www.imm.dtu.dk Also a special thank to M.Sc.Eng. Bo J. Pedersen for numerous fruitful discussions on fatigue loads as well as general aspects of the analysis of dynamic systems. Finally, a very special thank to my supervisor, Assoc. Prof. Niels Kjølstad Poulsen, IMM, DTU, for providing qualified and dedicated guidance on any matter during the project period.

No abstract

Active fault diagnosis (AFD) of parametric faults is considered in connection with closed loop feedback systems. AFD involves auxiliary signals applied on the closed loop system. A fault signature matrix is introduced in connection with AFD and it is shown that if a limited number of faults can occur in the system, a fault separation in the fault signature matrix can be obtained. Then the single elements in the matrix only depend of a reduced number of parametric faults. This can directly be applied for fault isolation. If it is not possible to obtain this separation, it is shown how the fault signature matrix can be applied for a dynamical fault isolation, i.e. fault isolation based on the dynamic characteristic of the fault signature matrix as function of the different parametric faults.

Abstract-In this work the problem of robust model predictive control (robust MPC) of a wind turbine in the full load region is considered. A minimax robust MPC approach is used to tackle the problem. Nonlinear dynamics of the wind turbine are derived by combining blade element momentum (BEM) theory and first principle modeling of the turbine flexible structure. Thereafter the nonlinear model is linearized using Taylor series expansion around system operating points. Operating points are determined by effective wind speed and an extended Kalman filter (EKF) is employed to estimate this. In addition, a new sensor is introduced in the EKF to give faster estimations. Wind speed estimation error is used to assess uncertainties in the linearized model. Significant uncertainties are considered to be in the gain of the system (B matrix of the state space model). Therefore this special structure of the uncertain system is employed and a norm-bounded uncertainty model is used to formulate a minimax model predictive control. The resulting optimization problem is simplified by semidefinite relaxation and the controller obtained is applied on a full complexity, high fidelity wind turbine model. Finally simulation results are presented. First a comparison between PI and robust MPC is given. Afterwards simulations are done for a realization of turbulent wind with uniform profile based on the IEC standard.

No abstract