kth.sePublications KTH
Change search
CiteExportLink to record
Permanent link

Direct link
Cite
Citation style
  • apa
  • ieee
  • modern-language-association-8th-edition
  • vancouver
  • Other style
More styles
Language
  • de-DE
  • en-GB
  • en-US
  • fi-FI
  • nn-NO
  • nn-NB
  • sv-SE
  • Other locale
More languages
Output format
  • html
  • text
  • asciidoc
  • rtf
Early fault diagnosis in rolling element bearings: comparative analysis of a knowledge-based and a data-driven approach
KTH, School of Industrial Engineering and Management (ITM), Production engineering.ORCID iD: 0000-0003-4240-4946
KTH, School of Industrial Engineering and Management (ITM), Production engineering.ORCID iD: 0000-0002-8222-503x
KTH, School of Industrial Engineering and Management (ITM), Production engineering.ORCID iD: 0000-0003-0045-2085
KTH, School of Industrial Engineering and Management (ITM), Production engineering.ORCID iD: 0000-0001-9185-4607
2024 (English)In: Journal of Intelligent Manufacturing, ISSN 0956-5515, E-ISSN 1572-8145, Vol. 35, no 5, p. 2327-2347Article in journal (Refereed) Published
Abstract [en]

The early identification of a defect that is developing in a bearing is crucial for avoiding failures in rotating machinery. Frequency domain analysis of the vibration signals has been shown to contribute to a better understanding of the nature of a developing defect. Early signs of degradation might be more noticeable in certain frequency bands. The advantages in identifying and monitoring these bandwidths are several: prevention of serious machinery damages, reduction of the loss of investments, and improvement of the accuracy in failure predicting models. This paper presents and compares two approaches for the diagnosis of bearing faults. The first approach was knowledge-based. It relied on principles of mechanics to interpret the measured vibration signals and utilized prior knowledge of the bearing characteristics and testing parameters. The second approach was data-driven whereby data were acquired exclusively from the vibration signal. Both approaches were successfully applied for fault diagnosis by identifying the frequencies of the vibration spectra characteristic for the bearing under study. From this, bandwidths of interest for early fault detection could be determined. The diagnostic abilities of both approaches were studied to analyze and compare their individual strengths regarding the aspects of implementation time, domain knowledge, data processing associated knowledge, data requirements, diagnostic performance, and practical applicability. The advantages, apparent limitations as well as avenues for further improvement of both approaches are discussed.

Place, publisher, year, edition, pages
Springer Nature , 2024. Vol. 35, no 5, p. 2327-2347
Keywords [en]
Fault diagnosis, Data-driven, Knowledge-based, Rolling elements bearings, Vibration, Degradation
National Category
Other Electrical Engineering, Electronic Engineering, Information Engineering
Research subject
Production Engineering; Computer Science
Identifiers
URN: urn:nbn:se:kth:diva-336688DOI: 10.1007/s10845-023-02151-yISI: 001010026300002Scopus ID: 2-s2.0-85162019317OAI: oai:DiVA.org:kth-336688DiVA, id: diva2:1797982
Funder
Vinnova, 2018-05033
Note

QC 20231023

Available from: 2023-09-18 Created: 2023-09-18 Last updated: 2026-05-13Bibliographically approved
In thesis
1. Towards Reliable Diagnostics Under Data Scarcity - Machine Learning for Manufacturing Equipment
Open this publication in new window or tab >>Towards Reliable Diagnostics Under Data Scarcity - Machine Learning for Manufacturing Equipment
2026 (English)Doctoral thesis, comprehensive summary (Other academic)
Abstract [en]

The ongoing digitalization of industrial systems is transforming maintenance practices by enabling continuous monitoring and data collection. This creates the foundation for data-driven approaches and enables advanced diagnostics through the use of machine learning methods in maintenance applications. To achieve reliable diagnostic performance, supervised machine learning methods rely on representative training data covering diverse operating conditions and characteristic signatures of different failure modes, with sufficient data quality, accurate labels, and proper annotations. However, in industrial maintenance, these requirements are commonly not satisfied due to heterogeneous operating regimes, harsh data acquisition conditions, and the inherently rare occurrence of faults. At the same time, maintenance applications are often safety-critical and associated with significant operational and economic risks, which motivates the need for reliable diagnostics even under data-constrained conditions.

This thesis treats data scarcity as an inherent and largely unavoidable constraint of industrial maintenance and develops a structured approach to characterize it, assess its effects on diagnostic reliability, and identify effective strategies to operate under such limitations. Data scarcity is formally defined as a multidimensional concept encompassing five dimensions: availability, coverage, representativeness, usability, and quality, establishing a framework for systematically assessing data-related limitations in industrial monitoring. Diagnostic reliability is characterized along three properties: accuracy, generalization, and robustness, which together define the basis for evaluating machine learning-based diagnostics under data scarcity.

The effects of data scarcity on diagnostic reliability are investigated through structured empirical studies that systematically vary controlled data scarcity factors, including data volume, fault sample ratio, and measurement degradation, across three transfer scenarios with increasing domain shift. Within these scenarios, classical machine learning and deep learning methods are combined with different knowledge transfer strategies, including domain adaptation, transfer learning, and joint learning, to examine how individual factors, their interactions, and the choice of learning strategy jointly determine diagnostic performance. This factorial approach connects the conceptual characterization of data scarcity directly to empirical evaluation, enabling quantitative assessment and analytical interpretation of model behavior under realistic industrial constraints.

