Vibration signals based fault severity estimation of a shaft using machine learning techniques

Estimation of fault severity of a shaft is important for scheduling the maintenance in rotating machines. Higher sensitivity towards faults and ease of use make vibration based fault detection techniques more preferable over other techniques. Experimental study is performed on a rotating shaft in a machine fault simulator with dual vibration sensors. Vibration signals acquired from the rotating machines is used to generate a classification and regression model for estimating the severity of fault in the rotating shaft using machine learning techniques. This paper attempts to classify the features extracted from raw vibration signal data with the help of decision tree classifier and then performing regression analysis using tree regression. In this process we first train the classifier and regression model with a set of statistical features such as kurtosis, skewness, standard deviation, mean, median etc. and build a model, which when provided with a new set of test data can identify the level of the fault severity associated with the rotating shaft. Fusing of data from two similar sensors has resulted in better results. (C) 2018 Elsevier Ltd. All rights reserved.