Condition-based monitoring system for diagnostics and prognostics of centrifugal pumps

Abstract

Centrifugal pumps are the most commonly used pumps in industry. Therefore, maintenance is a very important part of the pump's life cycle. This work presents a systematic robust conditionbased monitoring process for diagnostics and prognostics of centrifugal pumps. The process includes industrial design of experiments; data acquisition software and hardware; transient modeling using transfer functions, artificial intelligent control methods; and spectral analysis using Fast Fourier Transform. Industrial design of experiments used in this work includes orthogonal arrays or Taguchi method, full factorial design, and response surface methodology. As for the data acquisition system, NI DAQ (National Instruments) hardware were used interactively with LabVIEW software to collect the data from the different sensors. Transient modeling using time domain data was established using MATLAB's modeling toolbox, identification toolbox. Artificial intelligent control methods used were NARMA-L2 and Predictive Neural Networks, Fuzzy Logic control method, and Neuro-Fuzzy control method. Lastly, spectral analysis using Fast Fourier Transform was performed on the vibrations of the system.