A STUDY ON DEFROMATIONS DETECTION FOR CONCRETE BRIDGES USING A DEEP NEURAL NETWORK

Abstract

The number of bridges that have been constructed for more than 50 years is increasing every year. Furthermore, inspection engineers are decreasing in recent days and also the inspection costs for the maintenance and management of bridges are considerable. Hence, improving the efficiency of periodic inspections and introducing new technologies are urgent issues. In recent years, the application of a deep neural network (DNN), a type of machine learning, is considered to be an effective method for improving the inspection efficiency in the civil engineering field. In this study, a learning model was developed that can be used as a deformation detector for concrete bridges using the DNN as a machine learning. The developed learning model in this study trained using the photographs of the results of road bridges inspections conducted by Fukushima Prefecture. The deformation detector for of concrete members derived from the developed learning model was evaluated by the cross-validation method. As a result, the overall accuracy of the learning model ranged from 56% to 62%. In addition, the classification accuracy of each class tends to increase as the number of training data increased.