Exploratory analysis of the use of real-time 4D reconstruction during cold spray additive manufacturing

In-process spatio-temporal three-dimensional reconstruction, also known as 4D reconstruction, allows for early detection of deviations from the design in robotic additive manufacturing, thus providing the opportunity to rectify in an early stage to save both time and money. However, experiments have shown that in-process model reconstruction is made more difficult by the dynamic nature of the scene due to sensor and object movements, the three-dimensional growth of the build object, the textureless RGB quality of typical build surfaces, and computational complexity. This thesis proposes a framework to perform real-time, in-process 4D reconstruction for freeform additive manufacturing processes such as cold spray that deal with a real-time dynamic and evolving scene that includes object change. Temporal point clouds from three cameras in eye-in-hand configurations are acquired and segmented to extract the region of interest (build object). Then the multiview, multi-camera registration of segmented 3D points is addressed in real-time via combining geometrically constrained Fiducial marker tracking and plane-based registration without drift accumulation. Finally, the registered point clouds are fused via voxel fusion of growing parts to reconstruct the 3D model of the object with smoothened surfaces. The solution is deployed and verified in a robotic cold spray cell with different test scenarios and shape complexities. The results have shown a 3D registration accuracy of up to RMSE 0.923mm translation and 0.801◦ angular in real-time performance with a 60fps frame rate. Furthermore, the capability to reconstruct a high-quality model been demonstrated by using the low-cost sensor. In addition, the reconstructed 4D model has been used to analyse the deformation error of the cold spray deposited object.