20.500.12556/DKUM-46366 ANALIZA KALJENIH MATERIALOV S POMOČJO FRAKTALNE GEOMETRIJE ANALISYS OF HARDENED MATERIALS USING FRACTAL GEOMETRY V disertaciji proučujemo inteligentne sisteme in iskanje znanja, račcunalniške paradigme, ki so uporabni in koristni pri toplotni obdelavi materiala. Za razpoznavanje kompleksnosti različcno toplotno obdelanih vzorcev smo uporabili metodo fraktalne geometrije. Oblikovali smo inteligentni sistem s pomočjo katerega bomo napovedali topografske lastnosti materiala po toplotni obdelavi. Razvili smo nov algoritem 3D grafov vidljivosti. S pomočjo topološke lastnosti povezanost 3D grafov vidljivosti smo zgradili inteligentni sistem, s katerim bomo napovedali topografske lastnosti vzorcev po toplotni obdelavi. Fraktalno geometrijo lahko uporabimo za analiziranje kompleksnih struktur, ki se pojavljajo pri toplotni obdelavi materialov. Tako lahko s pomočcjo fraktalne geometrije pokažzemo prednosti laserske tehnike toplotne obdelave pred induktivnim in klasičcnim v kalilni peči. Fraktalna geometrija je nov pristop, ki temelji na karakterizaciji nepravilnih mikrostruktur in sluˇzi kot ocenjevalno orodje za določcitev strukturnih lastnosti in jo bomo uporabili pri analizi različno toplotno obdelanih materialih. Sama definicija pravi, da fraktalna geometrija temelji na ideji invariantnosti povečcave, kar pomeni, da je opazovana slika enaka ne glede na to, pod kakšno povečavo jo gledamo. Potrebno je poudariti, da fraktalna dimenzija v celoti ne karakterizira geometrije, ampak je le pokazatelj nepravilnosti objekta. Fraktalno geometrijo smo uporabili za določcitev topografskih lastnosti kaljenih materialov. Uvedli smo novo metodo za račcunanje fraktalne dimenzije 3D objektom. Z razvojem laserske tehnike na področcju toplotne obdelave materialov se je povečcala potreba po razvoju nove metode, s katero bi ugotovili (določili) večjo odpornost na obrabo, manjše trenje in večjo toplotno obstojnost materiala. Tako bomo z gradnjo inteligentnih sistemov povečali produktivnost na področju toplotne obdelave materialov. S pomočjo inteligentnega sistema bomo pokazali, katera tehnika in tehnologija toplotne obdelave je boljša. V disertaciji bomo predstavili štiri nove sestavljene hibridne metode strojnega učenja. Ti novi sestavljeni hibridni sistemi so: · sestavljeni hibrid genetski algoritmi - multipla regresija - nevronska mreža - multipla regresija. Poimenovali smo ga hibrid zanka. · sestavljeni hibrid genetski algoritmi - nevronska mreža - multipla regresija - nevronska mreža. Poimenovali smo ga optimalni hibrid zanka. · sestavljeni hibrid genetski algoritmi - nevronska mreža - multipla regresija - nevronska mreža - multipla regresija. Poimenovali smo ga ciklični hibrid. · sestavljeni hibrid genetski algoritmi - multipla regresija - nevronska mreža - multipla regresija - nevronska mreža. Poimenovali smo ga optimalni ciklični hibrid. Pri sestavljenih hibridih smo dosegli malce slabše rezultate od pričakovanih, zaradi pomanjkljivosti posameznih osnovnih metod. Metoda multipla regresija je najslabˇsa metoda in vpliva negativno na sestavljen hibrid. Novi omenjeni sestavljeni hibridi pa dajo boljše rezultate kot že znani sestavljeni hibridni sistemi. Želeli smo še izboljšati rezultate novih hibridov, zato smo zgradili kombinirane (sestavljene) nove hibridne sisteme, ki smo jih poimenovali hiper hibridi. Na koncu disertacije so podani še komentarji in ideja o dveh novih hibridih, ki smo jih poimenovali spiralni hibrid in optimalni spiralni hibrid. Ti metodi sta uporabni pri večjem številu osnovnih metod inteligentnih sistemov. Predlagamo še šest kombiniranih metod novih predstavljenih hibridov v hiper hibride. In this dissertation we study intelligent systems and the search for knowledge, computing paradigms that are useful and beneficial for the heat treatment of materials. To identify the complexity of different heat-treated samples, we used the fractal geometry method. We designed an intelligent system through which we announced