Fault detection and diagnosis using Principal Component Analysis of vibration data from a reciprocating compressor

This paper investigates the use of time domain vibration features for detection and diagnosis of different faults from a multi stage reciprocating compressor. Principal Component Analysis (PCA) is used to develop a detection and diagnosis framework in that the effective diagnostic features are selected from PCA of 14 potential features and a PCA model based detection method using Hotelling's T² and Q statistics is subsequently developed to detect various faults including suction valve leakage, inter-cooler leakage, loose drive belt, and combinations of discharge valve leakage with suction valve leakage, suction valve leakage with intercooler leakage and discharge valve leakage with intercooler leakage. A study of Q -contributions has found two original features: Histogram Lower Bound and Normal Negative log-likelihood which allow full classification of different simulated faults.