Data-Driven Diagnosis of Model-Plant Mismatch in MIMO Closed-Loop Control System.

Area of Science:

• Process Control
• Systems Engineering
• Model Validation

Background:

• Effective control system performance relies heavily on accurate process models.
• Quantifying the discrepancy between a process model and the actual plant is crucial for robust control design.
• Existing methods may not adequately isolate model-plant mismatch from other system variations.

Purpose of the Study:

• To propose a novel methodology for diagnosing model-plant mismatch within an internal model control (IMC) framework.
• To develop metrics for assessing overall process model quality, individual output accuracy, and specific input-output channel performance.
• To enable the isolation of model-plant mismatch from controller parameter variations.

Main Methods:

• Utilizing an internal model control (IMC) framework for diagnosis.
• Whitening system outputs and identifying an external disturbance model.
• Developing a process model residual based on identified disturbance and control models.
• Defining three distinct model quality indices.

Main Results:

• A novel methodology for model-plant mismatch diagnosis is presented.
• The method successfully identifies mismatched subchannels in control systems.
• Model-plant mismatch can be effectively isolated from changes in controller parameters.
• Demonstrated capability on a simulated column process and the Tennessee Eastman benchmark.

Conclusions:

• The proposed methodology provides a robust approach to quantifying and diagnosing model-plant mismatch.
• The defined model quality indices offer valuable insights into process model accuracy.
• This technique enhances the reliability of control systems by ensuring model fidelity.