ニューラルネットワークによる列車走行音からの線路内異常検知手法

抄録

In the case of driverless trains, the ability of abnormal-sound detection which traditionally has been based on crew's ears will be lost. To make up this, this paper proposes an abnormal-noise detection system using microphones under trains. The feature of the system is to determine the abnormalities on railway track not only based on the sound but also the velocity and the position of the train. To accomplish that, the system uses neural network which is able to predict normal sound level based on the velocity and the position.

In the system, when the running sound is extremely larger than the predicted normal sound, it would be determined as abnormal sound. To verify the effectiveness of the system, test running is conducted where the test train passes on a stone located on the rail. Through the test running, we have confirmed that the system would be able to detect the abnormal sound due to the stone in case the train passes on the stone at over 20km/h.