A global manufacturer of aluminum die-cast components wanted to automate 100% visual inspection and standardize it across all plants. Previously, each site used its own defect definitions, creating inconsistent quality standards. With Maddox AI, the company introduced a centralized cloud platform where defect classes, reference standards (golden samples), and inspection rules are defined once and then deployed as identical inspection logic to every location. Rare defects discovered at one plant are immediately added to model training, improving detection performance across all sites. The result is consistent global quality standards, faster rollouts, and significantly less rework and false scrap.
Product
Industry
Total savings / year
Our customer manufactures identical automotive die-cast parts at four plants (Western Europe, Eastern Europe, and North America) running three shifts. Despite the same drawings and requirements, quality inspection was long handled site by site:
Each plant had its own reference standards, defect catalogs, and “OK/NOK” interpretations
Decisions depended heavily on the inspector, shift, lighting, and local experience
Rare defects (e.g., specific cold-shut patterns) were addressed locally only
Surface variation increased false scrap and drove recurring re-sorting and re-inspection effort
With the Maddox AI inspection system, the goal is to reliably detect key defects—especially flash/burr, cold shut (flow lines), and near-surface porosity (pitting/pores)—to reduce customer complaints and manual re-inspections and to establish consistent quality standards across all plants.
Initial inspection relied on traditional rule-based systems, often followed by manual checks across all three shifts to re-verify scrap and false scrap.
100% inspection runs robustly through normal production variation (surface changes, reflections, minor process drift), significantly reducing false scrap.
Four lines meant four different standards—defect definitions, thresholds, and training were locally developed.
All plants and lines use the same inspection logic (same classes, same decision rules), deployed as a unified model/configuration version.
Rare defect patterns triggered customer complaints because they weren’t “known” at other sites.
When a rare defect appears at one plant, it is labeled once and used for training—then all other plants benefit immediately from the updated model.
Initial inspection relied on traditional rule-based systems, often followed by manual checks across all three shifts to re-verify scrap and false scrap.
Four lines meant four different standards—defect definitions, thresholds, and training were locally developed.
Rare defect patterns triggered customer complaints because they weren’t “known” at other sites.
100% inspection runs robustly through normal production variation (surface changes, reflections, minor process drift), significantly reducing false scrap.
All plants and lines use the same inspection logic (same classes, same decision rules), deployed as a unified model/configuration version.
When a rare defect appears at one plant, it is labeled once and used for training—then all other plants benefit immediately from the updated model.
Less pseudo scrap
Inspection criteria worldwide
Parts inspected per year (across all plants)
Would you like to learn more about our solution? – Test for free now!
We are ISO 27001 certified.
