A Distributed Sensing- and Supervised Deep Learning-Based Novel Approach for Long-Term Structural Health Assessment of Reinforced Concrete Beams

A Distributed Sensing- and Supervised Deep Learning-Based Novel Approach for Long-Term Structural Health Assessment of
Reinforced Concrete Beams

Access to significant amounts of data is typically required to develop structural health monitoring (SHM) systems. In this study, a novel SHM approach was evaluated, with all
training data collected solely from a validated finite element analysis (FEA) of a reinforced concrete (RC) beam and the structural health based on the tension side of a rebar under flexural loading. The developed SHM system was verified by four-point bending experiments on three RC beams cast in the dimensions of 4000 mm × 200 mm × 400 mm. Distributed optical fibre sensors (DOFS) were mounted on the concrete surface and on the bottom rebar to maximise sample points and investigate the reliability of the strain data. The FEA model was validated using a single beam and subsequently used to generate labelled SHM strain data by altering the dilation angle and rebar sizes. The generated strain data were then used to train an artificial neural network (ANN) classifier using deep learning (DL). Training and validation accuracy greater than 98.75% were recorded, and the model was trained to predict the tension state up to 90% of the steel yield limit. The developed model predicts the health condition with the input of strain data acquired from the concrete surface of reinforced concrete beams under various loading regimes. The model predictions were accurate for the experimental DOFS data acquired from the tested beams.