Intelligent Modeling for Batch Polymerization Reactors with Unknown Inputs.

Area of Science:

• Chemical Engineering
• Process Control
• System Identification

Background:

• Modeling batch polymerization reactors presents significant challenges for existing system identification techniques.
• Intelligent modeling approaches are crucial for accurate representation and control of these complex processes.
• Unknown inputs (UIs) can arise from modeling inaccuracies or process faults, complicating reactor modeling.

Purpose of the Study:

• To develop a novel recursive approach for identifying accurate models of batch polymerization reactors.
• To specifically address and estimate unknown inputs (UIs) within the reactor model.
• To enhance the precision and reliability of batch polymerization reactor models.

Main Methods:

• A recursive expectation-maximization (EM) algorithm is proposed, featuring an E-step and an M-step.
• The E-step recursively computes a Q-function using maximum likelihood estimation and Kalman filtering for state estimation.
• The M-step finds analytical solutions for UIs via local optimization of the recursive Q-function.

Main Results:

• The proposed recursive EM algorithm was applied to model batch polymerization reactors.
• Performance was evaluated against the augmented state Kalman filter (ASKF) using root mean squared errors (RMSEs).
• The recursive EM method achieved at least 6.52% lower RMSEs compared to the ASKF, demonstrating superior accuracy.

Conclusions:

• The novel recursive EM algorithm effectively models batch polymerization reactors, accurately estimating unknown inputs.
• The proposed method offers improved performance over existing techniques like the ASKF.
• This approach provides a robust tool for enhancing the understanding and control of polymerization processes.