A Feature‐Enhanced Artificial Neural Network‐Assisted Adaptive Truncated EWMA Control Chart for Robust Monitoring of Contaminated Processes

The use of conventional memory‐type control charts can be significantly hindered by the presence of outliers and contaminated observations, as these can lead to excessive false alarms and be difficult to detect true process shifts. To tackle the aforementioned problems, a Feature‐Enhanced Artificial Neural Network‐Assisted Adaptive Truncated Exponentially Weighted Moving Average (FEA‐ATEWMA) control chart for robust process mean monitoring is proposed. The proposed framework is designed to extract 5 informative features from each observation and use a feed‐forward artificial neural network to create an observation‐specific reliability weight. This weight also controls the observation weighting, the adaptive truncation, and the EWMA smoothing parameter, thus creating an integrated way to reduce the impact of contaminated observations, yet maintain sensitivity to process changes. The proposed chart performance is assessed by performing a large number of Monte Carlo simulations with different levels of contamination proportions, outlier sizes, and mean shifts and compared to the classical Truncated Adaptive EWMA (TAE) chart. The simulation results show that the proposed chart performs well and is stable in the in‐control state, is robust to contaminated state, and has competitive detection abilities for small to moderate process shifts. A real data application also demonstrates the application of the suggested monitoring scheme for industrial contaminated processes.