A FEniCS-based programming framework for modeling turbulent flow by the Reynolds-averaged Navier-Stokes equations

Finding an appropriate turbulence model for a given flow case usually calls for extensive experimentation with both models and numerical solution methods. This work presents the design and implementation of a flexible, programmable software framework for assisting with numerical experiments in computational turbulence. The framework targets Reynolds-averaged Navier-Stokes models, discretized by finite element methods. The novel implementation makes use of Python and the FEniCS package, the combination of which leads to compact and reusable code, where model- and solver-specific code resemble closely the mathematical formulation of equations and algorithms. The presented ideas and programming techniques are also applicable to other fields that involve systems of nonlinear partial differential equations. We demonstrate the framework in two applications and investigate the impact of various linearizations on the convergence properties of nonlinear solvers for a Reynolds-averaged Navier-Stokes model. © 2011 Elsevier Ltd.

Occlusive Components Analysis

We study unsupervised learning in a probabilistic generative model for occlusion. The model uses two types of latent variables: one indicates which objects are present in the image, and the other how they are ordered in depth. This depth order then determines how the positions and appearances of the objects present, specified in the model parameters, combine to form the image. We show that the object parameters can be learnt from an unlabelled set of images in which objects occlude one another. Exact maximum-likelihood learning is intractable. However, we show that tractable approximations to Expectation Maximization (EM) can be found if the training images each contain only a small number of objects on average. In numerical experiments it is shown that these approximations recover the correct set of object parameters. Experiments on a novel version of the bars test using colored bars, and experiments on more realistic data, show that the algorithm performs well in extracting the generating causes. Experiments based on the standard bars benchmark test for object learning show that the algorithm performs well in comparison to other recent component extraction approaches. The model and the learning algorithm thus connect research on occlusion with the research field of multiple-causes component extraction methods.

Low-rank optimization for distance matrix completion

This paper addresses the problem of low-rank distance matrix completion. This problem amounts to recover the missing entries of a distance matrix when the dimension of the data embedding space is possibly unknown but small compared to the number of considered data points. The focus is on high-dimensional problems. We recast the considered problem into an optimization problem over the set of low-rank positive semidefinite matrices and propose two efficient algorithms for low-rank distance matrix completion. In addition, we propose a strategy to determine the dimension of the embedding space. The resulting algorithms scale to high-dimensional problems and monotonically converge to a global solution of the problem. Finally, numerical experiments illustrate the good performance of the proposed algorithms on benchmarks. © 2011 IEEE.

Linear regression under fixed-rank constraints: A Riemannian approach

In this paper, we tackle the problem of learning a linear regression model whose parameter is a fixed-rank matrix. We study the Riemannian manifold geometry of the set of fixed-rank matrices and develop efficient line-search algorithms. The proposed algorithms have many applications, scale to high-dimensional problems, enjoy local convergence properties and confer a geometric basis to recent contributions on learning fixed-rank matrices. Numerical experiments on benchmarks suggest that the proposed algorithms compete with the state-of-the-art, and that manifold optimization offers a versatile framework for the design of rank-constrained machine learning algorithms. Copyright 2011 by the author(s)/owner(s).

Fixed-rank matrix factorizations and Riemannian low-rank optimization

Motivated by the problem of learning a linear regression model whose parameter is a large fixed-rank non-symmetric matrix, we consider the optimization of a smooth cost function defined on the set of fixed-rank matrices. We adopt the geometric framework of optimization on Riemannian quotient manifolds. We study the underlying geometries of several well-known fixed-rank matrix factorizations and then exploit the Riemannian quotient geometry of the search space in the design of a class of gradient descent and trust-region algorithms. The proposed algorithms generalize our previous results on fixed-rank symmetric positive semidefinite matrices, apply to a broad range of applications, scale to high-dimensional problems, and confer a geometric basis to recent contributions on the learning of fixed-rank non-symmetric matrices. We make connections with existing algorithms in the context of low-rank matrix completion and discuss the usefulness of the proposed framework. Numerical experiments suggest that the proposed algorithms compete with state-of-the-art algorithms and that manifold optimization offers an effective and versatile framework for the design of machine learning algorithms that learn a fixed-rank matrix. © 2013 Springer-Verlag Berlin Heidelberg.

