Data and computation efficient model-based fault detection for rolling element bearings using numerical differentiation

Abstract

A novel model-based approach for online condition monitoring of rolling element bearings based on efficient and robust numerical differentiation is proposed. The physical parameters may be identified using fault-free data. The main idea for fault detection is the design of signals, called residuals, sensitive only to faults using a simplified model of the bearing vibrations. These residuals are proportional to the force stemming from the faults and are easy to compute and implement. An approach for the design of a failure detection threshold is provided. This threshold can be designed to meet a desired probability of false alarms. The method is validated using four experimental data sets stemming from different research centers.