INCORPORATING PRODUCT ROBUSTNESS LEVEL IN FIELD RETURN RATE PREDICTIONS

Tekcan A. T., Kahramanoglu G., GÜNDÜZALP M.

EKSPLOATACJA I NIEZAWODNOSC-MAINTENANCE AND RELIABILITY, vol.14, no.4, pp.327-332, 2012 (SCI-Expanded) identifier identifier

  • Publication Type: Article / Article
  • Volume: 14 Issue: 4
  • Publication Date: 2012
  • Journal Name: EKSPLOATACJA I NIEZAWODNOSC-MAINTENANCE AND RELIABILITY
  • Journal Indexes: Science Citation Index Expanded (SCI-EXPANDED), Scopus
  • Page Numbers: pp.327-332
  • Keywords: reliability and return rate estimation, artificial neural networks, defining different failure types, product maturity level, product robustness level, field failures, product level testing, board level testing, design quality, ACCELERATED LIFE TESTS, RELIABILITY-PREDICTION, SYSTEM RELIABILITY, MODELS
  • Dokuz Eylül University Affiliated: Yes

Abstract

Reliability and return rate prediction of products are traditionally achieved by using stress based standards and/or applying accelerated life tests. But frequently, predicted reliability and return rate values by using these methods differ from the field values. The primary reason for this is that products do not only fail due to the stress factors mentioned in the standards and/or used in accelerated life tests. There are additional failure factors, such as ESD, thermal shocks, voltage dips, interruptions and variations, quality factors, etc. These factors should also be considered in some way when predictions are made during the R&D phase. Therefore, a method should be used which considers such factors, thus increasing the accuracy of the reliability and return rate prediction. In this paper, we developed a parameter, which we call Robustness Level Factor, to incorporate such factors, and then we combined this parameter with traditional reliability prediction methods. Specifically, the approach takes into account qualitative reliability tests performed during the R&D stage and combines them with life tests by using Artificial Neural Networks (ANN). As a result, the approach gives more accurate predictions compared with traditional prediction methods. With this prediction model, we believe that analysts can determine the reliability and return rate of their products more accurately.