ANN modeling and GA-Taguchi approach for multi-objective optimization of ultrasonic vibration-assisted micromachining of quartz

Abstract

Ultrasonic vibration-assisted electrochemical discharge machining (UV-ECDM) is frequently employed for micro hole fabrication of quartz, considering its capabilities to fabricate simple to complex geometries with high precision, minimal heat-affected zone, and improved flushing. However, the integration of nonlinear predictive modeling with evolutionary multi-objective optimization remains unaddressed. This study develops a unified modeling-optimization framework for the UV-ECDM of quartz. A Taguchi L₉ design was employed to evaluate the electrolyte concentration (20–25 wt.% NaOH), applied voltage (45–55 V), and tool feed rate (0.4830–0.6762 µm/s). UV-ECDM reduced hole diameter by 20.10% and improved depth by 7.42% compared to conventional ECDM. A feedforward artificial neural network (ANN) successfully captured the nonlinear process behavior with cumulative R² ≈ 0.97 and R ≈ 0.995, demonstrating high predictive fidelity. The ANN model was integrated with a multi-objective genetic algorithm (MOGA) that recommended optimal parameters at 48.5 V, 22.3 wt.% NaOH, 0.62 µm/s, predicting maximum depth of 1125 µm and minimum diameter of 1055 µm. Confirmatory trials verified the optimization framework, demonstrating an enhanced depth by 2.86% and 5.29% reduction in diameter. The proposed integrated ANN-MOGA methodology transforms UV-ECDM from a parameter-sensitive experimental process into a quantitatively optimizable micro-machining strategy for brittle insulating materials.