Search NASA⌕ Search

Engineering topics

Tartakovsky, Alexandre

Publications and source records attributed to Tartakovsky, Alexandre.

Differentiable modelling to unify machine learning and physical models for geosciences

Process-based modelling offers interpretability and physical consistency in many domains of geosciences but struggles to leverage large datasets efficiently. Machine-learning methods, especially deep networks, have strong predictive skills yet are unable to answer specific scientific questions. Here, in this Perspective, we explore differentiable modelling as a pathway to dissolve the perceived barrier between process-based modelling and machine learning in the geosciences and demonstrate its potential with examples from hydrological modelling. ‘Differentiable’ refers to accurately and efficiently calculating gradients with respect to model variables or parameters, enabling the discovery of high-dimensional unknown relationships. Differentiable modelling involves connecting (flexible amounts of) prior physical knowledge to neural networks, pushing the boundary of physics-informed machine learning. It offers better interpretability, generalizability, and extrapolation capabilities than purely data-driven machine learning, achieving a similar level of accuracy while requiring less training data. Additionally, the performance and efficiency of differentiable models scale well with increasing data volumes. Under data-scarce scenarios, differentiable models have outperformed machine-learning models in producing short-term dynamics and decadal-scale trends owing to the imposed physical constraints. Differentiable modelling approaches are primed to enable geoscientists to ask questions, test hypotheses, and discover unrecognized physical relationships. Future work should address computational challenges, reduce uncertainty, and verify the physical significance of outputs.

58 GEOSCIENCES↗

Enhanced physics-constrained deep neural networks for modeling vanadium redox flow battery

Numerical simulation has become indispensable in advancing cost-effective process optimization and control of flow batteries. We propose an enhanced version of the physics-constrained deep neural network (PCDNN) approach to provide high-accuracy voltage predictions in the vanadium redox flow batteries (VRFBs). The purpose of the PCDNN approach is to enforce the physics-based zero-dimensional (0D) VRFB model in a neural network to assure model generalization for various battery operation conditions. However, limited by the simplifications of the 0D model, the PCDNN cannot capture sharp voltage changes in the extreme SOC regions. To improve the accuracy of voltage prediction at extreme ranges, we introduce a second (enhanced) DNN to mitigate the prediction errors carried from the 0D model itself and call the resulting approach enhanced PCDNN (ePCDNN). By comparing with experimental data, we demonstrate that the ePCDNN approach can accurately capture the voltage response throughout the charge–discharge cycle, including the tail region of the voltage discharge curve. The loss function for training the ePCDNN is designed to be flexible by adjusting the weights of the physics-constrained DNN and the enhanced DNN. In conclusion, this allows the ePCDNN framework to be transferable to battery systems with variable physical model fidelity.

25 ENERGY STORAGE↗

Discovering new governing equations using ML

A hallmark of the scientific process since the time of Newton has been the derivation of mathematical equations meant to capture relationships between observables. As the field of mathematical modeling evolved, practitioners specifically emphasized mathematical formulations that were predictive, generalizable, and interpretable. Machine learning’s ability to interrogate complex processes is particularly useful for the analysis of highly heterogeneous, anisotropic materials where idealized descriptions often fail. As we move into this new era, we anticipate the need to leverage machine learning to aid scientists in extracting meaningful, but yet sometimes elusive, relationships between observed quantities.

97 MATHEMATICS AND COMPUTING↗

Comparison of Surface Tension Generation Methods in Smoothed Particle Hydrodynamics for Dynamic Systems

Developing robust numerical models of dynamic surface tension dominated multiphase systems is an ongoing challenge, especially in scenarios with large density and viscosity ratios. This is critical to the design and understanding of various physical and engineering systems, such as fluidized beds, fuel injectors, and drug delivery schemes. Much of the computational work in surface tension dominated multiphase flows has employed the continuum surface force method (CSF) of Brackbill et al. [1], which recasts surface tension from a surface force to a volumetric force that can be imposed in the vicinity of an interface. The CSF method produces accurate results across a variety of systems, however it relies on the identication of surface normals, which can be unreliable under certain conditions. Alternative methods of simulating surface tension have been proposed. Here the advantages and disadvantages of the CSF method in comparison to a pairwise forces (PF) method proposed by Tartakovsky and Panchenko [2] are explored. The CSF and PF methods are used in a smoothed particle hydrodynamics (SPH) framework to model dynamic systems. Results are compared to existing test cases from the literature and to analytic solutions derived from fundamental normal mode behavior of bubbles and droplets. A physical system for which the PF method is more stable and physically appropriate than the CSF method is identied.

