Analiza obratovalnih parametrov ležaja sesalne enote z metodami strojnega učenja

Benedik, Blaž (author), Duhovnik, Jožef (mentor), Tavčar, Jože (comentor)

Jedro doktorske naloge je empirično modeliranje dobe trajanja ležaja sesalne enote. Glavne vplive na dobo trajanja smo razložili s statistično prepoznanimi vplivi širšega nabora domnevnih vplivov. Postavljeni napovedni model ocenjuje dobo trajanja ležaja za 50 odstotno verjetnost odpovedi L_50. Le-to določajo temperatura ležaja, hitrostni faktor, ekvivalentna obremenitev, količina masti, vrsta oljne osnove, vrsta ležajne kletke, vrsta tesnila, tolerančni razred in položaj ležaja v sesalni enoti. Nabor empiričnih podatkov je sprva predstavljalo 4672 populacij z 38.000 sesalnimi enotami. Stroge filtrirne zahteve so rezultirale v končnem seznamu 170-ih populacij za izgradnjo Weibullove podatkovne baze. Rezultat multiple linearne regresije je dobljen empirični model, ki skupaj z zgrajeno Weibullovo bazo predstavljajo doprinos k znanosti na področju napovedovanja dobe trajanja ležaja.

• Language: Slovenian
• Keywords: sesalna enota, odpovedi ležaja, razpad masti, Weibullova porazdelitev, cenzurirani podatki, filtriranje testov, doba trajanja ležaja linearna regresija
• Work type: Doctoral dissertation
• Organization: FS - Faculty of Mechanical Engineering
• Year: 2018
• PID: 20.500.12556/RUL-106150
• Publication date in RUL: 01.02.2019

BENEDIK, Blaž, 2018, Analiza obratovalnih parametrov ležaja sesalne enote z metodami strojnega učenja [online]. Doctoral dissertation. [Accessed 18 April 2025]. Retrieved from: https://repozitorij.uni-lj.si/IzpisGradiva.php?lang=eng&id=106150

Secondary language

• Language: English
• Title: Bearing operational parameters analysis in vacuum cleaner motor using machine learning methods

Abstract

The focus of the thesis is the empirical modelling of bearing life. An approach towards the prediction of life through range of different conditions was used. The model estimates bearing life for 50% probability of survival - L_50, which is determined with bearing temperature, speed factor, equivalent load, grease fill, type of oil, type of bearing cage, type of seals, tolerance class and side of the motor. The empirical data initially consisted of 4672 different test populations, involving 38.000 vacuum cleaner motors. Strict filtering requirements resulted in 170 final populations selected for building Weibull databank for learning final models. The obtained empirical models gained with multiple linear regression are together with obtained databank a contribution to the science in field of bearing life forecasting.

• Keywords: vacuum cleaner motor, bearing failure, grease deterioration, Weibull probability function, censored data Filtering of tests Bearing life Linear regression