A Study on Multi-sensor Data Transformation for Directed Energy Deposition Process

A Directed Energy Deposition (DED) process is a crucial additive manufacturing (AM) technology that enables metal part fabrication and repair. However, real-time monitoring and quality assessment remain challenging because the molten pool is sensitive to process parameters. This study proposed a sensor data transformation AI model to enhance immersive digital twin-based monitoring of the DED process. A 3D Convolutional Autoencoder was employed to transform vision-based molten pool images into temperature data, eliminating the need for direct temperature measurement. To improve model accuracy, a 2D Gaussian Weighted Squared Error loss function was introduced, focusing on the center features of the molten pool. The proposed AI model was validated through comparative evaluations, showing an improvement in temperature prediction accuracy. Such AI-driven data transformation could be integrated into a digital twin system to enable real-time visualization and remote monitoring without dedicated temperature sensors. This approach can enhance data efficiency, reduce hardware dependency, and facilitate immersive monitoring in VR/AR environments. Future work needs to extend the model’s applicability to various materials and process conditions to improve its generalization capability.