Development of an image-based procedure for quality control of graphite components

Komutator je pomemben in zelo občutljiv del komutatorskega elektromotorja, ki pritrjen na os motorja skrbi za periodično menjavanje smeri električnega toka in tako omogoča delovanje motorja. Kakovost izdelave komutatorja je zato odločilnega pomena za kakovost elektromotorja. Ročni nadzor kakovosti je časovno zahteven in nezanesljiv, zato je v ključnih korakih proizvodnje komutatorjev smiselna uvedba avtomatskega nadzora kakovosti. V proizvodnji grafitnih komutatorjev, ki jih sestavljata grafitna ploščica in bakrena osnova, je ključen korak spajanje teh dveh sestavnih delov. Diplomsko delo obravnava razvoj vgradne aplikacije za avtomatski nadzor kakovosti grafitnih komutatorjev po spajanju grafitne ploščice z bakreno osnovo. Namen aplikacije je prepoznavanje štirih vrst napak, do katerih pride v postopku spajanja. Najprej z metodami strojnega vida iz slik komutatorjev pridobimo atribute, nato pa na njihovi podlagi s strojnim učenjem zgradimo odločitvena drevesa, ki omogočajo določanje napak na komutatorjih. Na koncu preizkusimo še druge metode učenja in njihove rezultate primerjamo z rezultati odločitvenih dreves.A commutator is an important and very sensitive part of the commutator electric motor. Located on the motor axis, it periodicaly changes the direction of the electric current, enabling the motor to run. For this reason, the quality of commutator production is crucial for the quality of the electric motor. Manual quality control is time-consuming and unreliable, therefore it is reasonable to introduce automated quality control in the key steps of commutator production. The graphite commutator consists of two main parts, a metalized graphite disc and a copper base. One of the crucial steps in graphite commutator production is soldering of these parts. This thesis deals with the development of an embedded application for automated inspection of the commutator quality after soldering of the metalized graphite disc and the copper base. The goal of the application is to detect four types of defects occurring during the soldering process. Methods of machine vision are used first to acquire attributes from the commutator images. From these attributes decision trees are then constructed through machine learning that make it possible to determine defects on commutators. Finally, other learning methods are tested and their results compared with the results of decision trees