SysML v2 and AI: A field continues to develop

SysML v2 has triggered a little hype and has the potential to give MBSE a new boost. At the same time, artificial intelligence (AI) is still the hype topic par excellence. These two topics not only complement each other perfectly, there is also a lot of activity in this area.

So today at SE-Trends we take a look at formal modeling with Imandra, two AI-based free SysML v2 generators and a brief outlook.

By the way: I have presentations on this and related topics at Embedded Word Conference (12.3.), ReConf (7.5.), MESConf (22.5.) MBSE Summit (2.6.). Maybe I'll meet one or two readers there?

A field develops

Since ChatGPT triggered the AI hype at the end of 2022, many experts have been exploring the potential for MBSE. I have also been looking into this intensively and ReConf 2023 presented a prototype developed by me. This showed that:

• Generative AI can generate initial models quite well, but has difficulties with revision and change management.
• Tactically, AI can already be used for smaller aspects of a model, but is not yet scalable to larger systems.

Although there is a lot of research, many results are not yet suitable for practical use. An interesting work in this area is AI Assisted MBSE with SysML by Tim Weilkins, Doug Rosenberg and Brian Moberley, which also explicitly addresses SysML v2 (I am planning a Review).

AI and formal modeling with Imandra

Imandra is a company from Austin, Texas, that offers "Reasoning as a Service". Reasoning here refers to analysis based on formal models. Imandra uses the Imandra Modeling Language (IML), which initially has nothing to do with SysML. IML is a subset of the functional programming language OCaml.

A model based on SysML v2 is a formal model. The Kernel Modeling Language (KerML) is the basis for formal semantics and enables formal verification. Imandra uses these possibilities to convert SysML v2 models into IML using a special translation and to check the models for reliability and security using AI-supported analysis tools.

By integrating the Imandra engine, models can be systematically checked for possible weaknesses, errors and inconsistent states. Imandra combines mathematically precise verification with the ability to automatically identify faulty scenarios and provide suggestions for improvement. The results show that AI is not only able to analyze models more efficiently, but also to optimize them specifically for robustness and rule conformity.

2 free SysML v2 generators

In addition to professional solutions, there are now also a number of free tools, although these are more in the marketing sector.

1st ChatGPT-based SysML v2 generator from the HOOD Group

The HOOD Group has a ChatGPT-based tool which promises the generation of SysML v2-compliant state machines. Well, there is some room for improvement. I first had a state machine generated for a traffic light. During validation in the Jupyter Notebook but unfortunately I got a few error messages.

I then set the targets a little lower and used the example prompt provided. This time the SysML was valid, but the state machine was rather disappointing:

2 Yeshat: High promises, sobering results

The company Yeshat offers a free SysML v2 generator that can supposedly create, maintain and visualize entire model series. Unfortunately, the reality is different:

• No valid model could be generated
• The promised visualization did not work
• Instead of a graphical model, only an (incorrect) ASCII graphic was output

Conclusion

SysML v2 and AI are a promising duo that could change MBSE in the long term. There is already some interesting research work and initial applications, but most AI-supported tools are not yet fully developed. Free generators in particular have so far only delivered limited results.

Nevertheless, the topic remains exciting: as development progresses, AI could play an increasingly important role in the modelling, verification and optimization of SysML v2 models. It remains to be seen which solutions will prevail in practice.