A Hybrid Bayesian-Gaussian Machine Learning Approach for Modelling Cutting Dynamics in Drilling of Al-TiB 2 Composites

The present study considers a physics-informed machine learning (PIML) framework to analyse the mechanical cutting dynamics during radial drilling of Al-TiB 2 metal matrix composites. During machining, metal matrix composites (MMCs) show high tool vibration along with fluctuating cutting forces due to their heterogeneous structure and abrasive reinforcement particles. These effects make it difficult to achieve consistent dimensional accuracy in radial drilling. Sixty fractional factorial drilling trials with varying process parameters, i.e., the feed rate and the spindle speed, are conducted on Al-TiB 2 composites having reinforcements up to 5.5 wt.%. Cutting force signals from these runs are processed to get four target variables, viz., maximum entry impact force, steady-state cutting force, active force variance, and diametric hole expansion. Instead of treating this process as a purely data-driven one and directly using conventional algorithms, a separate prediction of the deterministic mechanics and random micro-chatter during drilling is done. A modified version of the Kienzle power law is used for pre-processing the process parameters. On these parameters, Bayesian ridge regression is used to predict the physical baseline. The remaining non-linear chatter is predicted using a decoupled multi-kernel Gaussian process regression (MK-GPR) with a Matern covariance function. The whole process was conducted within a strict Leave-One-Out Cross-Validation (LOOCV) loop so that no data leak happens across the 60 run dataset. The final error metrics are tested against conventional algorithms like Random Forest, XGBoost, SVR, and MLP. The proposed framework achieves an average R 2 of 0.75 and outperforms the other methods by cutting absolute prediction errors (RMSE) by 14% to 39%. The relationship between steady-state cutting force and process parameters is also studied using the dataset. This variation is validated using the micrographs of the chip morphology. A combined use of statistical modelling and machining physics allows more reliable and interpretable optimisation of drilling parameters in MMCs.