Weighted minimum variance based on-line performance assessment of multivariate processes: A data-driven approach

Area of Science:

• Process Control
• Industrial Automation
• Data-Driven Modeling

Background:

• Traditional minimum variance benchmarking lacks on-line applicability in industrial settings due to insufficient process knowledge from routine data.
• Existing methods often require significant prior process knowledge, hindering real-time industrial implementation.

Purpose of the Study:

• To develop a data-driven approach for on-line performance assessment of multivariate processes without prior knowledge.
• To propose a practical scheme for on-line benchmark estimation when prior process knowledge is available.
• To validate the proposed methods using a case study.

Main Methods:

• Directly estimating the weighted minimum variance benchmark from closed-loop output data under routine operating conditions.
• Utilizing the normalizability of the first non-zero impulse response coefficient for benchmark estimation when process knowledge is available.
• Applying the methods to the "Shell" heavy oil fractionator for verification.

Main Results:

• Successful on-line estimation of benchmarks/indices for multivariate processes.
• Demonstrated effectiveness in reducing estimation errors and time costs for on-line updates.
• Validation of the data-driven approach's applicability in industrial scenarios.

Conclusions:

• The proposed data-driven approach enables effective on-line performance assessment for multivariate processes, even without prior knowledge.
• The method offers a practical solution for industrial implementation, improving efficiency and accuracy.
• The study successfully verified the approach's effectiveness on a real-world industrial case study.