Data-driven prediction of nominal composition of a steel to copper multi material using hyperspectral in-situ monitoring of the Directed Energy Deposition process

Laser-based Directed Energy Deposition (DED-LB) enables the fabrication of multi-material and compositionally graded metallic components. Yet, ensuring accurate local composition during printing remains a major challenge, especially critical for systems with strongly dissimilar physical and thermo-physical properties. In this study, a hyperspectral in-situ monitoring approach is proposed to detect and quantify compositional variations during the DED-LB fabrication of 316L to CuCr 1 Zr multi-material structures. Hyperspectral imaging captures the spatially resolved spectral emission of the melt pool, providing a high-dimensional representation of its thermal and radiative state that is sensitive to changes in material composition. A data-driven model based on Convolutional Neural Networks (CNN) is trained to correlate hyperspectral signatures with the local compositional ratio under varying processing conditions, enabling real-time classification of compositional states during deposition. The proposed method demonstrates high accuracy in identifying compositional transitions and detecting deviations from the designed gradient, even in the presence of melt pool instabilities inherent to the Fe–Cu system. Post-process characterization confirms a strong agreement between hyperspectral predictions and measured elemental composition. These results highlight the potential of hyperspectral in-situ monitoring and CNN classification as a powerful tool for real-time composition sensing and closed-loop control in multi-material DED-LB, paving the way toward more reliable fabrication of functionally graded metallic components.