Scientific Machine Learning in Industrial and Manufacturing Pipelines

Marta D’Elia, Pasteur Labs

Abstract:

Scientific machine learning (SciML) has shown great promise in the context of accelerating classical physics solvers and discovering new governing laws for complex physical systems. However, while the SciML activity in foundational research is growing exponentially, it lags in real-world utility, including the reliable and scalable integration into industrial and manufacturing pipelines. SciML algorithms need to advance in maturity and validation, which in the context of traditional and advanced manufacturing settings, requires operating in cyber- physical environments marked by large-scale, three-dimensional, streaming data that is confounded with noise, sparsity, irregularities and other complexities that are common with machines and sensors interacting with the real, physical world.  In this talk, I will highlight some of the current challenges in applying SciML in industrial contexts. By using a practical example, the heat exchanger simulation and design, I will discuss why these are necessary bottlenecks to break through and describe possible strategies.

Special attention will be on the generation of fast and flexible surrogates for heat exchange problems; I will present a comparison of current SciML methods and their improved variants with emphasis on the graph-based Neural Operators we have been building. The talk will conclude describing how to use these operators in the context of industrial design of a heat exchanger and, more in general, manufacturing pipelines, addressing the earlier bottlenecks and alluding to outsized opportunities for advancing manufacturing tools and processes.