Effective Thermal Conductivity Estimation Using a Convolutional Neural Network and Its Application in Topology Optimization
Published in Energy and AI, 2024
In this study, a convolutional neural network (CNN) model for predicting effective thermal conductivity inspired by the VGG networks is proposed. Trained using 130,000 unique binary images, the model achieves high predictive accuracy. The model is blind to physics, thus the predictions are purely based on the microstructure and data acquired through training. Nonetheless, it reaches unprecedented accuracy with a mean absolute percent error (MAPE) of 0.35% in testing at a thermal conductivity ratio of 10, and the accuracy drops to a MAPE of 2.35% when the thermal conductivities considered are that of aluminum and water.
This model proves quite useful when applied in topology optimization tasks, especially efficient in comparison to conventional CFD models in such tasks. The prediction time is 3 - 5 orders of magnitude faster, and the small MAPEs don’t have a negative effect in the overall optimization process. This work is a proof of concenpt that predictive models based on computer vision can be extremely accurate and reliable for optimization tasks.
Two GitHub repositories stem from this work, as well as a dataset publication. They are as follows:
Recommended citation: Andre Adam, Huazhen Fang, Xianglin Li, Effective thermal conductivity estimation using a convolutional neural network and its application in topology optimization, Energy and AI, Volume 15, 2024, 100310, ISSN 2666-5468, https://doi.org/10.1016/j.egyai.2023.100310.
Download Paper