advantages of parallel processing, self-adaptation, robustness, learnability and fault tolerance. Artificial neural networks based on "deep learning" have been successfully applied to fault diagnosis in various chemical processes. This article summarizes the principle and development process of artificial neural networks, and analyzes the research progress and application status of deep neural networks in chemical process fault diagnosis based on cases. Finally, it is pointed out that deep neural network in the field of chemical process fault diagnosis is of great significance in solving the impact of less fault data and system state changes on the fault detection rate, and promoting the industrial application of fault diagnosis models.
Read full abstract