Abstract
Valve stiction is a common and persistent fault of industrial loops in process control. The loop oscillations caused by valve stiction pose economic and safety risks to the production process. Detecting valve stiction is crucial for control loop performance evaluation, yet current methods suffer from computational complexity, limited generalizability, and low accuracy. Over the past two decades, recurrence analysis based on phase space reconstruction has emerged as a powerful tool to deal with complex nonlinear systems, particularly for detecting subtle changes in signals and systems. Despite this, a highly interpretable and accurate method for extracting features from recurrence plots (RPs) for valve stiction detection has not yet been developed. In this paper, recurrence analysis of loop signals is conducted from the perspective of phase space reconstruction. By combining the stiction formation mechanism, the statistic and distribution feature indexes are defined to characterize the stiction features in RPs. Based on this, a stiction detection method using the feature index (FI) in RPs is proposed. The performance of the RPs-FI detection method is validated in simulation cases and industrial cases. The impact of different recurrence definitions on stiction detection is discussed. The detection results show that the performance of the RPs-FI is superior to most detection methods. Notably, the proposed method achieves comparable generalization and accuracy to the latest machine learning-based detection methods without extensive data training and complex parameter tuning. This detection method demonstrates the great potential of recurrence analysis in valve stiction detection and has the reference value for online monitoring and performance evaluation of industrial loops.
Published Version
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call