딥러닝 기법을 이용한 펨토초 레이저 가공 홀의 깊이 예측

초록

This paper proposes an artificial intelligence learning method using deep learning to analyze the correlation between 2D images measured through machine vision and the depth of laser-drilled holes. Machined holes measured through machine vision become darker as the depth increases due to the keyhole effect. The deep learning model predicts the depth of the drilled hole from the diameter and brightness data of the hole by analyzing the 2D image data. The 2D image data of the machined hole is processed to be suitable for deep learning model training by adjusting the size of the image, pixel values, and removing outliers. The deep learning model's architecture, learning algorithm, and hyperparameters are selected to predict the depth of the hole by training the machined hole's image and diameter data. Subsequently, the performance of the trained deep learning model is evaluated through performance metrics during the test process, and the optimal model is determined. The proposed model in this paper is expected to predict the real-time machining status with high accuracy, reduce process time and operating costs, and increase productivity in manufacturing sites.