Search results

Global optimization in electrical engineering using stochastic methods requires usually a large amount of CPU time to locate the optimum, if the objective function is calculated either with the finite element method (FEM) or the boundary element method (BEM). One approach to reduce the number of FEM or BEM calls using neural networks and another one using multiquadric functions have been introduced recently. This paper compares the efficiency of both methods, which are applied to a couple of test problems and the results are discussed.

The process of identifying unknown hidden objects by taking advantage of electromagnetic effects becomes more and more important. Clearing of mines or finding electrical conductors in concrete should be mentioned here. Magnetisation and eddy currents are the phenomena which are used in general. In this case, the layout and arrangement of the excitation coils and receiving coils influences the effectiveness and accuracy crucially. This design optimization process can be done by simulating the electromagnetic field with a 3D finite element method. Once a satisfying configuration has been found, the question arises, which quantities of the measured (and hence simulated) signals contain the most reliable information? Since the 3D finite element calculations are very time consuming, the inverse problem (detecting the ferrous object from some measured signals) is performed by approximating the corresponding electromagnetic signal by a neural network. Investigations on a ferrous conductive rod will be described in the paper.

Many engineering optimisation problems are difficult to describe mathematically and as such can not be easily optimised. Recently attention has focussed on developing methods to create approximations of the real object function using numerical model data instead. The approximated function can then be optimised using a suitable optimisation method. This paper describes the extraction of derivative information from a neuro‐fuzzy system. Subsequently, this permits the application of classic deterministic optimisation methods in order to identify the global minimum of any approximated objective function. For non‐differentiable functions this approach is of great benefit. Results from an analytical optimisation example, in which the objective function and the solution are known, and a two variable loudspeaker optimisation problem are discussed. In both cases, the neuro‐fuzzy system worked well to model the physical problem and the extracted derivative served to locate the minimum.

The conventional starting point for the design of an electrical machine (or any low-frequency electromagnetic device) is known as “sizing”. In this process, a simple magnetic circuit is used to estimate the main geometric parameters. This does not work for many devices, particularly where eddy currents and non-linearity dominate. The purpose of this paper is to investigate an approach using a neural network trained on a large database of existing designs as a general sizing system.

The approach is based on a combination of a radial basis function neural network and a database of stored performances of electrical machines. The network is trained based on a set of typical performance requirements for a machines design problem. The resulting design is analyzed using finite elements to determine if the design performance is acceptable.

The number of neurons in the network was varied to determine the approximation and generalization capabilities. The finite element analysis showed that the network produced initial design parameters which resulted in an appropriate performance.

The research has looked at only one class of machine. Further work is needed on a range of machines to determine how effective the approach can be.

The approach can provide a good initial design and thus can reduce overall design time significantly.

The paper proposes a novel, fast and effective generalized approach to sizing low frequency electromagnetic devices.