Predicting the Remaining Useful Life of Ball Bearing Under Dynamic Loading Using Supervised Learning

Abstract

Rolling element bearing is one of the most critical components of rotating machinery. Its failure can be catastrophic and often results in both human and material losses. This paper presents a machine learning model to predict the wear process phenomena and remaining useful life of a bearing element using classification and regression techniques respectively. An algorithm is developed to recognize the underlying mapping function directly from the data using machine learning principles. Pearson correlation methodology is used to track the important features associated with the evolution of wear and understand its progression. Further, backward elimination technique with ordinary least squares regression results was used to track features for predicting the remaining useful life. The proposed approach is illustrated on a bearing failure data set from the national aeronautics and space agency. This study will be useful in forecasting the fault status of the bearing before it causes any major loss.