(Machine learning-assisted optimization of multifunctional Epoxy/MWCNT–nanoclay nanocomposites for fire-safe, self-sensing coatings)

Abstract

This study develops a validated machine learning-assisted design framework for multifunctional epoxy/MWCNT/nanoclay hybrid coatings integrating fire safety, mechanical robustness, electrical conductivity, UV shielding, and self-sensing capability. A four-factor Box–Behnken design was coupled with response surface methodology (RSM), Random Forest (RF) modeling, SHAP interpretability, and multi-objective desirability optimization to identify a balanced formulation window rather than optimizing individual properties separately. Compared with RSM, the RF model provided substantially improved predictive accuracy, achieving R2 values of 0.986, 0.995, and 0.998 for LOI, tensile strength, and log₁₀(electrical conductivity), respectively. The optimized hybrid coating exhibited LOI of 28.0 ± 0.2%, reduced pHRR of 720 ± 9 kW m⁻2, tensile strength of 80.0 ± 1.4 MPa, flexural modulus of 2.80 ± 0.14 GPa, electrical conductivity of ~1.0 × 10⁻4 S m⁻1, thermal sensing response of 8.0 ± 0.5%, and NH₃ response of 12.0 ± 0.9%, with prediction errors generally below 5%. The machine learning-assisted workflow reduced the experimental effort by approximately 50–65% compared with full-factorial exploration while maintaining experimental reliability through repeated cross-validation and independent validation. Structural analyses confirmed that intercalated/exfoliated nanoclay domains, CNT conductive pathways, and strengthened epoxy–filler interfaces jointly govern the multifunctional performance. This work demonstrates a reproducible data-driven strategy for designing fire-safe, durable, and self-sensing epoxy nanocomposite coatings.