ABSTRACT Classical memory‐type control charts such as Adaptive EWMA (AE) and Truncated Adaptive EWMA (TAE) are usually formulated on the basis of normality. Even though these charts can be used on non‐normal data and work round the impact of extreme data by truncation, this does not necessarily work well in the situation where there are real process changes. In order to overcome this drawback, the current paper will present a new Artificial Neural Network (ANN)‐supported Truncated Adaptive EWMA (ANN‐TAE) control chart where anomaly scores produced by an ANN‐based model will be employed in order to adjust the weight of incoming observations prior to the subsequent update of the EWMA statistic. This adaptive weighting system provides resilience to outliers whilst being sensitive to changes in the process mean that matter. The efficiency of the proposed chart is evaluated with Monte Carlo simulations and different contamination levels and magnitude of the shift. These findings indicate that the ANN‐TAE chart always offers more stable average run length behavior, low false‐alarm rates and moderate process shifts are detected faster by the ML‐TAE chart than the TAE charts. Moreover, the proposed ANN‐based monitoring framework has its applicability, which are supported by a real‐life case study of a polluted environment of the process itself.
Zaagan et al. (Fri,) studied this question.