Estimation of unmeasured structural responses of submerged floating tunnels using pattern model trained via long short-term memory

Structural conditions are typically investigated using a significant amount of structural response data. Compared with conventional infrastructures such as long-span bridges, submerged floating tunnels in deep-water offshore fields are restrictive in terms of sensor installation and data acquisition owing to their structural and environmental characteristics. Hence, we propose an effective method for estimating unmeasured structural responses using a structural-pattern model. The pattern model, which is used to estimate the responses, is established by a deep learning algorithm using structural response data obtained via hydrodynamics-based simulation. The proposed method is validated numerically. For the deep learning algorithm, long short-term memory (LSTM) is used to reflect the sequential characteristics of the structural behavior. In this study, we establish the input, hidden, and output layers of the LSTM. Additionally, we specify the tether tension and bending moments of the tunnel as responses that should be estimated and the acceleration as the response that could be measured. The proposed method is evaluated based on structural behaviors under various waves with different wave directions, heights, and periods. The result shows that the proposed method can precisely estimate the tether tension and bending moment of the tunnels using only the tunnel acceleration. In particular, the method can effectively estimate the responses under various directional waves without requiring wave information in advance.