The Model-plant Mismatch Detection For Industrial Closed-loop Control Systems

With the increasing demands for product quality and operation safety,control performance monitoring(CPM)has played a more and more important role for the closedloop industrial systems.In the operation of industrial systems,performance degradation depends on such factors as controller tuning,equipment degradation,unmeasured disturbance,model-plant mismatch(MPM),etc.The model-based controller usually runs normally during the early design stage.However,as time goes on and the process dynamic changes,the model mismatch may appear and become more and more serious,which brings safety problems.Hence,model mismatch detection is very necessary,and the research of an effective MPM detection method has become an important subject.For the model-based closed-loop control systems,most of the existing research results focus on a specific control system,which is not universal,and there are other problems such as the need for additional excitation signals,the inability to distinguish mismatches between disturbance model and process model,etc.In order to solve the mentioned problems,three MPM detection methods aiming at different model mismatch problems are studied in this thesis,and their dominance and universality are also gradually improved.The details of this thesis are as follows:A new MPM detection method based on the orthogonal projection for model predictive control(MPC)system is proposed.Firstly,based on the input/output data of the MPC system,the system residual is directly calculated,and the disturbance innovation is estimated by the orthogonal projection method.Then,a model quality index(MQI)based on the system residual is designed.The index just needs the closed-loop data,and can be related to the model parameters that have physical significance.Finally,the practicability of the proposed method is verified by the Wood-Berry distillation column.An MPM detection method based on the parity space is proposed.Firstly,based on the state space model,the parity space-based system residual and performance criteria are established.Then,the residual evaluation function and its corresponding threshold,which form the detection index,are derived by optimizing the performance criteria.The updtate value of the index can be recursively obtained by finding the minimum quadratic form of the unknown input.Finally,the proposed method is demonstrated bythe paper machine process with head box controller.An MPM detection method based on the subspace identification is proposed.Firstly,the extended process observability matrix,which is independent of the disturbance model,is estimated under the closed-loop condition by the subspace identification method.Then,its nuclear norm is selected as the detection index,and the control limit corresponding to the index is calculated by the kernel density estimation(KDE)method.Finally,the effectiveness and practicability of this method are verified by three different simulation examples.The proposed method is mainly based on the closed-loop identification theory,which can avoid the influence of external disturbance changes from the MPM.Besides,the proposed method is suitable for any closed-loop control systems,which is more generalizable to complex industrial systems.