10 challenges for AI in MBSE and the 8 skills required to meet them

At the INCOSE symposium last year, Mohammad Chami, Nabil Abdoun and Jean-Michel Bruel presented an interesting "vision paper" on the challenges for AI in MBSE. I have been of the opinion for a while that AI has the potential to bring MBSE into the mainstream.

Below are the most important contents of this well-structured paper.

The Paper was published by Wiley and is available free of charge for INCOSE (or GfSE) members.

Why

The authors see new technologies and competition as drivers. Furthermore, products are becoming increasingly multi-layered and complex. It is becoming more difficult to translate customer needs into product features. In addition to unpredictable costs, some products also pose a risk to life and limb.

MBSE has long been recognized as a way of mastering complexity. The topic of AI in product development is not new and was already investigated in the 1980s.

The authors have divided the main part of the paper into two parts. In the first part, they identify a total of ten challenges for the application of AI in MBSE; in the second part, they examine the required capabilities of AI to solve these challenges.

10 MBSE challenges

The authors did find information on MBSE in the literature, but without analyzing its dependencies. This is a significant contribution of this research paper. Figure 2 (in the paper) shows these dependencies mapped to the typical project phases. Of particular interest, of course, is how the 8 capabilities help to overcome the 10 challenges. More on this at the end of the article. Just this much in advance: The skills are a useful list for using AI in MBSE. But the skills alone are not enough for success.

Here is the list of the 10 challenges:

1st challenge: Upfront investments

As with any new method, MBSE requires a considerable upfront investment. One of the biggest challenges is therefore an effective investment strategy with a comprehensible ROI for decision-makers.

2nd challenge: Implementation strategy

Companies can decide to introduce MBSE in a sandbox environment or use it directly in production. The first option costs (additional) money, as does the second. However, with the second option, the risk is considerably higher, but it also promises to be able to use the added value of MBSE immediately.

Challenge 3: Definition of purpose and scope

The "why" and the "what" must be clarified and coordinated with the implementation strategy. But this is easier said than done. The teams must start with changes and should be refined accordingly in the development phase and used to measure the benefits during implementation.

4th challenge: Sensitization and resistance to change

The human factor plays a central role in the changeover. Especially in the initial phase, it is often the case that the key players are at different levels of MBSE knowledge and there is not enough time for training. We have to expect resistance and a lack of skills. Employees need a new way of thinking and new approaches to collaboration.

5th challenge: Support from the management level

Awareness of MBSE is slowly getting around at management level - albeit very slowly. As a result, there are still regular conflicts of objectives. Unfortunately, a short-term solution often wins, which sabotages the MBSE implementation process.

6th challenge: Definition and expansion of the method

Three things are important for the implementation of MBSE: modeling languages, modeling tools and a modeling method. The first two points are relatively simple, but the modeling method is where it gets difficult. Even well-documented methodologies must be adapted, documented and supported by the team with appropriate modeling rules, guidelines, tool adaptations and training materials.

7. challenge: modularity and reusability

Many companies still follow an opportunistic and isolated reuse approach ("copy & paste"). But one of the big advantages of MBSE (if used correctly) is better reusability. If we don't do this properly in the early stages of MBSE implementation, a major benefit of MBSE could be lost. On the other hand, we must also be careful not to Ivory tower models generate.

8th challenge: Dealing with complexity

Mastering complexity is an important driver for the introduction of MBSE. The drivers of complexity include software and networking, which drastically increases the number of dependencies. This level of complexity often pushes existing methods and tools to their limits.

9th challenge: Tool dependency and integration

Companies must be prepared for Tools and train their employees accordingly. This decision has major long-term consequences, as a change can be very costly. Furthermore, a modern product requires several tools that integrated have to be made.

10th challenge: Visualization of large models

Data is not the same as information. Therefore, extracting information from models is an important task that is not trivial. Navigating the model alone can be difficult. Many stakeholders will not want to work in the modeling tool, requiring additional effort to create views or reports.

AI4MBSE: AI capabilities for MBSE

Under the term AI4MBSE the authors have collected eight capabilities that AI systems must have in order to effectively support MBSE. As we will see later, these capabilities alone are not enough to mitigate all the challenges described above.

In the paper itself, the authors also go into further detail about the dependencies between the individual capabilities. This can be helpful, for example, when prioritizing an implementation.

Capability 1: Competencies regarding AI in MBSE

As the application of AI in MBSE is still at an early stage, we need to build up expertise here. This task lies with universities and other educational institutions.

Capability 2: Automation

An AI system should be able to relieve the team of work. This can even be complete automation. However, there are a few things to bear in mind, as we will see in the next skill.

Capability 3: Control of the application

In MBSE in particular, uncontrolled misapplication of results can lead to catastrophic outcomes. We therefore need to control the use of AI, for example by intervening when results are obviously incorrect.

Worth reading is this articlewhich, among other things, is fundamentally about the trade-off between automation and augmentation when using AI.

Ability 4: Graphic visualization

Since the representation of models plays a central role in communication, the authors see graphic visualization as an important skill in its own right. The visualization may also be adapted to the method used.

Capability 5: Support for AI extensions

Just as we need a customized method (Challenge 6), a method may also need a specific AI extension. This requires metamodels that perform a mapping between the modelling language, the AI algorithms and the graphical visualization elements.

Skill 6: Data retrieval techniques

The purpose of MBSE is, among other things, to answer questions from system engineers about the model content. An AI-based system must support users with appropriate semantic query techniques.

Capability 7: Ontologies

The effective use of AI in the context of MBSE is supported by a Ontology for the modeling of language. This is particularly important for dealing with natural language.

Ability 8: Acquiring knowledge

AI can support us in capturing knowledge. It can collect the knowledge of domain experts, analyze this knowledge and formally extract it into the system model and validate this knowledge with the help of the ontology.

Relationship between challenges and skills

The eight capabilities have the potential to mitigate the challenges in MBSE. The authors have illustrated a traceability matrix in Figure 4.

Almost more important than the mapped relationships are the missing relationships. In particular, challenges 5 (management support) and 9 (tool dependency and integration) are not supported by any of the AI capabilities, which is understandable.

In practice, however, these dependencies are of secondary importance. After all, the number of skills is manageable and all skills contribute to the successful use of AI in MBSE.

Conclusion

An exciting paper on a highly topical subject. Thanks to its holistic view, this paper is also suitable for people who need to make a statement on the use of AI in product development in their company.

In the future, the authors plan to validate their results by means of surveys. I am looking forward to further publications.

Image source: FreePic