Bayesian optimization for bladeless turbine computational fluid dynamics simulations

Mentors:
Lead: Annie Booth (Statistics, NCSU) Collaborator: James Braun (Mechanical Engineering, NCSU)

Intellectual merit and significance:
Turbines are essential sources of energy production, and a new class of bladeless turbines has the potential to improve energy output and efficiency for supersonic and hypersonic flow [1]. New turbine designs featuring an array of shock waves and separation regions are studied through intricate, steady, and unsteady computational fluid dynamics simulations. Such simulations are computationally costly, resulting in a limited amount of geometry evaluations through traditional evolutionary optimization algorithms. We seek to identify the optimal turbine configuration (e.g., geometry configurations, inlet characteristics, and flow properties) that maximize energy efficiency and minimize flow losses.

Objectives:
Students will deploy a strategic, sequential Bayesian optimization procedure using Gaussian process (GP) emulators [2] to identify the optimum configuration with as few evaluations of the costly computer simulation as possible.

Outcomes:
Identification of optimal input configurations, with comparisons to existing expert knowledge. Codes will be published in GitHub repositories for later expansion to more complex simulations.

References:
1. Braun J, Paniagua G, Falempin F, Le Naour B. Design and experimental assessment of bladeless turbines for axial inlet supersonic flows. J Eng Gas Turbines Power. 2020;142:041024.
2. Jones D, Schonlau M, Welch W. Efficient global optimization of expensive black-box functions. J Glob Optim. 1998;13:455-492.