Optimising the manufacturing of a β-Ti alloy produced via direct energy deposition using small dataset machine learning

Abstract Successful additive manufacturing involves the optimisation of numerous process parameters that significantly influence product quality and manufacturing success. One commonly used criteria based on a collection of parameters is the global energy distribution (GED). This parameter encapsulates the energy input onto the surface of a build, and is a function of the laser power, laser scanning speed and laser spot size. This study uses machine learning to develop a model for predicting manufacturing layer height and grain size based on GED constituent process parameters. For both layer height and grain size, an artificial neural network (ANN) reduced error over the data set compared with multi linear regression. Layer height predictions using ANN achieved an R2 of 0.97 and a root mean square error (RMSE) of 0.03 mm, while grain size predictions resulted in an R2 of 0.85 and an RMSE of 9.68 μm. Grain refinement was observed when reducing laser power and increasing laser scanning speed. This observation was successfully replicated in another α + β Ti alloy. The findings and developed models show why reproducibility is difficult when solely considering GED, as each of the constituent parameters influence these individual responses to varying magnitudes.