Time-Accurate Predictions for a Fully Cooled High-Pressure Turbine Stage—Part II: Methodology for Quantifications of Prediction Quality

Abstract

The aerodynamics of a fully cooled, axial, single stage high-pressure turbine operating at design corrected conditions of corrected speed, flow function, and stage pressure ratio has been investigated experimentally and computationally and presented in Part I of this paper. In that portion of the paper, flow-field predictions obtained using the computational fluid dynamics codes Numeca’s FINE/TURBO and the code TURBO were obtained using different design methodologies that approximated the fully-cooled turbine stage in different ways. These predictions were compared to measurements obtained using the Ohio State University Gas Turbine Laboratory Turbine Test Facility, in a process that was essentially a design methodology validation study, instead of a computational methodology optimization study. The difference between the two is that the designers were given one chance to use their codes (as a designer would normally do) instead of using the existing data to fine-tune their grids/methodologies by doing grid studies and changes in the turbulence models employed. Part I of this paper showed differing results from the two solvers, which appeared to be mainly dependent on the differences in grid resolution and/or modeling features selected by the code users. Examining these occurrences points to places where the design methodology could be improved, but it became clear that metrics were needed to compare overall performance of each approach. In this part of the paper, three criteria are proposed for measuring overall prediction quality of the unsteady predictions, which include the unsteady envelope size, envelope shape, and power spectrum. These measures capture the main characteristics of the unsteady data and allow designers to use the criteria of most interest to them. In addition, these can be used to track how well predictions improve over time as grid resolutions and modeling techniques change.