S. Joe Qin
Model Predictive Control and Performance Assessment
Process Monitoring and Fault Diagnosis
Modeling and System Identification
Semiconductor, Chemical and Pulp and Paper Processes
Due to increased complexity, quality, and environmental requirements in process operations, advanced process control strategies often have to be integrated to fulfill the control needs. The emergence of model predictive control, optimization, and process monitoring provides a tremendous opportunity to improve process efficiency and optimality. My research interest is to develop theory and new methods in process monitoring, modeling, and control.
Model predictive control (MPC) is by far the most effective advanced control strategy in process industries. An overview of model predictive control technology indicates that there are more than 2200 industrial applications to date. Recent academic research has built the connection between MPC and the linear quadratic regulator problem in the 1960's. Remaining challenges lie in the area of nonlinear MPC and fault tolerant MPC. When the process to be controlled is nonlinear that requires a nonlinear model in the control loop, nonlinear programing has to be used in solving the control problem on-line, which is extremely demanding computationally. Our objective is to develop simplified nonlinear MPC strategies so that the computational cost can be greatly reduced. Another objective is to maintain the control system integrity even if some of the sensors and/or actuators fail to function normally. This task involves the integration of sensor monitoring and model predictive control.
System identification for multivariable processes is very important for applying model predictive control and process monitoring. Current industrial practice almost always relies on system identification to derive the process model. Existing system identification problems include process input collinearity, parallel process outputs, and corruption of unmeasured disturbances. Our objective is to develop new methods based on subspace identification methods, partial least squares and advanced statistical methods to overcome these problems.
Process monitoring and fault diagnosis play a significant role in safe process operation and control. We make use of principal component analysis, partial least squares, dynamic models, and wavelet theory to monitor abnormal process conditions and sensor faults. These approaches are also applicable for data reconciliation and missing value replacement. The objective and challenge are to identify the root cause of any abnormal situations that could happen in a process. Current work in this area involves the diagnosis of polyester film processes and semiconductor processes.
Semiconductor processes monitoring and control represent an emerging area of process control where tremendous needs are identified for automatic process control and monitoring. Our objective is to develop generic run-to-run control and monitoring methods for the semiconductor industry. Current work in this area involves an on-going collaboration with manufacturing companies to develop adaptive monitoring and run-to-run control methods for RTP/RPA tools and plasma etching reactors.