Use of artificial intelligence to predict the performance of recycled aggregate concrete
Résumé
This chapter reviews artificial intelligence (AI) algorithms in classification composition of recycled aggregate (RA) and the prediction of the performance of recycled aggregate concrete (RAC). Recent AI algorithm developments have greatly improved RA classification and prediction models. Hyperspectral imaging systems have demonstrated good material identification accuracy when paired with partial least-squares discriminant analysis. In real-time CDW classification, deep learning models, such as YOLOV7, have attained a mean average precision of up to 70%. Convolutional neural networks have proven to be 97%–98% accurate in classifying and mass-estimating photographs of recycled aggregate. In other aspects, artificial neural networks (ANNs) have consistently outperformed standard regression approaches for predicting the properties of RAC, with R2 values frequently above 0.9. Prediction accuracy has been further enhanced by hybrid models that combine particle swarm optimization with machine learning algorithms. Furthermore, using AI-driven methods, promising results have also been observed in predicting durability attributes, including carbonation depth and chloride diffusion. AI technology in CDW recycling procedures may boost material recovery, raise the quality of the final cement-based product, and promote more environmentally friendly building techniques.