The results establish that model effectiveness is regime-dependent: no single learning strategy is universally optimal, and diagnostic performance is governed by the interaction between data characteristics, domain conditions, and the learning strategy employed. The findings are consolidated into practical insights for scarcity-aware machine learning, providing actionable guidance for data acquisition, model selection, and the design of reliable fault detection systems for condition-based maintenance in industrial environments.

Abstract [sv]

Den pågående digitaliseringen av industriella system förändrar underhållspraxis genom att möjliggöra kontinuerlig övervakning och datainsamling. Detta skapar grunden för datadrivna metoder och möjliggör avancerad diagnostik genom användning av maskininlärning i underhållstillämpningar. För att uppnå tillförlitlig diagnostisk prestanda kräver övervakade maskininlärningsmetoder representativa träningsdata som täcker olika driftförhållanden och karakteristiska signaturer för olika feltyper, med tillräcklig datakvalitet, korrekta etiketter och korrekt annotering. I industriellt underhåll är dessa krav dock ofta inte uppfyllda på grund av heterogena driftregimer, svåra förhållanden vid datainsamling och den inneboende sällsyntheten av felförekomster. Samtidigt är underhållstillämpningar ofta säkerhetskritiska och förknippade med betydande operativa och ekonomiska risker, vilket motiverar behovet av tillförlitlig diagnostik även under databegränsade förhållanden.

Denna avhandling behandlar databrist som en inneboende och till stor del oundviklig begränsning inom industriellt underhåll och utvecklar ett strukturerat tillvägagångssätt för att karaktärisera den, bedöma dess effekter på diagnostisk tillförlitlighet och identifiera effektiva strategier för att hantera sådana begränsningar. Databrist definieras formellt som ett flerdimensionellt begrepp som omfattar fem dimensioner: tillgänglighet, täckning, representativitet, användbarhet och kvalitet, vilket etablerar ett ramverk för systematisk bedömning av datarelaterade begränsningar inom industriell övervakning. Diagnostisk tillförlitlighet karaktäriseras längs tre egenskaper: noggrannhet, generaliseringsförmåga och robusthet, vilka tillsammans utgör grunden för utvärdering av maskininlärningsbaserad diagnostik under databrist.

Effekterna av databrist på diagnostisk tillförlitlighet undersöks genom strukturerade empiriska studier som systematiskt varierar kontrollerade databristfaktorer, inklusive datavolym, andel felprover och mätningsdegradation, över tre överföringsscenarier med ökande domänskift. Inom dessa scenarier kombineras klassiska maskininlärningsmetoder och djupinlärningsmetoder med olika kunskapsöverföringsstrategier, inklusive domänanpassning, överföringsinlärning och gemensam inlärning, för att undersöka hur enskilda faktorer, deras interaktioner och valet av inlärningsstrategi gemensamt bestämmer diagnostisk prestanda. Detta faktoriella tillvägagångssätt kopplar den konceptuella karaktäriseringen av databrist direkt till empirisk utvärdering, vilket möjliggör kvantitativ bedömning och analytisk tolkning av modellbeteende under realistiska industriella begränsningar.

Resultaten fastställer att modelleffektivitet är regimberoende: ingen enskild inlärningsstrategi är universellt optimal, och diagnostisk prestanda styrs av interaktionen mellan dataegenskaper, domänförhållanden och den använda inlärningsstrategin. Resultaten sammanställs till praktiska insikter för databristmedveten maskininlärning, vilka ger handlingsorienterad vägledning för datainsamling, modellval och utformning av tillförlitliga feldetekteringssystem för tillståndsbaserat underhåll i industriella miljöer.

Place, publisher, year, edition, pages
Sweden: KTH Royal Institute of Technology, 2026. p. 175
Series
TRITA-ITM-AVL ; 2026:12
National Category
Production Engineering, Human Work Science and Ergonomics Reliability and Maintenance Artificial Intelligence
Research subject
Production Engineering
Identifiers
urn:nbn:se:kth:diva-381136 (URN)978-91-8106-625-8 (ISBN)
Public defence
2026-06-05, F3, Lindstedtvägen 26-28, Stockholm, 14:00 (English)
Opponent
Supervisors
Available from: 2026-05-13 Created: 2026-05-13 Last updated: 2026-05-13Bibliographically approved

Open Access in DiVA

fulltext(3738 kB)533 downloads
File information
File name FULLTEXT01.pdfFile size 3738 kBChecksum SHA-512
ab260fbdc38c684a2add6282f2e33f630d772140c8f4bb81e2ddc40012f3772b42d8988261adcb610a71ecf4c070f94b3918ffbbf3905d0f60ce23fe72d9748e
Type fulltextMimetype application/pdf

Other links

Publisher's full textScopus

Authority records

Iunusova, EleonoraGonzalez, MonicaSzipka, KárolyArchenti, Andreas

Search in DiVA

By author/editor
Iunusova, EleonoraGonzalez, MonicaSzipka, KárolyArchenti, Andreas
By organisation
Production engineering
In the same journal
Journal of Intelligent Manufacturing
Other Electrical Engineering, Electronic Engineering, Information Engineering

Search outside of DiVA

GoogleGoogle Scholar
Total: 534 downloads
The number of downloads is the sum of all downloads of full texts. It may include eg previous versions that are now no longer available

doi
urn-nbn

Altmetric score

doi
urn-nbn
Total: 762 hits
CiteExportLink to record
Permanent link

Direct link
Cite
Citation style
  • apa
  • ieee
  • modern-language-association-8th-edition
  • vancouver
  • Other style
More styles
Language
  • de-DE
  • en-GB
  • en-US
  • fi-FI
  • nn-NO
  • nn-NB
  • sv-SE
  • Other locale
More languages
Output format
  • html
  • text
  • asciidoc
  • rtf