Exploring failure regression for bearing degradation

Bogie bearings performances directly influences the safety of the train. Diagnosis aims to detect the bearing failure, prognosis aims to predict the remaining life of a bearing. This study concerns on bogie bearings prognosis using survival function. Bearing health monitoring providing information of the system state evolved rapidly over the past decade. Root mean square is an efficient indicator of bearing degradation for inspection and maintenance. Numerous degradation models have been developed. However, the failure regression for bearing degradation is not yet well explored. The residual life provides a prediction for future operation has grown as a significant issue for bearing health condition. This article explores the failure regression modeling for bearing degradation. In the industry, bearings operate at different rotation and load conditions. The developed models for bearing residual’s life are limited to a single operating condition. Failure regression considers both rotation and load as operating conditions. Root mean square or kurtosis (time domain features) are extracted and used as bearing degradation indicators. Femto technology developed experimental platform that provides time to failure in specific conditions of operating. horizontal and vertical vibrations are measured every 10 s at 25.6 KHz by accelerometers. The time to failure is observed for bearing under three speed and three load conditions. The speed and load are considered as covariates in failure regression. The failure regression yields the bearing reliability, and this conditional reliability is used to calculate the bearing residual life. In the practice, bearing can be operated at different speeds, loads and operating conditions. The effect of internal and external covariates needs to be explored in order to provide better residual life prediction and to optimize to maintenance strategy. Failure regression can be considered as a tool for bearing residual life prediction in presence of internal and external covariates.