Fault Detection of Bearing using XGBoost Algorithm and Data Visualization using t-distributed stochastic neighbor embedding (t-SNE) Method

Abstract

For Industry, it is essential to correctly detect the faults of the rotating machinery which are caused by the rotor body defects, bearing defects, and gear defects. The majority of times breakdown of the machines is caused by the bearing defects hence, in this paper, we have considered the bearing defects. The data is collected from the test-rig and the collected data usually will be more hence, it becomes extremely difficult to solve the problems of high dimensionality attributes of the features, that needs to be extracted from the vibrational signals, once the extraction is performed which helps us in to classify the faults correctly. To determine high dimensional attributes, t-distributed stochastic neighbor embedding (t-SNE) is presented to visualize the data. This paper employs an XGBoost algorithm to classify the faults, which implements a gradient decision tree. XGBoost model is known for its faster calculations and good efficiency and along with this XGBoost model is compared to other models like SVM and Adaboost algorithms. The results from XGBoost shows better accuracy in comparison with results from other models.