Contact us

Other

AI Quality Prediction on Equipment Data — Injection, Heat Treatment and Welding

A production process that relied on experienced operators was moved onto lot-level datasets and AI quality prediction. Across three processes, defect loss cost fell by ₩5.6M and good-part output rose 13%.

Industry
Other
Solutions applied
a2b2laba2lab

  1. Problem

    Real-time monitoring was in place, but there was no way to analyze defect causes or predict quality. Because production depended on the knowledge of experienced operators, quality varied with the individual, and equipment setting errors during staff changes and generational turnover produced defects. Data was being collected, but with no infrastructure to process and extend it into AI, management could not build strategy on it.

  2. Data collected

    Equipment, production and quality data from three processes — injection, heat treatment and welding — were linked into lot-level datasets.

  3. Modules applied

    B²LAB processed and linked the data, while A² produced defect root cause, quality prediction and equipment optimization models. Quality prediction was implemented inside the existing MES, with real-time alarms when equipment deviates from operating standards.

  4. Result

    Defect causes can now be analyzed against equipment operating conditions. Judgments that rested on operator experience now sit on data, so the standard holds when staff change. Because results appear inside the existing MES, no separate system has to be opened on the floor.

Data present, but unusable

This site was not short of data. Real-time monitoring screens were already in place. The data simply was not in an analyzable form.

Equipment data accumulated on the equipment side, production data on the production side, quality results on the quality side. Asking why a particular lot went wrong meant digging through three places and matching them up by hand.

Linked at lot level

The first step was connecting the three streams at lot level. From there, which equipment conditions produced which quality outcome could be read on a single line.

Defect root cause, quality prediction and equipment optimization models went on top. The results were placed inside the existing MES rather than in a new system — screens people already use are the ones that actually get used.

The turnover problem

Each time operators changed, the equipment setting baseline moved with them, and that difference kept turning into defects. Real-time alarms on deviation from operating standards let less experienced operators work within the baseline.

View allContact us