Análisis de clústeres para simulaciones de mecánica granular mediante algoritmos de aprendizaje automático

DOI: https://doi.org/10.31908/19098367.2058
Palabras clave: simulaciones granulares, aprendizaje automático, análisis de clasificación, análisis de rendimiento

Resumen

Las simulaciones de dinámica molecular (MD) en colisiones de granos permiten incorporar propiedades complejas de interacciones de polvo. Realizamos simulaciones de colisiones de granos porosos, cada uno con muchas partículas, utilizando el software LAMMPS de MD. Las simulaciones consistieron en un grano de proyectil que golpeó un grano objetivo inmóvil más grande, con diferentes velocidades de impacto. La desventaja de este método es el gran costo computacional debido a que se modela una gran cantidad de partículas. Machine Learning (ML) tiene el poder de manipular grandes datos y construir modelos predictivos que podrían reducir los tiempos de simulación MD. Usando algoritmos ML (Support Vector Machine y Random Forest) podemos predecir el resultado de las simulaciones MD con respecto a la formación de fragmentos, después de varios pasos más pequeños que en las simulaciones MD habituales. Logramos una reducción de tiempo de al menos un 46%, para una precisión del 90%. Estos resultados muestran que SVM y RF pueden ser herramientas poderosas pero simples para reducir el costo computacional en simulaciones de fragmentación de colisiones.

Descargas

La descarga de datos todavía no está disponible.

Biografía del autor

Daniela Noemi Rim, Universidad Nacional de Cuyo

Bachelor’s degree in Basic Sciences with Orientation in Physics (Faculty of Exact and Natural Sciences, National University of Cuyo, Mendoza, Argentina, emitted April 17 th 2019). Thesis title: ‘Machine Learning techniques applied to Numerical Simulation of Granular Porous Materials’. Currently studying an MS in South Korea (‘MS in Information Technology’, Department of Information and Communication Engineering, Handong Global University, Pohang, Republic of Korea, since February 2020). Member of research group MILab (Machine Intelligence Lab, Handong Global University, Pohang, Republic of Korea). The topic of research focuses in Deep Learning applied to Natural Language Processing tasks (such as Neural Machine Translations) and audio signal compressions using Variational Autoencoders.

Emmanuel N. Millán, Universidad Nacional de Cuyo

Received his Ph.D. degree from Universidad Nacional de San Luis (UNSL), Argentina in 2016, and a BSc. in Software Engineering from Universidad del Aconcagua, Argentina, in 2010. He is a researcher within CONICET. He is interested in the implementation of parallel problems in hybrid clusters including Graphics Processing Units (GPUs), with applications in Molecular Dynamics, Machine Learning, Cellular Automata and Monte Carlo methods.

María Belén Planes, Universidad de Mendoza

2016 “Licenciada en Ciencias Básicas con orientación en Física” FCEN – Universidad Nacional de Cuyo – Mendoza, Argentina. 2018 “Profesora de grado universitario en Ciencias Básicas con orientación en Física” FCEN – Universidad Nacional de Cuyo – Mendoza, Argentina. 2020 last-year student “Doctorado en Astronomía” FCEFN - Universidad Nacional de San Juan – San Juan, Argentina. Dust aggregate collisions have been studied through complex molecular dynamics simulations with a focus on astrophysical topics that are not currently understood, such as the formation of planets in their early stage, high speed collisions in debris discs and the evolution of dust emitted by comets in their internal coma. [Planes M. B., et al, A&A 607, A19 (2017)] [Planes M. B., et al, MNRAS: Letters 487, L13 (2019)] [Planes M. B., et al, MNRAS 492, 1937 (2020)]. SIMAF – Universidad de Mendoza. CONICET – Argentina Research areas: planetary formation – comets – granular mechanics.

Eduardo M. Bringa, Universidad de Mendoza

Ph.D. Physics. 2000. University of Virginia (UVa), Charlottesville, USA. 1994. Licenciado en Física, Instituto Balseiro, Bariloche, Argentina. After obtaining his Ph.D., he was a postdoctoral researcher at the Astronomy Department (UVa, 2000-2001), and then postdoctoral researcher at Lawrence Livermore National Laboratory (LLNL, 2001-2003), where he later became part of the permanent research staff. In 2008 he returned to Argentina, where he currently is Principal Researcher in CONICET at the “Universidad de Mendoza”, and full Professor at “Universidad Nacional de Cuyo”. He is a member of AFA (Argentinean Physical Society) and APS (American Physical Society). Research area: simulations in physics, astrophysics and materials sciences, including Molecular Dynamics, Spin Dynamics and Monte Carlo simulations.

