Search results

This work aims to present a multilevel optimization strategy based on manifold‐mapping combined with multiquadric interpolation for the coarse model construction.

In the proposed approach the coarse model is obtained by interpolating the fine model using multiquadrics in a small number of points. As the algorithm iterates the response surface model is improved by enriching the set of interpolation points.

This approach allows to accurately solve the TEAM Workshop Problem 25 using as little as 33 finite element simulations. Furthermore, it allows a robust sizing optimization of a cylindrical voice‐coil actuator with seven design variables.

Further analysis is required to gain a better understanding of the role that the initial coarse model accuracy plays in the convergence of the algorithm. The proposed model allows to carry out such analysis by varying the number of points included in the initial response surface model. The effect of the trust‐region stabilization in the presence of manifolds of equivalent solutions is also a topic of further investigations.

Unlike the closely related space‐mapping algorithm, the manifold‐mapping algorithm is guaranteed to converge to a fine model optimal solution. By combining it with multiquadric response surface models, its applicability is extended to problems for which other kinds of coarse model such as lumped parameter approximations for instance are tedious or impossible to construct.

This study aims to focus on the application of the stochastic algorithms for optimal design of electrical machines. Among them, the authors are interested in particle swarm optimization and teaching–learning-based optimization.

The optimization process is realized by the coupling of the above methods to finite element analysis of the electromagnetic field.

To improve the performance of these algorithms and reduce their computation time, a coupling with the artificial neuron network has been realized.

The proposed strategy is applied to solve two optimization problems: Team workshop problem 25 and switched reluctance motor with flux barriers.

The purpose of this paper is to present a constructive algorithm to design multilayer perceptron neural networks used as approximation models of electromagnetic devices.

The proposed procedure allows automatic determination of both the number of neurons and the synaptic weights of networks with a single hidden layer. The approximation model is used in design optimization problems. The inputs of the neural network correspond to the design parameters whereas the output corresponds to the objective function of the optimization problem. The neural model is then inverted in order to determine which input is associated to a prefixed output.

The performance of the algorithm has been tested on analytical function and on the TEAM workshop problem 25.

As the reliability of the optimum solution is strongly affected by the accuracy of the neural approximation model, the approximation error is kept as low as possible, especially in the maximum/minimum points.

This paper aims to deal with the development of a newly improved version of teaching learning based optimization (TLBO) algorithm.

Random local search part was added to the classic optimization process with TLBO. The new version is called TLBO algorithm with random local search (TLBO-RLS).

At first step and to validate the effectiveness of the new proposed version of the TLBO algorithm, it was applied to a set of two standard benchmark problems. After, it was used jointly with two-dimensional non-linear finite element method to solve the TEAM workshop problem 25, where the results were compared with those resulting from classical TLBO, bat algorithm, hybrid TLBO, Nelder–Mead simplex method and other referenced work.

New TLBO-RLS proposed algorithm contains a part of random local search, which allows good exploitation of the solution space. Therefore, TLBO-RLS provides better solution quality than classic TLBO.

This study aims to improve the bat algorithm (BA) performance for solving optimization problems in electrical engineering.

For this task, two strategies were investigated. The first one is based on including the crossover technique into classical BA, in the same manner as in the genetic algorithm method. Therefore, the newly generated version of BA is called the crossover–bat algorithm (C-BA). In the second strategy, a hybridization of the BA with the Nelder–Mead (NM) simplex method was performed; it gives the NM-BA algorithm.

First, the proposed strategies were applied to solve a set of two standard benchmark problems; then, they were applied to solve the TEAM workshop problem 25, where an electromagnetic field was computed by use of the 2D non-linear finite element method. Both optimization algorithms and finite element computation tool were implemented under MATLAB.

The two proposed optimization strategies, C-BA and NM-BA, have allowed good improvements of classical BA, generally known for its poor solution quality and slow convergence rate.

This paper compares three different radial basis function neural networks, as well as the diffuse element method, according to their ability of approximation. This is very useful for the optimization of electromagnetic devices. Tests are done on several analytical functions and on the TEAM workshop problem 25.

The purpose of this paper is to develop a novel multi‐objective optimization algorithm which takes into account the uncertainty in design parameters by using a reduced resolution for their representation, thus implementing a simple form of robustness. Additionally, the number of function evaluations should be minimized.

The proposed approach is based on an elitist evolutionary algorithm coupled with a reduction in the number of significant figures used to represent design parameters. In effect, this becomes a filter in the optimization process and allows the system to avoid extremely sharp optima within the search space. By reducing the resolution of the search and maintaining a full archive of previous solutions, the number of evaluations of the objective functions, each of which may require an expensive numerical solution, is reduced.

The algorithm was tested both on an algebraic test function and on two TEAM Workshop Problems (22 and 25). The results demonstrated that it is stable; can emerge from deceptive fronts; and find optimal solutions which match those previously published at a relatively low‐computational cost.

The originality of this paper lies in the concept of using a low‐resolution representation of the design parameters. This results in a finite size search space and increases the speed of the algorithm while avoiding non‐manufacturable solutions.

To present a neural network‐based approach to the design of electromagnetic devices.

A neural model is created which reproduces the relationship between the design parameters of the device and the performance parameters, typically field values.

The neural model is a single hidden layer MLP network, trained by using a set of cases calculated, for example, by means of a finite element analysis. The design problem can be solved by fixing the performance values at the output of the network and by calculating the corresponding input values. The relationship between the input and the output of the neural network is represented by three equations systems. By means of these three systems, we can forward the domain of the input, and we can back propagate the desired output throughout the network layers. In such a way, both the domain of the design parameters and the domain of the desired performances values can be projected in the same space. Whatever point inside the intersection between the two projected domains corresponds to a solution of the design problem.

Presents a procedure which is able to find a point belonging to such an intersection.

Factors affecting convergence of the simulated annealing method are investigated using an actual model. The convergence characteristics of various optimization methods are examined using the contour line of objective function. Two kinds of combination methods with the simulated annealing method and the Rosenbrock’s method are investigated.

Manifold‐mapping (MM) is an efficient surrogate‐based optimization technique aimed at the acceleration of very time‐consuming design problems. In this paper we present two new variants of the original algorithm that make it applicable to a broader range of optimization scenarios.

The first variant is useful when the optimization constraints are expressed by means of functions that are very expensive to compute. The second variant endows the original scheme with a trust‐region strategy and the result is a much more robust algorithm.

Two practical optimization problems from electromagnetics eventually show that the proposed variants perform efficiently.

The original MM algorithm is extended with two new variants. Therefore, the MM approach is applicable to a much larger set of design situations.