Advanced Computational Tools for the Analysis and Design of Complex Metamaterials and Structures
Abstract
The traditional design approach has predominantly relied on trial and error. However, this intuitive approach is not only expensive but also time-consuming, thereby significantly limiting the design process for technological applications. A more effective approach is to adopt a systematic design procedure, for which computational tools have proven to be effective. These tools must not only accurately represent faithfully the physics of the problem but also efficiently discretize and solve (potentially coupled) partial differential equations within an iterative optimization process. Artificial intelligence also holds immense potential for design. However, despite significant advancements in language processing, voice/image recognition, and signal processing, the application of machine learning (ML) in computational design remains in its nascent stages. Therefore, harnessing innovations in ML- based models and integrating them with precise physics-based modeling is the key to novel groundbreaking design methodologies.
This presentation will discuss current developments on novel physics-based and machine learning-driven models for the analysis and design of metamaterials and structures. We will first discuss advancements in enriched finite element methods (e-FEMs), which augment standard finite element analysis by incorporating in- formation about discontinuities by means of enrichment functions. These sophisticated methods completely decouple finite element meshes from the geometry of the problem, facilitating fully automated and accurate analysis. Over the years, we have developed e-FEMs to address a wide range of intricate problems in solid mechanics, including the modeling of weak discontinuities (e.g., perfectly-bonded material interfaces) and strong discontinuities (fracture). Additionally, we have extended their application to immersed boundary (or fictitious domain) problems and to tie numerical interfaces (mesh coupling) and to solve highly-nonlinear contact problems. Beyond analysis, e-FEMs have proven valuable in computational design through the application of enriched topology optimization (e-TO). We demonstrate the application of e-TO in mitigating the likelihood of fracture in brittle solids designing chocolate edible unit cells with extreme fracture anisotropy and maximizing band gaps in phononic crystals and photonic devices. Finally, we present some developments in the domain of machine learning. Specifically, we discuss recent advancements in neural reparameterization, wherein we employ a neural network as a generative design technique completely alternative to topology optimization.
Short bio
Dr. Aragon is currently Associate Professor in Computational Design and Mechanics in the Mechanical Engineering faculty of Delft University of Technol- ogy. He holds a BSc degree from Argentina, and both an MSc (as a Fulbright Scholar) and PhD degrees from the University of Illinois at Urbana-Champaign (UIUC), USA. Dr. Aragon held postdoctoral roles at UIUC and at EPFL.
He has more than 15 years of experience developing pioneering computational techniques for analysis and design in solid mechanics. His research focuses on developing enriched finite element and topology optimization technology to address design challenges across biomimetic and composite mate- rials, acoustic/elastic metamaterials, photonic and phononic crystals, edible metamaterials, and origami. He holds two patents in noise attenuation applications using acoustic/elastic metamaterials and phononic crystals. He is the lead author of the book “Fundamentals of Enriched Finite Element Methods” (Elsevier, 2023). Finally, he is director of the Machine Intelligence Advances for Materials (MACHINA) lab, which integrates machine learning into computational design.
Sandwiches are offered at the end of the seminar.
The traditional design approach has predominantly relied on trial and error. However, this intuitive approach is not only expensive but also time-consuming, thereby significantly limiting the design process for technological applications. A more effective approach is to adopt a systematic design procedure, for which computational tools have proven to be effective. These tools must not only accurately represent faithfully the physics of the problem but also efficiently discretize and solve (potentially coupled) partial differential equations within an iterative optimization process. Artificial intelligence also holds immense potential for design. However, despite significant advancements in language processing, voice/image recognition, and signal processing, the application of machine learning (ML) in computational design remains in its nascent stages. Therefore, harnessing innovations in ML- based models and integrating them with precise physics-based modeling is the key to novel groundbreaking design methodologies.
This presentation will discuss current developments on novel physics-based and machine learning-driven models for the analysis and design of metamaterials and structures. We will first discuss advancements in enriched finite element methods (e-FEMs), which augment standard finite element analysis by incorporating in- formation about discontinuities by means of enrichment functions. These sophisticated methods completely decouple finite element meshes from the geometry of the problem, facilitating fully automated and accurate analysis. Over the years, we have developed e-FEMs to address a wide range of intricate problems in solid mechanics, including the modeling of weak discontinuities (e.g., perfectly-bonded material interfaces) and strong discontinuities (fracture). Additionally, we have extended their application to immersed boundary (or fictitious domain) problems and to tie numerical interfaces (mesh coupling) and to solve highly-nonlinear contact problems. Beyond analysis, e-FEMs have proven valuable in computational design through the application of enriched topology optimization (e-TO). We demonstrate the application of e-TO in mitigating the likelihood of fracture in brittle solids designing chocolate edible unit cells with extreme fracture anisotropy and maximizing band gaps in phononic crystals and photonic devices. Finally, we present some developments in the domain of machine learning. Specifically, we discuss recent advancements in neural reparameterization, wherein we employ a neural network as a generative design technique completely alternative to topology optimization.
Short bio
Dr. Aragon is currently Associate Professor in Computational Design and Mechanics in the Mechanical Engineering faculty of Delft University of Technol- ogy. He holds a BSc degree from Argentina, and both an MSc (as a Fulbright Scholar) and PhD degrees from the University of Illinois at Urbana-Champaign (UIUC), USA. Dr. Aragon held postdoctoral roles at UIUC and at EPFL.
He has more than 15 years of experience developing pioneering computational techniques for analysis and design in solid mechanics. His research focuses on developing enriched finite element and topology optimization technology to address design challenges across biomimetic and composite mate- rials, acoustic/elastic metamaterials, photonic and phononic crystals, edible metamaterials, and origami. He holds two patents in noise attenuation applications using acoustic/elastic metamaterials and phononic crystals. He is the lead author of the book “Fundamentals of Enriched Finite Element Methods” (Elsevier, 2023). Finally, he is director of the Machine Intelligence Advances for Materials (MACHINA) lab, which integrates machine learning into computational design.
Sandwiches are offered at the end of the seminar.
Practical information
- Informed public
- Free
Organizer
- Prof. Olga Fink (IMOS), Prof. Alexandre Alahi (VITA), Prof. Dusan Licina (HOBEL), Prof. Alain Nussbaumer (RESSLab)
Contact
- Jean-François Molinari