Luis G. Moyano, Universidad Nacional de Cuyo

Is an adjunct researcher at CONICET and associate professor at Instituto Balseiro (UNCuyo/Comisión Nacional de Energía Atómica). He is invited professor at Facultad de Ciencias Exactas y Naturales, UNCuyo. He graduated in Physics from Instituto Balseiro (Bariloche, Argentina, 2000) and holds a PhD also in Physics from CBPF/UFRJ (Rio de Janeiro, Brazil, 2006). Dr. Moyano was research staff member at IBM Research Brazil until 2016. Prior to working at IBM, he was leader data scientist at BBVA Data & Analytics (Madrid, Spain). He was also staff researcher at Telefónica Research (2008-2013, Madrid/Barcelona). He is currently a permanent member of the Statistical and Interdisciplinary Physics Group at Centro Atómico Bariloche. Dr. Luis Gregorio Moyano research interests lie in the intersection of physics and machine learning. His lines of work include network representation learning, with applications to biological and social systems

Citas

C. Ringl, E. M. Bringa, D. S. Bertoldi, and H. M. Urbassek, “Collisions of porous clusters: A granular-mechanics study of compaction and fragmentation,” The Astrophysical Journal, vol. 752, no. 2, p. 151, Jun 2012. [Online]. Available: http://dx.doi.org/10.1088/0004-637X/752/2/151.

V. Botu and R. Ramprasad, “Adaptive machine learning framework to accelerate ab initio molecular dynamics,” International Journal of Quantum Chemistry, vol. 115, no. 16, pp. 1074–1083, dec 2014.

M. Rupp, A. Tkatchenko, K.-R. Muller, and O. A. von Lilienfeld, “Fast and accurate modeling of molecular atomization energies with machine learning,” Physical Review Letters, vol. 108, no. 5, jan 2012.

D. S. Glazer, R. J. Radmer, and R. B. Altman, “Combining molecular dynamics and machine learning to improve protein function recognition,” in Pacific Symposium on Biocomputing. Pacific Symposium on Biocomputing. NIH Public Access, 2008, p. 332.

B. Lantz, Machine learning with R. Packt Publishing Ltd, 2013.

A. D. Whizin, J. Blum, and J. E. Colwell, “The physics of protoplanetesimal dust agglomerates. VIII. microgravity collisions between porous SiO2aggregates and loosely bound agglomerates,” The Astrophysical Journal, vol. 836, no. 1, p. 94, feb 2017.

M. B. Planes, E. N. Millán, H. M. Urbassek, and E. M. Bringa, “Dust-aggregate impact into granular matter: A systematic study of the influence of projectile velocity and size on crater formation and grain ejection,” Astronomy & Astrophysics, jun 2017.

C. Ringl and H. M. Urbassek, “A lammps implementation of granular mechanics: Inclusion of adhesive and microscopic friction forces” Computer Physics Communications, vol. 183, no. 4, pp. 986–992, Apr2012. [Online]. Available: http://dx.doi.org/10.1016/j.cpc.2012.01.004.

C. Ringl and H. M. Urbassek, “A simple algorithm for constructing fractal aggregates with pre-determined fractal dimension,” Computer Physics Communications, vol. 184, no. 7, pp. 1683–1685, jul 2013.

A. Seizinger, R. Speith, and W. Kley, “Compression behavior of porous dust agglomerates,” Astronomy & Astrophysics, vol. 541, p. A59, apr 2012.

J. Blum, “Experiments on sticking, restructuring, and fragmentation of preplanetary dust aggregates,” Icarus, vol. 143, no. 1, pp. 138–146, jan 2000.

E. N. Millán, C. Ringl, C. S. Bederián, M. F. Piccoli, C. G. Garino, H. M. Urbassek, and E. M. Bringa, “A gpu implementation for improved granular simulations with lammps,” in VI Latin American Symposium on High Performance Computing HPCLatAm 2013, C. G. Garino and M. Printista, Eds., 2013, pp. 89–100. [Online]. Available: http://hpc2013.hpclatam.org/papers/HPCLatAm2013-paper-10.pdf.

A. Kassambara, Practical guide to cluster analysis in R: Unsupervised machine learning. STHDA, 2017, vol. 1.

E. N. Millán, C. A. Ruestes, N. Wolovick, and E. M. Bringa, “Boosting materials science simulations by high performance computing,” in Actas de ENIEF 2017: Mecánica Computacional Vol. XXXV, AMCA. Asociación Argentina de Mecánica Computacional, Nov. 2017. [Online]. Available: https://cimec.org.ar/ojs/index.php/mc/issue/archive.

E. N. Millán, N. Wolovick, M. F. Piccoli, C. G. Garino, and E. M. Bringa, “Performance analysis and comparison of cellular automata GPU implementations,” Cluster Computing, apr 2017.

H.-P. K. J. S. Ester, Martin and X. Xu, “A density-based algorithm for discovering clusters in large spatial databases with noise.” vol. 96, no. 34, 1996, pp. 226–231.

Publicado
2020-12-21
Cómo citar
Rim, D., Millán, E., Planes, M., Bringa, E., & Moyano, L. (2020). Análisis de clústeres para simulaciones de mecánica granular mediante algoritmos de aprendizaje automático. Entre Ciencia E Ingeniería, 14(28), 81-86. https://doi.org/10.31908/19098367.2058
Número
Vol. 14 Núm. 28 (2020)
Sección
Artículos
Derechos de autor 2021 Entre Ciencia e Ingeniería