RUB » Bibliotheksportal

Maria Tauscher, Abdulmagid Bendalla, Guy Chayeb

• The presence of sensor faults in wireless structural health monitoring can significantly affect the quality of structural monitoring, and if remaining undetected, compromises the accuracy and reliability of the assessment conducted on a structure monitored. Fault diagnosis involves four stages: detection, isolation, identification, and accommodation. Approaches based on analytical redundancy using machine learning methods for fault detection and isolation have yielded positive outcomes and proven to be reliable. This paper focuses on the identification stage by identifying certain fault patterns. Some sensor fault types alter the sensor signal according to a certain pattern, making an identification based on statistical indicators possible and allowing back-tracing to the original sensor data. However, other fault types distort the signal significantly by adding a random component or destroy the signal completely, which prevents back-tracing to the original signal. Based on this, in this paper a computationally efficient approach for fault identification based on statistical indicators is presented.This approach enables automated and decentralized fault identification. The clear distinction between various sensor faults on-board enables reliable remote diagnosis and thus a qualified assessment of the maintenance measures required on a SHM system.