Machine learning and Weibull-based reliability analysis of impact-resistant geopolymer concrete with construction demolition waste and natural fibers

This study presents a novel impact-resistant geopolymer concrete (GPC) incorporating construction and demolition waste powder, brick waste powder, slag, recycled concrete aggregates (RCA), and natural fibers (banana, jute, and pineapple) to simultaneously enhance its mechanical performance, impact resistance, and sustainability. Unlike previous studies that primarily focused on compressive strength, this study integrates experimental evaluation, microstructural characterization, machine learning, and Weibull reliability analysis to assess both the impact strength performance and reliability of GPC. By addressing the brittleness and performance variability associated with high RCA content, this study provides a practical strategy for developing durable, low-carbon concrete within the framework of a circular economy. This study examined the strength of natural-fiber-based geopolymer concrete using slag, construction demolition waste powder, brick waste powder, and RCA (0, 25, 50, 75%, and 100%). Banana, pineapple and jute fibers were used at 0.75% by volume. These combinations were mixed and tested for the compression and impact strengths of a total of 20 mixtures. In addition, microstructural analysis was performed on the GPC. Machine learning and the Weibull distribution were used to analyze the performance and reliability of the impact strength. The results showed that increasing RCA content substantially reduced impact resistance, with reductions of 45.6–57.3% in first-crack impact resistance and 54.4–62.7% in failure impact resistance for the fiber-reinforced mixtures. Pineapple-fiber-reinforced GPC exhibited comparatively stable post-cracking performance, with the ductility index remaining within 1.81–2.01 across the investigated RCA levels. The Weibull analysis demonstrated that fiber incorporation reduced the variability of impact resistance and provided more stable reliability characteristics under repeated impact loading. The XGBoost model showed good predictive capability, achieving a coefficient of determination of 0.80 on the testing dataset, with root mean square error, mean absolute error, and mean absolute percentage error values of 11.05, 9.39, and 19.54%, respectively. Shapley additive explanations analysis indicated that natural coarse aggregate and pineapple fiber were the most influential variables affecting the predicted impact resistance, followed by superplasticizer dosage, RCA content, and fiber volume fraction.