Detection of decoupled input/output pairs in multivariable systems

Area of Science:

• Process Systems Engineering
• Control Theory
• Artificial Intelligence in Chemical Engineering

Background:

• Accurate system identification requires selecting relevant input/output (IO) pairs.
• Traditional methods often identify IO pairs post-model identification, limiting pre-identification analysis.
• Multiple Input Multiple Output (MIMO) systems present challenges in selecting optimal IO pairs.

Purpose of the Study:

• To propose a novel method for detecting uncoupled IO pairs in open-loop systems.
• To enable IO pair selection during the pre-identification stage, optimizing system modeling.
• To enhance the efficiency of system identification and subsequent model-based control.

Main Methods:

• Utilizes IO cross-correlation analysis to assess signal relationships.
• Employs signal filtering techniques to isolate relevant dynamic information.
• Integrates fuzzy logic analysis for robust detection of no-model IO combinations.
• Applies the method in the pre-identification phase of system modeling.

Main Results:

• Successfully detected no-model IO pairs in a simulated 7x6 Fluid Catalytic Cracking (FCC) system.
• Demonstrated the influence of detecting no-model IO pairs on identification accuracy and Model Predictive Controller (MPC) performance for a 2x2 distillation column.
• Validated the method using real-world data from an FCC unit in a petrol refinery.

Conclusions:

• The proposed method effectively identifies uncoupled IO pairs in MIMO systems during the pre-identification stage.
• Early detection of no-model IO pairs improves system identification and control performance.
• This technique offers a valuable tool for optimizing process modeling and control strategies in the chemical industry.