Utilisation of artificial neural networks to rationalise processing windows in directed energy deposition applications. (15th January 2021)

• Record Type: Journal Article
• Authors: Feenstra, D.R.; Molotnikov, A.; Birbilis, N.
• Is Part Of: Materials & design. Volume 198(2021)
• Publication Date: 2021-01-15
• Subjects: Directed energy deposition -- Neural network -- Process optimisation -- SS316L -- Hastelloy X -- Inconel 625; Materials -- Periodicals; Engineering design -- Periodicals; Matériaux -- Périodiques; Conception technique -- Périodiques; Electronic journals; 620.11
• DOI: 10.1016/j.matdes.2020.109342
• Languages: English
• ISSNs: 0264-1275

Abstract

Abstract: The application of Directed Energy Deposition (DED) when using new materials or new instruments, requires significant empirical testing to define a suitable or optimum process operation window. Determining the ideal DED parameters is challenging due to the complexity of the deposition process being dynamic in nature, with a multitude of parameters being highly influential on the resultant melt pool dimensions and the subsequent evolution of solidification. The present study seeks to rationalise the notion of a processing window by using artificial neural networks (ANN) to elucidate the complex interaction between the input parameters - specifically the relationship between the energy density of the laser and material deposition rate on the shape of single-track deposits. Herein, cross-sectional data was collected from single tracks of Inconel 625, Hastelloy X and stainless steel 316 L deposited onto mild steel substrates; whilst using a matrix of process parameters. The ANN was used to model the interplay between laser power, scan speed, laser beam diameter, material deposition rate and material type. The network was then used to visualize a theoretical relationship between the volumetric energy density and the energy required to melt a specific amount of the supplied powder. Graphibal abstract: Unlabelled Image Highlights: The melt pool dimensions of 297 single-track deposits of hastelloy, inconel and stainless steel were measured. The interplay of the key input parameters was modelled using an artificial neural network. Effects of laser power, scan speed, powder deposition rate and beam diameter on melt pool dimensions were visualised with contour maps. A relationship between energy density and powder deposition rate was ascertained and used to rationalise trends. An energy ratio was developed that could predict the level of track dilution.