Construction
of High Throughput Multiscale Material Simulation and Optimization Platform
Meng
Guo1, Dingwang Yuan2, Moubin Liu3, Meihong
Yang1, Zheng Qin2, Yiwu Ma2 and Wangyu Hu2*
1
Shandong Computer Science Center (National Supercomputer Center in Jinan), 1768
Xinluo Ave., High-tech Zone, Jinan, Shandong 250101, China
2
Hunan University，2
Lushan South Road, Yuelu District, Changsha, Hunan 410082, China
3
Peking
University，5 Yiheyuan Road, Haidian District, Beijing 100871, China
ABSTRACT: Large scale
computational simulation and data analysis have become an indispensable means for
new materials research and development. The requirement of computing resources
in material research and application is higher and higher. However, the
existing material computing platform has single mode, small scale and low
resource utilization efficiency. The material design is difficult to couple in
the same scale and between different scales, and the process is difficult to be
automated, as well as the result data is lack of effective integration,
analysis and expression. Based on the heterogeneous supercomputing grid
environment, we realize the integration, optimization and process automation of
multi-scale material computing software resources. We develop the dynamic data
acquisition and intelligent analysis technology for material simulation, and
form an open and shared resource system and database, so as to build a cloud
based platform for high-throughput multi-scale material simulation and
performance optimization, so as to leverage the requirement of computational
material and the supply of computing resources. It provides a platform for the
transformation from the traditional mode of "experience guided
experiment" to the new mode of "theoretical prediction then
experimental verification". According to the development concept of
material genetic engineering, we built an intelligent service cloud platform
for high throughput and multi scale material simulation and performance
optimization based on cloud computing technology, and developed the whole
process work system of material design, calculation and screening, and
customized the whole process workflow for typical material characteristics and
requirements, and provided it to users in the form of cloud service; the cloud
platform access the supercomputing grid resources and special database system
through a unified interface. Based on grid computing technology, we aggregate
multiple supercomputing resources and developed a large-scale resource
management, load balancing and high throughput task scheduling system to manage
the heterogeneous environment. We integrated and ported a series of scale
computational materials software including first principles, molecular
dynamics, Monte Carlo, phase diagram calculation, phase field simulation and
finite element analysis on several supercomputing platforms supporting x86,
Shenwei and Feiteng etc., which provide unified resource services for cloud
platform. According to the data characteristics of the whole process of
material simulation and simulation, we use the methods of multi-source
heterogeneous data modeling, knowledge discovery and knowledge mapping to
realize dynamic data perception, intelligent acquisition and data coupling
between processes, and establish distributed material database and knowledge
map database. Based on the developed computing platform, we have realized the
whole chain process framework of material design based on the scientific and
engineering requirements of typical materials such as aluminum alloy and
tungsten metal, and realized the process demonstration of high throughput multi
scale material simulation and performance optimization. In the future, we will
further aggregate advanced computing resources, integrate material computing
software, enrich material databases, expand high-throughput multi-scale
material computing processes and cases, open the platform for trails, and
explore the operation mode of computing platform suitable for the development
of material genetic engineering, and formulate the service mechanism and
transaction process of computing resources, data resources and scientific
research achievements of material genetic engineering, which will lead to
providing a better platform support for the development and application of new
materials.
Keywords: Material Genetic Engineering; High Throughput Multi Scale Computing; Computational Material Workflow; Grid Computing; Material Genetic Database.
Wangyu Hu, professor and doctoral supervisor of Hunan University, has been selected as the excellent cross century talents of the Ministry of Education and the new century 121 talent project of Hunan Province. He is the chief scientist of the National Key R&D Program High Throughput Multiscale Material Simulation and Optimization Platform. He has presided over and completed the sub project of Basic Research on The Application of Liquid Lithium Wall in Future Fusion Devices, 11 National Natural Science Funds projects and a major special project of Hunan Science and Technology Program. He is committed to the development of Analytical Embedded Atom Model (EAM) and multiscale parallel algorithm for material properties research and new material and device development. The monograph Embedded Atom Method Theory and Its Application in Materials Science: Atomic Scale Material Design Theory won the 14th China Book Award. He won the first prize of science and technology progress award of Hunan Province, the third prize of science and technology progress award of Ministry of Machinery and Electronics Industry, and the second prize of science and technology progress award of National Bureau of Machinery Industry twice. He has published more than 300 papers which are cited more than 5000 times with the H factor of 36.