Long-term structural health monitoring for bridge based on back propagation neural network and long and short-term memory

Abstract

Bridges are critical components of transportation infrastructure. To ensure the long-term performance of bridges and the safety of the public, regular inspections are required during their service. Structural performance assessments are subject to various conditions. Structural deterioration is caused by complex environmental and operational conditions (EOCs) including temperature changes, truckloads, chemical corrosion, etc. In this study, an in-site structural health monitoring (SHM) system is designed and deployed on the Caohekou Bridge in China with a series of sensors installed, providing continuous real-time data for 4 years. Crack width, vertical deformation, concrete strains, temperature, longitudinal displacement and acceleration are monitored and assessed. Time-history monitoring data are comprehensively analysed to advance our understanding of structural deterioration caused by time and temperature. It is very common to lose data during the monitoring process, especially in the long-term run. To overcome the challenge of missing data package and to realize early warning, methods including an SHM systems face, a back propagation neural network and a long short-term memory are proposed to predict the bridge responses under the change of EOCs. It has been proved that the performance of predicted crack widths is close to that of the measured value, and the trend of change is consistent. Such results indicate that approaches proposed that quantitatively assess in-service structure are promising. Therefore, effective and efficient maintenance decisions can be made to ensure an immediate response, long-term safety and serviceability of bridge structures.