Machine learning prediction and hybrid GRA–AHP optimization of AWJM parameters of ultrasonic assisted stir cast Al6061–B$$_4$$C–ZrO$$_2$$ composites

Abstract

This study investigates the abrasive water jet machining (AWJM) performance of an Al6061-0.5 wt.% B[Formula: see text]C-1 wt.% ZrO[Formula: see text] hybrid composite fabricated using ultrasonic-assisted stir casting. A Taguchi L27 orthogonal array was adopted to systematically evaluate the effects of five machining parameters, namely abrasive flow rate (AFR), water jet pressure (WJP), abrasive jet cutting speed (AJCS), stand-off distance (SOD), and abrasive grit size (GS) each at three levels. Material removal rate (MRR), surface roughness (Ra), and kerf taper angle (KTA) were considered as the key performance responses. The experimental results revealed that MRR varied from 7.86 to 15.24 mm[Formula: see text]/min, Ra ranged between 3.220 and 3.980 μm, and KTA varied from 0.142[Formula: see text] to 0.309[Formula: see text]. Analysis of variance (ANOVA) was performed to identify the statistically significant machining parameters influencing the responses, indicating that AFR was the dominant factor affecting MRR, whereas AJCS predominantly governed Ra and KTA. Furthermore, a hybrid Grey Relational Analysis-Analytic Hierarchy Process (GRA-AHP) multi-criteria optimization approach was employed to simultaneously maximize MRR and minimize Ra and KTA. The optimal machining condition was obtained at an AFR of 430 g/min, WJP of 280 MPa, AJCS of 80 mm/min, SOD of 1.5 mm, and GS of 120 mesh. The findings demonstrate the effectiveness of the proposed hybrid optimization framework in enhancing AWJM performance of advanced aluminium-based hybrid composites. Also, machine learning (ML) models, including Support Vector Regression (SVR), Random Forest (RF), and Multi-layer perceptron (MLP), were developed to predict machining output responses - MRR, Ra, and KTA based on AWJM process parameters. Among the developed models, the RF model demonstrated superior predictive capability for MRR and Ra with maximum [Formula: see text] values of 0.9882 and 0.9919, respectively, whereas the SVR model achieved the highest prediction accuracy for KTA with an [Formula: see text] value of 0.9955. The low RMSE and MAPE values further confirmed the robustness and reliability of the developed ML models for AWJM response prediction. The models exhibited high predictive accuracy with strong agreement between experimental and predicted results, even with a limited dataset. These results demonstrate the effectiveness of ML-based approaches as reliable tools for performance estimation and process optimization in AWJM of hybrid aluminium matrix composites.