Automating Print Settings for Confident Results – AMAIZE AI by NEOFORGE

We sat down with Omar Fergani, CEO and co-founder of NEOFORGE to talk about why metal additive manufacturing is still stuck in trial and error, what physics-driven AI actually means, and how AMAIZE turns the traditional design-produce-test-learn cycle on its head.

The Enormous Cost of Failure

What is the problem in metal printing today?

Metal additive is a critical technology for some subdomains like space and defense. However, although the promise of metal AM is so huge, we don’t see it scaling and we are not seeing adoption all over the world as we thought we would be by now.

There are multiple problems. The most obvious one is that it is an expert-demanding industrial technology. Most people think it is like FDM. I buy a machine, I have a design, I push the button and I print. As a matter of fact, it requires a huge amount of expertise, and the only way to train those experts is to make them go through a lot of failures. But the cost of these failures can be pretty huge, considering the multi-million dollar machine time and production materials like titanium or Inconel. So there aren’t many companies that can afford such personnel training.

The second issue is that getting the prints right the first time, even with experienced operators, is difficult. When you fail many times, you are wasting time on an expensive machine, wasting expensive material, creating delays. That adds to the cost structure.

And the third is that the use cases that would benefit most from AM are not yet widely adopted.

Thermal FEA has been talked about as a solution for print success. What do you think about that?

Simulating instead of going to the machine and trying and failing is very smart. More simulation, less physical iteration is always very wise. But is finite element the answer? From my perspective, no.

Manufacturing is not an analysis job. Manufacturing engineers are never going to adopt finite element. It is very complex, adding another level of expertise. And the level at which the finite element needs to be used to solve manufacturing problems is on the toolpath. Simulating 2,000 miles of toolpath with finite element would take probably 24 million years of compute. So saying that FEA is the answer is just not realistic.

However, the good news is that with AI we moved from deterministic PDE solving to a probabilistic, statistical-based approach. With that we are able to do 2,000 miles in a few minutes. That is the opportunity that machine learning brought.

What made you and your co-founder Katharina start NEOFORGE?

Katharina and I have complementary backgrounds, she brings deep expertise in AI and physics-based modeling, I bring years of end-user and machine-OEM experience watching the same problems repeat across customers. That combination is the founding bet of the company.

We came together around a specific conviction: the additive manufacturing industry was trying to solve a fundamentally probabilistic problem with deterministic tools. Finite element methods, no matter how sophisticated, were never going to scale to the toolpath-level resolution that real production requires — the compute cost is prohibitive, and the user experience is wrong for manufacturing engineers who are not analysts.

What we saw, before anyone was talking seriously about AI for manufacturing, was that machine learning trained on physics-grade data could collapse the compute problem from years to minutes. That insight, that the right paradigm was probabilistic and learning-based, not deterministic and solver-based is what AMAIZE is built on. Founding the company was the only way to pursue it at the depth and speed it needed.

Simulations that Reveal Risks and Provide Solutions

What is AMAIZE?

At the core of AMAIZE there is a foundational model that understands everything about laser-material interaction. We use this model for simulations. If you have a very large component that will take weeks of printing and the risk of failure is high, we will simulate that in a few minutes. We see where the risk of failure is, provide feedback to the designers, provide solutions to the manufacturing engineers so they can do all the learning upfront before even finishing the design and printing.

Think of it this way. The whole engineering production testing learning cycle has always gone the same way – design, production, testing, learning. We now have the possibility to bring the learning to the beginning. We tell engineers, don’t finish your design, don’t produce, just do as much learning as possible. It is possible now because it is computationally efficient. Once you have done all these digital iterations that are near-zero cost, you finish the design, produce, test in the physical world. That is the paradigm shift we bring.

And this is not just a digital technology swimming in the air. AMAIZE is the first technology in the world that was actually implemented in industrial-grade machines. Even more interestingly, we are the first company that actually drives the scanner, not just the machine. We integrated our AI in the laser scanner, SCANLAB for example. So it is a technology that goes from the design to the iteration to the execution on the machine.

So how does it actually work?

Whatever an expert would have done, the software will do now by running tons of simulations, nesting the part, putting the supports only where needed, optimizing the scan strategy to achieve the highest quality, the best surface finish, slicing and hatching to give you a G-code that is ready to be sent to the machine. Full end-to-end. The human sees what the technology is doing, can analyze it, can extract the data, can question it. But the reality is we are using compute to do the whole work.

Some of these objectives are competing, obviously. So we do multiple optimizations and show the customer different options represented by different G-codes. The software thinks that these three are most interesting because one optimizes for cost, another for quality, and a third for production time. It is your decision in the end which one you want to go with. And once you choose, you can literally just push the file and click “print” because we are integrated with your OEM.

What if my design just is not printable? What happens then?

When you drop the file, we are not only providing you with the G-code. We show you the result of where we see the issues. So you as a designer can see your design in 3D as it is – where the high critical areas are, where even with our optimization we could not reduce risk.

Think of a manifold. Imagine you are printing that. The zero-angle overhangs, no matter how much I optimize, are going to be very hot. The software will show you that you still have an issue in your design. That is the first benefit of our tool: design for manufacturing.

Example of a maniford part without AMAIZE in this orientation is not possible without huge support and defect

Example of a maniford part without AMAIZE in this orientation is not possible without huge support and defect

We show you where the issue is, we optimize it, it does not work, you need to work on your design. And here we saved you a lot of time and money because you did not go to the machine, you did not see the failure, you did not have to do monitoring. Near-zero dollar learning.