PaperlayerView exact revision on arXiv

Source-linked AI summary

Cross-Organizational SysML Model Integration: A Survey of Challenges and AI-Supported Tasks

Zirui Li, Torsten Brix, Stephan Husung

arXiv:2609.37000v1cs.SEcs.AI

TL;DR

Cross-organizational SysML model exchange remains difficult, and practitioner perspectives on integration challenges and AI’s role remain insufficiently explored. This paper surveys 29 MBSE stakeholders about integration challenges and AI-supported tasks. Respondents describe integration as multidimensional, with perceptions varying by role and experience, and value AI analysis support.

  • Problem

    Practitioner perspectives on cross-organizational model exchange and integration, including emerging AI’s role, remain insufficiently explored.

    Source: §I · p. 2

  • Method

    An online questionnaire gathered 29 stakeholders’ assessments of integration challenges, AI-supported tasks, and collaboration modes.

    Source: §I · p. 2

  • Results

    Stakeholders report semantic, behavioral, traceability, and format-related integration difficulties that vary by role and experience, and value AI for model understanding, conflict detection, and traceability analysis.

    Source: §VI · p. 9

  • Takeaways & Limitations

    The findings support AI as an embedded, assistive MBSE tool-chain component under human oversight, with future work directed toward role-sensitive integration support.

    Source: E. Limitations and future work · p. 8, §VI · p. 9

    The small, voluntary sample and self-reported ordinal ratings limit generalizability and reflect subjective perceptions rather than objective integration performance.

    Source: E. Limitations and future work · p. 8

Abstract

from arXiv · show

Cross-organizational collaboration is widely regarded as a key promise of SysML-based Model-Based Systems Engineering (MBSE), yet practitioners still face persistent challenges when exchanging and integrating system models. In parallel, Large Language Models (LLMs) raise expectations for AI-assisted model understanding and integration, while reliability and required human oversight continue to pose challenges. This paper reports the results of an online questionnaire survey with 29 MBSE stakeholders involved in cross-organizational collaboration. Respondents rated eight predefined integration challenge categories and six AI-supported task types on five-point Likert scales. The results indicate that stakeholders perceive model integration as a multi-dimensional alignment problem across semantics, behavior, traceability, and exchange interoperability. These perceptions vary by organizational role and frequency of integration involvement. AI is rated highly useful for analysis tasks such as semantic structure analysis and inconsistency detection, and respondents predominantly prefer human-in-the-loop use with mandatory verification. These findings motivate AI support that enhances, rather than replaces, engineering responsibility in SysML-based integration.

I. INTRODUCTION

Cross-organizational SysML model integration remains difficult amid heterogeneous methods, tools, and domain practices, while practitioner perspectives and AI’s role remain underexplored. The paper frames this gap through MBSE integration challenges and emerging SysML v2 capabilities.

I. INTRODUCTION

  • Cross-organizational collaboration is increasingly important, but heterogeneous modeling methods, tool ecosystems, and domain practices make SysML model exchange and integration challenging –.Source: §I · p. 2
  • Practitioner perspectives on cross-organizational model exchange and integration, including the role of emerging technologies such as AI, remain insufficiently explored.Source: §I · p. 2
  • The study asks how stakeholders assess integration challenges across roles and involvement levels, and how useful they find AI-supported tasks and collaboration modes.It uses a questionnaire with five-point Likert ratings and open-ended responses.Source: §I · p. 2

A. Model-Based Systems Engineering

  • SysML v1’s profile-based semantics, organization-specific extensions, and differing abstraction levels complicate standardized analysis and model alignment.Source: A. Model-Based Systems Engineering · p. 3
  • Profile-based variability can produce semantic drift, naming inconsistencies, mismatched abstraction levels, and divergent structures that increase manual reconciliation effort.Source: A. Model-Based Systems Engineering · p. 3
  • Although XMI supports model exchange, tool-specific implementations and metadata handling still require complex, context-dependent transformations and maintenance.SysML v2’s KerML semantic foundation supports consistency checking, transformation, and formal analysis, opening possibilities for cross-tool and cross-company collaboration.Source: A. Model-Based Systems Engineering · p. 3

B. Cross-company collaboration scenario in MBSE

Cross-company MBSE collaboration involves an OEM–supplier model exchange workflow with distinct alignment and interoperability challenges. The paper also frames AI assistance as a range of analysis, generative, and autonomous tasks that require careful consideration of reliability and accountability.

