A physics and data co-driven surrogate modeling approach for temperature field prediction on irregular geometric domain

Kairui Bao,Wen Yao,Xiaoya Zhang,Wei Peng,Yu Li

DOI: https://doi.org/10.48550/arXiv.2203.08150

2022-03-15

Abstract:In the whole aircraft structural optimization loop, thermal analysis plays a very important role. But it faces a severe computational burden when directly applying traditional numerical analysis tools, especially when each optimization involves repetitive parameter modification and thermal analysis followed. Recently, with the fast development of deep learning, several Convolutional Neural Network (CNN) surrogate models have been introduced to overcome this obstacle. However, for temperature field prediction on irregular geometric domains (TFP-IGD), CNN can hardly be competent since most of them stem from processing for regular images. To alleviate this difficulty, we propose a novel physics and data co-driven surrogate modeling method. First, after adapting the Bezier curve in geometric parameterization, a body-fitted coordinate mapping is introduced to generate coordinate transforms between the irregular physical plane and regular computational plane. Second, a physics-driven CNN surrogate with partial differential equation (PDE) residuals as a loss function is utilized for fast meshing (meshing surrogate); then, we present a data-driven surrogate model based on the multi-level reduced-order method, aiming to learn solutions of temperature field in the above regular computational plane (thermal surrogate). Finally, combining the grid position information provided by the meshing surrogate with the scalar temperature field information provided by the thermal surrogate (combined model), we reach an end-to-end surrogate model from geometric parameters to temperature field prediction on an irregular geometric domain. Numerical results demonstrate that our method can significantly improve accuracy prediction on a smaller dataset while reducing the training time when compared with other CNN methods.

Machine Learning

What problem does this paper attempt to address?

This paper attempts to solve the problem of temperature field prediction on irregular geometric domains. Specifically, in the entire aircraft structure optimization process, thermal analysis plays a very important role. However, when directly applying traditional numerical analysis tools (such as the finite difference method, the finite element method, the finite volume method, etc.) for thermal analysis, a severe computational burden will be faced, especially when each optimization involves parameter modification and repeated thermal analysis. In recent years, with the rapid development of deep - learning technology, some surrogate models based on convolutional neural networks (CNNs) have been introduced to overcome this obstacle. However, for temperature field prediction on irregular geometric domains (TFP - IGD), traditional CNNs are difficult to be competent, because most CNN models are originally designed to handle regular image data.

To solve this problem, this paper proposes a new surrogate modeling method driven by the synergy of physics and data. This method mainly includes the following steps:

1. **Geometric parameterization**: Use Bezier curves to parameterize the irregular boundaries and generate coordinate transformations from the irregular physical plane to the regular computational plane through the volume - adaptive coordinate mapping technique.
2. **Mesh generation surrogate model**: Use the partial differential equation (PDE) residuals as the loss function to build a physics - driven CNN surrogate model for rapid mesh generation.
3. **Temperature field prediction surrogate model**: Based on the multi - level reduced - order method, propose a data - driven surrogate model aiming to learn the temperature field solutions on the regular computational plane.
4. **Combined model**: Combine the mesh position information provided by the mesh generation surrogate model with the scalar temperature field information provided by the temperature field prediction surrogate model to form an end - to - end surrogate model, from geometric parameters to temperature field prediction on irregular geometric domains.

The experimental results show that this method can significantly improve the prediction accuracy on a relatively small data set while reducing the training time, and has obvious advantages compared with other CNN methods. In conclusion, this paper develops a new surrogate model driven by the synergy of physics and data, which can quickly and accurately predict the temperature field from geometric shape parameters, especially when dealing with problems of irregular geometric domains.