Abstract
In this study, we propose a matching method for a variety of the magnetostrictive vibration power generation device that are suitable for respective ambient vibrations by using the Self-organizing map (SOM). SOM is a kind of clustering methods that can classify the given data without any external criteria. The ambient vibration data are converted into the input signal for SOM by way of filtering and linear prediction processes. The data are then used to train SOM by which the data are grouped according to their similarity. Subsequently, the vibration responses of the generators were put through the trained SOM in order to determine into which cluster the data were grouped. The results showed the possibility of clustering ambient vibration data and matching the power generators to the appropriate vibration sources by using SOM. However, the classification accuracy was not enough due to the bias of the dominant frequency components in both the ambient vibration data and the generators. The natural frequency and its harmonic components were significant in the latter case. It is necessary to introduce the additional data with dominant frequencies that are not covered by the environmental data employed in this study, and to create variations of generator which do not overlap in response frequencies.
