Demonstration of Real-Time Autonomous Metal Additive Manufacturing

Laser-directed energy deposition offers unique opportunities for fabricating and repairing high value metallic components. However, major barriers do exist for widespread adoption due to costly, modeldependent, trial-and-error experimentation-based qualification protocols. In this study, we demonstrated an autonomous framework for process optimization that integrates in-situ X-ray imaging with Gaussian Process-based active learning. Thin-wall structures of the Ni-base superalloy Mar-M247 were fabricated under systematically varied laser power and scan speed, and build quality was quantified using a new metric combining geometrical and structural conformity. Crack identification was achieved through a machine-learning pipeline by employing modern data science, AI and in-situ sensing. The autonomous system rapidly converged to optimal process parameters within 14 iterations, significantly fewer than conventional design-of-experiments approaches. Ex-situ x-ray diffraction (XRD) and scanning electron microscope (SEM) characterizations confirmed that cracking tendency and microstructural evolution are strongly correlated with thermal gradients and melt pool geometry. Thermomechanical simulations further validated the observed "Goldilocks" regime of laser power and scan speed required to minimize cracking while maintaining geometrical conformity. Importantly, we show that geometrical and structural conformity are coupled phenomena, governed by the same underlying thermal gradients and stress evolution. While current sensor resolution limits defect detection at the sub-micron scale, the framework demonstrates a scalable pathway for rapid qualification.