PortugalManufacturing2020 – ongoingCorticeira Amorim
Deep-learning classification of cork stoppers from photo and X-ray images
Amorim said deep learning raised the repeatability of its cork stopper classification from 60% to 80%
Problem
Amorim Cork's factories produce about 25 million stoppers a day. Its older classification code ran to more than 10,000 lines and relied on 70 adjustable parameters, which made it dependent on scarce specialist staff.
Approach
Amorim introduced machine learning around 2020, using photographs and X-ray images. It then worked with Sentinel, a start-up from Guimarães, on a deep-learning model that continues earlier work with 3Control. The aim was to need only two or three parameter changes. Each stopper is now checked four times, giving 12 decisions per unit.
Results
Takeaway
Amorim measured the change as decision repeatability on the same image and gave no figure for defects reaching customers. It stated the goal of cutting tuning to two or three parameters but did not say whether it got there.
Categories
- Size
- Large enterprise
- Industry
- Manufacturing
- Business function
- Production and quality
- Innovation type
- Process innovation
- AI technology
- Computer visionPredictive analytics and ML
- Adoption
- Partner co-development
- Evidence
- Company's own claim
Sources
Source summaries
Amorim describes its move from rule-based stopper classification to deep learning built with the start-up Sentinel. It gives repeatability figures for the machine and for human graders. The company did not explain how the 80% and 90% figures relate.
Full details are in the original sources.
