reduced order framework for 2d heat transfer simulation under variations of boundary conditions based on deep learning algorithms

Due to the high computational cost of the direct numerical simulation methods of the governing equations of some natural phenomena, surrogate models based on machine learning methods such as deep learning algorithms have been commonly interested in modeling these phenomena. this paper proposes a reduced-order model based on a deep-learning algorithm to simulate temperature changes in a two-dimensional field. this model is developed using three different methods, including a framework based on convolutional neural networks, a physics-informed loss function of the phenomenon, and a reduced-order model using the autoencoder method. the model outcomes were compared with the results obtained from a high-resolution finite difference method. the results show that the reduced-order model (with an accuracy of 2.528×10-6 °c) has higher accuracy than the other two models. meanwhile, the model-based physics-informed loss is superior to the other two models in terms of steady-state temperature data consumption (only 400 data of size 8×8).

reduced order framework for 2d heat transfer simulation under variations of boundary conditions based on deep learning algorithms

due to the high computational cost of the direct numerical simulation methods of the governing equations of some natural phenomena, surrogate models based on machine learning methods such as deep learning algorithms have been commonly interested in modeling these phenomena. this paper proposes a reduced-order model based on a deep-learning algorithm to simulate temperature changes in a two-dimensional field. this model is developed using three different methods, including a framework based on convolutional neural networks, a physics-informed loss function of the phenomenon, and a reduced-order model using the autoencoder method. the model outcomes were compared with the results obtained from a high-resolution finite difference method. the results show that the reduced-order model (with an accuracy of 2.528×10-6 °c) has higher accuracy than the other two models. meanwhile, the model-based physics-informed loss is superior to the other two models in terms of steady-state temperature data consumption (only 400 data of size 8×8).