On developing robust MEWMA charts with Stahel–Donoho estimation: applications in chemical engineering

Efficient monitoring of chemical engineering processes such as Continuous Stirred Tank Reactors (CSTR) and Hydrogen Production Processes (HPP) is critical for ensuring product quality, operational stability, and safety. However, the multivariate nature of these systems and the frequent occurrence of outliers often challenge conventional statistical process control tools. This study introduces a robust Multivariate Exponentially Weighted Moving Average (MEWMA) control chart that incorporates the Stahel Donoho Robust Estimator (SDRE) to overcome these limitations. The proposed chart operates across two phases: in Phase I, the SDRE is applied to historical process data to compute robust estimates of the mean vector and variance–covariance matrix, producing contamination resistant control limits; in Phase II, these estimates are deployed for real-time online monitoring to detect process shifts reliably. By down weighting atypical observations without discarding information, the proposed chart provides stable parameter estimation, enhances sensitivity to small and moderate shifts, and maintains false alarm control even under contaminated data. Extensive simulation experiments and run-length based performance metrics confirm its superiority over traditional MEWMA, Hotelling T²[jls-end-space/], and existing robust charting methods across both stationary and non-stationary process environments. Applications to real-world CSTR and HPP datasets further demonstrate the effectiveness of the chart in recovering accuracy losses caused by outliers and enabling reliable early fault detection. The findings establish the SDRE based MEWMA chart as a powerful tool for robust multivariate monitoring in chemical engineering, with potential to improve process reliability, stability, and overall product quality.