Simultaneous Recovery Model for Missing Multiple-Source Structural Health Monitoring Data of a Quayside Container Crane

Abstract

Structural health monitoring (SHM) data encompasses vital information that provides a comprehensive understanding of the structural health condition. However, data loss may occur due to faults in acquisition equipment or sensors, and it is essential to reconstruct missing data to ensure the integrity of the monitoring information. Although extensive researches have been conducted on the topic of data recovery, a suitable missing data recovery method that can effectively address the missing data for multiple-source monitoring variables has not been identified yet. In this study, we proposed a novel missing data recovery model based on a deep learning framework to recover the missing strain and acceleration data simultaneously for SHM of the quayside container crane (QCC). The framework combines dual-tree complex wave transform (DTCWT) and bidirectional long short-term memory with attention mechanism (Att-BiLSTM). The multiple-source monitoring data are decomposed into several subtime series using the dual-tree complex wavelet, then, the Att-BiLSTM network is used to assign different weights to each subsequence in order to capture valuable information from the complete data. The effectiveness of the proposed model is verified by case studies, and the comparison results of missing data recovery under different miss rates show that the proposed model simultaneously improves the accuracy of missing data recovery for multiple-source monitoring variables.