topographical properties of the material after heat treatment. We have also developed a new algorithm for 3D graph visibility. With the help of the topological properties density of 3D graphs, we have built an intelligent system which can predict the topographic characteristics of the samples after heat treatment. Fractal geometry can be used to analyse complex structures that occur in the heat treatment of materials. Thus, the use of fractal geometry demonstrates the advantages of laser heat treatment techniques over the inductive, classical and the hardening furnace. Fractal geometry is a new approach, based on the characterisation of irregular microstructures, and serves as an assessment tool for determining structural properties. It can be used in the analysis of different heat-treated materials. Fractal geometry is based on the idea of invariant magnification, which means that the observed image is not the same regardless of how strong the microscope is. It should be noted that the fractal dimension does not fully characterise the geometry, but is rather an indication of irregularities. Fractal geometry was used here to determine the topographical properties of hardened materials . We have introduced a new method for calculating the fractal dimension of a 3D object. With the development of laser technology in the field of heat treatment of materials there is an increased need to develop new methods with which to determine (set) better resistance of material, lower friction and better heat resistance of material. We therefore aim to build intelligent systems to increase productivity in the field of heat treatment of materials. With the help of the intelligent system we intend to show which technique of heat treatment is best. In this dissertation we present four new composite hybrid methods: · composite hybrid genetic algorithms - multiple regression - neural networkmultiple regression (we call it a hybrid loop). · composite hybrid genetic algorithm - neural network - multiple regression-neural network (we call it the optimal hybrid loop). · composite hybrid genetic algorithm - neural network - multiple regressionneural network - multiple regression (we call it the cyclic hybrid). · composite hybrid genetic algorithm - multiple regression - neural network -multiple regression - neural network (we call it the optimal linear hybrid). Composite hybrid performances were slightly worse than expected, because of the shortcomings of the individual basic methods. The multiple regression method is the worst method and adversely affected the composite hybrid. The new composite hybrids give better results than existing composite hybrid systems, however. We wan’t to improofe results of new hybrid system, thus we built new composite hybrib, hyiper hybrid. At the end of the dissertation further comments are made and a two new hybrid systems proposed which we call the spiral hybrid and optimal spiral hybrid. This method are useful when a large number of basic methods are employed. We also propose combining (pooling) the six new hybrid methods presented in the new hyper hybrids. inteligentni sistemi algoritmi hibridni sistemi strojno učenje fraktalna geometrija teorija grafov topografija materiala po toplotni obdelavi intelligent system algorithms hybrid system machine learning fractal geometry graph theory topography of materials after heat treatment true true false M. Babič] Slovenski jezik Angleški jezik Doktorska disertacija 2014-09-24 07:29:14 2014-09-24 08:17:36 2022-05-17 03:11:35 2021-05-02 03:24:52 2014 0 [Maribor 0 0000-00-00 NiDoloceno NiDoloceno NiDoloceno 0000-00-00 0000-00-00 0000-00-00 004.896:530.191(043.3) 18099990 URN:SI:UM:DK:S0NIFSG5 DR_Babic_Matej_2014.pdf DR_Babic_Matej_2014.pdf 1 ACA2720E858AE86A146D695880E92EEB cb26515c075deed6a1923b99800376cbcde0b3bf534f39a2bf2332f1d9ffd205 4d5b49b4-7c0b-11eb-bb7a-00155d0001ca https://dk.um.si/Dokument.php?lang=slv&id=67818 Fakulteta za elektrotehniko, računalništvo in informatiko 0 0 0