Laser Powder Bed Fusion Melt Pool Prediction and Process Window Optimisation

In laser powder bed fusion (LPBF) the melt-pool depth and the defect mode (keyhole porosity, lack of fusion, balling) decide whether a part becomes dense. Using the public MeltpoolNet tables - process parameters and measured melt pools compiled from published papers (1,252 laser-PBF depth rows from 61 papers and 21 alloys, 1,168 defect- mode rows) - this project compares physics baselines (Rosenthal moving point source, normalised enthalpy) with gradient boosting and a Gaussian process on physics features, classifies the defect mode, and searches the fastest safe process window with Bayesian optimisation. Every model is evaluated with whole papers held out: no fitted model is significantly better than a calibrated Rosenthal line across papers, while a random-row split would report 19 % instead of 78 % MAPE. Unlike the related projects (A12 milling tool wear and PR3 turning chatter use signals, A13 steel surface defects uses images), PR4 is the only additive-manufacturing project and works on literature-derived process tables. A FastAPI web application with five interchangeable databases predicts melt pools, draws P-v maps, optimises the window and calibrates a machine with a few measured tracks.

Category: Production Engineering

Production Engineering Project PR04 from ProTechSC Academy, with graphical abstract, video demonstration and project guidance.