New strategy for fault detection and classification in wind turbines based on doubly-fed induction generators

A novel robust diagnostic system based on a linear fractional transform (LFT) representation combined with a static redundancy approach is proposed to design a residual generator for fault detection and localization in a wind system using the doubly fed induction generator (DFIG). As a result, faults in DFIG-based grid-connected wind systems can be grouped into three classes of faults, namely, model uncertainty-related faults (FLMU), set point disturbance-related faults (FLDS) and parameter uncertainty-related faults (FLPU). Based on the parity-space residual generations, an artificial neural network (ANN) structure has been combined with the classification to enable the assessment of hidden, indistinguishable or small amplitude faults. The training validation with two data sizes of 1278*4 and 1278*1 respectively at the inputs and outputs of the proposed ANN, presents better performance for a mean squared error value (MSE = 3.0532e⁻⁹), and a good correlation between outputs and targets for a regression value (R = 1). It emerges that the proposed robust and complete diagnostic system for the optimal and sustainable integration of wind turbines into the grid, offers very great advantages, particularly with regard to the precise and rapid detection of faults, and the assessment of hidden faults and/or ambiguous fault states in the wind system based on DFIG. In addition, the proposed approach allows the use of a reduced number of data, sensors and actuators required. Consequently, the system maintenance difficulties, complexity and cost of the diagnostic system are reduced.