EUVIMEDEUVIMED entdecken
Open Access · CC BY 4.0

Large Language Model Hallucination Mitigation in Three Industrial Use Cases

Petri Tikka, Jaakko Karjalainen, Andrea Alesani, Vladimir Goriachev

IEEE Access · 2026 · 14 · 25564-25576

DOI: 10.1109/access.2026.3659997 ↗

Kurz gefasst

Fachartikel in der Zeitschrift IEEE Access, 2026. Verfasst von Petri Tikka und 3 weiteren Autor:innen. Die Arbeit wurde bisher 3-mal zitiert (Crossref) und nennt 36 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0). Gefördert von: Business Finland, VTT Technical Research Centre Ltd.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Large-language models (LLMs) can produce factually false or ungrounded information. This occurs when the language model, being fundamentally a statistical model, generates inaccurate or non-existent information with high confidence in response to a given query. This phenomenon is particularly dangerous in safety-critical systems and involves risks when LLMs are used to access or control hardware such as robots, sensors, and other internet-of-things (IoT) applications. Additionally, in automation-rich industrial environments, effective human-machine cooperation depends on maintaining a shared and adaptive understanding of complex situations. Distributed Situation Awareness (DSA) provides a framework for how awareness emerges collectively across networks of operators, robots, sensors, digital interfaces, and LLM based artificially intelligent systems. LLMs can fuse fragmented data streams into coherent, actionable context, enabling Extended Reality (XR) technologies to strengthen DSA by embedding digital cues into physical workflows. While this process improves coordination and adaptability, it also makes reliable and verifiable model outputs essential, as hallucinations can erode operator trust and compromise distributed decision-making. Therefore, mitigation of hallucinations becomes essential for sustaining stable human–AI teaming. We present three industrial projects where LLMs are at the forefront, highlighting practical approaches for hallucination mitigation and demonstrating a transition from online to offline model. Scope is to demonstrate through architectural means how established mitigation mechanisms can be used in industrial systems.

Förderung

  • Business Finland
  • VTT Technical Research Centre Ltd

Zitieren

APA

Tikka, P., Karjalainen, J., Alesani, A., & Goriachev, V. (2026). Large Language Model Hallucination Mitigation in Three Industrial Use Cases. IEEE Access, 14, 25564-25576. https://doi.org/10.1109/access.2026.3659997

Vancouver

Tikka P, Karjalainen J, Alesani A, Goriachev V. Large Language Model Hallucination Mitigation in Three Industrial Use Cases. IEEE Access. 2026;14:25564-25576. doi:10.1109/access.2026.3659997

BibTeX
@article{Tikka2026,
  author = {Tikka, Petri and Karjalainen, Jaakko and Alesani, Andrea and Goriachev, Vladimir},
  title = {Large Language Model Hallucination Mitigation in Three Industrial Use Cases},
  journal = {IEEE Access},
  year = {2026},
  volume = {14},
  pages = {25564--25576},
  doi = {10.1109/access.2026.3659997}
}
Kostenlos starten →