BAYESIAN UPDATING OF WOODEN STRUCTURAL PARAMETERS AND SEISMIC DAMAGE CLASSIFIER USING NEURAL NETWORK

Structural health monitoring is an effective way to quickly and efficiently detect earthquake damage to a building. However, its applications to wooden buildings are still limited. One of the challenges in predicting the seismic response of a wooden structure is that the effect of variance in its material and structural properties is relatively significant. To handle this problem, we propose a method to update the damage classifier using Bayesian system identification. The damage classifier is constructed using neural network (NN) and machine learning techniques. The method utilizes the weighted loss function for the NN to reflect the updated probability distribution of the structural parameters. The NN is tuned using backpropagation, where the training data are consequently weighed by the weighed loss function, directly using the value of the posterior probability, according to the data obtained from the numerical model of a target structure with the corresponding parameters. Using the result of the Bayesian updating, the tuned NN damage classifier achieved a higher accuracy than before tuning by reflecting the posterior probability distribution of the structural parameters.