TY - JOUR

T1 - An intelligence parameter classification approach for energy storage and natural convection and heat transfer of nano-encapsulated phase change material

T2 - Deep neural networks

AU - Ghalambaz, Mohammad

AU - Edalatifar, Mohammad

AU - Moradi Maryamnegari, Sara

AU - Sheremet, Mikhail

N1 - Funding Information: This research of Mohammad Ghalambaz and Mikhail Sheremet was supported by the Tomsk State University Development Program (Priority-2030).

PY - 2023/9

Y1 - 2023/9

N2 - A deep neural network is utilized to classify the parameters of a natural convection heat transfer of a nano-encapsulated phase change material suspension using the isotherm images for the first time. A natural convection flow and heat transfer simulation dataset were created and used as a training and validation tool. Then, a deep neural network, consisting of three parts, was used for the classification task. The first part was made of several conventional layers, and a rectified linear unit activation layer supported each layer. The second part was a preparation layer for reshaping from 2D images to 1D classification. The third layer was made of a classifier layer. The results showed that the impact of the Rayleigh number and volume concentrations of nanoparticles could be classified by 99.8 and 93.32% accuracy, respectively. However, the Stefan number was classified weakly. As a part of the current research, a transfer learning approach was used to improve accuracy. The learning transfer approach was quite effective and improved the accuracy of the Stefan number classification by 16.6%.

AB - A deep neural network is utilized to classify the parameters of a natural convection heat transfer of a nano-encapsulated phase change material suspension using the isotherm images for the first time. A natural convection flow and heat transfer simulation dataset were created and used as a training and validation tool. Then, a deep neural network, consisting of three parts, was used for the classification task. The first part was made of several conventional layers, and a rectified linear unit activation layer supported each layer. The second part was a preparation layer for reshaping from 2D images to 1D classification. The third layer was made of a classifier layer. The results showed that the impact of the Rayleigh number and volume concentrations of nanoparticles could be classified by 99.8 and 93.32% accuracy, respectively. However, the Stefan number was classified weakly. As a part of the current research, a transfer learning approach was used to improve accuracy. The learning transfer approach was quite effective and improved the accuracy of the Stefan number classification by 16.6%.

KW - Deep learning

KW - Nano-encapsulated phase change suspension

KW - Natural convection heat transfer

KW - Physical characteristics classification

UR - http://www.scopus.com/inward/record.url?scp=85163883896&partnerID=8YFLogxK

U2 - 10.1007/s00521-023-08708-5

DO - 10.1007/s00521-023-08708-5

M3 - Article

AN - SCOPUS:85163883896

VL - 35

SP - 19719

EP - 19727

JO - Neural Computing and Applications

JF - Neural Computing and Applications

SN - 0941-0643

IS - 27

ER -