Data-Driven Health Assessment and Remaining Useful Life Prediction for Selective Catalytic Reduction Systems
Redlig, Frank (2025)
2025
All rights reserved. This publication is copyrighted. You may download, display and print it for Your own personal use. Commercial use is prohibited.
Julkaisun pysyvä osoite on
https://urn.fi/URN:NBN:fi:amk-2025060621209
https://urn.fi/URN:NBN:fi:amk-2025060621209
Tiivistelmä
This thesis examines the use of data to analyse and model the deterioration of selective catalytic reduction (SCR) systems, with the aim of developing a concept for estimating the system's health state and remaining useful life (RUL). The mechanisms of SCR catalyst deterioration are complex, and assessment has previously been conducted through spot checks and laboratory tests.
The implementation proposed was a data-driven method that first estimated the health state based on complete historical data from two SCR systems, in the form of health indices, a combined health score, and anomaly detection. Then, it used a Weibull survival model, a parametric degradation model, and a machine learning-based model for the RUL prediction. The result was presented as a generated report for the unit(s) under analysis.
The analysis was done solely on the existing sensor suite and historical data. The results show that it is possible to detect deterioration and potential issues of the SCR system. The RUL prediction provided basic concept models, and the thesis provides ideas for future improvement. Although the health state analysis and RUL prediction showed promise, the validation of the results was not conducted within the thesis and is left as future work.
The implementation proposed was a data-driven method that first estimated the health state based on complete historical data from two SCR systems, in the form of health indices, a combined health score, and anomaly detection. Then, it used a Weibull survival model, a parametric degradation model, and a machine learning-based model for the RUL prediction. The result was presented as a generated report for the unit(s) under analysis.
The analysis was done solely on the existing sensor suite and historical data. The results show that it is possible to detect deterioration and potential issues of the SCR system. The RUL prediction provided basic concept models, and the thesis provides ideas for future improvement. Although the health state analysis and RUL prediction showed promise, the validation of the results was not conducted within the thesis and is left as future work.