Physics-based surrogate modeling of parameterized PDEs for optimization and uncertainty analysis

This paper presents physics-based surrogate modeling algorithms for systems governed by parameterized partial differential equations (PDEs) commonly encountered in design optimization and uncertainty analysis. We first outline unsupervised learning approaches that leverage advances in the machine learning literature for a meshfree solution of PDEs. Subsequently, we propose continuum and discrete formulations for systems governed by parameterized steady-state PDEs. We consider the case of both deterministically and randomly parameterized systems. The basic idea is to embody the design variables or uncertain parameters in additional dimensions of the governing PDEs along with the spatial coordinates. We show that the undetermined parameters of the surrogate model can be estimated by minimizing a physics-based objective function derived using a multidimensional least-squares collocation or the Bubnov-Galerkin scheme. This potentially allows us to construct surrogate models without using data from computer experiments on a deterministic analysis code. Finally, we also outline an extension of the present appraoch to directly approximate the density function of random algebraic equations.

[1]  Dimitrios I. Fotiadis,et al.  Artificial neural networks for solving ordinary and partial differential equations , 1997, IEEE Trans. Neural Networks.

[2]  P. Nair.  On the theoretical foundations of stochastic reduced basis methods , 2001 .

[3]  Andrew R. Barron,et al.  Universal approximation bounds for superpositions of a sigmoidal function , 1993, IEEE Trans. Inf. Theory.

[4]  U. Kirsch.  Reduced basis approximations of structural displacements for optimaldesign , 1991 .

[5]  Prasanth B. Nair,et al.  Equivalence Between the Combined Approximations Technique and Krylov Subspace Methods , 2002 .

[6]  Carsten Franke,et al.  Solving partial differential equations by collocation using radial basis functions , 1998, Appl. Math. Comput..

[7]  Dimitris G. Papageorgiou,et al.  Neural-network methods for boundary value problems with irregular boundaries , 2000, IEEE Trans. Neural Networks Learn. Syst..

[8]  Juan J. Alonso,et al.  Airfoil design optimization using reduced order models based on proper orthogonal decomposition , 2000 .

[9]  Thomas J. Santner,et al.  Design and analysis of computer experiments , 1998 .

[10]  Mark A Fleming,et al.  Meshless methods: An overview and recent developments , 1996 .

[11]  A. Keane,et al.  Improved first-order approximation of eigenvalues and eigenvectors , 1998 .

[12]  E. Jaynes.  On the rationale of maximum-entropy methods , 1982, Proceedings of the IEEE.

[13]  D. Mackay,et al.  Bayesian methods for adaptive models , 1992 .

[14]  Andy J. Keane,et al.  Some Greedy Algorithms for Sparse Nonlinear Regression , 2001, International Conference on Machine Learning.

[15]  L. Watson,et al.  Reasonable Design Space Approach to Response Surface Approximation , 1999 .

[16]  B. Spencer,et al.  On the numerical solution of the Fokker-Planck equation for nonlinear stochastic systems , 1993 .

[17]  Christopher M. Bishop,et al.  Neural networks for pattern recognition , 1995 .

[18]  Vladimir Vapnik,et al.  The Nature of Statistical Learning , 1995 .

[19]  Yi Lin.  Tensor product space ANOVA models , 2000 .

[20]  E. Kansa.  Multiquadrics—A scattered data approximation scheme with applications to computational fluid-dynamics—I surface approximations and partial derivative estimates , 1990 .