Physics-Constrained Learning determines thermoacoustic instability boundaries with reduced evaluations

Summary

The paper introduces the physics-constrained learning (CPL) method, which efficiently identifies thermoacoustic instability boundaries with very few evaluations and outperforms traditional regression approaches.

Why it matters

Thermoacoustic instability prediction using high-fidelity simulations is computationally intensive, limiting the exploration of the design space. CPL offers a cost-effective pre-screening tool that maintains physical consistency, significantly reducing the number of expensive simulations required. This can accelerate combustion design cycles and enable more comprehensive stability assessments.

Critical take

The study validates CPL only on a generic reduced-order model; its performance on high-fidelity LES data from real-world scenarios has not yet been tested.