“No lightweight design without digitization”
Optimum lightweight structures are created when processes in development and production are optimally coordinated. Thus, lightweight design plays a key role in the introduction of digital development chains, as Peter Middendorf from the University of Stuttgart and Wolfgang Seeliger from Leichtbau BW explain in an interview.
lightweight.design: Industry 4.0, digital process chain, digital twin. Mr. Middendorf, are you still up to date in this regard?
Middendorf: We actually have many acronyms: the digital twin, more recently also the digital shadow, Industry 4.0. Here, it’s always about knowing what happens in production and continuing to work with this data.
You now also involve digital prototypes in your research work.
Middendorf: Exactly. For the moment, it’s more on the development side and not so much on the production side. Of course, there are a lot of interfaces, though: For the digital prototypes, I work with classic design tools such as CAD and CAE. We’re expanding the whole thing with CAM systems into the area of production technology. From there, it’s not far to production. Therefore, the digital prototype is more complementary to the digital twin that is then used in production.
“You become extremely fast in terms of time-to-market.”
First prototype, then twin...
Middendorf: … and in the next phase, we go even further: with the digital fingerprint. This corresponds to the completely digitally recorded workpiece and extends through development and production right through to the use phase.
For all the forensic scientists: what does the digital fingerprint involve?
Middendorf: Interesting data: how is a component used, what loads and how many load cycles does it experience during use? Such information can then be reflected online back into the production and development process in order to improve the product. For example, we can develop maintenance-on-demand concepts. Then I might not have to go to the workshop every 20,000 km to replace a component, but instead only when the component has actually reached its end of life.
What role does all this play in lightweight design?
Middendorf: A vital one. Lightweight design doesn’t just consist of the classic lightweight construction of materials. We also have functional lightweight design where, for example, we can make components more intelligent by means of sensor technology. In addition, we have structural lightweight design, which includes topology optimization and general interpretation methods, for example. And here is where it gets particularly interesting: for road traffic, railways and aviation, for example, we have to provide safety buffers. Some of these safety buffers are simply due to uncertainty in terms of the design. The more predictive I am in design, the more weight I can save without having to develop a new material, and without having to perform new shape optimization or new production process. I see tremendous added value in this regard.
Until then, however, a lot of research and development work lies before us. Does that mean that the subject is of interest for business?
Seeliger: Yes, we’re already trying to incorporate it into business models. One of our partners always says: without digitalization, no lightweight design. That’s because in lightweight design, we push the limits of the material. You have to test them, digitally if possible; otherwise, it’s too expensive. The test results then have to be reincorporated back into the design. This no longer works with paper and pencil; it happens much more often and much faster digitally. On the one hand, you optimize the product and on the other hand, you become extremely fast in terms of time-to-market. The requirement to continuously digitalize all development processes really first appeared with lightweight design. That’s why the topic of lightweight design is now the focus. Ultimately, however, all other industries benefit from it, as well.
Conversely, in the development of lightweight structures, there are acute problems that can’t be solved with existing techniques.
Middendorf: No, they’re not acute problems. Of course, we see that it’s not enough to design parts digitally and then throw them over the fence to the production department. In turn, production processes affect the material, tolerances and quality. That’s why we no longer just perform classic finite element design or the like; instead, we also model manufacturing processes digitally. For example, we’re currently making this very prominent for fiber composite materials.
Why for fiber composites in particular?
Middendorf: Because, on the one hand, it’s lightweight design par excellence and, on the other hand, the material of course only comes into being upon production. Before that, I only have my fibers and plastic and I only bring them together during production. If we think that through consistently, I already have an unlikely amount of data in the development process. Then comes testing, so-called virtual testing. More and more data is being generated there, too, which I hand over to the next production step, to the next manufacturing step, to quality assurance. Then, a lot of questions come up that originally didn’t result from lightweight construction at all: data formats, interfaces, big data.
Isn’t that a problem? In this country, we understand the development of material and production technology; innovative data technology usually comes from the U.S. Can we handle complex data or should we ask Silicon Valley about it?
Middendorf: Everyone has already made the pilgrimage to Silicon Valley. Of course, a lot comes from there in terms of software, algorithms and purely digital products. In the end, though, our product isn’t digital; it’s real. I think we’re skilled when it comes to issues like machine learning.
Seeliger: Our production expertise is an incredible asset that doesn’t exist in this form anywhere else. The second asset is development expertise. We’re often unaware of the specific expertise we have in complex development processes. This is precisely what we need for the digitalization of the development process. However, I also have to say: digitalization, Industry 4.0 — these are hype topics. I sometimes have the impression that, at least in Germany, we forget that we live from the physical product. We earn money with the product, not necessarily with digitalization. Our philosophy in lightweight design is that we optimize the product that the customer buys afterwards. However, what we need then is a digital development chain, an integrated process chain and networking with the production processes.
As a result, we have the necessary expertise for the digital process chain.
Middendorf: Yes, but we have to ensure that we bring together different disciplines — in research as well as in industry. Classically, we either have the mechanical engineer or the computer scientist. It is a task for training and the business to make progress here. I believe that we’re already well on the way.
And what exactly does the path look like?
