Digitala Vetenskapliga Arkivet

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
Applying Machine Learning in the Process Industry: A Quality Management Perspective
Linköping University, Department of Management and Engineering, Logistics & Quality Management. Linköping University, Faculty of Science & Engineering.ORCID iD: 0000-0003-4181-9816
2025 (English)Licentiate thesis, comprehensive summary (Other academic)
Abstract [en]

Is there room for improvement in your production process? Would it not be great if we could assign an algorithm to solve all those problems? Unfortunately, production processes tend to be complex and no algorithm that potent has been identified within the scope of this thesis. But maybe there are ways to use algorithms so that they can provide a bit of help.

Quality Management & Quality Technology is the research area that for decades has worked on the continuous improvement of industrial and other processes. A cornerstone of the methods from Quality Management & Quality Technology is to approach problems in these industrial and other processes in a scientific manner. A big part of this approach is to make decisions based on data: and thereby, statistical methods are an integrated part of the approach in order to understand process behavior. The statistical approach has proven to be very effective, but the specific methods used were developed during the 20th century, for the industry context of that time and within the technical abilities that then existed.

Machine learning (ML) is the concept of data-driven algorithms learning patterns to enable prediction, classification, and decision making. It has exploded as a research topic in recent years and, in the midst of the fourth industrial revolution, a belief is expressed that together with the rapidly growing amounts of industrial data, ML will drastically change industry by making the production processes increasingly efficient and supporting decision making, etc. The big issue is that it is hard to find how ML is to be applied in industry. Would it not make sense to combine the knowledge from the research area that has applied numerical methods and benefited industry for decades and combine that with the new type of numerical methods? In this thesis, the aim has been to do just that. In essence, how can you combine the traditional concepts and knowledge from Quality Management & Quality Technology together with the more novel algorithms from ML? This thesis advances concept generation by assessing the use of statistics in Quality Technology. It highlights the limitations of traditional methodologies in data-intensive environments and proposes ML algorithms as more effective solutions for managing large datasets. Within Root Cause Analysis (RCA), the thesis proposes a specific approach that is useful in paperboard manufacturing. With appropriate adaptations depending on the specific technical case, it is likely that this could also benefit other process industries. Additionally, it explores how ML can enhance statistical process control, offering insights into its potential to deliver more precise guidance and operate effectively across different organizational levels.

Abstract [sv]

Finns det utrymme för förbättring i er produktionsprocess? Skulle det inte vara fantastiskt om man kunde tilldela en algoritm uppgiften att lösa alla dessa problem? Tyvärr så är produktionsprocesser i regel komplexa och inom ramen för den här avhandlingen så har ingen så potent algoritm kunnat identifierats. Men, det kanske finns sätt att använda algoritmer på ett sätt så dom kan bidra till uppgiften?

Forskningsområdet Kvalitetledning & Kvalitetsteknik har i decennier arbetat med ständiga förbättringar av processer, framför allt industriella processer men även andra processer. En hörnsten har varit att angripa de problem som funnits i processerna på ett vetenskapligt sätt. Detta har till stor del handlat om att bygga beslut på data, och därmed har statistiska metoder varit en integrerad del av forskningsområdet då man genom statistik kan skapa en större förståelse för processen. Detta angreppsätt är idag välbeprövat och har bevisats vara väldigt effektivt. Metoderna utvecklades dock under 1900-talet, och är därmed anpassade för den industriella kontext och de tekniska möjligheter som fanns under den tiden.

Maskininlärning är datadrivna algoritmer lär sig mönster för prediktion, klassificering och beslut. Inom den fjärde industriella revolution, som pågår i skrivande stund, tror man att med Maskininlärning och den allt större mängden industriella data som finns tillgänglig drastiskt kommer att kunna förändra industrin genom att göra den mer effektiv, ska möjligheter till bättre beslut, osv. Maskininlärning är ett forskningsområde som har expanderat kraftigt under de senaste åren. Dock är den väldigt svårt att hitta någon som beskriver hur den ska appliceras i industrin.

Skulle det då inte vara rimligt att försöka kombinera den kunskap som förvärvats från det forskningsområde som applicerat och skapat värde genom att använda numeriska metoder inom industrin och kombinera denna kunskap med de nya typerna av numeriska metoder?

Att kombinera dessa två forskningsområden har därför varit målet med denna avhandling. Det vill säga, hur kan man kombinera de koncept och den kunskap som finns inom forskningsområdet Kvalitetledning & Kvalitetsteknik med de algoritmer som finns inom Maskininlärning?

