Application of Statistical Pattern Classification Methods for Damage Detection to Field Data

摘要

The field of Structural Health Monitoring (SHM) has received considerable attention for its potential applications to monitoring civil infrastructure. However, the damage detection algorithms that form the backbone of these systems have primarily been tested on simulated data instead of full-scale structures because of the scarcity of real structural acceleration data. In response to this deficiency in testing, we present the performance of two damage detection algorithms used with ambient acceleration data collected during the staged demolition of the full-scale Z24 Bridge in Switzerland. The algorithms use autoregressive coefficients as features of the acceleration data and hypothesis testing and Gaussian Mixture Modeling to detect and quantify damage. While experimental or numerically simulated data have provided consistently positive results, field data from real structures, the Z24 Bridge, show that there can be significant false positives in the predictions. Difficulties with data collection in the field are also revealed pointing to the need for careful signal conditioning prior to algorithm application.