| Abstract Scope |
Rapid production of large-scale components has emerged as an advantage of additive manufacturing (AM). Compared to longer lead times seen with forgings and castings, additive manufacturing techniques such as wire-arc directed energy deposition (DED) achieve deposition rates up to 4kg/hour, enabling the production of large structures within hours. Wire-arc DED employs a robotic cell and the gas metal arc welding process to print complex, near-net-shape components. This robotic cell is integrated with a digital twin to program toolpaths and simulate the pre-deposition process. The printing of a complex component with the robotic cell used in this work is demonstrated with the printing of a flange. An AI tool predicting the optimal stepover distance for AM is used for toolpath development. A low droplet contact period waveform was used for smooth deposition and a low wire feed speed was used for low heat input and reduced porosity. Determining these toolpaths and welding parameters has been largely dependent on human experience and heuristic decisions. However, an optimal toolpath is essential for printing a quality component, as various scanning strategies result in distinct microstructures and mechanical properties. Thus, this work develops AI-driven strategies for determining the optimal printing strategies (raster, contour, etc.) for AM and cladding applications within the digital twin environment. Using Design of Experiments (DOE) and statistical mapping techniques, 32 experiments are fitted to regression models, then used to generate 500 synthetic data. A neural network is then trained on the augmented data set with the objective of learning the toolpath strategies for optimizing productivity and deposition quality based on the welding parameters (wire feed speed and travel speed). Finally, a Python-based graphical user interface (GUI) is developed, allowing users to input shape, wire feed speed, and travel speed and receive an output of the optimal toolpath strategy. |