Outlier Screening<\/a> sorts can be built from a reference set of ‚acceptable‘ parts only.\u00a0 We don’t have to define, identify, or collect specific defective samples.\u00a0 Outlier screening has shown to be a very effective screening approach – reducing field failures for many customers across multiple industries.\u00a0 One customer had a ‚failure mode pareto chart‘ that showed that more than 50% of their field issues were casting shrink porosity.\u00a0 X-ray (and double X-ray, and more focused X-ray) were not effective at reducing the field failures, but PCRT Outlier Screening was.\u00a0 AND, the ‚other 50% of their field issues\u2019 were also dramatically reduced.\u00a0 Parts with ‚one-off‘ kinds of defective conditions can be hard to collect, particularly in a specific period of time.\u00a0 Sometimes, simulated versions of target defects can be modeled, but this brings other challenges and resource requirements to the project.<\/p>\n ZScore Plot: Training an Outlier Screening Sorting Module.<\/em><\/p>\n <\/p>\n We’ve also learned that just because ‚all of the failed parts had shrink porosity‘ (for example) doesn’t mean that ‚all shrink porosity will cause a part to fail‘.\u00a0 When a part is manufactured with a low % defect, it typically has ‚more differences‘ than just what we can see with our inspection.\u00a0 If we see shrink porosity, the material cooled at a different rate than ’normal‘, and this can lead to differences that we can’t readily see, like different material properties and different stress states than ’normal‘.\u00a0 These ‚extra differences‘ often seem to affect part performance, \u00a0and these extra differences certainly can make the part stand out more in resonance inspection – causing them to be a PCRT Outlier.\u00a0 Vibrant and its customers have been nearly as successful specifically targeting conditions like Shrink Porosity as we have just screening populations for outliers, and the start-up for an Outlier Screening application is much easier. (see case study here)<\/a><\/p>\n To get started, we collect data on parts that we make with our (in-control, capable) process.\u00a0 These might be parts that we are evaluating or qualifying with additional inspection or testing, if this is a new part that hasn\u2019t been made before.\u00a0 We can treat all parts as ‚acceptable‘, or we can come back later and update the part classification, based on the results of other inspections or testing.\u00a0 Then we can create 6-sigma, or 99% Tolerance Interval (or any other number) limits around the PCRT measures, to assure that future parts align with this original sample.\u00a0 If\/when parts fall outside of these limits, they can be investigated further, and used in future updates of the reference set \/ limits.\u00a0 When our other inspections have evaluated enough parts to produce a more robust sample of defective parts, we can use these to further enhance our inspection with ‚Targeted Defect‘ capabilities.<\/p>\n Talk with us about the options we have to help you get started with PCRT.<\/p>\n![\"\"](\"https:\/\/www.vibrantndt.com\/wp-content\/uploads\/2022\/05\/screen-shot-2022-05-26-at-10.43.01-am-300x152.png\")
VIPR Plot: Training a Targeted Defect Sorting Module<\/em><\/p>\n![\"\"](\"https:\/\/www.vibrantndt.com\/wp-content\/uploads\/2022\/05\/screen-shot-2022-05-26-at-10.43.22-am-300x223.png\")
<\/p>\n