Determination of parameters for controlled laser cleaving of silicate glasses using regression, neural network and fuzzy models

This study proposes a solution to the applied research problem of predicting the characteristics of laser cleaving of silicate glasses. The results of a numerical experiment conducted in APDL (Ansys Parametric Design Language) were used to build regression, neural network and fuzzy models for the controlled laser cleaving of silicate glasses. The processing speed, radius, and power of the laser beam were considered as variable factors, whereas the maximum temperature and thermoelastic tensile stresses in the laser-treated area were regarded as responses. The regression model for the responses of laser cutting of glass plates at a specified significance level was estimated using the findings from the face-centered version of the central composite design experiment. Artificial neural networks that exhibit response dependence on input factors were created and trained. The most effective neural network models of the maximum temperature and thermoelastic tensile stresses in the laser-treated area were determined using MAPE (mean absolute percentage error) heat maps. Fuzzy modeling of controlled laser cleaving of silicate glasses was conducted according to the developed linguistic variables of input and output parameters. An evaluation was performed to compare the results of regression, neural network, and fuzzy modelling based on accuracy criteria, ultimately identifying the most effective model. The research findings can be suggested for practical application in approximating the maximum temperature and thermoelastic tensile stress in the laser-treated area.