Full Text Available
Note: Clicking the button above will open the full text document at the original institutional repository in a new window.
State-based decision support for condition shifting of multimodal continuous processes
Thesis (MEng)--Stellenbosch University, 2021.
Saved in:
| Main Author: | |
|---|---|
| Other Authors: | |
| Format: | Thesis |
| Language: | en_ZA |
| Published: |
Stellenbosch : Stellenbosch University
2021
|
| Subjects: | |
| Tags: |
No Tags, Be the first to tag this record!
|
| _version_ | 1867614024570503168 |
|---|---|
| access_status_str | Open Access |
| author | Noelle, Francois Frieder |
| author2 | Louw, Tobias M. |
| author_browse | Louw, Tobias M. Noelle, Francois Frieder |
| author_facet | Louw, Tobias M. Noelle, Francois Frieder |
| author_sort | Noelle, Francois Frieder |
| collection | Thesis |
| dc_rights_str_mv | Stellenbosch University |
| description | Thesis (MEng)--Stellenbosch University, 2021. |
| format | Thesis |
| id | oai:scholar.sun.ac.za:10019.1/110074 |
| institution | Stellenbosch University (South Africa) |
| language | en_ZA |
| last_indexed | 2026-06-10T12:45:27.799Z |
| license_str | Other — see source repository |
| provenance_str_mv | Harvested via OAI-PMH from SUNScholar — Stellenbosch University Repository |
| publishDate | 2021 |
| publishDateRange | 2021 |
| publishDateSort | 2021 |
| publisher | Stellenbosch : Stellenbosch University |
| publisherStr | Stellenbosch : Stellenbosch University |
| record_format | dspace |
| source_str | SUNScholar — Stellenbosch University Repository |
| spelling | oai:scholar.sun.ac.za:10019.1/110074 State-based decision support for condition shifting of multimodal continuous processes Noelle, Francois Frieder Louw, Tobias M. Bradshaw, S. M. Auret, Lidia Stellenbosch University. Faculty of Engineering. Dept. of Process Engineering. Multimodal continuous processes Steady state analysis Electronic data processing -- Distributed processing Decision support systems Process control -- Automatic control Gaussian processes UCTD Thesis (MEng)--Stellenbosch University, 2021. ENGLISH ABSTRACT: Automated distributed control systems are able to keep controlled variables at their set points and thus are able to maintain process operation within an operating mode (quasi-steady state). These automated systems may however not be informed of complex issues within the process, as a result, human supervision is required within the control loop. These supervisors are therefore required to perform manual actions to allow a process to settle to safer or more profitable operating conditions. Modern industrial continuous processes thus undergo frequent state shifts either due to set point changes or sustained disturbances. Fundamental process models may assist supervisors in the evaluation of process conditions beyond their empirical experience, however, the development of such models is difficult and may require significant effort. Due to the existence of distributed control systems, many process plants have large historical databases of past sensor measurements. Data-driven approaches to process monitoring are therefore applicable. This investigation aims to discover the various steady states conditions (process modes), their economic performance, and switching conditions from the continuous process’s multimodal historical data. This knowledge can then be leveraged to provide decision support to supervisors, which would allow them to operate the process more profitably. Since historical process data is mostly not classified into its various states, an unsupervised data-driven approach is imperative. Specifically, the approach is developed and evaluated on synthetic multimodal data obtained from a propylene glycol reactor simulation, and finally implemented on actual industrial milling circuit data. The developed state-based decision support model made use of principal component analysis, stationarity analysis, K-means clustering, Gaussian mixture models, and key performance indicators in an integrated manner. Stationarity analysis was able to effectively detect and thus remove transient states from the simulated CSTR data, however, considerable process knowledge was required in setting the algorithm hyperparameters. K-means clustering was utilized to provide initial parameter estimates for the Gaussian mixture model fitting. The best model configurations were selected based on the lowest Bayesian information criterion, however, the suggested best model usually overfit the data. Additional model refinement was therefore required such that each process mode or steady-state was described by a single Gaussian within the model. These refining procedures made use of the data sequence, Euclidean distance between Gaussians, and their prior probabilities. The results showed that even if transient states are not removed prior to the analysis, relatively good monitoring performance can be achieved with the developed approach. Further, contribution plots were utilised to identify the key variables that may have resulted in a transition. As a result, useful decision support can be provided to process supervisors. An algorithm was developed which summarises high dimensional correlated historical process data into a two-dimensional process state “map”, which can effectively assist supervisors in navigating complex multimodal continuous processes. Further, expert knowledge can continually be leveraged to refine the process map and its corresponding model since the data-driven approach emphasizes