A Machine-Learning-Based Approach to Critical Geometrical Feature Identification and Segmentation in Additive Manufacturing

Abstract

Additive manufacturing (AM) processes offer a good opportunity to manufacture three- dimensional objects using various materials. However, many of the processes, notably laser Powder bed fusion, face limitations in manufacturing specific geometrical features due to their physical constraints, such as the thermal conductivity of the surrounding medium, the internal stresses, and the warpage or weight of the part being manufactured. This work investigates the opportunity to use machine learning algorithms in order to identify hard-to-manufacture geometrical features. The segmentation of these features from the main body of the part permits the application of different manufacturing strategies to improve the overall manufacturability. After selecting features that are particularly problematic during laser powder bed fusion using stainless steel, an algorithm is trained using simple geometries, which permits the identification of hard-to-manufacture features on new parts with a success rate of 88%, showing the potential of this approach.

Keywords

• SLM
• LPBF
• scanning strategies
• machine learning
• segmentation