Physics-guided deep reinforcement learning for flow field denoising

Abstract

A multi-agent deep reinforcement learning (DRL)-based model is presented in this study to reconstruct flow fields from noisy data. A combination of the reinforcement learning with pixel-wise rewards (PixelRL), physical constraints represented by the momentum equation and the pressure Poisson equation and the known boundary conditions is utilised to build a physics-guided deep reinforcement learning (PGDRL) model that can be trained without the target training data. In the PGDRL model, each agent corresponds to a point in the flow field and it learns an optimal strategy for choosing pre-defined actions. The proposed model is efficient considering the visualisation of the action map and the interpretation of the model performance. The performance of the model is tested by utilising synthetic direct numerical simulation (DNS)-based noisy data and experimental data obtained by particle image velocimetry (PIV). Qualitative and quantitative results show that the model can reconstruct the flow fields and reproduce the statistics and the spectral content with commendable accuracy. These results demonstrate that the combination of DRL-based models and the known physics of the flow fields can potentially help solve complex flow reconstruction problems, which can result in a remarkable reduction in the experimental and computational costs.

Deep Reinforcement Learning-Augmented Spalart–Allmaras Turbulence Model: Application to a Turbulent Round Jet Flow

“…In the present work, we use the local surrounding of a point as input data, comparable to the methods of Xu et al [20] and Zhou et al [21]. We combine this approach with the concept of multi-agent reinforcement learning that was also used by Novati et al [24] and Yousif et al [46]. Our method is configured in an open-loop design that is comparable to those of Viquerat et al [58] and Ghraieb et al [59].…”

“…Other than Novati et al [24], they surpass the performance of conventional subgrid-scale models by only relying on local variables utilizing a convolutional neural network architecture. Yousif et al [46] apply a multi-agent DRL-based model to reconstruct flow fields from noisy data. They incorporate the momentum equations and the pressure Poisson equation and, therefore, create a physics-guided DRL model.…”