CLUSTERING METHODS FOR CLASSIFICATION OF ELECTRONIC DEVICES BY PRODUCTION BATCHES AND QUALITY CLASSES

Abstract

Authors consider problem of electronic units packaging with highly reliable
electronic components and preventing ingress of low-grade counterfeit products that
does not meet the requirements for reliability. When making any electronic circuits,
it is desirable to use electronic and radio components with the same characteristics
which is most likely achieved using components (chips, transistors, diodes, capacitors,
relays, crystals, resistors, etc.) manufactured as a single production batch. If the
production method is not exactly known, only affordable way to improve the quality is
the comprehensive testing of the delivered production batches. The paper discusses the
problem of identifying a production batch of electronic and radio components delivered
from a provider based on the test results. The problem is reduced to a series of problems
of cluster analysis a special genetic algorithm is applied for. In addition, the testing
problem of electronic and radio products is presented as pattern recognition without a
teacher. A new algorithm for data classification in the multidimensional feature space
is given. It was proposed to group objects on the basis of the distances analysis, i.e.,
the algorithm does not require knowledge about a number of classes in contrast to the
majority of well-known algorithms for taxonomy.