Research Of Structual Damage Identification Based On Information Fusion And XGBoost

In the course of long-term use of building structure,various forms of structural damage will occur.However,long-term accumulation of damage caused by the structure has a great potential risk.If we fail to detect the damage of the building structure in time,it will probably lead to the destruction of the whole structure.Therefore,it is of great practical significance to study the structural damage identification technology which is simple and effective and can be applied to engineering practice.Based on the problem of building structure damage identification and combined with the Benchmark model proposed by the International Association for Structural Control and the American Society of Civil Engineers,a structural damage identification method based on information fusion and eXtreme Gradient Boosting(XGBoost)is discussed in this dissertation.The main research contents are divided into the following parts:First,this dissertation aims at the structural damage identification problem of single sensor information.As for the problem that the frequency characteristic vectors obtained from normalized relative energy are not sensitive to the damage information,a structural damage identification method based on normalized relative energy of wavelet packet node energy feature and combined with Support Vector Machine(SVM)is proposed.This method can effectively make the difference between different damage conditions concentrated in the sensitive frequency band and improve the accuracy of damage identification.Second,the structural damage identification problem based on multi-sensor information fusion is taken as the research object.Aiming at the different levels of information fusion,the damage identification method based on random vibration response cross-correlation function for data level information fusion and the damage identification method based on feature full fusion for feature level information fusion are analyzed.The experimental results show that the method of structural damage identification based on multi-sensor information fusion can effectively solve the multi-classification problem of structural damage relative to single sensor information.On this basis,a method of damage identification based on random forest feature selection is proposed.The experimental results show that this method can further improve the structural damage identification accuracy.Third,taking XGBoost-based structural damage identification as the research object,aiming at the problem that the SVM classifier is not accurate in solving some multi-classification problems and combining the feature-layer fusion and random forest feature selection,a structural damage identification method based on XGBoost is proposed in this dissertation.The comparative analysis of XGBoost and SVM,stochastic forest,gradient enhancement decision tree algorithm and adaptive lifting algorithm shows that XGBoost has better model accuracy,robustness and computational efficiency.