Long-term structural response prediction models for concrete structures using weather data, fiber-optic sensing, and convolutional neural network

This study proposes a long-term strain prediction model for concrete structures using weather data. In the proposed model, the relationship between weather and the strain data for a concrete structure is defined by a convolutional neural network (CNN), which is a machine learning technique, based on the strong correlation between the two types of data. The weather data collected from a weather station located near the monitored structure are used in the input layer of the CNN; the strain data measured by fiber-optic sensors (FOSs) at the structure are used in the output layer of the CNN. The trained CNN can predict the strain using only weather data in the case of sensor malfunctions or data loss. Various types of weather data, including the air temperature, relative humidity, and wind speed, are used to determine the environmental factors that are valid for predicting the long-term deformation of concrete structures. Six prediction models are proposed, in which the three types of weather data are used individually or jointly in the input layer of the CNN. The proposed models are applied to predict the strain of a footbridge located at Princeton University. To build the prediction models, the strain data measured at the bridge over a long-term period and the weather data obtained from a nearby local weather station are used. The performance of the prediction models is verified through long-term strain prediction. Furthermore, the prediction performance of the analyzed models is compared, and the weather data types that are significant for predicting the long-term deformation of concrete structures are elucidated.