pith. sign in

arxiv: 2606.03255 · v1 · pith:FTJLF4GQnew · submitted 2026-06-02 · 💻 cs.CE

Multi-Agent Framework Leveraging Knowledge Graphs for Virtual Commissioning Models

Max Diekmann , Jonas Nitzler , Jan Fischer , Hans-J\"urgen Pfisterer , Dirk Hartmann This is my paper

Pith reviewed 2026-06-28 08:13 UTC · model grok-4.3

classification 💻 cs.CE
keywords virtual commissioningknowledge graphsmulti-agent systemsPLC engineeringkinematic simulationTIA PortalNX MCDdiscrete manufacturing systems
0
0 comments X

The pith

A knowledge-graph multi-agent framework extracts data from PLC and simulation tools to semi-automate virtual commissioning model tasks.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper presents a framework that converts engineering data from incompatible sources into a shared graph database through a deterministic extraction process. A hierarchical multi-agent system then supports three recurring tasks in early virtual commissioning: answering questions about system behavior in natural language, generating executable simulation scripts, and suggesting mappings between control variables and simulation objects. The approach targets the manual cross-domain work that arises when PLC programs and kinematic models are built separately. Evaluation on a laboratory-scale manufacturing system indicates lower interpretation effort and more direct task support. The central claim is that grounding agents in the graph makes these engineering steps more reliable and less dependent on individual expertise.

Core claim

The paper's central claim is that a deterministic setup process can transform structured data from Siemens TIA Portal PLC projects and NX MCD kinematic models into a shared graph database, after which a hierarchical multi-agent architecture can deliver grounded natural-language queries, template-guided NX Open journal script generation, and ranked cross-domain mapping suggestions for virtual commissioning models.

What carries the argument

Hierarchical multi-agent architecture operating on a shared graph database built from extracted TIA Portal and NX MCD data.

If this is right

  • Manual cross-domain interpretation effort for virtual commissioning models decreases.
  • Recurring tasks of system understanding, simulation component generation, and signal mapping become more directly actionable.
  • Engineers gain natural-language access to combined engineering knowledge without switching tools.
  • Generation of executable NX Open journal scripts follows provided templates.
  • Mapping suggestions between PLC variables and NX MCD objects are produced with ranking.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The extraction and graph construction steps could be adapted to other PLC or simulation platforms if similar structured exports exist.
  • The same graph could later support additional automated checks such as consistency verification between control logic and kinematics.
  • Scaling the multi-agent tasks to full production lines may require new agent coordination rules not tested in the laboratory case.
  • Combining the graph with external data sources could extend the framework beyond initial model creation into ongoing maintenance.

Load-bearing premise

The deterministic setup process extracts complete and accurate structured data from Siemens TIA Portal and NX MCD and transforms both into a shared graph database without significant information loss or manual correction.

What would settle it

Apply the extraction process to a second laboratory or industrial system and check whether the resulting graph requires substantial manual fixes or loses critical PLC-to-simulation relationships before any agent tasks can run.

Figures

Figures reproduced from arXiv: 2606.03255 by Dirk Hartmann, Hans-J\"urgen Pfisterer, Jan Fischer, Jonas Nitzler, Max Diekmann.

Figure 2
Figure 2. Figure 2: Two-phase methodology pipeline. Phase I (setup) runs deterministic [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 1
Figure 1. Figure 1: Overview of the manufacturing under study [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 3
Figure 3. Figure 3: Architectural overview of the NX MCD data extraction pipeline, [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figure 5
Figure 5. Figure 5: Representative induced subgraph from the NX MCD domain [PITH_FULL_IMAGE:figures/full_fig_p005_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Agent roles and interaction pathways in the multi-agent system, [PITH_FULL_IMAGE:figures/full_fig_p005_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Compact example of a system-understanding query workflow from [PITH_FULL_IMAGE:figures/full_fig_p007_7.png] view at source ↗
read the original abstract

