Work

Fault Characterisation System based on Vibration Signal Analysis Using Machine Learning Techniques

Year: 2025
Type: article
Abstract: Vibration measurements have been used widely in industry to determine the state of health of machinery, especially rotating equipment. In particular, they have been used for various forms of fault detection including damage to bearings, rotating unbalance and misalignment. Appropriate use of vibration monitoring can result in early detection of impending problems thus preventing unscheduled shutdowns of operations. However, the effectiveness of vibration based condition monitoring is integrally tied to the use of an expert to interpret the data provided. In this work, an automated machine learning approach for detecting and characterising faults in rotating machinery based on accelerometer readings is presented. The vibration waveforms are preprocessed using Fourier transforms and wavelets. Key statistical parameters of these curves are explored including the skewness and the kurtosis; these are then used to isolate identifying features for various common faults in rotating machinery. These features are then fed into a supervised machine learning algorithm and a fault classification system is built. The method described herein is validated with vibration data collected from a laboratory fault simulator. The performance of the system is discussed and its applicability to industrial situations is described.
Language: English
Open Access status: diamond
OpenAlex ID: W7217174588
FWCI:
Cites: 7
Cited by:
Related to: 10