20.500.12556/RUNG-5116-fe911660-ae15-86a9-1f7f-2da62f5f42a8
Zaznavanje in lokalizacija poškodb ležajev v rotacijskih strojih
On-line fault diagnosis of bearings in rotational machines
Ležaji so pomemben del rotacijskega stroja, saj prenašajo silo vrtečega se dela na ohišje. Zaradi te funkcije so velikokrat podvrženi prekomernim silam in posledično poškodbam. Odkritje poškodbe v zgodnji fazi je zelo pomembno, saj lahko na ta način preprečimo draga popravila strojev in izpad prihodkov zaradi zaustavitve. V literaturi najdemo veliko postopkov za diagnostiko napak na ležajih, ki praviloma temeljijo na zaznavanju sprememb v karakterističnih veličinah, ki jim pravimo značilke. Prvi pomemben korak pri diagnostiki je detekcija, kjer je potrebno zaznati, ali je prišlo do spremembe v vrednosti značilke. Klasične značilke se dobro obnesejo pri konstantnih pogojih obratovanja, slabše pa, če se ti pogoji spreminjajo. Težko je namreč ločiti med spremembami v značilkah vsled poškodbe in spremembami zaradi spremenljivih obratovalnih pogojev. Tradicionalni načini zaznavanja sprememb v značilki temeljijo na ugotavljanju, ali je ta presegla nek prag. Težava, ki se pri tem pojavlja, je, da optimalnega praga ni enostavno sistematično izbrati, poleg tega pa se lahko pojavijo problemi s pogostimi prehodi značilke čezenj in nazaj, kar ima za posledico pogosto vklapljanje in izklapljanje alarma. Temu pravimo diagnostična nestabilnost.
Namen magistrskega dela je bil raziskati možnosti za odpravo težav z diagnostično nestabilnostjo pri diagnosticiranju poškodb ležajev s statističnimi koncepti, in sicer konkretno z uporabo Jensen-Renyijeve divergence. Delovanje pristopa smo najprej raziskali na simulacijskih primerih, nato pa ga uporabili na realnih meritvah iz baze podatkov IMS Bearing Dataset. Opisali smo celoten postopek detekcije, tudi določitev optimalnega praga poškodbe po hevristični metodi. Ker je ideja dokaj nova, smo morali za vse pripraviti ustrezne algoritme. Ugotovili smo, da Jensen-Renyijeva divergenca ob pravilni nastavitvi parametrov deluje zelo dobro. Vrednosti značilk naraščajo monotono, posledično je manj možnosti za lažni alarm. Poleg tega pa spremembo zazna že ob najmanjšem povišanju, brez zakasnitve.
Bearings are an important part of the rotational machinery that transfer axial and radial loads to the supporting structure. This feature causes bearings to be subjected to excessive force and consequently damage. Detecting faults in an early-stage is very important to prevent expensive machinery repairs and loss of revenues due to downtimes. Many procedures for diagnosing bearing faults are based on evaluating changes in the characteristic quantities called features. The first important step in diagnostics is detection. It is necessary to detect whether a change in the feature compared to the fault-free case has occurred. Classical features perform well under constant operating conditions, but might turn inappropriate if these conditions change. Namely, in such a case, it is difficult to distinguish between changes in features due to fault and changes due to variable operating conditions. Conventional change detection is based on assessing whether a feature value exceeds a threshold. This can have some drawbacks. Foremost, it is not easy to select an optimal threshold and what is more, we can have problems with frequent feature transitions over the threshold, resulting in alarms being triggered too often. This is called diagnostic instability.
The purpose of the thesis was to explore the possibility of eliminating diagnostic instability when diagnosing bearing faults by using Jensen-Renyi divergence. The behaviour of the approach was first investigated on simulation cases and then applied to real measurements from the IMS Bearing Dataset. The entire detection procedure is described, including the determination of the optimal fault threshold using heuristic approach. The idea is fairly new, so we had to come up with all the algorithms for the method to work. Jensen-Renyi divergence turned to work very well and is easy to tune. Under mild conditions, the evolution of the feature value is smooth, resulting in reliable detection with minimal chance for false alarm. In addition, it turned rather sensitive to the incipient faults, which makes it attractive for industry.
ležaji
diagnostika
Jensen-Renyijeva divergenca
Fourierjeva transformacija
bearings
diagnostics
Jensen-Renyi divergence
Fourier transform
true
true
false
Slovenski jezik
Angleški jezik
Magistrsko delo/naloga
2020-04-14 09:40:57
2020-06-30 05:27:26
2023-06-13 14:26:07
0000-00-00 00:00:00
2020
0
Nova Gorica
0
0000-00-00
NiDoloceno
NiDoloceno
NiDoloceno
0000-00-00
0000-00-00
0000-00-00
21033475
URN:SI:UNG:REP:HBTLUQ5E
Patrik_Persic.pdf
Patrik_Persic.pdf
1
2C984E9DC02D2069D3E618EB8A25B4D5
41382612d1f3f3beb942e18af76ab5350fb62a62496739ff5e8d45ad7753a3ad
7d0a5706-05cf-11ee-9c48-5ef991fed68f
https://repozitorij.ung.si/Dokument.php?lang=slv&id=19684
Poslovno-tehniška fakulteta
0
0
0