Research Papers: Gas Turbines: Structures and Dynamics

Wind Turbine Gearbox Fault Detection Using Multiple Sensors With Features Level Data Fusion

[+] Author and Article Information
Y. Lu

Department of Mechanical Engineering,  University of Connecticut, 191 Auditorium Road, Unit 3139, Storrs, CT 06269

J. Tang1

Department of Mechanical Engineering,  University of Connecticut, 191 Auditorium Road, Unit 3139, Storrs, CT 06269jtang@engr.uconn.edu

H. Luo

 Global Technical Leader – Wind, Machinery Diagnostics Services, GE Energy Services, 1 River Road, Schenectady, NY 12345


Corresponding author.

J. Eng. Gas Turbines Power 134(4), 042501 (Jan 25, 2012) (8 pages) doi:10.1115/1.4004438 History: Received May 05, 2011; Revised May 09, 2011; Published January 25, 2012; Online January 25, 2012

Fault detection in complex mechanical systems such as wind turbine gearboxes remains challenging, even with the recently significant advancement of sensing and signal processing technologies. As first-principle models of gearboxes capable of reflecting response details for health monitoring purpose are difficult to obtain, data-driven approaches are often adopted for fault detection, identification or classification. In this paper, we propose a data-driven framework that combines information from multiple sensors and fundamental physics of the gearbox. Time domain vibration and acoustic emission signals are collected from a gearbox dynamics testbed, where both healthy and faulty gears with different fault conditions are tested. To deal with the nonstationary nature of the wind turbine operation, a harmonic wavelet based method is utilized to extract the time-frequency features in the signals. This new framework features the employment of the tachometer readings and gear meshing relationships to develop a speed profile masking technique. The time-frequency wavelet features are highlighted by applying the mask we construct. Those highlighted features from multiple accelerometers and microphones are then fused together through a statistical weighting approach based on principal component analysis. Using the highlighted and fused features, we demonstrate that different gear faults can be effectively detected and identified.

Copyright © 2012 by American Society of Mechanical Engineers
Your Session has timed out. Please sign back in to continue.



Grahic Jump Location
Figure 1

Gearbox dynamics testbed with multiple sensors

Grahic Jump Location
Figure 2

Typical time domain sensor signals (unit: volt). (a) Accelerator #1; (b) Accelerator #2; (c) Microphone #1 (d) Microphone #2 (e) Tachometer.

Grahic Jump Location
Figure 3

(a) A typical vibration signals collected from an accelerometer and its time-frequency representations using (b) the harmonic wavelet, (c) Daubechies 4 wavelet and (d) Morlet wavelet, respectively

Grahic Jump Location
Figure 4

A tachometer signal and the motor speed contour converted from it

Grahic Jump Location
Figure 5

(a) Speed profile mask constructed from the tachometer signal. Solid contours correspond to the gear pair between input and intermediate shafts, while dotted contours correspond to the gear pair between intermediate and output shafts. (b) Time-frequency features highlighted using the constructed speed profile mask.

Grahic Jump Location
Figure 6

Harmonic wavelet map from (a) a representative accelerometer (#1) and (c) a representative microphone (#1), along with (c) the fused wavelet map for all four sensors

Grahic Jump Location
Figure 7

(a) Normally distributed random noise simulating the signal from a failed accelerometer. (b) Wavelet features extracted from the signal and highlighted using the speed profile masking as if the sensor is normal.

Grahic Jump Location
Figure 8

Wavelet features fused from multiple sensors, including the failed accelerometer, using (a) simple averaging and (b) the PCA based method

Grahic Jump Location
Figure 9

Fault diagnosis process based on statistical analysis of baseline library and test features

Grahic Jump Location
Figure 10

Absolute T scores calculated using wavelet features from the healthy gearbox

Grahic Jump Location
Figure 11

Absolute T scores calculated using wavelet features from a faulty gearbox: eccentricity

Grahic Jump Location
Figure 12

Absolute T scores calculated using wavelet features from a faulty gearbox: surface wear




Some tools below are only available to our subscribers or users with an online account.

Related Content

Customize your page view by dragging and repositioning the boxes below.

Related Journal Articles
Related eBook Content
Topic Collections

Sorry! You do not have access to this content. For assistance or to subscribe, please contact us:

  • TELEPHONE: 1-800-843-2763 (Toll-free in the USA)
  • EMAIL: asmedigitalcollection@asme.org
Sign In