AN ARTIFICIAL NEURAL-NETWORK-BASED APPROACH TO FAULT-DIAGNOSIS AND CLASSIFICATION OF FLUID-POWER SYSTEMS

In this paper, multilayer perceptron (MLP) type neural networks are us ed to detect leakages in an electrohydraulic cylinder drive. Both sing le-leakage and multiple-leakage type faults are investigated. The perf ormance of MLPs is examined relating to the level of leakage flowrate and it was found that MLPs perform well for line leakages but for acro ss-cylinder seal leakages they could only detect leakage over 1.01/min . The generalization tests on non-training leakage flowrate and workin g temperature are also included. A novel feature is the use of system state variables for network training, including additional terms to ac celerate convergence. The approach has also made a significant contrib ution to multiple-fault detection, particularly for the complex three- fault case.