A novel hybrid deep learning algorithm for estimating temperature-dependent thermal conductivity in transient heat conduction problems

Abstract

Thermal conductivity is a fundamental parameter in heat transfer, and effectively identifying the thermal conductivity property of materials is crucial for engineering applications. A novel deep learning framework combining bidirectional long short-term memory (Bi-LSTM) networks and multi-head self-attention (MSA) mechanisms is proposed to estimate temperature-dependent thermal conductivity for transient inverse heat conduction problems. The training data is obtained through finite element method (FEM). The temperature fields are utilized as inputs to train the network, enabling it to predict unknown thermal conductivity. A dynamic learning rate decay adjustment strategy is adopted to improve the performance of the model. In the proposed novel hybrid models, Bi-LSTM captures both forward and backward dependencies in the input data, while MSA enhances the learning ability of the model in complex nonlinear relationships by processing input sequences in parallel with different attention weights. Numerical examples analyze the effects of noise and the proportion of training samples on the prediction results, and the results show that the proposed network is less sensitive to noise. Moreover, comparison with other deep learning models highlights the superiority of the proposed framework. It demonstrates accuracy and effectiveness of the proposed method in identifying temperature-dependent thermal conductivity in 2D and 3D models.