Virtual commissioning models (VCMs) of discrete manufacturing systems are used to validate automation behavior before physical deployment, but creating and maintaining them remains labor-intensive. Relevant engineering information is distributed across programmable logic controller (PLC) engineering projects, such as Siemens TIA Portal, and kinematic simulation models, such as Siemens NX Mechatronics Concept Designer (NX MCD), where it is stored in incompatible, tool-specific data structures. In practice, IEC 61131-3-based PLC programs and variables are engineered separately from rigid-body and kinematic simulation objects such as parts, joints, sensors, and actuators. As a result, understanding system behavior, generating simulation components, and mapping PLC variables to corresponding simulation objects require cross-domain expertise and remain largely manual. This paper presents a knowledge-graph-grounded multi-agent framework for semi-automated VCM development. A deterministic setup process extracts structured data from Siemens TIA Portal and Siemens NX MCD and transforms both sources into graph-based representations within a shared graph database. The framework uses a hierarchical multi-agent architecture to support three task classes in early-stage VCM development: system understanding, simulation component generation, and cross-domain signal mapping. It provides grounded natural-language access to engineering knowledge, template-guided generation of executable NX Open journal scripts, and ranked mapping suggestions between PLC variables and NX MCD simulation objects. Evaluation on a laboratory-scale discrete manufacturing system shows that the approach reduces manual cross-domain interpretation effort and makes recurring VCM engineering tasks more actionable.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

1 major / 0 minor

Summary. The paper presents a knowledge-graph-grounded multi-agent framework for semi-automated virtual commissioning model (VCM) development. A deterministic setup process extracts structured data from Siemens TIA Portal PLC projects and Siemens NX MCD kinematic models into a shared graph database. A hierarchical multi-agent architecture then supports three task classes: system understanding via grounded natural-language queries, template-guided generation of executable NX Open scripts, and ranked cross-domain mapping of PLC variables to simulation objects. Evaluation on a laboratory-scale discrete manufacturing system is reported to reduce manual cross-domain interpretation effort and make recurring VCM tasks more actionable.

Significance. If the extraction process produces complete and accurate graphs and the agent tasks demonstrably reduce effort with measurable gains, the work could provide a practical bridge between incompatible engineering data sources in industrial automation. The combination of knowledge graphs with hierarchical agents for grounded access and script generation offers a concrete engineering application that could be extended to other multi-tool workflows. The manuscript supplies no quantitative metrics, error analysis, or protocol details, so the significance remains prospective rather than established.

major comments (1)
  1. [Abstract (setup process paragraph)] Abstract, paragraph on setup process: The central claim that the framework reduces manual cross-domain effort rests on the assertion that the deterministic extraction 'transforms both sources into graph-based representations without significant information loss.' No description is given of how custom blocks, user-defined types, or non-standard kinematic constraints in the proprietary TIA Portal and NX MCD formats are parsed or preserved; if such elements are dropped or mis-mapped, the downstream agent tasks operate on an incomplete model and the reported effort reduction does not follow.

Simulated Author's Rebuttal

1 responses · 0 unresolved

Thank you for the opportunity to respond to the referee report. We have carefully considered the major comment and outline our planned revisions below.

read point-by-point responses

  1. Referee: [Abstract (setup process paragraph)] Abstract, paragraph on setup process: The central claim that the framework reduces manual cross-domain effort rests on the assertion that the deterministic extraction 'transforms both sources into graph-based representations without significant information loss.' No description is given of how custom blocks, user-defined types, or non-standard kinematic constraints in the proprietary TIA Portal and NX MCD formats are parsed or preserved; if such elements are dropped or mis-mapped, the downstream agent tasks operate on an incomplete model and the reported effort reduction does not follow.

