Sarcouncil Journal of Engineering and Computer Sciences

Abstract: Hydraulic systems are integral components in many modern industrial applications, and early fault detection is critical to maintain operational efficiency and minimize costly downtimes. This paper develops a robust framework for early fault detection in hydraulic systems using pressure signal analysis. The study employs an axial piston pump as an industrial benchmark, simulating critical faults-internal leakage, partial outlet blockage, and valve failure-via mathematical modeling and FluidSIM simulations. The orifice flow equation, derived from Bernoulli’s principle, dynamically models fault impacts by adjusting parameters (e.g., leakage coefficient (Ct), orifice area (A orifice)). Three hydraulic systems (lift, drilling, transport) are simulated under normal and faulty conditions, generating pressure-time profiles characteristic of each failure mode. Synthetic pressure signals (600 samples, 100 time steps each) are generated to train a 1D Convolutional Neural Network (CNN) for fault classification. The CNN architecture comprises convolutional, pooling, and dense layers, achieving 99.5% accuracy on test data with F1-scores exceeding 0.98. Results demonstrate that pressure signatures, when processed by deep learning, enable precise, non-intrusive fault diagnosis. This approach bridges physics-based simulation with data-driven AI, offering significant potential for industrial predictive maintenance. Future work should validate the framework with real sensor data in complex systems like construction or manufacturing.