Chapter 6

Reverse Design of Heat Exchange Systems Using Physics-Informed Machine Learning

Chang He

Chang He

Sun Yat-Sen University, School of Chemical Engineering and Technology, No. 135, Xingang Xi Road, Zhuhai, Guangdong, 519082 China

The Key Laboratory of Low-carbon Chemistry & Energy Conservation of Guangdong Province, Guangdong Engineering Center for Petrochemical Energy Conservation, 132 Waihuan East Road, University City, Panyu District, Guangzhou, 510275 China

Search for more papers by this author
Yunquan Chen

Yunquan Chen

Sun Yat-Sen University, School of Materials Science and Engineering, No. 135, Xingang Xi Road, Guangzhou, 510275 China

Search for more papers by this author
First published: 13 December 2024

Summary

Geometry optimization is crucial for heat exchanger design, but it is often constrained by the high computational time and cost required. By utilizing physics-informed neural network (PINN) to parameterize the geometric and operating inputs, this chapter proposes a new inverse design method that starts with the desired objectives and works backward to find the optimal designs. For this purpose, specialized PINN structures are functionally designed to approximate the governing equations and construct PINN-derived surrogate models, which can be coupled with multi-objective optimization and decision-making algorithms. Results are presented for two examples: (i) a 3D finned heat sink system and (ii) a tubular air cooler system, demonstrating the ability to not only accelerate the search for the Pareto-optimal designs but also visualize real-time field distributions for better physical inspection.

The full text of this article hosted at iucr.org is unavailable due to technical difficulties.