An expert system utilizing time waveform data for fault diagnosis of rotating machinery

The study presented in this thesis provides an alternative to the current methods which are used to analyze large quantities of vibration data. This system is designed to perform diagnostics on the rotating machinery that are commonly found in industrial production facilities. Current automated techniques generally analyze only spectral data, but the approach presented herein utilizes a machine's time waveform vibration signature. The algorithms that are presented calculate a number of waveform parameters that characterize certain visible fault patterns that are associated with defects in rotating machinery. A set of diagnostic rules was developed that use the calculated parameters as inputs. The calculation of the waveform parameters and their subsequent input into the diagnostic rules results in a diagnosis as to whether or not the machine from which data was taken is operating within acceptable limits. The diagnostic system does not formulate a precise diagnosis, but simply tells the user if a given machine requires further analysis. Thus, the waveform diagnostic system serves as a screening tool for a vibration analyst when examining a large volume of vibration data. The results of using this technique to analyze data that was collected on over one hundred machines for each of six routine measurement surveys was found to have a potential to reduce an analyst's diagnostic and reporting time by a factor of one-half. This would potentially allow an analyst skilled in the use of this diagnostic system to monitor twice the number of machines that was previously included in a condition monitoring program.