Leveraging Computational Design with nTop to Drive the Energy Transition with Siemens Energy
Interview with Bradley Rothenberg of nTop & Markus Lempke of Siemens Energy
In this interview with Markus Lempke of Siemens Energy and Bradley Rothenberg of nTop, we explore the symbiotic relationship between computational design and additive manufacturing.
Highlighting their upcoming joint presentation at CDFAM in Berlin, Lempke and Rothenberg discuss the practical impacts of computational approaches for the optimization and serial production of high-performance metal components, underscoring the crucial role of software like nTop in enhancing efficiency and innovation within the energy industry.
The conversation covers the origin of nTop through to the application of computational design in real-world manufacturing scenarios, shedding light on the iterative journey of adopting new design paradigms, emphasizing the significance of customer feedback and cross-software interoperability in advancing the field.
Origins of nTop
Brad, could you provide a brief overview of the motivation behind founding nTop and the specific challenges it aims to address?
I started nTop almost a decade ago because I was frustrated by mainstream CAD software’s ability to model complex shapes really easily and quickly. From working with 3D printing, I knew that this introduced a level of complexity that made the current tools grind to a halt.
I had this eye-opening on-site mtg in the early days of nTop at GE Aviation: I asked one of their top mechanical designers to walk me through their process of modeling a heat exchanger: he had this complex model chopped up and split into 12 different sub-models running on 12 different machines.
He would go to each machine, punch in a few numbers to the parametric model defined as a BREP, and his CAD system would crunch away, sequentially updating the model on a single CPU core for hours at a time on each machine.
Due to rebuild errors, the models on several of the machines would fail to update. This was a process that would go on for weeks at a time in order to get new geometry, suitable to 3D print and test.
That same afternoon, I sat with the same engineer, and in an hour, we built a computational model of that same heat exchanger in nTop that we could then punch in new parameters to watch the entire geometry generate in real time, ready to get sliced for 3D printing.
It was at this point I knew engineering tools were due for a paradigm shift – there was a class of parts that the world now needed that the CAD tools were just not built to handle – this required more than an incremental improvement to the BREP based CAD models.
Bradley Rothenberg
nTop is short for nTopology, which means “any shape” – I started nTop to build the most advanced computational design software that enables you, an engineer, to build a model that captures not just one version of a design, but all possible designs within a given design space driven by inputs that you define through our block system.
Input requirements for a design, and nTop software computes the shape, if the input requirements change, the shape is recomputed.
The modeling system is built on an implicit model representing a Signed-Distance Field (SDF) – it doesn’t need to be computed sequentially, rather, the entire model is computed in parallel across multiple CPUs or in real-time w/ GPUs.
We launched nTop in 2018 with General Atomics as our first customer. Since then, we’ve landed in over 400 accounts and are used by engineers worldwide to model some of their most advanced designs, from FAA certified lightweight aero-structures, heat exchangers and manifolds, to FDA certified medical implants. Our software is used across aerospace, automotive, medical, energy, and consumer products / sporting goods, mostly for the design of 3D printed parts & products.
Adoption of nTop at Siemens Energy
Markus, could you share the journey of Siemens Energy’s adoption of nTop? What is your role, how did you get started with nTop, and how has it aligned with and supported the initial objectives that led to its integration into your workflows?
I am a heat transfer design engineer within the central metal Additive Manufacturing (AM) organization serving both our own internal, mostly gas turbine market, as well as supporting our growing external customer base.
At Siemens Energy we don’t just provide printed AM parts. We also provide our engineering services for external customers to leverage, which is quite unique within the world of AM.
Markus Lempke
One product group that is of high interest right at this moment are customized and highly efficient heat exchangers, as well as other heat management applications where high surface to volume ratios are paramount.
My initial interest in nTop was in the exploration and application of TPMS structures within thermal management applications.
The ability of nTop to utilize implicit modeling to develop lattice structures is so strong. Indeed, nTop has helped us explore a variety of conceptual heat exchanger designs over a large range of boundary conditions as well as modules for the direct capture of CO2 from the atmosphere.
Working jointly with my US-based design colleague Andrew Kappers, I have now assumed the role of a “power-user”, driving the adoption of nTop both inside Siemens Energy as well as for external projects, constantly exploring new as well as existing fields of applications in all sectors of industry.
