Accuracy Improvement of Damage Classifier for a Wooden Building Considering Pulse-like Seismic Ground Motions

Abstract

Rapid damage assessment by a structural health monitoring system is expected to prevent casualties caused by the collapse of wooden buildings due to repeated seismic motions. While neural networks are known to be a powerful tool for determining the risk of collapse based on the response records, they are not good at classifying untrained damage patterns. This study proposed a method to consider the diversity of damage patterns in the construction of a damage classifier for wooden buildings. This is achieved by using pulse-like simulated ground motions as the input ground motions for seismic response simulation of a target building model in the development of the training dataset. The effectiveness of the proposed method was validated by comparing its accuracy on test data.