CNN-Based Wire Monitoring in LW-DED: Correlation with Surface Metrology

Abstract Laser Wire Directed Energy Deposition (LW-DED) is a high-precision additive manufacturing method known for material efficiency and high deposition rates; however, real-time monitoring remains challenging due to complex interactions between the laser, melt pool, and wire, especially during multilayer deposition. This study presents a real-time wire monitoring approach for multilayer LW-DED of Inconel 625 to predict surface waviness. Using a constant linear energy density, 10-layer wall structures were fabricated. A convolutional neural network model achieved 81.11% mAP50–95 and over 59 frame per second for wire detection, while an artificial neural network, using wire features and process parameters, predicted Wp10 waviness with a 33.54 µm RMSE and R 2 greater than 80%. The results confirm the system’s effectiveness in monitoring and surface quality prediction, offering a promising solution for quality control in multilayer LW-DED.