For me as one of the main developers of AdditiveLab software, I am always interested to validate our software's performance with real manufacturing cases. For a long time, the VTT Pipe Case Study has been on my bucket list and I finally found some time to do it. For those who are not familiar with the VTT pipe case study, the study contains an in-depth description of the additive manufacturing process of a pipe geometry, as well as measurements post-manufacturing. Therefore, this study makes an excellent candidate to benchmark AM process simulation software such as AdditiveLab.
The figures above show the pipe test geometry provided by VTT and can be accessed from their website. The pipe design was manufactured with H13 tool steel via SLM technology.
The pipe is particularly a cool case since the manufactured geometry showed several issues including distortions and shrinkages which could compromise the design for its usage. Therefore, I was really curious about how AdditiveLab software predicts regions in the design with such issues. There are several ways of how to simulate the additive manufacturing process, but I was specifically interested in benchmarking macro-layer based methods; the inherent strain and the thermo-mechanically coupled analysis method. (Please note: In this article three different simulation methods will be presented; (1) a thermo-mechanical scanning path analysis that will be used to determine inherent strains for the (2) macro-layer based inherent strain simulation of the pipe, and a (3) thermo-mechanical macro-layer based simulation of the pipe).
Thermo-mechanical scanning path simulation:
All right, so where did I start? For the macro-layer based inherent strain method, the first thing that I needed to do was to determine the inherent strains that are necessary for AdditiveLab to simulate the displacements during and after the manufacturing process. This can be done with two different methods; the first one is to use a calibration procedure within the software that utilizes measured data from a calibration print. The second one is to create a microscopic scanning path analysis and measure the residual strains.
Considering there was no calibration geometry available in the VTT study, I chose to determine the strains via a thermo-mechanical, scanning path analysis within AdditiveLab. Since this analysis requires temperature-dependent, thermal and mechanical material data for H13 (steel that was used by VTT to create the design), I went online, started googling the material and was able to find data that I could use for the simulation. The rest of the info that I needed (e.g. process parameters), I have taken from a second study presented by VTT. (I put the references of the material and the link to the second study of VTT at the end of this article)
To create the scanning path model and to analyze the residual strains, I used the AdditiveLab Python API. There are several ways of creating scanning path analyses ranging from single trajectories to entire hatching path analyses. However, for the sake of simplicity, I decided to do a single, short trajectory model and measure the strains at a representative location. The results of this simulation are illustrated in the figure below:
The figures above show the thermo-mechanical scanning-path analysis performed in AdditiveLab; the material phases (left), and the total strain (TOSTRAIN) along the scanning path (right).
The figure above shows the total strain vs. distance of the principal components (EXX, EYY, EZZ) perpendicular to the scanning path (measured in the center) .
I measured the simulated residual strains perpendicular to the scanning path; now, here is where it becomes a little tricky. There are different philosophies out there about which strains (elastic, plastic, thermal, total) and which components to use as representative strains for the macro-layer based inherent strain simulation. Googling simulation related inherent strain research publications will give you more insight if you want to understand the different options. However, when modeling a single track, I have found that looking at the total strain (elastic + plastic) at the principal components (EXX, EYY, EZZ) reflects the inherent strains needed for the macro-layer analysis pretty well. (We have validated this several times with the calibration method and actual test prints). However, besides our company's and my personal experience, there is also a more scientific explanation of why these components are suitable, but this will not be addressed in this article. (You can contact us if you need more info about that).
Simulation of the pipe using the inherent strain method:
After the scanning path simulation, I could use the determined inherent strains, define a new material in AdditiveLab and run the macro-layer based simulation. The results of the simulation are shown below:
The figures above show the deviation of the printed part measured by VTT (left) and simulated with the inherent strain Method in AdditiveLab (right).
The above results illustrate the experimental results produced by VTT and the according results produced by AdditiveLab software. When comparing the deviations of the manufactured part with the original CAD geometry, the measured and simulated results match considerably well. Particularly the shrink-line introduced at the lower part of the flange (indicated by arrow) can be nicely predicted and is within the same range of displacements. However, there are regions that do not match perfectly which may be contributed by:
Material data The material data was gathered from sources cited below. However, the temperature-dependent data was only provided up to ~600 degrees Celsius. Therefore, the scanning path simulation may not properly reflect the behavior of the material at higher temperature ranges, which subsequently affects the determination of the inherent strains. If one is not satisfied with the quality of the results, a calibration procedure to determine the inherent strains could be performed.