Middendorf: When I can’t do something, I collaborate. For example, the topic of artificial intelligence isn’t my area. But there are a lot of colleagues I work with who are experts in that area. I may have the data, while others have the algorithms. Then we have to work together to create added value. A research factory like Arena 2036 is a good place for that. The same is true for companies: we have great companies, including many start-ups, that are active in the areas of machine learning, artificial intelligence, big data and cyber security.
Seeliger: We shouldn’t underestimate that we’re very good in the digital domain. Obviously, all of these CAD systems come from France or the U.S. and are owned by the company there. But they’re developed in Leinfelden, in Vaihingen or in Möhringen, because on the one hand, there is the engineering expertise, while at the same time there are also the digitalization skills to bring these things together.
About 15 years ago, mechatronics was the cutting edge of mechanical engineering and electrical engineering. Is the interface between mechanical engineering and computer science now leading to mechanical/computer science?
Middendorf: It’s just being developed. We have the Cyber Valley initiative, where precisely such customized degree courses are being developed. At the same time, of course, we aren’t giving up our focus on engineering. Instead, we are complementing relevant teaching content there. We can and have to adapt the curriculum. This isn’t easy, because we still want to get the basic knowledge in design, calculation and production within the same training period.
In engineering training, students deal primarily with the laws of physics and with making everything predictable. With all its mountains of data, though, digital networking requires the use of non-transparent black-box models.
Middendorf: This is very interesting. Of course, we’re reluctant to assign responsibility to a machine-learning algorithm that doesn’t know what elastoplastic material behavior is. That will come, though, because I can’t do that comprehensively just with our physically based models. It will be a merging of both, and we’ll see how far we get. I still hope that it will not make human intelligence redundant. Artificial intelligence and machine learning will play a big role, though, if there are adequate amounts of data available that are used properly.
Industry 4.0 in particular is an issue for OEMs that want to combine a wide range of capabilities under their roof and to make their processes more efficient. What interest do small companies have in making their data available for this chain?
Seeliger: That’s a big problem. Three years ago, we had a discussion with a big OEM about introducing the digital development chain. A large number of people were involved. However, technology was never talked about, not even digitalization! The conversation was always about who owns the data, who protects my data, and how can I retain ownership of it? Organizational and legal questions: those are the topics. As far as I know, they are still unclear and present the biggest obstacles to digital networking.
“I still hope that it will not make human intelligence redundant.”
What does a business model that addresses the concerns of all the partners look like?
Who develops the joint standards?
Middendorf: In our digital prototype project, we try to develop quasi-standards together with software companies and other networks. After all, it doesn’t help if we develop an Arena 2036 standard format here and apply different standards in Wolfsburg or Munich. We’re developing a standard that we also pass on and which is based largely on neutral data formats that software companies can adopt. Multiple interfaces are no longer created; instead, everything is saved in the same interface format.
Will this lead to the next DIN standard?
Middendorf: Something like that can’t be dictated, since a DIN standard can’t be developed; it would take too long. It has to evolve in the market; anyone who can convert to a neutral format has a competitive advantage.
What are the next steps in the development towards the digital twin?
Middendorf: We’ll see an increased use of sensors to generate even more data. These sensors will not only be there to record a specific pattern of behavior. For example, a crash sensor integrated into the component is nothing more than an acceleration sensor. And it may also be able to provide me with information during the production process as well as read certain data during operation. The use of sensors of all kinds, be it individual sensors, planar sensors or multifunctional materials, will do a great deal in the near future. I see a lot of potential in the linkage with the topic of lightweight design and digitalization.
Seeliger: In addition to sensors, I see the topic of machine learning and artificial intelligence. Introducing this to development processes in such a way that we shorten development cycles will take us a giant step further. Compare that with autonomous driving. Taking the picture itself is easy. But the difficulty is interpreting what you see there. And that has to come, we need it for autonomous driving, but we also need it for such digital development. On the one hand, we need the sensors and, behind them, the intelligence.
It sounds like an engineer’s dream: he always has all the data available, and he can control everything. When will the digital twin reach its limit?
Middendorf: As soon as people can no longer maintain an overview of it. Once it’s no longer clear for the developer or the production engineer why an algorithm acts in a certain way. That will be the point when we reach the limit. We’ll then be leaving the area where an engineer would still play a role. Nobody will be able to answer for the mistakes that arise after that. I don’t know if or how quickly we’ll reach this limit.
Seeliger: The limit is an ethical one. However, the fears that thousands of engineers will suddenly be unemployed are unfounded. We should see it the other way round: we can shorten development cycles, time-to-market and so on. Today, a car is developed within 48 months. If we can do that in 24 hours, it’s a huge opportunity. Highly intelligent and well-educated people who now spend weeks optimizing a wiper motor could, in the future, develop a new car every day. They could put their brainpower into things that we need to change. New visions could be developed and tested immediately, their costs could be estimated, and so on. We then suddenly reach a completely different dimension of development work. It’s a really big opportunity. Also for tackling a lot of problems that are faced by mankind. If we use it correctly.
Professor Middendorf, Dr. Seeliger, thank you for the interview.
Interview: Thomas Siebel