B. Cross-company collaboration scenario in MBSE

  • OEMs share subsystem-relevant model fragments with suppliers, who return detailed subsystem models for integration, verification, and iterative refinement.Source: B. Cross-company collaboration… · p. 3
  • A common modeling language does not ensure shared understanding; abstraction and naming differences, methods, and interface specifications create distinct alignment and interoperability challenges –.Source: B. Cross-company collaboration… · p. 3

C. AI in MBSE

  • Proposed LLM applications include requirements and modeling assistance, consistency checking, and SysML v2 API-enabled querying, modification, and element derivation,,.Source: C. AI in MBSE · p. 3
  • AI in MBSE includes analysis support, generative assistance, and autonomous integration decisions, with six task types evaluated in this study.Source: C. AI in MBSE · p. 3
  • Data quality, semantic deviations, hallucinated content, and limited transparency support positioning AI as a reviewed support tool rather than a fully autonomous agent,,,.Source: C. AI in MBSE · p. 3

III. SURVEY DESIGN AND DATA ANALYSIS METHOD

The study uses an online questionnaire to examine practical perceptions of SysML model exchange and integration. Its design combines structured ratings with open-ended responses and checks intended to support consistent interpretation.

A. Data collection approach and Survey Design

  • Three questionnaire parts covered respondent backgrounds and integration involvement, model-exchange challenges, and AI-supported tasks and collaboration modes.Items drew on literature themes and combined five-point Likert ratings, multiple-choice questions, and open-ended questions.Source: A. Data collection approach and Survey… · p. 4
  • The questionnaire was distributed through academic and industrial Systems Engineering communities primarily from May to July 2025, yielding 29 retained responses after consistency and relevance checks.Source: A. Data collection approach and Survey… · p. 4
  • Illustrative examples accompanied challenge categories to support consistent understanding, while open-ended answers complemented and contextualized Likert-scale findings.Source: A. Data collection approach and Survey… · p. 4

B. Data Analysis Methods

The analysis combines descriptive subgroup comparisons with qualitative coding of open-ended responses. The results section covers respondent context, integration challenges, AI-task usefulness, subgroup differences, and additional themes.

B. Data Analysis Methods

  • Descriptive statistics and subgroup comparisons characterize Likert responses by organizational role and integration involvement, with exploratory interpretation rather than statistical or causal inference.Source: B. Data Analysis Methods · p. 4
  • Open-ended responses were coded using Mayring’s qualitative content analysis, with literature-based initial categories refined through pilot coding and inductive subcategories.The finalized scheme was applied across responses to identify practical challenges, expectations, and risks.Source: B. Data Analysis Methods · p. 4

IV. SURVEY RESULT

  • The results section reports integration challenges, AI-task usefulness, differences by engagement and role, and themes from open-ended responses.Source: §IV · p. 4

A. Respondent background and modeling context

B. Challenges rating in model exchange and integration

Stakeholders rated integration challenges across semantics, behavior, traceability, and formats, while AI was viewed as most useful for analysis and less useful for generative tasks; respondents favored mandatory human verification.

B. Challenges rating in model exchange and integration

  • A, D, F, and G each had median 4, about 60% high ratings, and means of 3.55–3.97; B also had median 4 but fewer high ratings (51.7%).The categories represent naming and semantic conflicts, behavioral and logical inconsistencies, missing traceability, format differences, and structural or abstraction challenges.Source: B. Challenges rating in model exchange… · p. 5, Table I · p. 5
  • Interface and port integration (C), methodological and SysML profile integration (E), and metadata-related issues (H) each had median 3; high ratings ranged from 34.5% to 39.3%, and means were approximately 3.11 to 3.25.The passage characterizes their response distributions as more dispersed.Source: B. Challenges rating in model exchange… · p. 5

