Impact load identification method based on artificial neural network for submerged floating tunnel under collision

Abstract

In order to minimize economic and human losses due to collisions that may cause serious problems, such as severe damage and collapse, in a submerged floating tunnel (SFT), an impact load identification method that can estimate the magnitude, as well as the location of the impact as the point, is proposed. The proposed method is based on an artificial neural network, specifying multi-layered perceptron (MLP), and its performance is verified by numerical simulation and experimental tests. The structural response data in the time domain obtained from the SFT when collisions occur within an ocean current is used to train the neural network model. The MLP-based identification system is investigated to establish each purpose of impact points and magnitude estimation considering the classification and regression model. In the numerical simulation, a finite element model of the SFT simplified as the beam on an elastic foundation is established in ABAQUS and the time history of impact is used to simulate the impact as a collision in hydraulic resistance. The time-series acceleration responses of the simulation are extracted to convert to input variables. The experiment with an SFT model, a set of accelerometers, and an impact hammer is conducted for the identification system validation. Only using the time-series acceleration as input variables for the neural network shows an accurate estimation of the impact points and the magnitude over 97%, and provides reasonable results using the minimal amount of data.