About this Abstract |
Meeting |
2021 Annual International Solid Freeform Fabrication Symposium (SFF Symp 2021)
|
Symposium
|
Process Development
|
Presentation Title |
Image Registration and Matching Error in 2D and 3D for Laser Powder Bed Fusion |
Author(s) |
Andrew Lang, Cesar Ortiz Rios, Joseph Newkirk, Robert Landers, James Castle, Douglas Bristow |
On-Site Speaker (Planned) |
Andrew Lang |
Abstract Scope |
This work outlines a method to register 2D and 3D images taken post-process and in situ from 301L stainless steel parts printed by Laser Powder Bed Fusion. The process uses DREAM.3D, an open source software that provides for data transport in a non-proprietary format. The Robust Automatic Threshold selection technique is used to create a boundary point cloud of the part from each image. The Iterative Closest Point technique is applied to the point clouds for both 2D images and 3D image stacks to create an affine transformation matrix for registration. Multiple 2D SEM images of the same sampled layer are taken under different settings and imaging conditions and registered to a common target. Images from post-process X-ray Computed Tomography and an in situ short-wave infrared camera create 3D image stacks, which are directly registered in 3D space. Registration accuracy is validated by creating a correspondence list of the closest point in the registered point clouds and the matching error is calculated using mean average error. Mean average error is computed using point-to-point and point-to-plane methods; the point-to-plane method is shown to be more reliable. Finally, the registered 3D images are down sampled to the lower resolution image dimensions and fused to the nearest point to create an array containing corresponding in situ and post-process data. |
Proceedings Inclusion? |
Definite: Post-meeting proceedings |