Mechanical properties analysis of geopolymer concrete based on the sugarcane bagasse ash using machine learning

Abstract

Geopolymer concrete (GPC) became the substitute for conventional Portland cement-based concrete in recent years. In this work, fly ash (FA) and sugarcane bagasse ash (SBA) are used to create geopolymer concrete. The percentages of fly ash replacement with SBA at 0%, 10%, 20%, 30%, 40%, and 50%, respectively, are indicated by the designations FS0, FS10, FS20, FS30, FS40, and FS50. Whereas FS0 indicates the traditional mix with no fly ash replacement, FS50 indicates a 50% replacement of fly ash. Compressive strength, flexural strength, and split tensile strength were among the mechanical characteristics that were tested. At 12 M NaOH, the geopolymer concrete FS30 had compressive strengths of 46.88 MPa after 28 days. This effect is consistent with fly ash’s geopolymerization reaction, which helps to generate more binding phases and produces denser and stronger concrete. Random forest (RF) algorithm demonstrated the most robust performance, achieving strong testing R² values of 0.8368, 0.7861, and 0.8199 for compressive, flexural, and split tensile strength, respectively. In contrast, the XGBoost model was, evidenced by its near-perfect training R² scores of approximately 0.9999, which plummeted to significantly lower testing R² values of 0.7378, 0.7136, and 0.7140 for the three strength properties. Similarly, the ANN also showed with high training R² values around 0.94, but a poor ability to generalize, resulting in considerably lower testing R² values of 0.4836, 0.5199, and 0.5424.