Machine Learning-Based Multi-Fault Diagnosis of Deep Groove Ball Bearings Using Vibration Signal Analysis under Variable Speed Conditions

The most common type of rolling element bearing used in the power and chemical industry, in addition to being used in the automotive industry, is called a deep groove ball bearing (DGBB). These bearings will help machines to operate in a smooth manner with as little noise and friction, but over time they will be affected by high load and contact stresses caused by the motion of the bearings and the loads on those bearing, thus making them very vulnerable to failure as a result of high stresses and fatigue from having a load on the bearing. Even though it may be difficult to detect a minor defect in the bearing by visual inspection, it is very important to discover and repair any defects in a DGBB before they cause a major failure of the bearing and lead to a substantial amount of cost associated with repairing the machine. Using vibration analysis as a technique to identify fault conditions within a bearing has been utilized for many years. There are many patterns of characteristic vibrations generated due to bearing faults in the inner race, outer race or rolling elements of the bearing prior to full bearing failure. Characteristic patterns of vibration can provide valuable information on the condition of a bearing. Vibration signals are frequently analysed using a Fast Fourier Transform (FFT) analyser to convert the time response of these vibration signals into frequency response. The frequency spectrum provides an avenue for determining the existence of faults and their location based upon the fact that each fault type produces vibrations at a distinct frequency or characteristic frequency. By examining the frequency spectrum, one may thus establish frequencies associated with various fault conditions for a bearing. In this study, vibration signals from a healthy (non-defective) bearing were first recorded. After that, different types of defects were intentionally introduced into various components of the bearing. The analysis clearly showed that each defect generated excitation in the system at its own frequency, resulting in peaks in the FFT spectrum for each defect. Additionally, machine learning algorithm was performed to validate how vibration parameters change with increasing speed. It was observed that parameters like unbalance and misalignment become more noticeable as the speed increases.