C. AI support usefulness and oversight preferences

  • Semantic structure analysis support and identifying semantic conflicts or inconsistencies were among the most consistently highly rated tasks, each with a median rating of 5 and high-rating proportions above 80%.Source: C. AI support usefulness and oversight… · p. 5
  • Model content comprehension and model-element mapping suggestions each received over 75% high ratings, with mean scores of 4.28 and 4.21, respectively.These tasks were rated highly, though less consistently than the leading semantic-analysis tasks.Source: C. AI support usefulness and oversight… · p. 5, Table II · p. 5
  • Integration-workflow generation and cross-methodology transformation suggestions remained positively rated but had median ratings of 4 and lower high-rating proportions.Overall, respondents rated analysis, interpretive, and conflict-detection tasks more favorably than generative tasks.Source: C. AI support usefulness and oversight… · p. 5, Table II · p. 5
  • 83% preferred AI support with mandatory human verification, while 17% preferred AI suggestions with the primary work remaining manual; neither extreme workflow was selected.The two unselected extremes were fully manual workflows and highly automated modes with optimal review.Source: C. AI support usefulness and oversight… · p. 5

D. Engagement-Dependent Differences in Challenge Perception

Challenge ratings varied with reported engagement frequency: higher-engagement respondents rated behavioral misalignment, format gaps, and metadata issues higher, while several other categories were rated slightly higher by the low-engagement group.

D. Engagement-Dependent Differences in Challenge Perception

  • The comparison grouped 14 respondents reporting frequent or occasional involvement as high engagement and 15 reporting rare or no involvement as low engagement.Figure 1 reports group-level mean ratings for the eight integration challenges.Source: D. Engagement-Dependent Differences in… · p. 5, Figure 1 · p. 6
  • Higher-engagement respondents tended to rate behavioral misalignment, format gaps, and metadata issues as more important, while naming conflicts, interface and port issues, and traceability gaps were slightly higher for the low-engagement group.Figure 1 compares perceived challenge severity between the two engagement groups.Source: D. Engagement-Dependent Differences in… · pp. 5–6, Figure 1 · p. 6
  • Engagement-group differences were more evident for dynamic consistency and exchange artifacts, while ratings for most basic modeling challenges remained relatively similar.Differences between groups remained small for structural abstraction and methodology or profile integration.Source: D. Engagement-Dependent Differences in… · pp. 5–6

E. Role-Specific Differences in Challenge Perception

Role-based comparisons show broadly similar challenge ratings with differences on particular aspects; open-ended responses, analyzed through qualitative content analysis, add practical challenges and AI requirements beyond the predefined categories.

E. Role-Specific Differences in Challenge Perception

  • Challenge A rates highly across roles, while suppliers report greater difficulty on B, C, and E; tool vendors rate F higher, and suppliers and tool vendors rate G and H higher than OEMs.Role groups partially overlap because respondents could select multiple roles, so these are ratings by role experience rather than comparisons of mutually exclusive groups.Source: E. Role-Specific Differences in… · p. 6
  • Overall, challenge ratings are similar across roles, with differences concentrated in individual aspects.Figure 2 presents role-specific perceptions for challenges A–H.Source: E. Role-Specific Differences in… · p. 6, Figure 2 · p. 7

F. Themes from open-ended responses

  • Respondents added interoperability concerns including restricted model-data access, unstable exchanges, and practical difficulties with standardized formats.The analysis used qualitative content analysis based on Mayring.Source: F. Themes from open-ended responses · p. 6
  • Additional challenges span change and consistency management, process and governance constraints including document-based contractual deliverables, and mismatches in abstraction level or disciplinary perspective.These themes cover model-version changes, consistency with lifecycle artifacts, late exchange planning, and insufficient integration into information-management strategies.Source: F. Themes from open-ended responses · p. 6
  • Promising AI uses include model understanding and conflict detection, repetitive-task automation, model generation or transformation, and traceability and impact analysis across lifecycle artifacts.Examples include natural-language model queries, notation correction, diagram creation, and transformations across methodologies or SysML versions.Source: F. Themes from open-ended responses · p. 6
  • Respondents cautioned against using AI to compensate for poor data or modeling discipline and urged caution with full automation.They also identified knowledge structuring, natural-language access, standardized APIs, write-back, and integration with ALM, PLM, and Digital Twin environments as relevant requirements or opportunities.Source: F. Themes from open-ended responses · pp. 6–7

V. DISCUSSION

The discussion relates questionnaire findings to existing MBSE and AI-assisted Systems Engineering research, interpreting challenge prominence, role and experience variation, and implications for AI-assisted integration.

  • The discussion connects empirical survey findings with prior research and considers why challenges are prominent, how perceptions vary, and what they imply for AI-assisted SysML integration.Source: §V · p. 7

A. Core integration challenges in SysML-based MBSE

The discussion characterizes SysML model integration as a multi-layer alignment challenge, while noting that formal semantics and exchange infrastructure alone do not resolve cross-organizational coordination needs.

