Industrial application of enhanced exponentially and homogenously double-weighted moving average control charts for outliers' detection

Abstract

Abstract When control charts monitor processes with unknown parameters, preliminary samples are required to estimate the process parameters. These estimates are instrumental to constructing the control charts limits. The variability in the samples used by practitioners and the presence of extreme values in those samples affect the efficiency of the resulting charts in the monitoring stage. This study aims to produce efficient and sensitive double-weighted moving average charts, exponentially (DEWMA) and homogenously (DHWMA) in the presence of outliers for location monitoring. The study enhances the DEWMA and DHWMA charts with outliers’ detection models. It evaluates the effect of practitioners’ variabilities of sample sizes used for estimation in phase-I, studies the negative impacts of outliers present in such samples on DEWMA and DHWMA charts, and then incorporates some outliers’ detection models in the structures of the two charts, as a remedy to outliers’ effect. The study employed Monte-Carlo simulations to evaluate the effectiveness of the proposed control charts based on outliers’ detection models with their run length properties as performance measures. The study concludes by applying the proposed charts on real-life data extracted from the carbon-fiber manufacturing industry and comparing the DEWMA and DHWMA charts. The results showed that the enhanced DEWMA and DHWMA charts for outliers’ detection are quicker and more sensitive in detecting anomalies than their counterparts. The results further proved the proposed charts require less amount of sample data to perform efficiently. The DEWMA chart also performs better than DHWMA in most of the scenarios. We recommend the proposed control charts incorporating outliers’ detection models to monitor processes prone to outliers efficiently and adequately.