2. News
  3. Stories
  4. 2018
  5. 02
  6. Resilience reaps success

We use cookies to enable you to make the best possible use of our website and to improve our communications with you. We take your preferences into regard and process data for analytics and personalization only if you give us your consent by clicking on "Agree and continue" or if you make a specific selection by clicking on "Set cookie preferences". You can revoke your consent at any time with effect for the future. Information on the individual cookies used and the possibility of revocation can be found in our privacy policy and in the cookie policy.

Agree and continue Set cookie preferences

Resilience reaps success

Deep Virtuality

Very practical work on the future of information technology is carried out at Volkswagen’s Data:Lab Munich. The best example: the startup DeepVirtuality of engineer Erich Payer.

When Erich Payer departs to catch the train to Munich every Monday morning, he leaves his beloved running shoes behind in the hallway. That’s a pretty big sacrifice for the passionate runner. But he travels light. Everything he’ll need in the next few days fits into a small trolley: Clothes, a laptop – and a DIN A5 leather diary. It contains just about Erich Payer’s whole world: Drawings, ideas, sketches, postcards, screenshots of simulation results, AI predictions, and fragments of e-mails to his youngest daughter Elisa, who studies in San Francisco.

In 2016, Payer founded the company DeepVirtuality, which develops artificial neural networks that enable designers and engineers to make sophisticated predictions on the most important mechanical properties of products. The company embodies a combination of 30 years of experience in engineering and deep learning, i.e. artificial intelligence (AI) and artificial neural networks. “How everything has now come together was somewhat by chance,” enthuses the 56-year-old. “The way we are combining the disciplines has enormous potential.”

Erich Payer

The volumes of data handled by DeepVirtuality when a “typical” neural network is being trained – so as to be able to predict the stiffness, strength, vibration or acoustics of vehicle components – is in the “high gigabyte range.” That rises to terabytes for more extensive AI predictions, such as for how vehicles behave in a crash. So the quantity of data is immense, but always depends on the task at hand. “In that respect, I hope that Data:Lab will provide me with fresh stimulus,” says the Austrian. He has three months’ time.

Payer is an engineer through and through. He grew up in a region of Styria where the steel industry dominated daily life until the beginning of the 1980s. The first stainless steel was developed in 1912 in Mürzzuschlag, the home of Viktor Kaplan, the inventor of the turbine that bears his name. Elfriede Jelinek, winner of the Nobel Prize in Literature, was also born there. People lived in Mürzzuschlag in order to work in the steel mill. Payer’s parents were no different. But his mother wanted him to “become something.” And so he was the first from the housing estate to gain the Matura, the general qualification for university entrance in Austria. He then went on to study mechanical engineering, establish an engineering office, and cooperate with the automotive industry for the first time. The main focus even back then: simulation methods.

It’s a product development software created by engineers for engineers. AI is a revolutionary foundation in this form.

And he still deals with simulation methods to this day. However, he had a roller-coaster career between then and now. After a setback in his profession, the engineer redefined himself: He vowed to work only on solutions that captured his full interest. For instance, he developed a finite element software that allows the physical behavior of any product to be simulated realistically on the computer under actual operating conditions. One that – to put it simply – answers questions like: How does a tennis racket react when it strikes the ball with full force?

Payer made his software available in the web as an open source project in 2011. He soon noticed what one of his young employees was up to, toying around with AI in the field of image recognition. And in 2016 he combined his engineering knowledge and the software he had developed with the AI know-how of his employee Michael Pucher. “Our new AI solution is based on that triumvirate. It’s a product development software created by engineers for engineers. AI is a revolutionary foundation in this form.”

Payer encounters mavericks and people driven by a passion in the Startup Collaboration Space of Volkswagen’s Data:Lab. Its aim is to harness hardware and software solutions from young companies. The startups are sponsored with €30,000, discuss things with their mentors and meet experts from the Data:Lab – and networking within the group is initiated. Payer presents his innovation on this cheerless November day: The world market for simulation software has a volume of more than five billion euros and is dominated by a handful of players. “With our AI solution, we could cater for a large slice of it – because we’re faster and more precise. But to do that we need the leverage of a company like Volkswagen.”

“I can best persuade others if I’m able to apply our AI solutions in pilot projects. The results then speak for themselves.” In the first weeks, Payer and his Data:Lab mentor Daniel Weimer set about defining the contact persons and arranging meetings. They travel to Wolfsburg and Ingolstadt. Payer is given the unique opportunity to present himself to the startup community with people such as Ellen Gehrlach from the Data:Lab Munich. There are a lot of presentations, calling cards are exchanged, and yet – one week before Christmas – the founder has the feeling that time is running out. Payer had hoped to persuade others quickly with pilot projects. Disenchantment sets in. “There’s still not anything concrete. Whenever I speak with engineers, they’re always fascinated by our AI solutions, but we’ve still not been able to acquire concrete projects.”

