STATISTICAL PATTERN RECOGNITION FOR STRUCTURAL HEALTH MONITORING USING ESN FEATURE EXTRACTION METHOD

Jianxi Yang, Gaocen Sha, Yingxin Zhou, Guiping Wang, and Boren Zheng

Keywords

Statistical pattern recognition, nonlinear feature extraction, ESN, dynamic response signal, damage sensitivity index, bridge monitoring

Abstract

For structural dynamic response signal, as the traditional feature extraction methods based on statistical pattern recognition are linear and insensitive, this paper proposes a new signal feature extraction method based on the echo state network (ESN). The proposed method is suitable for analysis of signal with non-linear characteristics. It is more sensitive on the damage identification than the existing methods based on statistical pattern recognition. The proposed method collects structural dynamic response signals under different conditions and uses the ESN network for system identification. The output weights are served as the damage feature values of the bridge structure. A number of experiments are performed by using the non-linear vibration finite element model and a real bridge scale model under environmental motivation. The experimental results show that the proposed non-linear feature extraction method based on ESN is more sensitive than the traditional auto-regressive model. There are obvious differences in the damage sensitivity index value between the healthy case and the damage case. The damage sensitivity index increases linearly with the structural degradation. Moreover, this paper constructs the damage sensitive index by using the Euclidean distance which is consistent with the evolution trend of the structural state. The proposed method offers a more appropriate theoretical and technical support for the existing bridge monitoring data processing and the structural damage evolution trend assessment.

Important Links:



Go Back