Bridge structural failure happens as the lack of monitoring. The existence of bridge structural health monitoring system is necessary for bridge maintenance due to its ability to process data and provide the information of structural health level. This research is performed to design a deep neural network model for classifying structural integrity with high accuracy. The model requires input data in the form of F-statistic, which is derived from structural vibration data. In the current approach, the vibration data are obtained from numerical analysis by means of the finite element methods. As much as 17.493 vibration cases are generated for five levels of structural integrity, namely, healthy conditions and conditions of 1%, 5%, 10%, 20% damage level. The neural network model consists of one input layer of 20 neurons, six hidden layers with 12 neurons per layer, and one output layer of 5 neurons. The model is trained by using Adam optimizer. The results show that the model is able to accurately classify the structural damage at 83.3% accuracy, and the majority of the false predictions occur in differentiating the healthy structural condition from those of 1% damage.
展开▼
