Fault diagnosis of reciprocating compressors using revelance vector machines with a genetic algorithm based on vibration data

This paper focuses on the development of an advanced fault classifier for monitoring reciprocating compressors (RC) based on vibration signals. Many feature parameters can be used for fault diagnosis, here the classifier is developed based on a relevance vector machine (RVM) which is optimized with genetic algorithms (GA) so determining a more effective subset of the parameters. Both a one-against-one scheme based RVM and a multiclass multi-kernel relevance vector machine (mRVM) have been evaluated to identify a more effective method for implementing the multiclass fault classification for the compressor. The accuracy of both techniques is discussed correspondingly to determine an optimal fault classifier which can correlate with the physical mechanisms underlying the features. The results show that the models perform well, the classification accuracy rate being up to 97% for both algorithms.