An algorithmic approach to outlier detection and parameter estimation in Phase I for designing Phase II EWMA control chart

TESTİK M. C., Kara O., Knoth S.

COMPUTERS & INDUSTRIAL ENGINEERING, vol.144, 2020 (SCI-Expanded) identifier identifier

  • Publication Type: Article / Article
  • Volume: 144
  • Publication Date: 2020
  • Doi Number: 10.1016/j.cie.2020.106440
  • Journal Indexes: Science Citation Index Expanded (SCI-EXPANDED), Scopus, ABI/INFORM, Aerospace Database, Applied Science & Technology Source, Business Source Elite, Business Source Premier, Communication Abstracts, Compendex, Computer & Applied Sciences, INSPEC, Metadex, DIALNET, Civil Engineering Abstracts
  • Hacettepe University Affiliated: Yes


Industry 4.0 vision is bringing more challenges to quality practitioners. With advances in data acquisition systems and lowered costs, it is not uncommon in production systems to have more quality characteristics to be monitored. Once a control chart design is given, monitoring of a quality characteristic can be simplified by using a computer program that does the control statistic calculations and plots these together with the control limits over time. Then, operators trained on operational use of control charts can look for root causes of alarms triggered and take corrective actions. Nevertheless, design of a control chart requires considerable amount of time and sound knowledge on statistical methods. To eliminate the labor and automate the design of a control chart, algorithmic approaches are required to be implemented as instructions for computer programs. To serve this purpose, Shewhart individuals (I-) chart with various design choices is used in the following to algorithmically detect and eliminate outliers in the Phase I sample for estimating unknown process parameters. Hence, our approach is not restricted with the assumption of exclusively in-control observations in the sample as typically considered in the literature for evaluating the effects of parameter estimation. Estimated parameters are then incorporated into the design of exponentially weighted moving average (EWMA) chart, which is used in Phase II for process monitoring. Average and standard deviation of average run length performances of the EWMA chart are used as metrics, these are computed by using integral equations, and the results are assessed by an algorithm proposed for systematically generating knowledge on how to estimate process parameters and design the EWMA chart. The approach is general and can be implemented with alternative Phase I and Phase II control charts.