A Closer Look at Bearing Fault Classification Approaches

Abstract

Rolling bearing fault diagnosis has garnered increased attention in recent years owing to its presence in rotating machinery across various industries, and an ever increasing demand for efficient operations. Prompt detection and accurate prediction of bearing failures can help reduce the likelihood of unexpected machine downtime and enhance maintenance schedule, averting lost productivity. More recent technological advances have enabled monitoring the health of these assets at scale using a variety of sensors, and predicting the failures using modern machine learning (ML) approaches. Deep neural network approaches that has demonstrated superior performance across computer vision and natural language processing tasks are increasingly considered for failure prediction using vibration sensors. Vibration data has been collected using accelerated run-to-failure of overloaded bearings, or by introducing known failure in bearings, under a variety of operating conditions like rotating speed, load on the bearing, type of bearing fault, and data acquisition frequency. However, in using deep neural networks or other ML models for predicting failure using vibration data, there is a lack of consistency in the choice of how these approaches are evaluated, what portion of data from run-to-failure experiments are considered for training failure prediction models, or even how these problems are formulated as machine learning classification problems. An understanding of the impact of these choices is important to reliably develop models, and deploy them in practical settings. In this work, we demonstrate the significance of these choices on the performance of the models using publicly-available vibration datasets and discuss the implications of these choices for the models to be useful in real world scenarios.