Convolutional Neural Networks for Steady Flow Approximation

Abstract

In aerodynamics related design, analysis and optimization problems, flow fields are simulated using computational fluid dynamics (CFD) solvers. However, CFD simulation is usually a computationally expensive, memory demanding and time consuming iterative process. These drawbacks of CFD limit opportunities for design space exploration and forbid interactive design. We propose a general and flexible approximation model for real-time prediction of steady non-uniform laminar flow in a 2D and 3D domain based on convolutional neural networks (CNNs). We explored alternatives for the geometry representation and the network architecture of CNN. We show that convolution neural networks can estimate the velocity field two orders of magnitude faster than a GPU-accelerated CFD solver or four orders of magnitude faster than a CPU-based CFD solver at a cost of a low error rate. Our results show that we can reduce the average time to generate a fully converged CFD result from 82 seconds on a single core CPU to 7 milliseconds by leveraging both CNN and GPU at the cost of a low 1.98% to 2.69% error rate. This approach can provide immediate feedback for real-time design iterations at the early stage of design. Compared with existing approximation models in the aerodynamics domain, CNN enables an efficient estimation for the entire velocity field. Furthermore, designers and engineers can directly apply the model in their design space exploration algorithms without training lower-dimensional surrogate models.