Thanks to their kind support of CDFAM Berlin you can use the code NTOP for a 20% discount on tickets at checkout
nTop’s Collaboration with Siemens Energy
Brad, could you describe how nTop collaborates with companies like Siemens Energy to expedite their realization of value from using nTop?
Computational design is a new paradigm for engineers – i.e. it’s a different process to design a model in nTop than to design a part in CAD.
When you are building a computational model in nTop, you are essentially defining a set of rules & relationships between geometry and data (usually represented as fields, but also sometimes discrete points, values, vectors etc.).
Bradley Rothenberg
This computational model represents not just one design, but all possible designs within a given set of a parameters – for example, you might have a model of a lower level design feature, like an isogrid structure that can be applied to any surface to add additional stiffness to it, or you might have a computational model for a part-level design, like a 3-domain heat exchanger that can be sized to a specific set of requirements / locations of inlets & outlets, all the way to a computational model for the entire structure of a robotic assembly, like the Ocado series 600 warehouse robot.
I mention the above because building a computational model usually takes a little bit more time upfront, but once it’s built enables the fastest possible iteration enabling customers to improve their designs much faster than having to manually iterate / manually model changes.
When we work with companies, usually it is during the first phase to help them build out a computational model, while teaching them the fundamentals of implicit modeling through this lens.
It’s in the rapid-fire iteration phase that the value of a computational model is realized (though you learn a lot in building the model too – I find that part really fun).
We’ve built a customer success team that includes some of the best engineers in the industry to work closely with our customers to realize the value of nTop – One of their programs is onboarding customers and during that period, we can teach building computational models via2-day on site workshop or through remote onboarding – We are also in beta right now for classroom sessions which multiple customers can join regularly. I prefer going on site, as I’ve built relationships over the last several years, and have made really good friends in the process of doing this.
Following on from this, could you discuss how this engagement has advanced your objectives and the process, from the initial exploratory projects, to extending nTop’s use across Siemens Energy?
Markus: As Brad just mentioned, computational design, and in particular implicit modeling, require a different way of thinking compared to traditional CAD software. Thus, proper training and onboarding is paramount in the initial phase.
But even beyond this we have greatly profited from the ongoing interaction with the customer success team and the solution engineers.
Even as “power-user”, I can confidently say that I have yet to fully explore the functionality of nTop.
Equally, for Siemens Energy to uncover new customers and develop industry-leading functionality our partnership with nTop has been crucial. It has resulted in new business for Siemens Energy and led to several ideas and feature requests which nTop are now developing.
Leveraging the “voice of the customer” to guide software development into a direction that is beneficial for the engineering community as-a-whole, is both motivating and rewarding for me personally too.
One aspect that is paramount in the wider adoption of nTop within Siemens Energy is the interoperability with other CAx software packages. We were actively engaged in the very first use case of the “Implicit Interop” interface: printing a heat exchanger directly from the implicit file format. Also in this matter we try to leverage our voice to advocate for the adoption of the implicit file format through our existing contacts with other software companies that we rely on in our development, manufacturing and quality assurance processes.
The most pleasant surprise in the journey to extend nTop usage within Siemens Energy, was the high interest shown by the engineering teams even from our more traditional product lines where we already have firmly established manufacturing processes (e.g. injection molding, casting) and well-developed design tool chains. This really underlines the strengths of implicit modeling for certain geometries, regardless of the manufacturing process.
Brad: I agree w/ Markus above in that I want to talk through what it takes for our customers to take their most advanced designs from nTop to production. Of course, since it’s CDFAM, I also want to show some of the new features that are part of nTop 5.
nTop + Siemens Energy CDFAM Presentation Overview
Could both of you provide a preview of the topics you’ll be covering in your joint presentation at the CDFAM Symposium in Berlin?
Markus: My intent is to highlight that computational design and implicit modeling have very practical uses within the world of additive manufacturing. nTop is so much more than a software tool to create fancy prototypes for exhibitions.
At Siemens Energy our focus is the serial production of high-performance metal components. Typical manufactured batch sizes range from dozens to thousands of parts per year. In this context I want to highlight how nTop helps us create robust parametric models for optimization and how we leverage field-driven design to simplify and “tune” our heat exchanger designs for end-customer use. Furthermore, I will showcase how computational design can help to reduce cost in the AM manufacturing process, by optimizing support structures based on process simulation data from ANSYS Additive.
