knowlEdge · Project

AI Tools That Let Factory Teams Train and Share Smart Manufacturing Models Across Sites

manufacturingPilotedTRL 7

Imagine every machine on your factory floor could learn from its own data — spotting defects, predicting breakdowns, optimizing output — but that intelligence stayed locked in one corner of the plant. knowlEdge built a system that lets AI models train right at the machine (the "edge"), test safely inside a digital copy of your production line, and then get shared or sold to other factories through a marketplace. Think of it like an app store, but instead of phone apps, you're downloading ready-trained AI brains for industrial equipment.

By the numbers

Manufacturing sectors validated (dairy, automotive plastics, gear machines)

Consortium partners across the project

Countries represented in the consortium

Major innovations delivered (edge AI, digital twin, data management, human-AI collaboration, model standardization, AI marketplace)

Total project deliverables produced

Industry partners in the consortium

SMEs among the consortium partners

The business problem

What needed solving

Most manufacturers know AI could improve their production — catching defects earlier, predicting machine failures, optimizing processes. But the barriers are brutal: you need data scientists you can't hire, your production data is sensitive and can't just be shipped to the cloud, and every AI model built for one line or one plant is useless everywhere else. The result is that factory AI stays stuck in expensive one-off proof-of-concepts that never scale beyond a single use case.

The solution

What was built

The project built an edge-to-cloud AI platform with 6 core components: AI services that run directly on factory equipment, a digital twin for safe testing, a privacy-preserving data management layer, human-AI collaboration tools for non-technical operators, standardized AI model packaging for cross-site reuse, and a marketplace to trade trained AI models. All components were validated through 3 full industrial pilots at Parmalat, Kautex Textron, and Bonfiglioli, each with documented final assessments across 49 total deliverables.

Audience

Who needs this

Food and dairy processors running continuous production lines with quality variabilityAutomotive component manufacturers (especially plastic injection molding) needing better defect detectionIndustrial machinery manufacturers looking to optimize assembly and reduce scrapMulti-plant manufacturing groups wanting to share AI models across locationsProduction managers evaluating Industry 4.0 tools but lacking in-house data science teams

Business applications

Who can put this to work

Food & Dairy Processing

enterprise

Target: Mid-to-large dairy and food manufacturers

If you are a dairy processor dealing with inconsistent product quality or unplanned line stoppages — this project developed AI services tested directly at Parmalat's milk production facilities. The system trains models on live production data at the machine level, uses a digital twin to validate changes before they hit the real line, and lets your domain experts guide the AI without needing data science degrees. Pilot results with final assessment are available from the Parmalat milk industry pilot.

Automotive Parts Manufacturing

mid-size

Target: Plastic injection molding and automotive component suppliers

If you are an automotive parts manufacturer struggling with defect detection or process variability in plastic molding — this project ran a full pilot at Kautex Textron, a major plastic parts supplier for the car industry. The AI models run at the edge (right on the production equipment), reducing latency and keeping sensitive production data on-site. A final assessment of the Kautex Textron pilot confirmed results across their plastic parts production.

Industrial Machinery & Gearbox Manufacturing

enterprise

Target: Manufacturers of mechanical drive systems, gearboxes, and power transmission equipment

If you are a gear machine or power transmission manufacturer facing pressure to reduce scrap rates and improve process consistency — this project completed a pilot at Bonfiglioli, a global gear machine manufacturer. The system enables human-AI collaboration where your experienced operators inject their know-how into the AI, and the digital twin lets you test optimizations risk-free before deploying them on the shop floor.

Frequently asked

Quick answers

What would it cost to implement this in our factory?

The project did not publish pricing or implementation cost data. Since knowlEdge was a Research and Innovation Action (RIA), the outputs are pre-commercial. Any deployment would likely require a customization engagement with one or more consortium partners such as VTT or the industrial partners involved in the pilots.

Can this scale to a multi-site manufacturing operation?

Yes — the architecture was specifically designed for edge-to-cloud distribution, meaning AI models can train locally at each site and share knowledge across locations. The AI model marketplace and standardization mechanisms were built to enable exactly this kind of cross-site, cross-company scaling. The system was validated across 3 different manufacturing sectors.

Who owns the IP and how can we license it?

As an EU-funded RIA project with 17 partners, IP is distributed among consortium members. VTT (Finland) coordinated the project. Licensing arrangements would need to be negotiated with the relevant partners depending on which components you need. The AI marketplace concept itself may offer a commercial licensing path.

How does this handle data privacy and security on the shop floor?

The project built a dedicated data management layer running from edge to cloud that specifically addresses data quality, privacy, and confidentiality. They call it a 'data safe fog continuum' — in practice, it means sensitive production data can stay on your premises while only sharing AI model parameters (not raw data) with the cloud or marketplace.

Do we need data scientists on staff to use this?

No — one of the 6 core innovations is Human-AI Collaboration and Domain Knowledge Fusion tools. These are designed for domain experts (your experienced machine operators and engineers) to inject their knowledge into the system without coding. The AI then uses that input to adapt automatically to production changes.

What is the timeline from first contact to a working system?

The project ran from January 2021 to March 2024 and produced 49 deliverables including final pilot assessments. The technology has been validated in 3 industrial settings but is not yet a turnkey product. Based on available project data, expect a pilot engagement of several months to adapt the platform to your specific production environment.

Does it comply with EU regulations on AI and data?

The project was designed with EU data governance principles in mind, including data privacy and confidentiality built into the edge-to-cloud architecture. It also emphasizes human-in-the-loop AI, which aligns with the EU AI Act's requirements for human oversight in industrial applications.

Consortium

Who built it

The knowlEdge consortium is unusually strong for business adoption: 17 partners from 8 countries, with an industry ratio of 47% (8 industrial partners out of 17). This is well above the typical EU research project. It includes 4 SMEs, 5 research organizations, and 3 universities, coordinated by VTT — Finland's leading applied research center with a strong track record of turning research into industrial tools. The presence of major end-users like Parmalat, Kautex Textron, and Bonfiglioli as pilot hosts means the technology was shaped by real production requirements, not just academic ambition. The geographic spread across Belgium, Germany, Greece, Spain, Finland, Italy, Slovakia, and the UK provides a broad European market entry point.

TEKNOLOGIAN TUTKIMUSKESKUS VTT OYCoordinator · FI
UNIVERSITAET MUENSTERparticipant · DE
NEXTWORKS SRLparticipant · IT
ETHNIKO KENTRO EREVNAS KAI TECHNOLOGIKIS ANAPTYXISparticipant · EL
PARMALAT SPAparticipant · IT
FERNUNIVERSITAT IN HAGENparticipant · DE
INFORMATION CATALYST SLthirdparty · ES
INFORMATION CATALYST FOR ENTERPRISE LTDparticipant · UK
DIMECC OYparticipant · FI
UNIVERSITAT POLITECNICA DE CATALUNYAparticipant · ES
FONDAZIONE LINKS - LEADING INNOVATION & KNOWLEDGE FOR SOCIETYparticipant · IT
BARCELONA SUPERCOMPUTING CENTER CENTRO NACIONAL DE SUPERCOMPUTACIONparticipant · ES

How to reach the team

VTT Technical Research Centre of Finland — search for knowlEdge project lead at VTT.fi

Order a report on this consortium See TEKNOLOGIAN TUTKIMUSKESKUS VTT OY profile →

Next steps

Talk to the team behind this work.

Want to explore how knowlEdge AI tools could fit your production line? SciTransfer can connect you with the right consortium partner for your sector and use case.

Order a report on this topic More in Manufacturing & Industry 4.0