Sponsored by: Novem
Wind turbines, especially offshore turbines are shutdown for a prolonged period of time due to failures and degradation of components under severe weather conditions. Repairing offshore wind turbines is costly. Model-based fault detection and controller reconfiguration are two new, partly still experimental techniques, that have the potential to detect faults quickly and minimize the impact of faults in a more efficient way than the currently used classical condition monitoring and control methods can. The combination of fault detection and controller reconfiguration is called fault-tolerant control. Fault tolerant control increases system availability.
A case study was performed to investigate the feasibility of model-based fault detection and controller reconfiguration of wind turbines. The case study was focused on failures of the sensors and actuators in the primary pitch-to-vane control loop and to changes in the gain of this control loop that can be viewed as being due to the occurrence of faults or degradation of components. Two major bottlenecks in the use of fault detection methods for wind turbines are: 1) The operation of a wind turbine in a closed-loop control configuration, which leads to correlation between the input and output signals of the wind turbine; 2) The fact that the closed-loop wind turbine system is driven by an unknown input signal, the rotor effective wind speed. In practice, the wind speed cannot be measured accurately, and therefore it is assumed to be unknown.
The combination of the closed-loop operation and the unknown driving input leads to a complicated setting for performing fault detection. Most currently available fault detection methods are not suited for this scenario. Within this feasibility study dedicated fault detection methods have been developed that can deal with the closed-loop setting and the unknown wind input. These methods are based on a Kalman filter and use a model of the deterministic dynamic behavior of the wind turbine and models of the faults that are to be expected (for example multiplicative actuator faults). The developed methods not only estimate the faults, but also the unknown input signal.
Next: Neuro-fuzzy modeling in model-based fault Up: Controller design Previous: The standard predictive control problem
Last modified: 24 March 2005, 10:16 UTC
||Search Site map|