AI-Integrated Framework for Enhancing High Level Architecture

Design Across System Lifecycle Stages

Tianxiao Xu

1,2 a

, Néjib Moalla

, Mohand Lounès Bentaha

, Hazal Aktekin

and Claudia Agostinelli

Université Lumière Lyon 2, INSA Lyon, Universite Claude Bernard Lyon 1, Université Jean Monnet Saint-Etienne,

DISP UR4570, 69007 Lyon, France

IVECO GROUP, 69200 Vénissieux, France

IVECO GROUP, 10156 Turino, Italy

{tianxiao.xu, nejib.moalla, mohand.bentaha}@univ-lyon2.fr,

Keywords: Artificial Intelligence (AI), Model-Based Systems Engineering (MBSE), Multidisciplinary Design Analysis

and Optimization (MDAO), Digital Twin (DT), Digital Continuity.

Abstract: AI technology is increasingly being introduced into the automotive industry to support the product design

process and address the challenges arising from growing product complexity. Systems Engineering is an

interdisciplinary approach and methodology aimed at designing, developing, and managing complex systems

throughout entire system lifecycle. The development of Model-Based Systems Engineering (MBSE)

significantly enhances complexity management and requirement traceability in conceptual design phases. In

the design and analysis phases, the use of Multidisciplinary Design Analysis and Optimization (MDAO)

effectively addresses challenges in multidisciplinary problems, identifies optimal solutions, and supports

decision-making. Digital Twin (DT) technology is extensively studied and applied to monitor, analyse, and

predict operational system behaviour. Integrating AI into system design, along with its combination with

MBSE, MDAO, and DT technologies, not only addresses design challenges but also creates new opportunities

to advance systems engineering. This paper focuses on how high-level architecture design supports different

stages of system lifecycle. The study explores the roles AI can play in the process, as well as its integration

with related technologies, and proposes an AI-integrated framework to ensure digital continuity throughout

system lifecycle stages.

1 INTRODUCTION

With the increase in automotive functionalities and

components, systems have become more complex,

and advanced features such as autonomous driving

systems and intelligent connectivity further

complicate development challenges. To address these

challenges, modern technologies such as Artificial

Intelligence (AI), Model-Based Systems Engineering

(MBSE), Multidisciplinary Design Analysis and

Optimization (MDAO), and Digital Twin (DT) have

been introduced. Their integration enhances

efficiency and optimizes decision-making across all

lifecycle stages, from design to operation.

https://orcid.org/0009-0009-4211-2960

https://orcid.org/0000-0003-4806-0320

https://orcid.org/0000-0001-6564-3435

Systems Engineering (SE) is a transdisciplinary

and integrative approach to enable the successful

realization, use, and retirement of engineered systems,

using systems principles and concepts, and scientific,

technological, and management methods (INCOSE,

2023). Starting with requirements analysis, it covers

conceptual design, system integration, verification

and validation, and eventually operation and disposal.

The primary objective is to manage complexity,

ensuring that all functionalities and requirements are

aligned, ultimately delivering a product that meets

stakeholders’ expectations.

MBSE is a model-centric approach to performing

systems engineering (Douglass et al., 2022). MBSE

Xu, T., Moalla, N., Bentaha, M. L., Aktekin, H. and Agostinelli, C.

AI-Integrated Framework for Enhancing High Level Architecture Design Across System Lifecycle Stages.

DOI: 10.5220/0013444200003896

Paper published under CC license (CC BY-NC-ND 4.0)

In Proceedings of the 13th International Conference on Model-Based Software and Systems Engineering (MODELSWARD 2025), pages 407-419

ISBN: 978-989-758-729-0; ISSN: 2184-4348

407

helps organize requirements and rapidly create and

evaluate various architectural solutions. For instance,

MBSE is applied in autonomous vehicles to ensure

the integration of functionalities across software,

hardware, and communication systems (Raza et al.,

2024), drive co-simulation of different disciplinary

analysis models (Zhao et al., 2024), and optimize the

thermal management system of vehicles by analyzing

energy flow and heat dissipation efficiency in

advance (Habermehl et al., 2022).

MDAO is a methodology used during the design

phase to foster collaboration across various

disciplines such as aerodynamics, structural

mechanics and cost optimization, and identify

optimal system designs (Simpson et al., 2011).

MDAO addresses challenges in managing

interactions among disciplines while ensuring

optimized results across different design objectives.

(Moerland et al., 2020) applied MDAO in the next

generation of aircraft to provide significant

reductions in aircraft development costs and time to

market. Through MDAO analysis, the design

optimization of fuel consumption, take-off weight,

and rotor dynamics parameters was completed,

achieving an integrated process of system design and

optimization (Qi et al., 2024).

DT refers to a virtual representation of a physical

entity that is interconnected with its real-world

counterpart through continuous and bidirectional data

exchange in real time (Singh et al., 2021). DT can

monitor vehicle conditions in real time and predict

potential issues, and can optimize logistics vehicle

routes and maintenance plans in fleet management

(Alexandru et al., 2022). DT is also is beneficial to

enhance the traditional product design and

development processes (Tao et al., 2018).

AI simulates human intelligence through

technologies such as Machine Learning (ML), Deep

Learning (DL), and Natural Language Processing

(NLP) (Aurélien, 2019). Its primary methods include

supervised learning, unsupervised learning, and

Reinforcement Learning (RL). AI enhances system

design efficiency, such as through automated

requirement analysis and model generation (Zhao et

al., 2021), and supports design decision-making via

optimization algorithms (Mirjalili et al., 2020).

This concept of AI4MBSE refers to leveraging AI

to enhance MBSE. AI4MBSE focuses on improving

tasks like requirements engineering, model

generation, and decision-making by using ML and

NLP. Examples include automated traceability of

requirements and intelligent model recommendations

to streamline systems engineering workflows. AI

helps handle complexity and reduce errors in MBSE

processes, especially in domains like transportation

and aerospace (Li et al., 2022). Automates repetitive

and error-prone MBSE tasks, such as requirements

extraction and traceability (Ghanawi et al., 2024).

Enhances decision-making through AI-driven model

analysis (Raz et al., 2021). Improves MBSE modeling

capabilities based on domain knowledge (Zhang et al.,

2024).

