Enhancing Manufacturing Quality: A Multivariate Framework for Robust Outlier Detection in Part Average Testing

Abstract

This paper investigates the development and efficacy of Part Average Testing (PAT) in identifying outliers within manufactured components, which account for between 1 and 10% of production and are critical to ensuring product quality and reliability. The paper proposes an integrated testing framework that acknowledges the interdependencies between various manufacturing parameters, moving away from traditional sequential testing methods towards a robust, multivariate approach. This approach leverages statistical metrics such as mean and standard deviation, advocating for robust measures like median and inter-quartile range to counter the influence of outliers on these statistical assumptions. Additionally, the study critically evaluates the application of these methodologies across different manufacturing sectors, using a Monte Carlo simulation to empirically compare the effectiveness of standard and robust outlier detection methods across various probability distributions. The findings underscore the superiority of robust methods, particularly in handling data distributions with extreme values or skewness.