master's thesis
Beno Šircelj (Author), Miha Moškon (Mentor), Aljaž Košmerlj (Co-mentor)

Abstract

In this work, we show how Probabilistic Soft Logic (PSL) can be used to build production forecasting models, which to the best of our knowledge has not been done before. This type of forecasting is trying to predict the state of the processes in dependence of the current state of the system, such as any failures or yields. The models are created and analyzed based on a real use case in the petroleum industry. The models using PSL represent the characteristics of the underlying processes by assigning the appropriate weights to a set of PSL rules. We show how such a rule set can be constructed and how the weights can be set using four different weight learning methods. For comparison, we build three standard machine learning models. In our experiments, the models built with PSL are inferior to the other methods in terms of accuracy and computation time, making them inefficient in their current form. We discuss the possibilities to increase the applicability of the proposed implementation to the production forecasting using digital twins.

Keywords

modeling;machine learning;smart factories;probabilistic soft logic;logic networks;computer science;computer and information science;master's degree;

Data

Language: English
Year of publishing:
Typology: 2.09 - Master's Thesis
Organization: UL FRI - Faculty of Computer and Information Science
Publisher: [B. Šircelj]
UDC: 004.85:685.34.02(043.2)
COBISS: 69237507 Link will open in a new window
Views: 355
Downloads: 73
Average score: 0 (0 votes)
Metadata: JSON JSON-RDF JSON-LD TURTLE N-TRIPLES XML RDFA MICRODATA DC-XML DC-RDF RDF

Other data

Secondary language: Slovenian
Secondary title: Modeliranje proizvodnih procesov z matematično logiko
Secondary abstract: V tem delu pokažemo, kako je mogoče uporabiti verjetnostno mehko logiko (PSL) za izgradnjo modelov za napovedovanje proizvodnje, kar po našem najboljšem vedenju še ni bilo storjeno. Ta vrsta napovedovanja poskuša napovedati stanje procesov v odvisnosti od trenutnega stanja sistema, na primer morebitne izpade ali donose. Modeli so ustvarjeni in analizirani na podlagi dejanskega primera uporabe v naftni industriji. Modeli, ki uporabljajo PSL, predstavljajo karakteristike pripadajočih procesov z dodelitvijo ustreznih uteži množici pravil PSL. Pokažemo, kako je mogoče sestaviti tak nabor pravil in kako je mogoče določiti uteži z uporabo štirih različnih metod učenja uteži. Za primerjavo sestavimo tri standardne modele strojnega učenja. V naših poskusih so modeli, zgrajeni s PSL, slabši od drugih metod glede natančnosti in časa izračuna, zaradi česar so v sedanji obliki neučinkoviti. Obravnavamo tudi možnosti za povečanje uporabnosti predlagane rešitve za napovedovanje proizvodnje z uporabo digitalnih dvojčkov.
Secondary keywords: modeliranje;pametne tovarne;verjetnostna mehka logika;logične mreže;računalništvo in informatika;magisteriji;Strojno učenje;Proizvodni procesi;Simbolična in matematična logika;Računalništvo;Univerzitetna in visokošolska dela;
Type (COBISS): Master's thesis/paper
Study programme: 1000471
Embargo end date (OpenAIRE): 1970-01-01
Thesis comment: Univ. v Ljubljani, Fak. za računalništvo in informatiko
Pages: XII, 39 str.
ID: 13085122