Tribological Optimization of AZ91/TiB₂ Magnesium Matrix Composites Using Stacked Ensemble Learning and Bayesian Optimization

This study investigates the tribological performance of AZ91 magnesium matrix composites reinforced
with TiB₂ particles, synthesized using the stir casting technique. TiB₂ reinforcements were incorporated at
2, 4, 6, and 8 wt.% to enhance wear resistance. The wear behavior was systematically analyzed by
varying normal load, sliding speed, and reinforcement content, structured through a Taguchi L18
orthogonal array. A Stacked Ensemble Learning (SEL) model, integrating multiple regression
architectures, was employed to predict wear rate and friction coefficient with high accuracy, ensuring
robust generalization across varying test conditions. Additionally, Bayesian Optimization Algorithm was
implemented for multi-objective optimization (MOO), effectively balancing wear resistance and
mechanical stability while leveraging probabilistic modeling for improved search efficiency. The
optimized results indicate that 6 wt.% TiB₂, under moderate loading conditions and controlled sliding
speeds, yields significant improvements in wear resistance due to the formation of a load bearing
tribolayer and increased dislocation density induced by TiB₂ particulates. The presence of TiB₂ acts as a
solid lubricant and hinders the plastic deformation of the soft AZ91 matrix, leading to reduced material
loss during sliding contact. Comparative validation between experimental and predicted results confirms
the model’s reliability, achieving a prediction error of <5%. The proposed hybrid methodology
integrating predictive analytics with Bayesian-driven optimization presents a computationally efficient
and experimentally validated framework for designing high-performance Mg-based composites with
superior tribological properties. This approach advances the development of lightweight structural
materials, reinforcing their applicability in aerospace, biomedical, and automotive sectors.