Batch Variability in Accelerated-Degradation Testing

• Autores: Ming Li, Necip Doganaksoy
• Localización: Journal of quality technology: A quarterly journal of methods applications and related topics, ISSN 0022-4065, Vol. 46, Nº. 2, 2014, págs. 171-180
• Idioma: inglés
• Enlaces
  • Texto completo
• Resumen

  • Problem: The study that motivated this article concerns the development of a new seal to be used in a high-pressure application. A weakened seal would not only adversely affect the overall device performance but it could also lead to potential safety concerns. The seal was designed to perform for 10 years in the application environment. The development team had undertaken a temperature accelerated test to study the degradation of seal strength. The ad hoc procedure used to fit an engineering model to the data did not readily allow estimation of lower quantiles of the strength distribution and obtaining statistical confidence bounds on quantities of interest. The authors were approached for advice on statistical analysis and modeling of the data in order to guide engineering and management decision making. Approach: In the initial analysis of the data, the authors used a standard statistical model based on the engineering relationship for seal strength and temperature. The analysis of residuals from the maximum-likelihood fit of the model revealed an additional source of variation in the data that had not been accounted for. This finding cast doubt on the inferences that resulted from the standard statistical model due to the inadequacy of the model in representing the data. The standard model was extended to accommodate this new source of variability through nonlinear mixed effects modeling. Results: The extended model was shown to adequately account for the excess source of variation in the data. The inferences based on this model showed greater uncertainty and highlighted the need to identify and mitigate the root cause of the variation. A narrow consulting question led to a significant discovery that paved the way for a meaningful engagement with the seal team. The case study shows the importance of fully understanding the experimental process and the usefulness of incisive residuals plots in statistical analysis and modeling.