High-dimensional data analysis for early fault detection in wind turbines
Abstract »In this study, high-dimensional data analysis methods to address the problem of fault detection of wind turbines are proposed. Fault detection requires continuous monitoring and processing of Big Data generated by the wind turbines and recorded by the supervisory control and data acquisition (SCADA) system. In this paper, random matrix theory and nonparametric kernel distribution are used to model the oscillations of the nacelles of wind turbines, from a wind farm located in Denmark, prior to the occurrence of a major fault in one of the turbines. We establish universality by referring to the asymptotic distribution of the empirical spectral density (ESD) of the sample covariance matrix deviating from the Marchenko – Pastur (MP) law almost surely due to the occurrence of the fault. We establish a nonparametric estimate of the empirical spectral density of the sample covariance matrix with a given kernel function and a bandwidth parameter. We further analyze the empirical eigenvalue density function of the sample covariance matrix and compare with the MP law prior to the fault, during the fault and after the fault has occurred. The results of this study show that deviation from the MP universality law with probability close to one can be seen prior to the occurrence of the fault. To the best of our knowledge, this study is the first one that demonstrates the applications of high-dimensional signal processing techniques such as random matrix theory towards fault detection in wind turbines.
Biography: Prof. Esmaeil S. Nadimi is an associate professor in signal processing at the Maersk Mc-Kinney Moller Inst. Faculty of Engineering, University of Southern Denmark. His main field of research is in Big Data analysis and non-invasive medicine and robotics. https://piseg.sdu.dk
Affiliation: University of Southern Denmark (SDU) Maersk Mc-Kinney Moller Inst. (Electrical Engineering) 5230 Odense DenmarkHerp, JurgenHerp, Jurgen