Increasingly, offshore wind farms are being constructed in deeper water and at greater distances offshore, and comprise larger turbines. However, routine inspection and maintenance is more difficult for offshore wind turbines in comparison to those onshore. Further, devices larger than 1MW are than three times more likely to fail than turbines smaller than 500kW. Consequently, monitoring and diagnostics of wind turbines play an increasingly important role in their competitive operation.
Condition monitoring systems provide knowledge about the past and current condition of wind turbines, and are used to improve their performance, reliability and availability. The data generated by wind turbine monitoring systems can be used to obtain models of turbines operating under different conditions, and hence predict the measured output based on known inputs. A model-based condition monitoring system can be implemented by comparing actual output data with the predicted output for given input signals. Any differences between the output signals will be due to changes in the process, possibly due to the occurrence of faults.
A quasi-linear state dependent parameter (SDP) model structure is proposed in this paper to model a wind turbine. The parameters of the SDP model vary as functions of the state variables. SDP models have been used extensively as the basis of non-linear controllers; however, the research described in this paper represents the first occasion for which they have been employed for a model-based condition monitoring system.
In order to demonstrate the SDP model-based condition monitoring method, a simulation of a grid-connected wind turbine has been constructed using PSCAD software in the paper. A model of the relationship between wind speed and generated power is used to detect grid faults by comparing the power output data from the simulation with the predicted output for given wind speed signals. The SDP methodology is further demonstrated to detect a potential fault in the gearbox bearing of a faulty wind turbine by comparing SCADA (Supervisory Control and Data Acquisition) data from this turbine with the model prediction. An adaptive threshold for the gearbox bearing temperature has been obtained from the output predicted by the SDP model, and it is envisaged that this will form the basis of an early warning system.
It is relatively straightforward to derive SDP models off-line for deployment on-line; it is necessary only to store the current input signal and the delayed system variables, together with the arithmetic expressions required to calculate the state dependent parameters. Indeed, in comparison to artificial neural network-based models, SDP models require less computing power to implement.