Classification Accuracy Analysis of Machine Learning Algorithms for Gearbox Fault Diagnosis

Abstract

AbstractMost of the current research problems are related to early fault diagnosis of various rotating machines. A very small fault in gear teeth increases the vibration response of the gearbox which leads to catastrophic failure. Machine learning algorithms are used to diagnose the faults. This paper presents a comparative analysis of different machine learning algorithms used for gearbox fault diagnosis. The comparison is done by classification accuracy of algorithms on the basis of vibration signals of chipped, eccentric, and broken gear teeth. All the data of various signals (around 2,66,666 samples of each dataset) is performed on different machine learning algorithms. Each data of samples is trained and tested (70 to 30 in ratio). Five machine learning classifiers (Naive Bayes, KNN, decision tree, random forest, and SVM) are considered here for finding the best suitable method. Some of the above algorithms are not used for gearbox fault diagnosis; however, they are also explored in the present work.KeywordsGearbox fault diagnosisMachine learning algorithmsSignal processingVibration analysis