A novel predictive model for abrasive waterjet deep hole drilling on AL7075 T6 using machine learning and...

Abrasive Waterjet (AWJ) is a promising non-traditional method for precision cutting of aerospace materials like AL7075 T6. This work explores AL7075 T6 AWJ-based Deep Hole Drilling (DHD) using a full factorial design with accurate modeling and optimization performed through machine learning and evolutionary algorithms. The objective is to investigate the influence of process parameters and to model, and predict the optimal AWJ-DHD settings, such as waterjet pressure, standoff distance, and abrasive mass flow rate, on drilling qualities including geometrical and dimensional precision (kerf angle, kerf ratio), surface roughness, and drilling efficiency. Four machine learning models, Adaptive Boosted Regression (ABR), Extreme Gradient Boosting (XGB), Decision Tree (DT), and Random Forest (RF) were developed with experimental data to enhance prediction accuracy and process efficiency. Among the developed models, RF had the lowest testing error value for all responses with root mean square values of 0.046 (kerf angle), 0.0078 (kerf ratio), 0.044 (surface roughness), and 0.027 (drilling rate). Moth-Flame Optimization (MFO), Differential Evolution (DE), and Sine Cosine Algorithm (SCA) were used for multi-response optimization of AWJ deep hole drilling parameters. The optimal algorithm for each response was selected using Deng's similarity-based ranking. The ranking revealed SCA algorithm outperformed MFO and DE. The SCA algorithm discovered optimal parameter setting for AWJ-DHD as a water pressure of 350 MPa, standoff distance of 1.5 mm, and an abrasive mass flow rate of 300 g/min. Under these conditions, the predicted responses were a kerf angle of 0.048⁰, kerf ratio of 0.011, a surface roughness of 1.438 μm, and a drilling rate 0.769 mm/s. The validation trials using optimized parameters yielded a kerf angle of 0.047⁰, a kerf ratio of 0.066, a surface roughness of 1.40 μm, and a drilling rate of 0.769 mm/s, with percentage variations of 2.08%, 3.03%, 2.14%, and 2.65%, respectively, thereby demonstrating the efficiency of the developed machine learning model and optimization technique. The integrated machine learning and evolutionary algorithm framework improved drilling efficiency and hole quality by minimizing surface roughness.