Comparison of fault detection methods for a hydraulic accumulator loading circuit

This paper presents a comparison of some fault detection methods for a hydraulic accumulator loading circuit that can be used as automated condition monitoring tools in a cyber-physical system. Measuring the loading time and the accumulators pressure enables a categorization of different failure types derived from different combinations of a pump or an accumulator failure. The commonly used approach of observing only the loading time is compared to a manual feature engineering approach, a statistical feature extraction and selection approach and an image transformation approach. Tests on a digital twin of the hydraulic accumulator loading circuit show that all of the presented methods outperform the commonly used approach significantly. However, the statstical feature extraction and selection approach delivers the best accuracy.