AI-Based Hydraulic and Electrical Fault Identification in Direct-Driven Hydraulic Systems

Abstract

Direct drive hydraulic (DDH) systems present a promising alternative to traditional systems in off-road mobile machinery (NRMM) due to their energy efficiency, improved controllability, and lower maintenance costs. Such systems combine both electrical and hydraulic components that are subject to wear and tear and require fault identification, both individually and in combination. Since all system elements are interconnected hydraulically, mechanically, or electrically, changes in the behavior of some components directly affect the operation of others. Consequently, this paper explores an AI-based fault diagnosis method that uses only electrical signals from an electric motor, while simultaneously identifying both hydraulic and electrical faults in DDH systems, eliminating the need for supplementary sensors. This method was tested in an experiment that combines a physical emulation of the hydraulic part of the system based on mathematical modeling and work with an electric motor with real faults. The study covers various faults such as inter-turn winding short circuits, internal cylinder leakage, and gear pump and check valve faults. The obtained data underwent multi-stage preprocessing, as well as machine learning algorithms, mainly support vector machine (SVM) with different kernels, resulting in an overall classification accuracy above 84% in three-fourths of the tested scenarios, with about half exceeding 90%. A masking effect was also observed, where an inter-turn short circuit fault partially obscures the signatures of some hydraulic faults. The obtained results prove the effectiveness and prospects of the proposed fault diagnosis method in both DDH and other electro-hydraulic systems. Such a method can increase the probability of timely fault detection, thereby reducing downtime and maintenance costs while significantly increasing the efficiency of NRMM.