Unsteady CFD analysis and shape optimization of a hydraulic turbine

Authors

Varvara G. Asouti  - National Technical University of Athens (NTUA), School of Mechanical Engineering, Parallel CFD & Optimization Unit, 15772 Athens, Greece
Xenofon S. Trompoukis - National Technical University of Athens (NTUA), School of Mechanical Engineering, Parallel CFD & Optimization Unit, 15772 Athens, Greece
Kyriakos C. Giannakoglou - National Technical University of Athens (NTUA), School of Mechanical Engineering, Parallel CFD & Optimization Unit, 15772 Athens, Greece
Evgenia A. Kontoleontos - Andritz Hydro GmbH, 4030 Linz, Austria

Abstract

The increased energy demands and grid regulation services led the hydropower suppliers to extend the machines operating range, pushing hydraulic turbines to operate in a far more dynamic way, at flow conditions not experienced in the past. Αt these flow conditions, dynamic phenomena such as pressure pulsations regularly appear; the latter are decisive for their smooth and safe operation. Thus, the importance of fluctuating loads increases and the unsteady operational behaviour receives more attention. This paper is concerned with the CFD-based analysis and shape optimization of a hydraulic turbine runner at rated head, where pressure pulsations between the runner and the guide vanes need to be minimized. Next to this target, constraints for keeping the desirable flow rate and avoiding cavitation are imposed. The unsteady flow analysis is carried out using the in-house GPU-accelerated flow solver PUMA, [1], which solves the Unsteady Reynolds Averaged Navier Stokes (URANS) equations in the time domain coupled with the Spalart-Allmaras turbulence model. In PUMA, all flow quantities are computed in double precision, the left-hand-side is stored in single precision, to reduce memory requirements and increase efficiency. Interfaces between guide vanes and runner are modelled via the sliding plane technique. For the shape optimization of the turbine, the runner’s blade geometry is parameterized using a new in-house, turbomachinery-oriented, free-form deformation tool handling multi-row geometries. This tool is based on volumetric NURBS, maintains hub and shroud axisymmetry and allows the boundary mesh nodes to slide over the boundary surfaces by incorporating a volumetric model with an intermediate coordinate system transformation. For this optimization, a control lattice encapsulating the runner’s blade is created and the coordinates of (some of) its control points are the design variables; ~200 design variables in total are used. The optimization is carried out by means of an evolutionary algorithm assisted by surrogate evaluation models (metamodel-assisted evolutionary algorithm; MAEA) and the principal component analysis (PCA), [2], the EASY platform. PCA is used to tackle the so-called "curse of dimensionality", since ~200 design variables is already a high number of design variables for a stochastic population-based search method. In specific, PCA is implemented during the application of the evolution operators and/or the metamodel training, to reduce the number of input units and, thus, get more accurate predictions. EASY may be deployed in an HPC environment and use its resources for evaluating the candidate solutions during the optimization run. The optimized geometry is compared with the baseline one. This work has received funding from the European High-Performance Computing Joint Undertaking under Grant Agreement No 956560 (REGALE). [1] Asouti et al., International Journal for Numerical Methods in Fluids, 67, 232 (2011). [2] Kapsoulis et al., Applied Soft Computing, 64, 1 (2018).