PROGNOSTICS OF HIGH SPEED BEARING USING SOFT COMPUTING METHODS

Abstract

Condition monitoring and fault diagnosis of rolling element bearings (REBs) are very important to make sure the steadiness of industrial and domestic machinery. An attempt has been made in this report to reflect the different causes because of which bearing defect occur. There are various method for the analyses of the bearing like thermal analysis, vibration analysis etc for reviewing the condition of the bearing. For extracting the features from the vibration signal different Signal processing technique in time domain, frequency domain and time frequency domain are used for fault diagnosis of the bearing. Various classifiers are examined, including artificial neural networks (ANN), support vector machines (SVM), decision trees, and K closest neighbour (KNN), all of which are used to determine the type of bearing fault. In this case, a deep learning method such as CNN was applied. Bearing prognosis is carried out by extracting the vibration signal and extracting various time domain and spectral kurtosis features. The feature's rank was computed using monotonicity and the feature was picked based on the rank, and the feature's dimensionality was decreased using PCA. To determine the severity of the degradation, an exponential degradation model was fitted to the health indicator and ANN was applied to the selected feature. Different machine learning methods, such as least square support vector regression, were used for comparison, and the root mean square error was calculated.