Numerical simulations of unsteady wet steam flows with non-equilibrium condensation in the nozzle and the steam turbine

Numerical techniques for non-equilibrium condensing flows are presented. Conservation equations for homogeneous gas-liquid two-phase compressible flows are solved by using a finite volume method based on an approximate Riemann solver. The phase change consists of the homogeneous nucleation and growth of existing droplets. Nucleation is computed with the classical Volmer-Frenkel model, corrected for the influence of the droplet temperature being higher than the steam temperature due to latent heat release. For droplet growth, two types of heat transfer model between droplets and the surrounding steam are used: a free molecular flow model and a semi-empirical two-layer model which is deemed to be valid over a wide range of Knudsen number. The computed pressure distribution and Sauter mean droplet diameters in a convergent-divergent (Laval) nozzle are compared with experimental data. Both droplet growth models capture qualitatively the pressure increases due to sudden heat release by the non-equilibrium condensation. However the agreement between computed and experimental pressure distributions is better for the two-layer model. The droplet diameter calculated by this model also agrees well with the experimental value, whereas that predicted by the free molecular model is too small. Condensing flows in a steam turbine cascade are calculated at different Mach numbers and inlet superheat conditions and are compared with experiments. Static pressure traverses downstream from the blade and pressure distributions on the blade surface agree well with experimental results in all cases. Once again, droplet diameters computed with the two-layer model give best agreement with the experiments. Droplet sizes are found to vary across the blade pitch due to the significant variation in expansion rate. Flow patterns including oblique shock waves and condensation-induced pressure increases are also presented and are similar to those shown in the experimental Schlieren photographs. Finally, calculations are presented for periodically unsteady condensing flows in a low expansion rate, convergent-divergent (Laval) nozzle. Depending on the inlet stagnation subcooling, two types of self-excited oscillations appear: a symmetric mode at lower inlet subcooling and an asymmetric mode at higher subcooling. Plots of oscillation frequency versus inlet sub-cooling exhibit a hysteresis loop, in accord with observations made by other researchers for moist air flow. Copyright © 2006 by ASME.

An experimental and numerical study of an oscillating transonic shock wave in a duct

A combined experimental and numerical study of a transonic shock wave in a parallel walled duct subject to downstream pressure perturbations has been conducted. Experiments and simulations have been carried out with a shock strength of M∞ = 1.4 for pressure perturbation frequencies in the range 16-90 Hz. The dynamics of unsteady shock motion and the interaction structure between the unsteady transonic shock wave and the turbulent tunnel floor boundary layer have been investigated. It is found that the (experimentally measured) dynamics of shock motion are generally well predicted by the computational scheme, especially at relatively low (≈ 40 Hz) frequencies. However, at higher frequencies (≈ 90 Hz), some subtle differences between the shock dynamics measured in experiments and those predicted by Computational Fluid Dynamics (CFD) exist. There is evidence from experiments that variations in shock / boundary layer interaction (SBLI) structure caused by shock motion are responsible for a change in the nature of shock dynamics between low and high frequency. In contrast, numerical results at low and high frequencies do not differ significantly and this suggests that the numerical method is not fully capturing the physics of the unsteady flow. Possible reasons for this are considered and a number of areas where CFD is unable to replicate experimental observations are identified. Significantly, CFD predicts changes in SBLI structure due to shock motion that are much too large and this may explain why none of the subtle effects on shock dynamics seen in experiments occur in CFD. Further work developing numerical methods that demonstrate a more realistic sensitivity of SBLI structure to unsteady shock motion is required. Copyright © 2010 by P.J.K. Bruce.

