Technology Highlights in GeoDict 2022
Welcome to my talk about the technology highlights in our new release GeoDict2022. It is a pleasure for me to present the new GeoDict version and I want to thank the audience for attending this year's GeoDict User Meeting. My name is Erik Glatt, and I am the CTO of Math2Market, the company behind the software GeoDict. Before our team spun-off the company I worked at the Fraunhofer ITWM as a researcher. I hold a PhD in theoretical Physics from the Technical University in Darmstadt.
Before I start with the technology highlights, I want to point out that the GeoDict2022 release is out now. We recommend to download and test it, as we think GeoDict2022 is the best solution for digital material analysis and design. Furthermore, you can explore all the new features I am going to present in the following slides. What we mean with digital material analysis and design is shown in the workflow on this slide: It often starts with the import of a 3D image of a real material. Afterwards the image is analyzed to understand the material and its properties, this step alone can already be very useful.
If one understands the materials geometric properties one can create its statistical Digital Twin. Having this twin one can start to design or engineer a next generation material with improved properties. In this way, GeoDict can help to design improved materials faster and cheaper. This workflow will aid me now as guideline to walk you through the technology highlights in GeoDict2022. In this presentation, I can only cover a selection of improvements and new features. In many of the more specific talks of my colleagues, you will get more detailed information.
This year we look, at first, at the highlights regarding the property simulation, which are used in the Analyze, the Validate and the Design steps: It is possible now to simulate 3-point flexural bending test in ElastoDict. The nonlinear mechanics simulations are much faster. With the new version, it is possible to track objects and contacts during large deformations. The computation speed for demanding Stokes and Navier-Stokes flow simulations, for example for complete filters, is strongly improved.
Battery-LIR can do fast and memory efficient simulations with porous binder now. Direct half-cell battery simulation and complex charging profiles for batteries are possible in BatteryDict. The 2-phase flow is strongly improved allowing, for example, a constant flow rate and forced imbibition simulations. Now, I show these highlights in a bit more detail. To allow direct simulations of flexural bending tests in GeoDict was wished for many years. Now with GeoDict2022, such simulations are possible.
One can specify the bending loads either as Flexural Deflection or as Flexural Force. The result is the flexural modulus, the force-deflection curve, and corresponding local displacement, strain and stress fields. On this slide, the flexural test simulation is performed on a laminate structure, were the bending in x- and y- direction is compared.
The flexural modulus of the example structure is much higher in x-direction, because the top and bottom sheet layers are oriented in this direction. The color fields show the amount of plastic deformation in the matrix material for both directions. With GeoDict2022 it is possible to use the conjugate gradient methods also for nonlinear simulations. This often leads to much faster simulations and a better convergence of the solution.
Here an open-cell Aluminum foam, where the struts can deform plastically, is simulated. The simulation results are shown in the video. Comparing the run-times, one sees that with GeoDict2022 the simulation is 3 times faster. It is also possible now to track the object information during a large deformation simulation. This means that one knows which voxel belongs to which object in each deformation step.
Here this is demonstrated for a non-woven which is compressed by 30%. After the compression, the number of fiber contacts is nearly 2 times larger as at the beginning. This means that the material is much stiffer in z-direction. The tracking of objects also allows to study how the shape of the objects change during the deformation simulation.
For the LIR solver in FlowDict, the new Acceleration Method Krylov is available. This option reduces the runtime for challenging flow simulations drastically. Challenging simulations could be fast flows for complete filters or Stokes flow through low porosity rocks. The reduced runtime is paid by an increased memory consumption.
In the default solver settings, this new option is used automatically, but it can be enabled or disabled by the user. On this slide the effect on the runtime is shown for a tight sand-stone. Compared to GeoDict2021 the run-time is decreased by a factor of 5. With GeoDict2022, it is now possible to simulate porous binder also with the Battery-LIR solver. For the battery charging simulation shown on this slide, this leads to a runtime reduction of factor 6 and an immense memory reduction compared to simulations with the BEST solver. This means that one can compute much bigger and more realistic batteries with GeoDict2022.