Lack of process information Even though VTT provided in-depth information about the manufacturing process, there was some information missing that had to be assumed in the scanning path model. For example, heat losses at the top surface of the scanned powder-bed are predominantly driven by convection and radiation depending on the conditions in the build chamber e.g. gas flow can have a very strong effect on the top-surface cooling but was not specified in the VTT study.
Simplification via the inherent strain Method It definitely needs to be mentioned that the inherent strain method provides a phenomenological simplification of the actual thermo-mechanical AM process. Even though the inherent strain method allows for decent and very fast predictions of AM outcomes, it sometimes can be limited in accuracy for some geometrical features in AM designs.
Nevertheless, as seen in the result pictures, the magnitudes and directions of geometrical deviations are within the same range, and critical regions suffering from shrinkage could be detected.
Simulation of the pipe using the thermo-mechanical analysis:
Another way to get more insight into the complex behavior of the design during and after the manufacturing process is to utilize a thermo-mechanical analysis. Within AdditiveLab there is also the possibility to perform a macro-layer based, thermo-mechanically coupled analysis. Similar to the inherent strain analysis it allows the prediction of deformations, stresses, and strains. However, thermo-mechanical analyses are often more accurate since they allow for consideration of complex physics that reflect the manufacturing process better. Anyhow, I was curious as well what the result of the thermo-mechanical simulation would look like and performed it on the pipe geometry as well. The results are depicted below:
The figures above show the deviation of the printed part measured by VTT (left) and simulated with the thermo-mechanical Method in AdditiveLab (right).
The above results illustrate the experimental results produced by VTT and the according results produced by AdditiveLab software with the macro-layer based, thermo-mechanical simulation method. The results of this simulation match the experiment considerably better than the ones from the inherent strain method. (E.g. At the front- and back-side of the side flanges and top surface of the top flange).
After completing this study and comparing the results, I finally did the benchmark and could check it off my bucket list. Furthermore, as one of the main developers of AdditiveLab, I am very happy how well our software performs. With the integrated Python API in AdditiveLab, I could generate all the simulation models within a few minutes (I used a tutorial template which I modified accordingly to match the study). The actual simulation times of the models (162k nodes, 132k (hex-)elements) on my laptop were:
thermo-mechanical scanning path simulation, 25 minutes
inherent strain macro-layer simulation (pipe geometry), 20 minutes
thermo-mechanical macro-layer simulation (pipe geometry), 8 hours
Inherent strain versus thermo-mechanical model
After seeing the results of the two methods, one might wonder why not use the thermo-mechanical simulation method all the time? Well, especially in an early design and build preparation stage, quick feedback about potential regions causing problems is of importance. Users often do not want to wait several hours until they can look at simulated outcomes. Once AM designs and build preparations have been optimized, it makes sense to give it a final check with a thermo-mechanical analysis that runs e.g. overnight to get a more detailed feeling of what the actual outcome of the process will be.
What else you should know
When we develop simulation methods, we quite often use research papers for guidance and use them as a reference for validation purposes. However, this is quite theoretical and one must combine evaluating the performance of the developed methods with industrial applications. For example, when we developed the inherent strain analysis for the mechanical simulation module, we used publicly available research publications to help us calibrate our numerical physics. Furthermore, we use industrial experiments for validation. If you have a limited number of production builds that can be used for validation, then I recommend utilizing publicly available data. There are organizations that make AM data of manufactured test geometries publicly available, and even beyond that, provide an in-depth description of each step performed to create them. Amongst others, there are two organizations that I want to give a big shout out to; the American institution NIST (https://www.nist.gov/) and the Finnish VTT research center (https://www.vttresearch.com/). Both provide excellent data that allow AM software and hardware developers to validate their approaches with in-depth documented data.
A special thanks to VTT allowing me to publish their data.
Resources
Material
https://www.hudsontoolsteel.com/technical-data/steelH3
High temperature behaviour of H13 steel, C. Tekmen, M. Toparli, I. Ozdemir, I. M. Kusoglu, K. Onel, Zeitschrift für Metallkunde · December 2005
http://www.matweb.com/search/datasheet_print.aspx?matguid=e30d1d1038164808a85cf7ba6aa87ef7
Second study of VTT
Laakso, Petri, et al. "Optimization and simulation of SLM process for high density H13 tool steel parts." Physics Procedia 83 (2016): 26-35.
Author
Christian Rossmann
Comments