Datos No Estacionarios

Research Article
Received 20 May 2011, Revised 26 June 2011, Accepted 26 June 2011 Published online in Wiley Online Library

(wileyonlinelibrary.com) DOI: 10.1002/asmb.911

Nonstationarity in statistical process control — issues, cases, ideas
Bart De Ketelaerea,b * † , Kristof Mertensa , Frank Mathijsa , Daniel Sabin Diaza and Josse De Baerdemaekera
Statistical process control (SPC) is apowerful framework that is used in many industries to decrease process variability and to pinpoint special cause variation. Although a broad range of techniques have been developed to do so, often the real-life situation does not fully comply with the basic assumptions that are made in SPC resulting in poor results. One of the main violations against the assumptions is the fact that industrialprocesses rarely behave in a stationary manner — this is evidently the case for biological processes but is also an important issue when monitoring industrial processes. Besides, the ever increasing amount of data, with a clear shift towards multivariate and even multiway quality control, makes the classical univariate approach not feasible anymore. These two observations pose important challenges tostatisticians to develop novel SPC algorithms that are broadly applicable in modern industries. In this contribution we discuss both issues and use two very different case studies to show the reader recent directions and developments in the SPC landscape. Copyright © 2011 John Wiley & Sons, Ltd. Keywords: statistical process control; nonstationarity; synergistic control

1. IntroductionEngineering is a rapidly changing discipline that is characterized by an ever increasing amount of automation. The higher degree of automation and related increasing production speeds together with the strong expectations of the customers have led to the development of a broad range of novel sensor technologies that all produce mass data in a fraction of a second, at a cost-efficient price. One can thinkof cheap camera technologies, vibration sensors and optical sensor technologies to name a few. Also, computer power has increased significantly so that engineers have available a broad range of tools for attaining their quality goals. These developments have as a consequence that the type of data and the way they are collected are subject to change. Whereas the engineer would have had informationabout one single quality characteristic of a limited sample of the total production batch some decades ago, nowadays he or she may have collected hundreds or even thousands of different characteristics of each and every sample, possibly even with repeated measures over time. These developments seem to have found their way in many application fields, where process sensors are widely implemented andform the basis of the business management strategy to improve the quality of process outputs by identifying and removing the causes of defects (errors) and minimizing variability. This idea of better understanding the process and minimizing variability is central to widely accepted strategies such as Six Sigma and, for the food and pharmaceutical industries, Process Analytical Technologies.Monitoring processes to gain insights and to detect special cause variation are the subject of a field that is referred to as statistical process control (SPC), although the term statistical process monitoring might be a better description because the actual control step (in an engineering sense) is not part of it. Key tools in SPC are control charts, first developed back in the 1920s by Shewhart whoconcluded that while every process displays variation, some processes display controlled variation that is natural to the process (common cause variation), while others display uncontrolled variation that is not

Universiteit Leuven, Department of Biosystems, Division Mechatronics, Biostatistics and Sensors, Kasteelpark Arenberg 30, 3001 Heverlee, Belgium b Katholieke Universiteit Leuven, Leuven...