Automated valve condition classification of a reciprocating compressor with seeded faults:experimentation and validation of classification strategy

Abstract

This paper deals with automatic valve condition classification of a reciprocatingprocessor with seeded faults. The seeded faults are considered based on observationof valve faults in practice. They include the misplacement of valve and springplates, incorrect tightness of the bolts for valve cover or valve seat, softening ofthe spring plate, and cracked or broken spring plate or valve plate. The seededfaults represent various stages of machine health condition and it is crucial to beable to correctly classify the conditions so that preventative maintenance can beperformed before catastrophic breakdown of the compressor occurs. Considering thenon-stationary characteristics of the system, time–frequency analysis techniques areapplied to obtain the vibration spectrum as time develops. A data reductionalgorithm is subsequently employed to extract the fault features from the formidableamount of time–frequency data and finally the probabilistic neural network isutilized to automate the classification process without the intervention of humanexperts. This study shows that the use of modification indices, as opposed to theoriginal indices, greatly reduces the classification error, from about 80% down toabout 20% misclassification for the 15 fault cases. Correct condition classificationcan be further enhanced if the use of similar fault cases is avoided. It is shownthat 6.67% classification error is achievable when using the short-time Fouriertransform and the mean variation method for the case of seven seeded faults with 10training samples used. A stunning 100% correct classification can even be realizedwhen the neural network is well trained with 30 training samples being used.