A deep learning method for estimating thermal boundary condition parameters in transient inverse heat transfer problem

Estimating thermal boundary condition parameters in the transient inverse heat transfer problem (IHTP) is characterized by instability and non-uniqueness of the solution. This study formulated a simple method based on deep learning, which can realize multi-dimensional real-time prediction of thermal boundary condition parameters. The proposed model combining convolutional neural network (CNN) and long short-term memory networks (LSTM) allowed estimating multiple time-varying parameters based on the time-varying temperature field image of the target. The data-driven model used the computational fluid dynamics (CFD) method to obtain the numerical data, and the influence weight of the parameters was introduced in the training process to improve the generalization ability of the model. An experiment on the cubic cavity was made to verify the reliability of the proposed model to estimate time-varying parameters. The studies we have performed showed that the proposed hybrid models outperformed the standalone models (CNN, LSTM) in estimating multiple time-varying parameters. In the experimental results, the relative errors of air temperature and humidity were only 2.33% and 4.33%, respectively. These attempts of introducing the deep learning method to the IHTP in the present study were successful and it was significant for the study of the transient inverse heat transfer problem. (c) 2022 Elsevier Ltd. All rights reserved.

Automated summary

This study proposed a simple method based on deep learning to estimate thermal boundary condition parameters in the transient inverse heat transfer problem. By combining convolutional neural network (CNN) and long short-term memory networks (LSTM), real-time prediction of multiple time-varying parameters can be achieved. Experimental results showed that the proposed model outperformed standalone models in estimating multiple time-varying parameters.