Session: Rising Stars of Mechanical Engineering Celebration & Showcase

Paper Number: 148671

148671 - Machine Learning-Assisted Calibration of Force Fields for Atomistic Simulation of Polymer Nanocomposites 

Polymer nanocomposites (PNCs) have proved to be a pivotal technology with significant potential to enhance the prosperity, security, and global competitiveness of the United States. These advanced materials are becoming increasingly relevant to the economic performance of major industrial sectors such as aerospace, manufacturing, biomedical, and civil infrastructure. PNCs offer tunable properties, whereby alterations to their constituents, processing conditions, and microstructure enable creating products with specific, unique functions. The development of multiscale models that accurately represent the properties of PNCs and the simulation of these models under various loading conditions is crucial. Such modeling is essential for establishing precise processing-statistics-structure-property (PSSP) relations, which are foundational to the innovative use of PNCs. However, the task of developing these models is challenging due to the extensive computational effort required. This difficulty arises from the need to accurately represent the diverse compositions, phenomena, and interactions that PNCs feature across multiple scales. Our approach integrates atomistic simulations, specifically molecular dynamics (MD), to address these complexities with advanced computational techniques such as metaheuristic optimization and machine learning. This integration aims to calibrate coarse-grained force fields (CG-FFs) for polymers, which is a critical step in constructing hierarchical models. These models are adept at predicting the mesoscale mechanical properties of PNCs, which can significantly accelerate both MD and adaptive kinetic Monte Carlo simulations. In this research, we use polyvinyl chloride (PVC) reinforced with carbon nanotubes (CNTs) as a model system to demonstrate the effectiveness of our methodologies. We introduce a Δ-learning approach, incorporating a machine-learning surrogate model that refines predictions of PVC’s mechanical properties and density derived from classical all-atom and coarse-grained MD simulations. This model is then integrated with the particle swarm optimization technique to refine the CG-FF for PVC iteratively. The refined CG-FF's robustness and applicability are validated by its performance in predicting the mechanical properties in expansive configurations of PVC enriched with CNTs. Furthermore, we employ three machine-learning algorithms to predict the strain-stress behavior of PVC nanocomposites accurately. This predictive capability is vital for applying PNCs in real-world scenarios. Overall, our work not only enhances the understanding and application of PNCs in critical industrial sectors but also pushes the boundaries of what is possible with these innovative materials by integrating cutting-edge computational techniques. This research opens new avenues for the customized design of PNCs tailored for specific applications, leading to advancements in technology and materials science that will promote sustainability, multi-functionality, and economic viability.

Presenting Author: Hessam Yazdani University of Missouri

Presenting Author Biography: Hessam Yazdani, PhD, PE, is an associate professor of civil and environmental engineering and the director of the Sustainable Infrastructure, Geotechnics and Materials (SIGMa) Research Lab at the University of Missouri. Dr. Yazdani specializes in geotechnical engineering, experimental and computational multiscale mechanics of materials (particularly polymer nanocomposites), reliability analysis, machine learning, and optimization. His multidisciplinary research centers on fostering sustainability and resilience in civil and marine infrastructure through addressing the geotechnical aspects of renewable energy systems, designing high-performance and multifunctional materials (e.g. for self-monitoring structures), incorporating risk assessment, sustainability, and optimization into the design of geotechnical and structural systems, and estimating and monitoring the performance of engineering systems using machine learning techniques. Dr. Yazdani has (co)authored over 70 publications in books, journals, conference proceedings, and reports. He has received several competitive international and national awards, including the NSF CAREER Award, the DURIP Award from the Air Force Office of Scientific Research (AFOSR), and the DFI 2013 Best Student Paper Award. He has also been recognized several times by former students as the Faculty, Mentor, and Advisor of the Year. Dr. Yazdani is a registered professional engineer in Michigan.

Authors:

Hamid Ghasemi Howard University
Hessam Yazdani University of Missouri