Estimating the weight and the failure load of a spaghetti bridge: a deep learning approach

Abstract

ABSTRACT In this article, the ability of estimating the weight and the failure load of a structure through image processing has been investigated. To this end, the well-known civil engineering practice, spaghetti bridge, has been used as a test structure. To set up uniform experiments and to simplify the construction process, only 2 dimensional bridges were considered. By defining the failure load as the load that breaks the bridge, in the process of construction and testing, only bridges that were broken have been added to the database. The developed database was employed to train and validate the artificial neural network with three hidden layers particularly designed for this research. Four different activation functions were tested. The results obtained from Logistic sigmoid activation function were comparatively better. The designed artificial neural network was optimised through genetic algorithm. The benefit of using genetic algorithm was that several solutions were obtained. The artificial neural network with the lowest error for both training and validation data was selected. The results indicate that it is possible to estimate the weight and the failure load of a bridge with an acceptable degree of accuracy just by using the image of the bridge.