Pattern recognition methods for detection of degradation in bridge structures.

摘要

A new approach to signature analysis for bridge structures is introduced. The approach utilizes pattern recognition techniques to detect relatively small cracks in girders. The signatures used in this research are the time-domain response signal and the frequency response function (FRF).;To perform the signature analysis, available pattern recognition techniques are studied. In addition, the following techniques are developed: adaptive template matching, waveform chain code, Distortion Identification Function (DIF), Normalized Frequency Ratio (NFR), Signature Assurance Criterion (SAC), and Cross Signature Assurance Criterion (CSAC).;The data used in this research are obtained from a small simply supported bridge. The structure is asymmetric and weighs about one ton. Acceleration response signals due to impact excitation on the lower flange of the steel girders are used. The excitation is performed using a modally tuned hammer, and the response is picked up using an accelerometer. Degradation is simulated in a typical interior girder by cutting through splice plates using a jigsaw.;Repeatability of results is established using two types of tests. First, by repeating the same cut three times at the same crack location. Second, by performing the experiment with another cut at a different location on the same girder. To test recognition techniques with more realistic signals, signatures of a two-span highway bridge are used. As an alternative source of characteristic signals, strain response sensed by strain gages mounted on the girder flange, is considered.;To study the robustness of recognition techniques, synthetic white noise with uniform probability distribution is added to signatures. To reduce the effect of noise, filtering and smoothing techniques are used.;Using FRF signatures with no synthetic noise, adaptive template matching, waveform chain code, DIF, and SAC detected 1/4-inch long cuts in a four-inch wide flange. When noise is added, adaptive template matching, waveform chain code, and SAC are shown to be robust. All techniques were capable of detecting cuts from both acceleration and strain signals.