Predicting the damage to cementitious composites due to acid attack and evaluating the effectiveness of eggshell powder using interpretable artificial intelligence models

Abstract

The resistance of cement-based composites (CBCs) to the intervention of aggressive ions is the primary determinant of their durability. H2SO4 is a highly corrosive acid capable of causing structural damage. The corrosive properties of H2SO4 can pose a significant threat to building materials, resulting in accelerated deterioration and structural failure. This study employed advanced computing tools to assess the compressive strength loss (CSL) in mortar modified with eggshell powder (ESP-CM) subjected to acid conditions. The modeling process utilized an experimental dataset of six different raw materials as inputs, and the output variable was the CSL. ESP-CM samples were subjected to a 5% H2SO4 solution during the experimental work. Two machine learning interpretable models, including gene-expression programming (GEP) and multi-expression programming (MEP), were developed for the said purpose. In addition, sensitivity analysis (SA) and SHapley Additive exPlanations (SHAP) analysis were used to assess the influence and correlation of raw inputs. The established models for predicting the CSL of ESP-CM due to acid attack corresponded well with the actual results. The accuracy of constructed models was evaluated using statistical analyses and analyzing the difference amongst target and model-estimated results. Comparing the R2 of the models, it was determined that the MEP model performed better than the GEP model in predicting the CSL of ESP-CM after an acid attack, with an R2 of 0.90 versus 0.87. The interaction diagrams derived from the SHAP study exhibited that a replacement level of ESP between 40 and 60 kg/m3 provided the greatest resistance to acid attack.