Optimization of Process Parameters for Binder Jetting 3D Printing of GH2132 Superalloy Based on Machine Learning

<title>Abstract</title> <p> Binder jetting 3D printing (BJ3DP) offers significant advantages in fabricating complex metal components, yet the relative density and dimensional accuracy of green parts depend heavily on process parameters. This study investigates the effects of layer thickness, white ink concentration, roller traverse speed, and roller rotation speed on the relative density and three-dimensional deviations of GH2132 superalloy green parts. Orthogonal experiments and range analysis yield a preliminary optimum (50 µm, 50%, 45 p/ms, 60 p/ms), achieving 56.19% density. Five machine learning models are compared, with ANN selected as the surrogate (average R ²  = 0.9164). SHAP interpretation on the independent test set reveals nonlinear segmented characteristics of white ink concentration, identifying its optimal range as 30%–50%. Feature contributions to density rank as layer thickness &gt; white ink concentration &gt; roller rotation speed &gt; roller traverse speed. Bayesian optimization with normalized Euclidean distance mapping yields the recommended window (50 µm, 40%, 40 p/ms, 70 p/ms), predicting 57.89% density. Validation experiments show that the measured relative density is 58.3%, with X‑, Y‑, and Z‑axis deviations of 0.36, 0.24, and 0.06 mm, respectively, all within 2% relative to the nominal dimensions, confirming the effectiveness and engineering applicability of the proposed optimization framework. </p>