Confidence assessment of quality prediction from process measurement in sequential manufacturing processes

Nong Ye, Qiu Zhong, Gregory E. Rahn

Research output: Contribution to journalArticlepeer-review

6 Scopus citations


Sequential manufacturing processes are common in many manufacturing industries. We can use an artificial neural network (ANN) technique to build a model of manufacturing processes to capture the relationship between process measures and production unit quality. Using the ANN model, the quality value of a production unit can be predicted for its process measures. In addition to the predicted quality value, we need to assess our confidence in the predicted quality value through a confidence interval (prediction interval). Little work exists to assess the confidence in ANN outputs. This paper presents a comparative study of several methods in assessing the confidence in ANN outputs for quality prediction. We investigate two new methods: the variance estimation method and the distribution estimation method, in comparison with two existing methods: the error estimation method and the linear regression method. With respect to prediction accuracy and power, our results show that the distribution estimation method appears very promising among the four methods. However, the distribution estimation method requires much computation and storage during training and prediction, which may create difficulty in real-time quality monitoring and control. For the consideration of computation and storage requirement, the variance estimation method provides an acceptable solution. The error estimation method produces the worst performance among the four methods.

Original languageEnglish (US)
Pages (from-to)177-184
Number of pages8
JournalIEEE Transactions on Electronics Packaging Manufacturing
Issue number3
StatePublished - Jul 2000

ASJC Scopus subject areas

  • Industrial and Manufacturing Engineering
  • Electrical and Electronic Engineering


Dive into the research topics of 'Confidence assessment of quality prediction from process measurement in sequential manufacturing processes'. Together they form a unique fingerprint.

Cite this