The Condition Based Monitoring for Bearing Health

Abstract

Bearing is a small component that widely uses in industries, either in rotary machines or shafts. Faulty in bearing might cause massive downtime in the industries, which lead to loss of revenue. This paper intends to find the consequential statistical time-domain-based features that can be used in classification from accelerometry signals for the bearing condition. An accelerometer was used as the data logger device to attain the condition signals from the bearing. Machinery Failure Prevention Technology (MFPT) online dataset has three different bearing conditions: baseline condition, inner faulty condition, and outer faulty condition. Extraction of eight statistical time-domain features was done, which is root-mean-square (RMS), minimum (Min), maximum (Max), mean, median, standard deviation, variance, and skewness. The identification of informative attributes was made using a filter-based method, in which the scoring is done by using the Information gain ratio. For the extracted features, the data splitting of training data to testing data was set to the ratio of 70% and 30%, respectively. The selected feature for classification is then fed into various types of classifiers to observe the effect of this feature selection method on the classification performance. From this research, six features were identified as the significant features: variance, standard deviation, Min, Max, mean, and RMS. It is said that the classification accuracy of the training data and the testing data using the filter-based feature selection method is equivalent to the classification accuracy of all the features selected.