Scrutinizing the Performances of Hybrid ANN Models for Forecasting Condition of Bridges

Abstract Bridges all over the world are vulnerable to severe deterioration agents meanwhile their maintenance budgets are being tightened. This state of affairs necessitates the establishment of an autonomous deterioration model to predict the performance condition of bridges. This research paper explores the implementation of a set of intelligent data driven models to analyze the future condition ratings of bridge decks. These models comprise support vector machines, Gaussian process regression, regression tree, back propagation artificial neural network, Elman recurrent neural network, cascade forward neural network, long short‐term memory network and deep convolutional neural network. The performance comparison is carried out using five evaluation metrics of root mean squared percentage error, mean absolute percentage error, root mean squared error mean absolute error and relative absolute error. The models herein are developed and validated using the structural deterioration information retrieved from the National Bridge Inventory (NBI). It can be argued that the developed deterioration model could be implemented by departments of transportation to analyze and monitor the performance condition behavior of bridge components over their useful lifetime.