Robust process capability performance : An interpretation of key indices from a nonparametric viewpoint

Purpose – The purpose of this paper is to propose a set of process capability indices (PCIs) which are based on robust and agile statistics such that they may be applicable irrespective of the process status. Design/methodology/approach – The four popular PCIs – Cp, Cpk, Cpm and Cpmk – are reconstructed to improve location and dispersion predictions by introducing robust estimators such as the median and the interquartile range. The proposed PCIs are sequentially evaluated in partitioned regions where fluctuations are inspected to be not significant. The runs test playing the role of a detector permits marking those regions between two consecutive appearances of causes that disrupt data randomness. Wilcoxon's one-sample test is utilized to approximate PCI's central tendency and its confidence interval across all formed partitions. Findings – The Cpmk depicted the most conservative view of the process status when tracking the magnesium content in a showcased aluminum manufacturing paradigm. Cp and Cpk were benchmarked with controlled random data. It was found that the proposed set of robust PCIs are substantially less prone to false alarm in predicting non-conforming units in comparison to the regular PCIs. Originality/value – The recommended method for estimating PCIs is purely distribution-free and thus deployable at any process maturity level. The advantageous approach defends vigorously against the influence of intruding sources of unknown and unknowable variability. Therefore, the predicament here is to protect the monitoring indicators from unforeseen data instability and breakdown, which are conspicuous in wreaking havoc in managerial decisions.