Source-linked AI summary
Physics guided machine learning using simplified theories
Suraj Pawar, Omer San, Burak Aksoylu, Adil Rasheed, Trond Kvamsdal
TL;DR
Data-driven models face challenges predicting unseen configurations, while high-fidelity flow simulations can be computationally expensive. This paper adds simplified-physics features to intermediate neural-network layers and reports more accurate, lower-uncertainty airfoil predictions within the panel method’s supported angle range.
Problem
Data-driven models face challenges predicting unseen configurations, while high-fidelity simulations can be computationally expensive for repeated design-space exploration.
Source: pp. 1, 3Method
PGML couples simplified-theory predictions with neural-network latent variables as intermediate features, using panel-method estimates for airfoil forces.
Source: pp. 2–3Results
PGML gives significantly more accurate predictions with reduced uncertainty for angles between -10 and +10 degrees on both NACA23012 and NACA23024 airfoils.
Source: p. 4Takeaways & Limitations
The authors propose PGML for data-scarce physical systems, including costly wind-farm or three-dimensional wing optimization settings.
Source: p. 4The study is a proof of concept starting with XFOIL; extending PGML to complex, high-dimensional flow problems remains future work.
Source: p. 5
Abstract
from arXiv · showhide
Recent applications of machine learning, in particular deep learning, motivate the need to address the generalizability of the statistical inference approaches in physical sciences. In this letter, we introduce a modular physics guided machine learning framework to improve the accuracy of such data-driven predictive engines. The chief idea in our approach is to augment the knowledge of the simplified theories with the underlying learning process. To emphasise on their physical importance, our architecture consists of adding certain features at intermediate layers rather than in the input layer. To demonstrate our approach, we select a canonical airfoil aerodynamic problem with the enhancement of the potential flow theory. We include features obtained by a panel method that can be computed efficiently for an unseen configuration in our training procedure. By addressing the generalizability concerns, our results suggest that the proposed feature enhancement approach can be effectively used in many scientific machine learning applications, especially for the systems where we can use a theoretical, empirical, or simplified model to guide the learning module.