Turbine blade aerodynamic performance can be accurately predicted using one-dimensional mean-line and two-dimensional through-flow solvers. These predictions have been achieved by coupling one and two-dimensional ideal flow equations with blade aerodynamic loss and flow deviation angle models. These loss and deviation models are largely generated using classical cascade testing which have limitations and constraints. These limitations are associated with testing in general and include scope, time, resources, geometric and operating parameter space, data scatter and uncertainty. The models largely ignore interaction effects and can be subjective. The loss and deviation models also do not incorporate blade features associated with modern turbine blades. The objective of this paper is to study the feasibility of conducting these experiments numerically using three-dimensional turbine blade and constructing global blade performance and loss models.
The study looks at a number of competing surrogate modeling techniques and evaluates their performance for optimum blade loss and deviation prediction. The effect of Reynolds number, performance parameter definition, operating condition specification along with the use of extended parameters are investigated to further enhance the surrogate models. The performance map generated using the optimized surrogate models is then validated using a 1.5 stage axial turbine. The results show that numerically generated surrogate models can be used to accurately predict the CFD based axial turbine performance.