1. Real-time Localization of Dynamic Impact Load on Plate Structure using Deep Learning
- Author
-
Pan, Yuxin, Li, Teng, Ventura, Carlos, Zhang, Shunduo, Liang, Xiaoxi, Song, Xiaofan, Pan, Yuxin, Li, Teng, Ventura, Carlos, Zhang, Shunduo, Liang, Xiaoxi, and Song, Xiaofan
- Abstract
Identification and localization of a dynamic impact load applied to a structure are crucial for monitoring its health and safety. Traditional methods based on inversion techniques, either in time-domain or frequency-domain, require large computational cost which cannot achieve real-time implementation. This study proposes a novel deep learning algorithm for accurately localizing a dynamic impact load applied to a structure by deploying a limited number of sensors. The development of the deep learning algorithm involves: 1) time history analysis of a structural finite element model subjected to random impact loading to generate training dataset; 2) development of a deep neural network model to extract and regress the inherent relationship in multivariate time series between input loading and output responses. The proposed approach is verified on a rectangular plate structure subjected to simulated impact loading at a total of 88 possible locations. A high accuracy rate of 96% achieved in 1 millisecond demonstrate the superiority of the proposed deep neural network in achieving real-time localization of structural impact load.
- Published
- 2022