On anomaly detection in bioreactor processes

Heikkilä, Rasmus

On anomaly detection in bioreactor processes

Files

master_Heikkilä_Rasmus_2019.pdf (3.04 MB)

Perustieteiden korkeakoulu | Master's thesis

Unless otherwise stated, all rights belong to the author. You may download, display and print this publication for Your own personal use. Commercial use is prohibited.

Authors

Heikkilä, Rasmus

Date

2019-12-17

Major/Subject

Systems and Operations Research

Mcode

SCI3055

Degree programme

Master’s Programme in Mathematics and Operations Research

Language

en

Pages

75+9

Abstract

Applying statistical methods for analyzing and monitoring complex industrial pro- cesses has been a popular area of research over the past couple of decades. One particularly attractive field of application is bioreactor processes, which involve liv- ing organisms, such as bacteria or animal cells, and have complex dynamics with significant batch-to-batch variation. These processes are particularly essential in pharmaceutical manufacturing, in which early detection of process anomalies and equipment failure is vitally important. This thesis studies anomaly detection methods for bioreactor processes, including not only detection but also identification of anomalies. In addition, state estimation methods for inferring system state from imperfect measurements are studied. The focus is on extensions of two widely popular statistical methods: principal component analysis and Kalman filtering. The thesis presents two experiments which are performed using data from a simulated penicillin manufacturing process. The first experiment involves estimating the concentrations of cells and penicillin with an Unscented Kalman Filter. While the filter performs reasonably well with normal data, it diverges when a process fault occurs. The filter could be improved further by applying delayed sensor fusion to utilize infrequent cell and penicillin concentration measurements. In the second experiment, a PCA-based model is developed for anomaly detection and identification. The model incorporates dynamic behavior and does not rely on distributional assumptions. In addition, the number of principal components included in the model is determined by cross-validation. The model is able to detect various process faults and correctly identify the variables associated with the fault. Possible directions of further research include robust PCA models and nonlinear models.

Tilastollisten menetelmien soveltaminen monimutkaisten teollisuusprosessien analysointiin ja seurantaan on ollut tärkeä tutkimuskohde viimeisten vuosikymmenten aikana. Bioreaktoriprosessit muodostavat erityisen kiinnostavan sovellusalueen, koska elävät organismit ovat niissä keskeisessä roolissa, mikä tekee niiden käyttäytymisestä hankalasti ymmärrettävää ja aiheuttaa merkittävää eräkohtaista vaihtelua. Nämä prosessit ovat erityisen tärkeitä lääketeollisuudessa, jossa prosessipoikkeamien ja laiterikkojen tunnistaminen mahdollisimman aikaisin on äärimmäisen tärkeää. Tässä työssä tutkitaan poikkeamantunnistusmenetelmien soveltamista bioreaktoriprosesseihin. Työ kattaa sekä poikkeamatilanteiden havaitsemisen että poikkeamaan liittyvien muuttujien tunnistamisen. Lisäksi työssä tutkitaan tilaestimoinnin menetelmiä, joita käytetään systeemin tilan estimointiin epätäydellisten mittausten perusteella. Työ keskittyy pääkomponenttianalyysiin ja Kalman-suotimeen pohjautuvien mallien tarkasteluun. Työ sisältää kaksi koetta, jotka on suoritettu simuloidulla aineistolla. Simulaattori perustuu penisilliinin valmistusta kuvaavaan dynaamiseen malliin. Ensimmäisessä kokeessa pyritään estimoimaan solujen ja penisilliinin konsentraatio systeemissä käyttäen Kalman-suotimeen perustuvaa mallia. Tilaestimaattien tarkkuus on varsin hyvä, mutta poikkeamatilanteissa estimaatit erkaantuvat todellisista arvoista. Suodinta voisi parantaa hyödyntämällä harvoin otettavia solu- ja penisilliinimittauksia. Toisessa kokeessa kehitetään pääkomponenttianalyysiin perustuva poikkeamantunnistusmalli, joka mallintaa systeemin dynaamista käyttäytymistä eikä tee jakaumaoletuksia. Malliin sisällytettävien pääkomponenttien lukumäärä valitaan ristiinvalidointiin perustuen. Malli kykenee havaitsemaan monenlaisia prosessivikoja ja tunnistamaan vikaan liittyvät muuttujat. Mahdollisia kehityssuuntia ovat robustit pääkomponenttianalyysiin perustuvat mallit sekä epälineaariset mallit.

Supervisor

Ilmonen, Pauliina

Thesis advisor

Salmenkaita, Jukka-Pekka
Liski, Antti

Keywords

process monitoring, anomaly detection, state estimation, principal component analysis, bioreactor processes

Permanent link to this item

https://urn.fi/URN:NBN:fi:aalto-201912226595

Collections

[dipl] Perustieteiden korkeakoulu / SCI

Show all metadata

On anomaly detection in bioreactor processes

Files

URL

Journal Title

Journal ISSN

Volume Title

Authors

Date

Department

Major/Subject

Mcode

Degree programme

Language

Pages

Series

Abstract

Description

Supervisor

Thesis advisor

Keywords

Other note

Citation

Permanent link to this item

Collections