Vibration analysis based predictive maintenance approach for prediction of remaining useful life of bearings

Predictive maintenance is a method that enables the early prediction of failures by analyzing measurements obtained from equipment through sensors. This maintenance approach involves various techniques for accurately monitoring the health condition of equipment. These techniques encompass methods that can operate locally or in cloud-based environments and incorporate data analysis and machine learning algorithms. Predictive maintenance offers cost advantages compared to unplanned and periodic maintenance. Elevators are an indispensable part of modern society, and their safe and uninterrupted operation is of great importance. However, unexpected failures and interruptions jeopardize user safety and can lead to significant financial losses for business owners. Bearings are one of the most frequently malfunctioning components in elevators. Bearings are components that reduce friction between moving parts or facilitate rotational or linear movement by restricting motion in the desired direction. They are exposed to challenging operating conditions such as excessive loads, high speeds, and inadequate lubrication. Under these conditions, factors such as increased clearance, friction force, and excessive heating can result in performance degradation and failures. Failure to detect these failures in a timely manner and neglecting proper maintenance can lead to bearing deterioration. In this thesis study, a predictive maintenance model has been developed to predict potential failures in bearings by analyzing vibration data obtained from five bearings with the same characteristics attached to the shaft of an asynchronous motor. The dataset related to bearing failures was obtained from the study titled "A Hybrid Prognostics Approach for Estimating Remaining Useful Life of Rolling Element Bearings" by Biao Wang, Yaguo Lei, Naipeng Li, Ningbo Li, published in IEEE Transactions on Reliability, vol. 69, no. 1, pp. 401-412, 2020", which is openly available to researchers. The vibration signals obtained from bearings were processed using Fast Fourier Transform from the time domain to the frequency domain and Wavelet Transform from the frequency domain to the time-frequency domain in the study. In the time domain, features such as mean, standard deviation, RMS, peak-to-peak, kurtosis, and crest factor were calculated. The peak frequency feature was utilized in the frequency domain. Additionally, the average peak frequency feature was obtained in the time-frequency domain. Subsequently, a selected number of features based on monotonicity levels were combined as a health indicator using principal component analysis. This health indicator represents the health condition of the bearing with a single value. Finally, using this health indicator, the remaining useful life of bearings was predicted in the verification dataset. This study aims to employ predictive maintenance methods for the early prediction of bearing failures in elevators. The obtained results could be a significant step towards optimizing elevator maintenance processes, preventing unexpected failures, and reducing operational costs. Furthermore, this study can serve as an example demonstrating the application of data analysis and machine learning techniques in industrial settings.