Ensemble of extreme leaming machines for diagnosing bearing defects in non-stationary environments under class imbalance condition

Two practical inevitabilities for diagnostic systems are the abilities of incremental learning in non-stationary environments and diagnosing under the dass imbalance condition. The dass imbalance condition has been widely occurred in real applications where system usually works in the normal state and it is not easy to collect the representative patterns of faulty dasses. This work aims to adapt two state-of-the-art ensemble-based techniques for incremental learning and diagnosing faults in non-stationary environments under the dass imbalance. These techniques train several extreme learning machines to create the ensemble which can incrementally learn the relation between features and faults in various dass-imbalanced chunks of data collected from non-stationary environments. These diagnostic schemes are applied to diagnose bearing defects in induction motors.