Failure Predictability Model Based on Reliability Assessment of Prestressed Concrete Bridge Beams Using Artificial Neural Networks

Abstract

This paper presents an approach utilizing neural networks for structural engineering and reliability assessment, with a focus on demonstrating their efficacy in predicting stresses and failure probabilities of prestressed concrete bridge beams within the concept of limit state design. Previous studies have highlighted the robust capabilities of neural networks in solving complex problems and their wide utilization across a range of applications in civil engineering. The study aims to assess the applicability of neural networks to predict flexural stress for a prestressed bridge beam. Subsequently, evaluate their reliability for the bending failure criterion, specified in the French code of prestressed concrete design, BPEL 91 revised 99. Furthermore, the paper outlines a methodology that combines neural networks for prediction and the First Order Reliability Method for reliability evaluation. To illustrate the efficacy of the proposed approach, the predicted stress is compared with the resulting values through finite element analysis and the response surface method. The neural network learning process is based on a collected design dataset of multiple prestressed bridges sourced from technical studies offices, construction companies, and the Ministry of Equipment and Water in Morocco.