Metoda ekstrakcije audiosignala u dijagnostici rada regulacijske sklopke transformatora

doctoral thesis

• Title (english): Audio signal extraction method in transformer on-load tap changer diagnostics
• Author: Adnan Sečić; MBZ: 40459
• Mentor: Igor Kuzle (mentor)
• Committee member: Juraj Havelka (predsjednik povjerenstva)
• Committee member: Kristian Jambrošić (član povjerenstva)
• Committee member: Žarko Janić (član povjerenstva)
• Granter: University of Zagreb; Faculty of Electrical Engineering and Computing; Zagreb
• Defense date and country: 2024, Croatia
• Scientific / art field, discipline and subdiscipline: TECHNICAL SCIENCES; Computing
• Universal decimal classification (UDC): 004 - Computer science and technology. Computing. Data processing

Abstract

U radu se razmatra problematika dijagnostike regulacijske sklopke transformatora na osnovu audio zapisa. Ovaj pristup dijagnostici donosi nekoliko izazova: automatska segmentacija audio zapisa, odvajanje korisnog zvuka od pozadinske buke i ekstrahiranje dijagnostički relevantnih informacija iz audio zapisa. Problematika automatske segmentacije rješava se primjenom konvolucijskih dubokih neuronskih mreža, čija je efikasnost ispitana na stvarnim audio zapisima zabilježenima tijekom rada regulacijske sklopke. Za rješavanje problematike razdvajanja signala također se koristi pristup zasnovan na dubokom učenju, gdje se duboka neuronska mreža uči da razdvoji signale na temelju ocjene vremensko-frekvencijske maske koja se primjenjuje na vremensko-frekvencijski spektar signala. Ispitivanje metodologije provedeno je na stvarnim signalima, primjenjujući tehniku razdvajanja na kontaminirane signale prije njihove segmentacije. Dodatno, u radu je razmatrana sposobnost razdvojenih signala da prenesu dijagnostički relevantne informacije. U svrhu potvrde učinkovitosti metode razdvajanja signala, provedena je ekstrakcija nekoliko odabranih značajki iz signala u vremenskoj domeni, koje su nakon toga upotrijebljene u postupku klasifikacije signala. Za klasifikaciju signala korišten je algoritam slučajne šume, koji se odlikuje svojom jednostavnošću i efikasnošću. Dobiveni rezultati pokazali su značajan potencijal, ukazujući na robusnost i učinkovitost primijenjene metodologije.

Abstract (english)

This paper addresses the diagnostics of transformer on-load tap changers based on audio recordings. This diagnostic approach presents several challenges: automatic segmentation of audio recordings, separation of useful sound from background noise, and extraction of diagnostically relevant information from the audio data. The issue of automatic segmentation is tackled using convolutional deep neural networks, whose efficiency has been tested on real audio recordings captured during the operation of the tap changer. To solve the signal separation problem, a deep learning approach is also employed, where a deep neural network is trained to separate signals based on the estimation of a time-frequency mask applied to the time-frequency spectrum of the signals. The methodology was tested on real signals by applying the separation technique to contaminated signals before their segmentation. Additionally, the paper explores the ability of the separated signals to convey diagnostically relevant information. To validate the effectiveness of the signal separation method, several selected features were extracted from the time-domain signals and subsequently used in the signal classification process. A random forest algorithm, known for its simplicity and efficiency, was used for signal classification. The obtained results demonstrated significant potential, indicating the robustness and effectiveness of the applied methodology.

• Keywords: regulacijska sklopka; transformator; neinvazivna dijagnostika; audio dijagnostika; duboko učenje
• Keywords (english): OLTC; transformer; non-invasive diagnostics; audio-based diagnostics; deep learning
• Language: croatian
• URN:NBN: urn:nbn:hr:168:030964
• Promotion: 2024
• Study programme: Title: Computer Science; Study programme type: university; Study level: postgraduate; Academic / professional title: Doktor znanosti (Doktor znanosti)
• Type of resource: Text
• Extent: 171 str.
• File origin: Born digital
• Access conditions: Open access
• Created on: 2024-08-29 11:31:10