Paper

Mechanical System Fault Detection using Intelligent Digital Signal Processing


Authors:
Aaron R. Rababaah; Joseph Arumala; Ibibia K. Dabipi; Kenny Fotouhi; Gurdeep Hura; Gurdeep Hura; Avinash Dudi
Abstract
This paper presents an intelligent method for failure detection based on digital signal analysis in offshore wind turbines. The primary goal of this research is to build an intelligent method which can detect mechanical failure in offshore wind turbines. The proposed method is a multi-stage process including signal acquisition, pre-processing, model training, testing, defect pattern detection and result verification. Two primary stages were investigated for design choice: signal characterization which includes feature selection and extraction, and defect classification. For the characterization stage, two techniques were investigated: discrete cosine transform (DCT) and fast Fourier transform (FFT). A pre-examination of the similarity measure among the resulting vectors of the two techniques was conducted and results indicate that a Euclidian-based similarity measure was superior to a correlation-based similarity measure of the signal vectors by a significant factor. FFT was chosen over DCT because FFT naturally relates to the frequency domain of a signal and signifies a direct interpretation of the primary harmonics of the original signal. For the classification stage, two implementations of the process were executed and compared: one implementation did not utilize an intelligent agent and the other utilized a neural network model to classify signal vectors into healthy and damaged classes. The difference between the two implementations was very significant: the intelligent agent demonstrated a very reliable classification accuracy > 90% while the other demonstrated an accuracy of only 53%.
Keywords
Wind Turbine Health Monitoring; NREL Gearbox Data; Artificial Neural Networks; Fourier Transform; Randomized Sampling
StartPage
27
EndPage
39
Doi
Download | Back to Issue| Archive