Publication
Fault diagnosis using interval data-driven LPV observers and structural analysis
Conference Article
Conference
IFAC Symposium on Fault Detection Supervision and Safety for Technical Processes (SAFEPROCESS)
Edition
12th
Pages
25-30
Doc link
https://doi.org/10.1016/j.ifacol.2024.07.188
File
Abstract
This paper presents a Fault Detection and Isolation (FDI) method that combines Structural Analysis (SA) and machine learning data-driven techniques. The adaptive network fuzzy inference system (ANFIS) is used to obtain a model of the real system to be monitored using historical data from non-faulty scenarios. The structure of the model is given by the SA by means of graphical or textual description of the system. The obtained model is formulated as an observer where unknown but bounded process and sensor noises are considered. Then, LPV observers are used to carry out the fault diagnosis using model residuals. Once observers detect a residual inconsistency, a fault alarm is raised and the fault isolation triggered. Fault isolation is carried out through the transformation of residuals using the Kramer function and the Dempster-Shafer theory. The proposed fault isolation method considers the plausibility of the activation of each residual and the SA information to provide the most probable fault. Finally, a well-known four-tanks system illustrates the performance and results of the proposed method.
Categories
control theory.
Scientific reference
X. Fang, J. Blesa and V. Puig. Fault diagnosis using interval data-driven LPV observers and structural analysis, 12th IFAC Symposium on Fault Detection Supervision and Safety for Technical Processes, 2024, Ferrara, Vol 58 of IFAC Papers Online, pp. 25-30.
Follow us!