Full Text Available
Note: Clicking the button above will open the full text document at the original institutional repository in a new window.
Investigating the performance of time-constrained clustering of multi-modal chemical process data
Olivier, J. F. 2025. Investigating the performance of time-constrained clustering of multi-modal chemical process data. Unpublished masters thesis. Stellenbosch: Stellenbosch University [online]. Available: https://scholar.sun.ac.za/items/5df85a8e-234d-4a22-90ac-7ce2b0b6f59d
Saved in:
| Main Author: | |
|---|---|
| Other Authors: | |
| Format: | Thesis |
| Language: | English |
| Published: |
Stellenbosch : Stellenbosch University
2025
|
| Subjects: | |
| Tags: |
No Tags, Be the first to tag this record!
|
| _version_ | 1867613897372991488 |
|---|---|
| access_status_str | Open Access |
| author | Olivier, Jacobus Frederik |
| author2 | Louw, Tobias Muller |
| author_browse | Louw, Tobias Muller Olivier, Jacobus Frederik |
| author_facet | Louw, Tobias Muller Olivier, Jacobus Frederik |
| author_sort | Olivier, Jacobus Frederik |
| collection | Thesis |
| dc_rights_str_mv | Stellenbosch University |
| description | Olivier, J. F. 2025. Investigating the performance of time-constrained clustering of multi-modal chemical process data. Unpublished masters thesis. Stellenbosch: Stellenbosch University [online]. Available: https://scholar.sun.ac.za/items/5df85a8e-234d-4a22-90ac-7ce2b0b6f59d |
| format | Thesis |
| id | oai:scholar.sun.ac.za:10019.1/132532 |
| institution | Stellenbosch University (South Africa) |
| language | English |
| last_indexed | 2026-06-10T12:43:26.328Z |
| license_str | Other — see source repository |
| provenance_str_mv | Harvested via OAI-PMH from SUNScholar — Stellenbosch University Repository |
| publishDate | 2025 |
| publishDateRange | 2025 |
| publishDateSort | 2025 |
| publisher | Stellenbosch : Stellenbosch University |
| publisherStr | Stellenbosch : Stellenbosch University |
| record_format | dspace |
| source_str | SUNScholar — Stellenbosch University Repository |
| spelling | oai:scholar.sun.ac.za:10019.1/132532 Investigating the performance of time-constrained clustering of multi-modal chemical process data Olivier, Jacobus Frederik Louw, Tobias Muller Stellenbosch University. Faculty of Engineering. Dept. of Chemical Engineering. Chemical processes -- Mathematical models Time-series analysis Algorithms Multivariate analysis UCTD Olivier, J. F. 2025. Investigating the performance of time-constrained clustering of multi-modal chemical process data. Unpublished masters thesis. Stellenbosch: Stellenbosch University [online]. Available: https://scholar.sun.ac.za/items/5df85a8e-234d-4a22-90ac-7ce2b0b6f59d Thesis (MEng)--Stellenbosch University, 2025. ENGLISH ABSTRACT: Unsupervised pattern recognition methods such as data clustering has great potential to analyse visited operational modes and create pseudo-labels of modes and faults for supervised classification algorithms used in process monitoring systems of chemical processes. Conventional clustering algorithms, however, assume the independence of data observations, an assumption violated by chemical time series data. Constrained clustering, specifically time-constrained clustering, offers a promising solution by integrating user knowledge to better account for the autocorrelation in chemical data. Thus, this study developed three time-constrained clustering algorithms based on the conventional K-means, DBSCAN and GMM algorithms. The algorithms were implemented in three different case studies and compared with the conventional algorithms. The algorithms were termed time-constrained K-means (TCK-means), time-constrained DBSCAN (TCDBSCAN) and time-constrained GMM (TCGMM). The first case study evaluated the performance of the algorithms on isolated mean and variance changes in time series data. Step changes in the means and variances were implemented at specific time intervals to create two modes. The DBSCAN and TCDBSCAN algorithms performed unsatisfactorily in obtained accuracy, computational requirement and simplicity of hyperparameter initialisation. TCGMM outperformed the other algorithms by never obtaining lower than 88 % accuracy in correctly identifying the modes. The TCGMM and GMM algorithms were also tested in another dataset where combinations of mean and variance shifts are randomly instantiated based on a conditional probability table (CPT). TCGMM outperformed conventional GMM by obtaining a best accuracy of 89.2 % versus 61.5 % and learned the CPT with an average difference in the main diagonal entries of 1.85 % and an average difference in the off-diagonal entries of 0.653 %. The second case study consisted of a simple mixing tank simulation which was used to assess the performance of TCGMM on a more practical example of operational modes that consist of variance changes. The mixing tank was controlled with level and inventory control that cause the variance shift in the feature space. The control philosophy switched between level and inventory control based on a CPT that represents the randomised changes in control modes calculated by a real-time optimiser. The best unsupervised performance of TCGMM and GMM was 85.9 % and 51.2 % overall accuracy, respectively. The TCGMM algorithm learned the CPT with an average difference in the transition probabilities of 0.04 %. The final case study consisted of a non-isothermal CSTR with 2 normal modes, 5 different faults and 6 secondary faults resulting in 13 operating modes. The model was simulated for six days to create a historical database, and the modes were randomly simulated using a CPT. The TCGMM algorithm managed to achieve higher than 50 % accuracies for 8 of the 13 modes, including four of the six simultaneous faults. TCGMM learned the CPT with an average absolute difference in the main diagonal entries of 0.942 % and an average difference in the off-diagonal entries of 0.118 %. The investigation concluded that the TCGMM algorithm seems to outperform the conventional algorithms in clustering accuracy and seems a valid preprocessing algorithm to label chemical process modes. AFRIKAANSE OPSOMMING: Geen opsomming beskikbaar. Masters 2025-06-10T13:38:10Z 2025-06-10T13:38:10Z 2025-03 Thesis https://scholar.sun.ac.za/handle/10019.1/132532 en Stellenbosch University xii, 115 pages : illustrations application/pdf Stellenbosch : Stellenbosch University |
| spellingShingle | Chemical processes -- Mathematical models Time-series analysis Algorithms Multivariate analysis UCTD Olivier, Jacobus Frederik Investigating the performance of time-constrained clustering of multi-modal chemical process data |
| title | Investigating the performance of time-constrained clustering of multi-modal chemical process data |
| title_full | Investigating the performance of time-constrained clustering of multi-modal chemical process data |
| title_fullStr | Investigating the performance of time-constrained clustering of multi-modal chemical process data |
| title_full_unstemmed | Investigating the performance of time-constrained clustering of multi-modal chemical process data |
| title_short | Investigating the performance of time-constrained clustering of multi-modal chemical process data |
| title_sort | investigating the performance of time constrained clustering of multi modal chemical process data |
| topic | Chemical processes -- Mathematical models Time-series analysis Algorithms Multivariate analysis UCTD |
| url | https://scholar.sun.ac.za/handle/10019.1/132532 |
| work_keys_str_mv | AT olivierjacobusfrederik investigatingtheperformanceoftimeconstrainedclusteringofmultimodalchemicalprocessdata |