Session: 03-01-01: General Topics of Aerospace Materials

Paper Number: 109301

109301 - Theory-Guided Machine Learning for Analysis of Multi-Physics Problems and Its Recent Applications to Aerospace Composites 

Multi-physics problems, which involve the analysis of complex interactions across various time and length scales, are a common occurrence in science and engineering. These problems present significant challenges as well as opportunities for research and innovation. One such problem is the fabrication, testing, and certification of aerospace composite materials, such as carbon fiber reinforced polymers used in the production of Boeing 787 and Airbus A350 components. To address process uncertainties and material variabilities, current methodologies for testing and numerical simulations often rely heavily on expert knowledge and experience and may involve trial-and-error approaches. To streamline this process and mitigate risks related to testing errors, human errors, and simulation errors, advanced analytics and machine learning techniques can be integrated with traditional, theory-based methods. In recent years, various data-driven methodologies have been developed to address these complexities, including theory-guided machine learning (TGML), physics-informed ML, and/or scientific AI. TGML techniques have the advantage of creating physically-consistent surrogate models with small datasets and can successfully extrapolate and predict the behavior of complex systems outside of their training zones. This presentation will discuss TGML techniques and their advantages over theory-agnostic machine learning methods, and will review some recent applications at the University of Washington, including the acceleration of testing and simulation campaigns for processing and failure analysis of advanced composites and the development of real-time and autonomous tools for assisting with the certification of aerospace structures.

Presenting Author: Navid Zobeiry University of Washington

Presenting Author Biography: Navid Zobeiry is an assistant professor of Materials Science & Engineering at the University of Washington. His research focuses on the intersection of materials science, data science, and manufacturing technologies. He has worked with aerospace and automotive manufacturers and materials suppliers on a variety of topics, including material and process characterization, process simulation and optimization, damage characterization and failure analysis of composites, and theory-guided machine learning. Using machine learning approaches, Zobeiry has developed methods for accelerated testing and analysis of materials, smart and self-optimizing manufacturing techniques, and accelerated qualification and certification of aerospace composites.

Authors: