Lost data recovery for structural health monitoring based on convolutional neural networks

• DOI: 10.1002/stc.2433
• Publication Date: 2019-08-09T03:35:31Z
• Authors: Gao Fan; Jun Li; Hong Hao
• Keywords: Robustness; Structural Health Monitoring

Signal transmission loss of using wireless sensors for structural health monitoring is a usual case, which undermines the reliability conditions. The measured vibration data with high ratio can hardly be used analysis, that is, modal identification, as it will lead to significant errors in results. This paper proposes novel approach based on convolutional neural networks recovering lost monitoring. network fully feed-forward bottleneck architecture and skip connection, constructs nonlinear relationships between incomplete signal from complete true signal. trained extracts robust higher representation features signals compression layers expands those gradually throughout reconstruction recover obtain signals. long-term Dowling Hall Footbridge are employed validate effectiveness robustness proposed recovery. Two case studies conducted recovery accuracy single-channel multiple-channel cases, respectively. effect sampling rate also investigated. exhibits outstanding capability recovery, even when have severe ratios up 90%. To further demonstrate recovered identification results by different show very good agreement obtained data.