A deep learning framework for predicting the effect of surface topography on thermal contact resistance

Efficient heat transfer across contacting surfaces is essential for effective thermal management; however, it is often hindered by thermal contact resistance resulting from complex surface topography. Here we present an interpretability analysis utilizing deep learning based to predict and visualize the key features influencing thermal contact resistance. We developed a convolutional-neural-network-based model trained on an extensive dataset generated using surface fractal theory. The model's predictive performance was validated against experimental data, in which surface topography and thermal resistance were measured for ground and turned steel specimens. Our model accurately predicts thermal contact resistance and estimates the actual contact area. Moreover, interpretability and visualization techniques reveal that both direct contact spots and non-contact regions of the surface topography significantly affect heat transfer, surpassing the explanatory power of traditional roughness parameters. This approach provides a robust methodology to enhance the fundamental understanding and predictive capabilities regarding thermal contact resistance.