Leakage Monitoring and Feature Importance Analysis of Hydraulic Systems Based on a Hybrid SVM-RF Algorithm

Hydraulic systems serve as a pivotal component in modern industrial settings, where leakage faults significantly impact the system's safety and operational efficiency. This paper introduces a leakage detection method for hydraulic systems based on a hybrid SVM-RF algorithm. By integrating the strengths of Support Vector Machines (SVM) and Random Forest (RF) algorithms, the accuracy and stability of leakage prediction are enhanced. Leveraging a hybrid SVM-RF algorithm further improves the prediction stability of the system across diverse environmental conditions. Experimental results demonstrate that both the hybrid SVM-RF algorithm and the individual SVM and RF models exhibit high prediction accuracy. However, the hybrid SVM-RF algorithm significantly outperforms in terms of stability, making it more suitable for long-term monitoring in practical production environments. Finally, the Random Forest algorithm is utilized to conduct an importance analysis of the key features in the hydraulic system, providing a solid basis for fault diagnosis. The study reveals that characteristics such as flow rate, pressure, and efficiency factors play crucial roles in leakage prediction.