Structural damage identification and experiment based on FBG sensors and PCA-KNN approach

Abstract

Structural damage has the characteristics of concealment and low response, therefore, how to improve the accuracy and stability of damage identification has always been a challenge for the researchers. An innovative approach (PCA-KNN) for structural damage identification and anomaly localization was proposed based on monitoring data. Firstly, traditional principal component analysis (PCA) was completed on the monitoring data matrix with taking into account the contribution of error subspaces. Secondly, the sigmoid function was assigned corresponding weights in terms of the sensitivity of the principal components. Finally, the comprehensive differential index was developed with the K-Nearest Neighbor (KNN) algorithm for less noise interference. In terms of sensor monitoring, the wide range FBG strain sensor and FBG tilt sensor were developed to perceive structural mechanical parameters. The innovative approach and sensors were applied to the benchmark model of Base Excited 3-Story Structure and shaking table testing to implement the comprehensive index calculation, which could effectively identify structural damage with abnormal monitoring data from the sensor nearby. Through comparison and analysis, the new technology could promote the ability of quantitative damage identification and localization.