Machine Learning for Scanning Path Prediction in Laser Forming - Application of Structured Patterns and CNN.

For prototype design and small batch production, laser forming has more advantages for manufacturing metal sheet products than conventional processes. However, predicting and optimizing the required conditions for the product's final shape is difficult, especially in multi-stage forming, where many governing factors affect the shape intricately. In this study, the convolution neural network (CNN) is proposed to simulate the correlations between the scanning paths and the final deformations of a metal sheet. The imaginary data, which used values to present the distribution of deformation height of metal sheets, was used to examine the feasibility of applying CNN. On the other hand, the simulated images of a structured pattern projected on the sheet surface were used to train and test the CNN. The results demonstrate that CNN can use imaginary data in the training dataset to predict the scanning path determined by two points on the edge of the metal sheet with high accuracy. Although the performance of the test dataset needed to be better to prove the general-purpose ability, this research validates the feasibility of applying CNN for scanning path prediction in laser forming. [ABSTRACT FROM AUTHOR]