This concept of MBSE4AI applies MBSE

principles to design and develop AI systems

themselves. It ensures that AI solutions are integrated

systematically, considering requirements, constraints,

and validation across the system lifecycle. MBSE4AI

emphasizes structuring AI system designs to ensure

safety, reliability, and interoperability in complex

environments (Anton et al., 2023). Systematically

integrates AI solutions into larger systems while

managing risks and ensuring transparency, and

enables continious validation and verification of AI

components against system requirements (Torkjazi

and Raz, 2024).

AI is used to enhance the efficiency and accuracy

of MDAO workflows. Accelerating optimization

algorithms by using AI models to reduce computation

time. Assisting in exploring diverse design spaces

through AI-guided sampling and evaluation methods

(Karali et al., 2024). This approach addresses

challenges like high computational costs and the

difficulty of identifying optimal solutions in complex

systems.

AI supports DT by improving real-time data

analysis, anomaly detection, and predictive

maintenance. It enhances simulation accuracy by

using AI models to fill gaps in physical modeling or

to simulate scenarios that are computationally

intensive with traditional methods (Rasheed et al.,

2020). In operation, AI enables better system

monitoring and decision-making, such as in fields

like manufacturing, healthcare, and transportation.

This research aims to explore the roles of AI in

early stages of the system lifecycle based on their

diverse capabilities. It focuses on leveraging AI to

enhance MBSE capabilities, such as in requirements

modeling and architecture modeling. Additionally, it

seeks to improve the efficiency of MDAO processes

using AI while accelerating design space exploration.

Furthermore, the study investigates the application of

AI to enhance the accuracy of DT models, as well as

their updating and generation capabilities. Ultimately,

this work aims to support digital continuity

throughout the system lifecycle using AI technologies.

This research investigates the integration of AI

into system design processes, emphasizing its role in

enhancing system design and aligning with specific

MBSE-AI Integration 2025 - 2nd Workshop on Model-based System Engineering and Artiﬁcial Intelligence

408

capabilities needs. By integrating AI into MBSE the

study aims to improve modeling efficiency, maintain

model consistency, and provide heuristic alternatives

for decision-making. In the context of MDAO, AI is

leveraged to enhance computational efficiency,

reduce computational burdens, and enable

multifidelity analysis for more effective design

optimization. For DT applications, AI contributes to

the generation of predictive models, enabling

accurate system monitoring, forecasting, and

optimization. Finally, an AI-integrated framework is

proposed to facilitate interaction across lifecycle

stages, ensuring cohesive system development and

lifecycle management.

This research addresses several problems

regarding the integration of AI into system design

processes. First, it should explore the current

capabilities of AI and its developmental trends,

investigating the roles AI can play in system design

and the improvements it can bring to the process. In

the field of MBSE, the study should examine pressing

challenges that AI could address, identifying gaps in

its current applications and unresolved issues. For

MDAO, it investigates existing directions of AI

application and explores potential areas for further

enhancement. Within the DT technology, the research

should explore into AI’s current applications and

evaluates how AI can introduce new capabilities and

improve performance. Finally, it needs to consider the

interrelationships between different stages of the

system lifecycle, analyzing how AI can enhance

iteration across these stages to ensure a more efficient

and integrated system design process.

This paper focuses on how AI technology can

support system design processes by enhancing

various aspects of the system lifecycle. These include

supporting MBSE models during the conceptual

design phase, MDAO models during the design and

analysis phase, and DT models during the operational

phase. By leveraging AI to ensure digital continuity

across design activities, this research aims to improve

the efficiency and quality of product development.

The integration of AI with MBSE, MDAO, and DT

technologies has been widely studied and applied

across various fields. While these applications share

the use of AI, the field of AI encompasses a diverse

array of methods and models, resulting in diverse

solutions for each research focus or application case.

This paper investigates how AI can support a cross-

lifecycle design framework for system development,

exploring the roles AI can play within this framework

and the potential outcomes that can be achieved.

However, the application of MBSE to AI systems for

specific design problems and case studies is beyond

the scope of this paper.

This paper discusses the roles that AI technology

can play in system design processes, highlighting the

improvements it can bring to areas such as MBSE,

MDAO, and DT applications. Based on the current

state of AI technology, it further elaborates on how

high-dimensional system architectures can be applied

effectively across different stages of the system

lifecycle.

This paper defines the application of AI in system

design and explores how AI can be applied to

different aspects of the system lifecycle. It addresses

AI's roles in various stages, such as the MBSE model

in the conceptual phase, the MDAO model in the

design phase, and the DT model in the operational

phase. The paper provides a definition of AI's role in

these stages and presents an extendable AI supported

framework that supports the interaction and

integration of these three domains across the lifecycle.

This framework ensures digital continuity in system

design solutions throughout different stages of the

lifecycle.

2 STATE OF THE ART

2.1 Artificial Intelligence

AI technology is increasingly being introduced into

the automotive industry to support the product design

process and address the challenges posed by growing

product complexity. Machine Learning (ML) is a

subset of AI, where machines learn from data to

perform pattern recognition, prediction, and decision-

making without explicit programming. ML provides

solutions for many complex fields and drives the

application of AI.

One of the most popular ML algorithms today is

Deep Learning (DL), where the ML model consists of

an Artificial Neural Network (ANN) (Ian et al., 2017).

These ANNs are widely used for tasks such as image

recognition, natural language processing, and

predictive analytics.

Reinforcement Learning (RL) is a semi-

supervised learning model in which an agent

continuously makes decisions and adjusts its actions

through trial and error based on the environment's

responses. The agent is the core component of the RL

model, and it determines which actions to take based

on a policy function (Sutton and Barto, 2018). Most

RL algorithms can use an ANN, a method known as

Deep Reinforcement Learning (DRL).

AI-Integrated Framework for Enhancing High Level Architecture Design Across System Lifecycle Stages

409

The goal of Transfer Learning (TL) is to leverage

the knowledge of a pre-trained model to solve new

but similar problems. This method can adapt the

retrained model to a new problem by adding layers to

the ANN (Wang and Chen, 2023). This approach

significantly reduces the training time. TL is also a

highly effective technique when there is limited data.

Progressive Learning (PL) is a form of TL. ML

models in TL are gradually retrained to solve

increasingly complex tasks. PL can significantly

reduce training time and, consequently, decrease the

required computational resources (Fayek et al., 2020).

