17 July 2017 in Archives of Computational Methods in Engineering
Metamodeling, the science of modeling functions observed at a finite number of points, benefits from all auxiliary information it can account for. Function gradients are a common auxiliary information and are useful for predicting functions with locally changing behaviors. This article is a review of the main metamodels that use function gradients in addition to function values. The goal of the article is to give the reader both an overview of the principles involved in gradient-enhanced metamodels while also providing insightful formulations. The following metamodels have gradient-enhanced versions in the literature and are reviewed here: classical, weighted and moving least squares, Shepard weighting functions, and the kernel-based methods that are radial basis functions, kriging and support vector machines. The methods are set in a common framework of linear combinations between a priori chosen functions and coefficients that depend on the observations. The characteristics common to all kernel-based approaches are underlined. A new \(\nu \)-GSVR metamodel which uses gradients is given. Numerical comparisons of the metamodels are carried out for approximating analytical test functions. The experiments are replicable, as they are performed with an opensource available toolbox. The results indicate that there is a trade-off between the better computing time of least squares methods and the larger versatility of kernel-based approaches.
28 November 2022 in Structural and Multidisciplinary Optimization
26 April 2012 in Computational Mechanics
In the course of designing structural assemblies, performing a full optimization is very expensive in terms of computation time. In order or reduce this cost, we propose a multilevel model optimization approach. This paper lays the foundations of this strategy by presenting a method for constructing an approximation of an objective function. This approach consists in coupling a multiparametric mechanical strategy based on the LATIN method with a gradient-based metamodel called a cokriging metamodel. The main difficulty is to build an accurate approximation while keeping the computation cost low. Following an introduction to multiparametric and cokriging strategies, the performance of kriging and cokriging models is studied using one- and two-dimensional analytical functions; then, the performance of metamodels built from mechanical responses provided by the multiparametric strategy is analyzed based on two mechanical test examples.
10 March 2023 in IEEE Transactions on Magnetics
Density-based topology optimization methodologies in magnetostatics interpolate either the magnetic permeability μ or the magnetic reluctivity ν to define intermediate materials and compute the problem sensitivities. These choices may lack physical interpretation and are not suited to model realistic permanent magnets. This work proposes the interpolation of the vector magnetic polarization m as a more general alternative, which can model any magnetic material. It is then applied to the multi-material topology optimization of the rotor of a permanent magnet synchronous machine. The case of an ideal permanent magnet with a constant magnetic polarization is compared to that of AlNiCo. The results show that considering the non-linearity of the permanent magnet can significantly impact the optimized geometry.
30 April 2013 in Computers & Structures
This paper deals with the building of a gradient-based metamodel using a dedicated strategy for solving structural assemblies problems. This work is the first part of a two-levels global optimization strategy. The general objective is to reduce computation costs; here, we focus on the costs which are associated with the generation of the metamodel. Our goal is achieved through the introduction of two main elements: what we call a “multiparametric strategy” based on the LATIN method, which reduces the computation costs when the parameters vary, and the use of a cokriging metamodel taking gradients into account. Several examples illustrate the efficiency of these two elements.
6 October 2017 in Comptes Rendus Mécanique
In this paper, we characterize random eigenspaces with a non-intrusive method based on the decoupling of random eigenvalues from their corresponding random eigenvectors. This method allows us to estimate the first statistical moments of the random eigenvalues of the system with a reduced number of deterministic finite element computations. The originality of this work is to adapt the method used to estimate each random eigenvalue depending on a global accuracy requirement. This allows us to ensure a minimal computational cost. The stochastic model of the structure is thus reduced by exploiting specific properties of random eigenvectors associated with the random eigenfrequencies being sought. An indicator with no additional computation cost is proposed to identify when the method needs to be enhanced. Finally, a simple three-beam frame and an industrial structure illustrate the proposed approach.