University of Twente Student Theses

Login

Model-based Probabilistic Diagnostics of Cyber-Physical Systems

Grievink, Guus (2022) Model-based Probabilistic Diagnostics of Cyber-Physical Systems.

[img] PDF
664kB
Abstract:With model-based diagnostics, a machine is modelled with the intention to trace back the cause of a particular machine failure. Where traditionally this is done with logical statements, a probabilistic approach can be used by translating a diagnostic problem into a Bayesian network. The research in this paper expands on the work of the TNO’s Embedded Systems Institute (ESI) which, among other things, researches the Bayesian approach to model-based diagnosis. In the Bayesian models of ESI, a uniform prior probability is assumed on the health states of components. However, this does probably not hold in practice, as some components might be more likely to fail than others. The research tries to determine whether including probabilistic information reflecting the real-world scenario improves the diagnostic capabilities of Bayesian diagnostic models. A simple theoretical system consisting of water pipes is used to achieve this. A network containing probabilities of component failure reflecting the used data is compared to a uniform model. Experiments on the test data show that the adjusted model, on average, performs better than the uniform model. However, other observations indicate that the used structure for the Bayesian network might not be optimal for including probabilities of component failure from real-world data. Suggestions are made to mitigate these issues, which are up for future research.
Item Type:Essay (Bachelor)
Faculty:EEMCS: Electrical Engineering, Mathematics and Computer Science
Subject:30 exact sciences in general, 50 technical science in general, 54 computer science
Programme:Business & IT BSc (56066)
Link to this item:https://purl.utwente.nl/essays/91937
Export this item as:BibTeX
EndNote
HTML Citation
Reference Manager

 

Repository Staff Only: item control page