MESConf Open Space: How AI is revolutionizing MBSE and focusing on compliance
There was no way around AI at this year's MESConf either. In addition to a keynote on the topic, I led an open space on the subject of "MBSE support with AI". For the participants - and anyone interested - here are the results for you to read. To anticipate an exciting topic: The participants chose compliance as the most important topic regarding the use of AI with MBSE.
Keynote speech by Jan Seyler (Festo)
Festo is popular and widely used for professional tools in the skilled trades. But Festo has also long since made a name for itself in robotics. In his keynote speech, Jan Seyler showed many now familiar examples from the field of generative AI. What was new, however, was that he also showed results from AI systems that were older. This enabled him, for example, to compare current and older Pictures from Dall-E with each other. We were able to see that a few months made a quantum leap in quality possible. This was a vivid illustration of the rapid pace of development.
Open Space on AI at MBSE
In this Open Space we want to collect constructive, specific ideas for the use of AI in MBSE. The session was well attended with around 20 people.
First of all, we looked at the question of what AI actually is. Even if at the moment almost all the talk is about generative AI There are many different techniques and tools that we quickly lose sight of, such as image recognition or natural language processing.
However, the focus of this session should not be the "how", but the "what" and "why": What should an AI do in MBSE and why does it make sense?
With this objective in mind, we created a total of 17 use cases for AI in MBSE. The complete list can be seen below in the form of photographed flipcharts.
Once we had drawn up the list, we evaluated the individual points in the group. There were some interesting results and also a few surprises. The scale went from 1 (superfluous use case) to 10 (the absolute hammer).
Result 1: AI for compliance with MBSE is the highest rated use case
Only one use case achieved ten points: The topic of compliance support. We identified a number of specific use cases for this topic:
- Find relevant standards
- Implement "things" required by the standard (requirements, etc.)
- Indications of missing elements required for compliance
- Verification management (possibly partially automated
- Configuration (e.g. whether the microcontrollers were selected to be ASIL-compliant)
On our own behalf: For our AI quality assistant Semiant we developed a corresponding concept some time ago. I would like to exchange ideas with readers in order to validate the concept.
Result 2: "Copilot" for MBSE :Analyze Body of Knowledge
The AI assistant Copilot from gitHub may not be as well known as chatGPT, but it has made waves in the world of software development. Copilot programs on instruction and converts human speech into code. This use case was rated 9 out of 10 by the participants.
Copilot was able to use billions of lines of code from open source code as learning material. Unfortunately, MBSE does not have such a wealth of experience. In this respect, it is likely to be difficult to implement this use case in the near future.
Nevertheless, such an assistant can be useful even without "the world's modeling knowledge", especially for companies that have to do a lot of reuse. These include vehicle manufacturers (many variants) or plant manufacturers (each product is unique and customer-specific).
If the text notation of SysML v2 should gain popularity, even the user interface could look similar.
Result 3: Chat and content generation not very attractive
I've now seen half a dozen demo videos where the user describes something and the AI then generates a handful of requests (Valispace, trace.space, Visure, to name a few). This use case was only rated 4 out of 10.
The answer is understandable: Sure, something like that is impressive and gives something in a demo, but: (1) There are comparatively few new requirements are written. During development, existing requirements are primarily developed further. (2) This works well for a familiar product (e.g. a drone), but as the AI obtains its knowledge from the internet, the results are likely to be rather meagre for novel or exotic products.
Chat also only had a score of 4 out of 10, although this is also a frequently cited and impressive use case. Here, however, it could be that the result was negatively influenced by the participants: After all, everyone present was a modeling expert. I could imagine that Chat would score much better with decision makers or support staff.
Conclusion: Two candidates for the killer app of AI in MBSE
The remaining use cases were in between with scores between 5 and 8 points, without any major anomalies.
I am sure that we will see exciting developments in the use of AI with MBSE over the next few years. Whether this will result in a "co-pilot for MBSE" is a big question, but there will certainly be a lot happening in the area of compliance. I will continue to report on this.
Image: Maximilian Willert
