EN
Back to the archive

The encyclopedia · Engineering & Operations · Technical decision · 2023–2026

HD Hyundai turns master welders' know-how into robots that novices run

HD Hyundai encoded master welders' know-how into robots that beginners supervise; LUG output jumps from 100 to 440 parts a day with fewer workers.

HD Hyundai Heavy Industries

The solution

Shipbuilding's bottleneck is skilled welders: thick steel plate joints demand years of judgment. HD Hyundai Heavy Industries started in 2023 deploying welding robots that carry the know-how of masters with 20+ years' experience.

A worker with about one year on the job now monitors two robots, with six or more planned via autonomous electric-rail robots by October 2026. The robots recognize the weld target and correct position in real time; data is logged for quality control.

In LUG production — the parts used to lift hull blocks — manual teams of six made 100 pieces a day; the AI-robot system is expected to reach 440 a day with one or two workers, roughly what two 5-to-10-year veterans used to do, with about 70% higher productivity.

In August 2026 HD Hyundai began a pilot at Huntington Ingalls' Ingalls shipyard in Mississippi, exporting the system to US naval shipbuilding.

Why it worked

  • Capturing tacit expertise removes the dependency on scarce veterans.
  • One supervisor operating many robots changes the labor economics of welding.
  • Data logging turns welding into a measurable, improvable process.
  • Digital-twin quality checks let managers inspect without entering the yard.
What it achievedBottle the expert; let beginners superviseclever

What can be applied

When a craft bottleneck is human experience, don't replace the worker — encode the expert's tacit rules into machines that anyone can supervise.

Aftermath

The system is being enhanced so AI reads drawings and generates weld conditions and sequences, with people managing the whole process; HD Hyundai is expanding the pilot's scope with Huntington Ingalls.

Sources

spotted an error? The archive wants to know.

Related cases