  • SysML model integration is a multi-dimensional alignment challenge spanning semantics, behavior, traceability, and exchange formats, rather than one technical bottleneck.The reported high ratings across these aspects suggest challenges emerge across multiple model layers simultaneously.Source: A. Core integration challenges in… · p. 7
  • Naming and semantic differences, abstraction and structural mismatches, behavioral misalignment, traceability gaps, and format issues complicate integration across organizations.The discussion links these challenges to divergent interpretations, asynchronous model evolution, and lifecycle information management.Source: A. Core integration challenges in… · p. 7
  • SysML v2’s formal semantics can support ambiguity reduction and automated consistency checking, but formalized semantics alone are insufficient without explicit alignment mechanisms.Source: A. Core integration challenges in… · p. 7
  • Fragile XMI exchanges and tool-specific transformations motivate robust exchange approaches; standardized APIs may support controlled model access and continuous integration.Source: A. Core integration challenges in… · p. 7
  • Stakeholders see AI primarily as assistance for model understanding, semantic analysis, and conflict detection, while expressing caution toward generative or autonomous transformations.The discussion frames AI as complementary to engineering work within semantic, organizational, and technical integration frameworks.Source: A. Core integration challenges in… · p. 7

B. Role- and experience-specific perspectives

Challenge perceptions vary with role and integration involvement; respondents favor analytical AI assistance under human review, alongside coordinated advances in semantics, infrastructure, and tooling.

B. Role- and experience-specific perspectives

  • Perceived integration challenges vary with organizational role and involvement, reflecting practical exposure and responsibility in collaboration workflows.Source: B. Role- and experience-specific… · pp. 7–8
  • Less-involved respondents rate naming, semantic, and traceability issues slightly higher, while highly engaged respondents emphasize behavior, formats, methods, profiles, and metadata.Differences in structural abstraction and interface integration are comparatively small across engagement groups.Source: B. Role- and experience-specific… · p. 8
  • Role-specific concerns include OEM attention to semantics, supplier difficulty with abstraction and methodology alignment, and tool-provider sensitivity to traceability and formats.Researchers show comparatively balanced ratings, with somewhat greater attention to semantic, behavioral, and traceability issues.Source: B. Role- and experience-specific… · p. 8

C. Insights for AI-assisted integration

  • Stakeholders value AI more for model comprehension, semantic analysis, and inconsistency detection than for generating workflows or cross-methodology alignment suggestions.The latter tasks receive lower and more heterogeneous ratings, possibly reflecting caution about AI shaping modeling decisions or methods amid responsibility boundaries.Source: C. Insights for AI-assisted integration · p. 8
  • Most respondents prefer human-led AI assistance with mandatory human review, rather than fully manual work or highly automated modes with minimal oversight.Source: C. Insights for AI-assisted integration · p. 8

D. Insights from Open-Ended Responses

  • Open-ended responses call for explicit alignment beyond SysML v2 semantics, API-centric tool-independent exchange, and support for lifecycle artifacts and version management.They place structurally invasive or generative AI decisions under human control.Source: D. Insights from Open-Ended Responses · p. 8

E. Limitations and future work

Stakeholders identify integration challenges across semantics, behavior, traceability, and formats, while valuing AI for analysis more than generative or highly automated tasks. The study’s findings are constrained by a small, voluntary sample and self-reported perceptions; future work will explore AI-assisted alignment under explicit human control.

E. Limitations and future work

  • The small voluntary sample may introduce selection bias and limits generalizability; self-reported ordinal Likert responses reflect subjective experience rather than objective integration performance.Source: E. Limitations and future work · p. 8
  • Role-based comparisons are constrained by overlapping role assignments and uneven group sizes.Source: E. Limitations and future work · p. 8
  • Future work will develop and evaluate model-alignment mechanisms addressing semantic, behavioral, and traceability consistency across heterogeneous tools and abstraction levels, with AI support under explicit human control.Source: E. Limitations and future work · p. 8

VI. CONCLUSION

  • Integration difficulties span semantic, behavioral, traceability, and format-related issues, with their relative importance varying by organizational role and integration experience.Source: §VI · p. 9
  • Stakeholders value AI for model understanding, semantic conflict detection, and traceability analysis, but are more cautious about generative transformations and highly automated workflows.Source: §VI · p. 9
Loading 2609.37000v1…