Fault Diagnosis in Chemical Reactors with Data-Driven Methods.

Area of Science:

• Chemical Engineering
• Process Control
• Fault Diagnosis Systems

Background:

• Continuous stirred-tank reactors (CSTRs) are crucial in chemical processes.
• Effective fault diagnosis is essential for CSTR safety and efficiency.
• Traditional methods can be disrupted by changing process parameters.

Purpose of the Study:

• To investigate fault diagnosis (detection, isolation, estimation) in a CSTR.
• To evaluate data-driven methods (RF, KNN) and model-based approaches.
• To develop a robust fault diagnosis framework resilient to parameter shifts.

Main Methods:

• Experimental data from a CSTR for 3-picoline oxidation.
• Application of Random Forest (RF) and k-nearest neighbors (KNN) for fault diagnosis.
• Utilized Isolation Forest (IF) for anomaly detection and model recalibration.

Main Results:

• RF and KNN successfully detected, isolated, and estimated faults under nominal conditions.
• A shift in heat transfer coefficient disrupted both data-driven and model-based methods.
• IF algorithm enabled model recalibration, restoring performance; updated datasets improved RF/KNN adaptability.

Conclusions:

• Data-driven methods (RF, KNN) show scalability and adaptability in CSTR fault diagnosis.
• Model-based methods require recalibration strategies for robustness against parameter variations.
• A combined framework integrating data-driven and model-based approaches ensures robust fault diagnosis.