How AI and ML will shape the future of mechanical design

In the latest episode of Lexicon, Interesting Engineering (IE)Now we turn to Dr. Jaroslaw Rzepecki, Chief Technology Officer (CTO) at Monomo.

“Our mission is to improve the efficiency of motor systems in unprecedented ways, enabling them to use precious resources more sustainably.” Monumo explainsMonumo is working hard to achieve that goal, using proprietary datasets and machine learning techniques to build one of the world’s first “large-scale engineering models” (LEMs).

As it matures, this model could work like Engineering R&D Midjourney or Dall-E, helping engineers create plans for components or entire machines on demand—a quantum leap in computer-aided design (CAD), if you like.

While the interface is not as “simplified” as you’d expect from a large-scale language model (LLM) like ChatGPT, it still leverages our engineers’ time to create the best kit they can imagine – and the potential is huge.

Meet Jaroslaw Rzepecki, Chief Technology Officer at Monumo

Jaroslaw Rzepecki leads the company’s technology development, oversees the hardware and software development pipeline, and directs machine learning (ML) research.

Prior to joining Monumo, Jaroslaw was a key member of the Codemasters team that developed the racing video games Grid and Dirt 2. He also worked as a software engineer at Siemens and held senior positions at Microsoft Research and ARM.

During an interview, he told Interesting Engineering that he spends some of his free time practicing martial arts, specifically kickboxing. I asked him if martial arts has helped him in his professional life.

“Well, it’s a bit like my professional journey, so I’ve tried my hand at different disciplines, I’ve changed clubs and, of course, I’ve changed my style a bit,” Jaroslaw said.

Don’t underestimate the power of physical training

“I’ve done a lot of different sports. My favourite sport is kickboxing. I’ve probably been doing it the longest out of any other sport, but whether it helps or not, it does help. I think it helps me concentrate better. It helps clear my mind,” he added.

“Afterwards you’ll probably feel physically tired. You’ll feel a lot more energized. You’ll have more energy to do something that day than if you skipped training the day before, so I’d say it definitely helps,” Jaroslaw said.

After a wide and varied career that included academia, computer game design, and software engineering at Siemens and ARM, Jaroslav realised Monomo’s potential and left to become the company’s second employee, after which he rose through the ranks to become head of the company’s engineering department.

When asked if this was a big risk for him, Jarosław replied, “Well, whenever you change something there’s always some risk involved, right? But without risk it’s no fun, right? So, yes, I think there was some risk involved. But, well, like I said, I calculated the risk and I thought it was OK.”

Using machine learning to create new machines

The main goal of Monumo’s research is to combine knowledge of physics and engineering with machine learning (ML) and artificial intelligence (AI) to build computer models that help sketch new models of machines. The idea is that with enough data and training, such models could be used to create novel designs that were previously unimaginable.

Moreover, it will be data- and expert-driven, and not just for everyone to use – this is mainly because Monomo intends to own the software proprietaryly, but also because the software is essentially a complex, multi-disciplinary physics model.

This combines data and understanding from different engineering disciplines and physics, and potentially integrating a variety of other disciplines. This includes atomic physics, nanotechnology, biology, geology, etc. If it can be measured or modeled, it can be integrated into the model.

“We have a one-sentence headline here, but I think anyone in the engineering community listening to this podcast will understand how challenging it is when you’re designing a complex engineering system to find the right balance between its various components,” Jaroslaw says.

We are an army

“It’s a hard problem. I like challenges, and I like hard problems. And when you apply deep tech to engineering, it automatically becomes a multidisciplinary problem because, naturally, you have to combine the latest developments in algorithmic optimization in computer science, mathematics and physics,” he added.

But LEM is a long-term goal. For now, they Answer Model It is important not only to be able to generate the model, but also to be able to provide training data for the LEM later. Monumo is focused on making electric motors as energy efficient as possible.

When pressed on the issue of LLM and hallucinations, Jaroslaw explained that Anser, and ultimately LEM, are not affected by this, because the designs generated are “sense-checked” with mechanical engineering tools to assess their feasibility.

If it doesn’t pass, the software flags the issue and the user goes back to the drawing board to revise the design accordingly. The entire design process is just like in real life, going through multiple stages to arrive at the final piece.

It’s a collaborative approach, like tweaking the parameters of a Midjourney or Dall-E to get the image you want. Anser can also incorporate specific customer considerations and constraints into the design, based on the customer’s needs.

Walk before you run

With so much of our modern world using energy in one form or another, even small improvements in energy efficiency could save millions of dollars around the world – less energy wasted is a win-win for the planet.

“Therefore, any improvements that can be made to electric motors will have a huge positive impact on the ecosystem and help move society towards a greener way of living,” Jaroslaw said.

The company chose electric motors because it’s a simple yet complex enough problem that if the Anser proves itself on this stuff, with enough data and training, it could be used for essentially anything (within common sense).

“The techniques we’re applying and the simulations we’re building are multiphysics simulations, so they can be applied to other engineering disciplines. So, yes, we’re laying the foundations and building simulations that are flexible enough,” he explained.

“LLMs (large-scale language models) are what power today’s AI models to mimic human capabilities with words and pictures. In the future, LEMs (large-scale engineering models) will produce solutions that go beyond what humans have achieved so far. By combining our ability to run and store massive amounts of simulations with our optimization intelligence, we are already ready to build these valuable data sets and train new models,” Monumo explains.

Your job is safe, but improve your skills just in case

And don’t worry that such models will take away your engineering job: Jaroslaw explained that Anser and its descendants should be thought of as new, capable computer-aided design (CAD) software.

“I don’t think we need to worry about engineers losing their jobs. We’ll always need engineers. You know, these are all tools and we’ve seen it in principle every time a new tool is developed,” he said.

“Humanity has a choice: we can use these amazing new tools to do the same thing we’ve always done, but with fewer people, or we can use these new tools and all the people we have to do more. And our attitude has always been to do more,” he added.

So, it may be time to hone your AI and ML expertise.