**Figure 9.**

*Open architecture 3D metal printer PANDA.*

These experiments will serve two purposes:


Experiments will be conducted in an open-structure metal AM printer, PANDA 11 (**Figure 9**). This open structure allows the system to monitor the build system and track temperature data for each layer. This system also enables closed-loop feedback operated online monitoring and control system. This is a part of the ongoing research that the authors are working currently. Details of the project can be found in recent publication of this research group [1]. This project involves understanding the physics of defect generation in metal powder bed fusion and using machine learning (ML) algorithm to implement the knowledge in automated process parameter selection to minimize printing defects. The machine learning algorithm under consideration is a graph based spatio-temporal convolutional neural network (ST-GCN) that will be trained using the results obtained in CFD and FEA modeling. The code will also incorporate genetic algorithm and/or game theoretic model to optimize the process parameters in order to minimize the defect generation. Once the ML code is trained and tested, it will be implemented using the device driver of the Open additive machine (shown in **Figure 9**). During operation, an online monitoring system using IR camera will be used to track the thermal history. This spatio-temporal temperature data will work as input to the ML algorithm, and finally using optimization theory, the device driver will receive information on optimized combination of process parameters, that will be activated for the next layer of printing. This is a novel idea for controlling the defect generation in metal additive manufacturing using process parameters and physics-based understanding of the process.
