Optimizing the Information Needed for Wind Turbine Health Monitoring

Abstract

Offshore wind is a promising energy resource to face the climate change and to reduce the gas emissions. Over the last decade, this sector has experienced a very rapid growth, wind turbines have become much powerful and wind parks became bigger. However, the costs of operations and maintenance (O&M) still represent an important part of the cost of energy (about 30 %). Structural health monitoring (SHM) can play a major role to reduce these costs, by detecting coming failures before they occur and predicting the failure time. On a wind turbine (WT), hundreds of sensors can be found at different location and these sensors measure, for example, temperatures, output power, rotational speed, pitch and yaw angles, etc. Their primary functions are performance optimization and load control, but they are also used for condition monitoring. Unfortunately, a high number of sensors lead to more complexity and more sensor faults. As a result, relationships, correlations and redundancies are not used enough, even sometimes not used at all, in WT SHM. Moreover, through a high number of sensors, a wind turbine may be subjected to more and more frequent short stops that are not always necessary (false alarms). The main goal of this research is to optimize the sensor network and to improve its use for SHM. In this framework, several steps have to be considered. First, a theoretical analysis needs to be conducted to find what the minimum information is needed to understand a WT behavior. The WT behavior varies with respect to the environmental and operational conditions. Indeed, loads, rotation speed, power output, internal temperatures, etc. depend on these conditions. Assessing the health of a WT needs then to be performed in accordance to the working state of the WT. In the present research, the WT working states are described and classified into six main categories ('stopped', 'starting', 'running', 'transition around rated power', 'shutdown' and 'idling'). The possible previous and following states of each state have been determined, as well as the changes expected in the main signals. It is proven that a WT working state can be determined with only six sensors.