Two-step fault diagnosis framework for rolling element bearings with imbalanced data based on GSA-WELM and GSA-ELM

Abstract

Rolling element bearings constitute the key parts on rotating machinery, and their fault diagnosis is of great importance. In many real bearing fault diagnosis applications, the number of fault data is much less than the number of normal data, i.e. the data are imbalanced. Many traditional diagnosis methods will get low accuracy because they have a natural tendency to favor the majority class by assuming balanced class distribution or equal misclassification cost. To deal with imbalanced data, in this article, a novel two-step fault diagnosis framework is proposed to diagnose the status of rolling element bearings. Our proposed framework consists of two steps for fault diagnosis, where Step 1 makes use of weighted extreme learning machine in an effort to classify the normal or abnormal categories, and Step 2 further diagnoses the underlying anomaly in detail by using preliminary extreme learning machine. In addition, gravitational search algorithm is applied to further extract the significant features and determine the optimal parameters of the weighted extreme learning machine and extreme learning machine classifiers. The effectiveness of our proposed approach is testified on the raw data collected from the rolling element bearing experiments conducted in our Institute, and the empirical results show that our approach is really fast and can achieve the diagnosis accuracies more than 96%.

Network-combined broad learning and transfer learning: a new intelligent fault diagnosis method for rolling bearings

“…To simulate unbalanced data conditions for the collected data of the rolling bearings, source domain datasets were constructed according to different values of the ratio (R), defined as the ratio of the number of faulty data to the number of normal data [43]. If the number of samples is set to 4950, the R is set to 1, 2/3, 1/2, and 1/10 for which R = 1 signifies a balanced dataset; R = 2/3 a slightly unbalanced dataset, R = 1/2 a moderately unbalanced dataset, and R = 1/10 an extremely unbalanced dataset.…”

An Oversampling Method of Unbalanced Data for Mechanical Fault Diagnosis Based on MeanRadius-SMOTE

“…From the perspective of the classifier, scholars mainly set the cost matrix, and change the loss function or network structure to make the classifier aware of this imbalance [ 35 ]. These classifiers are often only suitable for identifying faults in stationary parts, such as gears or bearings [ 36 ].…”

Neighborhood graph embedding interpretable fault diagnosis network based on local and non-local information balanced under imbalanced samples

“…Chen et al 25 used a multi-classifier integrating random sampling and weighted balance distribution to complete rolling bearing unbalance fault diagnosis. Lan et al 26 used a two-stage fault diagnosis framework of extreme learning machine and its variants to cope with rolling bearing fault diagnosis in data imbalance conditions. Razavi-Far et al 27 used an interpolated oversampling method for hybrid fault diagnosis of asynchronous motor bearings under unbalanced conditions.…”