Fault Detection and Isolation of Wind Turbines - A Real Field Data Approach

Abstract

The future of wind energy passes through the installation of o shore wind farms. In such locations a non-planned maintenance is highly costly, therefore, a fault-tolerant control system that is able to maintain the wind turbine connected after the occurrence of certain faults can avoid major economic losses. The purpose of this Master's thesis is to design a Fault Detection and Isolation (FDI) system, which is responsible of detecting the wind turbine faults and identify their origin. In this sense, a robust fault detection based on system identi fication and adaptive threshold generation is proposed. Real fi eld data is used to identify the nominal model that produces the estimated output for residual generation. To avoid long term deviations, this estimated output is computed from the nominal model and an observer that follows the so-called Luenberger scheme. Moreover, an adaptive threshold based on model error modelling that take into account the nominal model uncertainty i.e. makes the FDI system robust is presented. Since the wind turbine is a highly non-linear system with a complex operating range, all these techniques are extrapolated to the entire wind turbine range using Linear Parameter Varying (LPV) models. Finally, an analysis based on residual sensitivity is developed with the aim of making the FDI system able to isolate the faults.