Joint Research Conference
June 24-26, 2014
In the traditional SPC approach, the retrospective analysis in Phase I, provides the parameter estimates required for the control chart construction, under the assumption of a relatively long sequence of iid data from the “in control” state of the process. This paradigm is problematic in cases that online monitoring is required or when any of the assumptions is violated, causing questionable control limits that will seriously affect phase II testing.
In this work we propose a Bayesian alternative, using the idea of the sequentially updated predictive distribution. This non-static control chart will make use of available prior information regarding the underline parameter of the Normal process and allow online inference from the very first observation. A comparison to frequentist based self starting methods will be provided and a case study from internal quality control management of a medical lab will be presented.
Keywords:
Bayesian statistical process control; Outlier detection; Online inference; Short runs