Automated Model Discovery for Soft Matter

Stanford University, Stanford CA

Investigators

Abstract

In solid mechanics, a constitutive relation describes a material’s response to external stimuli, such as forces. Constitutive modeling and parameter identification are the cornerstones of the mechanics of materials and structures. The current gold standard in constitutive modeling is to first select a model and then fit its parameters to data. However, the scientific criteria for model selection are poorly understood and depend largely on user experience and personal preference. This award seeks to democratize constitutive modeling through automated model discovery and make it accessible to a more inclusive and diverse community. The main deliverable is an open source discovery platform that will discover the best model and parameters, entirely without human interaction. This open source platform will feature a new family of neural networks, data, models, and parameters. It will be freely available to a wide range of users, regardless of their institutional or financial resources. As such, automated model discovery will lower the barrier of entry into the STEM fields and foster a more inclusive and diverse scientific community. This project has broad scientific, social, and economic impacts. It will democratize constitutive modeling, stimulate discovery in the mechanics of materials and structures, establish machine learning tools to characterize, create, and functionalize soft matter, and train the next generation of civil, mechanical, and manufacturing innovators to use these new technologies. The goal of this research is to establish neural networks that autonomously discover models for soft matter systems. Instead of using classical neural networks that provide no insight into the underlying physics, this project designs its own constitutive neural networks. To train, test, and validate these networks, this project will generate an open source library with benchmark data from dozens of living and engineered materials. All networks, data, models, and parameters of this project will be freely available to promote engineering education and advance scientific knowledge. This project has the potential to induce a paradigm shift in constitutive modeling, from user-defined model selection to automated model discovery. This could forever change how we simulate materials and structures. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.