Fault Diagnosis For Closed-loop Control Systems Based On Parity Space Transformation And Deep Neural Network

Closed-loop control is one of basic concepts of cybernetics,referring to a control relationship that the controlled output is returned to the input as a control in a certain way and exerts a control influence on the input.Therefore,the closed-loop control system is also called the feedback control system.Because it can improve the robustness of the system and reduce the interference of external factors on the system effectively,the closed-loop control system has been widely used in the field of chemical control and aircraft control and so on.One of the difficult problems in the field of closed-loop control system is fault detection and identification.Due to the inherent feedback adjustment mechanism,a closed-loop control system is generally very robust to external disturbances and internal noises,which makes it more difficult for fault detection in the closed-loop control system.Closed-loop control systems often encourage faults to propagate inside the systems,which may lead to the consequence that faults amplitude becomes smaller and fault characteristics difference becomes more inapparent,which makes it more difficult for fault identification in the closed-loop control system.Aiming at the above two problems and staring from the fault and propagation mechanism of closed-loop control system,this paper studies the fault detection and identification method applicable to closed-loop control system based on data driving and knowledgedriven theory.Moreover,the typical closed-loop control systems,such as Tennessee Eastman Process(TEP)system and the Satellite Attitude Control System(SACS),are used to verify the methods proposed in this paper.And the main researches are as follows:(1)According to the characteristics of the closed-loop control system,the concept and system characteristics of the closed-loop control system are given.Then,the characteristics difference between the open-loop control system and the closed-loop control system is compared.On this basis,the mathematical models of the two control systems are given.Finally,the fault propagation mechanism of the closed-loop control system is analyzed theoretically based on this mathematical model.(2)Aiming at the difficulty in fault detection of closed-loop control system,a fault detection method based on parity space is proposed.The principle of dual spatial transformation is analyzed.The method of obtaining stable kernel matrix is given and the existence of stable kernel matrix is proved.On the basis of this,combined with the characteristics of closed-loop control system,a stable kernel matrix suitable for closed-loop control system is given.Finally,the fault-to-noise ratio is increased by the multi-time residual sequence,thereby improving the fault detection rate of the closed-loop control system.(3)Aiming at the difficulty in fault identification of closed-loop control system,a fault identification method based on deep neural network is proposed.The principle of deep neural network is analyzed.Then,deep neural network is applied to find fault characteristics difference between different data modes,and a sliding window is used to amplify the fault-to-noise ratio and characteristics difference,with an aim to increase the identification performance,which makes up for the deficiency of traditional neural networks not suitable for dynamic systems.(4)The fault detection and identification methods of the closed-loop control systems proposed in this paper are verified through the typical closed-loop control systems such as Tennessee Eastman Process system(Stationary System)and Satellite Attitude Control System(Non-stationary System).Compared with the traditional fault detection and identification methods,the experimental results show that the proposed methods improve the fault detection rate and identification rate of the closedloop control system.