Session: DAC-04-02: Data-Driven Design

Paper Number: 116429

116429 - Heat Sink Design Optimization via Gan-Cnn Combined Deep-Learning 

This work proposes a combined deep learning based ap- proach to improve thermal component heat sinks involving tur- bulent fluid flow. Random ellipse based heat sink designs are gen- erated using a python library called ellipse packing. A Generative Adversarial Network (GAN) is trained to learn and recreate the new ellipse based heat sinks. OpenFoam 7 (Open Source Com- putational Fluid Dynamics software) along with high throughout computing are used to efficiently generate temperature data for new designs. To improve the speed of design evaluation, a Con- volutional Neural Network (CNN) is trained to predict the entire temperature field for a given design. The trained CNN is able to predict the entire temperature field for the design with a Mean Average Error of 0.4 degrees kelvin in 0.04 seconds (22,500 times faster than the simulation). Later, the CNN and GAN are com- bined together to be able to create and simulate new designs. The combined model is then used to optimize the latent representation of 64 random designs on a Graphical Processing Unit (GPU) in ten minutes. Optimized designs outperform the training data by five degrees on average. The combined GAN and CNN framework has applications to a variety of data driven engineering problems to produce new and optimal designs.

Presenting Author: Nathan Flynn University of Wisconsin, Madison

Presenting Author Biography: Research Assistant
Master's student

Authors:

Nathan Flynn University of Wisconsin, Madison
Xiaoping Qian University of Wisconsin, Madison