Problem
Eight people across day and night shifts were assigned to a single machine. Everything from loading work-in-progress to stacking was manual, keeping productivity low and generating human error. When defects occurred, neither the count nor the type was captured as data — operators entered it by hand — so there was no basis for root cause improvement or prevention.
Data collected
Equipment, production and quality data from the vertical injection process were collected, standardized and linked to the existing MES. Defect information previously entered by hand was switched to automatic capture.
Modules applied
C² (automatic collection) and D² (standardization and data lake storage) built the foundation, with A² providing the AI analysis on top. Preprocessing, variable correlation and quality prediction modules were deployed, together with a 3D monitoring system for the vertical injection process.
Result
Inspection automation changed the staffing structure so that one person covers two machines. Defect data now accumulates automatically, giving process improvement a factual basis, and 3D monitoring makes process status visible at a glance.
- 50.2s → 35.11sCycle time
- 4.8 → 10.6 (+121%)Output per minute
- 11% → 2%Quality defect rate
A process that needed a lot of people
The problem here predated AI. Eight people per machine, across day and night shifts, were loading, removing and stacking work-in-progress by hand.
The bigger issue was that nothing was being recorded. When a defect appeared, an operator wrote it down, which meant even the defect count was unreliable. There was no data to base improvement on.
We kept the order
First we made the data accumulate automatically (C², D²), then connected it to the existing MES, and only then added quality prediction (A²). Attaching AI before the data foundation would have produced results nobody could trust.
Introducing the inspection automation also changed the staffing. What had been several people per machine became one person covering two machines.