Information entropy: An effective approach for wind turbine condition monitoring

Increasing deployment of large wind turbines offshore and in remote areas urgently requires reliable and efficient condition monitoring techniques to achieve high turbine availability and the expected economic return. Much effort has been made in the past a few years to develop advanced signal processing technologies to improve wind turbine condition monitoring signal interpretation capability. However, despite these achievements a reliable condition monitoring of wind turbines has not yet been reached in the practice. The reasons are complex, but major reasons are: (1) Advanced signal processing techniques involve complex calculations unsuited for online use; and (2) These techniques improve signal processing accuracy but are still load-dependent as a consequence their condition monitoring results are still affected by external turbine loads. This raises difficulties in making the right judgement on the turbine health simply by observing fault-related frequency amplitudes. In view of these issues, a new, online, load-independent, condition monitoring technique is developed in this paper based on the concept of information entropy. Following investigation of the responses of information entropy to different fault scenarios, an entropy-based condition monitoring strategy has been developed. The proposed technique is verified experimentally by applying it to interpreting vibration and electrical signals from a specially designed wind turbine condition monitoring test rig, and then to analysing vibration signals from a real 750kW wind turbine gearbox. Experiments show that the proposed technique is efficient and reliable for detecting wind turbine mechanical and electrical faults.