Health Assessment and Fault Classification for Centrifugal Pump Using Logistic Regression

Real-time health monitoring of industrial components and systems that can detect, classify and predict impending faults is critical to reducing operating and maintenance cost. This paper presents a logistic regression based prognostic method for on-line health assessment and failure modes classification. System condition is evaluated by processing the information gathered from access controllers or sensors mounted at different points in the system, and maintenance is performed only when the failure/ malfunction prognosis indicates instead of periodic maintenance inspections. The wavelet packet decomposition and fast Fourier transform (FFT) technique is used to extract features from non-stationary vibrations signals, wavelet package energies and fundamental frequency amplitude are used as features and Principal Component Analysis (PCA) is used to features reduction. Reduced features are input into logistic regression (LR) models to assess machine health condition and identify possible failure modes. The maximum likelihood method is used to determine parameters of LR models. The effectiveness and feasibility of this methodology have been illustrated by applying the method to a real centrifugal pump.