It is now also possible to directly charge half-cells in BatteryDict. This means one can import an electrode or create one with GrainGeo, for example a cathode with the corresponding predefined script. Afterwards, in BatteryDict one selects “Charge Half Cell”, sets the parameters, and can run the charging simulation without the need to design a complete battery. So, it is much easier to perform half-cell simulations.
It is also possible to set complex charging profiles. In this way one can charge and discharge the battery in multiple steps. In the charging plot this is shown for a half-cell simulation, however this works also for complete batteries. For the computation of 2 phase-flow we have developed the „Dynamic PMM“ to extend the quasi-static approach with a dynamic fluid interface movement.
This dynamic method features the following: - the computation of non-monotonic capillary pressure curves. - the computation of forced imbibition. - the consideration of mixed wettability.
The method now predicts critical pore filling events, as published by Jung et al. in 2016, as can be seen in the animation. Details and validation of this feature highlight can be seen in the presentation of Sven Linden. Here, as an example where both, the fast flow-solver and the dynamic PMM, are used the relative permeability computation of a huge mixed-wet digital sandstone sample. The sample has 1,500 voxels on each side and thus, more than 3 billion voxels in total.
The entire relative permeability computation required a runtime of less than 7 days and less than 200 GB of working memory on a 32 cores Linux computer. Now, we come to selected improvements for the analysis of 3D images which is also needed in the Validate step: The percolation path and geodesic tortuosity computation is more accurate. With GeoDict-AI, you can train neuronal nets for image analysis. In FiberFind, an enhanced post-processing is available for the identified fibers. To get a more accurate percolation path length or a more accurate geodescic tortuosity the option „Optimize Path“ is available in GeoDict2022. With this option, the paths are pulled tight after the computation on the voxel grid, resulting in more accurate paths lengths.
Here, this is shown for a single path where this option leads to a 4% reduction of the computed path length. The path before the optimization is shown in black and the optimized path in yellow. You have a 3D image, and you want to find the cracks in a rock, or fibers with a specific shape in a non-woven, or the warp threads in a woven structure, you can do this with GeoDict-AI in GeoDict2022. Just provide the training data, what often can be done via a relatively simple GeoDict script, train you network, validate the results, and if you a satisfied with the nets analysis quality apply the net to the images you want to analyze.
The example on this slide shows the result of a net which was trained to find the warp and weft threads in a weave. The training data for this net was created with WeaveGeo. Applying the net the solid voxels get different material Ids for warp or weft threads. Let us continue with the feature highlights regarding material modelling, which is needed for the validation and, especially, for the design step: One can model foams with (tri-)angular struts and a specified SVP.
It is possible to generate grains and fibers with more than one material ID per object. It is possible to generate strut bases lattices and triply periodic minimal surfaces for additive manufacturing in GridGeo. One can directly edit selected gad-objects via the context menu in the Visualization area. One can model super-quadric particles as GAD objects. In the generation of convex polyhedra one can match the enclosing object volumes.
To easily create statistical Digital Twins of foams, one needs to model triangular struts. Furthermore, one needs to directly match a given solid volume percentage. With GeoDict2022, both is possible in FoamGeo and was used to create the Digital Twin of an open-cell aluminum foam shown on this slide. The pore size distribution and pore orientation was measured with „Identify Pores“ in PoroDict and the strut diameter distribution was computed with FiberFind. Afterwards, one enters the statistical description of the foam in the FoamGeo GUI and creates the Digital Twin.
Until now, a GAD object like a fiber or a grain, was restricted to one material ID. With GeoDict2022, this has changed and multiple material IDs per object with specific patterns are possible. This option is available in GadGeo and the „GAD Objects“ content menu. Choose „Change Material ID Model ...“ and create structures as shown on this slide. Here, we see grains with stripes, trilobal fibers with radial stripes and combined objects, where each sub-object has its own material ID. With this new feature one has countless new options for more realistic structure modelling in GeoDict.
