1000 Kelvin – How AI Copilots are Enabling the AM Industry Scaling
Interview with CEO + Co-Founder Omar Fergani
Developing and qualifying machine parameters for metal laser powder bed systems has been an incredibly time consuming and expensive endeavor for all, with duplicated effort and closely guarded secret ‘special sauce’ further holding back advancement, adoption and scaling of metal AM.
Once a company has their parameters ‘dialed in’ for a specific machine and material using test coupons, geometry specific requirements add another level of complexity that is often addressed using simulation to predict and compensate for thermal distortion and reduce build failures.
1000 Kelvin promises to bypass both the trial and error, AND the simulation process using a hardware and geometry agnostic machine learning process, to provide an AI powered co-pilot to get to ‘first time right’ parts.
Co-founder and CEO Omar Fergani will be presenting case studies on the use of their software at CDFAM in Berlin so we asked a few questions about their AMaize software capabilities and what people need to get started with it.
Could you begin by introducing 1000 Kelvin and detailing the software your company offers?
1000 Kelvin is an AI software company that enables engineers in Additive Manufacturing (AM) to get qualified parts to market faster at lower cost.
1000 Kelvin’s AMAIZE software uses AI models to predict print challenges and correct them directly in the print file, increasing the number of prints that are perfect on the first try.
Process simulation for metal AM has advanced considerably in the last 5-7 years. How does the integration of AI/ML technologies complement or compete with these advancements, and what does AI/ML offer that simulation cannot?
I respectfully disagree, the FEA-based simulation product has not been adopted by the end user as we had hoped, and except for some enhancements to the meshing technology, it has not advanced technically, and there are strong reasons for this.
First, I believe the underlying methods used for these simulations are not truly physics-driven; they are based on an extensive and difficult-to-control calibration procedure due to the nature of the inherent strain method.
One of the limitations of AM process simulations is the massive amount of time required to achieve a single prediction, and when it is fast, it means that substantial computing investments are necessary.
Furthermore, most of the simulations focus on macro-level aspects such as distortions and do not predict process-related challenges at the level of the melt pool and track. Computationally, this is not feasible using FEA.
Finally, once you have your prediction, FEA technology cannot provide optimizations and actionable solutions. Most engineers are able to conduct a few physical trials to gain the same insights.
Leveraging AI takes the approach to new heights. With our AMAIZE prediction, we can perform detailed analyses of the print file containing all vectors and process parameters. Moreover, we provide a platform for optimization of the job file. Lastly, thanks to hyper-fast computing and integration with machine OEMs, you can download your file. These capabilities allow engineers to save a significant amount of time and solve their problems swiftly. This is a game-changer, without any domain for comparison.
Could you discuss the origins of your initial training data concerning machine parameters and outcomes, specify what data a customer needs to provide to utilize your software effectively, and explain the measures you take to guarantee that proprietary data provided by customers remains confidential and is not shared with others?
We don’t use customer data for our training. Our models are geometry and machine-independent, but material-dependent. Customers need to provide AMAIZE with their print file that they will use on the machine (e.g., .SLM, .CLI, Openjz, etc.). We perform our inference and optimization on the print file.
We offer the highest level of IT security and work collaboratively with our customers’ IT departments to meet their requirements. For example, we can deploy on GovCloud to be ITAR compliant. We exclusively use AWS and deploy on servers as specified by the customer, typically in Germany and the USA. Additionally, for some OEMs, we have implemented additional encryption on the print files. For instance, when you analyze and correct a print file on AMAIZE, you will specify the machine on which the file must be executed; hence, it cannot be opened on other machines.
Could you provide some real-world examples illustrating how your software has been applied, including specific applications, and detail the time and cost savings achieved through its use?
We have observed our solution being adopted for multiple applications:
- AMAIZE is aiding the energy sector in printing complex spare parts that were originally intended to be cast, specifically valves and impellers, without any modification for additive manufacturing (MfAM). The significant impact for our customers is that they have increased the number of spare parts available with a much shorter lead time compared to casting (reduced from months to days, thanks to AMAIZE).
- In the automotive industry, we have clients utilizing AMAIZE for printing functional prototypes with far fewer supports (resulting in a 55% reduction for turbocharger housings), aiding in the elimination of post-processing machining and reducing their time to market for development projects. This has become a critical metric in the automotive sector today.
