Simon Dehn

The automation of product development is increasingly enabled by agents, which extend large language models (LLMs) from requirements engineering to computer-aided engineering (CAE). An agent can be understood as an LLM embedded in a software harness that provides tools, memory, execution loops, and feedback channels. Model-based systems engineering (MBSE) maintains the system architecture and may therefore provide an important source of context and state for such agents. This scoping review investigates which artifacts provide context and maintain state in the reported approaches, and whether the MBSE system architecture is among them. A systematic search of arXiv, IEEE Xplore, Scopus, and Web of Science on 1 August 2026 covered three search areas: MBSE and agents, MBSE and CAE, and agents and CAE. Of 14 584 records published from 2022 onward, title and abstract screening retained 495, of which 329, available in full text, were classified along the dimensions of their search area. Screening and classification were carried out by an LLM under author-defined criteria and closed vocabularies, reported along the PRISMA-trAIce checklist, and every evidence passage was verified against the full text. The review positions the MBSE system architecture within the agent harness and identifies how far the literature supports this role.

Future cyber-physical systems (CPSs), integrating subsystems of the mechanical, electrical and software domains, are becoming increasingly interconnected and complex. As complexity grows, testing effort increases as well. This includes the test-case definition step, where the test targets and boundary conditions are specified. With rising system complexity, the effort required to ensure that all relevant conditions for each test target are identified increases. Manual test-case definition remains the norm, creating effort bottlenecks in ensuring systematic coverage and compliance with standards such as ISO 26262 and ISO 29119. This paper explores how large language models (LLMs) can support the identification of complex boundary conditions for CPS test cases through detailed requirement analysis. The impact of performing taxonomy-guided, structured requirement mapping prior to test-case generation was evaluated by comparing it with a version without this guidance. Furthermore, the influence of supplying a Model-Based Systems Engineering (MBSE) system model as context information via Graph RAG is examined. The results show that structured, stepwise reasoning significantly improves reliability and consistency over unguided generation, while system-model information provides valuable contextual insight but has a minor impact in the chosen example. These findings outline a scalable framework for AI-assisted test-case generation.

The complexity of modern cyber–physical systems is steadily increasing as their functional scope expands and as regulations become more demanding. To cope with this complexity, organizations are adopting methodologies such as model-based systems engineering (MBSE). By creating system models, MBSE promises significant advantages such as improved traceability, consistency, and collaboration. On the other hand, the adoption of MBSE faces challenges in both the introduction and the operational use. In the introduction phase, challenges include high initial effort and steep learning curves. In the operational use phase, challenges arise from the difficulty of retrieving and reusing information stored in system models. Research on the support of MBSE through artificial intelligence (AI), especially generative AI, has so far focused mainly on easing the introduction phase, for example by using large language models (LLMs) to assist in creating system models. However, generative AI could also support the operational use phase by helping stakeholders access the information embedded in existing system models. This study introduces an LLM-based multi-agent system that applies a Graph Retrieval-Augmented Generation (GraphRAG) strategy to access and utilize information stored in MBSE system models. The system’s capabilities are demonstrated through a chatbot that answers questions about the underlying system model. This solution reduces the complexity and effort involved in retrieving system model information and improves accessibility for stakeholders who lack advanced knowledge in MBSE methodologies. The chatbot was evaluated using the architecture of a battery electric vehicle as a reference model and a set of 100 curated questions and answers. When tested across four large language models, the best-performing model achieved an accuracy of 93 percent in providing correct answers.

The complexity of modern cyber-physical systems is steadily increasing as their functional scope expands and as regulations become more demanding. To cope with this complexity, organizations are adopting methodologies such as Model-based Systems Engineering (MBSE). By creating system models MBSE promises significant advantages such as improved traceability, consistency, and collaboration. On the other hand, the adoption of MBSE faces challenges in both the introduction and the operational use. In the introduction phase, challenges include high initial effort and steep learning curves. In the operational use phase, challenges arise from the difficulty of retrieving and reusing information stored in system models. Research on the support of MBSE through Artificial Intelligence (AI), especially Generative AI, has so far focused mainly on easing the introduction phase, for example by using Large Language Models (LLM) to assist in creating system models. However, Generative AI could also support the operational use phase by helping stakeholders access the information embedded in existing system models. This study introduces an LLM-based multi-agent system that applies a Graph-Retrieval-Augmented-Generation (GraphRAG) strategy to access and utilize information stored in MBSE system models. The system’s capabilities are demonstrated through a chatbot that answers questions about the underlying system model. This solution reduces the complexity and effort involved in retrieving system model information and improves accessibility for stakeholders who lack advanced knowledge in MBSE methodologies. The chatbot was evaluated using the architecture of a battery electric vehicle as a reference model and a set of 100 curated questions and answers. When tested across four large language models, the best-performing model achieved an accuracy of 93 percent in providing correct answers.

