A surrogate model-based method to obtain optimal design in spiral casing of Francis turbine

Abstract

Numerical simulations of fluid flow through various spiral casings like accelerated, free vortex and decelerated type with different aspect ratios (AR) are carried out to construct surrogates. These surrogates are utilised for analysing design sensitivity of spiral casing to obtain its optimal design. Responses like spiral velocity coefficient, total pressure loss and average radial velocities obtained from numerical computations are used for surrogates' construction. Different surrogate models considered are Kriging, polynomial response surface, support vector regression and weighted average surrogate. Surrogates are validated using average error analysis for the selection of best surrogate. Weighted average surrogate performs well in most of the cases among all responses. Near optimal solutions obtained from the best surrogates are proposed.