Non-stationary spectral estimation for wind turbine induction generator faults detection

Development of large scale offshore wind and marine current turbine farms implies to minimize and predict maintenance operations. In direct- or indirect-drive, fixed- or variable-speed turbine generators, advanced signal processing tools are required to detect and diagnose the generator faults from vibration, acoustic, or generator current signals. The induction generator is traditionally used for wind turbines power generation. Even if induction machines are highly reliable, they are subjected to many types of faults. The aim then, is to detect them at an early stage in order to prevent breakdowns and consequently ensure the continuity of power production. In this context, this paper deals with wind turbines condition monitoring using a fault detection technique based on the generator stator current. The detection algorithm uses a recursive maximum likelihood estimator to track the time-varying fault characteristic frequency and the related energy. Furthermore, a decision-making scheme and a related criterion are proposed. The feasibility of the proposed approach has been demonstrated using simulation data issued from coupled magnetic circuits induction generator model driven by a wind turbine for both electrical asymmetry and mechanical imbalance.