MVHFC-Net: AI-driven multi variate hydraulic fault detection and classification network in hydraulic systems

Hydraulic systems find extensive applications in various industrial domains, including aerospace, automotive, and manufacturing. These systems rely on precisely controlling and managing fluid pressure, flow rates, temperatures, and other physical parameters to ensure optimal performance and reliability. However, the complexity and dynamic nature of hydraulic systems poses significant challenges in detecting and diagnosing faults accurately, leading to potential operational inefficiencies and safety concerns. This research addresses the problem of fault detection and classification in hydraulic systems using a novel methodology. This work proposes a Multi-Variate Hydraulic Fault Classification Network (MVHFC-Net) approach comprising a multi-label multi-label dataset. Initially, the dataset contains inputs such as motor power, pressure, temperature, volume flow, vibration, and virtual parameters like cooling efficiency and power, where faults are identified based on these input parameters. Subsequently, data preprocessing ensures data quality and prepares it for further analysis. In addition, an Extra Trees Classifier (ETC) is also employed to capture relevant information from the raw data effectively. Further, the Feature Importance Ranking (FIR) procedure is then utilized for optimal feature selection, enhancing the discriminative power of the classification model. Finally, the Hierarchical Extreme Learning Machine (HELM) classification model is employed to classify outcomes such as the condition of the cooler and valve, severity of internal pump leakage, pressure level of the hydraulic accumulator, and stability indicators. The proposed methodology offers a comprehensive framework for fault detection and classification in hydraulic systems, improving operational efficiency and reducing downtime. The proposed MVHFC-Net achieved an accuracy of 100%, precision of 100%, recall of 100%, and F1-score of 100% on all targets such as cooler condition, value condition, internal pump leakage, hydraulic accumulator, and stable flag. Experimental results demonstrate the approach’s effectiveness in accurately diagnosing system faults and providing actionable insights for maintenance and optimization.