Author(s): Albert Kindl; Vladimir Haban; Petr Konas; Martin Hudec; Pavel Rudolf

Linked Author(s):

Keywords: No Keywords

Abstract: In this article, we present the basic possibilities for determining a fault on a hydraulic machine using artificial intelligence and machine learning methods. Signals are obtained from pressure, acceleration, microphone, and acoustic emission sensors, which are suitably placed on the observed hydraulic machine. The sampling frequency on all sensors was set to 200kHz, and the measurement time for one file was 4 s. The centrifugal pump was measured under laboratory conditions in a fault-free state and in a state with a fault that was artificially induced by drilling one of the blades. The measured signals were subsequently processed into a MEL-spectrogram. These MEL-spectrograms were used for AI (artificial intelligence) training, both for supervised and unsupervised cases. The article presents the results of supervised AI training.

DOI: https://doi.org/10.1088/1755-1315/1561/1/012045

Year: 2025