Deep material network for creating process-guided multiscale database of short fiber reinforced composites

Z. Liu
Livermore Software Technology, an ANSYS company,
United States

Keywords: multiscale computation, homogenization, physics-based building block, transfer learning, plasticity

Summary:

In the automotive industry, injection-molded short fiber reinforced composites have been identified as a key technology for reducing energy consumption in light-weight vehicle design. The molding process induces spatially varying microstructures across various length scales, while the resulting strongly anisotropic and nonlinear material properties are still challenging to be captured by conventional modeling approaches. In the paper, we present an integrated data-driven modeling framework for short fiber reinforced composites based on process simulation, material homogenization, mechanistic machine learning, and concurrent multiscale structural analysis. To prepare the linear elastic training data for our machine learning tasks, representative volume elements (RVEs) with different fiber orientations and volume fractions are generated through stochastic reconstruction and analyzed using the Ls-Dyna RVE package. More importantly, we utilized the recently proposed Deep Material Network (DMN) [1-2] to learn the hidden microscale morphologies from the data. With essential mechanics embedded in its building blocks, the data-driven material model can be extrapolated to describe nonlinear material behaviors efficiently and accurately. Through the transfer learning of DMN [3], we create a unified process-guided material database that covers a full range of geometric descriptors for short fiber reinforced compotes. Finally, this unified DMN database is implemented and linked to Ls-Dyna to enable concurrent multiscale structural simulations. From our perspective, the proposed data-driven material modeling framework is promising in many emergent multiscale engineering systems, such as metallic additive manufacturing and compressive molding. 1. Z. Liu, C.T. Wu, M. Koishi. A deep material network for multiscale topological learning and nonlinear modeling of heterogeneous materials. Computer Methods in Applied Mechanics and Engineering 345 (2019): 1138-1168. 2. Z. Liu, C.T. Wu. Exploring the 3D architectures of deep material network in data-driven multiscale mechanics. Journal of Mechanics and Physics of Solids 127 (2019), 20-46. 3. Z. Liu, C.T. Wu, M. Koishi. Transfer learning of deep material network for seamless structure-property predictions. Computational Mechanics (2019), 1-15.