human-machine interactions. The decision support system worked well on simulated CSTR data, however, more advanced procedures are required to “diagnose” the causes of transitions within industrial process data. AFRIKAANSE OPSOMMING: Geoutomatiseerde verspreide beheerstelsels kan veranderlikes by hul setpunte beheer en kan prosesbedryf binne ’n bedryfsmodus handhaaf (quasi-bestendige toestand). Hierdie geoutomatiseerde stelsels kan egter nie kennis neem van komplekse probleme binne die proses nie, en as ’n gevolg, word menslike toesig benodig om aksies met die hand uit te voer om ’n proses na veiliger of meer winsgewende bedryfskondisies te bring. Moderne industriële aaneenlopende prosesse gaan dus gereelde toestand veranderinge deur as gevolg van setpuntveranderinge, of aanhoudende steuringe. Fundamentele prosesmodelle kan toesighouers assisteer in die evaluasie van proseskondisies verder as hul empiriese ondervinding, maar die ontwikkeling van sulke modelle is moeilik en mag beduidende moeite vereis. As gevolg van die bestaan van verspreide beheerstelsels, het baie prosesaanlegte groot historiese databasisse van vorige sensormates. Datagedrewe benaderinge tot prosesmonitering is daarom toepaslik. Hierdie ondersoek mik om die verskillende bestendige toestande (prosesmodus), hul ekonomiese doeltreffendheid, en omruilkondisies van die aaneenlopende proses se multimodale historiese data, te ontdek. Hierdie kennis kan dan gebruik word om besluitondersteuning aan toesighouers te verskaf, wat hulle dan sal toelaat om die proses meer winsgewend te bedryf. Aangesien historiese prosesdata meestal nie in sy verskeie toestande geklassifiseer is nie, is ’n ongekontroleerde datagedrewe benadering noodsaaklik. Meer spesifiek, die benadering is ontwikkel en geëvalueer op sintetiese multimodale data verkry van ’n propileen-glikolreaktorsimulasie, en uiteindelik geïmplementeer op ʼn werklike industriële malery se stroomdata. Die ontwikkelde toestand-gebaseerde besluit ondersteuning model het gebruik gemaak van hoofkomponentanalise, stasionariteitanalise, K-gemiddelde groepering, Gauss-mengselmodelle, en sleutel doeltreffendheidsindikators op ’n geïntegreerde manier. Stasionariteitanalise het oorgangstoestande effektief opgespoor en dus van die gesimuleerde KGTR-data verwyder, maar aansienlike proseskennis was vereis om die algoritme hiperparameters te stel. K-gemiddelde groepering is gebruik om aanvanklike parameterberamings te verskaf vir die Gauss-mengselmodelpassing. Die beste modelkonfigurasies is gekies gebaseer op die laagste Bayesiaanse informasie kriteria, maar die voorgestelde beste model het gewoonlik die data oorgepas. Addisionele modelverfyning was daarom nodig sodat elke prosesmodus of bestendige toestand deur ’n enkel Gauss-kromme binne die model beskryf kon word. Hierdie verfyningprosedures het gebruik gemaak van die datareekse, Euklidiese afstand tussen Gauss-krommes, en hul vorige waarskynlikhede. Die resultate het gewys dat selfs as oorgangstoestande nie voor die analise verwyder is nie, kan relatief goeie moniteringsdoeltreffendheid bereik word met die ontwikkelde benadering. Verder, verspreidingsplotte is gebruik om die sleutel veranderlikes te identifiseer wat ’n oorgang tot gevolg kon hê. As resultaat, kan bruikbare besluitondersteuning verskaf word aan prosestoesighouers. ’n Algoritme is ontwikkel wat hoë dimensionele, gekorreleerde, historiese data opgesom het en in ’n twee-dimensionele prosestoestand gekarteer het, wat effektiewelik gebruik kan word om toesighouers te assisteer om komplekse multimodale aaneenlopende prosesse te navigeer. Verder, deskundige kennis kan deurentyd gebruik word om die proseskaart en sy korresponderende model te verfyn aangesien die 3 datagedrewe benadering mens-masjien interaksie beklemtoon. Die besluit ondersteuning sisteem het goed gewerk op gesimuleerde KGTR-data, maar meer gevorderde prosedures is nodig om die oorsake van oorgange binne industriële prosesdata te diagnoseer. Masters 2021-03-08T11:08:39Z 2021-04-21T14:39:04Z 2021-03-08T11:08:39Z 2021-04-21T14:39:04Z 2021-03 Thesis http://hdl.handle.net/10019.1/110074 en_ZA Stellenbosch University 137 pages application/pdf Stellenbosch : Stellenbosch University |
| spellingShingle | Multimodal continuous processes Steady state analysis Electronic data processing -- Distributed processing Decision support systems Process control -- Automatic control Gaussian processes UCTD Noelle, Francois Frieder State-based decision support for condition shifting of multimodal continuous processes |
| title | State-based decision support for condition shifting of multimodal continuous processes |
| title_full | State-based decision support for condition shifting of multimodal continuous processes |
| title_fullStr | State-based decision support for condition shifting of multimodal continuous processes |
| title_full_unstemmed | State-based decision support for condition shifting of multimodal continuous processes |
| title_short | State-based decision support for condition shifting of multimodal continuous processes |
| title_sort | state based decision support for condition shifting of multimodal continuous processes |
| topic | Multimodal continuous processes Steady state analysis Electronic data processing -- Distributed processing Decision support systems Process control -- Automatic control Gaussian processes UCTD |
| url | http://hdl.handle.net/10019.1/110074 |
| work_keys_str_mv | AT noellefrancoisfrieder statebaseddecisionsupportforconditionshiftingofmultimodalcontinuousprocesses |