Integrated Data-Driven Process Monitoring and Explicit Fault-Tolerant Multiparametric Control

Area of Science:

• Chemical Engineering
• Control Systems Engineering
• Machine Learning Applications

Background:

• Active fault-tolerant control is crucial for maintaining process stability and performance.
• Model predictive control (MPC) offers advanced control capabilities but can be computationally intensive.
• Integrating machine learning with MPC can enhance robustness and adaptability.

Purpose of the Study:

• To develop a novel active fault-tolerant control strategy.
• To combine machine learning-based process monitoring with explicit/multiparametric model predictive control (mp-MPC).
• To ensure smart operation through rapid adaptation to process faults.

Main Methods:

• Utilized Support Vector Machine (SVM) for data-driven fault detection and diagnosis.
• Employed a nonlinear, kernel-dependent SVM for feature selection and ranking.
• Applied Random Forest algorithm for data-driven fault magnitude estimation.
• Developed an explicit/multiparametric MPC using the PAROC framework.

Main Results:

• Generated explicit control strategies as affine functions of system states and fault magnitudes.
• Demonstrated the framework's effectiveness using a penicillin production semibatch process.
• Enabled rapid, intelligent switching between pre-computed control actions based on continuous monitoring.

Conclusions:

• The proposed framework effectively integrates machine learning and mp-MPC for active fault tolerance.
• Continuous process monitoring facilitates smart operation and rapid control strategy adaptation.
• The strategy enhances operational efficiency and safety in dynamic processes.