Low-Permeability bentonite–ash–cement composites: Optimization using Taguchi design, machine learning, and microstructural characterization

This study explores the development of sustainable bentonite–ash–cement composites for low-permeability barrier applications through a Taguchi L27 experimental design. Five parameters—bentonite content (0–10%), incineration ash content (0–10%), water-to-cement ratio (0.40–0.50), curing period (7–28 days), and compaction pressure (10–20 MPa)—were systematically varied to assess their effects on unconfined compressive strength (UCS), hydraulic conductivity (k), and plasticity index (PI). The optimized mix (L26: B3A3W2C1P2) achieved superior performance, with a UCS of 35.47 MPa, k = 3.20 × 10−7 cm/s, and PI = 12.44%, whereas the least effective mix (L1) recorded a UCS of 22.85 MPa, k = 5.00 × 10−7 cm/s, and PI = 10.01%. Microstructural analysis via SEM-EDS and XRD revealed that the optimal mix developed a dense matrix enriched with well-formed C–S–H gels and pozzolanic reaction products, while the poorest mix showed unreacted ash and a porous structure. Additionally, machine learning models were employed to predict UCS, with the Random Forest model demonstrating the highest accuracy (R2 = 0.97, RMSE = 0.62 MPa). The findings highlight the synergistic effect of bentonite and ash in enhancing mechanical strength and impermeability, and underscore the potential of machine learning to streamline mix design.