On-Line Failure Detection of Vibrating Structures

摘要

Nondestructive evaluation of vibrating structures for failure detection is an area of interest at the Lawrence Livermore National Laboratory. Recent advances in signal processing and analysis of vibrating structures has been applied to this problem with promising results. The techniques used are recursive on-line algorithms which include stochastic models for noise sources. An experiment was designed to investigate the feasibility of signal processing procedures for failure detection. A given structure was excited and its shock response measured using an accelerometer. The response data was digitized and processed. This failure detection problem was approached from the stochastic estimation theory viewpoint. First, we characterize the structural model from experimental data, then construct a failure detector based on this model. A nonlinear identification algorithm (using an extended Kalman filter) is utilized for estimating modal parameters from vibration data. Using this model of the structure, a signal processing estimator is designed to increase the output signal-to-noise ratio. The estimator is followed by a decision device to detect structural failures or anomalies. The results of this feasibility study indicate reasonable performance of these techniques for model characterization and failure detection. Examples and graphical illustrations are presented. (ERA citation 06:029155)