Vibration Data Analysis for Fault Detection in Manufacturing Systems - A Systematic Literature Review

Early fault detection in manufacturing systems is vital to avoid unintended downtime of production and loss of revenue. To enable early fault detection, vibration data analysis is widely used as it provides rich health information and can be acquired without disrupting the production flow. With this paper, we summarise and categorise existing approaches for fault detection in vibration signals, which mainly deal with bearings in rotational equipment. We conducted a pivotal literature review to identify the state of the art in terms of vibration data analysis to detect faults in manufacturing systems. Then, we compared the different approaches based on their benefits and limitations and discussed the appropriate use cases and requirements for each approach. Fault detection in vibration signals can be divided into two sub-areas: traditional signal processing, such as time/frequency domain analysis and Wavelet Transform, and classification approaches using a variety of Artificial Intelligence strategies. Based on our findings, each approach serves well under certain circumstances, e.g., neural network methods have high accuracy but require large amounts of data compared to other classification methods.