Division of Information Technology, Engineering and the Environment
School of Electrical and Information Engineering
School of Electrical and Information Engineering
Magnetic Hysteresis Models Using Knowledge Based Techniques,
Research Development Grant, University of South Australia, 1996-97.
The project initiated under a previous research development grant has
led to development of a novel technique to describe magnetisation
process by means of artificial neural networks (NN). A multi-layer
feed-forward NN architecture was used with the back propagation
algorithm and single layer of hidden processing units with nonlinear
sigmoid processing function.
A custom made C++ program was written to train NN by configuring with
suitable weights and to generate resulting mapping of magnetisation
variables. The major hysteresis loop and the set of five first order
transition curves was used for training and testing in the static case.
In the dynamic case, dynamic 50 Hz hysteresis loops were used for NN
modelling. In both cases, the accuracy of approximation was good once a
near optimum NN configuration was found by adjusting number of hidden
processing units and number of iteration steps.
The extension of the project targets intelligent generation of NN
structure and its optimisation which leads to the advancement of this
novel approach to the description of the magnetisation process.
