diplomska naloga
Jan Dobnikar (Avtor), Goran Turk (Mentor), Vlatko Bosiljkov (Član komisije za zagovor), Janko Logar (Član komisije za zagovor), Vlado Stankovski (Komentor)

Povzetek

Diplomska naloga obravnava problem razvrščanja lesa v trdnostne razrede. Vsak material, uporabljen v konstrukciji, mora biti razvrščen v trdnostni razred in ustrezno označen z oznako CE. Za razvrstitev v določen trdnostni razred mora izpolnjevati minimalne zahteve tega razreda glede mehanskih karakteristik. Les je nehomogen anizotropen material. Trdnost lesa znotraj posamezne vrste je med drugim odvisna tudi od dela debla, s katerega odreţemo preizkušanec, od nahajališča, okolice, prsti, starosti, poškodovanosti, grčavosti itd. Edini način za točno določitev trdnosti lesa je porušitev preizkušanca. Trenutno je v Sloveniji uveljavljeno le vizualno ocenjevanje lesa, ki pa je zelo konzervativno, zato je veliko dobrega lesa prodanega pod zasluţeno ceno. V nalogi smo z uporabo metod strojnega učenja razvili sistem, ki je osnova za boljše razvrščanje lesenih elementov v razrede. Analizirano je bilo več kot 5000 lesenih preizkušancev, na katerih so bile opravljene tako neporušne kot porušne preiskave. Rezultati preiskav so bili obdelani s programoma WEKA Toolkit in IBM SPSS 20. Poleg osnovnih statističnih testov so bile za obdelavo uporabljene nevronske mreţe in regresijska drevesa. Pokazali smo, da je bil izbrani vzorec dovolj velik, ter da so nevronske mreţe primerno orodje za določevanje parametrov za klasificiranje lesa na osnovi rezultatov neporušnih analiz.

Ključne besede

gradbeništvo;diplomska dela;UNI;trdnostni razred;les;dinamični odziv;elastični modul;strojno učenje;regresijsko drevo;nevronska mreža;

Podatki

Jezik: Slovenski jezik
Leto izida:
Tipologija: 2.11 - Diplomsko delo
Organizacija: UL FGG - Fakulteta za gradbeništvo in geodezijo
Založnik: [J. Dobnikar]
UDK: 004.85:624.011.1(043.2)
COBISS: 5809249 Povezava se bo odprla v novem oknu
Št. ogledov: 2347
Št. prenosov: 433
Ocena: 0 (0 glasov)
Metapodatki: JSON JSON-RDF JSON-LD TURTLE N-TRIPLES XML RDFA MICRODATA DC-XML DC-RDF RDF

Ostali podatki

Sekundarni jezik: Angleški jezik
Sekundarni naslov: ǂThe ǂuse of machine learning methods for the grading of timber strength
Sekundarni povzetek: This work addresses the problem of grading timber into strength classes. All materials, used in the construction, have to be graded and marked with a CE mark. For the classification into a certain strength class, the material has to adhere to minimal mechanical requirements of the class. Timber is an anisotropic, non-homogeneous material. The timber strength within a certain species depends, among other factors, on the part of the log that the sample has been cut off, the environment of its growth, the type of soil, the timber age, the damage to the timber, the presence of knots etc. The only accurate method for determining timber strength is destructive testing. Presently, the only method considered relevant in Slovenia is the visual evaluation of the timber, which is a very conservative method and results in lower prices of good quality timber. The goal of the presented work was to develop a system for timber grading on the basis of results of non-destructive methods, particularly by using the machine learning methods. The analysis sample consists of more than 5000 timber samples, which have been tested with non-destructive, as well as destructive methods. Results of the performed tests were analysed by using the machine learning algorithms contained in the WEKA Toolkit and the statistical methods of the IBM's SPSS 20 program. In order to analyze the collected data, standard statistical tests were performed for the analysis and artificial neural networks, as well as regression trees, were also used. The presented results indicate that the number of the samples was sufficient and that the formalism of the artificial neural networks is the appropriate tool for determining the timber grading parameters on the basis of results of non destructive testing methods.
Sekundarne ključne besede: graduation thesis;civil engineering;strength clas;timber;dynamic response;module of elasticity;machine learning;regression tree;artificial neural network;
Vrsta datoteke: application/pdf
Vrsta dela (COBISS): Diplomsko delo
Komentar na gradivo: Univ. v Ljubljani, Fak. za gradbeništvo in geodezijo
Strani: IX, 63 str.
ID: 8725518