Condition assessment of RC beams using artificial neural networks

Abstract

In this study, an imaged-based methodology for condition assessment of Reinforced Concrete (RC) road bridge components is presented. The study is divided into experiments on RC rectangular beams and scaled (1:12) T-beams, validation of numerical models for T-beams and training of corresponding artificial neural networks (ANNs). In the experimental work, Digital Image Correlation (DIC) was used as a virtual sensor for data extraction. Rectangular RC beams of size 1800 mm × 150 mm × 200 mm and scaled (1:12) RC T-beams were tested under four-point flexural loading on a 100-ton dynamic testing machine. The experimental stress-strain curves obtained from the compression test on prism specimens at 28 days were used as input data for material model parameters in finite element model (FEM) software SAP2000. To assess the condition of structural components, a local damage index (LDI) was developed. Validation of FEM results with experimental results enables derivation of moment-curvature backbone curve for full-scale bridge girders, which enables further quantification of damage and residual moment capacity of full-scale bridges designed by Ministry of Road Transport and Highways (MoRTH). The correlation between the experiments, simulation and ANN predictions was found to be very satisfactory.