Data-driven prediction of next-layer melt pool temperatures in laser powder bed fusion based on co-axial high-resolution Planck thermometry measurements. (July 2022)

• Record Type: Journal Article
• Authors: Kozjek, Dominik; Carter, Fred M.; Porter, Conor; Mogonye, Jon-Erik; Ehmann, Kornel; Cao, Jian
• Is Part Of: Journal of manufacturing processes. Volume 79(2022)
• Page Start: 81
• Page End: 90
• Publication Date: 2022-07
• Subjects: Additive manufacturing -- Laser powder bed fusion -- Melt pool temperatures prediction -- Machine learning; Production management -- Data processing -- Periodicals; Manufacturing processes -- Periodicals; Procestechnologie; Productietechniek; Production -- Gestion -- Informatique -- Périodiques; Fabrication -- Périodiques; Manufacturing processes; Production management -- Data processing; Periodicals; 670.5
• DOI: 10.1016/j.jmapro.2022.04.033
• Languages: English
• ISSNs: 1526-6125

Abstract

Abstract: Uncontrolled process variability, stemming from geometry, machine, or parameter variation, can lead to metallurgical defects such as keyhole porosity and lack of fusion as well as geometrical defects such as increased surface roughness or increased deformation in the produced parts. This lack of control is a pressing problem in laser powder bed fusion (L-PBF) processes. One way to reduce this variability is to use model-based predictive control. Process parameters such as laser power and scan speed can be adjusted during the process based on in-situ measurements of process conditions such as melt pool size or temperature. In this paper, a predictive model that is an essential element in a larger predictive control ecosystem for L-PBF is developed and tested. The proposed machine learning-based regression model is trained using high-resolution co-axial melt pool temperature measurements from the previous layers. The machine learning model can predict the melt pool temperatures along the toolpath for the next layer assuming processing parameters remain the same. The paper describes the development of the machine learning-based prediction model and presents the guidelines for the design and selection of the features in feature vector. The estimation of the prediction performance based on real physical data is presented followed by suggestions of future work. The main limitation of the current approach is the relatively high computational cost. Some guidelines for implementation and possible improvements are given in the discussion of results.