Decision support system for predictive maintenance of wind farms

The member states of the European Commission are committed to reducing primary energy consumption by 20% by 2020. In order to achieve this objective, it has become essential to develop and integrate renewable energy sources (RES), in particular wind power, into conventional production networks, while ensuring that two objectives are met:

• increase the output of RES, particularly wind farms, by improving their availability and reliability;

• cut production costs by reducing maintenance costs and limiting the consequences of faults affecting the normal operation of SER components.

In order to achieve these two objectives, it is essential to equip wind turbines with effective monitoring tools that can detect faults early and reliably, and estimate their criticality and remaining service life, so that adjustments or repairs can be carried out as quickly and cost-effectively as possible.

Wind turbine fault diagnosis is a very difficult task, not least because of the high variability of wind speed and turbulence around the rotor plane, the non-linearity of wind turbine dynamics, the appearance of certain faults (e.g., blade pitch angle pivoting actuator faults) in operating conditions (power optimization region) where the consequences of these faults are hidden, control actions that compensate for the effects of faults, and the low volume of data describing faults compared with normal operating data.

There are numerous methods for diagnosing wind turbine faults in the literature. These methods are based on the use of a model characterizing the modes or behaviors of normal and/or faulty operation. In general, these methods can be classified into two main categories: analytical model-based methods and signal processing/artificial intelligence-based methods. In the first class, a mathematical or analytical model (quantitative and/or qualitative) is built using a priori knowledge of the system's dynamics and/or structure. In the second class, the model is built by learning using a set of data on the system's behavior. Approaches belonging to these two categories have their advantages and disadvantages, depending on the available knowledge of the system's behavior, its complexity, the mechanism by which faults appear and their development dynamics.

In this article, the general principle of wind turbine fault diagnosis methods will be presented. Next, the best-known wind turbine fault diagnosis methods in the literature will be studied and compared using several examples. The aim is to demonstrate their ability to meet the challenges of developing and implementing a predictive maintenance support system for wind farms.