We have developed an innovative 3D data visualization and simulation framework that combines generative modeling, real-time simulation and GPU-based analysis. This holistic system makes it possible not only to visualize complex data, but also to analyse it dynamically and integrate it into existing data pipelines. M2 has been awarded the BSFZ seal of approval for this scientific and technical innovation expertise.
Article by Christoph Maurer, Data Science & AI and Visual Design Lead at M2
As a data analytics and consulting company with a strong focus on business intelligence, we see it as our mission not only to prepare complex data, but also to present it in a form that creates the greatest possible added value. We have been integrating BI solutions into companies in a wide range of industries for years and have noticed that many challenges relating to process optimization or the simulation of processes and scenarios now require more than “classic” 2D data visualizations.
The step into the third dimension
Three years ago, we therefore launched an ambitious research project. The goal: to develop a comprehensive framework that redefines generative and analytical 3D data visualization and simulation and can be seamlessly integrated into existing data pipelines.
The innovation behind this has now been officially recognized: We have received the BSFZ seal of approval for our three-year research project in accordance with the Research Allowance Act (FZulG). The Research Grant Certification Office (BSFZ) has examined our project and confirmed its scientific and technical novelty. This seal confirms our scientific and technical novelty and underlines the fact that we are setting a real milestone in 3D data processing.
Our research focus: What makes the framework so special?
1. Generative 3D structures & high-performance simulation
We have chosen a procedural approach that combines all work steps from the data ETL process to generative model creation and various types of simulation. This allows almost any type of 3D object to be created, analyzed and adapted in real time: Differential and procedural generation (e.g.: reaction-diffusion, differential growth), particle and fluid simulation, solid-state calculations and network algorithms (force-directed graphs).
Some of these modules are GPU-accelerated in order to deliver the necessary performance and scalability for large data volumes.
2. Analytical processes seamlessly integrated
Instead of having to apply pure 3D graphics “on spec”, we use systematic analyses. Using Python libraries such as cuDF and cuML (GPU-based), large amounts of data can be processed efficiently and integrated directly into our 3D pipelines. Clustering methods (e.g. DBSCAN, HDBSCAN or K-Means) create the basis for well-founded findings in real time.
3. High-End-Visualization in Houdini & Omniverse
Our development stack combines the flexibility of SideFX Houdini - known for its procedural workflows - with the powerful NVIDIA Omniverse platform, which enables real-time ray tracing and physically correct simulation on multi-GPU hardware. The link: USD (Universal Scene Description). With USD, we store all assets and simulation data in a non-destructive form, which also simplifies collaboration and repeated iterations many times over.
4. Focus on Cloud- & GPU-performance
Omniverse's cloud capability, combined with our modular architecture, opens up completely new fields of application. This allows us to execute and visualize huge data streams or extensive simulations in the shortest possible time - whether for industrial metaverse, digital twins, robotics training or scientific research.
Why is our approach innovative?
• Holistic workflow: Instead of using isolated solutions for data cleansing, simulation and visualization, our framework integrates all process steps in one pipeline.
• High performance & scalability: GPU acceleration in every step and specialized implementations in VEX, OpenCL or CUDA enable us to achieve high computing speed even with very large data sets.
• Universal storage format: With USD, we have a central data hub to efficiently handle metadata, variants and complex 3D scenes.
• Extensible modules: We develop HDAs (Houdini Digital Assets) and Omniverse extensions that allow our solutions to be flexibly integrated into existing pipelines.
These features have not yet been implemented in this combination in the specialist world. Classic simulation tools are usually monolithic and only CPU-based, while GPU-accelerated modules are often purely specialized applications. Merging them into a holistic, generative and analytical 3D data framework is a decisive step towards closing the gap between data science and 3D visualization.
Risks & how we master them
The development of such a framework brings with it challenges that we have tackled in countless tests:
• GPU vs. CPU calculations:
Not every algorithm can be efficiently outsourced to the GPU (keyword FP64 accuracy). Through comprehensive benchmark tests, we find the balance of when a process runs better on the CPU or on the GPU.
• Interoperability of different simulations:
Different resolutions for volumes, polygon meshes or particle systems require careful parameterization and data exchange processes. We have introduced a multi-stage validation process to ensure that the simulation pipeline is stable and reproducible.
• Data volumes & storage load:
As we can use USD to log all steps and save variants, the volume of data increases. With specially developed import and export interfaces, we have very granular control over what information is written to the USD file, when and in what resolution.
It is precisely these technical and scientific risks that make up the research character of our project - and ensure that we can offer a robust product in the end.
Diagrams as epistemic tools
Our 3D data visualization goes far beyond mere “eye-candy” representations. We follow a layered approach, as defined by the “Grammar of Graphics”, by dividing visualizations into clearly defined layers (data, transformations, geometries, scales) and thus achieve consistent diagram forms and precise data communication.
At the same time, we see diagrams as active tools that make implicit knowledge explicit and relationships visible. At their core are relational structures - in our framework, for example, through force-directed graphs and network algorithms. Metaphors such as “flow” or “space” also facilitate the intuitive understanding of abstract variables. Diagrams can not only depict the current state, but also simulate future scenarios. Our non-destructive workflow turns visualizations into ongoing fields of experimentation and “generators of insight”.
Through the synthesis of Grammar of Graphics and Diagrammatics, our 3D visualizations become thinking tools in which data science and 3D representation merge in such a way that new knowledge is created relationally, iteratively and scalably.
Fields of application & future prospects
• Industrial Metaverse & Digital Twins:
Complex IoT data streams can be analyzed and simulated in real time in an interactive 3D environment.
• Robotics & Automation:
AI training data sets can be efficiently expanded in procedurally generated 3D worlds - a quantum leap in sim-to-real transfer.
• Scientific Research & Education:
Large-scale simulation projects (e.g. fluid mechanics, astrophysics) can be GPU-accelerated and collaboratively processed in the cloud.
Our framework remains modular and expandable - we are already working on new components for AI-supported rendering, Physics-Informed Neural Networking (PINN) and AR and VR integrations.
Conclusion
The award of the BSFZ seal in accordance with the Research Grants Act not only confirms the innovative strength of our project, but also demonstrates the benefits such a framework can have for a wide range of industries and research sectors. The work by no means ends here - we see our solution as a starting point for closing the gap between classic data analysis and advanced 3D simulation.
We look forward to supporting companies and research institutions alike with our framework. Because anyone working on complex issues today - whether in Industry 4.0, digital twins or scientific model simulations - needs a holistic, high-performance approach. This is exactly what we provide.
Interested?
If you would like to find out more or accelerate your projects with our framework, please get in touch. Let's shape the future of the 3D data world together!