Explainable ensemble learning with Bayesian Optimization for microstructure-driven tensile strength prediction and design optimization of Al–SiC metal matrix composites
5 affiliations
- 1Department of Research and Innovation, Saveetha School of Engineering, SIMATS, Chennai - 602105, Tamil Nadu, India
- 2Department of Computer Science and Engineering, Vemu Institute of Technology, P.Kothakota, Chittoor District, Andhra Pradesh -517112
- 3Department of Mechanical Engineering, KCG College of Technology, Karapakkam, Chennai-600097, Tamil Nadu, India
- 4Division of Research and Development, Lovely Professional University, Jalandhar - Delhi G.T.Road, Phagwara, Punjab (India) – 144411
- 5Lloyd Institute of Engineering & Technology, Plot No. 3, Knowledge Park II, Greater Noida, Uttar Pradesh 201306, India
Abstract
The advancement of Al–SiC metal matrix composites (MMCs) is crucial for modern structural and aerospace applications because of their high stiffness, thermal stability, and lightweight characteristics. However, conventional optimization and experimental approaches used to achieve maximum mechanical strength are labor-intensive, expensive, and often limited by nonlinear interactions among microstructural parameters. To address these limitations, this study proposes the Metal Matrix Composite Optimization Network (MMCO-Net), a hybrid artificial intelligence-driven ensemble framework integrating Random Forest (RF) and Extreme Gradient Boosting Regression (XGBR) for accurate prediction of Ultimate Tensile Strength (UTS). The proposed MMCO-Net employs a weighted fusion mechanism that adaptively combines model contributions through cross-validation, achieving high predictive accuracy with R 2 = 0.98, RMSE = 4.6 MPa, and MAE = 3.8 MPa. Bayesian Optimization (BO) was incorporated to optimize the key microstructural parameters, namely SiC weight fraction (x 1 ) and porosity (x 2 ), in order to maximize UTS. The optimization process identified an optimal configuration of 38 wt% SiC and 1.6% porosity, resulting in a predicted peak UTS of 326.4 MPa. Furthermore, SHAP-based explainability analysis revealed that SiC weight fraction contributed most positively toward tensile strength, whereas porosity exhibited the strongest negative influence. Comparative analysis demonstrated that MMCO-Net outperformed conventional Random Forest, Linear Regression, and CatBoost models in prediction accuracy and stability. The proposed MMCO-Net framework significantly reduces dependence on exhaustive experimental iterations and establishes a strong foundation for future multi-property prediction and intelligent composite material engineering.
Related
- Explainable CatBoost–TabPFN Hybrid for Mode-Resolved Interlaminar Fracture Toughness Prediction in Fiber-Reinforced Polymer Composites
- Study of the Microstructure and Ferroelectric and Magnetic Properties of Three-Component Multiferroic Composites.
- Investigation of the influence of nutrient additives and the role of microstructure on the unconfined compression behaviour of mycelium composites
- ExNCDIA: an explainable ensemble learning framework for predicting non-chemotherapy drug-induced agranulocytosis and interpreting structural associations
- Predicting continuous load-settlement trajectories of bored piles: A GA-optimized weighted ensemble framework with explainable AI
BibTeX
@article{natrayan2026explainable,
title = {{Explainable ensemble learning with Bayesian Optimization for microstructure-driven tensile strength prediction and design optimization of Al–SiC metal matrix composites}},
author = {Lakshmaiya Natrayan and Putta Nageswara rao and P. Hari Chandra Prasad and Seeniappan Kaliappan and Ramya Maranan and Anand Rajendran},
journal = {Journal of Materials Research and Technology},
year = {2026},
month = jul,
doi = {10.1016/j.jmrt.2026.07.030},
url = {https://doi.org/10.1016/j.jmrt.2026.07.030},
note = {CC BY-NC-ND},
}
RIS
TY - JOUR
TI - Explainable ensemble learning with Bayesian Optimization for microstructure-driven tensile strength prediction and design optimization of Al–SiC metal matrix composites
AU - Lakshmaiya Natrayan
AU - Putta Nageswara rao
AU - P. Hari Chandra Prasad
AU - Seeniappan Kaliappan
AU - Ramya Maranan
AU - Anand Rajendran
PY - 2026
DA - 2026/07/01
JO - Journal of Materials Research and Technology
DO - 10.1016/j.jmrt.2026.07.030
UR - https://doi.org/10.1016/j.jmrt.2026.07.030
AB - The advancement of Al–SiC metal matrix composites (MMCs) is crucial for modern structural and aerospace applications because of their high stiffness, thermal stability, and lightweight characteristics. However, conventional optimization and experimental approaches used to achieve maximum mechanical strength are labor-intensive, expensive, and often limited by nonlinear interactions among microstructural parameters. To address these limitations, this study proposes the Metal Matrix Composite Optimization Network (MMCO-Net), a hybrid artificial intelligence-driven ensemble framework integrating Random Forest (RF) and Extreme Gradient Boosting Regression (XGBR) for accurate prediction of Ultimate Tensile Strength (UTS). The proposed MMCO-Net employs a weighted fusion mechanism that adaptively combines model contributions through cross-validation, achieving high predictive accuracy with R 2 = 0.98, RMSE = 4.6 MPa, and MAE = 3.8 MPa. Bayesian Optimization (BO) was incorporated to optimize the key microstructural parameters, namely SiC weight fraction (x 1 ) and porosity (x 2 ), in order to maximize UTS. The optimization process identified an optimal configuration of 38 wt% SiC and 1.6% porosity, resulting in a predicted peak UTS of 326.4 MPa. Furthermore, SHAP-based explainability analysis revealed that SiC weight fraction contributed most positively toward tensile strength, whereas porosity exhibited the strongest negative influence. Comparative analysis demonstrated that MMCO-Net outperformed conventional Random Forest, Linear Regression, and CatBoost models in prediction accuracy and stability. The proposed MMCO-Net framework significantly reduces dependence on exhaustive experimental iterations and establishes a strong foundation for future multi-property prediction and intelligent composite material engineering.
SN - 2214-0697
SN - 2238-7854
ER -