Abstract
Valves are crucial components of a hydraulic system that enable reliable fluid management. Hydraulic valves actuated by a solenoid are prone to degradation in their switching behavior, which may induce undesirable fluctuations in the fluid pressure and flow rate, thereby impairing the system performance and limiting its predictability and reliability. Therefore, it is imperative to monitor the switching behavior of solenoid-actuated hydraulic valves. First, recurrence quantification analysis (RQA) has been applied to the experimental flow signals from a hydraulic circuit to understand the complex switching behavior of the valve. Using RQA, the monotonicity of six recurrence-based parameters has been assessed. In addition, two more nonlinear features, namely, Higuchi and Katz fractal dimensions have been extracted from the flow signals. Based on these eight features (six RQA-derived features and two nonlinear features) a feature matrix is formulated. Second, in a parallel approach, eight different statistical features are extracted from the flow signal to construct another feature matrix. Subsequently, different machine learning methods namely Ensemble learning, K-Nearest Neighbor (KNN), and support vector machine (SVM) have been trained on these two feature sets to predict the valve switching characteristics. A comparison between two feature sets shows that ensemble learning gives better prediction accuracy (99.95% versus 92.2% using statistical features) when fed with RQA features combined with fractal dimensions. Therefore, this study demonstrates that by utilizing the recurrence plots and machine learning techniques on the flow rate signals, the degradation in the switching behavior of hydraulic valves can be monitored effectively, with a high-prediction accuracy.