Nonparametric SPC: Solving complex industrial monitoring tasks with low computational cost
Diego Carvalho do Nascimento, Anderson O. Fonseca, Marcelo B. Fonsêca, Moisés Rodrigues Silva, Paulo H. Ferreira · Array · 2026
Quality assurance is essential for an organization’s success today, but it is also a great challenge in the Big Data era. Statistical Process Control (SPC), focusing on control charts of monitoring variability, is a fundamental tool, though outliers, skewed information, and temporal dependence may influence the ones based on the process average. Nonparametric techniques stand out when it is needed to represent the process dynamism with robustness and derive statistical tests based on, e.g., the median instead of the mean. This work approaches the univariate Nonparametric Process Monitoring, such as Sign and Signed-Rank statistics, which are useful for run-type signaling, monitoring spread or joint (considering the known parameter of the specified tolerant interval, case K, or unknown parameter of in-control, case U). This study illustrates some mining industry monitoring tasks through nonparametric control charts, allowing the suggestion of a preventive maintenance plan for different scenarios. Due to the scarcity of implementations of nonparametric methods in the context of SPC, this study developed a new package for the R software, named npSPC , which shows extremely low computational cost. The package provides nonparametric control charts (alternative to the mean parametric charts of Shewhart, CuSum, and EWMA), enabling statistical monitoring in a wider range of problems under minimal assumptions (such as not the response variable being continuous or symmetric) with fast inferences.