Fault diagnosis in reciprocating compressor bearings: an approach using LAMDA applied on current signals

Abstract

Condition monitoring is one of the most important activities to implement predictive maintenance in industrial processes and perform fault diagnosis. Vibration is the most used signal for this purpose, however current signals arise as a non-intrusive alternative to condition monitoring. On the other hand, data driven approaches becomes as a way to develop fault classifiers by using Machine Learning. This paper proposes the development of a fault classifier for diagnosing failures in the bearings of a reciprocating compressor by using the current signals measured from the induction machine that power the mechanical device. The proposal applies cluster validity assessment for feature selection, and a LAMDA-based model for classification. Results show that this proposal can diagnose three failure modes with a precision over 90%.