smoothed particle hydrodynamics, multiphase flow↗

Sliced-Inverse-Regression--Aided Rotated Compressive Sensing Method for Uncertainty Quantification

Compressive-sensing-based uncertainty quantification methods have become a pow- erful tool for problems with limited data. In this paper, we use the sliced inverse regression (SIR) method to provide an initial guess for the alternating direction method, which is used to enhance sparsity of the Hermite polynomial expansion of stochastic state variables. The sparsity improvement increases both the efficiency and accuracy of the compressive-sensing-based uncertainty quantification method. We demonstrate that the initial guess from SIR is more suitable for the cases when the available data is very limited (Algorithm 4). We also propose another algorithm (Algorithm 5) that performs dimension reduction first with sliced inverse regression, then constructs a Hermite polynomial expansion of the reduced model. This method allows us to approximate the statistics accurately with even less available data. Both methods are non-intrusive and require no a priori information of the sparsity of the system. We demonstrate the effectiveness of these two methods (Algorithms 4 and 5) using problems with up to 500 random dimensions.

Yang, Xiu↗

Smoothed particle hydrodynamics model for Landau-Lifshitz-Navier-Stokes and advection-diffusion equations

We propose a novel smoothed particle hydrodynamics (SPH) discretization of the fully coupled Landau-Lifshitz-Navier-Stokes (LLNS) and stochastic advection-diffusion equations. The accuracy of the SPH solution of the LLNS equations is demonstrated by comparing the scaling of velocity variance and the self-diffusion coefficient with kinetic temperature and particle mass obtained from the SPH simulations and analytical solutions. The spatial covariance of pressure and velocity fluctuations is found to be in a good agreement with theoretical models. To validate the accuracy of the SPH method for coupled LLNS and advection-diffusion equations, we simulate the interface between two miscible fluids. We study formation of the so-called “giant fluctuations” of the front between light and heavy fluids with and without gravity, where the light fluid lies on the top of the heavy fluid. We find that the power spectra of the simulated concentration field are in good agreement with the experiments and analytical solutions. In the absence of gravity, the power spectra decay as the power −4 of the wavenumber—except for small wavenumbers that diverge from this power law behavior due to the effect of finite domain size. Gravity suppresses the fluctuations, resulting in much weaker dependence of the power spectra on the wavenumber. Finally, the model is used to study the effect of thermal fluctuation on the Rayleigh-Taylor instability, an unstable dynamics of the front between a heavy fluid overlaying a light fluid. The front dynamics is shown to agree well with the analytical solutions.

37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CH↗

Discrete models of fluids: spatial averaging, closure and model reduction

We consider semidiscrete ODE models of single-phase fluids and two-fluid mixtures. In the presence of multiple fine-scale heterogeneities, the size of these ODE systems can be very large. Spatial averaging is then a useful tool for reducing computational complexity of the problem. The averages satisfy exact balance equations of mass, momentum, and energy. These equations do not form a satisfactory continuum model because evaluation of stress and heat flux requires solving the underlying ODEs. To produce continuum equations that can be simulated without resolving microscale dynamics, we recently proposed a closure method based on the use of regularized deconvolution. Here we continue the investigation of deconvolution closure with the long term objective of developing consistent computational upscaling for multiphase particle methods. The structure of the fine-scale particle solvers is reminiscent of molecular dynamics. For this reason we use nonlinear averaging introduced for atomistic systems by Noll, Hardy, and Murdoch-Bedeaux. We also consider a simpler linear averaging originally developed in large eddy simulation of turbulence. We present several simple but representative examples of spatially averaged ODEs, where the closure error can be analyzed. Based on this analysis we suggest a general strategy for reducing the relative error of approximate closure. For problems with periodic highly oscillatory material parameters we propose a spectral boosting technique that augments the standard deconvolution and helps to correctly account for dispersion effects. We also conduct several numerical experiments, one of which is a complete mesoscale simulation of a stratified two-fluid flow in a channel. In this simulation, the operation count per coarse time step scales sublinearly with the number of particles.

