Session: 16-04-03: AM Bench Plenaries III

Paper Number: 173872

The Trace of Heat: Models, Benchmarks, Optimizations, and Metrics 

Recent benchmark efforts have revealed astonishing insight into the range of validity of the transient non-linear diffusion equation to model powder bed fusion of metals (PBF-LB/M). This presentation will report on the latest insights into what such a diffusion model is capable of, what it is not capable of, and which metrics might be a good choice to compare simulations with experiments in transient scenarios.  

Specifically, we will first rehearse results of a blinded comparison of the prediction of the dimensions of 200 melt pools generated by Gaussian and non-Gaussian laser beam shapes. The comparison will include multi-physics melt pool models in the form of smoothed particle hydrodynamics. Such multi-physics models undoubtedly are very valuable to generate insight into the physical behavior of the process e.g. to predict the formation of keyhole- modes and phenomena such as humping or balling. However, even if such states cannot be computed by diffusion models, they can accurately deliver melt pool dimensions and indicate that undesirable limit states are reached. Depending on the application, such predictions prove to be very valuable.

A good example of such an application is beam shaping. The task is: given a desired melt pool, compute is the corresponding beam shape that generates such a desired melt pool. This talk will shortly rehearse such beam shaping methodologies including their validation for stationary cases. 

Clearly, little about the PBF-LB/M process is stationary. Given the fact that current multi-physics models are computationally too involved to be used in optimization scenarios even at melt pool level, it is of interest to evaluate the validity of the transient non-linear diffusion equation in truly transient situations such as laser turn-around scenarios. To our own surprise a comparison gives surprisingly accurate results even for such transient scenarios judged by the eye norm, when the post processing is done right. 

This leads to the serious question of how to compare results of transient experiments against transient computations. Corresponding metrics must consider both space and time and can by construction not be unbiased. The presentation will lay out selected metrics and investigate which highlight or hide what kind of differences and similarities between computations and experiments. Although somewhat trivial, it is valuable to keep in mind that both, the choice of the model and the choice of the metric determine the accuracy of a comparison especially in transient scenarios.

Presenting Author: Stefan Kollmannsberger Bauhaus-Universität Weimar

Presenting Author Biography: 2009 PhD in Computational Modeling of Fluid-Structure Interaction at Technical University Munich
2010-2024 Senior Researcher, later Adjunct Teaching Professor for Computational Mechanics at Technical University Munich
2024-date Full Professor of Data Engineering at Bauhaus-University Weimar

Selected roles relevant to the community of Additive Manufacturing:
-Chairman of the ECCOMAS special topic conference on Simulation in Additive Manufacturing (since 2014)
-Advisor to the AM-Bench Committee

Authors:

Stefan Kollmannsberger Bauhaus-Universität Weimar
Vijaya Holla Technische Universität München
Philipp Kopp Bauhaus-Universität Weimar