Modern vehicles are complex cyber-physical systems with numerous customer functions and operating states, requiring careful management during development. Ensuring that all potential states and transitions are rigorously defined at the requirement stage is critical for vehicle safety and reliability. Due to stringent regulatory requirements and competitive market pressures, it is essential to articulate these requirements unambiguously, free of redundancy, and correctly. However, even a formally correct set of requirements can be underspecified, leading to unintended operating states that compromise vehicle functionality. To ensure requirement completeness, various approaches have been proposed to generate operating states and transitions allowed by a given set of requirements, ensuring that only intended behaviors are permitted. Existing methodologies often rely on natural language requirements, prone to ambiguities and errors. Furthermore, several methods formalize requirements using temporal logic languages, which can be checked algorithmically. Because of the complexity of modern vehicles, Model Based Systems Engineering (MBSE) has become state of the art in vehicle requirements engineering, as MBSE enables rigorous traceability of requirements and early model-based validation of requirement fulfillment. Currently a gap remains in integrating the comprehensive generation of operating states and transitions with the formalization in temporal logic languages, specifically for the use in MBSE development environments. This paper introduces a multi-step approach that formalizes requirements using Linear Temporal Logic (LTL) and generates operating states and transitions, which are then modeled in a SysML v2 state machine diagram, enhancing both the development process and the system’s reliability.

Successful products at least precisely meet the customers’ expectations and, in the best case, exceed them. To develop successful products, customer expectations must be translated into requirements. With the increasing functionalities of products in recent years, the customers’ expectations regarding product interaction and its behavior in different environmental conditions have also become more extensive. Current approaches of model-based systems engineering (MBSE) enable developing complex mechatronic products seamlessly from requirements to functions and solutions on a parameter level. However, there is a lack of approaches that systematically translate complex customer expectations into functional and design requirements as a starting point for further development.In this contribution we present a method and a corresponding meta-model that allows to systematically formalize the dependencies of different stakeholders and their expectations as well as different environmental conditions and constraints. From these dependencies, operating states are elicited that represent a set of simultaneously valid stakeholder expectations with their corresponding constraints. From these operating states, functional and design requirements are systematically derived as a basis for the model-based design of the system under development. Our meta-model is compatible to the established modeling language SysML, thus, existing approaches for the function-oriented model-based system development can benefit directly from these formally modeled requirements.Our publication signposts the potential for systematic and formal translation of customer expectations into operating states as well as requirements and thus enables a targeted, customer-centric and function-oriented development of mechatronic systems. We applied our method in an interdisciplinary, industrial project using the example of a thermal management system of a battery electric vehicle.

Managing an ever-growing number of requirements is a time consuming and highly complex activity in product development projects. With the rise of Artificial Intelligence (AI) in the last decades, algorithms are now capable of supporting Requirements Engineering (RE) tasks. The algorithm’s ability to intelligently process natural language data as well as the application in RE have been published extensively. In established RE processes however, it is often unclear, where AI algorithms can be utilized. The result is a significant imbalance between state-of-the-art AI algorithms and their application rates in real RE processes. One of the reasons is, that current models of RE processes aren’t able to convey the needed information for the identification of suitable tasks. Therefore, this paper aims to provide a framework of finite and standardized process steps with a data processing-oriented perspective, that can be used to systematically identify the points in the RE processes where AI algorithms can be applied. With the standardized process steps of this framework, established RE processes can be abstracted and expressed in a way that is compatible with the scope of AI algorithms. Due to the finite number of standardized process steps, libraries of suitable AI algorithms for the defined process steps can be built, so that their applicability can be efficiently assessed. The presented framework was developed collaboratively during a research project with industry participation.

In the course of an ever-increasing complexity of today's cyber-physical products, a shift of traditional systems engineering (SE) towards model-based system engineering (MBSE) is taking place. Many MBSE approaches use central system models to seamlessly link relevant development artifacts. The underlying methodologies are based on function-oriented procedure models of traditional SE. To support function-oriented development, various tools have been developed. These tools include so-called design catalogs, which collect known technical solutions to make domain specific knowledge broadly available. For conceptual development phases, especially the catalogs of Koller and Roth are commonly used. These catalogs provide a collection of fundamental physical effects for the fulfillment of elementary functions. A transfer of the analog, document-based effect catalogs to digital, model-based effect libraries has the potential to make product development more efficient. On one hand, finding the most suitable solution for a given system function could be efficiently supported via algorithms. On the other hand, an easy integration as well of reusable modeled effects as well as an early fundamental validation of possible solution concepts could be achieved. The digitalization of individual physical effects and their integration into model-based solution concepts has already been demonstrated. However, it has not yet been possible to create digital effect libraries that enable the seamless reuse of digital physical effect models. This paper therefore aims to identify existing deficits within the formalization of physical effects. To achieve this, requirements for digital physical effect libraries are determined. Based on these requirements, existing deficits are derived by comparing the requirements with the analog effect catalogs and existing MBSE modeling methodologies. Finally, recommendations for action and possible approaches to overcome the identified deficits are presented.