Digital Engineering: The Convergence of AI and MBSE

Let’s delve into the fascinating intersection of Artificial Intelligence (AI) and Model-Based Systems Engineering (MBSE). My team here at Microsoft calls this combination Digital Engineering. In this blog article, we’ll explore why AI matters in MBSE, common challenges, and approaches to tackle them.

If you are unfamiliar with MBSE I wrote a quick primer to get you up to speed. You can read it here: Model-Based Systems Engineering (MBSE): A Primer (drjoeshepherd.com)

Why AI in MBSE Matters

Ever since GenAI was released to the masses, customers have been looking for ways, sometimes any way, to infuse generative AI into their businesses. Product design and manufacturing is no different. In many cases manufacturing customers are already faced with a pivotal digital transformation effort as they adopt MBSE practices. It makes sense then that they would also explore ways they can accelerate their capabilities by adding AI into the mix. Afterall, what's one more completely disruptive and transformational initiative? You might as well go all in, right?

Artificial Intelligence has the potential to revolutionize MBSE by enhancing efficiency, effectiveness, and resilience across the systems engineering processes. If that isn't enough to get you moving maybe you should consider the cost of doing nothing. If your competition, and the entire market for that matter, is investing in AI then it simply becomes table stakes to play the game. Standing still amounts to falling behind. Here are just a few of the potential use cases where AI can improve the digital engineering process.

Intelligent Requirement Analysis

Requirements are the cornerstone of all Systems Engineering (SE) efforts. Its largely recognized that if a vendor can elevate the quality of their requirements early on, the value will permeate the rest of the downstream processes. Some benefits include reduced manual effort, improved accuracy, and comprehensive requirement capture.

Automated Model Generation

Models are the basis of MBSE and SysML is the language in which they are expressed. The problem is that models are complex and the bigger the product the more complex they become. Imagine a battleship or a satellite and how complicated it must be to understand and model these systems of systems. The ability of AI to generate models from requirements, and requirements from models, can be a real competitive advantage.

Intelligent Model Verification

Systems engineers spend an exorbitant amount of time simply verifying that a given model or SysML diagram is accurate. The ability of AI to understand, analyze and validate models based on requirements can ensure what gets built is valid.

Defect Analysis

We are seeing customers use AI to reduce the impact of defects throughout the engineering process. While defect detection can be a challenge, AI can help you quickly determine what to do once a defect is detected. This can quickly reduce the amount of time needed for remediation.

Common Challenges

While AI offers immense potential, the road ahead isn't easy by any means. Below are a few challenges facing many of my customers.

Legacy Data Estate

AI relies heavily on large, quality data sets. What's the primary can that customers consistently kick down the road? That's right, their data estate. Many customers struggle to access their data and when hey access they struggle with data quality and consistency. This is foundational challenge clocking access to large, diverse data sets for training purposes.

Model Building

There's building a model and building the right model. This requires expertise in selecting the appropriate algorithms, architectures, and hyperparameters for the job, balancing model complexity with interpretability.

Engineering Fundamentals

Most AI projects begin with experiments, and, in my experience, the experiments never really stop. It's easy to keep thinking of AI as something you play with in the lab but in order to achieve scale you have to apply engineering fundamentals at some point. These are the principles that allow a company to operationalize their AI efforts, creating a consistent and scalable AI Factory if you will.

Ethics and Confidentiality

It seems like every other week we hear about data breach. Just wait till those breaches start including not only data, but data that is the result of poorly trained, or unethically trained models. When these breaches start propagating hallucinations into the wild, we may see some strand outcomes.

Approaches to Address Challenges

There are however a few things customers can do to help overcome these challenges.

Selectively Modernize the Data Estate

You don’t want to try and tackle the data estate all at once. You'll never get anywhere. Trust me, I've seen too many try and fail. Instead start with a single use case and a single need. Modernize or even greenfield the program. See my post about Pathways and Horizons for ideas. The goal here is to get a win on the board and accelerate learning.

Collaborate Across Disciplines

It's a mistake to treat an AI effort as a purely data science effort. There is a lot that goes into these projects that you'll miss. I find its always best to treat these efforts as products. That means you need Product Owners to ensure you build the right thing and engineers to ensure you build it the right way. The engineering fundamentals like xOPS alone are worth the collaboration.

Continuous Learning and Adaptation

Your data and models will evolve over time. You'll learn new things, gather new data, and situations will change. Your ability to learn and adapt is heavily dependent on implementing sound mechanism for model retraining.

Conclusion

In summary, AI in MBSE holds immense promise, but successful integration requires a holistic approach, addressing technical, ethical, and practical aspects. By leveraging AI effectively, we can build better, more resilient systems that meet the complex demands of our interconnected world.

References:

1. Natural Language Processing for Systems Engineering: Automatic Generation of Systems Modelling Language Diagrams1

2. A First Step towards AI for MBSE: Generating a Part of SysML Models from Text Using AI2

Feel free to explore these references for deeper insights! 😊Title