Measurement of reinforcement corrosion in concrete by integrated simulations, ultrasonic tests, and artificial neural networks

Current methods for measuring reinforcement corrosion in concrete have limitations, including low efficiency, poor accuracy, and ambiguous interpretation. High-accuracy nondestructive measurement of reinforcement corrosion was realized in this study through the integration of three technical approaches: (1) Ultrasonic wave propagation in reinforced concrete (RC) with different rebar diameters was simulated using COMSOL to establish the relationship between rebar corrosion (reflected by diameter reduction) and ultrasonic characteristics; (2) Accelerated corrosion experiments were performed on RC specimens using a 5 % sodium chloride solution, monitoring corrosion evolution over 40 h. Three distinct corrosion stages were identified: slow development, rapid expansion, and stable deterioration. Additionally, ultrasonic propagation time was found to delay and amplitude increase with intensifying corrosion, while concrete compressive strength decreased rapidly after 8 h of corrosion; (3) A back propagation-artificial neural network (BP-ANN) optimized by the AdaBoost algorithm was developed, with ultrasonic amplitude, crack width, and compressive strength as input parameters. To enhance model training stability and avoid overfitting, 10-fold cross-validation was adopted for data partitioning, and the Gray Wolf Optimization (GWO) algorithm was integrated to optimize the weights and thresholds of the BP-ANN. The AdaBoost-BP model achieved high prediction accuracy for rebar corrosion indicators: 93.75 % for mass loss rate and 97.67 % for section loss rate. The integrated method, combining coda wave interferometry (CWI), ultrasonic testing, and ANN, aims to overcome the limitations of existing techniques. A reliable framework for nondestructive evaluation of reinforcement corrosion was provided, enabling precise monitoring of RC structure durability. This foundation can be extended to complex real-world structures (e.g., bridges, coastal buildings) and promote the development of intelligent corrosion detection systems. • Integrate COMSOL simulations, ultrasonic tests and AdaBoost-BP-ANN to realize early detection of rebar corrosion via Coda Wave Interferometry. • Enables early corrosion detection (80–200 μm depth) via coda wave interferometry, capturing subtle rebar diameter reductions before visible cracks. • The optimized model achieves 93.75 % and 97.67 % accuracy for rebar mass loss and section loss prediction, outperforming traditional methods.