Multiscale modeling, discrete models, fluid flow↗

Modeling and Simulation of Pore Scale Multiphase Fluid Flow and Reactive Transport in Fractured and Porous Media

In the subsurface fluids play a critical role by transporting dissolved minerals, colloids and contaminants (sometimes over long distances), by mediating dissolution and precipitation processes and enabling chemical transformations in solution and at mineral surfaces. Although the complex geometries of fracture apertures, fracture networks and pore spaces may make it difficult to accurately predict fluid flow in saturated (single-phase) subsurface systems, well developed methods are available. The simulation of multiphase fluid flow in the subsurface is much more challenging because of the large density and/or viscosity ratios found in important applications (water/air in the vadose zone, water/oil, water/gas, gas/oil and water/oil/gas in oil reservoirs, water/air/non-aqueous phase liquids (NAPL) in contaminated vadose zone systems and gas/molten rock in volcanic systems, for example). In addition, the complex behavior of fluid-fluid-solid contact lines, and its impact on dynamic contact angles, must also be taken into account, and coupled with the fluid flow. Pore network models and simple statistical physics based models such as the invasion percolation and diffusion-limited aggregation models have been used quite extensively. However, these models for multiphase fluid flow are based on simplified models for pore space geometries and simplified physics. Other methods such a lattice Boltzmann and lattice gas models, molecular dynamics, Monte Carlo methods, and particle methods such as dissipative particle dynamics and smoothed particle hydrodynamics are based more firmly on first principles, and they do not require simplified pore and/or fracture geometries. However, they are less (in some cases very much less) computationally efficient that pore network and statistical physics models. Recently a combination of continuum computation fluid dynamics, fluid-fluid interface tracking or capturing and simple models for the dependence of contact angles on fluid velocity at the contact line has been used to simulate multiphase fluid flow in fracture apertures, fracture networks and pore spaces. Fundamental conservation principles - conservation of momentum, and conservation of mass (or conservation of volume for incompressible fluids) and conservation of energy, as well as symmetries (Galilean invariance and isotropy) are central to the physics of fluids and the models used to simulate them. In molecular and mesoscale models observance of these conservation principles and symmetries at the microscopic level leads to macroscopic fluid dynamics that can be represented by the Navier Stokes equation. The remarkable fact that the flow of all simpe fluids, irrespective of their chemical nature, can be described by the Navier-Stokes equation is a result of these conservation principles and symmetries acting on the molecular level.

99 GENERAL AND MISCELLANEOUS↗

Diffuse-Interface Model for Smoothed Particle Hydrodynamics

Diffuse-interface theory provides a foundation for the modeling and simulation of microstructure evolution in a very wide range of materials, and for the tracking/capturing of dynamic interfaces between different materials on larger scales. Smoothed particle hydrodynamics (SPH) is also widely used to simulate fluids and solids that are subjected to large deformations and have complex dynamic boundaries and/or interfaces, but no explicit interface tracking/capturing is required, even when topological changes such as fragmentation and coalescence occur, because of its Lagrangian particle nature. Here we developed an SPH model for two component single phase fluids that is based on diffuse-interface theory. In the model, the interface has a finite thickness and a surface tension that depend on the coefficient, k, of the gradient contribution to the Helmholtz free energy functional and the density dependent homogeneous free energy. In this model, there is no need to locate the surface (or interface) or to compute the curvature at and near the interface. One- and two-dimensional SPH simulations were used to validate the model.

71 CLASSICAL AND QUANTUM MECHANICS, GENERAL PHYSIC↗