Publication Date

2008

Document Type

Dissertation

Committee Members

Charles Cross (Committee Member), Ramana Grandhi (Advisor), Ravi Penmetsa (Committee Member), Mo-how Shen (Committee Member), Joseph Slater (Committee Member)

Degree Name

Doctor of Philosophy (PhD)

Abstract

Design of structural components is constrained by both iteration time and prediction uncertainty. Iteration time refers to the computation time each simulation requires and controls how much design space can be explored given a fixed period. A comprehensive search of the space leads to more optimum designs. Prediction uncertainty refers to both irreducible uncertainties, such as those caused by material scatter, and reducible uncertainty, such as physics-based model error. In the presence of uncertainty, conservative safety factors and design margins are used to ensure reliability, but these negatively impact component weight and design life. This research investigates three areas to improve both iteration time and prediction uncertainty for turbomachinery design. The first develops an error-quantified reduced-order model that predicts the effect of geometric deviations on airfoil forced response. This error-quantified approximation shows significant improvements in accuracy compared to existing methods because of its bias correction and description of random error. The second research area develops a Probabilistic Gradient Kriging approach to efficiently model the uncertainty in predicted failure probability caused by small sample statistics. It is shown that the Probabilistic Gradient Kriging approach is significantly more accurate, given a fixed number of training points, compared to conventional Kriging and polynomial regression approaches. It is found that statistical uncertainty from small sample sizes leads to orders of magnitude variation in predicted failure probabilities. The third research area develops non-nominal and nominal mode Component Mode Synthesis methods for reduced-order modeling of the geometric effects on rotor mistuning. Existing reduced-order methods approximate mistuning with a nominal-mode, or design intent, basis and airfoil modal stiffness perturbation. This assumption introduces error that can be quantified when compared to a finite elment model prediction of a geometrically perturbed rotor. It is shown that the nominal-mode approach can produce significant errors, whereas the non-nominal approach accurately predicts blade-to-blade mistuned response.

Page Count

257

Department or Program

Ph.D. in Engineering

Year Degree Awarded

2008


Included in

Engineering Commons

Share

COinS