Canonical Variate Residuals-Based Fault Diagnosis for Slowly Evolving Faults

Xiaochuan Li (1) , David Mba (2) , Demba Diallo (3) , Claude Delpha (4)

1 DMU - De Montfort University [Leicester, United Kingdom]
2 School of Engineering - Cranfield University
3 GeePs - Laboratoire Génie électrique et électronique de Paris
4 L2S - Laboratoire des signaux et systèmes

This study puts forward a novel diagnostic approach based on canonical variate residuals (CVR) to implement incipient fault diagnosis for dynamic process monitoring. The conventional canonical variate analysis (CVA) fault detection approach is extended to form a new monitoring index based on Hotelling’s T2 , Q and a CVR-based monitoring index, Td . A CVR-based contribution plot approach is also proposed based on Q and Td statistics. Two performance metrics: (1) false alarm rate and (2) missed detection rate are used to assess the effectiveness of the proposed approach. The CVR diagnostic approach was validated on incipient faults in a continuous stirred tank reactor (CSTR) system and an operational centrifugal compressor.

Mots clés: slowly evolving faults; fault detection; fault identification

Domaines: Energie électrique; Systèmes et contrôle [cs.SY]

Xiaochuan Li, David Mba, Demba Diallo, Claude Delpha. Canonical Variate Residuals-Based Fault Diagnosis for Slowly Evolving Faults. Energies, 2019, Fault Diagnosis and Fault-Tolerant Control, 12 (4), pp.726. ⟨10.3390/en12040726⟩. ⟨hal-02084506⟩