Date of Award


Document Type


Degree Name

Doctor of Philosophy (PhD)

Legacy Department

Civil Engineering


Atamturktur, Sez

Committee Member

Juang , Hsein

Committee Member

Khan , Abdul

Committee Member

Ravichandran , Nadarajah

Committee Member

Wagner , John


In the field of wind energy, modeling and simulation techniques provide an efficient and economical alternative to experimentation for studying the behavior of wind turbines. Numerical models however are approximations of reality, thusly making it crucial to evaluate various sources of uncertainties that influence the model predictions. Credibility of a numerical model rests on the model's ability to replicate existing experimental data, widely known as fidelity-to-data. This dissertation advocates that fidelity-to-data, while necessary, is insufficient to claim credibility of a numerical model. Herein, the objective is to develop numerical models that not only provide agreement to experimental data, but also remain consistent (robust) as unavoidable uncertainties are considered.
The focus in this dissertation is on the development of models that are simplified yet consistent with experiments, which offer the possibility of large scale simulations for rapid prototyping and prognostics. This dissertation presents a completely integrated Verification and Validation (V&V) procedure that includes the solution and code verification, sensitivity analysis, calibration, validation, and uncertainty quantification in the development of a finite element (FE) model of the CX-100 wind turbine blade that is simplified yet consistent with experiments. This integrated V&V procedure implements a comprehensive evaluation of uncertainties, including experimental, numerical, and parametric uncertainties, to evaluate the effect of assumptions encountered in the model development process. Mesh refinement studies are performed to ensure that mesh size is chosen such that the effect of numerical uncertainty does not exceed experimental uncertainty. A main effect screening is performed to determine and eliminate the model parameters that are least sensitive to model output, reducing demands on computational resources to only calibrate parameters that significantly influence model predictions. Model calibration is performed in a two-step procedure to de-couple boundary condition effects from the material properties: first against the natural frequencies of the free-free experimental data, and second against the natural frequencies of the fixed-free experimental data. The predictive capability of the calibrated model is successfully validated by comparing model predictions against an independent dataset. Through the V&V activities, this dissertation demonstrates the development of a FE model that is simplified yet consistent with experiments to simulate the low-order vibrations of wind turbine blades.
Confidence in model predictions increases when the model has been validated against experimental evidence. However, numerical models that provide excellent fidelity to data after calibration and validation exercises may run the risk of generalizing poorly to other, non-tested settings. Such issues with generalization typically occur if the model is overly complex with many uncertain calibration parameters. As a result, small perturbations in the calibrated input parameter values may result in significant variability in model predictions. Therefore, this dissertation posits that credible model predictions should simultaneously provide fidelity-to¬-data and robustness¬-to-uncertainty. This concept that relies on the trade-off between fidelity and robustness is demonstrated in the selection of a model from among a suite of models developed with varying complexity for CX-100 wind turbine blade in a configuration with added masses. The robustness to uncertainty is evaluated through info-gap decision theory (IGDT), while the fidelity to data is determined with respect to the experimentally obtained natural frequencies of the CX-100 blade.
Finally, as fidelity and robustness are conflicting objectives, model calibration can result in multiple plausible solutions with comparable fidelity to data and robustness to uncertainty, raising concerns about non-uniqueness. This dissertation states that to mitigate such non-uniqueness concerns, self-consistency of model predictions must also be evaluated. This concept is demonstrated in the development of a one dimensional simplified beam model to replace the three dimensional finite element model of CX-100 wind turbine blade. The findings demonstrate that all three objectives, fidelity-to-data, robustness-to-uncertainty and self-consistency are conflicting objectives and thus, must be considered simultaneously. When all three objectives are considered during calibration it is observed that the fidelity optimal model remains both least robust and self-consistent, suggesting that robustness and self-consistency are necessary attributes to consider during model calibration.