Soleiman Hosseinpour, Witold Kinsner, Nariman Sepehri

Abstract: This paper proposes a novel approach for fault detection in an electro-hydrostatic actuation (EHA) system, focusing on detecting system leakage. Bayesian optimization is integrated within a neural network framework for tuning the hyperparameters. This new approach enhances the network's capability to classify faults with higher accuracy. To detect faults effectively, we utilize a polyscale complexity measure known as variance fractal dimension (VFD), which extracts critical features from the signal data. These features are fed into the Bayesian-optimized neural network, forming an effective fault detection model. We compare the performance of our Bayesian-optimized neural network against traditional classification methods, including support vector machines, decision trees, and random forests. The results demonstrate that our approach not only improves fault detection accuracy but also outperforms these conventional methods. This establishes its potential as a reliable technique for fault detection in hydraulically actuated systems.

Keywords: Electro-hydrostatic actuation system, fault detection, internal leakage, Bayesian optimization, artificial neural network, polyscale complexity measures.

Date Published: November 21, 2024
DOI: 10.11159/jmids.2024.016

View Article