Bayesian optimisation improves engineering designs while preserving key system states
Bayesian Optimisation under State-Preservation Constraints
Machine Learning
Summary
Optimizing designs that rely on complex systems, like physical processes described by equations, is hard because some important parts must stay nearly the same. This paper presents a method that predicts how small changes will affect these key parts and uses that to focus the search on safe improvements. The authors show this approach works well for optimizing a component in a nuclear fusion device while keeping the plasma boundary stable.
What this means in practice
- •For engineering design teams: Optimize complex engineered systems while maintaining critical operational parameters with fewer tests.
- •For fusion reactor developers: Improve Tokamak divertor designs safely without destabilizing the plasma boundary during iterative testing.
Authors
Gabriel Diaz-Aylwin, Vignesh Gopakumar, Omkar Myatra, David Moulton, Lorenzo Zanisi, David S. Leslie, Henry B. Moss
Abstract
In many engineering design problems, the objective and constraints depend on the state: the solution of a PDE determined by the design parameters. We consider improving a design while holding selected state observables near trusted values, which we call state preservation constraints. Constrained Bayesian optimisation handles these with a learnt feasibility model, but struggles with this problem's highly anisotropic feasible set. Our central idea is to pre-compute the set of controls whose linearised constraint response stays within tolerance, thereby pulling back the state-space constraint into design space. This linearisation defines an ellipsoid from which we can efficiently draw a large number of well-spread candidates. The underlying linear response map is refined online, and the ellipsoid is rebuilt accordingly. We demonstrate the method end-to-end on our key application - Tokamak divertor optimisation under plasma-boundary preservation.