Fault Detection of Anti-friction Bearing using Ensemble Machine Learning Methods

Anti-Friction Bearing (AFB) is a very important machine component and its unscheduled failure leads to cause of malfunction in wide range of rotating machinery which results in unexpected downtime and economic loss. In this paper, ensemble machine learning techniques are demonstrated for the detection of different AFB faults. Initially, statistical features were extracted from temporal vibration signals and are collected using experimental test rig for different input parameters like load, speed and bearing conditions. These features are ranked using two techniques, namely Decision Tree (DT) and Randomized Lasso (R Lasso), which are further used to form training and testing input feature sets to machine learning techniques. It uses three ensemble machine learning techniques for AFB fault classification namely Random Forest (RF), Gradient Boosting Classifier (GBC) and Extra Tree Classifier (ETC). The impact of number of ranked features and estimators have been studied for ensemble techniques. The result showed that the classification efficiency is significantly influenced by the number of features but the effect of number of estimators is minor. The demonstrated ensemble techniques give more accuracy in classification as compared to tuned SVM with same experimental input data. The highest AFB fault classification accuracy 98.12% is obtained with ETC and DT feature ranking.