Accelerated ML-driven Multiscale Prediction of Nonlinear Mechanical Response in Composites
Abstract
Approved for Public Release.#Naval aviation weapon platforms and systems often require uniquely advanced, lightweight, and high-performance composites. In the ongoing effort to develop and design such high-performing composites, it is essential to thoroughly understand the nonlinear mechanical behavior of these materials at the microscale. These nonlinear microstructural responses are critical as they dictate the initiation and propagation of damage, ultimately affecting composite performance. Moreover, advancements in composite manufacturing, such as automated fiber placement, have expanded the design space, offering engineers a variety of complex composite layups and patterns. However, current engineering tools often fail to efficiently predict the intricate relationships between mi-crostructure, properties, and performance, compromising the design process. To fully capitalize on advancements in composite science for Naval aviation, it is essential to have a robust understanding ofthe nonlinear relationships between a composite#s microstructure and its performance.This proposal will create an accelerated data- and mechanics-based method that predicts non-linear mechanical responses at various scales of composites and elucidates relationships between microstructure-property-performance. The crux of the proposed work is to create and leverage ML-driven mechanics methods that will enable massively efficient models of full-scalecomposite structures. These models will include the effects of microstructural details, such as defects, under arbitrary applied loading scenarios. The innovation stems from integrating appropriate ML meth-ods with mechanics principles and geometrical interpretation to enhance learning efficiency and significantly reduce the required training data. The project leverages the computational power of GPUs for data-driven simulations of complex composites. The 3-D data-driven models will tackle the complexities associated withsimulating fiber weaves, non-standard fiber placement patterns and lay-ups, and ply drop-offs, thereby facilitating advanced composite design.The scale-up will be successful because preliminary results have shown that the training data can be reduced by at least 50% compared to state-of-the-art methods. By incorporating realistic microstructural variability, material properties, and initial defects into the modeling process, these novel ML tools will reduce uncertainty in composite response predictions. The developed 3-D ML tools will drastically accelerate composite modeling and, therefore, enable the creation of highly tailored composite layups. This enhancement will facilitate the design, manufacturing, and analysis of printed composites with complex patterns. The success of this project will illuminate the response of advanced composites in naval applications and support the design of composites that meet both the current and future demands of Navy weapon platforms and systems.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Apr 10, 2025
- Source ID
- N000142512233
Entities
People
- Maryam Shakiba
Organizations
- Office of Naval Research
- Regents of the University of Colorado
- United States Navy