In additive manufacturing, a given space is filled with a periodic lattice-structure. Common examples are strut-based lattices and periodic cells based on triply periodic minimal surfaces. In GeoDict2022, one can directly create such lattices in GridGeo, where four examples are shown on this slide. The lattice for a specific application must be tailored to specific performance characteristics. In GeoDict, one can create such a lattice and directly compute its properties. For example,
with ElastoDict, when a specific mechanical performance is required. At last, we come to some general innovations, which are useful for all steps in the design work-flow: It is possible now to freely adjust the units and the style of all result plots. The main GUI was improved to allow, for example, better measure options in the 2D view.
New interpolation options for the volume field visualization are available. The export to Excel via Python is possible, for example, also on Linux computers. One can create 3D images from surface meshes loaded in GeoDict. We have improved cloud computation options for flexible and unlimited computations. Equations and References are available in many solver dialogs. It is possible now to crop volume fields together with the loaded structure.
In GeoDict2021 the plots, for example in the result files, are already very nice and one has a lot of options to customize these plots. However, with GeoDict2022 we have further improved the customization options. One can freely adjust the ranges and units of such plots. Furthermore, one can change the title, color and style of each graph. The customization of plots can be done on all loaded gdr-files of a given command at the same time.
One can directly save a plot as image via the graph context-menu. After the gdr-file is closed the customized plots are saved in the file. All these steps can be recorded as GeoDict macro and thus the plot post-processing is completely scriptable. GeoDict supports different mouse interaction modes. These mouse modes can be selected with the tool-buttons shown on the top of this slide. In this way one can select GAD objects and voxels, draw in the current 3D image and now also measure distances and angles in the 2D view.
If a mouse mode is selected a corresponding widget is opened at the right side of the GUI. This widget shows the options for the selected mouse mode, for example one can choose if one wants to measure a distance or an angle. In the shown image, I measured 2 grain diameters and 2 angles between touching grains.
So, with GeoDict2022 it is much easier to manually study feature sizes and shapes in a loaded geometry for example from µCT-scan. In GeoDict2022, we have also improved the visualization. One example is the improved field interpolation. For battery simulations, one has often orders of magnitude lower Li concentration in the electrolyte compared to the active materials. In this case, one gets a wrong impression of the field values when an interpolation across the material interfaces is done. With GeoDict2022, it is possible now to interpolate per material ID or to interpolate only on the visible materials.
This interpolation on visible materials can also be shown on a smooth iso-surface. These options lead to a much more accurate visualization of the field values for cases, as described above. On this slide, the new interpolation options are shown for a mechanics simulation, where the stress values in the stiff fibers are much larger than in the matrix. The image on the left shows the classic linear interpolation, which leads to visualized stress values which are too small at the material interface.
The interpolation only for the visual fibers gives a much better result, which can be seen in the image in the middle and on the right. Starting with GeoDict2022, a third way to export gdr result-files to Excel exists. This option is called „Excel (generic) - Python“ und creates results like the „Excel (generic)“ option.
The results of both generic exports look very similar, but the new option has some advantages: No Excel is required on the computer the export is done, so it works also on Linux machines. The new export is faster and more robust, as one does not rely on visual basic. The example here shows the generic Python export for a stokes result from FlowDict. To get access to more computer power and to use computation resources more flexible the cloud is a very useful tool for simulations.
With GeoDict, one has multiple ways to use cloud resources. Interactive remote sessions are already available for a long time. Since last year we offer the web-based batch-operation together with KaleidoSim, the GeoCloud. This option will be available from the GeoDict GUI starting with GeoDict2023. As a new option, we offer the web-based interactive remote session.
One just logs-in to the web interface, starts and runs a virtual machine instance. Afterwards one can connect to the virtual machine to open an interactive session in a new tab in the web-browser. Now, one can run GeoDict and use it as you would locally: Generate structures, perform analyses and simulations, create images and videos… The project directory is an empty 1TB drive that you can use for structures and results. For the down- and upload of files, one can use for example WinSCP. In the future, it will be possible to download the project folder contents directly from the web interface. You can learn more about our cloud options in the talk of my colleague Rolf Westerteiger in the Image Processing and Image Analysis session of this year's User Meeting This is the end of my presentation.
I hope I could show you a few interesting improvements of our software. Thank you for your attention.