Deep learning virtual experiments for complex materials with non-periodic, undefinable, hierarchical, tangible structures

-Overcoming the limitations of conventional materials and process informatics-
Deep learning virtual experiments for complex materials with non-periodic, undefinable, hierarchical, tangible structures
Like

In our article “Virtual Experimentations by Deep learning on Tangible Materials” in Communications Materials, we propose a deep learning computational framework that can implement virtual experiments on non-periodic, undefinable, hierarchical, tangible structures. 

Artificial intelligence constructed from the database of materials is an emerging powerful technique for accelerated discovery of new materials with targeted properties. Such artificial intelligence techniques have been used in the fields of chemical compounds, solid-state inorganic materials, and materials with periodic crystalline structures. However, artificial intelligence techniques cannot be easily applied to tangible structures, such as fabrics, yarns, metal alloys, plastic/rubber composites, fine ceramics, and multi-materials. This is because their structures are highly complex, leading to the difficulty in the expression of the structures from elements, chemical bonds, unit lattice in periodic structures.

We have focused on recent advances in deep learning techniques for acquiring the information from complex features. By introducing the state-of-art deep learning techniques to create the novel framework for complex materials, we aimed to broaden the applicability of artificial intelligence techniques to a variety of materials. 

Figure 1
Figure 1. Schematic of our concept of artificial intelligence technique. Sequence, including material, process, structure, and process, in our real world is represented by deep learning techniques. Reproduction from our work, Honda*, Muroga*, Nakajima* (*contributed equally) et al., Communications Materials, 2, 88 (2021). DOI:10.1038/s43246-021-00195-2, open access with CC BY 4.0 license. 

Our concept, shown in Figure 1, can be summarized by following four important ideas.

  1. Sequence of experiments in our real world, including materials, process, structure, property, can be computationally mimicked by deep learning techniques.
  2. Trained conditional generative adversarial network (GAN) can generated the complex structure of materials at arbitrary composition.
  3. Combining multi-scale structures generated by multiple GAN models for different scales is effective to implement structural hierarchy.
  4. Artificial neural network of combined multi-scale structures is powerful to predict performance of final processed materials at arbitrary composition and process condition in a short time even for a large number of conditions (e.g., thousands of compositions, which is impossible for real experiments).
Figure 2
Figure 2. Artificial intelligence techniques applied to tangible carbon nanotube films. a, Carbon nanotube structures generated by conditional generative adversarial networks for different scales. b, Ashby map of predicted electrical conductivity against specific surface area of binary and ternary carbon nanotube mixtures. c, Construction of wall number phase diagrams of carbon nanotube films. Colors indicates single-wall (red), double-wall (green), triple-wall (magenta), and multi-wall (blue) carbon nanotubes. Reproduction from our work, Honda*, Muroga*, Nakajima* (*contributed equally) et al., Communications Materials, 2, 88 (2021). DOI:10.1038/s43246-021-00195-2, open access with CC BY 4.0 license.

In our paper, the proposed concept of the deep learning framework is demonstrated by carbon nanotube films, one of the typical materials with tangible and hierarchical structures. Our results of artificial intelligence techniques (Figure 2) for generating structures and predicting properties can be used as a versatile database. We have investigated the followings from the results of artificial intelligence techniques in the paper.

  • Construction of phase diagrams from artificial intelligence techniques
  • Proposal of cost-effective composition to achieve desired properties of processed materials
  • Inverse design of energy device (super-capacitor) aiming to target performance

Not surprisingly, our proposed method is not limited to the carbon nanotube film. Any kind of materials with such complex structures can by treated by our proposed scheme in a similar manner of carbon nanotube film. Our demonstration of carbon nanotube films is one of the clues to the challenging applications of artificial intelligence techniques to such complex structures seeking for further high-performance materials and devices.

We are currently developing further advanced and innovative artificial intelligence techniques applicable to any kind of highly complex materials and processes. We are strongly believing that our developed artificial intelligence techniques dramatically provide a rational route to overcome the limitations of previous materials informatics and process informatics. If you are interested in our work, please visit and see our paper in Communications Materials or contact us and look forward to our work in the near future on surprising and attractive artificial intelligence techniques contributing to the evolution of materials and their processes.

Shun Muroga


Paper (Open Access)

†Honda, T., †Muroga, S., †Nakajima, H., Shimizu, T., Kobashi, K., Morita, H., Okazaki, T., *Hata, K., “Virtual Experimentations by Deep learning on Tangible Materials”, Communications Materials, 2, 88 (2021), DOI:10.1038/s43246-021-00195-2, Published: August 30, 2021.

†: These authors contributed equally to the paper (Honda, T., Muroga, S., Nakajima, H.), *: Corresponding author of the paper (Hata, K.)

Link: https://www.nature.com/articles/s43246-021-00195-2

Supporting Dataset

Muroga, S. “Trained models of Generative Adversarial Networks for Carbon Nanotube Hierarchical Structures”. figshare (2021). DOI: 10.6084/m9.figshare.14872146

 

Citation of this Article in Nature Portfolio

Muroga, S., “Deep learning virtual experiments for complex materials with non-periodic, undefinable, hierarchical, tangible structures -Overcoming the limitations of conventional materials and process informatics- “, Nature Portfolio Device and Materials Engineering Community (2021). Link: https://devicematerialscommunity.nature.com/posts/deep-learning-virtual-experiments-for-complex-materials-with-non-periodic-undefinable-hierarchical-tangible-structures-overcoming-the-limitations-of-conventional-materials-and-process-informatics

 

Acknowledgement

This work was supported by a project (JPNP16010) commissioned by the New Energy and Industrial Technology Development Organization (NEDO).

 

Lab Website

Link: https://unit.aist.go.jp/cnta/index_en.html 

CNT-Application Research Center

National Institute of Advanced Industrial Science and Technology (AIST)

 

Author of this Article in Nature Portfolio Device and Materials Engineering Community

Dr. Shun Muroga

Researcher, CNT-Application Research Center

National Institute of Advanced Industrial Science and Technology (AIST)

Contact: Personal E-mail address is available from the Community of Nature Portfolio.

ORCID Link: https://orcid.org/0000-0002-6330-0436

ResearchGate Link: https://www.researchgate.net/profile/Shun-Muroga

researchmap Link: https://researchmap.jp/s_muroga?lang=en


 

Please sign in or register for FREE

If you are a registered user on Research Communities by Springer Nature, please sign in

Subscribe to the Topic

Electrical and Electronic Engineering
Technology and Engineering > Electrical and Electronic Engineering

Related Collections

With collections, you can get published faster and increase your visibility.

Chiral topological matter

This Collection brings together the latest advances in our understanding of quantum materials where the intertwining of chirality and topology gives rise to novel and unexpected phenomenology.

Publishing Model: Open Access

Deadline: Jan 01, 2024

Memristors and non-volatile memory devices

This Collection brings together the latest developments in the realization and optimization of memristive technologies for modern applications that take advantage of neural networks and neuromorphic computing.

Publishing Model: Open Access

Deadline: Apr 01, 2024