Guang Lin

Adaptive ANOVA decomposition of stochastic incompressible and compressible flows

Realistic representation of stochastic inputs associated with various sources of uncertainty in the simulation of fluid flows leads to high dimensional representations that are computationally prohibitive. We investigate the use of adaptive ANOVA decomposition as an effective dimension-reduction technique in modeling steady incompressible and compressible flows with nominal dimension of random space up to 100. We present three different adaptivity criteria and compare the adaptive ANOVA method against sparse grid, Monte Carlo and quasi-Monte Carlo methods to evaluate its relative efficiency and accuracy. For the incompressible flow problem, the effect of random temperature boundary conditions (modeled as high-dimensional stochastic processes) on the Nusselt number is investigated for different values of correlation length. For the compressible flow, the effects of random geometric perturbations (simulating random roughness) on the scattering of a strong shock wave is investigated both analytically and numerically. A probabilistic collocation method is combined with adaptive ANOVA to obtain both incompressible and compressible flow solutions. We demonstrate that for both cases even draconian truncations of the ANOVA expansion lead to accurate solutions with a speed-up factor of three orders of magnitude compared to Monte Carlo and at least one order of magnitude compared to sparse grids for comparable accuracy.

Predicting shock dynamics in the presence of uncertainties

We revisit the classical aerodynamics problem of supersonic flow past a wedge but subject to random inflow fluctuations or random wedge oscillations around its apex. We first obtain analytical solutions for the inviscid flow, and subsequently we perform stochastic simulations treating randomness both as a steady as well as a time-dependent process. We use a multi-element generalized polynomial chaos (ME-gPC) method to solve the two-dimensional stochastic Euler equations. A Galerkin projection is employed in the random space while WENO discretization is used in physical space. A key issue is the characteristic flux decomposition in the stochastic framework for which we propose different approaches. The results we present show that the variance of the location of perturbed shock grows quadratically with the distance from the wedge apex for steady randomness. However, for a time-dependent random process the dependence is quadratic only close to the apex and linear for larger distances. The multi-element version of polynomial chaos seems to be more effective and more efficient in stochastic simulations of supersonic flows compared to the global polynomial chaos method.

Uncertainty quantification via random domain decomposition and probabilistic collocation on sparse grids

Quantitative predictions of the behavior of many deterministic systems are uncertain due to ubiquitous heterogeneity and insufficient characterization by data. We present a computational approach to quantify predictive uncertainty in complex phenomena, which is modeled by (partial) differential equations with uncertain parameters exhibiting multi-scale variability. The approach is motivated by flow in random composites whose internal architecture (spatial arrangement of constitutive materials) and spatial variability of properties of each material are both uncertain. The proposed two-scale framework combines a random domain decomposition (RDD) and a probabilistic collocation method (PCM) on sparse grids to quantify these two sources of uncertainty, respectively. The use of sparse grid points significantly reduces the overall computational cost, especially for random processes with small correlation lengths. A series of one-, two-, and three-dimensional computational examples demonstrate that the combined RDD-PCM approach yields efficient, robust and non-intrusive approximations for the statistics of diffusion in random composites.

Multi-output separable Gaussian process: Towards an efficient, fully Bayesian paradigm for uncertainty quantification

Computer codes simulating physical systems usually have responses that consist of a set of distinct outputs (e.g., velocity and pressure) that evolve also in space and time and depend on many unknown input parameters (e.g., physical constants, initial/boundary conditions, etc.). Furthermore, essential engineering procedures such as uncertainty quantification, inverse problems or design are notoriously difficult to carry out mostly due to the limited simulations available. The aim of this work is to introduce a fully Bayesian approach for treating these problems which accounts for the uncertainty induced by the finite number of observations. Our model is built on a multi-dimensional Gaussian process that explicitly treats correlations between distinct output variables as well as space and/or time. The proper use of a separable covariance function enables us to describe the huge covariance matrix as a Kronecker product of smaller matrices leading to efficient algorithms for carrying out inference and predictions. The novelty of this work, is the recognition that the Gaussian process model defines a posterior probability measure on the function space of possible surrogates for the computer code and the derivation of an algorithmic procedure that allows us to sample it efficiently. We demonstrate how the scheme can be used in uncertainty quantification tasks in order to obtain error bars for the statistics of interest that account for the finite number of observations.

