Geometry-Based Radiation Prediction of Laser Exposure Area for Laser Powder Bed Fusion Using Deep Learning

Abstract

Laser powder bed fusion (LPBF) is a promising technique used to manufacture complex geometries in a layer-wised manner. Radiation during the LPBF process is influenced by the part geometry, e.g., the overhang angle and the wall thickness. Locally varying radiation can cause deformation of the product after manufacturing. Thus, the prediction of the geometry-caused radiation before the manufacturing can support the evaluation of the design printability to achieve first-time-right printing. In this paper, we present a framework to predict the geometry-based radiation information using a deep learning (DL) algorithm based on the part geometry from computer-aided design (CAD). The algorithm was trained using data from an LPBF-print job consisting of parts with varying overhang angles. Image data, which include the information of radiation, were captured with an optical tomography (OT) camera system that was installed on a LPBF machine used in a laboratory environment. For the DL algorithm, a U-Net based network with mean absolute error (MAE) loss was applied. The training input was binarized OT data representing the contour of the designed geometry. Complementary, the OT data were used as ground truth for the model training. For the application, the design contours of multiple layers were extracted from the CAD file. The result shows the applicability to predict the OT-like radiation by its contour, which has the possibility to show the anomaly due to the part geometry.

Keywords

• laser powder bed fusion
• laser radiation prediction
• deep learning
• U-Net