Skip to main content
NASA Advanced Supercomputing (NAS) Division World-class modeling, simulation, and supercomputing in support of the NASA mission

Machine Learning Surrogates and Bayesian Optimization in Turbomachinery Structural Design

AMS Seminar Series

Speaker: Kristopher Pierson, NASA Glenn Research Center

Presentation

Seminar Slide Deck (PDF-5.6MB)

Abstract

This seminar addresses two related topics: machine learning surrogate model accuracy in turbomachinery structural design and the application of Bayesian optimization. The first section examines how choices such as surrogate model type, the number of design variables, the training sample size, and the depth of neural network architectures affect predictive accuracy, with results validated against finite element analysis. The second section introduces Bayesian optimization, including an overview of three acquisition functions and demonstrations on two example problems. Bayesian optimization is then applied to the same rotor design problem as the first section, showing that superior results can be obtained with fewer finite element simulations than direct surrogate-based optimization, highlighting its effectiveness for computationally intensive design tasks.

Biography

Photo of Kristopher Pierson
Kristopher Pierson has been an Aerospace Engineer at NASA Glenn Research Center since 2022. He specializes in CFD and FEA simulation techniques with application to aeromechanics of turbomachinery fan blades. He also works in design optimization using machine learning surrogate models and has recently begun extending his CFD expertise to include LES modelling. From 2019-2022, Kristopher worked in the tire industry where he specialized in structural dynamics of tire/wheel assemblies. Kristopher received his Ph.D. in Mechanical Engineering from the University of Akron in 2019, where his dissertation focused on tribology topics.