Heatmap-based temporal analysis for valve stiction detection in industrial control loops

Area of Science:

• Process Control
• Industrial Automation
• Fault Detection

Background:

• Control valve stiction is a prevalent nonlinear fault in process industries.
• Stiction causes stick-slip motion, leading to sustained oscillations and reduced plant performance, product quality, and efficiency.
• Existing detection methods often lack robustness across diverse operating conditions, necessitating accurate and computationally efficient solutions.

Purpose of the Study:

• To propose a novel model-free, data-driven method for detecting control valve stiction.
• To analyze the dynamic interaction between controller output (OP) and process variable (PV) for stiction identification.
• To develop a practical and interpretable solution for near real-time stiction detection in industrial control systems.

Main Methods:

• Utilized time-augmented signal representations to create heatmaps highlighting stagnation-prone regions in PV relative to OP.
• Extracted a physically interpretable PV/OP heat ratio as a compact feature for classification.
• Employed threshold-based classification with Receiver Operating Characteristic (ROC) analysis for optimal parameter selection.

Main Results:

• A classifier trained on 58 industrial loops achieved 95% accuracy on a test set of 20 loops.
• Cross-validation yielded consistent performance (78-81%) across different dataset partitions.
• Sensitivity analysis confirmed the method's robustness and generalization capabilities under parameter variations.

Conclusions:

• The proposed data-driven framework offers a practical and interpretable approach for detecting control valve stiction.
• The method demonstrates high accuracy and robustness, addressing limitations of existing techniques.
• This approach facilitates near real-time stiction detection, improving industrial process control and efficiency.