    Authors: We thank the referee for this observation. The claim in the abstract regarding 'without significant information loss' is indeed not supported by a detailed description of the parsing logic for custom or non-standard elements in the current manuscript. The extraction is deterministic for the standard elements encountered in our laboratory-scale case study. To address this, we will revise the abstract to tone down the claim to 'into graph-based representations of the primary data structures' and add a new subsection in the methods describing the extraction rules for standard elements and explicitly stating the current limitations with custom blocks, user-defined types, and non-standard constraints. This will clarify the scope of the effort reduction achieved. revision: yes

Circularity Check

0 steps flagged

No circularity; descriptive framework with independent empirical evaluation

full rationale

The paper describes a practical multi-agent software framework for virtual commissioning that extracts data from TIA Portal and NX MCD into a shared graph database, then applies agents for system understanding, script generation, and mapping. No equations, fitted parameters, or predictions appear in the provided text. The central claim rests on an empirical evaluation on a laboratory-scale system rather than any derivation that reduces to its own inputs by construction. No self-citation chains, uniqueness theorems, or ansatzes are invoked. The extraction process is presented as a deterministic engineering step, not a self-definitional or fitted result. The derivation chain is therefore self-contained against external benchmarks.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 1 invented entities

The central claim rests on the assumption that tool data can be losslessly converted to graphs and that the multi-agent system can reliably perform the three task classes; no free parameters or invented physical entities are described.

axioms (1)
  • domain assumption Data from Siemens TIA Portal and NX MCD can be extracted and transformed into graph representations without significant loss.
    Invoked in the deterministic setup process described in the abstract.
invented entities (1)
  • Hierarchical multi-agent architecture no independent evidence
    purpose: Support three task classes in early-stage VCM development
    Introduced as the core of the framework; no independent evidence outside the paper.

pith-pipeline@v0.9.1-grok · 5803 in / 1228 out tokens · 27259 ms · 2026-06-28T08:13:53.721880+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

42 extracted references · 12 canonical work pages

  1. [1]

    VDI/VDE 3693 Blatt 1: Virtuelle Inbetriebnahme - Modellarten und Begriffe,

    Verein deutscher Ingenieure, “VDI/VDE 3693 Blatt 1: Virtuelle Inbetriebnahme - Modellarten und Begriffe,” 2025. [Online]. Available: https://www.vdi.de/mitgliedschaft/vdi-richtlinien/details/ vdivde-3693-blatt-1-virtuelle-inbetriebnahme-modellarten-und-begriffe

    2025

  2. [2]

    Virtual Commissioning Of Manufacturing Systems A Review And New Approaches For Simplification,

    P. Hoffmann, T. M. Maksoud, R. Schumann, and G. C. Premier, “Virtual Commissioning Of Manufacturing Systems A Review And New Approaches For Simplification,” inProceedings - 24th European Conference on Modelling and Simulation, ECMS, 2010, pp. 175–181. [Online]. Available: http://www.scs-europe.net/dlib/2010/ 2010-0175.htm

    2010

  3. [3]

    Concepts and trends of virtual commissioning – A comprehensive review,

    N. Striffler and T. V oigt, “Concepts and trends of virtual commissioning – A comprehensive review,”Journal of Manufacturing Systems, vol. 71, pp. 664–680, 2023. [Online]. Available: https://linkinghub.elsevier.com/ retrieve/pii/S0278612523002145

    2023

  4. [4]

    Real-time digital twins,

    D. Hartmann, “Real-time digital twins,”Zenodo, 2021. [Online]. Available: https://doi.org/10.5281/zenodo.5470479

    work page doi:10.5281/zenodo.5470479 2021

  5. [5]

    A Survey on Digital Twin: Definitions, Characteristics, Applications, and Design Implications,

    B. R. Barricelli, E. Casiraghi, and D. Fogli, “A Survey on Digital Twin: Definitions, Characteristics, Applications, and Design Implications,” IEEE Access, vol. 7, pp. 167 653–167 671, 2019. [Online]. Available: https://ieeexplore.ieee.org/document/8901113/

    arXiv 2019

  6. [6]

    Digital Twins - a golden age for industrial mathematics,

    D. Hartmann and H. Van Der Auweraer, “Digital Twins - a golden age for industrial mathematics,”Journal of Mathematics in Industry, vol. 15, p. 6, 2025. [Online]. Available: https://mathematicsinindustry. springeropen.com/articles/10.1186/s13362-025-00170-3

    work page doi:10.1186/s13362-025-00170-3 2025

  7. [7]

    IEC 61131-3: Pro- grammable controllers - Part 3: Programming languages,

    International Electrotechnical Commission, “IEC 61131-3: Pro- grammable controllers - Part 3: Programming languages,” 2025. [Online]. Available: https://webstore.iec.ch/en/publication/68533

