Harnessing Physics-Informed AI for Cold Spray Shape Prediction: A Path to Improved Additive Manufacturing and Repair

Cold Spray is a solid-state deposition technology that offers the ability to fabricate large deposits with minimal thermal effects, making it suitable for coating, repair and additive manufacturing of high-value components. Despite its advantages, a challenge to the wider application of cold spray is the difficulty in accurately predicting and, therefore, controlling the shape of the deposits it produces. Physical observations of the material build-up process during spraying have led to the development of analytical models that describe deposit shape through partial differential equations. Albeit accurate, such models require significant computational effort, particularly when simulating complex spraying trajectories, and depend on experimental data for the calibration of their parameters. In this study, Physics-Informed Neural Networks, a recent machine learning technique, were introduced to address the task of cold spray shape prediction. These networks can embed physical laws, represented through partial differential equations, while also estimating unknown parameters using limited experimental data. This approach has the potential to serve as a unified framework for simulating cold spray deposits across various working conditions, including different feedstock materials and substrate geometries. By doing so, it can enhance the optimization of both additive manufacturing and repair processes.