Initial phase modelling in numerical explosion applied to process safety

The utilisation of computational fluid dynamics (CFD) in process safety has increased significantly in recent years. The modelling of accidental explosion via CFD has in many cases replaced the classical Multi Energy and Brake Strehlow methods. The benefits obtained with CFD modelling can be diminished if proper modelling of the initial phase of explosion is neglected. In the early stages of an explosion, the flame propagates in a quasi-laminar regime. Proper modelling of the initial laminar phase is a key aspect in order to predict the peak pressure and the time to peak pressure. The present work suggests a modelling approach for the initial laminar phase in explosion scenarios. Findings are compared with experimental data for two classical explosion test cases which resemble the common features in chemical process areas (confinement and congestion). A detailed analysis of the threshold for the transition from laminar to turbulent regime is also carried out. The modelling is implemented in a fully 3D Navier-Stokes compressible formulation. Combustion is treated using a laminar flamelet approach based on the Bray, Moss and Libby (BML) formulation. A novel modified porosity approach developed for the unstructured solver is also considered. Results agree satisfactorily with experiments and the modelling is found to be robust. © 2013 The Institution of Chemical Engineers.

Direct numerical simulation of turbulent flows over superhydrophobic surfaces with gas pockets using linearized boundary conditions

Superhydrophobic surfaces are shown to be effective for surface drag reduction under laminar regime by both experiments and simulations (see for example, Ou and Rothstein, Phys. Fluids 17:103606, 2005). However, such drag reduction for fully developed turbulent flow maintaining the Cassie-Baxter state remains an open problem due to high shear rates and flow unsteadiness of turbulent boundary layer. Our work aims to develop an understanding of mechanisms leading to interface breaking and loss of gas pockets due to interactions with turbulent boundary layers. We take advantage of direct numerical simulation of turbulence with slip and no-slip patterned boundary conditions mimicking the superhydrophobic surface. In addition, we capture the dynamics of gas-water interface, by deriving a proper linearized boundary condition taking into account the surface tension of the interface and kinematic matching of interface deformation and normal velocity conditions on the wall. We will show results from our simulations predicting the dynamical behavior of gas pocket interfaces over a wide range of dimensionless surface tensions.

Joint experimental and numerical approach to three-dimensional shock control bump research

Previous studies of transonic shock control bumps have often been either numerical or experimental. Comparisons between the two have been hampered by the limitations of either approach. The present work aims to bridge the gap between computational fluid dynamics and experiment by planning a joint approach from the outset. This enables high-quality validation data to be produced and ensures that the conclusions of either aspect of the study are directly relevant to the application. Experiments conducted with bumps mounted on the floor of a blowdown tunnel were modified to include an additional postshock adverse pressure gradient through the use of a diffuser as well as introducing boundary-layer suction ahead of the test section to enable the in-flow boundary layer to be manipulated. This has the advantage of being an inexpensive and highly repeatable method. Computations were performed on a standard airfoil model, with the flight conditions as free parameters. The experimental and computational setups were then tuned to produce baseline conditions that agree well, enabling confidence that the experimental conclusions are relevant. The methods are then applied to two different shock control bumps: a smoothly contoured bump, representative of previous studies, and a novel extended geometry featuring a continuously widening tail, which spans the wind-tunnel width at the rear of the bump. Comparison between the computational and experimental results for the contour bump showed good agreement both with respect to the flow structures and quantitative analysis of the boundary-layer parameters. It was seen that combining the experimental and numerical data could provide valuable insight into the flow physics, which would not generally be possible for a one-sided approach. The experiments and computational fluid dynamics were also seen to agree well for the extended bump geometry, providing evidence that, even though thebumpinteracts directly with the wind-tunnel walls, it was still possible to observe the key flow physics. The joint approach is thus suitable even for wider bump geometries. Copyright © 2013 by S. P. Colliss, H. Babinsky, K. Nubler, and T. Lutz. Published by the American Institute of Aeronautics and Astronautics, Inc.