Development of a Deep Learning Model for Predicting and Correcting Sintering Deformation of Printed Metal Components

While a number of different technologies for 3D printing metal parts exist, all of them rely on forming parts sequentially out of individual metal powder particles. They are thus prone to producing parts that initially contain microscale gaps, bubbles and other weaknesses. In order to correct for this and produce reliable, production-quality parts, the printed components must be subjected to sintering. In this process, the parts are heated to near-melting, closing gaps and fusing discontinuities. As a result, the parts shrink and deform, often in complex nonlinear patterns. We propose research into developing a deep network model for predicting and compensating for this deformation. In phase I we intend to carry out a feasibility study using a single metal and printing technology, gathering data that can be used to develop a baseline model for the deformation process. This can then be used to develop a model that automatically pre-deforms part designs, such that the final sintered parts will precisely match the original design. In future work, more data can be collected in order to expand the model to more varied printing modalities.