Abstract
dc:description.abstractThis thesis advances data-driven process monitoring for detecting, locating, and interpreting faults in industrial processes. First, it introduces a fault detection architecture that reduces data from the plant and a reference model to fewer dimensions and compares the lower-dimensional data to detect faults despite plant--model mismatch. Second, it presents a knowledge-based method that uses the plant’s process flow diagram, sensor placement, and control scheme to define monitoring blocks. This decomposition helps locate faults within plant sectors, while a likelihood-based contribution map identifies abnormal variables and traces fault propagation. Third, it statistically evaluates autoencoders based on Kolmogorov--Arnold networks for unsupervised fault detection, identifying data-efficient variants when limited fault-free training data are available. Finally, it introduces a method that adjusts fault detection scores based on their recent behavior during persistent faults. The method maintains continuous alarms when these scores repeatedly cross the detection threshold, while retaining simple alarm logic.
Degree
thesis:*- Department dc:contributor.department
- Chemical Engineering
- Year dc:date.issued
- 2026
Author and committee
dc:creator, dc:contributor.*- Author dc:creator
-
- Luna Villagómez, Enrique
- Advisor dc:contributor.advisor
-
- Mahalec, Vladimir
Subjects
dc:subject × 4Rights
dc:rights- Statement dc:rights
-
- Attribution 4.0 International
- Licence dc:rights.uri
- Language dc:language.iso
- en
Identifiers
dc:identifier.*- Identifier URI
- https://doi.org/10.71548/25562