The concept of advisor agents is a relatively new

field, and new methodologies are still being explored.

Advisor agents are trained agents that can support the

training of a new primary agent. While agents and

multi-agent systems have a long history as a major

approach in distributed AI, the advisor agent

framework represents a more recent advancement,

offering a highly flexible architecture. The number of

advisor agents and their interactions with the primary

agent can be determined by the engineer (Zhang et al.

2021).

Surrogate models are widely used in optimization

and engineering design, aiming to accelerate the

computational process by replacing high

computational cost models with simpler or less

computationally expensive ones (Hao et al., 2022).

Surrogate models can also be created using

supervised ML algorithms. These ML models attempt

to identify certain trends in large amounts of training

data. When the trends generated by the underlying

high-fidelity simulations are too complex, ML

algorithms based on ANN can be selected.

2.2 AI and MBSE

International Council on Systems Engineering

(INCOSE) has initiated two key initiatives,

AI4MBSE and MBSE4AI, that explore the

integration of AI with MBSE. AI4MBSE focuses on

leveraging AI technologies to enhance MBSE

processes, improving aspects such as system

modelling, requirements analysis, and design

optimization. This initiative aims to make MBSE

more efficient and adaptive through the application of

AI. On the other hand, MBSE4AI examines how

MBSE methodologies can be used to support and

improve AI systems, especially in terms of model

validation, integration, and lifecycle management.

Together, these initiatives aim to bridge the gap

between AI and MBSE, creating a synergy that can

optimize system engineering processes and enable

smarter, more robust system designs.

2.2.1 MBSE4AI

Future system operations increasingly require the

integration and interoperability of multiple intelligent

systems driven by AI, which has become a core

element of the system, spanning the entire system

lifecycle. (Raz et al., 2021) discussed the challenges

faced by AI-driven aerospace systems in systems

engineering activities. To develop and implement AI

to meet the conceptual design needs of aerospace

systems, AI will be designed as one of the primary

functional elements. To train and develop AI models,

the system design process may undergo

corresponding modifications, such as the AI pipeline

(Blasch and Pokines, 2019). In this process, they

propose Systems Engineering as Data Curator for AI

to address challenges such as the availability of data,

the type of data, and the role of SE. Data architecture

is added to the MBSE model to establish relationships

with operational concepts, functional architecture,

physical architecture, and so on, to ensure that MBSE

supports the overall R&D process of integrating AI

systems.

AI systems also have varying degrees of

intelligence. (Torkjazi and Raz, 2024) based on the

steps of the Object-Oriented Systems Engineering

Method (OOSEM), utilized the Unified Architecture

Framework (UAF) to model autonomy integration.

They modelled autonomous systems through SE

technical processes. To reflect the differences in

autonomy between systems, they added accuracy

Technical Performance Measures (TPMs) to the

AI/ML components to assess different system

solutions.

A MBSE-Enhanced Long Short-Term Memory

(LSTM) Framework has been provided for Satellite

System Reliability and Failure Prediction, which

offers a case study on how MBSE can support the

design of AI-integrated systems (Alandihallaj et al.,

2024). The framework describes the predictive

system architecture using LSTM networks, an AI

technique, through the MBSE model. The integration

of AI with MBSE, as demonstrated in this study,

shows significant potential in enhancing the

reliability and longevity of satellite systems.

2.2.2 AI4MBSE

MBSE can be applied to support AI-integrated

systems, and conversely, AI can also enhance and

support the MBSE processes and activities. (Chami et

al., 2022) focus on the challenges faced by MBSE at

different stages of its application and analyse the

capabilities AI can offer and how these can be

allocated to address the corresponding MBSE

MBSE-AI Integration 2025 - 2nd Workshop on Model-based System Engineering and Artiﬁcial Intelligence

410

challenges. The paper provides an outlook on how AI

can support MBSE, relevant research has already

been conducted in different fields for specific

research questions.

The first capability of AI applied to MBSE is

ensuring data extraction for MBSE. MBSE models

can establish traceability management between

requirements and system design solutions. With the

application of NLP-based models, research and

applications for converting natural language into

machine-readable language are becoming

increasingly diverse.

Standard documents, as an important source of

requirements in various fields, serve as one of the key

references for requirement engineers in requirement

extraction. (Ghanawi et al., 2024) contribute to the

integration of AI with MBSE, focusing solely on the

extraction and transformation of medical standards

information from documents into SysML norm

models. They employed an open-source multimodal

classifier model and a proprietary Large Language

Model (LLM) to achieve this goal. Although the title

retrieval performed well in terms of recall across

different approaches, the precision was generally low.

Future improvements to the proprietary AI model are

needed to achieve better results, but this may increase

the training and subsequent usage costs.

(Chen et al., 2022) proposed an NLP-based

framework for information extraction under the

general condition that can automatically detect the

actors and their responsible actions. To validate the

performance of the developed model, the study

compared the NLP-generated report with the

manually created SysML model. The results showed

that the precision and recall rates for extracting roles

and responsibilities were 0.86 and 0.66, respectively,

indicating that this text-to-model framework has the

potential to accurately convert general policy

documents into SysML.

(Chami et al., 2019) also utilized NLP techniques

to train a Named Entity Recognition (NER) model.

However, their objective extended beyond just

extracting requirements, stakeholder roles, and

responsibilities. Their scope included identifying

system actors, use cases, associations, and blocks,

aiming to use AI technologies to generate parts of

SysML models from text. This approach allows the

model to be trained through label annotation,

enabling semi-structured text to be converted into

SysML model entities with minimal training effort.

In addition to research on the automated

supplementation and transformation of requirement

models in MBSE using AI, another emerging trend is

leveraging AI technology to enable the automatic

generation of system design solutions or provide

design references. Prior to the application of AI

technology, rule-based generative design methods

and tools already existed in various disciplines. These

design tools often address specific disciplinary

problems, enabling rapid design space exploration

and generating many solutions that meet specified

requirements.

With the application of AI and ML technologies,

data-driven generative techniques have also advanced.

(Zhang et al. 2021) proposed an MBSE modeling

process recommendation method based on domain

knowledge and SysML models by using a Global

Vectors for Word Representation (GLOVE) model

pre-trained on both domain-specific and general

knowledge, combined with the concept of a

recommendation system. This method not only

considers the influence of general and domain-

specific knowledge on the modeling process but also

utilizes SysML models as training data to provide

recommendations for subsequent modeling. These

recommendations are generated based on textual

training and proposed as suggested solutions derived

from the knowledge. The information structure that a

SysML model can encompass is closely related to the

data structure of the knowledge base.

For MBSE models, in addition to the logical

architecture used to describe the composition of the

system, the system architecture also includes other

types of information such as system requirements and

interfaces. To cover all the information in a SysML

model, it is necessary to continuously refine the data

structure of the knowledge base, which will also

impose higher demands on the methods and costs of

data training.

The emergence of LLM, such as OpenAI's GPT

series, has brought significant opportunities for

transformation across multiple domains, driving

industry professionals to explore their potential

applications. (Johns et al., 2024) integrated OpenAI's

GPT-4 Turbo with CATIA Magic for MBSE, creating

the AI Systems Modeling Enhancer (AI-SME) to

generate MBSE models. Compared to the time

required for human modeling, AI-SME offers

significant advantages. The results demonstrate that

the requirements, structural, and interface definitions

created by AI-SME maintain coherence and

consistency, but they are not complete. AI-SME can

serve as a modeling assistant to automate primary

tasks and improve the efficiency of prototype

architectural development.

AI algorithms can also solve specific design

problems. (Rudolph, 2024) introduced AI methods

using three artificial intelligence search algorithms

AI-Integrated Framework for Enhancing High Level Architecture Design Across System Lifecycle Stages

411

(PMSO, SA, and A*) to automate packing, piping,

and routing for arbitrarily complex 3D CAD

geometries. Packing, piping, and routing problems

are considered NP-complete problems. Models in

these specific disciplines often have higher

confidence but also require greater computational

power. In the field of MBSE, multidisciplinary

problems are often encountered, and solving these

problems typically requires integration, verification,

and validation. Solving NP problems in individual

disciplines is a prerequisite for ensuring the efficiency

of integration and verification.

2.3 AI and MDAO

Multidisciplinary Design Analysis and Optimization

(MDAO) is a methodology used for designing

complex engineering systems. During this process,

there is a need to integrate multidisciplinary analysis

models, which include many design variables,

objective functions, and state variables arising from

coupling relationships. These models exhibit high

complexity in both structure and processes. (Karali et

al., 2024) proposed an AI-driven multidisciplinary

conceptual design framework for UAVs to ensure the

MDAO process. They employed AI-based surrogate

models, using Latin Hypercube Sampling (LHS) to

optimize the design space exploration process, NN

(neural networks) for black-box model training, and

genetic algorithms to search for optimal solutions

along the Pareto front. By introducing and

implementing this intelligent conceptual design

algorithm, numerical data can be incorporated into

the early stages of the design process, significantly

reducing reliance on manual intervention.

In addition to optimizing the MDAO process

through AI, AI technologies are also applied to

address specific challenges arising from

multidisciplinary interactions, improving the

efficiency of MDAO analysis while ensuring the

accuracy of results. For example, (Wang et al., 2023)

leveraged AI technologies to tackle challenges

associated with Uncertainty-based Multidisciplinary

Design Optimization (UMDO). In this process, data

is first processed using unsupervised learning for

clustering analysis and dimensionality reduction.

Subsequently, supervised learning is employed to

train disciplinary models using pre-prepared datasets

to reduce computational costs. Finally, evolutionary

algorithms such as genetic algorithms are used to

search for optimal solutions.

In addition to optimizing the MDAO process and

improving solution efficiency, the integration of AI

technologies also provides new methods for surrogate

model generation. (Sisk et al., 2023) utilized

generative adversarial networks (GANs) to establish

a training framework for model generation,

addressing the challenges of urban air mobility

(UAM). This training framework is like the XDSM

structure commonly used in MDAO. By training

flight trajectories through a twin generator and

combining the generated model data with Deep

Neural Network (DNN) methods, surrogate models

are trained. These generated surrogate models can

then be re-applied in the MDAO process.

2.4 AI and DT

DT refers to the virtual representation of a physical

system that is connected in real-time to enable

monitoring, prediction, and optimization. The

development of AI technologies has significantly

contributed to the application of DT by providing

data-driven solutions. The process of constructing a

DT is highly complex, and it varies depending on the

application purpose and the selected technologies.

(Orlova, 2022) conducted a comprehensive analysis

of the design methods for DT of organizational and

technical systems, defining the various stages of DT

development as well as the relevant technologies that

can be applied during the Design and Engineering

phase. However, for complex organizational and

technical systems operating under conditions of

uncertainty, there is currently no comprehensive and

universal methodological approach to address these

challenges.

Although current trends suggest that DT will be

entirely controlled by AI, as with the integration of AI

and MBSE, DT and AI influence each other mutually.

(Bariah et al., 2024) explored the interrelationship

between AI and DT in practical applications. On one

hand, TL in AI can be used to address network

updates in distributed DT, thereby reducing the

training time required for model updates. On the other

hand, leveraging communication technologies, DT

can provide experience-driven learning methods and

integrate model-based learning approaches, such as

DT-enabled RL methods, to enhance reasoning

capabilities in AI algorithms. This approach achieves

a complementary combination of data-driven and

model-driven methods, leveraging the advantages of

both.

(Groshev et al., 2021) applied DT to Cyber-

Physical Systems (CPS) by defining AI agents and

provided the allocation relationships between

applications, AI technologies, and physical devices.

The AI agents were used for functional and

infrastructure applications, with detailed definitions

MBSE-AI Integration 2025 - 2nd Workshop on Model-based System Engineering and Artiﬁcial Intelligence

412

of their usable input data and expected output data.

The research results demonstrated that DT

applications employing AI agents can effectively

predict system dynamics. AI agents offer significant

opportunities to enhance the reliability, robustness,

and performance of DT.

As an application model during the system

operation phase, the DT model has stricter

requirements for computation time and, like

simulation analysis models, faces NP problems.

(Karali et al., 2023) developed a data generation

algorithm, which includes high-fidelity models based

on Computational Fluid Dynamics (CFD) methods

and low-fidelity models based on computational

aerodynamics methods. Then, using multi-fidelity

data, they generated a new surrogate model based on

TL. Thanks to this approach, the developed AI model

can use data from lower-fidelity models to more

accurately predict missing flow conditions in the

high-fidelity data.

2.5 AI and Digital Continuity

Digital Continuity refers to the ability to ensure that

all digital information remains consistent, accessible,

understandable, and usable throughout the entire

product lifecycle or business process (Ren et al.,

2020). We mentioned the support of AI for MBSE,

MDAO, and DT, which are situated at different stages

of the system lifecycle. Various fields have also

analysed the advantages and challenges brought by

the integration of these processes. The integration of

design activities at different stages of the system

lifecycle has also introduced the issue of digital

continuity.

2.5.1 Interaction Across Lifecycle Stages

MBSE is an effective approach for demonstrating

multidisciplinary coupling relationships needed to

meet specific requirements. By integrating MBSE

with MDAO, it is possible to improve system

architecting, streamline the development of agile

MDAO design systems, to trace analysis results back

to the corresponding requirements, revealing implicit

relationships between different requirements that

arise from the solution domain. (Fouda et al., 2024)

proposed a novel MBSE-driven MDAO process

design and implementation method to address the

wing shape optimization problem. By establishing a

metamodel of the MDAO process within the MBSE

model environment, data consistency is ensured. The

MDAO process is described using MBSE, which in

turn drives multidisciplinary simulations. The

integration of MBSE and MDAO provides significant

flexibility in adapting to changing requirements,

while also improving the traceability of design

decisions throughout the product development

lifecycle.

(Wu et al., 2022) proposed a new

multidisciplinary collaborative design method

supported by DT. To describe complex products in a

virtual environment, they further developed a

systematized multidisciplinary collaborative design

framework based on DT, integrating

multidisciplinary collaboration into three stages:

conceptual design, detailed design, and virtual

validation. By utilizing DT for parallel design across

different disciplines within the virtual environment,

this approach can reduce anomalies caused by

multidisciplinary integration. Although this

multidisciplinary integrated design method does not

directly employ AI technologies, it reveals that the

same disciplinary problems can be described using

both model-driven and data-driven methods, such as

DT models. This provides the potential for iterative

optimization between detailed design models and DT

models, as well as a foundation for integrating AI

technologies.

The integration of MBSE and DTs has also seen

significant development in various fields. For

example, (Lopez and Akundi, 2022) have explored

the use of MBSE in the development process of DT.

(Bordeleau et al., 2020) research demonstrates that

MBSE can manage heterogeneous models from

different disciplines. MBSE models include the

relationships between system components, enabling

the driving of other design processes and providing a

foundation for multidisciplinary simulation.

However, challenges exist regarding the accuracy and

sources of analysis models driven by MBSE, which

DT can address. (Purohit and Madni., 2022)

developed a DT prototype and obtained experimental

results, collecting data from physical systems in the

real world to update the DT model, enhancing

operational analysis and modelling. DT models are

also seen as an important application in supporting

V&V processes. By incorporating DT into MBSE

models, significantly reduced the time required for

early-stage V&V (Bouhali et al., 2024). (Madni et al.,

2019) have also considered DT technology as an

integral part of MBSE methodology and

experimentation testbeds.

2.5.2 Ai Integration

According to current research, systematic methods to

ensure digital continuity throughout the system's

AI-Integrated Framework for Enhancing High Level Architecture Design Across System Lifecycle Stages

413

lifecycle are still under exploration. The introduction

of AI technologies has also brought new capabilities

to address this issue.

(Erikstad) explored whether LLM, particularly

multi-agent LLM, can be combined with MBSE

principles to address the high development costs and

potential errors of optimization and simulation

models. The system to be optimized is captured as

classes and instances to serve as the syntax for the

narrative-to-model mapping, while the MBSE models

and views become the grammar. In the article,

ChatGPT is used to directly create different AI

agents, each of which can be considered as directly

replacing the corresponding designer roles. The

deliverables produced by these agents interact based

on business processes and roles, thus achieving a

complete design process. According to the research

findings, the team of agents can complete the work

and collaboratively develop solutions through

autonomous coordination of processes. However,

further in-depth research is needed to determine to

what extent the capabilities and specialization of each

agent should be enhanced.

(Scott et al., 2016) proposed another use of agents,

where a typed database is used as a knowledge

representation to create agents that utilize AI

techniques. These agents can check the quality of

information and provide feedback, integrated within

the tools of the lifecycle chain. This improves process

quality and helps with design activities.

2.6 Discussion

This paper aims at supporting activities across

different lifecycle stages of a system engineering

using AI technology, such as the digital continuity

between MBSE, MDAO, and DT. The application of

AI across various topics effectively demonstrates the

fundamental capabilities of AI and the challenges

faced within each topic, while also providing a

foundation for the digital continuity of activities at

different lifecycle stages. By integrating AI to ensure

digital continuity, it further advances the role of

MBSE throughout the system's lifecycle.

3 PROPOSED FRAMEWORKS

3.1 Roles of AI

For each system life cycle process, an Input-Process-

Output (IPO) diagram can illustrate the typical inputs,

process activities, and typical outputs, as shown in

Figure 1. Additionally, each activity includes controls

and enablers. We aim to use this format to classify

and define the capabilities AI can provide during the

system design process, allowing us to better position

AI within the entire life cycle.

Figure 1: Roles of AI in IPO diagram.

The AI-Connector functions at the input and

output ends of each process activity, ensuring data

continuity and simplifying data extraction processes

through AI algorithms, such as natural language

extraction and output. This category includes two

roles: importer and exporter. These two roles may

overlap from the perspective of different processes, as

the exporter of the previous process can also serve as

the importer for the next process.

 Importer: Serves as the input for data, ensuring

that the current activity has all the necessary

inputs to initiate.

 Exporter: Serves as the output for data,

ensuring the continuity and consistency of

output data.

The AI-Assistant operates within each design

process and aims to assist staff in design activities

using AI technology, enhancing work efficiency and

quality. This category includes two roles: generator

and accelerator.

 Generator: Directly generates part of the design

solution by providing multiple alternative or

optimal solutions based on given inputs and the

current context.

 Accelerator: Improves the efficiency and

quality of solution generation.

The AI-Controller is used to inspect and optimize

design activities to ensure the quality of the design

process. It can extract existing design standards to

verify design content and optimize results or provide

optimization suggestions. This optimization can

target individual design activities or support the

MBSE-AI Integration 2025 - 2nd Workshop on Model-based System Engineering and Artiﬁcial Intelligence

414

optimization of multiple design activities. This

category includes two roles: checker and optimizer.

 Checker: Inspects the content of design

activities.

 Optimizer: Optimizes single or multiple design

activities.

The AI-Enabler incorporates one or more methods

to ensure the execution of design tasks, providing

references for the design process. For AI-Enabler, this

can be implemented through advisor agents.

3.2 AI in MBSE

In the MBSE process, we can utilize NLP models to

establish the importer and exporter roles in the

requirements management process. NLP models can

extract relevant requirements from stakeholder needs

and narratives, transforming them into requirements

models. In this process, NLP or LLM can also serve

as generators, assisting in the creation of

requirements models. By training on standards and

norms related to requirements definition, a checker

can be established to automatically verify the syntax,

semantic, and conflicts within the model.

Table 1: AI roles in MBSE.

Roles Capabilities

Potential

metho

importe

Extract form needs

NLP, ML,

exporte

Transform requirements into

MBSE model

NLP, KE

MBSE

orte

Transform from MBSE to

MDAO

ML, NLP

Generator

Generate requirement model

from text

NLP, DL,

Generate behaviour from

narrative

NLP, ML

Generate model structure RL, KE

Advisor

Provide references and

alternatives

Checker Syntax and semantic review NLP

After obtaining itemized requirements, ML

models or NLP models can be used to automatically

transform requirements into SysML use cases.

Furthermore, using Knowledge Engineering (KE)

and RL methods, existing requirements and solution

data can be trained to automatically construct system

architectures from requirements, functioning as a

generator. During this process, checkers implemented

using NLP methods for SysML syntax and grammar

remain applicable.

Once the tasks within MBSE are completed, ML

can be employed to transform specific MBSE

information, such as the MDAO structure described

by a parametric diagram, into the corresponding

downstream design environment. To enable the

smooth progression of the design process, an advisor

can be implemented by adding LLM-based agents.

Advisors share similar capabilities with generators,

but specialized LLM training is more challenging,

albeit more powerful. If the models generated by

specialized LLM achieve a certain level of

certification accuracy, these agents can also be

applied as generators.

3.3 AI in MDAO

Table 2: AI roles in MDAO.

Roles Capabilities

Potential

metho

Importer

Convert DSM ML

Exporter

Transfer results ML

Transform SM ML, RL

Accelera-

Enhance optimizer algorithm RL, DL

Generator Generate SM for analyzer

ML, RL,

Advisor

Provide references and

solutions

In the MDAO process, the importer can transform

input content into the format required by the MDAO

workflow, such as converting the Design Structure

Matrix (DSM) defined in the MBSE model into a data

structure usable within the optimization environment,

while ensuring information completeness. During the

simulation process, the accelerator enhances

efficiency by replacing traditional optimization

algorithms with RL, thereby speeding up analysis and

optimization. The generator can leverage model order

reduction methods, also referred to as a trainer, to

train surrogate models using ML, reducing the

computational cost of discipline-specific analysis

models. At the end of the MDAO process, the

exporter can transfer the resulting data back to the

MBSE environment to verify whether the

requirements have been met. It can also transform the

generated surrogate models into the DT environment

to support system operation. Similarly, the MDAO

process can incorporate agents as advisors to provide

recommendations for solving multidisciplinary

problems.

AI-Integrated Framework for Enhancing High Level Architecture Design Across System Lifecycle Stages

415

3.4 AI in DT

In the context of DT, DT models are planned and

developed during the design phase. As shown in table

3, they can also be formed during the operational

phase in a data-driven manner to address emerging

issues. An importer functions as a tool to transform

models defined during the design phase into

operational formats within the DT space. A generator

in this process can also act as a model trainer,

continuously iterating and updating the DT model

using operational data through methods such as RL

or TL. High-fidelity DT models can replace

traditional discipline-specific analysis models and

participate in MDAO simulation and analysis

processes.

Table 3: AI roles in DT.

Roles Capabilities

Potential

metho

Importer Transform models in DT

environment

Exporter

Output

DT models

Generator Train and upgrade DT RL, TL

3.5 AI in Digital Continuity

This study aims to leverage AI technologies to

support digital continuity, facilitating the integration

of MBSE, MDAO, and DT. Beyond the roles that AI

can play in enhancing process efficiency and quality

within these workflows, this framework also

emphasizes the contribution of AI in ensuring digital

continuity throughout the system's lifecycle, the AI

integrated framework is defined as shown in Figure 2.

From a methodological perspective, MBSE

effectively describes the interactions between various

components of the system, providing essential inputs

for MDAO simulation structures. By ensuring digital

continuity between MBSE and MDAO through AI,

the establishment of MDAO simulation workflows

can be made more efficient. During the MDAO

process, AI can be used to generate lightweight, high-

fidelity surrogate models, reducing the computational

cost associated with integrated simulations.

Furthermore, AI algorithms can accelerate the

exploration of the design space, enabling the

identification of optimal solutions in a shorter

timeframe.

In the integration of MDAO and DTs, AI plays a

critical role. Some DT applications during the

operational phase are planned and developed during

the design phase. By utilizing AI training techniques,

surrogate models generated through MDAO can be

applied to certain DT applications. Through ML

methods such as TL, DT models can be continuously

updated and optimized using the vast amounts of data

generated during operation.

Models that represent the same object, such as

surrogate models generated from simulation data and

those developed from operational data, can contribute

to beneficial iterations in system design. By

leveraging operational data validated during runtime

to refine the analyzers in MDAO simulations, the

confidence in the models can be enhanced while

maintaining computational efficiency. This also

provides a new pathway for the reuse of historical

data and digital artifacts to meet new business

objectives during the system design process.

Figure 2: AI-Integrated Framework to support Digital Continuity.

MBSE-AI Integration 2025 - 2nd Workshop on Model-based System Engineering and Artiﬁcial Intelligence

416

4 CONCLUSIONS

Based on current development trends and the

capabilities offered by AI, we proposed an AI-

integrated digital continuity framework. This

framework connects MBSE, MDAO, and DT,

enabling the utilization of high-dimensional

architectural design information across different

stages of the system lifecycle, and establishes data

feedback loops for subsequent processes in the

traditional system lifecycle, such as detailed design

and system operational data, offering new options to

support early-stage design verification and validation.

Within this framework, we defined the roles that

AI can play and the capabilities it provides. The AI-

integrated digital continuity framework supports

object-oriented design methods, offering a holistic

methodology for the evolution and optimization of

systems at different design stages.

In future work, we will apply AI algorithms to

enhance the efficiency of each process while focusing

on the practical value brought by this cross-lifecycle

iterative design process in improving system design

quality and efficiency, as well as uncovering

opportunities for new solutions.

ACKNOWLEDGEMENTS

This paper presents results that are developed in

collaboration between the IVECO Group company

and the University Lumiere Lyon 2, DISP Lab. This

research is established under a CIFRE contract

(2024/0091). The content of this paper reflects an

R&D initiative promoted by IVECO Group.

Responsibility for the information and views

expressed in this paper lies entirely with the authors.

REFERENCES

INCOSE (Ed.). (2023). INCOSE systems engineering

handbook. John Wiley & Sons.

Douglass, B. P., & von Holst, C. (2022). Agile Model-

Based Systems Engineering Cookbook: Improve system

development by applying proven recipes for effective

agile systems engineering. Packt Publishing Ltd.

Raza, H., Deol, E., Hedge, S., & Tanvir, H. (2024). Model

Based Systems Engineering for Sustainable

Autonomous Vehicle Design and

Development. Journal of Knowledge Learning and

Science Technology ISSN: 2959-6386 (online), 3(4),

188-199.

Zhao, J., Liang, H., Gao, X., Xu, T., & Yan, H. (2024). Co-

simulation Architecture and Platform Establishment

Method for Cloud-Based Predictive Cruise Control

System. Automotive Innovation, 1-11.

Habermehl, C., Höpfner, G., Berroth, J., Neumann, S., &

Jacobs, G. (2022). Optimization workflows for linking

model-based systems engineering (MBSE) and

multidisciplinary analysis and optimization

(MDAO). Applied Sciences, 12(11), 5316.

Simpson, T. W., & Martins, J. R. (2011). Multidisciplinary

design optimization for complex engineered systems:

report from a national science foundation workshop.

Moerland, E., Ciampa, P. D., Zur, S., Baalbergen, E.,

Noskov, N., D'Ippolito, R., & Lombardi, R. (2020).

Collaborative Architecture supporting the next

generation of MDAO within the AGILE

paradigm. Progress in Aerospace Sciences, 119,

100637.

Qi, K., Luo, Z., Zhao, H., & Guo, Y. (2024, May). Linking

Model-Based Systems Engineering (MBSE) and

Multidisciplinary Analysis and Optimization (MDAO)

for Early Design of Helicopter Primary Parameters.

In Journal of Physics: Conference Series (Vol. 2755,

No. 1, p. 012003). IOP Publishing.

Singh, M., Fuenmayor, E., Hinchy, E. P., Qiao, Y., Murray,

N., & Devine, D. (2021). Digital twin: Origin to

future. Applied System Innovation, 4(2), 36.

Alexandru, M., Dragoș, C., & Bălă-Constantin, Z. (2022).

Digital Twin for automated guided vehicles fleet

management. Procedia Computer Science, 199, 1363-

1369.

Tao, F., Zhang, H., Liu, A., & Nee, A. Y. (2018). Digital

twin in industry: State-of-the-art. IEEE Transactions on

industrial informatics, 15(4), 2405-2415.

Zhao, L., Alhoshan, W., Ferrari, A., Letsholo, K. J., Ajagbe,

M. A., Chioasca, E. V., & Batista-Navarro, R. T. (2021).

Natural language processing for requirements

engineering: A systematic mapping study. ACM

Computing Surveys (CSUR), 54(3), 1-41.

Mirjalili, S., & Dong, J. S. (2020). Multi-objective

optimization using artificial intelligence techniques.

Springer.

Li, Z., Wang, G., Lu, J., Broo, D. G., Kiritsis, D., & Yan,

Y. (2022). Bibliometric analysis of model-based

systems engineering: past, current, and future. IEEE

Transactions on Engineering Management, 71, 2475-

2492.

Ghanawi, I., Chami, M. W., Coric, M., & Abdoun, N. (2024,

July). Integrating AI with MBSE for Data Extraction

from Medical Standards. In INCOSE International

Symposium (Vol. 34, No. 1, pp. 1354-1366).

Raz, A. K., Blasch, E. P., Guariniello, C., & Mian, Z. T.

(2021). An overview of systems engineering challenges

for designing AI-enabled aerospace systems. In AIAA

Scitech 2021 Forum (p. 0564).

Zhang, J., & Yang, S. (2024). Recommendations for the

Model-Based Systems Engineering Modeling Process

Based on the SysML Model and Domain

Knowledge. Applied Sciences, 14(10), 4010.

AI-Integrated Framework for Enhancing High Level Architecture Design Across System Lifecycle Stages

417

Anton, K., & Oleg, K. (2023). MBSE and Safety Lifecycle

of AI-enabled systems in transportation. International

Journal of Open Information Technologies, 11(8), 100-

104.

Torkjazi, M., & Raz, A. K. (2024, July). Model‐Based

Systems Engineering (MBSE) Methodology for

Integrating Autonomy into a System of Systems Using

the Unified Architecture Framework. In INCOSE

International Symposium (Vol. 34, No. 1, pp. 1051-

1070).

Karali, H., Inalhan, G., & Tsourdos, A. (2024). AI-Driven

Multidisciplinary Conceptual Design of Unmanned

Aerial Vehicles. In AIAA SCITECH 2024 Forum.

Rasheed, A., San, O., & Kvamsdal, T. (2020). Digital twin:

Values, challenges and enablers from a modeling

perspective. IEEE access, 8, 21980-22012.

Ian, G., Yoshua, B., & Aaron, C. (2017). Deep learning:

Adaptive computation and machine learning.

Sutton, R. S., & Barto, A. G. (2018). Reinforcement

learning: An introduction. MIT press.

Wang, J., & Chen, Y. (2023). Introduction to transfer

learning: algorithms and practice. Springer Nature.

Zhang, K., Yang, Z., & Başar, T. (2021). Multi-agent

reinforcement learning: A selective overview of

theories and algorithms. Handbook of reinforcement

learning and control, 321-384.

Hao, Z., Liu, S., Zhang, Y., Ying, C., Feng, Y., Su, H., &

Zhu, J. (2022). Physics-informed machine learning: A

survey on problems, methods and applications. arXiv

preprint arXiv:2211.08064.

Raz, A. K., Blasch, E. P., Guariniello, C., & Mian, Z. T.

(2021). An overview of systems engineering challenges

for designing AI-enabled aerospace systems. In AIAA

Scitech 2021 Forum (p. 0564).

Blasch, E., & Pokines, B. (2019, July). Analytical science

for autonomy evaluation. In 2019 IEEE National

Aerospace and Electronics Conference (NAECON) (pp.

598-605). IEEE.

Alandihallaj, M. A., Ramezani, M., & Hein, A. M. (2024).

MBSE-Enhanced LSTM Framework for Satellite

System Reliability and Failure Prediction.

In MODELSWARD (pp. 349-356).

Chami, M., Abdoun, N., & Bruel, J. M. (2022, July).

Artificial Intelligence Capabilities for Effective Model‐

Based Systems Engineering: A Vision Paper.

In INCOSE International Symposium (Vol. 32, No. 1,

pp. 1160-1174).

Chen, M., & Bhada, S. V. (2022, July). Converting natural

language policy article into MBSE model. In INCOSE

International Symposium (Vol. 32, pp. 73-81).

Chami, M., Zoghbi, C., & Bruel, J. M. (2019). A first step

towards AI for MBSE: generating a part of SysML

models from text using AI. A First Step towards AI.

Johns, B., Carroll, K., Medina, C., Lewark, R., & Walliser,

J. (2024, July). AI Systems Modeling Enhancer (AI‐

SME): Initial Investigations into a ChatGPT‐enabled

MBSE Modeling Assistant. In INCOSE International

Symposium (Vol. 34, No. 1, pp. 1149-1168).

Rudolph, S. (2024, February). On Some Artificial

Intelligence Methods in the V-Model of Model-Based

Systems Engineering. In MODELSWARD (pp. 386-

393).

Wang, C., Fan, H., & Qiang, X. (2023). A Review of

Uncertainty-Based Multidisciplinary Design

Optimization Methods Based on Intelligent

Strategies. Symmetry, 15(10), 1875.

Sisk, S., Yan, G., & Du, X. (2023). Surrogate-based

Optimal Multidisciplinary Takeoff Trajectory Design

for Electric Drones. In AIAA AVIATION 2023 Forum (p.

3594).

Orlova, E. V. (2022). Design technology and AI-based

decision making model for digital twin

engineering. Future Internet, 14(9), 248.

Bariah, L., & Debbah, M. (2024). The interplay of ai and

digital twin: Bridging the gap between data-driven and

model-driven approaches. IEEE Wireless

Communications.

Groshev, M., Guimarães, C., Martín-Pérez, J., & de la Oliva,

A. (2021). Toward intelligent cyber-physical systems:

Digital twin meets artificial intelligence. IEEE

Communications Magazine, 59(8), 14-20.

Karali, H., Inalhan, G., & Tsourdos, A. (2023). AI-Based

Multifidelity Surrogate Models to Develop Next

Generation Modular UCAVs. In AIAA SCITECH 2023

Forum (p. 0670).

Ren, Y., Qi, J., Cheng, Y., Wang, J., & Alfarraj, O. (2020).

Digital continuity guarantee approach of electronic

record based on data quality theory. Computers,

Materials & Continua, 63(3), 1471-1483.

Fouda, M., Willrodt, L., Almeida, H., Cortez, J., Hussein,

O., Castaneda, F., & Brook, R. (2024). AN MBSE

ENABLED MDAO APPROACH FOR THE

CONCEPTUAL DEVELOPMENT OF COMPLEX

SYSTEMS.

Wu, Y., Zhou, L., Zheng, P., Sun, Y., & Zhang, K. (2022).

A digital twin-based multidisciplinary collaborative

design approach for complex engineering product

development. Advanced Engineering Informatics, 52,

101635.

Lopez, V., & Akundi, A. (2022, April). A conceptual

model-based systems engineering (mbse) approach to

develop digital twins. In 2022 ieee international

systems conference (syscon) (pp. 1-5). IEEE.

Bordeleau, F., Combemale, B., Eramo, R., Van Den Brand,

M., & Wimmer, M. (2020). Towards model-driven

digital twin engineering: Current opportunities and

future challenges. In Systems Modelling and

Management: First International Conference, ICSMM

2020, Bergen, Norway, June 25–26, 2020, Proceedings

1 (pp. 43-54). Springer International Publishing.

Purohit, S., & Madni, A. M. (2022). Employing Digital

Twins Within MBSE: Preliminary Results and Findings.

In Recent Trends and Advances in Model Based

Systems Engineering (pp. 35-44). Cham: Springer

International Publishing.

Bouhali, I., Idasiak, V., Martinez, J., Mhenni, F., Choley, J.

Y., Palladino, L., & Kratz, F. (2024, October).

Leveraging Aerospace Industry Digital Transformation:

Bridging the gap from MBSE to MBD with Digital

Twin Simulation for Mechatronic Systems

MBSE-AI Integration 2025 - 2nd Workshop on Model-based System Engineering and Artiﬁcial Intelligence

418

Requirements Early Validation. In 2024 IEEE

International Symposium on Systems Engineering

(ISSE) (pp. 1-8). IEEE.

Madni, A. M., Madni, C. C., & Lucero, S. D. (2019).

Leveraging digital twin technology in model-based

systems engineering. Systems, 7(1), 7.

Erikstad, S. O. MULTI-AGENT LLMS AND MBSE FOR

DEVELOPING DESIGN OPTIMIZATION MODELS.

Scott, W., Shirvani, F., Arabian, G., & Campbell, P. (2016,

July). Case Study: Customised Enhancement of an

MBSE Environment for Transport Infrastructure

Projects. In INCOSE International Symposium (Vol. 26,

No. 1, pp. 1703-1714).

AI-Integrated Framework for Enhancing High Level Architecture Design Across System Lifecycle Stages

419