Prediction of air compressor condition using vibration signals and machine learning algorithms

Air compressors are widely used equipment in the modern world for their tremendous utilization in applications of both domestic and industrial sectors. The inbuilt mechanical parts are often prone to various failures due to the complexity in the construction of air compressors that affects the overall system process. Hence, it is essential to devise a methodology to identify the failures at the early stages of its operation to avoid the major causalities due to process breakdown and system seizure. In this study, a single-acting single-stage reciprocating air compressor was chosen. The fault conditions like inlet valve fluttering, outlet valve fluttering, valve plate leakage, and check valve fault were considered. The statistical, histogram and autoregressive moving average features were extracted from the raw vibration signals. The most dominating features were selected using a decision tree algorithm and those features were classified using machine learning classifiers like Lazy K Star, Decorate, and radial basis function networks. The classifier Lazy K Star on autoregressive moving average feature exhibits the highest fault classification rate of 99.67% in classifying various compressor conditions and the results were compared and presented.