E-diagnosis of physical processes based on high resolution methods, Application : Wind turbine machines

The expansion of systems using intelligent sensors has prompted the study of physical processes E- diagnosis based on high resolution methods. The automated control of modern wind machines requires proactive maintenance. We proposed several indicators measuring the performance level of a wireless protocol for routing data packets to the monitoring station. A study to design a diagnosis system entitled IESRCM for local or remote monitoring for the mentioned machines is achieved. A comparison has been realized to appreciate the performance of this system when it is integrated with GPRS or Wi-Max wireless modules. The obtained results by simulation using Proteus ISIS and OPNET software have favored the incorporation of Wi-Max module in the proposed system because its advantages over GPRS. The high resolution spectral estimation methods are effectively used for detecting electromechanical wind turbine faults. In front of the diversity of these methods, an investigation of each algorithm separately has been performed with a composite signal of stator current containing several types of defects and under different noisy environments. It was deduced in therein that the accuracy of the spectral estimation depends on the degree of the signal disturbance, the severity level of the faults, the frequency sampling and the number of data samples. The comparison with simulation in Matlab that we have made between these algorithms has proved the superiority of ESPRIT algorithm. However, this algorithm has a relatively large computing time and requires an important memory size to be executed. To overcome this problem, an improvement of ESPRIT-TLS technique has been proposed to make it applicable in real time. A new version of this method is developed in this thesis entitled Fast-ESPRIT. The proposed development is made by combining pass band recursive filtering technique IIR of Yule-Walker and decimation technique. The evaluation of the proposed technique for wind turbine fault detection of various types is performed. The analysis of the obtained results confirms that the Fast-ESPRIT algorithm provides a very satisfactory spectral accuracy in discriminating the studied faults harmonics. It resulted in a reduced complexity with an eligible ratio, a reduction of the required memory size for its implementation 5 times lower and a decrease of calculation time about 14,25 times less. This method provides better spectral resolution even in presence of a significant number of harmonics of different faults. However, this new method has some limitations because it does not recognize the type and the severity level of a detected fault. Therefore, another real time control approach has been proposed. It combines the developed Fast-ESPRIT method, the fault classification algorithm called CAFH and a fuzzy inference system interconnected with vibration sensors located on various wind turbine components. A new indicator of severity level for each studied fault type was formulated. It allows avoiding unnecessary alarms. Matlab simulation of this approach under four failure types with a noise shows that it provides a good robustness of faults classification.