Decentralized fault detection and isolation using bond graph and PCA methods

• Published in: International journal of advanced manufacturing technology Vol. 99; no. 1-4; pp. 517 - 529
• Main Authors: Said, Maroua, Fazai, Radhia, Ben Abdellafou, Khaoula, Taouali, Okba
• Format: Journal Article
• Language: English
• Published: London Springer London 01.10.2018; Springer Nature B.V
• Subjects: Algorithms; Architecture; CAE) and Design; Computer simulation; Computer-Aided Engineering (CAD; Diagnostic systems; Engineering; Fault detection; Industrial and Production Engineering; Mechanical Engineering; Media Management; Noise measurement; Original Article; Perturbation; Principal components analysis; Probability density functions; Residues; Subsystems; Fault detection and isolation; Large-scale system; Bond graph; Diagnostic bond graph; Principal component analysis
• ISSN: 0268-3768; 1433-3015
• DOI: 10.1007/s00170-018-2526-4

This paper proposes a new method for fault detection and isolation of a large-scale system by the bond graph (BG) model and the principal component analysis (PCA) technique in a decentralized architecture. The proposed method is entitled the BG-PCA technique. The main objective is to address the problem of monitoring large-scale system components by fault detection and isolation (FDI) methods. In the modeling framework, the BG model is presented by exploiting its structural and causal properties and the diagnostic bond graph (DBG). In measurement noise and perturbation conditions, the probability density function takes place to generate a clear decision procedure to detect the operating mode. In our approach, we firstly generate the structured residues. Furthermore, with a decentralized architecture, we can locate in what subsystem a fault is first detected. Finally, for isolation, the generation of the structured residues method from PCA is exploited to ensure localization. To validate the suggested BG-PCA algorithm, simulations are computed on a three-tank system to show the efficiency of the proposed FDI method, and satisfactory results are found.