Normalization Process Based on Kernel Ridge Regression Applied on Wind Turbine IAS Monitoring
Variable speed wind turbines use the available wind resource more efficiently than a fixed speed wind turbine, especially during light wind conditions. This enhancement forces the monitoring methods to deal with these large variations in speed and torque, since the conditions are seldom if ever stationary. The unsteady behavior of these wind turbines is also a difficulty in terms of long term diagnostic, since the comparison of successive measurements is usually performed under the same operating conditions. Normalization of the indicators according to well-chosen variables might bring a valuable tool regarding several aspects.
In this paper, the attention is focused on the regression process using classical machine learning tools. The difficulty is to design a process able to efficiently estimate the behavior of any indicator depending on the environmental conditions. Indeed, indicator multivariate laws are expected to present extremely varied shapes, and using common linear regression technique can hardly solve this issue. Kernel machines are therefore presented in this paper as an efficient solution to normalize the indicators, and will be shown to ease the health monitoring of the wind turbine shaft line on a practical case based on instantaneous angular speed signals. The example presents the distinctive feature to have a defect visible only specific operating conditions. This operating conditions being unknown a priori, this example clearly enlightens the need of such a regression tool.
KeywordsParametrization Kernel machine Support vector machine Non-stationary conditions Instantaneous Angular Speed
The authors would like to thank ENGIE Green for their generous financial support of the work. This work has also been supported by the French Institute Carnot Télécom et Société Numérique.
- 7.Saunders C, Gammerman A, Vovk V (1998) Ridge regression learning algorithm in dual variables. In: Proceedings of the 15th International Conference on Machine Learning, ICML 1998Google Scholar
- 11.Nabney IT (1999) Efficient training of RBF networks for classification. Technical report NCRG/99/002, Aston University, Birmingham, UKGoogle Scholar
- 14.André H, Bourdon A, Rémond D (2011) On the use of the instantaneous angular speed measurement in non-stationary mechanism monitoring. In: Proceedings of the ASME 2011 International Design Engineering Technical Conferences & 23rd Biennial Conference on Mechanical Vibration and Noise, IDETC/CIE 2011, Washington, DC, USAGoogle Scholar
- 15.Andre H, Khelf I, Leclere Q (2017) Wind turbine bearing fault detected with IAS combined with harmonic product spectrum. In: COMADEM, University of Central Lancashire, UKGoogle Scholar
- 16.Andre H, Khelf I, Leclere Q (2017) Harmonic product spectrum revisited and adapted for rotating machine monitoring based on IAS, Surveillance 9, Fs, MaroccoGoogle Scholar
- 18.Gomez J, Khelf I, Bourdon A, Andre H, Alloin L, Rmond D (2017) Analysis of IAS for condition monitoring by means of a simplified wind turbine dynamic model, WCCM, London, UKGoogle Scholar