Prediction of melt pool geometry in laser directed energy deposition considering non-steady-state melt pool behavior

Accurate prediction of melt pool geometry is critical for achieving high-quality deposition in laser-directed energy deposition (L-DED). Traditional statistical methods often fail to capture the complex, nonlinear relationships between process parameters and melt pool geometry, particularly under non-steady-state conditions, leading to limited prediction accuracy. To address this challenge, this study proposes a novel framework that quantifies melt pool fluctuations using standard deviation, leveraging coaxial and off-axis imaging to dynamically observe melt pool behavior, and integrating single-track cross-sectional measurements as primary data. A Bayesian calibration approach, enhanced by the Shapiro-Wilk test and Monte Carlo sampling, is employed to construct a prior distribution for standard deviation and refine the model with limited experimental data. The proposed Sequential Bayesian Calibrated Model (SBCM) is validated through thin-wall manufacturing experiments and thermal field simulations. Results demonstrate that SBCM significantly outperforms the statistical surrogate model, reducing the MAE by 80% − 90% for key metrics including melt pool height, dilution rate, and powder catchment efficiency. Specifically, the MAE for melt pool height decreased from 93.5 to 16.5 %, while temperature simulation accuracy improved by 5 %. This work highlights the potential of integrating non-steady-state melt pool behavior into predictive models, offering critical guidance for optimizing process parameters (e.g., layer spacing in thin-wall manufacturing) and advancing the precision of thermal field simulations in L-DED.