Fault detection of bearings based on AutoRegressive modelling and Support Vector Machine classification

Bearings are the common mechanical components and they have an authentic role generally in every kind of machines or applications. Therefore it becomes more and more important to understand properly the behaviour and the nature of the bearing and also their working conditions. For this reason, the data from the components on the machine as bearings in the modern industry; the importance of reducing machine downtime and machine costs, increasing machine performance and achieving a more stable operation is increasing day by day. For that purpose, different solutions have been provided to detect and diagnose faulty situations in bearings with different application-oriented approaches. In this study, by means of vibration based method; the vibration signals coming from the bearings are filtered by Wavelet analysis to avoid non-stationarities and then the filtered signal is modeled according to the AutoRegressive (AR) process. The coefficients of the related model is the basis for the Machine Learning algorithm Support Vector Machines (SVM) by optimizing with Bayesian Optimization to perform Fault Detection and Identification (FDI). This usage of AR process provides a useful forward-looking method for appropriate feature selection. In this study the data which was provided by Center for Intelligent Maintenance Systems is used and the results have been discussed and concluded.