Neural networks based identification for the Akinada Suspension Bridge with earthquake responses

A neural networks based identification approach with the direct use of actual incomplete time-domain dynamic responses under the Geiyo Earthquake (24 March, 2001) is developed for a suspension bridge, Akinada Bridge in Japan. As existing information, the velocity responses in the middle of the main span in longitudinal, transverse and vertical direction, the acceleration responses at the top and in the middle of the tower in transverse and longitudinal direction, and the acceleration responses at the basement were measured. Two neural emulators are constructed and trained with the first part of the dynamic responses to identify the transversal and vertical velocity at the deck in the middle of the main span of the Akinada Bridge. The two neural emulators can be treated as a nonparametric model for the bridge in the situation before the earthquake occurred. The performance of the trained emulator neural network models for the suspension bridge is evaluated by numerical simulation in which the forecast response from the neural emulator is compared with the measurement during different stages of the Geiyo Earthquake. The RRMS error between the forecast responses and the measurements in different stages are decided. Results show that the RRMS errors according to the transversal and vertical velocity in the middle of the main span have a variance. This results means that occurrence of damages in some structural members is possible. This analysis result is testified by inspection after the earthquake, it is found that four stay cables are broken during the earthquake. The proposed approach is a nonparametric system identification method, in which a prior information about the exact model is not needed, so it has significant advantage when dealing with real-word situations where the selection of a suitable parametric model for identification is usually a demanding task. And the proposed neural emulator is promising method for the design of the control system for a suspension bridge for which the modeling and control system design is difficult.