Prediction of Rolling Bearing Cage Dynamics Using Dynamic Simulations and Machine Learning Algorithms

Abstract

Cage instability or highly dynamic cage movement can have a strong influence on the performance of rolling bearings. In addition to very loud and disturbing noises (“squeal”), bearing failure due to cage fracture can occur. This article deals with two topics: the general classification of cage motions and the prediction of application-dependent cage motions to prevent cage instability during operation. The dependencies of the unstable cage movement on the bearing’s load and geometric characteristics of the cage are analyzed using a large number of sophisticated simulations, based on multibody dynamics. To evaluate the cage movements, first a key figure called the “cage dynamics indicator” (CDI) is introduced, which is used to classify the simulation results by means of quadratic discriminant analysis into three types “unstable,” “stable,” and “circling” (= classification of cage motion). Second, a machine learning algorithm trained and tested on the basis of more than 4,000 simulation results enables a time-efficient prediction of the physical correlations between bearing load and cage properties and the resulting cage dynamics (= prediction of cage motion). A comparison of the calculated cage dynamics with the results of an optical measurement of the cage dynamics rounds off this article. This comparison illustrates the high quality of the simulation models and the training data used for machine learning.