Heat Conduction Plate Layout Optimization Using Physics-Driven Convolutional Neural Networks

Abstract

Optimizing heat conduction layout is essential during engineering design, especially for sensible thermal products. However, when the optimization algorithm iteratively evaluates different loading cases, the traditional numerical simulation methods usually lead to a substantial computational cost. To effectively reduce the computational effort, data-driven approaches are used to train a surrogate model as a mapping between the prescribed external loads and various geometry. However, the existing model is trained by data-driven methods, which require intensive training samples from numerical simulations and do not effectively solve the problem. Choosing the steady heat conduction problems as examples, this paper proposes a physics-driven convolutional neural networks (PD-CNNs) method to infer the physical field solutions for randomly varied loading cases. After that, the particle swarm optimization (PSO) algorithm is used to optimize the sizes, and the positions of the hole masks in the prescribed design domain and the average temperature value of the entire heat conduction field is minimized. The goal of reducing heat transfer is achieved. Compared with the existing data-driven approaches, the proposed PD-CNN optimization framework predicts field solutions that are highly consistent with conventional simulation results. However, the proposed method generates the solution space without pre-obtained training data. We obtained thermal intensity results for holes 1, hole 2, hole 3, and hole 4 with 0.3948, 0.007, 0.0044, and 0.3939, respectively, by optimization PD-CNN model.