Primjena strojnog učenja u održavanju

master's thesis

• Title (english): Application of machine learning methods in maintenance
• Author: Marin Doko
• Mentor: Dragutin Lisjak (mentor)
• Granter: University of Zagreb; Faculty of Mechanical Engineering and Naval Architecture; Zagreb
• Defense date and country: 2018-07-12, Croatia
• Scientific / art field, discipline and subdiscipline: TECHNICAL SCIENCES; Mechanical Engineering

Abstract

Inteligentni sustavi rane dijagnostike kvarova unutar sustava održavanja strojne opreme sve više zamjenjuju konvencionalne metode dijagnostike koje uvelike ovise o ljudskom faktoru. Primjenom algoritama strojnog učenja unutar strategije prediktivnog održavanja povećava se efikasnost sustava, kvaliteta proizvoda i usluge te se smanjuje vrijeme zastoja rada. U ovom radu primijenjen je algoritam konvolucijskih neuronskih mreža za analizu vibracijskih signala prikupljenih sa senzora na eksperimentalnom postavu rotacijske opreme. Vibracijski signal ulazi u neuronsku mrežu bez prethodnih operacija ručnih izvlačenja značajki te kroz višeslojnu strukturu mreže prolazi kroz niz operacija kojima se značajke signala automatski uče, izlučuju i grupiraju. Točnost klasifikacije ovakvog algoritma iznosi oko 99.6% i time pokazuje potencijal unutar dijagnostike kvarova strojne opreme i nameće se kao centar daljnjeg istraživanja.

Abstract (english)

Intelligent fault diagnosis systems are replacing conventional diagnostic methods which are dependent on commonly unreliable human factor. System efficiency, products and services quality can all be increased by applying machine learning algorithms in predictive maintenance strategy and, in the same time, machine downtime can be minimized. In this thesis, Convolutional neural networks are applied for analysis of vibration signals, which are measured from experimental setup of rotational equipment. Vibration signal is imported in CNN without manual extraction of features and passed through multi-layer architecture where the features are automatically learned, extracted and grouped. Classification accuracy of this proposed algorithm is 99.6%, so it shows large potential for appliance in fault diagnostics and prediction field and it is becoming central part of further investigations.

• Keywords: Prediktivno održavanje; strojno učenje; Internet stvari; konvolucijske neuronske mreže; vibracije
• Keywords (english): Predictive maintanance; machine learning; Internet of things; convolutional neural networks; vibrations
• Language: croatian
• URN:NBN: urn:nbn:hr:235:655932
• Study programme: Title: Mechanical Engineering; Study programme type: university; Study level: graduate; Academic / professional title: magistar/magistra inženjer/inženjerka strojarstva (magistar/magistra inženjer/inženjerka strojarstva)
• Type of resource: Text
• File origin: Born digital
• Access conditions: Open access
• Created on: 2019-05-22 22:48:10