When you adopt a non convention reliability metric

414 views

Published on

Most commercial enterprises want some meaningful but simple metrics to measure, track and report their part and/or product reliability, which link well to their business and/or engineering processes. An internal reliability measurement may sometimes be "defined" for that purposes. However, without careful consideration initially, the adopted non-conventional reliability metric may be lack of scientific sense, causing a lot of confusion and misunderstanding in calculation, reporting and communication. This talk presents you a real case study, which turns out to be a good example of lesson learned.

Published in: Technology, Business
0 Comments
0 Likes
Statistics
Notes
  • Be the first to comment

  • Be the first to like this

No Downloads
Views
Total views
414
On SlideShare
0
From Embeds
0
Number of Embeds
20
Actions
Shares
0
Downloads
16
Comments
0
Likes
0
Embeds 0
No embeds

No notes for slide

When you adopt a non convention reliability metric

  1. 1. When You Adopt a Non‐ convention Reliability  Metric … (当你采用自定义的可靠性度 量 …) Dr. Wendai Wang (汪文岱) ©2011 ASQ & Presentation Wang Presented live on Mar 09th, 2011http://reliabilitycalendar.org/The_Reliability_Calendar/Webinars_‐_Chinese/Webinars_‐_Chinese.html
  2. 2. ASQ Reliability Division  Chinese Webinar  Series One of the monthly webinars  on topics of interest to  reliability engineers. To view recorded webinar (available to ASQ Reliability  Division members only) visit asq.org/reliability To sign up for the free and available to anyone live  webinars visit reliabilitycalendar.org and select English  Webinars to find links to register for upcoming eventshttp://reliabilitycalendar.org/The_Reliability_Calendar/Webinars_‐_Chinese/Webinars_‐_Chinese.html
  3. 3. When You Adopt a Non-convention Reliability Metric …当你采用自定义的可靠性度量 … 汪文岱博士 可靠性工程总监 GreenVolts, Inc. Fremont, California wendaiw@gmail.com
  4. 4.  Case Study: 案例研究 Failure Data Analysis: 失效数据分析 Field Reliability: 应用可靠性 Lesson Learned: 经验教训 Reliability Measures: 可靠性度量 Warranty Period: 保修期Key Words/Terminologies关键词/术语
  5. 5.  Reliability Metrics – theoretically & practically • define / represent; • be able to be designed in; • report out the reliability of your systems, subsystems and parts.Design in and Report out
  6. 6.  Some well-defined, commonly-used reliability indices • Reliability • Failure Rate • MTTF / MTBF • …Convention Reliability Metrics
  7. 7.  Annualized Failure Rate (AFR) to measure the parts reliability – What ?Annual Failure Rate ?
  8. 8.  Have been used in high-tech industries. Could be just a Failure Rate? But confused with the word “annualized”.Failure Rate ?
  9. 9.  Some studies / publications • “AFR: Problems of Definition, Calculation and Measurement in a Commercial Environment” by Jon Elerath • “The iFR Method for Early Prediction of Annualized Failure Rate in Fielded Products” by Bill LycetteResearch
  10. 10.  It’s unconventional • Different definitions (self defined) • No standard calculation method • Likely result in significantly different estimates by customers and supplies It could be • a failure percentage based or • a time based rate.Non-convention Metric
  11. 11.  The AFR definition we used: Total Failures from Units in the Denominator AFR = Total Units Shipped in past Warranty PeriodDefinition
  12. 12.  Apparently it’s simple • Easy to obtain (from data) • Seems easy to understand o Failure percentage based Seems meaningful for business process Seems well defined (in calculation) Not a failure rate !Good for Business
  13. 13.  Lack of theoretical base • What does it really mean? • How to convert the AFR to other reliability quantities? • How to quantify the confidence bounds? Confusion in calculations  How difference between methods? Lots of misunderstanding • Failure rateDisadvantages
  14. 14.  Always see reliability (in term of AFR) improvement for design changes • Not always match with reality • Not see reliability improvement in whole parts pool Discrepancy in an AFR value between our estimate supplier’s estimate.Case Study
  15. 15.  The AFR value highly depends on the interval over which the data is collected. • Warranty period is in the definition. • It may take as much as whole period before the improvement (or degradation) is observed. Someone couldn’t wait and reported out AFR values based on a short period (available data) !Misunderstand in Calculation
  16. 16.  The AFR we defined is not the Failure Rate ! BUT suppliers thought: our AFR = the Failure Rate. AFR numbers arrived at widely different figures even using the same data.Misunderstand in Terminology
  17. 17.  What’s the term we really defined? Tota lFa i l ures fromUni tsi nthe Denomi na to r AFR  Tota lUni tsShpi ppednPa s tWa rra ntyPeri od i N n  F t  i   i 1 N N Theoretically, it’s an average value of the unreliability function over the warranty period *. 1 T AFR   F (t )dt T 0The Language of Engineering is Math. * Depends on the real calculation method.
  18. 18.  Precisely, it’s an estimator of an average value of the unreliability function over the period. Probability of Failure AFR Value with 2- Part’s year data unreliability curve AFR Value with 1- Years year data 0 1 2 3 4Different Data Intervals
  19. 19.  Numerical example* • Failure Rate (constant) = 0.1 failures/year • New Order = 1,000 parts/year • AFR = 5.25% (using Year 2010 data only) • AFR = 9.56% (using 2-year data)Simple Explanation
  20. 20.  A percentage based metric A non-parametric estimate • matching moment estimation (MME) A better estimate (MLE) could be introduced • fully use information from the data • independent of data interval • be able to establish the confidence boundsBuild up Theoretical Base
  21. 21.  Be able to establish relationship between AFR and other reliability quantities. Example: Part - PN 123456 • 489 units shipped in past 2 years. • There were 56 failures among them. • AFR_MME = 56 / 489 = 11.5%, which can be converted to • FR = 13.47 FPMHConverted to Other Quantities
  22. 22.  Different estimate methods • AFR_MME = 11.5%  FR = 13.47 FPMH • AFR_MLE = 14.1%  FR = 17.69 FPMH Failure Rate directly from data • FR = 17.88 FPMH Confidence bounds for AFR (at 90% CL) • AFR_LL = 10.9% • AFR_UL = 17.8%Converted to Other Quantities
  23. 23.  Non-conventional reliability metrics were wisely defined for good reasons. A thorough study is often needed to make good sense out of it. Communication is imperative. Education still is a crucial task for reliability engineering.Summary

×