Machine Learning Based Dynamic Failure Criteria for Reliability Analysis of Bearings

Accurate estimation of the reliability for different mechanical components plays an important role in the design and maintenance of mechanical systems. In this regard, a new method is proposed for increasing the accuracy of reliability prediction of bearings by introducing a new approach to determine dynamic failure criteria. To be specific, a Bayesian network classifier is applied to establish a machine learning approach for the determination of failure criteria at each time step with varying working and physical condition. The resulted failure criteria at each time are utilized together with a Kriging estimator to express an updated limit state function. Consequently, the second order reliability method is used for the calculation of time- varying reliability. Finally, the presented method is applied for reliability analysis of rolling element bearings and the resulted reliability curve for both accelerated and normal working conditions are presented. The outcome of this work can result in a pertinent approach for further calculation of the reliability of complex mechanical systems.