NEURAL NETWORK MODEL FOR IDENTIFICATION OF MATERIAL CREEP CURVES USING CUDA TECHNOLOGIES
Ukrainian Journal of Information Technology · 2019
DOI: 10.23939/ujit2019.01.011 ↗Kurz gefasst
Fachartikel in der Zeitschrift Ukrainian Journal of Information Technology, 2019. Verfasst von Ya. I. Sokolovskyy und 3 weiteren Autor:innen. Die Arbeit nennt 26 Literaturangaben.
Abstract (Originalsprache)
This paper addresses the problem of identifying rheological parameters of wood using artificial neural networks with parallel learning algorithm using Python programming language, Chainer framework and CUDA technology. An intelligent system for identification of rheological parameters of wood has been developed. The system created contains the most user-friendly interface, all the necessary set of tools for automation of the process of visualization and analysis of data. In the process of creation of the intellectual system, the following tasks were envisaged: to carry out the analysis of artificial intelligence systems and the analysis of training of artificial neural networks, in particular multilayer neural networks of direct propagation, recurrent neural networks and the Kohonen neural network; examine the structure of the Chainer framework and its interaction with CUDA; to conduct existing cloud technologies to accomplish the task; to conduct the analysis of algorithms of studies of artificial neuron networks, their mathematical providing; to implement parallelization of learning algorithms and to develop the necessary software. Using Chainer allows you to create a memory pool for GPU memory allocation. To avoid memory allocation and erasure during computing, Chainer provides the ability to use the CuPy memory pool as a standard memory allocation without dealing with memory allocation. An intellectual system to determine the physical and mechanical parameters of a mathematical model of non-isothermal moisture transfer and viscoelastic deformation of capillary-porous materials was developed. It provides the opportunity to identify parameters of the kernels of creep and relaxation that is written as a linear combination of exponential operators. The proposed algorithm of approximation and obtained calculated ratios of rheological behavior of wood by means of multilayer neural network with exponential activation functions in hidden layers allows to increase the accuracy of approximation of experimental creep data. The developed mathematical models can be used to create an automated systems of finite-difference calculation of temperature and moisture content, stress components during the drying of capillary-porous materials with taking into account the technological parameters of the drying agent.
Zitieren
APA
Sokolovskyy, Y. I., Shymanskyi, V. M., Mokrytska, O. V., & Kharko, Y. V. (2019). NEURAL NETWORK MODEL FOR IDENTIFICATION OF MATERIAL CREEP CURVES USING CUDA TECHNOLOGIES. Ukrainian Journal of Information Technology. https://doi.org/10.23939/ujit2019.01.011
Vancouver
Sokolovskyy YI, Shymanskyi VM, Mokrytska OV, Kharko YV. NEURAL NETWORK MODEL FOR IDENTIFICATION OF MATERIAL CREEP CURVES USING CUDA TECHNOLOGIES. Ukrainian Journal of Information Technology. 2019. doi:10.23939/ujit2019.01.011
BibTeX
@article{Sokolovskyy2019,
author = {Sokolovskyy, Ya. I. and Shymanskyi, V. M. and Mokrytska, O. V. and Kharko, Ya. V.},
title = {NEURAL NETWORK MODEL FOR IDENTIFICATION OF MATERIAL CREEP CURVES USING CUDA TECHNOLOGIES},
journal = {Ukrainian Journal of Information Technology},
year = {2019},
doi = {10.23939/ujit2019.01.011}
}
EUVIMED