Selecting Impactful Applications at Siemens Energy
Markus, when considering computational design and advanced manufacturing techniques such as metal 3D printing, how do you identify the applications where they can deliver the most significant impact and provide return on investment?
Markus: There are certain factors that are already a good first indication for the likelihood of success:
Is it a clean-sheet design?
While supply chain resilience and lead time topics can also be a major motivation to shift to additive manufacturing, the additional design freedom that AM brings to the table usually remains untapped in these situations. Accordingly, the necessity for computational design of legacy parts is often quite limited. However, with full design authority (within conventional constraints such as bolt holes, connections, etc.), you can unleash the full potential of AM. This extends beyond the AM component itself to the overall assembly, e.g. because that enables more flexibility in the type and position of the internal interfaces.
Is part performance / functionality paramount?
At Siemens Energy our focus area in metal AM is the Laser Powder Bed Fusion (LPBF) process. This process is ideally suited for parts with intricate details under high thermomechanical loads. But AM can come at a cost. Therefore, the additional cost must be justified by an improvement in the functionality and performance of the part. To achieve this, it is also paramount to leverage computational design to quickly create and digitally assess design variants so we can print right first time.
Is the cost for the conventionally manufactured part also high?
As mentioned earlier, AM in most cases comes at a premium. However, some components are expensive to produce even with traditional manufacturing techniques. A good example here are complex assemblies with tight tolerances. Here the additional cost for the manufacturing part can very often be offset by savings in assembly cost or also by reliability due to the reduction of complexity. AM has the potential to achieve all of this whilst maintaining or even improving the functionality of the part.
Building on the topic of identifying and implementing high-impact applications, could you discuss how Siemens Energy evaluates and measures the performance outcomes against the initial investment? What specific criteria or bench marks do you use to ensure the chosen applications meet or exceed your performance and ROI expectations?
Markus: As a company with a long tradition, Siemens Energy has generations of products with no lack of benchmark example cases that we can choose from. From day one, we were under a lot of scrutiny, as we worked our way through a catalog of parts to replace in already well-established gas turbine products. Even now whilst extending towards a wider product range, more often than not, we find ourselves “qualifying-out” parts rather than “qualifying them in”.
The business case to move toward AM is a tricky one, as many parts have been designed the same way for decades and have been thoroughly optimized over that time. So, to ensure that we are focusing on the right opportunities we implemented the “Additive Manufacturing Component Implementation” (AMCI) board where a cross-functional expert team takes a rigorous look at the business case, technical risks/gaps and required budget for all candidate parts.
If the component passes this gate, we then follow the product development process (PDP), just like for the conventionally manufactured parts. This includes a lot of tracking and assessments (digitally as well as hardware tests) so that at the end we can confidently say whether we achieved the initially formulated goals.
We need to be this thorough to be able to deliver on the promise of AM: to reduce complexity, extend life, increase performance and all at a similar cost to the conventionally produced part. It’s a complex road and nTop has become one central tool in our toolbox that we leverage along this development chain.
Markus Lempke
Incorporating User Feedback into nTop
Brad, how does nTop incorporate feedback from Siemens Energy and similar clients to enhance the software’s capabilities for solving future challenges?
Our Sales, & R&D teams, along with myself, spend a large portion (sometimes the majority) of our time meeting with customers and listening to what’s working vs what’s bottlenecking their workflows – We spend a lot of time trying to really understand all of the hard problems customers are solving with nTop, and even more-so, what they are trying to achieve with nTop – it could be a more seamless feedback loop between the design and physics to more quickly understand how a parameter adjusting “waviness” reduces the pressure-drop, and ultimately improves the performance of a heat exchanger, or maybe how to gain a more intuitive understanding of all of the different parameters going into a model interactively?
Ultimately our customers only care about our tech (i.e the intricacies of implicit modeling or mesh-less simulation techniques) in relation to how this tech helps them solve a problem they couldn’t solve before.
Whenever we deliver a new feature that un-bottlenecks a customer workflow we have documented, we try to go back to that customer with a demo and sample workflow they can build off of.
