Vibration Based Condition Monitoring of Air Compressor Through Histogram Features and Dagging Classifier

Air compressors are one of the crucial mechanical equipments used in diverse applications like gaseous plants, thermal power plants and petroleum & petro-chemical industries. The uninterrupted (fault free components) working of such significant systems avoids the major losses due to critical causalities and system seizures. The main aim behind the study is to monitor the system conditions continuously to identify, locate and rectify the occurrence of failures at beginning stage. The vibration signals are measured as physical parameters from the compressor setup for six different test conditions namely, good, inlet valve fluttering, outlet valve fluttering, inlet& outlet valve fluttering, valve plate leakage and check valve fault. Histogram features are extracted from the vibration signals and the J48 decision tree algorithm was used for selecting the most prominent features among the given input features. Dagging under Meta classifier was used as a machine learning algorithm for classifying multiple fault conditions of the compressor and the results were presented based on its classification accuracy and computational time