I denna avhandling skapas nya koncept genom att utvärdera de fundamentala idéerna från användandet av statistik inom Kvalitetsteknik. Den beskriver klassiska metoders begränsningar när de sätts i en mer dataintensiv miljö och föreslår maskininlärningsalgoritmer för mer effektiva lösningar i denna typ av större data-set.

För Rotorsaksanalys föreslås ett specifikt tillvägagångsätt som framgångsrikt testats i kartongtillverkning. Det är rimligt att tro att den även kommer fungera på andra industrier, men beroende på de tekniska förutsättningarna kan metoden behöva anpassas.

Utöver det utforskas hur maskininlärning kan utveckla användandet av statistisk processtyrning genom att leverera bättre guidning för användare samt att kunna sprida tillvägagångssättet högre upp inom organisationen.

Place, publisher, year, edition, pages
Linköping: Linköping University Electronic Press, 2025. , p. 42
Series
Linköping Studies in Science and Technology. Licentiate Thesis, ISSN 0280-7971 ; 2016
Keywords [en]
Pulp, Paper, Machine Learning, Quality, Process Industry, Industry
Keywords [sv]
Papper, Massa, Maskininlärning, Kvalitet, Processindustri, Industri
National Category
Reliability and Maintenance
Identifiers
URN: urn:nbn:se:liu:diva-212476DOI: 10.3384/9789181180626ISBN: 9789181180619 (print)ISBN: 9789181180626 (electronic)OAI: oai:DiVA.org:liu-212476DiVA, id: diva2:1946012
Presentation
2025-04-03, C4, C-huset, Campus Valla, Linköping, 13:00 (English)
Opponent
Supervisors
Available from: 2025-03-20 Created: 2025-03-20 Last updated: 2025-03-24Bibliographically approved
List of papers
1. Prediction of Total Quality Performance
Open this publication in new window or tab >>Prediction of Total Quality Performance
2023 (English)In: Proceedings of 26th Excellence in Services International Conference (EISIC) 2023, Università di Verona , 2023Conference paper, Published paper (Refereed)
Abstract [en]

Background and purpose of the paper: In the context of the fourth industrial revolution, thepossibility of developing prediction models has garnered attention for their ability to anticipateproduct properties based on process settings. In paperboard making, in which you have lowsampling rates, with time delays to sampling and about 2000 process variables that affect theoutcome, it is easy to believe that these types of prediction models drastically can improveprocess yield by providing guidance to operators. However, paperboard making is evaluated bymultiple quality properties that intercorrelate with each other and the deployment of numerousprediction models without understanding of normal process variation can lead to an informationoverload, causing tampering and lead to a reduced quality output.Methodology: This paper investigates the feasibility of using machine learning models topredict total quality yield on a full-scale paperboard machine.Main Findings: By changing the focus to an aggregated level, our study aims to alleviate theburden of information overload and reduce the risk of tampering.Originality/Practical implications: The research shows that it is possible to predict the totalquality yield with reasonable accuracy. The approach seems promising, and it is believed thatit could provide insight for operators and support operational management.Type of paper: Full scale case study

Place, publisher, year, edition, pages
Università di Verona, 2023
Keywords
Prediction; Machine learning; Tampering
National Category
Mechanical Engineering
Identifiers
urn:nbn:se:liu:diva-210528 (URN)
Conference
26th Excellence In Services International Conference, 31st August and 1st September, 2023, University of the West of Scotland, Paisley Campus, Paisley, Scotland, UK 
Available from: 2024-12-17 Created: 2024-12-17 Last updated: 2025-03-24

Open Access in DiVA

fulltext(4815 kB)81 downloads
File information
File name FULLTEXT02.pdfFile size 4815 kBChecksum SHA-512
684919a7340bf3500a0f90376e26f8485de4af3ab3629553afeadea38fc5a3489da62b7c41ccdaae58e9db7e2e42bed9c797a271ce96381d4df5ebb599004fb2
Type fulltextMimetype application/pdf
Order online >>

Other links

Publisher's full text

Search in DiVA

By author/editor
Runosson, David
By organisation
Logistics & Quality ManagementFaculty of Science & Engineering
Reliability and Maintenance

Search outside of DiVA

GoogleGoogle Scholar
Total: 82 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
isbn
urn-nbn

Altmetric score

doi
isbn
urn-nbn
Total: 435 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