MLP 기반 솔레노이드 펌프의 고장 분류 연구

Abstract

Purpose: Fluid pumps are crucial in hydraulic and thermodynamic systems. They compel them for prolonged use leading to varous issues, such as fatigue, stress, contamination, and filter clogging. Vibration monitoring of hydraulic components has shown reliable efficiencies in fault detection and isolation (FD&I), considering that vibrational signals reflect operational conditions. Methods: The data were obtainined by operating the solenoid pumps under five conditions: filter clogging, high viscosity, unspecified input power, and normal state. After that, the data were extracted and selected based on statistical features and applied to machine learning to diagnose and classify faults. Results: We used the robust multi-layer perceptron classifier to perform diagnosis after discriminative feature selection. The results showed a reliable test accuracy of 95.6% with a minimal false alarm rate. Conclusion: Accurate diagnostics across operating conditions can be achieved using artificial intelligence by extracting statistical features from solenoid pump vibrational signals.