abstract
- Multivariate control charts are essential tools for quality monitoring in modern manufacturing and service processes, where several interrelated variables influence system performance. However, traditional assumptions of multivariate normality and known process parameters are rarely met in practice, which favors the use of nonparametric control charts. Additionally, the influence of the reference sample on the null distribution of the monitoring statistic ¿ particularly when Phase I sample sizes are small ¿ can lead to significant distortions in the chart's in-control behavior. In this research, we propose the first nonparametric multivariate CUSUM control chart that incorporates cautious learning and provides guaranteed in-control performance. The proposed chart is based on sequential ranks computed over distance metrics and does not rely on any distributional assumptions. It achieves robust monitoring by cautiously updating the reference sample, which helps stabilize the null distribution of the monitoring statistic and control the false alarm rate. Simulation studies conducted using R demonstrate that our approach performs effectively in detecting moderate to large location shifts, even with limited Phase I data, making it a practical solution for real-world applications.