Prediction of process instability by WAAM in-process monitoring and CTWD drift estimation

Résumé fourni par la source

Wire Arc Additive Manufacturing (WAAM) is a promising technology to produce large metallic components. However, drifts of the Contact Tube to Workpiece Distance (CTWD) are difficult to manage, which can induce process instability and defects like porosity. To safeguard material integrity, instability must be detected before it occurs, which is very challenging. This paper proposes an original predictive model for process instability caused by CTWD drift. In-process monitoring data were collected on a robotic cell and sixteen key features were extracted as potential monitoring criteria for CTWD drift estimation. Feature selection was then performed by several filtering methods, to identify the best indicator for CTWD drift estimation through regression model. ANOVA was used to identify which Key Process Parameters (KPP) should be considered for finetuning the linear regression model. Then, Support Vector Regression (SVR) model was developed to predict the boundary of instability T⁎, based on dedicated KPP identified by ANOVA. By combining the estimated CTWD drift and drift speed with the predicted boundary T⁎, the model predicts the Number of Remaining Layers (n RL ) before process instability onset. The proposed method was evaluated on a use-case of thin wall manufacturing. Instability predictions aligned with observations, demonstrating the model capability to anticipate defects. The instability consequences on material integrity were confirmed by metallographic examination. The combination of real-time monitoring, data-driven models and eXplainable AI, provides a robust and interpretable framework to improve WAAM control, while providing manufacturing experts with new insight on the KPP for instability prediction and proactive intervention. • Challenge of detecting instability before it occurs, to preserve material integrity • An original predictive model based on monitoring and CTWD drift estimation • New understanding of welding controller response to CTWD drift by physics-based features • Good results by eXplainable IA, including for unseen process parameters