Assessment of Compressive Strength of Concrete with Glass Powder and Recycled Aggregates Using Machine Learning Approaches

In the last decade, the importance of sustainable construction and artificial intelligence (AI) in civil engineering has been underlined in many studies. Numerous studies highlighted the superiority of AI techniques over simple and mathematical regression analyses, which suffer from relatively poor generalization and an inability to capture highly non-linear relationships among inputs and output(s) parameters. In this study, to evaluate the compressive strength of concrete with glass powder (GP) and recycled aggregates, 600 concrete samples were tested in the laboratory, and their results were evaluated. For intelligent assessment of concrete compressive strength (CCS), the study utilized an improved artificial neural network (ANN) with particle swarm optimization (PSO) algorithm and imperialist competitive algorithm (ICA). For training the models, the experimentally obtained data were used. The concrete ingredients formed the inputs of the AI-based predictive models of CCS. The experimental findings reveal that the implementation of recycled coarse aggregates in concrete from a sustainable construction point of view is advantageous and can enhance the CCS by 11.43%. Apart from that, findings indicate that utilization of 10% GP can lead to a nearly 20% increase in CCS (from 44.6 to 54.1 MPa). Additionally, the experimental observations show almost 40% improvement of CCS when 5% micro silica was used in the concrete mixture. Based on the findings, the study suggests the utilization of waste glass powder to partially replace cement in concrete, which can reduce the amount of cement production. This reduction from economic, energy-saving, and environmental (reduction in greenhouse gas emissions) points of view is of interest. On the other hand, the AI results show that the PSO-based ANN model outperforms the ICA-based ANN for the utilized dataset. According to the findings, the PSO-based ANN predictive model (with a coefficient of determination value of 0.939 and root mean square value of 0.113 for testing data) is a capable tool in predicting the CCS. Hence, this study recommends the implementation of AI-based models in CCS assessment.