We’re now able to slash the training times. Whereas three-and-a-half days used to be needed to train a neural network, that just takes 13 minutes now.

Now comes the final spurt. Payer is looking forward to being able to work at his desk in Munich for a few days and rub minds with other startups from the Collaboration Space. It’s very quiet, no mobile phones ring, there’s no noise from the foosball tables that, if the clichés are to be believed, are part and parcel of a startup lab. All the protagonists are sitting here in airy, bright offices, working away almost noiselessly. When Erich Payer puzzles over an idea, he sometimes jumps up all of a sudden, rushes to the glass wall and jots down formulas with a pen. It’s not long before someone comes out of an adjoining room – the men discuss a bit and Payer changes the formula. Working together with AI experts, the short distance to the Data:Lab to obtain tips and support, the open and keenly interested community – he enjoys all of that. He still doesn’t have an inkling that he’ll receive a crucial tip in one of the next lunch breaks.  

At that lunch in the Data:Lab shortly before Christmas, the Austrian is sitting together with AI specialists. He’s talking passionately about DeepVirtuality and how the product development process can be shortened radically with AI solutions. The subject is AI libraries: Software libraries that provide learning algorithms for neural networks, for example. They can then be linked with the user’s own software. A colleague asks him which of these libraries he uses for training his neural networks. PyBrain? Would he like to try out PyTorch? In the space of just two weeks, Payer’s employee in Austria, Michael Pucher, adapts the company’s own software tools and the new AI library to each other – and a surprising breakthrough is achieved.

When you believe in something, you need resilience. After all, not everyone sees the potential of revolutionary new technologies right away.

In the meantime he’s also been given the opportunity to implement a pilot project for PSW, a subsidiary of Audi from Ingolstadt. Payer received the design data for a vehicle component and was tasked with showing how quick and accurate the calculations with his system are. They were compared with existing results of PSW. “After we obtained the data from PSW, we had the result in a matter of hours. And there was even a lower target/actual deviation than expected.”

Payer is itching to demonstrate the advantages – both technologically and economically – of his AI solutions in further trial projects. By way of comparison: Development times can be cut by 50 to 70 percent with conventional simulation. However, studies show that they can even be reduced by up to 90 percent with the use of AI.

The collaboration with PSW gives Payer cause for optimism. There are also already promising contacts with other companies from the Volkswagen Group and with Group Research. And firms far removed from the automotive industry, such as an Austrian ski manufacturer, are considering whether to deploy the AI solutions. “Those are strategic decisions for companies. And so it’s only natural that they want to see that it really works,” states Payer. At the end of the day, what counts are facts: “If our AI predictions match the results of far more time-consuming simulations, that convinces engineers.”

By the time Payer and Pucher pack their things on their last day at the Data:Lab, they have created the foundation for market launch of their AI solutions. “When you believe in something, you need resilience. After all, not everyone sees the potential of revolutionary new technologies right away,” is how Payer analyzes his time at the Data:Lab. He had hoped to open more doors in the three months. However, he was taking two very valuable things back home with him: “The people I had the honor of getting to know in my time here. And the tip from the AI specialist in the canteen, which has made our solutions so incredibly efficient” – a boon than came from a completely different direction than expected.

Pucher, Payer and his daughter Elisa, who has traveled specially to the final presentation, then set off back to Styria. The running shoes are waiting there in the hallway for the runner. And if you’ve met Payer, you don’t have to be a prophet to guess that a whole lot of new ideas will also be hatched on his next runs.


The companies

DeepVirtuality (Graz/Austria and San Francisco/U.S.): AI solutions to boost the speed of prototype development. Artificial neural networks are used in that instead of simulation. They can be deployed at different points: After the idea has been born, after the CAD model has been created or after a first simulation model has been created. At each of these points, an accurate prediction for the prototype can be obtained by training the neural networks.

Relectrify (Melbourne/Australia): Recycling instead of disposal. The company with its German head Valentin Münzel has developed the world’s first plug-and-play technology that significantly increases the service life of second-life batteries. The engineers take used batteries from e-cars and show how they can be repurposed and kept in the economic cycle, for example as high-quality household batteries. With demand for batteries for e-vehicles set to increase hugely by 2025, the quantity of old batteries will likewise rise. Relectrify not only makes an important contribution to protecting the environment by reusing them, but also shows how they can be a source of value added.

Medopad (London/UK): Virtual doctor in the driver’s cab. The largest player of the three startups was only founded in 2011, but is already established in the UK in the field of healthcare technology. Medopad collects patients’ data and offers tailored support – in this case for professional drivers. An app prompts them to take a break or exercise or notices when a driver gets tired.


Important note

When you access this link, you leave the pages of Volkswagen AG. Volkswagen AG does not claim ownership of third-party websites accessible via links and is not responsible for their content. Volkswagen has no influence on the data that is collected, stored or processed on this site. You can find more detailed information on this in the data protection declaration of the provider of the external website.

Continue to page Cancel