Masters Thesis: Feedforward Learning Control for Individual Blade Pitch Control of Modern Two-Bladed Wind Turbines

Wind is arguably the most promising source of sustainable energy. Moreover, recent developments in wind turbine technology have achieved significant cost reductions in extracting energy from wind. The main reason for this has been the application of innovative and effective control strategies. Over time, turbine designs have focused on reducing the overall cost of energy by either maximizing energy yield or minimizing the effect of loads to increase the life of the wind turbine. Minimizing the effect of loads is the main focus in this study. A standard control strategy currently used in the industry involves changing the pitch angle for each individual blade to reduce the loads through a mechanism called Individual Blade Pitch Control (IPC). However, besides significant load reductions, this control strategy causes considerable wear and tear to the large pitch bearings of wind turbines due to high pitch control action. For that reason, a control strategy that not only reduces loads, but also provides smoother pitch control action has been investigated in this report. The loads experienced by the blades of the wind turbine are partly deterministic. They vary slowly over a fixed time period and more importantly, contribute largely to fatigue in the blades. The transient characteristics of these fatigue-causing loads have prompted the use of a feedforward Learning Controller (LC). Furthermore, the input space of the LC can be restricted to a certain class of basis functions. The basis functions are chosen to be sinusoidal signals as a representation of the fatigue-causing loads. The basis functions are then parametrized and these parameters are repetitively learned using a learning algorithm. The learning algorithm involves a technique where the parameters are calculated such that it enables the wind turbine to learn from its “mistakes” from the previous trial. Consequently, the controller adjusts its parameters after each trial during operation. The learning control strategy has been successfully applied to the IPC mechanism in this study. Various performance indicators indicating fatigue in the blades and pitch bearings have been used to compare the performance of the LC to the standard IPC approach. Results from these performance indicators show that the implemented LC indeed has smoother pitch control action compared to the standard IPC approach. Moreover, the load reduction capabilities are excellent under constant wind conditions. However, under turbulent wind conditions, the performance of the LC deteriorates which highlights some of its limitations. These limitations arise from the fact that the controller is based on a model linearized for Master of Science Thesis Rishabh Dev Sharma ii a particular set of operating conditions such as rotor speed. Hence, the performance of the controller deteriorates when the rotor speed varies under turbulent wind conditions. Moreover, it takes a few iterations for the control parameters to converge and if during this time the rotor speed changes, the parameters calculated are no longer optimal. Rishabh Dev Sharma Master of Science Thesis