Hole Depth Prediction in a Femtosecond Laser Drilling Process Using Deep Learning

Abstract

In high-aspect ratio laser drilling, many laser and optical parameters can be controlled, including the high-laser beam fluence and number of drilling process cycles. Measurement of the drilled hole depth is occasionally difficult or time consuming, especially during machining processes. This study aimed to estimate the drilled hole depth in high-aspect ratio laser drilling by using captured two-dimensional (2D) hole images. The measuring conditions included light brightness, light exposure time, and gamma value. In this study, a method for predicting the depth of a machined hole by using a deep learning methodology was devised. Adjusting the laser power and the number of processing cycles for blind hole generation and image analysis yielded optimal conditions. Furthermore, to forecast the form of the machined hole, we identified the best circumstances based on changes in the exposure duration and gamma value of the microscope, which is a 2D image measurement instrument. After extracting the data frame by detecting the contrast data of the hole by using an interferometer, the hole depth was predicted using a deep neural network with a precision of within 5 μm for a hole within 100 μm.

Keywords

• laser micromachining
• deep learning
• femtosecond laser
• silicon nitride
• blind hole