Data-driven Modeling of Parameterized Nonlinear Fluid Dynamical Systems with a Dynamics-embedded Conditional Generative Adversarial Network

Abdolvahhab Rostamijavanani,Shanwu Li,Yongchao Yang
2024-12-24
Abstract:This work presents a data-driven solution to accurately predict parameterized nonlinear fluid dynamical systems using a dynamics-generator conditional GAN (Dyn-cGAN) as a surrogate model. The Dyn-cGAN includes a dynamics block within a modified conditional GAN, enabling the simultaneous identification of temporal dynamics and their dependence on system parameters. The learned Dyn-cGAN model takes into account the system parameters to predict the flow fields of the system accurately. We evaluate the effectiveness and limitations of the developed Dyn-cGAN through numerical studies of various parameterized nonlinear fluid dynamical systems, including flow over a cylinder and a 2-D cavity problem, with different Reynolds numbers. Furthermore, we examine how Reynolds number affects the accuracy of the predictions for both case studies. Additionally, we investigate the impact of the number of time steps involved in the process of dynamics block training on the accuracy of predictions, and we find that an optimal value exists based on errors and mutual information relative to the ground truth.
Machine Learning,Chaotic Dynamics
What problem does this paper attempt to address?
The problem that this paper attempts to solve is: how to accurately predict the behavior of parameterized non - linear hydrodynamic systems, especially when the parameters of these systems (such as Reynolds number) change. Although traditional computational fluid dynamics (CFD) methods can provide high - precision results, their computational cost is high and time - consuming, and it is difficult to quickly iterate and explore the design space. In addition, existing data - driven models face challenges when dealing with parameterized non - linear systems because they need to simultaneously identify the time - dynamics and their dependence on system parameters from the measured data. To solve these problems, the author proposes an improved model based on the generative adversarial network (GAN) - the conditional generative adversarial network with dynamics embedding (Dyn - cGAN). By introducing a dynamics module into the modified conditional GAN, Dyn - cGAN can simultaneously identify the time - dynamics and their dependence on system parameters, thereby predicting the flow field more accurately. ### Specific problem description 1. **Computational resources and time cost**: - Traditional CFD simulations require a large amount of computational resources and time, especially in the design stage where parameters change frequently. 2. **Limitations of existing data - driven models**: - Existing data - driven models are mainly for non - linear systems with fixed parameters, and it is difficult to handle the time - dynamics of parameterized non - linear systems and their dependence on parameters. 3. **Lack of physical interpretability**: - Most existing models cannot directly describe the internal time - evolution process, resulting in the lack of interpretability of physical phenomena in the model and difficulty in generalizing to other scenarios. ### Solution The Dyn - cGAN model proposed by the author solves the above problems in the following ways: - **Dynamics module embedding**: A dynamics module is introduced into the generator of GAN, enabling the model to capture the time - dynamics and their dependence on system parameters. - **Conditional generative adversarial network**: Using the conditional GAN framework, the model can generate corresponding flow field predictions according to different system parameters. - **Recursive prediction**: By recursively predicting the flow field in multiple time steps through the dynamics module, the accuracy and generalization ability of the prediction are improved. ### Experimental verification To evaluate the effectiveness and limitations of Dyn - cGAN, the author conducted numerical studies, including flow around a cylinder and two - dimensional cavity problems, and examined the prediction accuracy under different Reynolds numbers. The experimental results show that Dyn - cGAN performs well in predicting the flow field in time and space, especially when dealing with parameterized non - linear systems. ### Key formulas The Navier - Stokes equation is used to describe fluid motion: \[ \frac{\partial u}{\partial t}+u\cdot\nabla u = -\frac{1}{\rho}\nabla p+\nu\nabla^{2}u + f \] where \(u\) is the velocity field, \(\rho\) is the fluid density, \(p\) is the pressure, \(\nu\) is the dynamic viscosity of the fluid, and \(f\) is the source term. The state - transfer equation of the dynamics module: \[ \phi_{t + 1}=A(\phi_{t}) \] where \(\phi_{t}\) is the latent coordinate at time step \(t\), and \(A\) is a non - linear transformation function. The loss function includes prediction loss and adversarial loss: \[ \lambda_{G}=\beta_{1}\lambda_{G_{\text{pred}}}+\beta_{2}\lambda_{G_{\text{adv}}} \] where: \[ \lambda_{G_{\text{pred}}}=\frac{1}{n}\sum_{i = 1}^{n}\|G(t;P_{i}^{\text{sim}})-s(t;P_{i}^{\text{sim}})\|_{\text{MSE}} \] \[ \lambda_{G_{\text{adv}}}=-\frac{1}{n}\sum_{i = 1}^{n}\log D(G(t;P_{i}^{\text{sim}})) \]