PHYSICS

Study and optimization of laser cleaving of crystalline quartz using genetic algorithm, neural network and neuro-fuzzy models

Yu. V. Nikityuk, L. N. Marchenko, A. N. Serdyukov

Francisk Skorina Gomel State University

Abstract: The study focuses on developing a metamodel for the laser cleaving process of crystalline quartz, encompassing modeling and optimization. Using a finite element model implemented in the APDL programming language, temperature fields and thermoelastic stress fields were determined. These fields arise in a monocrystalline quartz plate due to sequential laser heating and coolant exposure, analyzed for three distinct variants: I – analysis of the $ZY$-plane cross-section with laser beam movement along the $X$-axis; II – analysis of the $YX$-plane cross-section with laser beam movement along the $X$-axis; III – analysis of the $XY$-plane cross-section with laser beam movement along the $Z$-axis. A central composite design was employed to conduct a numerical experiment, where the factors included processing speed, geometric parameters of the elliptical laser beam, $\mathrm{CO}_2$ laser power, and quartz plate thickness. According to the experimental design, calculations were performed for $27$ factor combinations, determining the maximum temperature values ($T_1$, $T_2$, $T_3$) for three processing variants of a square quartz plate, along with three corresponding values of maximum tensile stress ($S_1$, $S_2$, $S_3$) acting perpendicular to the laser-induced crack fronts. The optimal artificial neural network architectures were identified for predicting maximum temperatures and thermoelastic stresses in the laser processing zone of crystalline quartz using TensorFlow. Neuro-fuzzy models were developed in the ANFIS framework, followed by a comparative analysis of neural network and neuro-fuzzy approaches. Furthermore, the most effective input parameters for laser cleaving of crystalline quartz were determined through optimization using the MOGA.

Keywords: laser cutting, Artificial Neural Networks, adaptive network-based fuzzy inference system, MOGA optimization genetic algorithm, ANSYS program.

UDC: 535.33:539.3.075:621.373.8:621.9.048.7:549.6

Received: 08.07.2025

Language: English

DOI: 10.54341/20778708_2025_4_65_29