    2025

  8. [8]

    Leveraging Large Language Models for Enhanced Digital Twin Modeling: Trends, Methods, and Challenges,

    L. Yang, S. Luo, X. Cheng, and L. Yu, “Leveraging Large Language Models for Enhanced Digital Twin Modeling: Trends, Methods, and Challenges,” 2025. [Online]. Available: https://arxiv.org/abs/2503.02167

    arXiv 2025

  9. [9]

    Knowledge Graphs as Enhancers of Intelligent Digital Twins,

    N. Sahlab, S. Kamm, T. M ¨uller, N. Jazdi, and M. Weyrich, “Knowledge Graphs as Enhancers of Intelligent Digital Twins,” in4th International Conference on Industrial Cyber-Physical Systems, ICPS, 2021, pp. 19–

    2021

  10. [10]

    Available: https://ieeexplore.ieee.org/document/9468219/

    [Online]. Available: https://ieeexplore.ieee.org/document/9468219/

    arXiv

  11. [11]

    Multi-Agent Systems for Manufacturing Digital Twins: A Perspective on Agency and Large Language Models,

    N. P. Greis, H. P. Cherukuri, and J. C. Outeiro, “Multi-Agent Systems for Manufacturing Digital Twins: A Perspective on Agency and Large Language Models,”IFAC-PapersOnLine, vol. 59, pp. 1612–1617,

  12. [12]

    Available: https://linkinghub.elsevier.com/retrieve/pii/ S2405896325010328

    [Online]. Available: https://linkinghub.elsevier.com/retrieve/pii/ S2405896325010328

  13. [13]

    Digital Twin: Origin to Future,

    M. Singh, E. Fuenmayor, E. Hinchy, Y . Qiao, N. Murray, and D. Devine, “Digital Twin: Origin to Future,”Applied System Innovation, vol. 4, p. 36, 2021. [Online]. Available: https://www.mdpi.com/2571-5577/4/ 2/36

    2021

  14. [14]

    Design of a material sorting digital twin system based on NX MCD,

    M. Li and D. Yang, “Design of a material sorting digital twin system based on NX MCD,”Manufacturing Review, vol. 11, p. 13,

  15. [15]

    Available: https://mfr.edp-open.org/10.1051/mfreview/ 2024010

    [Online]. Available: https://mfr.edp-open.org/10.1051/mfreview/ 2024010

    work page doi:10.1051/mfreview/

  16. [16]

    Towards Self-X cognitive manufacturing network: An industrial knowledge graph- based multi-agent reinforcement learning approach,

    P. Zheng, L. Xia, C. Li, X. Li, and B. Liu, “Towards Self-X cognitive manufacturing network: An industrial knowledge graph- based multi-agent reinforcement learning approach,”Journal of Manufacturing Systems, vol. 61, pp. 16–26, 2021. [Online]. Available: https://linkinghub.elsevier.com/retrieve/pii/S0278612521001643

    2021

  17. [17]

    A knowledge graph based construction method for Digital Twin Network,

    Y . Zhu, D. Chen, C. Zhou, L. Lu, and X. Duan, “A knowledge graph based construction method for Digital Twin Network,” in1st International Conference on Digital Twins and Parallel Intelligence, DTPI. IEEE, 2021, pp. 362–365. [Online]. Available: https://ieeexplore.ieee.org/document/9540177/

    arXiv 2021

  18. [18]

    Drath, Ed.,AutomationML: A Practical Guide

    R. Drath, Ed.,AutomationML: A Practical Guide. De Gruyter, 2021. [Online]. Available: https://doi.org/10.1515/9783110746235

    work page doi:10.1515/9783110746235 2021

  19. [19]

    Asset Administration Shell for industrial applications - Part 1: Asset Administration Shell structure,

    International Electrotechnical Commission, “Asset Administration Shell for industrial applications - Part 1: Asset Administration Shell structure,”

  20. [20]

