Image: MADE.
Machine works Johannes Pedersen Maskinfabrik and Deepvis have run a joint project to bring AI vision to weld inspection — and the result is a bottleneck removed and a door opened to bigger orders.
The problem: 5,000 parts, inspected by eye
Johannes Pedersen Maskinfabrik supplies parts including bicycle holders for Copenhagen’s S-trains. Every weld is inspected manually today, and a single bicycle holder can carry up to 12 welds.
“The operator gets very tired if he has to inspect 5,000 parts and just as many welds on the plate.”
— Falko Thuesen, Purchasing Manager, Johannes Pedersen Maskinfabrik
Large orders were arriving faster than manual inspection could clear them.
How the two met
Falko Thuesen met Deepvis at a network meeting through AddSmart, the EU co-financed digitalisation hub in North Jutland. The technology had not yet been tested in the metals industry, so the two agreed to apply for a MADE Kickstart project — the scheme that supports companies testing new technology.
“We have documented the effect we have had with many of our customers in the plastics industry. Now we are trying to transfer the same model know-how to the metals industry. If we can prove it does the same — and get it approved for visual inspection — it opens a new market where we can serve the metals industry.”
— Carl Gustav Fagerlund, co-founder of Deepvis
The vision system sits in a black box that travels back and forth on a rail; a hole in the box lets the carriage pass over the tube.
What the numbers look like
Trained on just 25 metal tubes, the system reaches confidence levels around 95% on the majority of assessments — its certainty that a weld is of approved quality. Lower confidence levels, down to roughly 75%, are used to surface the handful of critical welds where the uncertainty is greatest, so a human looks exactly where it matters.
With production running at 400–500 tubes a month, the incoming images were expected to sharpen the model further within the first 14 days.
Part of why it works with so little data is that the model learns what good looks like, rather than learning to recognise faults:
“With many varying defects — when you weld, for instance, no two welds look alike. It is a huge advantage that you simply define the good parts, and do not have to keep configuring every new defect.”
— Carl Gustav Fagerlund
Toward accreditation
The system is under accredited assessment at Nordjysk Svejse Kontrol A/S, which has already given a positive indication that it will be approved for inspection at the same level as a certified welding inspector.
“The advantage of automating quality control is that you secure consistent, good quality. We can document it, and we can take time pressure off our welding coordinators and improve the working environment, because he no longer has the very demanding job of looking extremely closely for very small holes in a weld.”
— Falko Thuesen
And on what it unlocks commercially:
“We can take in larger orders, because we no longer have a bottleneck called weld coordinates.”
— Falko Thuesen
What comes next
The next steps are testing welds in bent tubes, and finding a way to test for faults where the eye cannot reach — inside the weld itself — for an even stronger quality test.
About Johannes Pedersen Maskinfabrik. Founded in 1975 by Johannes Pedersen, headquartered in Viborg, Denmark. The company produces body parts, exhaust systems and oil and fuel tanks for the automotive industry, including for classic cars, and manufactures components for the rail industry. It is part of JP Group Holding a/s and operates globally with advanced machining and assembly facilities.
About AddSmart. A European digitalisation hub helping SMEs become more digital and innovative, started in October 2022 and co-financed by the EU. Partners include Erhvervshus Nordjylland, MADE, Digital Lead, UCN, TechCollege, NordDanmarks EU-Kontor, Aalborg University and Enterprise Europe Network.
Source: MADE — “AI-løsning fjerner flaskehals på maskinfabrik – kan nu tage større ordrer”, 3 February 2026. Also covered by Erhvervshus Nordjylland. Quotes translated from Danish.
