Inspection allocation models for flexible inspection systems

Abstract

Recent technological developments require new models for inspection allocation. The tightening of product tolerances and the application of computer technology to almost every aspect of manufacturing are some of the changes that affect the role that inspection has on quality output. Due to the tightening of tolerances, manufacturing processes are not capable of rendering an adequate quality level. As a result, screening inspection as a means of guaranteeing quality levels is commonplace in today's manufacturing environment. An example of new computer technologies is Automated Visual Inspection (AVI). AVI has taken the place of an inspector in deciding whether or not a part is manufactured according to specifications. These new technologies make possible a new inspection concept known as Flexible Inspection System (FIS). Such a system uses the historical information provided by the inspection operations of a single unit in order to select an inspection routine to be used at the current inspection station and to decide whether to scrap the unit under inspection in order to optimize a global objective. The development of an FIS model applied to a serial multi-stage production system and single inspection station is the objective of this dissertation. In the modelled system, the inspection may yield type I and II errors. These characteristics together with some independence assumptions, allow the quality of the unit under manufacture to be modelled as a partially observable Markov process. Thus, the inspection allocation becomes a special case of optimal control of a partially observable Markov process under time constraints where the quality of the unit under observation is a discrete time continuous state Markov process. Time constraints are introduced to account for the limited time available for inspection allocation.