Compressive sensing based machine learning strategy for characterizing the flow around a cylinder with limited pressure measurements

Compressive sensing is used to determine the flow characteristics around a cylinder (Reynolds number and pressure/flow field) from a sparse number of pressure measurements on the cylinder. Using a supervised machine learning strategy, library elements encoding the dimensionally reduced dynamics are computed for various Reynolds numbers. Convex L1 optimization is then used with a limited number of pressure measurements on the cylinder to reconstruct, or decode, the full pressure field and the resulting flow field around the cylinder. Aside from the highly turbulent regime (large Reynolds number) where only the Reynolds number can be identified, accurate reconstruction of the pressure field and Reynolds number is achieved. The proposed data-driven strategy thus achieves encoding of the fluid dynamics using the L2 norm, and robust decoding (flow field reconstruction) using the sparsity promoting L1 norm.

Point-wise hierarchical reconstruction for discontinuous Galerkin and finite volume methods for solving conservation laws

We develop a new hierarchical reconstruction (HR) method [17,28] for limiting solutions of the discontinuous Galerkin and finite volume methods up to fourth order of accuracy without local characteristic decomposition for solving hyperbolic nonlinear conservation laws on triangular meshes. The new HR utilizes a set of point values when evaluating polynomials and remainders on neighboring cells, extending the technique introduced in Hu, Li and Tang [9]. The point-wise HR simplifies the implementation of the previous HR method which requires integration over neighboring cells and makes HR easier to extend to arbitrary meshes. We prove that the new point-wise HR method keeps the order of accuracy of the approximation polynomials. Numerical computations for scalar and system of nonlinear hyperbolic equations are performed on two-dimensional triangular meshes. We demonstrate that the new hierarchical reconstruction generates essentially non-oscillatory solutions for schemes up to fourth order on triangular meshes.

Numerical Studies of Three-dimensional Stochastic Darcy's Equation and Stochastic Advection-Diffusion-Dispersion Equation

Solute transport in randomly heterogeneous porous media is commonly described by stochastic flow and advection-dispersion equations with a random hydraulic conductivity field. The statistical distribution of conductivity of engineered and naturally occurring porous material can vary, depending on its origin. We describe solutions of a three-dimensional stochastic advection-dispersion equation using a probabilistic collocation method (PCM) on sparse grids for several distributions of hydraulic conductivity. Three random distributions of log hydraulic conductivity are considered: uniform, Gaussian, and truncated Gaussian (beta). Log hydraulic conductivity is represented by a Karhunen-Loève (K-L) decomposition as a second-order random process with an exponential covariance function. The convergence of PCM has been demonstrated. It appears that the accuracy in both the mean and the standard deviation of PCM solutions can be improved by using the Jacobi-chaos representing the truncated Gaussian distribution rather than the Hermite-chaos for the Gaussian distribution. The effect of type of distribution and parameters such as the variance and correlation length of log hydraulic conductivity and dispersion coefficient on leading moments of the advection velocity and solute concentration was investigated.

Stochastic Computational Fluid Mechanics

Stochastic simulations in computational fluid dynamics let researchers model uncertainties beyond numerical discretization errors. The authors present examples of stochastic simulations of compressible and incompressible flows and provide analytical solutions for verifying these newly emerging methods for stochastic modeling

Dynamic-Feature Extraction, Attribution, and Reconstruction (DEAR) Method for Power System Model Reduction

In interconnected power systems, dynamic model reduction can be applied to generators outside the area of interest (i.e., study area) to reduce the computational cost associated with transient stability studies. This paper presents a method of deriving the reduced dynamic model of the external area based on dynamic response measurements. The method consists of three steps, namely dynamic-feature extraction, attribution, and reconstruction (DEAR). In this method, a feature extraction technique, such as singular value decomposition (SVD), is applied to the measured generator dynamics after a disturbance. Characteristic generators are then identified in the feature attribution step for matching the extracted dynamic features with the highest similarity, forming a suboptimal “basis” of system dynamics. In the reconstruction step, generator state variables such as rotor angles and voltage magnitudes are approximated with a linear combination of the characteristic generators, resulting in a quasi-nonlinear reduced model of the original system. The network model is unchanged in the DEAR method. Tests on several IEEE standard systems show that the proposed method yields better reduction ratio and response errors than the traditional coherency based reduction methods.

Enhancing sparsity of Hermite polynomial expansions by iterative rotations

Compressive sensing has become a powerful addition to uncertainty quantification in recent years. This paper identifies new bases for random variables through linear mappings such that the representation of the quantity of interest is more sparse with new basis functions associated with the new random variables. This sparsity increases both the efficiency and accuracy of the compressive sensing-based uncertainty quantification method. Specifically, we consider rotation-based linear mappings which are determined iteratively for Hermite polynomial expansions. We demonstrate the effectiveness of the new method with applications in solving stochastic partial differential equations and high-dimensional ( O ( 100 ) ) problems.