Supervised Classification Methods in Condition Monitoring of Rolling Element Bearings

Operational vibrational diagnostics is crucial for providing the reliability of mid and large scale combustion engine applications (e.g. railway, automotive heavy vehicles or electric generators). This work reports study presenting application of supervised learning and classification methods based on pattern recognition using different classifiers (e.g. logistic regression, k-nearest neighbor or normal density) in order to detect early warning diagnostic symptoms of malfunctioned rolling element bearings (REBs) in the presence of background disturbances from combustion diesel engine. The REB's malfunction type classification is based on time domain (RMS, peak to peak, Crest factor) as well as frequency domain signal processing methods like envelope analysis or modulation intensity distribution (MID) which allows to neglect the influence of background noise representing by non-stationary operating conditions and possible structural modifications (e.g. maintenance activities or parts replacing). The proposed data classification methods are compared and validated by using experimental measurements conducted on a dedicated combustion engine test bench for wide range of rotational speed and different levels of REB's radial load.