Beyond global energy density for direct energy deposition: machine learning for modelling and optimizing process parameters

Abstract Additive manufacturing (AM) has numerous process parameters that lead to significant variation in the success and quality of the manufactured product. The input parameter of ‘global energy distribution’ (GED) is used throughout the literature to describe the input energy onto the surface of a build due to the combination of laser power, laser scanning speed and laser spot size. This paper identifies more accurate modelling using machine learning for the GED constituent process parameters and their influence on the responses of manufacturing layer height, relative density and grain size on the build. The layer height was best modelled using an artificial neural network (ANN) which produced an R 2 value of 0.97 and a root mean square error (RMSE) over the data set of 0.03mm. An accurate prediction model was produced to aid manufacturers in setting the ‘layer height’ user parameter. The relative density was best modelled using multi linear regression (MLR) and produced an R 2 of 0.73 and a RMSE of 0.66%. The grain size was best modelled using an ANN and produced an R 2 of 0.85 and RMSE of 9.68µm. These models show why reproducibility is difficult when considering GED singularly, as each of the constituent parameters influence these individual responses to varying magnitudes. The methodology presented should aid industrial users with resource efficient process parameter response modelling with application in process/product optimization.