Fault diagnosis of air compressor set-up using decision tree based J48 classification algorithm

Abstract

This paper describes an air compressor fault diagnosis method based on the acquisition of audio signals pertaining to seven faulty and one healthy condition. These audio signals are processed using Local Mean Decomposition (LMD) signal processing technique. Further, statistical indicators have been evaluated for feature extraction considering the decomposed signals. From different statistical indictors mean, variance, root mean square (RMS), root mean amplitude (RMA), absolute mean amplitude (AMA), kurtosis, peak to peak index, waveform index, peak index, impulse index, margin index, skewness, Shannon entropy, standard deviation, log energy entropy, log detector and CPT has been selected for decision tree based J48 classification algorithm. Decision tree based J48 classification algorithm more accurately identified healthy and faulty bearing state in an air compressor with the help of all statistical indicators. Data set consists of 360 instances having 17 attributes with 2 classes (healthy and bearing fault). Higher classification accuracy of J48 algorithm (96.66 %) has been obtained for healthy and faulty bearing conditions. LMD along with decision tree based J48 classification algorithm is quite suitable for processing and monitoring in situ fault features in air compressor set-up.