Simultaneous Identification of Premodeled and Unmodeled Variation Patterns

• Autores: Daniel W. Apley, Ho Youn Lee
• Localización: Journal of quality technology: A quarterly journal of methods applications and related topics, ISSN 0022-4065, Vol. 42, Nº. 1, 2010, págs. 36-51
• Idioma: inglés
• Enlaces
  • Texto completo
• Resumen

  • One method often used to identifying root causes of variation in multivariate manufacturing data first models a set of variation patterns off-line, and then attempts to identify the presence or absence of the premodeled patterns in a sample of on-line data. A second method requires no premodeling and uses only on-line data to determine the nature of any variation patterns that are present. The method presented here combines these two approaches and eliminates some of the shortcomings of each. The new method premodels only patterns that are relatively easy to do. The characteristics of any unmodeled patterns that may be present in the on-line data are automatically detected and treated in a manner that does not affect the detection of the premodeled patterns. The approach is illustrated in an example from automobile body assembly.