Shape regulation of tapered microchannels in silica glass ablated by femtosecond laser with theoretical modeling and machine learning

Femtosecond laser processing is widely used in the micromachining of hard and brittle materials. Preparation of tapered microchannels with customizable cross-sections in silica glass using ultrafast lasers is of great significance in the field of microfluidic applications. In this paper, the width and depth of tapered microchannel in silica glass are predicted by combining theoretical modeling and machine learning. The functional relationship between laser processing parameters and microchannel width is obtained by theoretical modeling and introducing correction coefficients. The estimated model width is highly consistent with the experimental results. To solve the complex nonlinear mapping relationship between microchannel depth and processing parameters, a machine learning method based on a backpropagation neural network algorithm is proposed. By reasonably selecting model parameters, accurate prediction of microchannel depth is achieved with the corresponding average relative prediction error of 5.174%. Based on the proposed method, an effective parameter optimization strategy for achieving microchannels of specific sizes is developed. This method provides a new scheme for size prediction and controllable fabrication of silica glass microchannels with a femtosecond laser. Moreover, the proposed approach significantly reduces the time and cost of trial and error during actual processing and product development.

Automated summary

This paper predicts the width and depth of tapered microchannels in silica glass using a combination of theoretical modeling and machine learning. The study proposes a machine learning method to solve the complex nonlinear mapping relationship between microchannel depth and processing parameters. The developed approach provides an effective parameter optimization strategy for achieving microchannels of specific sizes.