Chatting with the Factory: Capturing Industrial Data Context for Flexible AI

EasyChair Preprint 16033

Abstract

Manufacturing companies generate large volumes of production and operational data, however many industrial AI solutions are limited to a narrow scope and built for a single predefined task, which makes the same underlying data hard to reuse when new questions arise. A key reason is that the knowledge needed to interpret production data, including machine semantics, process relationships, shift structures, and disturbance logs, is largely tacit and resides with experienced personnel rather than in any reusable form. This paper investigates how such contextual knowledge can be captured at low effort and shared across people and AI applications for analysis. Based on a case study at a heavy-duty automotive manufacturing site, we characterize the capture of industrial data context as a knowledge-capture problem and derive technical requirements for addressing it. We then propose a three-stage architecture in which experts' explanations of how the production data and operational records should be read are turned into structured form, held in a knowledge graph, and exposed to AI agents through a standardized interface. Three domain experts explained how the data should be read in roughly 3.5 hours of ordinary conversation, and an AI coding assistant turned those explanations into structured form under developer review, so no expert had to write a data model or a query. The approach is implemented in a machining center and evaluated via a user evaluation with manufacturing employees, a manual KPI cross-check, and a factual benchmark across several off-the-shelf language models. The results indicate that the captured context supports consistent KPI computation and reliable natural-language access for users with different backgrounds and analytical needs. The paper contributes a reframing of industrial data context and a proof-of-concept prototype of a context layer in industrial AI.

Keyphrases: Industrial AI, Industrial Data Contextualization, Manufacturing Analytics, industrial systems

@booklet{EasyChair:16033,
  author    = {Nicolas Leberruyer and Mats Ahlskog and Jessica Bruch and Sarmad Bashir and Natalie Agerskans and Stéphane Bouchereau},
  title     = {Chatting with the Factory: Capturing Industrial Data Context for Flexible AI},
  howpublished = {EasyChair Preprint 16033},
  year      = {EasyChair, 2026}}