PSI - Issue 52

Available online at www.sciencedirect.com Available online at www.sciencedirect.com ScienceDirect Structural Integrity Procedia 00 (2022) 000 – 000 Available online at www.sciencedirect.com ScienceDirect Structural Integrity Procedia 00 (2022) 000 – 000

www.elsevier.com/locate/procedia www.elsevier.com/locate/procedia

ScienceDirect

Procedia Structural Integrity 52 (2024) 391–400

© 2023 The Authors. Published by Elsevier B.V. This is an open access article under the CC BY-NC-ND license (https://creativecommons.org/licenses/by-nc-nd/4.0) Peer-review under responsibility of Professor Ferri Aliabadi Abstract The work describes a numerical procedure to study biaxial compression failure in brittle foams. The foam is modelled as 3D interpenetrating structure of spherical pores with pore size distribution identical to a real ceramic foam studied by micro-computed tomography scans. Multiple realizations of artificial foam microstructures are generated through a novel microstructure reconstruction algorithm for a given value of porosity and pore size distribution. A boundary value problem is solved for each microstructure realization to determine effective compression stress-strain behavior through finite element simulations. This work focusses on biaxial compression failure in which a strain-based loading is applied along two orthogonal directions on a 3D model. For different loading conditions, damage initiation and propagation mechanisms are studied in multiple microstructure realizations. The study of number of damaged regions as the loading is increased reveals interesting mechanisms of macroscopic fracture. Lastly, a neural network based surrogate model is developed that predicts the effective compression stress-strain behavior for different loading conditions. It is shown that the neural network-based prediction for a given size of foam volume element is improved if it is informed by a neural network of a smaller sized volume element. Since the computational expense for generating data on a smaller sized volume element is much lesser, the overall accuracy of the model is considerably improved at minimum increase in computational expense. © 2023 The Authors. Published by ELSEVIER B.V. This is an open access article under the CC BY-NC-ND license (https://creativecommons.org/licenses/by-nc-nd/4.0) Peer-review under responsibility of Professor Ferri Aliabadi Keywords: Biaxial compression failure; microstructure reconstruction; neural networks; foam modeling Fracture, Damage and Structural Health Monitoring Biaxial compression failure of brittle foams: A transfer learning based strategy Vinit Vijay Deshpande a , Romana Piat a * a Department of Mathematics and Natural Sciences, University of Applied Sciences,Schoefferstrasse 3, Darmstadt 64295, Germany Abstract The work describes a numerical procedure to study biaxial compression failure in brittle foams. The foam is modelled as 3D interpenetrating structure of spherical pores with pore size distribution identical to a real ceramic foam studied by micro-computed tomography scans. Multiple realizations of artificial foam microstructures are generated through a novel microstructure reconstruction algorithm for a given value of porosity and pore size distribution. A boundary value problem is solved for each microstructure realization to determine effective compression stress-strain behavior through finite element simulations. This work focusses on biaxial compression failure in which a strain-based loading is applied along two orthogonal directions on a 3D model. For different loading conditions, damage initiation and propagation mechanisms are studied in multiple microstructure realizations. The study of number of damaged regions as the loading is increased reveals interesting mechanisms of macroscopic fracture. Lastly, a neural network based surrogate model is developed that predicts the effective compression stress-strain behavior for different loading conditions. It is shown that the neural network-based prediction for a given size of foam volume element is improved if it is informed by a neural network of a smaller sized volume element. Since the computational expense for generating data on a smaller sized volume element is much lesser, the overall accuracy of the model is considerably improved at minimum increase in computational expense. © 2023 The Authors. Published by ELSEVIER B.V. This is an open access article under the CC BY-NC-ND license (https://creativecommons.org/licenses/by-nc-nd/4.0) Peer-review under responsibility of Professor Ferri Aliabadi Keywords: Biaxial compression failure; microstructure reconstruction; neural networks; foam modeling Fracture, Damage and Structural Health Monitoring Biaxial compression failure of brittle foams: A transfer learning based strategy Vinit Vijay Deshpande a , Romana Piat a * a Department of Mathematics and Natural Sciences, University of Applied Sciences,Schoefferstrasse 3, Darmstadt 64295, Germany

* Corresponding author. Tel.: +49-6151-16-30030; fax: +49-6151-16-8975. E-mail address: romana.piat@h-da.de * Corresponding author. Tel.: +49-6151-16-30030; fax: +49-6151-16-8975. E-mail address: romana.piat@h-da.de

2452-3216 © 2023 The Authors. Published by ELSEVIER B.V. This is an open access article under the CC BY-NC-ND license (https://creativecommons.org/licenses/by-nc-nd/4.0) Peer-review under responsibility of Professor Ferri Aliabadi 2452-3216 © 2023 The Authors. Published by ELSEVIER B.V. This is an open access article under the CC BY-NC-ND license (https://creativecommons.org/licenses/by-nc-nd/4.0) Peer-review under responsibility of Professor Ferri Aliabadi

2452-3216 © 2023 The Authors. Published by ELSEVIER B.V. This is an open access article under the CC BY-NC-ND license (https://creativecommons.org/licenses/by-nc-nd/4.0) Peer-review under responsibility of Professor Ferri Aliabadi 10.1016/j.prostr.2023.12.039