Bradley Rothenberg
Following up on how nTop integrates client feedback into software development, could you give us a glimpse into future versions of nTop? Specifically, how are the insights and requirements from teams like Markus’s shaping the upcoming features or enhancements we can anticipate?
I’m not sure if it’s obvious, but I hear it almost every day – simulation right now is a bottleneck – more specifically the time it takes to get an accurate simulation of the physics using traditional CAE tools is too slow.
nTop opens up the ability for massive design space exploration really fast, however, for each run, if there’s physics in the loop, even with automated meshing and analysis, the bottleneck is the time it takes to generate a simulation-ready mesh, the size of the mesh and thus the time it takes to solve the finite element problem.
We have customers, like Siemens Energy, that are designing heat exchangers that, when meshed for CFD, could be on the order of 1B elements. This takes days to run, even on GPUs in Star-CCM or OpenFoam on CPUs.
I’m really excited by the progress at a number of other startups tackling this problem, both on the solver side, like cloudFluid with their cloud GPU lattice boltzmann solver, or Luminary Cloud with their cloud GPU fluids solvers. In cloudFluid we’re seeing problems that used to take days now run in hours, plus there’s no pain of having to generate an unstructured mesh. Additionally, other startups are tackling this problem by training up AI models that can quickly predict the physics, like Neural Concept, Navasto, or Navier.
We are even seeing a number of our customers start to train up their own AI models from nTop implicits + analysis, like at Ocado for their Series 600.
On the Physics and AI side, I’m most excited about the partnerships we are developing with some of the other startups in the space to define this generation’s end-to-end real time engineering workflow.
We’ve launched nTop Core, our SDK for our partners to integrate nTops implicit model into their tools and we’ll be announcing a number of new integrations in addition to those we currently have into EOSPrint, Fusion360, CloudFluid, and Inkbit.
Additionally, we are continuing to improve our implicit modeling capabilities, with the launch of nTop 5 coming up, we have an entirely new core modeling technology (Inside nTop, we’ve been calling it Sequoia, because it’s essentially a big tree of operations that is compiled into lower level code that’s evaluated in parallel on CPUs or GPUs to spit out geometry) – Our kernel (not the new restaurant) is pure implicit, which means we know everything we want to know about a model. Nothing gets dumbed down to the voxel level unless you want to use voxels – which is analogous to cleaning up an illustrator graphic in Photoshop – The beautiful thing about this is that B-Rep geometry stays precise B-Rep round trip through implicit modeling and back to a CAD assembly. Not only does nTop evaluate the implicit blazingly fast at native CAD double precision, nTop also evaluates its gradients, any parametric derivative, and the gradients of such derivatives. That’s why our interaction is real-time. It powers field optimization and puts geometry in a black box for machine learning via nTop Core.
We’ll be launching the most advanced renderer that I’ve seen for implicits enabling our customers to better visualize large parts with thin features.
I’m really bullish on the new generation of engineering software, & I expect that we will see our customers iterating in near real-time via end-to-end computational models of advanced systems, like a full jet engine (including all of the physics in the loop), within the next 5 yrs – enabled by the massive parallel compute we now have access to along with implicit modeling and more advanced physics solvers.
What I don’t think we’ll see is a full AI that I just say “AI, make me a jet engine,” and it spits out the entire engine design…
Takeaways and Expectations from CDFAM
Finally, what are the key takeaways you both hope attendees will gain from your presentation, and what do you aim to learn or achieve at CDFAM?
Markus: As previously mentioned I really want to share examples with the community that show how we are using computational design and implicit modeling on real parts to solve problems with tremendous impact on the speed and success of the energy transition.
I think this can vastly contribute to the sense of purpose that a lot of us feel.
With so many software vendors being present, I would also like to use the “voice of the customer” to advocate for increased standardization and interoperability around implicit modeling.
And then last but not least I can’t wait for all the inspiration and creative ideas I will soak up from the community and that will help me tremendously in existing design challenges and those yet to come.
Brad: I hope to share with the community my experience of and learning what it takes to build industrial parts using new implicit modeling tools and I hope others can learn from what we’ve built and improve on / speed up current bottlenecks in our customer workflows.
I’m most excited to spend time with close friends and make new friends while all of us are sharing advanced new software and algorithms to solve design and engineering challenges.
Thanks to their kind support of CDFAM Berlin you can use the code NTOP for a 20% discount on tickets at checkout
