Perfectly Parallel Optimization for Cutback Trailing Edges

Abstract

Previous attempts have been made to optimize the performance of film-cooling slots for cutback trailing edges, but these involved the use of steady calculation methods, which have been shown to be inappropriate for accurately capturing the behavior of this class of flows. Here, an unsteady method (large-eddy simulation on a coarse grid, or very large-eddy simulation) is used to compute the flow. To take advantage of the enormous parallel capacity of modern supercomputers and distributed computing nets, as well as the relatively low cost of very large-eddy simulation, while at the same time mitigating its lower scope for significant parallelization, a perfectly parallel evolutionary optimization process was undertaken. A relatively crude optimization target of maximizing the adiabatic wall film-cooling effectiveness averaged over the entire exposed cutback surface was used as a proof of concept. The optimizing heuristic then used an evolutionary approach to design a turbulator planform, subject to some imposed design restrictions. Six hundred large-eddy simulation type simulations were carried out over 12 generations, and the best performing designs from the last generation are examined. The optimized design showed a considerable improvement in the target metric over the previous experimental geometries. The influence of various geometric parameters on several of the metrics of film cooling is also explored by mining data from the populations generated over the course of the optimization. In a targeted optimization exercise, it is likely that these data could be used to steer the course of the evolution down favorable paths more quickly. Nomenclature= mass flow rate Q = conserved variable p = static pressure p 0 = stagnation pressure t = time U = fluid speed u, v, w = components of velocity γ = ratio of specific heats Δ = grid spacing Δx , Δy , Δz = nondimensional wall distance Δx = finite length scale η aw = adiabatic wall film-cooling effectiveness ρ = density τ R ij = residual stress tensor ϕ = convected scalar

Eddy Resolving Strategies in Turbomachinery and Peripheral Components

“…Motivated by these encouraging results and also those of Martini et al [34], for this notionally Class A (more so when the internal wake inducing pins are included), flow Watson and Tucker [35] optimized internal geometry for adiabatic cooling effectiveness. The optimization involved over 600 ZLES simulations for different internal pin layouts and was genetic algorithm based.…”

“…The initial simulations in this section use the NEAT computational fluid dynamics program. Inspired by the work of Watson and Tucker [35], Dai et al [36] perform a well-resolved (average grid spacings around 10 in wall units) LES-based optimization of labyrinth seals with the α subgrid scale model [37]. The IBM method was used to facilitate rapid geometry changes that avoided remeshing.…”

Design optimisation of labyrinth seals using LES

“…For example, Pierret [17] explored the ability of GAs to optimise a compressor blade design, which was found to perform well. Watson and Tucker [18] optimised the performance of film-cooling slots for cutback trailing edges using a GA. Six hundred LES simulations were conducted. The optimised design showed a considerable improvement over the previous experimental geometries.…”

The Current State of High-Fidelity Simulations for Main Gas Path Turbomachinery Components and Their Industrial Impact

“…It has been shown in some cases, though, that adopting coarse grids to run Very Large Eddy Simulations can be sufficiently affordable to enable genetic-based optimizations of specific turbine aerodynamic features (e.g. [147]).…”