On-line failure detection of a vibrating structure: A model-based approach

Abstract

Model-based failure detection is based on the principle that the MBP for a normal or pristine structure is developed first and tuned during the calibration stage assuring a statistically validated processor. Once developed, the MBP is used as the integral part in a sequential detection scheme to decide whether or not the structure under investigation is operating normally. When an abnormality is detected, a failure alarm is activated and the type of failure is classified based on a library of potential failure modes. Here we use high-order parametric models to capture the essence of the structures over a limited frequency band known to be sensitive to structural changes. These estimated or identified models for normal operations are then used to develop the MBP which in this instance is a recursive Kalman filter. The filter is known to produce zero-mean/white residuals when optimally tuned to the data. Failure is declared when these properties are no longer valid. Once the detection is accomplished, the next step is to classify the type of failure mechanism and eventually the locations. Here we show results of the designs on both simulated and measured data.