Uncertainty Techniques Make Defect Classification More Intuitive

In many industrial process control situations the need to identify and classify defects is key to enabling process improvements. Inspection is used for this task. The process of constructing a classification scheme to correctly identify defects in a product is a difficult process. Many people assume that a description of what is to be identified exists. That is not necessarily true. Three types of uncertainty exist which make it difficult to define a complete model of an inspection environment. The domain knowledge is vague and nearly always incomplete, and the constraints on inspection equipment means the data output often produces ambiguous results. A method is required which will allow such uncertainties to be incorporated into an automated visual inspection system. Fuzzy Logic and Dempster Shafer have been used to aid the development of a classification scheme for identifying defects in plastic film. This paper discusses the implementation of the techniques to a real industrial problem.