Directly from surface to bulk: rapid prediction of internal densification in laser powder bed fusion additively manufactured nickel-based superalloy using machine learning

Abstract

For the quantitative evaluation of processing quality of laser additive manufacturing, traditional characterisation methods such as Archimedes method and computed tomography (CT) face limitations in testing volume-constrained components, while optical imaging methods require destructive sample preparation. The rapid characterisation of internal metallurgical quality and densification behaviour in complex components remains one of the key challenges restricting the even broad applications of additive manufactured parts. This study classified the surface roughness of the formed specimens using statistical methods, and combined the Gaussian process regression algorithm to achieve rapid and accurate prediction of the internal densification behaviour of the laser powder bed fusion fabricated GH4169 components. The research revealed that when energy input remains conventional thresholds, components exhibit nearly defect-free interiors with densities reaching 99.9%. Excessive energy input results in smooth surfaces but leads to the formation of spherical pores. Parts with coarse surfaces demonstrate unmelted holes caused by insufficient energy input, whereas parts with porous surfaces contain numerous irregular defects and unfused powder due to excessively low energy input. Furthermore, the predicted R2 can reach up to 0.939 through this method. This work provided novel methodology and theoretical foundation for rapid quality evaluation and characterisation of LPBF fabricated GH4169 complex components.