Reduced-Order Models for Ranking Damage Initiation in Dual-Phase Composites Using Bayesian Neural Networks
- PDF / 893,731 Bytes
- 11 Pages / 593.972 x 792 pts Page_size
- 87 Downloads / 194 Views
https://doi.org/10.1007/s11837-020-04387-y Ó 2020 The Minerals, Metals & Materials Society
AUGMENTING PHYSICS-BASED MODELS IN ICME WITH MACHINE LEARNING AND UNCERTAINTY QUANTIFICATION
Reduced-Order Models for Ranking Damage Initiation in Dual-Phase Composites Using Bayesian Neural Networks ADITYA VENKATRAMAN,1 DAVID MONTES DE OCA ZAPIAIN,1 and SURYA R. KALIDINDI1,2 1.—George W. Woodruff School of Mechanical Engineering, Georgia Technology, Atlanta, GA 30332, USA. 2.—e-mail: [email protected]
Institute
of
The design and development of materials with increased damage resilience is often impeded by the difficulty in establishing the precise linkages, with quantified uncertainty, between the complex details of the internal structure of materials and their potential for damage initiation. We present herein a novel machine-learning-based approach for establishing reduced-order models (ROMs) that relate the microstructure of a material to its susceptibility to damage initiation. This is accomplished by combining the recently established materials knowledge system framework with toolsets such as feedforward neural networks and variational Bayesian inference. The overall approach is found to be versatile for training scalable and accurate ROMs with quantified prediction uncertainty for the propensity to damage initiation for a variety of microstructures. The approach is applicable to a large class of challenges encountered in multiscale materials design efforts.
INTRODUCTION The design and development of materials with increased resilience to damage and failure is critical to the advancement of structural applications in many advanced technologies.1,2. The rich and complex details of the material’s microstructure play an important role in controlling these properties, especially those related to damage initiation.2 Recent advances in computational mechanics and materials have offered sophisticated tools3–7 for studying the highly nonlinear dependencies between the microstructure of a material and its damage resilience. In these approaches, representative volume elements (RVEs) of the material capturing the salient features of its microstructure are generated, and suitable damage-based mechanical constitutive laws are applied to the constituent phases of the material.4,8–11 Solutions to these highly challenging problems are often only possible using numerical approaches based on the finite element (FE) (Received June 28, 2020; accepted September 14, 2020)
method. Although these RVE-based FE approaches are highly versatile, they are often computationally expensive. Most importantly, they cannot be incorporated directly into design tools used to optimize the performance of a given engineering component. A much more practical approach is to first extract a high-fidelity reduced-order model (ROM) that captures the relationships between the microstructure of a material and its damage resilience, and then use the ROM in the optimal design of the material microstructure for the selected application. Prior efforts
Data Loading...
