Detection en temps reel de sous-performance energetique d'eoliennes

The project was conducted as part of a collaboration between BBA, Ecole de technologie superieure, the Technocentre Eolien (TCE) and Osisoft Canada. TCE has provided data for a period of one year for two turbines Repower MM92 on the site nordique experimental en eolien CORUS (SNEEC) wind farm. The objective is to develop and integrate tools for performance monitoring of wind turbines a branch of condition monitoring systems. The tools developed are Caselitz method, control charts and central moments method. These Tools are developed in the Matlab environment. Caselitz tools and control charts are also integrated into the OSIsoft PI system of TCE. They are developed from two predicting models of the power based on "bins" method and neural network. The models are trained with data spanning a period of 143 days using a 10 min average running on 2 min and a 1 min average.;Caselitz method was developed for "bins" models. Control charts have been developed with performance ratio and residue indicators to have a unit of measure of performance independent of the speed of operation. The limits are calculated with the standard deviation divided by speed "bins" of 0.5 m/s for data operation of good operation. Central moments method was determined by trial and error. The method of central moments with the residue indicator is one that fits best with the Sgurrtrend example.;The daily indicators: duration (%), losses (% and kWh) and distance (%) for central moments method are developed for each method. A correlation analysis determines the most effective methods. For the root cause analysis, 29 days with the most losses for one of the two wind turbines were studied. In order to achieve this, we used a graphical analysis of frost indicators as well as status codes from the wind turbine. The results are described in a summary table.