Streamlining Data-Centric ML-Ops in the Integrated Control System at ESS
(2025)Department of Automatic Control
- Abstract
- This thesis in collaboration with ESS proposes a workflow for developing, deploying, and integrating machine learning models into the control system. Machine learning offers a promising potential for control systems but often introduces technical debt, with the actual model code representing only a small fraction of the total code base. To address this a data-centric process is proposed for retrieving and selecting control time-series data, training ML models, and integrating them into the control system as process variables. The proposed workflow is tested and validated on the Cryogenic Moderator System (CMS). While the system primarily relies on PID controllers, their performance sometimes suffers from oscillations. Hence, the PID... (More)
- This thesis in collaboration with ESS proposes a workflow for developing, deploying, and integrating machine learning models into the control system. Machine learning offers a promising potential for control systems but often introduces technical debt, with the actual model code representing only a small fraction of the total code base. To address this a data-centric process is proposed for retrieving and selecting control time-series data, training ML models, and integrating them into the control system as process variables. The proposed workflow is tested and validated on the Cryogenic Moderator System (CMS). While the system primarily relies on PID controllers, their performance sometimes suffers from oscillations. Hence, the PID controllers need to be tuned. In this part ML is used to model the system, which enables the possibility to tune the PIDs.
The workflow was evaluated through a case study on the CMS at ESS. The results show that the workflow successfully enabled end-to-end ML integration, from data retrieval to deployment. By using parameter files to drive the process, the system allowed for efficient retraining and redeployment of models. (Less)
Please use this url to cite or link to this publication:
http://lup.lub.lu.se/student-papers/record/9207340
- author
- Gyllenstierna, Christoffer and Ritseson, Victor
- supervisor
- organization
- year
- 2025
- type
- H3 - Professional qualifications (4 Years - )
- subject
- report number
- TFRT-6278
- other publication id
- 0280-5316
- language
- English
- id
- 9207340
- date added to LUP
- 2025-08-08 15:11:17
- date last changed
- 2025-08-08 15:11:17
@misc{9207340,
abstract = {{This thesis in collaboration with ESS proposes a workflow for developing, deploying, and integrating machine learning models into the control system. Machine learning offers a promising potential for control systems but often introduces technical debt, with the actual model code representing only a small fraction of the total code base. To address this a data-centric process is proposed for retrieving and selecting control time-series data, training ML models, and integrating them into the control system as process variables. The proposed workflow is tested and validated on the Cryogenic Moderator System (CMS). While the system primarily relies on PID controllers, their performance sometimes suffers from oscillations. Hence, the PID controllers need to be tuned. In this part ML is used to model the system, which enables the possibility to tune the PIDs.
The workflow was evaluated through a case study on the CMS at ESS. The results show that the workflow successfully enabled end-to-end ML integration, from data retrieval to deployment. By using parameter files to drive the process, the system allowed for efficient retraining and redeployment of models.}},
author = {{Gyllenstierna, Christoffer and Ritseson, Victor}},
language = {{eng}},
note = {{Student Paper}},
title = {{Streamlining Data-Centric ML-Ops in the Integrated Control System at ESS}},
year = {{2025}},
}