Data Based Noninstrusive Detection Of Model-plant Mismatch For Closed-loop Industrial Systems

With the rapid development of science and technology,the industrial control system has become more and more complex and intelligent,and the control performance is more stringent.The control system can meet control and production targets of the company at the beginning of the operation.However,with the operation of the system,the uncertainty of the system will degrade the control performance.The main reason for degrading the control performance is model-plant mismatch.In order to achieve the efficient and safe operation of the industrial control system,the detection and diagnosis of the model-plant mismatch is essential.At present,the existing research methods detected the model-plant mismatch detection in an invasive way.They introduced external perturbations on the setpoints or the manipulated variables of the closed-loop control systems,which disturbed the normal operation of industrial processes inevitably and wasted the energy and raw materials.On the other hand,most of the model-plant mismatch diagnosis techniques are still for MPC control systems.However,there is model-plant mismatch in all of the model based controllers.Therefore,this paper proposes a nonintrusive model-plant mismatch detection method based on internal model control(IMC)structure using the routine closed loop industrial data.Compared with traditional detection methods,the proposed method does not introduce any external excitations and it is applicable to all of the model-based controllers.Besides,this method is not sensitive to controller tuning or the changes of disturbance model.The details of the paper are summarized as follows:A nonintrusive model-plant mismatch detection method of Exponentially Weighted Moving Average(EWMA)Run-to-Run(RtR)controller has been proposed in the paper.The orthogonal projection method is adopted to estimate the white noise.The disturbance model is identified by the recursive extended least squares(RELS)method with forgetting factor.A model quality variable is then defined and calculated using the estimated disturbance model,EWMA controller tuning parameter and the output error.The model evaluation index is proposed using the ratio between the variance of the estimated white noise and that of the model quality variable.The model-plant mismatch of the semiconductor manufacturing process is detected based on the model evaluation index.Besides,the influence of model-plant mismatch on the control performance of EWMA RtR control system is analyzed.A general closed loop control system with model based controllers has been considered.A nonintrusive model-plant mismatch detection method based on model evaluation index is proposed.Based on the internal model structure of the closed-loop control system,the disturbance model is identified by the improved adaptive least absolute shrinkage and selection operator(Lasso)method from routine closed-loop data.The model quality variable is obtained using the estimated disturbance model,process model and closed-loop data.The relationship between the model quality variable and the process disturbance innovation is discussed.The model evaluation index is proposed,which is the ratio of the variance of the estimated disturbance innovation and that of the model quality variable.Then,the index is proposed to detect the model-plant mismatch of Single Input Single Output(SISO)closed-loop control system.The nonintrusive detection of model-plant mismatch in Multi-Input Multi-Output(MIMO)closed loop control system has been studied in the paper.The process model residual and the process model residuals of the ouput channels are defined and calculated from the routine closed-loop input and output data.The linear regression model between process model residual of the individual output channel and the white noise is identified by the improved adaptive Lasso method.Then,three model evaluation indices are proposed,which are a global model quality index and two local model quality indices.The global model quality index detects the overall process model mismatch of MIMO closed-loop control system,and two local model quality indices determines the mismatched input-output channels.A method of online monitoring the process model-plant mismatch of SISO closed-loop system has been proposed.The asymptotically statistical characteristics of the model evaluation index are derived.It is demonstrated that the model evaluation index follows the normal distribution asymptotically.Besides,the statistical mean and variance of the model evaluation index are approximated by Taylor expansion.Finally,the Shewhart control chart and moving window method are combined to monitor the model evaluation index of time varying process in real time.And,the process model quality of several control systems is monitored online.The efficacy of the proposed approaches is validated through four processes,i.e.,semiconductor manufacturing process,continuous stirred tank process,Wood-Berry distillation column process and Tennessee Eastman process.In comparsion with other methods,the proposed methods show better performances.On the basis of this,some future research directions are discussed.