- Contract manufacturers are leveraging our solution to expedite their engineering processes and reduce non-recurring engineering (NRE) costs by up to 90% by utilizing our digital iteration workflow, as opposed to dealing with multiple print failures and iterations on the machines. Ultimately, they are not just cutting costs directly but also improving their delivery times, ensuring customer satisfaction, and fostering repeat business.
These are just a few examples that we find exhilarating because they go beyond the simple math of cost-saving. From our perspective, these applications are driving significant growth within our industry.
What indicators should a customer look for to determine when it’s time to move beyond trial and error or simulation methods and start using AMaize?
We have designed AMAIZE to be user-friendly, scalable, cost-effective, and computationally efficient, all with the singular aim of outperforming the economics of trial and error and traditional simulation software.
If you see yourself in any of these statements, please don’t hesitate to contact us:
- “I sometimes decline business opportunities because I anticipate that the complexity of the part will require too much engineering and iteration effort.”
- “I avoid using FEA simulation because it is too complex and time-consuming.”
- “My business is expanding, and I am struggling to recruit application engineers who have the necessary experience.”
- “I find that all these digital technologies are expensive, difficult to use, and often overpromise.”
- “I have solid business cases, but the cost per part is high, and I need to find a way to reduce it.”
These statements reflect common concerns we aim to address with AMAIZE.
What other genuine advancements and progress are being made in the application of AI/ML in engineering, distinguishing them from the often delusional marketing hype? How do you see these advancements impact design and engineering in practical, measurable ways?
The good news is that I believe we have reached the peak of the hype, and moving forward, everything should become more sensible for end users.
On a serious note, I foresee that CAx, as we know it, is evolving rapidly, and the engineering disciplines will undergo a significant transformation. I predict that the role of the engineer as we currently understand it will change dramatically over the next three years.
The time from product conception to market will decrease substantially. CAE and the physics solvers that support it will evolve into a data generation engine for many companies, which in turn will be used to create AI models. These models will be implemented at the design stage, significantly accelerating the design cycles. I have observed automotive OEMs with a wealth of simulation data speeding up their product innovation cycles. Companies like Navasto, Neural Concept, and others are already revolutionizing the industry.
In manufacturing, the most significant recent advancement was the introduction of computer numerical control (CNC), which led to a significant increase in productivity and improved quality and yield.
With today’s AI technology, I believe we are on the cusp of a second revolution, where production machines, both additive and subtractive, will no longer require human operators.
Co-pilot systems will bring an unprecedented level of speed and precision, as well as free up resources for manufacturing companies. This innovation will be propelled by AI models trained on physics and process data. Companies such as 1000 Kelvin, Productive Machines, and others will be instrumental in this transition.
My concern is that while these developments are already underway, I do not see how the training programs are evolving to meet this new reality. Furthermore, it is disconcerting that the adoption of these technologies is proceeding slowly, particularly in Western industries. It is critical to communicate this message, and I appreciate your efforts in this regard!
What are you most excited to work on or witness being developed in the application of AI/ML, specifically in the areas of design innovation and the process optimization of manufacturing?
I am witnessing and involved in a project thanks to my friend Matthias, the CEO of Navasto, which is showing massive progress towards GenAI for the design of engineered products.
Developments I once thought would take a decade to achieve may now occur much more rapidly with the integration of GPT, advanced mesh generation techniques, and domain expert GNN models. The concept of ‘make me a car’ might become a reality sooner than expected. This is truly exciting.
At 1000Kelvin, I am really excited about the recent work we’re doing in building recipes and helping our customers define the process routes to achieve not just impeccable geometry but also the desired material properties. Consider the significant positive environmental impact you could make by printing Ti6Al4V parts without the need for energy-intensive heat treatments.
Finally, what do you hope attendees will take away from your presentation at CDFAM in Berlin, and what are you looking to gain from participating in the event?
Firstly, I am extremely excited to attend CDFAM in my hometown of Berlin. This event is outstanding, with a selection of excellent speakers and a nice venue.
I anticipate a deep and open exchange and collaboration. I hope to effectively communicate that AI represents a tremendous opportunity for our industry and that we can harness its potential during the engineering phases to design and produce sustainable products.
Register to attend CDFAM Berlin to connect with Omar and other experts on the adoption of AI for design, engineering and manufacturing.
