ABSTRACT
Structural health monitoring (SHM) strategies often rely on data acquired from a single bridge seeing a broad range of environmental and operational loads over time. However, the emergence of wireless connectivity now allows multiple bridges in the same highway corridor to be monitored and their response to the same truck linked. In this work, time series forecast models are explored as a tool for jointly modeling the response of bridges on the same corridor and monitored under identical load conditions. More specifically, two types of models are built for the task at hand. First, an encoder-decoder architecture with two different cell types, namely gated recurrent unit (GRU) and long short-term memory (LSTM) are explored. The second type is the autoregressive with exogenous inputs (ARX) model. To evaluate the performance of the forecast models in taking the output of one bridge to predict the response of another under the same truck load, finite element models are built for two real-world bridges and a simulation dataset containing 2,100 pairs of bridge responses to the same truck load created. The two encoder-decoder models provided accurate prediction capabilities of bridge response compared to the ARX baseline approach.
