Machine learning-driven digital twin for real-time temperature prediction and adaptive process control in large-scale additive manufacturing

Temperature management in large-scale additive manufacturing (LSAM) plays an important role in interlayer bonding strength and print quality. This study develops a digital twin which predicts the real-time temperature and provides adaptive control of printing parameters in robotic LSAM of polymer structures. A finite-difference model was developed to simulate the thermal behavior under various print velocities, feed rates, and specimen dimensions and the results were validated against experimental temperature measurements using a thermal camera. Simulation data were used to train a machine learning model to predict the temperature of the newly deposited layer in real time based on input data of print geometry and process parameters. It was demonstrated how geometry and printing parameters can significantly impact interlayer temperature and, consequently, the overall printing quality and the interlayer bonding strength through experiments. Another machine learning model is trained to predict printing process parameters such as print speed and feeding rate based on the user input parameters (e.g. print geometrical parameters, etc.) to achieve the desired print quality. A digital twin is developed integrating the joint data from the robotic system to visualize the extruder trajectory, predict the temperature distribution of the newly deposited layer in real time as well as to adaptively control the printing process parameters during the printing process to achieve the proper interlayer temperature and desired print quality. The developed digital framework provides a foundation for real-time LSAM process monitoring, control and adjustment, offering new capabilities for improving print quality and minimizing defects in real time. • Finite-difference thermal model for LSAM validated against experiments with 5% error • Interlayer temperature strongly influences interlayer bonding and print quality. • ML model was developed for real-time temperature prediction in digital twin. • Data-driven digital twin enables real-time temperature vision and adaptive control. • Digital twin achieves desired interlayer temperature and raster width.