A data-driven multiplicative fault diagnosis approach for automation processes.

Area of Science:

• Automation Engineering
• Process Control
• Data Analytics

Background:

• Traditional fault diagnosis often focuses on additive faults.
• Multiplicative faults, which increase process variability, are critical for performance degradation in automation.
• Existing methods may not adequately address multiplicative fault diagnosis.

Purpose of the Study:

• To develop a novel data-driven method for diagnosing multiplicative key performance degradation in automation processes.
• To differentiate from and improve upon existing additive fault diagnosis approaches.
• To identify low-level components responsible for increased process variability and performance decline.

Main Methods:

• Feature extraction for multiplicative faults directly from process data.
• Evaluation of each process variable's contribution to performance degradation to pinpoint root causes.
• Validation through a numerical example and Monte Carlo simulations for statistical robustness.
• Case study application on the Tennessee Eastman process for practical relevance.

Main Results:

• Successfully extracted features indicative of multiplicative faults.
• Quantified the impact of faults on individual process variables.
• Demonstrated statistical effectiveness and practical applicability of the proposed method.
• Identified specific components causing performance degradation in the case study.

Conclusions:

• The proposed data-driven method effectively diagnoses multiplicative faults in automation processes.
• This approach offers a valuable alternative to traditional additive fault diagnosis.
• The method provides a robust framework for identifying root causes of performance degradation, enhancing process reliability.