Machine learning-assisted dimensional accuracy control in DLP-based fabrication of ceramic cores

Ceramic cores are essential components in the precision casting of turbine blades, making the study of 3D printing technology crucial for the development of complex ceramic cores. However, achieving dimensional accuracy in ceramic cores remains a significant challenge. Digital light processing (DLP) based vat photopolymerization (VPP) enables high-precision molding. This study introduces an innovative approach by integrating the XGBoost model with DLP printing to develop a predictive model for three-dimensional printing errors, aiming to achieve exceptional precision. A comprehensive dataset of 300 experimental trials was generated by systematically varying four key process parameters: UV light intensity, ceramic particle size, solid content, and workpiece height. After optimization, the XGBoost model demonstrated the following performance: in the horizontal direction (XY), the coefficient of determination (R²) = 0.9931 and mean absolute error (MAE) = 0.0137; in the vertical direction (Z), R² = 0.9528 and MAE = 0.0148. As a result, the XY width error and Z height error of the ceramic core green body was maintained within 0.01 mm, demonstrating a significant improvement over traditional compensation method. Furthermore, machine learning enables the prediction of printing errors, significantly reducing design and production cycle times. This study underscores the versatility and significant potential of combining machine learning technology with high-precision DLP ceramic printing technology, providing an effective data-driven framework solution for high-precision green-body fabrication in silica-based ceramic 3D printing.