Fault Diagnostics of Wind Turbine Drive-Train using Multivariate Signal Processing R. Uma Maheswari Anna University, Assistant Professor (Senior Scale), Rajalakshmi Institute of Technology, Chennai-600124, India. R. Umamaheswari Velammal Engineering College, Chennai-600066, India. (Received 5 June 2018; accepted 18 February 2019) The vibration measured from wind turbine drivetrain components is a mixture of multiple frequency modes. In practice, in wind turbine drivetrain condition monitoring systems, multiple accelerometer sensors are used to mea- sure the vibration. Inter-channel common modes are not processed in the standard single-channel empirical mode decomposition (EMD) and it suffers from mode mixing and mode misalignment. Inter-channel correlation implies the causation of vibration mode shapes. Multivariate EMD (MEMD) possesses an enhanced spatial and spectral coherence. The mode alignment property of MEMD is used to process the inter-channel common modes, thus MEMD overcomes the limitation of mode misalignment in single-channel EMD. Still, MEMD exhibits a degree of mode mixing. White noise powers are added in separate channels to lessen the mode mixing. In this research, a novel multivariate signal processing technique, noise-assisted multivariate empirical mode signal decomposition (NA-MEMD) with a competent nonlinear Teager-Kaiser energy operator (NLTKEO), is proposed and tested for truthful extraction of instantaneous frequency and instantaneous amplitude features, and thereby ensures superior fault diagnosis performance. The dyadic filter bank structure of the proposed NA-MEMD decomposes the non- stationary vibrations effectively. The proposed method is used to predict the surface damage pattern embedded in multi-source vibrations at a low-speed planetary gear stage. The effectiveness of the proposed algorithm is tested with NREL GRC wind turbine condition monitoring benchmark datasets. 1. INTRODUCTION The contribution of wind power in the renewable energy sec- tor is growing exponentially. The cost of wind energy is pro- portional to the operation and maintenance (O&M) cost of the wind farms. Unscheduled downtime of the turbines increases O&M cost and also affects their reliability. Predictive condi- tion monitoring is a promising maintenance strategy that re- duces the maintenance costs and optimizes the availability of wind energy. The condition-based maintenance strategy is ca- pable of predicting the system component failure at the incipi- ent stages, thereby abrupt faults can be prevented. With the re- cent advent of sensor technology, the condition monitoring of wind forms shows excellent capabilities in measuring the rotat- ing machine vibration with more precision. Vibration analysis is one of the condition-based monitoring techniques that has been practiced in the industry at large. Time domain and fre- quency domain analysis of measured vibration signals are ex- hausted since these traditional signal processing algorithms are appropriate only for the stationary operating conditions. Wind turbines are nonlinear systems operating at non-stationary op- erating conditions; thus, more sophisticated techniques are re- quired for continuous monitoring and fault diagnosis. Li, et al. proposes fault diagnosis of planetary gears based on multiscale symbolic dynamic entropy. Bandwidth-based envelope inter- polation is proposed for rolling bearing fault diagnostics. 1–4 Various decomposition techniques like intrinsic scale decom- position and local scale decomposition have been studied in both Yu and Liu and Yu and Lv to extract the weak fault fea- tures from rolling bearings. 5, 6 The local mean decomposition is a variant of EMD, which has been applied to decompose the vibration signal into the product of envelope and frequency modulated signal. From this product function, the instanta- neous frequency was estimated to detect the gear crack fault frequencies. 7 To eliminate the mode mixing problem in EMD, ensemble empirical mode decomposition (EEMD) was used by the en- ergy separation algorithm for fault diagnosis. 8 Yu, et al. ap- plied the EMD method to detect the faults of roller bearings. The wavelet-based de-noising was employed on a vibration signal envelope to extract fault patterns. 9 Diagnosing bearings faults by using EMD with variants was studied in Yu, et al., and in Rai and Mohanty. 10, 11 The extracted energy of each intrinsic mode function (IMF) was used as a feature to train the ANN. 10 A Hilbert transform and FFT were used to extract the enve- lope spectrum of computed IMF to diagnosis the bearing fault signatures. 11 Yang, et al. proposed the condition monitoring system for wind turbines by applying EMD on the vibration signal. That amplitude modulation technique on resultant IMF was studied to correlate different types of faults. 12 The major drawbacks in standard single-channel EMD are mode mixing and mode misalignment problems. Mode mix- ing means different oscillation modes are present in a single IMF and mode misalignment corresponds to the appearance of the same mode across different IMFs. 13 In wind turbine vi- bration monitoring, the data are collected at multiple locations with multiple accelerometer sensors and the acquired signals are multivariate in nature. As in simple single-channel ap- proaches, if these vibration signals sensed at different loca- tions are analysed separately, the location information could 334 https://doi.org/ijav.2019.24.21527 (pp. 334342) International Journal of Acoustics and Vibration, Vol. 24, No. 2, 2019