Bearing fault diagnosis based on cross-machine statistical features generalization and improved extreme learning machine

Asset reliability is among the primary objectives in technological advancements and effective maintenance is essential to guarantee optimal performance of machineries while upholding safety requirements.Intelligent models based on machine learning and deep learning techniques have been extensively suggested for advanced maintenance procedures.In recent times, there has been a trend in fault diagnosis studies towards cross-machine diagnosis which involves multiple machines.Therefore, this paper proposes a cross-machine bearing fault diagnosis trained without faulty data of target machine; based on selected generalized statistical vibration features and improved extreme learning machine.This work utilized an online bearing dataset from a source machine and experimental datasets from a target machine.The statistical vibration features were derived from both datasets (online and experimental) and subsequently chosen based on distinctive characteristics in features.Next, specific characteristics will be input into the improved extreme learning machine (ELM) technique for the purpose of fault categorization.The suggested model demonstrated substantial cross-machine classification ability, with an accuracy rate of up to 98.9%.