Abstract
In this study, we propose a classification system that matches the most appropriate power generation device to an arbitrary environmental vibration. Firstly, 40 types of ambient vibration data, such as machine vibration, window vibration, etc., are prepared. On the other hand, three types of magnetostrictive vibration power generation devices with different resonant properties are prepared. Secondly, a Convolutional Neural Network (CNN) that is built in the numerical analysis software MATLAB was used to construct the matching system. CNN is particularly effective for pattern recognition and data classification, which is to be dealt with in this research. The 30 types of ambient vibration data are used as input data, whereas the device identifier that shows the maximum open-circuit voltage is used as the teaching data. CNN learns the relationship between the characteristics of ambient vibration and the devices that showed the maximum open-circuit voltage. As a result, the accuracy of the CNN was 70% for 10 types of unknown environmental vibrations.
