ライブラリ登録: Guest

ISSN Online: 2377-424X

ISBN Print: 978-1-56700-474-8

ISBN Online: 978-1-56700-473-1

International Heat Transfer Conference 16
August, 10-15, 2018, Beijing, China

AN IMPROVED CONVOLUTION NEURAL NETWORK MODEL TO PREDICT THE EFFECTIVE THERMAL CONDUCTIVITY OF COMPOSITE MATERIALS

Get access (open in a dialog) DOI: 10.1615/IHTC16.nmt.022090
pages 6891-6901

要約

The thermal transport in composite materials has attracted research attentions for the past many years. Numerous methods, such as direct solution of governing equation and effective medium theory (EMT) are generally applied to conduct such investigations. With the recent development of machine learning methods, it is now possible to develop models based on statistical data to obtain the effective thermal conductivity of composite materials. In this work, an improved convolution neural network (CNN) model is developed to predict the effective thermal conductivity of composite materials, and the results are compared with other machine learning methods including support vector regression (SVR) and Gaussian process regression (GPR). Different network structures are developed to find an optimal model for the training data obtained from the lattice Boltzmann method (LBM). To further improve CNN model in predicting thermal conductivity of composite materials, a new Concat layer is added to consider the thermal conductivity values of matrix and inclusion. Since the improved CNN model can include the detailed distribution of inclusions, a better predicting performance compared with the predictions by SVR and GPR is obtained. This work demonstrate the potential of using machine learning method to develop surrogate models for predicting effective properties of composite materials.