A research group led by Masanobu Horie at RICOS Co. Ltd. In collaboration with Assistant Professor Naoto Mitsum of the University of Tsukuba, in the use of artificial intelligence to achieve high-accuracy, high-speed predictions of water and air flow and other phenomena. This technology achieves a complex balance between accuracy and computation time (measured on the same computer) that has not been achievable using existing physical simulations and other AI methods. The paper has been published on arXiv Prepress server.
Physical simulations are the dominant methods for predicting flow phenomena. However, there is a trade-off between accuracy and computation time; High-accuracy analysis of phenomena requires a long calculation time, and simplifying the process to shorten the computation time reduces the prediction accuracy. In recent years, extensive research has been done on building models that predict physical phenomena using a basic artificial intelligence technique known as machine learning. However, this approach was often not applicable to simulations under complex conditions as handled in conventional physical simulations, and there were issues in terms of reliability and versatility.
By combining physical simulation with machine learning, this research group has realized a high-speed prediction model that guarantees reliability and versatility, while leveraging the strengths of machine learning to make predictions based on existing data. The group achieved high-speed predictions without significantly compromising accuracy compared to conventional physical simulations by having the model learn from pre-prepared high-resolution simulation data. In addition, this newly developed technique theoretically proves that prediction accuracy does not deteriorate, while prediction accuracy decreases with existing machine learning technology when observing the same phenomenon from a different perspective.
In physical simulations of the flow phenomenon, the boundary conditions of the phenomena are given, such as considering the parts of the “holes where air enters” and “walls that do not allow air to pass”. However, current machine learning technology cannot take these specific circumstances into account. The new technology successfully combines machine learning algorithms with rigorous processing of boundary conditions by modeling the correspondence between the input physical conditions and those in the high-dimensional abstract data space processed by the machine learning algorithms.
This was achieved by embedding the computational methods of physical simulation into a machine learning algorithm, which is a unique feature of this technique. This time, the research group succeeded in showing that machine learning can have the same versatility as traditional physical simulations without losing the advantages of machine learning.
This technology is expected to speed up the evaluation process by simulating the flow phenomenon, which can be a hurdle in design and manufacturing, and improve the efficiency of the entire design and manufacturing processes. It may also be an important step in increasing the accuracy of weather forecasts and improving the efficiency of ventilation systems to prevent the spread of infectious diseases caused by droplets.
Masanobu Horie et al., Physics Embedded Neural Networks: Graph of PDE Neural Solutions with Mixed Boundaries, arXiv (2022). doi: 10.48550/arxiv.2205.11912
Provided by the Japan Science and Technology Agency
the quote: Rapid and Highly Accurate Prediction of Flow Phenomena (2023, April 3) Retrieved April 3, 2023 from https://phys.org/news/2023-04-speedy-highly-accurate-phenomena.html
This document is subject to copyright. Apart from any fair dealing for the purpose of private study or research, no part may be reproduced without written permission. The content is provided for informational purposes only.