Smart Phone Based Bridge Seismic Monitoring and Vibration Status Realization by Time Domain Convolutional Neural Network

Date: 4/24/2019

Time: 09:15 AM

Room: Elliott Bay

In this study, a bridge seismic monitoring system was developed using smart phones as sensor nodes. Connected smart phones together in the same project as a sensor network, this system provides an easy way for bridge vibration sensing method. A long-term field test of this system was performed on a PC bridge in Japan. Smart phones with the monitoring application were deployed at six different locations inside the bridge’s box girder. The system was connected to a cloud server for real-time data uploading and remote control. Inconsistent sampling rates and faulty sensor readings were identified to be two major problems with the use of the proposed system. A few seismic acceleration response records of the Takamatsu Bridge were captured during the more than one-year operation period. Dynamic properties extracted from this system were compared with those of reference seismometers to verify its viability and accuracy. To recognize vibration wave form from those sensor faulty data included data, a simple neural network was trained with data measure from the bridge. Combining with original time domain data as input layer, 2 one dimensional Convolutional Neural Network layers, Pooling layer, and 2 Fully Connected Layers, the proposed model serves fast vibration classification of device faulty, transportation vibration, earthquake event and ambient noise in very high accuracy.

Presenting Author: Ji Dang

Authors