Control System Performance Assessment And Fault Diagnosis Based On Closed-loop Data

There are many control loops in process industry system. How to monitor these loops in a simple and effective way? Whether the measurement index of control loops can be obtained through performance assessment or not? How to keep control loops in good condition through fault diagnosis? All these problems have attracted many researchers. While control system performance assessment and fault diagnosis are very important in process industry, they require a lot of prior knowledge, such as the time delay of different channels and the interactor for MIMO systems, and so on. It would be meaningful if performance assessment and fault diagnosis can be applied based on closed-loop operating data.This thesis focuses on closed loop data based control system performance assessment and fault diagnosis. The main work and the research results are as follows:(1) By using fast sampling based closed-loop PEM identification, a stable model can be obtained for system without external excitation. It can be seen through closed-loop identifiability analysis that an appropriate fast sampling rate can reduce the identification error and thus obtain better identification results.(2) A new solution strategy of interactor matrix is proposed: it is independent on prior knowledge and can provide the calculation result for checking and amending of the unitary interactor matrix. A modular implementation structure is given, which is beneficial to programming application.(3) A new polynomial division method: Ratio and Displacement algorithm, is proposed ,which can be used to calculate the time delay coefficient matrix. A process-oriented description of the algorithm principle is given, and the advantages of this algorithm are obviously seen from the comparison with other two methods: the calculation time and the consumption of the computer's memory are both small. (4) A control performance assessment and fault diagnosis method based on closed-loop data is presented: firstly obtain a model by fast sampling PEM identification with closed-loop data, then calculate interactor matrix, following by the control performance index. When the performance is deteriorated, implement fast sampling PEM identification again in order to find the root reason: the controlled process, the controller parameters or the system set points has changed, which is beneficial to confirming possible and excluding impossible fault.As aforementioned, the control performance assessment and fault diagnosis approach presented in this paper is a practical way to comply with process industry. In order to take this method into action, there are many detailed problems worthy of further research, such as, whether there is a simple and effective method to determine the various identification parameters in order to identify well with closed-loop data.