Vibration analysis for early detection of bearing failures

• Main Author: Gam, Kheng Shiang
• Format: Final Year Project / Dissertation / Thesis
• Published: 2024
• Subjects: TJ Mechanical engineering and machinery; TS Manufactures
• Online Access: http://eprints.utar.edu.my/6868/1/ME_2100685_FYP_Report_%2D_KHENG_SHIANG_GAM.pdf; http://eprints.utar.edu.my/6868/

Bearing failures are a leading cause of rotating machinery failures in industries such as oil and gas, manufacturing, and power generation. Early detection of bearing failures using Predictive Maintenance (PdM) is critical for minimizing downtime and optimizing maintenance strategies. The aim of this study is to investigate the vibration characteristics of defective bearings by developing a reliable vibration monitoring system. The vibration monitoring algorithm utilizes time-domain parameters, frequency domain analysis, and envelope analysis to assess bearing conditions. The vibration indicators utilized in this study include K-factor, root mean square (RMS) acceleration, peak acceleration, crest factor, kurtosis, RMS velocity, RMS displacement, and peak-to-peak displacement. This research specifically focuses on types of bearing damage, including lubricant contamination, chemical corrosion, and mechanical damage. The key findings of this study highlight several important observations in bearing fault detection. First, the use of the envelope spectrum has proven to be highly valuable in visualizing specific bearing fault frequencies, which makes it a powerful tool for fault detection. The successful detection of the Ball Pass Frequency Outer (BPFO) was achieved with an error margin ranging from 0.31% to 1.24%. It was also observed that the K-factor, RMS acceleration, and peak acceleration are sensitive to operating speed, which may pose challenges in variable-speed applications. For indicators without speed dependency, thresholds are established based on the two-sigma criteria. The crest factor threshold is set at 7.49, while the kurtosis threshold is established at 4.11. The algorithm's analysis results were consistent with the physical inspection of disassembled bearings. In comparison to a commercial monitoring system, the crest factor was identified as the most consistent indicator for evaluating bearing condition, demonstrating a percentage difference ranging from 2.63% to 65.19%. In summary, the combination of RMS acceleration and kurtosis is proposed as a reliable method for identifying bearing faults.