A study on assessing the damage of a bearing with multilayer neural network using the acceleration response of a bridge

After an earthquake, quickly identifying damage on a bridge is crucial to early recoveries. However, current method of structural health monitoring on bridges relies on cameras, cable networks, and manpower. This method requires high costs as well as slower recovery effort. In order to address these issues, a health monitoring model of a bridge bearing using a multilayer neural network and accelerometers were proposed. Such a model ensures real time evaluation of bearing damage while using only the acceleration responses of a bridge. A multilayer neural network was proposed with acceleration responses of steel girder and bearing being the input while the output being bearing displacement. The responses were obtained by applying seismic motions to the structural model with TDAP III, a general purpose three dimensional dynamic analysis computer code. Several different combinations of seismic motions were considered as the learning data set. The results showed that using a large number of learning data sets for the multilayer neural network returned a high applicability. This study demonstrated that the bearing displacement, as well as the bearing damage could be precisely estimated.