| 研究課題/領域番号 |
17K17825
|
|---|
| 研究機関 | 京都大学 |
|---|
研究代表者 |
MOLINA JOHN 京都大学, 工学研究科, 助教 (20727581)
|
|---|
| 研究期間 (年度) |
2017-04-01 – 2022-03-31
|
| キーワード | machine learning / constitutive relations / multi-scale simulations |
|---|
| 研究実績の概要 |
We have developed a Machine Learning (ML) method that is capable of learning the constitutive relation for the stress of soft matter materials from microscopic training data. This is crucial because Soft Matter is characterized by a hierarchy of length- and time-scales, which makes them incredibly challenging to model. Our method can be applied to any Soft Matter system in order to accelerate the currently used state-of-the-art multi-scale simulation methods. We tested our method by learning the constitutive relation for polymer melt flows, for which there exist well-established theoretical and computational models, in contrast to the more recent and phenomenological work on cells and cellular tissues. In particular, we verified our approach by learning the constitutive relation for an ensemble of non-interacting Hookean dumbbells, for which the exact solution is known, i.e., the Maxwell model. Our results are in excellent agreement with Multi-Scale simulations, but they require only a small fraction of the computational resources. We thus achieve the best of both worlds: the accuracy of multi-scale descriptions with the speed of macroscopic descriptions.
These results have been published in two recent papers, which explore the use of ML techniques within standard physics simulations, with the goal of developing physics informed machine-learning approaches.
|
| 現在までの達成度 (区分) |
現在までの達成度 (区分)
2: おおむね順調に進展している
理由
Our recent (and originally unplanned) work on Machine Learning for Soft Matter has been incredibly successful and fruitful. We think it opens up many exciting applications, including, but not limited to, learning the constitutive relations of cellular tissues. To do this, we largely paused the work on the cell model itself, but we think the benefits far outweigh any downside.
|
| 今後の研究の推進方策 |
We will continue to develop our Machine Learning approach (to consider more complex microscopic models), and investigate how to apply it to cellular tissues. In parallel, we will continue our study on the dynamics of cells, and how to control them, using the original phase-field model developed at the beginning of the project. The long-term goal will be to learn the constitutive relation for an active tissue, using training data generated from the sub-cellular phase-field model. This would allow us to consider much larger systems than we thought possible when we began this study.
|
| 次年度使用額が生じた理由 |
Due to the COVID-19 outbreak, we were not able to carry out our planned activities. Initially we had planned to attend the APS march meeting in 2020, and after this was cancelled, we made plans to attend the 2021 edition. Unfortunately, it was still impossible to travel, so this latest edition was held as an online conference.
While the COVID situation is improving, we are not optimistic about the state of travel during this fiscal year, so we will not make any conference travel plans at this moment. Instead, we will use our remaining budget to purchase computing time on the university supercomputer, in order to run additional simulations/analysis.
|
