CPS: Medium: Collaborative Research: AI-Enabled Modeling, Sensing, Digital Twins, and Autonomous Control for Smart, Sustainable, and Scalable Big Area Additive Manufacturing

The ability to produce large, high-performance metal components is key for the operational readiness of the US energy, aerospace and naval systems. Laser wire directed energy deposition is a relatively new large-scale 3D printing technology for rapid production of large, complex metal components. Despite its potential, the industrial adoption of this technology is hampered by concerns related to quality, reliability, and safety. The process may introduce defects that can cause sudden and premature failure of the component during operation. To mitigate this problem, this project introduces a new intelligent manufacturing approach for detecting and correcting defects in large and complex metal parts made using the laser wire directed energy deposition process. New artificial intelligence (AI) algorithms, powered by fast computer simulations and data from sensors inside the machine will detect, suppress and correct defects in the part on-the-fly as it is being manufactured. This research will thus enable US manufacturing industry to deploy laser wire directed energy deposition to manufacture parts first-time-right with zero defects, and as a result strengthen the domestic manufacturing and supply of critical components across defense, energy, and aerospace sectors. Additionally, the project will train a future workforce at the intersection of AI, materials science and advanced manufacturing. The goal of this project is to establish an unified cyber-physical systems framework for the real-time prediction and control of microstructure and residual stresses in large metal parts manufactured using the laser wire directed energy deposition process. In pursuit of this goal, this inter-disciplinary research will integrate the following three elements encompassing advanced manufacturing, physics-based modeling, materials science, in-situ sensing, machine learning, and control theory: - Mirror-as-you-build: Physics-informed machine learning models will be developed to rapidly predict process outcomes and optimize process parameters, such as laser power and scan speed, and toolpaths before fabrication. This enables feedforward control to achieve desired microstructure and minimize residual stress. - Qualify-as-you-build: Digital twin models will combine process simulations with real-time sensor data (thermal, optical, and acoustic) to predict microstructure and residual stresses during fabrication. - Correct-as-you-build: Autonomous closed-loop control strategies will adjust process parameters in real-time to correct for stochastic variations. Defective regions that cannot be rectified by adjusting parameters, will be removed and re-deposited using hybrid additive–subtractive manufacturing. The scientific contribution of this work is a unified, interpretable framework for linking process parameters, process physics, real-time sensing data, and material outcomes, enabling reliable, first-time-right manufacturing of high-quality and operationally safe components using laser wire directed energy deposition. The methods developed are generalizable and can be extended to similar material fusion-based additive manufacturing processes. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.