In an industrial field a technique called a vibrate-acceleration method which translates vibration of bearing to electrical signal has been widely used for detection of abnormality of a ball-bearing. However, any authorized rule to decide whether a bearing is abnormal or has not been yet established. Presently, final decision dependents in the intuition of an expert.
Then this paper proposes a technique to automatically decide abnormality of ball-bearings by using a two dimensional graphic plane with axis of frequency and time and neural-networks. Effectiveness of the technique is confirmed experimentally. In the experiment, fractal dimension of the graphic plane is firstly measured, and we confirm it has strong correlation with R. M. S. of the vibration-accelation signal. Then we construct a neural network as it makes an input plane for neural-networks the graphical spectal plane. We could have justify accuracy more than 70 percent for nine samples involving abnormal ones.
J-STAGEがリニューアルされました! https://www.jstage.jst.go.jp/browse/-char/ja/
