Roketinec, Ivica (2001) Primjena diskriminacijske analize u proizvodnim procesima. = Discriminant analysis application in production processes. Scientific master's thesis , Sveučilište u Zagrebu, Fakultet strojarstva i brodogradnje, UNSPECIFIED. Mentor: Šakić, Nikola.
Full text not available from this repository.Abstract (Croatian)
Za najveći je broj suvremenih proizvodnih sustava karakterističan proizvodni program relativno većeg asortimana i varijanti proizvoda u isto tako relativno brojnim količinama istovrsnih proizvoda. Upravljanje proizvodnim procesima u ovakvim je sustavima vrlo složeno jer tipična rješenja poput onih za pojedinačnu, maloserijsku iii pak masovnu proizvodnju nisu adekvatno primjenjiva. Ipak, ideja je formirati takvo rješenje koje ce u što je moguće većoj mjeri iskoristiti njihove prednosti, prije svega visoku proizvodnost masovne te visoku fleksibilnost pojedinačne i maloserijske proizvodnje. U to se svrhu proizvodi grupiraju u grupe koje ce između sebe biti što je moguće heterogenije a unutar što je moguće homogenije spram unaprijed definiranih obi1ježja, primjerice tehnoloških parametara. Proizvodi iste grupe izrađivati ce se prema istim iii sličnim tehnološkim postupcima. Osnovni je problem kako razvrstati nove varijante proizvoda u postojeće grupe tehnološki sličnih proizvoda kad im tehnologija izrade nije poznata. Ideja je iz već ranije formiranih grupa na osnovu poznatih tehnoloških podataka o proizvodima iz prijašnjih serija razviti postupak, metodu ili model po kojemu bi se neki sasvim novi proizvod za kojeg je tehnologija izrade nepoznata mogao svrstati u neku od postojećih grupa tehnološki sličnih dijelova pa onda u ovisnosti o grupi u koju je upao, razraditi za njega tehnološki postupak. Informacije prema kojima bi se ocjenjivala mogućnost klasifikacije tog proizvoda u neku od prethodno definiranih grupa tehnološki sličnih proizvoda bili bi isključivo konstrukcijski parametri tog proizvoda koji se nalaze na konstrukcijskom crtežu. Kao alat kojim se ova hipoteza nastoji poduprijeti izabrana je diskriminacijska analiza kao jedna iz skupine zavisnih multivariatnih statističkih metoda. U ovom je radu spomenuta metoda detaljno razrađena, dok su preostale predstavljene samo informativno. Opisani su ciljevi koji se tom metodom postižu, data su osnovna ograničenja, njene prednosti i nedostaci. Također, istaknuti su i preduvjeti koji moraju biti zadovoljeni da bi se metoda uopće mogla primijeniti. Posebna je pogodnost metode što iz izvornih, tj. manifestnih varijabli formira nove, tzv. latentne varijable kao linearne kombinacije manifestnih. Ova se pogodnost s jedne strane očituje u redukciji varijabli problema, a s druge strane, u novim se, latentnim varijablama ostvaruje maksimalna moguća separacija ovih prethodno definiranih grupa. Time se ujedno postiže njihovo minimalno preklapanje pa ce i klasifikacijski postupak koji ce koristiti latentne varijable dati najmanju moguću grešku. U radu je provedena analiza na realnim podacima o strojnim dijelovima koji su već prethodno bili razvrstani u grupe prema njihovoj tehnološkoj sličnosti. Diskriminacijskom je analizom pokazano da se usprkos niza nepovoljnih okolnosti doista mote na osnovu konstrukcijskih parametara dosta dobro procijeniti u koju bi grupu tehnološki sličnih strojnih dijelova promatrani subjekt trebao upasti. Važno je napomenuti da se mnoge od ovih nepovoljnih okolnosti mogu izbjeći, ukloniti ili umanjiti. U svakom slučaju, dobiveni su rezultati dobar putokaz kojim bi se moglo krenuti u analizu te u konačnici postići što veću efikasnost u upravljanju proizvodnim procesima.
Abstract
The production program with relative wider range and variant of products in also relative larger quantities of the similar products is characteristic for the greatest number of modern production systems. Production processes management in such systems is very complex because the typical solutions known job shop, small batch or repetitive production aren't adequate applicable. Though, the idea is to form such solution that will as the most as possible use their advantages, at first high productivity of repetitive, and high flexibility of job shop and small batch production. From this reason, the products are grouped into groups that will be among themselves as the most heterogeneous and inside themselves the most homogenous as possible according to in advance defined attributes, the technological parameters, for example. The products from the same group will be manufactured according to the same or similar routings. The basic problem is how to classify the new variants of products into existing technologically similar groups of products when their routings aren't known. The idea is to develop from earlier formed groups which are composed on the basis of known technological data about earlier manufactured products, the procedure, method or model according to some completely new product, for whom the routing is not known, could be classified into some of existing technologically similar groups of products, and after that in dependence of group into which it has been classified the routing for this product is elaborated. The information according to the possibility of classification of some product into one of predefined groups of technologically similar products would be evaluated, would be exclusively the construction parameters of this product which could be found on product construction drawing. As a tool with which this hypothesis is tried to be upheld the discriminant analysis as one from the group of dependent multivariate statistical methods is chosen. In this paper the noted method has been detailed elaborated while the others are represented only briefly. The aims that by this method are obtained have been described, the basic constrains, and also its advantages and disadvantages are given. Although, the prerequisite which has to be fulfilled in order to the method could even be applicable has been pointed out. Particular convenience of method is that from original, i.e. manifest variables the new, latent variables as the linear combinations of manifest variables are formed. This convenience is manifested in one hand in variables number reduction, and in the other hand, by using the new, latent variables the maximal possible predefined groups separation is achieved. Hereby the minimal overlap of the groups is accomplished so the classification procedure that uses the latent variables will give the smallest possible classification error. In this paper the analysis using the real data about machine parts that were beforehand classified into groups according to their technological similarities has been performed. By discriminant analysis has been shown that despite for a number of unsuitable circumstances verily on the construction parameters basis could be good enough estimated into which group of technologically similar machine parts an observed subject should be classified. It's important to remark that many of that unsuitable circumstances could be avoid, eliminated or reduced. In any case, the gained results are good signpost in which direction the analysis could be started and finally achieve even larger efficiency in production processes management.
| Item Type: | Thesis (Scientific master's thesis) |
|---|---|
| Uncontrolled Keywords: | diskriminacijska analiza; separacija grupa; klasifikacija; proizvodni procesi; konstrukcijski i tehnološki parametri; manifestne i latentne varijable |
| Keywords (Croatian): | discriminant analysis; group separation; classification; production processes; constructional and technological parameters; manifest and latent variable |
| Divisions: | 700 Department of Industrial Engineering |
| Date Deposited: | 22 Sep 2014 18:00 |
| Last Modified: | 16 Oct 2015 13:19 |
| URI: | http://repozitorij.fsb.hr/id/eprint/173 |
Actions (login required)
 |
View Item |
Citations:
Citations: