Seeded Fault Testing and Classification of Dynamically Loaded Floating Ring Compressor Bearings

Abstract

This paper investigates a variety of signalmonitoring and datadriven processing techniques to classify seed faults imposed on floating ring main crankshaft compressor bearings. Simulated main bearing shaft motion using an adaptation of the mobility method is first applied to demonstrate the plausibility of the method. Condition monitoring for three different fault types is experimentally investigated through seeded fault testing. A novel method for feature extraction utilizes a fast Fourier frequencydomain transformation coupled with a binning method that uses information across the entire frequency range. A principal component transformation process is then applied to reduce the dimension of the frequencybased feature vector to a small set of generalized features. A Bayesian classifier on the generalized features designed through seeded fault training data is shown to have excellent classifier performance across all fault types. A singleaxis position measurement of the crankshaft shows the most promising results compared to a traditional accelerometer on the bearing housing and a novel accelerometer on the crankshaft. The singleaxis measurement provides a costefficient alternative method to the twoaxis orbit measurement typically used for traditional journal bearings.