Horizontal axis wind turbines (HAWTs) experience yaw misalignments due to the physical limitations of yaw controllers and various novel active yaw controls. Moreover, the motion of floating offshore wind turbines (FOWTs) accelerates yaw misalignment. The blade element momentum (BEM) method is widely used due to its computational efficiency for the design of HAWTs. Momentum theory, the basis of BEM, assumes steady flow and uniform induction field at the disc. Those assumptions are relaxed by engineering models to capture yaw and unsteady effects. Current yaw engineering models, however, are inaccurate since they do not capture the asymmetric wake expansion effect. Dynamic inflow models have been developed for non-yawed flow. Furthermore, the AVATAR project shows that BEM using fully coupled engineering models, the current yaw, dynamic inflow and various engineering models, suffers from significant deficiencies. This purpose of this paper, therefore, is to investigate dynamic effects for yawed flow, and determine if current dynamic inflow models are applicable in yawed conditions. The Glauert’s modified momentum theory is applied to dynamic inflow models to couple the two models. Among all coupled models, Øye, Yu PWVM and Yu FWVM DIM can capture asymmetric trends. However, the results show the significant deficiencies in phase delay on the actuator disc.

This paper presents an efficient strategy for the Bayesian calibration of parameters of aerodynamic wind turbine models. The strategy relies on constructing a surrogate model (based on adaptive polynomial chaos expansions), which is used to perform both parameter selection using global sensitivity analysis and parameter calibration with Bayesian inference. The effectiveness of this approach is shown in two test cases: calibration of airfoil polars based on the measurements from the DANAERO MW experiments and calibration of five yaw model parameters based on measurements on the New MEXICO turbine in yawed conditions. In both cases, the calibrated models yield results much closer to the measurement data, and in addition they are equipped with an estimate of the uncertainty in the predictions.