OpenImpala: OPEN source IMage based PArallisable Linear Algebra solver
OpenImpala: OPEN source IMage based PArallisable Linear Algebra solver
复制标题
DOI：
10.1016/j.softx.2021.100729
复制
发表时间：
2021-06-04
期刊：
影响因子：
3.4
通讯作者：
Kramer, Denis
中科院分区：
文献类型：
--
作者：
Le Houx, James;Kramer, Denis
基于图像的建模已经成为锂离子电池建模领域内的流行方法，这是由于其能够表示多孔电极的异质性。基于图像的建模的常见挑战是3D断层扫描数据集的大小，其可以是数十亿体素的量级。以前，不同的近似方法已被用来简化计算问题，但每一个都有相关的限制。在这里，我们开发了一个数据驱动的，完全并行的，基于图像的建模框架，称为OpenImpala。微X射线计算机断层扫描（CT）用于非破坏性地从样品中获得3D显微结构数据。然后，这些3D数据集直接用作基于有限差分的直接物理建模的计算域（例如，直接在CT获得的数据集上求解扩散方程）。然后OpenImpala计算给定微观结构的等效均匀传输系数。这些系数被写入参数化文件，以便与两种流行的连续电池模型直接兼容：PyBamm和DandeLiion，促进不同尺度的计算电池建模之间的联系。OpenImpala已被证明可以很好地扩展分布式内存架构上越来越多的计算核心，使其适用于现代断层扫描的典型大型数据集。(C)2021作者由爱思唯尔公司出版
Image-based modelling has emerged as a popular method within the field of lithium-ion battery modelling due to its ability to represent the heterogeneity of the porous electrodes. A common challenge from image-based modelling is the size of 3D tomography datasets, which can be of the order of several billion voxels. Previously, different approximation methods have been used to simplify the computational problem, but each of these come with associated limitations. Here we develop a data-driven, fully parallelisable, image-based modelling framework called OpenImpala. Micro X-ray computed tomography (CT) is used to obtain 3D microstructural data from samples non-destructively. These 3D datasets are then directly used as the computational domain for finite-differences based direct physical modelling (e.g. to solve the diffusion equation directly on the CT obtained datasets). OpenImpala then calculates the equivalent homogenised transport coefficients for the given microstructure. These coefficients are written into parameterised files for direct compatibility with two popular continuum battery models: PyBamm and DandeLiion, facilitating the link between different scales of computational battery modelling. OpenImpala has been shown to scale well with an increasing number of computational cores on distributed memory architectures, making it applicable to large datasets typical of modern tomography. (C) 2021 The Authors. Published by Elsevier B.V.