Abstract
Health monitoring for rotating machines is investigated through two kinds of mock-up experimental data analysis. One is an anomaly mock-up test of roll bearing type rotating machine. Here, inner ring defect anomaly is simulated and its operating data are measured by both attached type accelerometer sensor and non-attached type microphone. Three kinds of signal pre-processing methods, frequency spectrum, principal component analysis and cepstrum, are applied to discriminate normal and abnormal states using several different classification algorithms, such as adaboost or random forest. Through analysis of their performance with the help of receiver operating characteristic (ROC) curve, the importance of diversified health monitoring methods is discussed. Another mock-up experiment is an accelerated test of roll bearing wear. Here, acoustic emission counts, accelerometer signal and wear particle number in lubricating oil are measured. Using these observation data, we make clear the relationships between deterioration mechanisms of bearing and behaviour of different observations.
