Fault Detection in Open-Loop Controlled Structures

Abstract

In the past decades, various algorithms have been proposed to detect failed structural members, sensors, and actuators. Most of these works make the assumption that the fault has already been isolated to either the sensors, actuators, or the structure itself. This work provides a novel technique for system-failure detection and classification using the eigensystem realization algorithm and outlier analysis. The technique is based on the comparison of the state-space-form realizations between the current system and the healthy system. Three different failure types of a dynamic system (sensor failure, actuator failure, and structural failure) are studied and compared to determine the unique statistical characteristics of each type of failure mode. Six coefficients representing these patterns are extracted from the realizations of the current system and compared with the generalized benchmarks of the healthy system so that system failure can be diagnosed. An expert system is also developed to automate the diagnosis process. Simulations and experiments successfully validate this work.