A dimensionless group-incorporating artificial neural network (DI-ANN) model for single-track depth prediction of SS316L for laser-directed energy deposition (L-DED)

In this work, a dimensionless group-incorporating artificial neural network (DI-ANN) model is established to determine the relative influence of factors affecting single-track depth for laser-directed energy deposition (L-DED), specifically as it correlates with two types of porosity (keyhole mode and lack-of-fusion voids). The DI-ANN model’s input features include process parameters, process signatures acquired from experiments, and multiphysics simulation-based dimensionless groups. Experimental data were acquired for SS316L straight single-track samples, prepared using a continuous-wave laser with a top-hat profile. The feature influence on single-track depth was analysed using Spearman’s correlation coefficient before training the DI-ANN model, which was tested using 5-fold cross-validation. Laser power, scanning speed, powder feed rate, melt pool width, melt pool area, maximum temperature, maximum Marangoni number, mean and maximum Grashof number, and mean solidification time are selected for training the model. The predictive quality of this model for single-track depth is 6.1% mean absolute percent error (MAPE) and 11.7% maximum absolute percent error (MXAPE). Compared with recent literature, this study shows an approximately 30% improvement in the MAPE and a 70% improvement in MXAPE. The presented model can be used for process parameter optimisation for porosity reduction and be a valuable foundation for feedback control algorithms for heat penetration.