A Novel Two‐Stage Cluster‐Based Robust Standard Deviation Control Chart

ABSTRACT This study proposes a robust standard deviation estimator for Phase I control charts. The method uses a two‐stage cluster‐based outlier screening mechanism that combines K‐means clustering with silhouette coefficient analysis. The first stage screens at the group level, and the second stage screens individual observations. Both stages include mechanisms to protect extreme values. We evaluate the method through Monte Carlo simulations under four contamination scenarios: symmetric variance, asymmetric variance, localized variance, and diffuse mean disturbances. Results show that and perform best in Phase I parameter estimation. These variants also show strong robustness in Phase II control chart implementation. The proposed control charts maintain average run length (ARL) values close to nominal specifications under contamination. They show superior detection capabilities for process shifts compared to established robust estimators.