    Available: https://webstore.iec.ch/en/publication/65628

    [Online]. Available: https://webstore.iec.ch/en/publication/65628

  21. [21]

    An Architecture for Knowledge Graph based Simulation Support,

    F. G. Listl, J. Fischer, and M. Weyrich, “An Architecture for Knowledge Graph based Simulation Support,” in28th International Conference on Emerging Technologies and Factory Automation, ETFA, 2023, pp. 1–8. [Online]. Available: https://ieeexplore.ieee.org/document/10275514/

    arXiv 2023

  22. [22]

    LLM-Based Digital Twin Water Conservancy Knowledge Graph Construction,

    Y . Yang, F. Ye, D. Xu, X. Zhang, and J. Xu, “LLM-Based Digital Twin Water Conservancy Knowledge Graph Construction,” in 36th International Conference on Tools with Artificial Intelligence, ICTAI. IEEE, 2024, pp. 656–662. [Online]. Available: https: //ieeexplore.ieee.org/document/10849426/

    arXiv 2024

  23. [23]

    LLMasMMKG: LLM Assisted Synthetic Multi-Modal Knowledge Graph Creation For Smart City Cognitive Digital Twins,

    S. Mandal and N. E. O’Connor, “LLMasMMKG: LLM Assisted Synthetic Multi-Modal Knowledge Graph Creation For Smart City Cognitive Digital Twins,”Proceedings of the AAAI Symposium Series, vol. 4, pp. 210–221, 2024. [Online]. Available: https: //ojs.aaai.org/index.php/AAAI-SS/article/view/31795

    2024

  24. [24]

    php/AAAI/article/view/12028/11887

    K. Shuster, S. Poff, M. Chen, D. Kiela, and J. Weston, “Retrieval Augmentation Reduces Hallucination in Conversation,” inFindings of the Association for Computational Linguistics: EMNLP, 2021, pp. 3787–3803. [Online]. Available: https://doi.org/10.18653/v1/2021. findings-emnlp.320

    work page doi:10.18653/v1/2021 2021

  25. [25]

    Retrieval-Augmented Generation for Knowledge- Intensive NLP Tasks,

    P. Lewis, E. Perez, A. Piktus, F. Petroni, V . Karpukhin, N. Goyal, H. K ¨uttler, M. Lewis, W. tau Yih, T. Rockt ¨aschel, S. Riedel, and D. Kiela, “Retrieval-Augmented Generation for Knowledge- Intensive NLP Tasks,”Advances in neural information processing systems, NeurIPS, vol. 33, pp. 9459–9474, 2020. [Online]. Available: https://arxiv.org/abs/2005.11401

    Pith/arXiv arXiv 2020

  26. [26]

    Retrieval-Augmented Generation for AI-Generated Content: A Survey,

    P. Zhao, H. Zhang, Q. Yu, Z. Wang, Y . Geng, F. Fu, L. Yang, W. Zhang, J. Jiang, and B. Cui, “Retrieval-Augmented Generation for AI-Generated Content: A Survey,”Data Science and Engineering, pp. 1–29, 2026. [Online]. Available: https://doi.org/10.1007/s41019-025-00335-5

    work page doi:10.1007/s41019-025-00335-5 2026

  27. [27]

    Graph-based Approaches and Functionalities in Retrieval-Augmented Generation: A Comprehensive Survey,

    Z. Zhu, T. Huang, K. Wang, J. Ye, X. Chen, and S. Luo, “Graph-based Approaches and Functionalities in Retrieval-Augmented Generation: A Comprehensive Survey,”ACM Computing Surveys, vol. 58, pp. 1–38,

  28. [28]

    Available: https://dl.acm.org/doi/10.1145/3795880

    [Online]. Available: https://dl.acm.org/doi/10.1145/3795880

    work page doi:10.1145/3795880

  29. [29]

    From Local to Global: A Graph RAG Approach to Query-Focused Summarization,

    D. Edge, H. Trinh, N. Cheng, J. Bradley, A. Chao, A. Mody, S. Truitt, D. Metropolitansky, R. O. Ness, and J. Larson, “From Local to Global: A Graph RAG Approach to Query-Focused Summarization,” 2024. [Online]. Available: https://arxiv.org/abs/2404.16130

