Durability assessment and machine learning-based prediction of coconut shell and silica fume concrete

Abstract

The increased demand for sustainable building materials has promoted the use of industrial and agricultural by-products in concrete construction. The paper provides an experimental and machinelearning-based study on durability performance of M30-grade concrete. The partial replacement of conventional coarse aggregate was done with coconut shell (CS) and silica fume (SF) was used as a supplementary cementitious material (SCM). Eight concrete mixtures were experimented upon: the control mix, silica-fume-modified concretes (5%–15% SF replacement) and combined CS-SF concretes (10%–40% coconut shell replacement). Durability properties were evaluated using Water absorption and Effective porosity testing (ASTM C642), Rapid Chloride Permeability Test (ASTM C1202), Sorptivity testing and Sulphate and Acid Attack resistance tests according to CEB-FIP guidelines for durability assessment. The silica fume had a significant effect on the permeability related durability, lowering the Rapid Chloride Permeability Test (RCPT) value from 1120 C for the control mix to 548 C for the optimum CS-SF mix, which is about 51% less penetrable to chloride ions. The refinement of pore resulted in reduction of the sorptivity of silica-fume-modified mixes by approximately 4-5 percent. Coconut shell aggregates were porous and took more water, to a maximum of 6.7% at 40% replacement. Nonlinear correlations of mix parameters and durability performance were modelled using machine learning. Twelve models were used through grid search optimisation. A pipeline-based LOOCV strategy was employed to ensure limited data is assessed in an unbiased manner. R 2 , RMSE and MAE were used to evaluate model performance. The XGBoost model demonstrated predictive ability (R 2 = 0.96). SHAP analysis showed permeability-related properties are determined by silica fume content, while absorption and acid resistance are controlled by coconut shell content. Based on a small dataset, the proposed experimental-AI structure displays a proof-of-concept of explainable durability prediction and performance-based mix design of sustainable concrete.