Damage Identification Algorithm For Truss Structures Based On Dynamic And Static Response

With the development of science and technology and industry,damage location identification and damage degree determination of truss structures have become an important research direction of truss structure health inspection and safe use.Truss structure will cause damage during service,if not timely detection,may lead to sudden failure of truss structure,resulting in economic loss and even lead to danger.Therefore,using health inspection and damage identification method to prevent the failure risk of truss structure has scientific significance in theory and application value in engineering.Gradient Boosted Tree(GBDT)is an integrated learning algorithm that uses parallel computation in model prediction to output prediction results,and can effectively perform truss structure damage identification based on truss damage database.Empirical Modal Decomposition(EMD)can deal with nonlinear and non-smooth problems in the signal,and can also perform local characterization of the signal.The isolated forest(IForest)algorithm is an outlier detection algorithm based on a tree model.The Missing Forest(MF)algorithm is a nonlinear modeling data complementation algorithm.Principal Component Analysis(PCA)is a data dimensionality reduction algorithm.Fast Fourier Transform(FFT)is a method of time domain frequency domain transform analysis.In this paper,a comprehensive analysis of truss damage data structure and the respective advantages of algorithms are presented,and a truss structure damage identification method based on EMD,FFT,PCA,IForest,MF and GBDT is proposed.The main research contents of this thesis are as follows.1.Establishing truss structure damage database.The truss damage database is divided into dynamic response inherent frequency and static response deflection.The damage deflection of truss is obtained by using displacement meter,and the collected truss vibration damage signal is decomposed by using empirical modal decomposition,and the frequency domain features are extracted by fast Fourier transform.The damage deflection and frequency domain features of the truss structure are composed into a truss damage database.2.The combination of principal component analysis,isolated forest,missing forest and gradient boosting tree is used to study the truss structure damage identification problem.The principle of principal component analysis,isolated forest,missing forest and gradient boosted tree is given to identify the damage of truss structure by using the model of combined dynamic and static response,and the corresponding damage identification method of truss structure is established.The method makes up for the deficiencies of traditional machine learning damage recognition and effectively improves the accuracy of truss damage recognition.3.Gradient lifting tree classification algorithm is used to identify the damage location of truss structures,and Gradient lifting tree regression algorithm is used to determine the damage degree of truss structures.The truss damage data set was optimized through dimensionality reduction,anomaly detection and data completion processing of the truss damage database,and the position and degree of damage of the truss structure were determined by gradient lifting tree.4.EMD,FFT,PCA,IFOREST,MF and GBDT are used for truss structure damage identification.The results show that the algorithm can accurately identify the damage of truss structure,and the accuracy of identifying the damage location of truss structure reaches 90%,and the error of predicting the damage degree of truss is within 2%,so the EMD,FFT,PCA,IForest,MF and GBDT truss structure damage identification methods established in this thesis have reference value in engineering applications.