    Pith/arXiv arXiv 2024

  30. [30]

    MADTwin: a framework for multi-agent digital twin development: smart warehouse case study,

    H. Marah and M. Challenger, “MADTwin: a framework for multi-agent digital twin development: smart warehouse case study,” Annals of Mathematics and Artificial Intelligence, vol. 92, pp. 975–1005, 2024. [Online]. Available: https://link.springer.com/10.1007/ s10472-023-09872-z

    2024

  31. [31]

    Empowering digital twins with large language models for global temporal feature learning,

    Y . Sun, Q. Zhang, J. Bao, Y . Lu, and S. Liu, “Empowering digital twins with large language models for global temporal feature learning,”Journal of Manufacturing Systems, vol. 74, pp. 83–99,

  32. [32]

    Available: https://linkinghub.elsevier.com/retrieve/pii/ S0278612524000372

    [Online]. Available: https://linkinghub.elsevier.com/retrieve/pii/ S0278612524000372

  33. [33]

    Agents4PLC: Automating Closed-loop PLC Code Generation and Verification in Industrial Control Systems using LLM-based Agents,

    Z. Liu, R. Zeng, D. Wang, and G. Peng, “Agents4PLC: Automating Closed-loop PLC Code Generation and Verification in Industrial Control Systems using LLM-based Agents,”IEEE Transactions on Software Engineering, TSE, 2026. [Online]. Available: https: //doi.org/10.1109/TSE.2026.3667895

    work page doi:10.1109/tse.2026.3667895 2026

  34. [34]

    Benchmarking and validation of prompting techniques for AI-assisted industrial PLC programming,

    K. Adnyana and A. Schwung, “Benchmarking and validation of prompting techniques for AI-assisted industrial PLC programming,” Machine Learning with Applications, vol. 23, p. 100804, 2026. [Online]. Available: https://doi.org/10.1016/j.mlwa.2025.100804

    work page doi:10.1016/j.mlwa.2025.100804 2026

  35. [35]

    Festo Didactic CP Factory Complete System 404-1

    Festo, “Festo Didactic CP Factory Complete System 404-1.” [On- line]. Available: https://ip.festo-didactic.com/Infoportal/CPFactoryLab/ hardware/details.php?model=CP-L-404-1&lang=en

  36. [36]

    Neo4j Graph Database,

    Neo4j, “Neo4j Graph Database,” 2026. [Online]. Available: https: //neo4j.com/

    2026

  37. [37]

    Normalizing Property Graphs,

    P. Skavantzos and S. Link, “Normalizing Property Graphs,”Proceedings of the VLDB Endowment, vol. 16, pp. 3031–3043, 2023. [Online]. Available: https://dl.acm.org/doi/10.14778/3611479.3611506

    work page doi:10.14778/3611479.3611506 2023

  38. [38]

    LangChain framework,

    LangChain, “LangChain framework,” 2026. [Online]. Available: https://www.langchain.com/

    2026

  39. [39]

    GPT-4 Technical Report,

    OpenAI et al., “GPT-4 Technical Report,” Tech. Rep., 2023. [Online]. Available: https://arxiv.org/abs/2303.08774

    Pith/arXiv arXiv 2023

  40. [40]

    Holland and K

    M. Holland and K. Chaudhari,Manufacturing Letters, vol. 40, pp. 100–103, 2024. [Online]. Available: https://doi.org/10.1016/j.mfglet. 2024.03.010

    work page doi:10.1016/j.mfglet 2024

  41. [41]

    Classification and Review of Multi-agents Systems in the Manufacturing Section,

    G. Andreadis, P. Klazoglou, K. Niotaki, and K.-D. Bouzakis, “Classification and Review of Multi-agents Systems in the Manufacturing Section,”Procedia Engineering, vol. 69, pp. 282–290,

  42. [42]

    Available: https://doi.org/10.1016/j.proeng.2014.02.233

    [Online]. Available: https://doi.org/10.1016/j.proeng.2014.02.233

    work page doi:10.1016/j.proeng.2014.02.233 2014