Successive ridge detection framework: A novel method for valve stiction detection

Oscillations are common in industrial control loops, and valve stiction is one of their primary causes. Existing nonlinear-based stiction detection methods usually rely on signal decomposition to identify fundamental and harmonic components from process variables (PVs). However, their performance is often degraded by mode mixing and weak harmonics being obscured by noise disturbances. To address these limitations, this work introduces ridge detection (RD) into valve stiction detection. RD directly tracks ridges, i.e., mode instantaneous frequencies, on the short-time Fourier transform (STFT) spectrogram, thereby avoiding signal decomposition and mitigating mode mixing. Building on this idea, we propose a successive RD framework (SRDF) for automatic and robust stiction detection. SRDF first removes trend components from PV signals using the Hodrick-Prescott filter to alleviate spectral leakage. It then performs successive ridge extraction with a physically guided masking radius derived from the spectral standard deviation, thereby suppressing pseudo-ridges and missed detections. Stiction is finally confirmed by verifying the fundamental-harmonic relationships between the extracted ridges. In addition, task-oriented STFT parameter selection rules are established to ensure reliable ridge extraction. Experiments on the International Stiction Database benchmark and a self-constructed industrial dataset from a phosphoric acid plant show that SRDF outperforms advanced decomposition-based methods and baseline RD, achieving detection accuracies of 88.46% and 85.00%, respectively. These results demonstrate the effectiveness and practical applicability of the proposed method.