Intelligent Additive Manufacturing - Metals

Texas Research Institute Austin (TRI) and the University of Texas at Austin (UT) will implement real-time process control within an existing metal AM process to improve the consistency of fabrication conditions and the mechanical properties of sintered structures. The proposed approach is based on a surrogate modeling control system developed by Professor Joseph Beaman’s lab at UT, and uses in-situ infrared temperature measurements to create surrogate sintering models that predict optimal laser power with real-time data. The surrogate models are dynamically generated since they use information acquired from each layer to improve their predictive capabilities. Using real-time data to update these models allows the control system to maintain optimal laser power by adapting to the dynamic fabrication environment during the sintering process. However, the surrogate modeling approach requires location-specific information to train the model because the ideal laser conditions are time- and location-specific. Therefore, to improve the surrogate model and increase the uniformity of AM metal parts, we will adapt the feed-forward technique to include an artificial neural network (NN) that will allow the IR measurements acquired in one location to be generalized and applied to other positions while fabrication is underway. This approach will reduce the amount of data required to train the system and produce a more robust controller.