Applying Self-organizing Maps to Master Data Making in Automatic Exterior Inspection

Abstract

In modern electronics and the electronic device industry, the manufacturing process has been changed tremendously by introducing surface mountain technology (SMT). Many automatic machines for inspecting exteriors have been added into the assembly line, in order to find automatically those products with exterior defects. Usually image processing technology and equipment are used in automatic exterior inspection due to the requirement of high inspection speed and accuracy. The pattern-matching method is the most frequently used method for image processing in exterior inspection, in which, a reference must be made as a representative image of the object to be inspected, the so-called master data. How the master data should be generated is a very important issue for increasing the inspection accuracy. In this chapter, we propose a method of making master data by using the self-organizing maps (SOM) learning algorithm and prove that such a method is effective not only in judgement accuracy but also in computational feasibility. We first apply the SOM learning algorithm to learn the image’s feature from the input of samples. Secondly, we discuss theoretically the learning parameters of SOM used in the new master data making process. Thirdly, we propose an indicator, called continuous weight, as an evaluative criterion of learning effects in order to analyze and design the learning parameters. Empirical experiments are conducted to demonstrate the performance of the indicator. Consequently, the continuous weight is shown to be effective for learning evaluation in the process of making the master data.

This chapter is organized as follows. Section 13.1 introduces our motivation for the research. Section 13.2 presents how to make the master data. In Section 13.3 the sample selection methods are defined in detail. In Section 13.4 comparison experiments are presented and discussed. Concluding remarks are given in Section 13.5.