Multiple Principle Component Statistics for Bearing Condition Monitoring

This paper presents a novel approach for detecting rolling element bearing damage in industrial electric motors utilizing an embedded Micro-Electro-Mechanical Systems (MEMS) vibration sensor system, focusing on the comprehensive processing of spectral data from multiple sensor axes directions. A baseline measurement is performed during the deployment phase, when bearings are assumed undamaged and driven loads are connected to the motor. The period for the baseline measurement is selected to capture all expected motor loading conditions and associated vibrations. Spectral information from the baseline measurement is dimensionally reduced using Principal Component Analysis (PCA). The principal components are subsequently utilized to form a combined Hotelling T-squared statistic, from which an upper control limit for future bearing condition monitoring is calculated. As the motor operates in the field, new vibration data is continuously projected onto the established principal components to monitor bearing conditions. The proposed technique was validated experimentally and demonstrated enhanced sensitivity and accuracy in early detection of bearing damage due to lubrication starvation and bearing discharge current damage, promising improved reliability and maintenance efficiency in electric motor applications.