Modal Parameter Automatic Identification For Structures Under Ambient Excitation And Algorithm Optimization

Experimental modal analysis has important engineering application in structuralhealth monitoring, damage identification and dynamic characteristic analysis, etc. Itsanalyzing steps mainly includes data collection, data processing, parameteridentification, and model verification, in which the modal parameter identification isthe core content of the experimental modal analysis. During data collecting, the data ofall freedoms is usually collected at a single time, however the data should be collectedin different groups when the number of acquisition channels is less than the number ofstructure’s freedoms. SSI(Stochastic subspace identification) is the most widely usedtime-domain method in single measurement analysis; ERA(eigensystem realizationalgorithm) is a commonly used method in multi-component measurement analysis;while CWT (continuous wavelet transform)-based modal parameters identification isgood at identifying low-frequency close modal, all of them are studied detailedly inthis thesis. Today, order selection and modal selection for SSI and ERA needs manualparticipation, the automation degree is low, so the research about automatic modalparameter identification is urgently needed. Moreover, with the constantly increasingof analysis data amount, Data-SSI(data driven stochastic subspace identification) andCWT-based modal parameters identification exhibit a low efficiency in identification,so the algorithm of the modal parameter identification needs to be improved. Theexperimental modal systems used in some Chinese important domains are mainlydeveloped by foreign corporations, domestic experimental modal systems are still indevelopment stage, and the gap between domestic systems and foreign maturecommercial systems is still obvious, so developing experiment modal analysis systemwith self-propery right and perfect functions to narrow this gap is very important. Thisthesis studies the automatic modal parameter identification and algorithm optimizationfor some time-domain method such as stochastic subspace identification method, theeigensystem realization algorithm, and a analysis system with self-propery right andperfect functions has also been developed. The main contents and conclusions of thethesis are as follows:①Stochastic subspace identification which is a most widely used time-domainmethod in single measurement analysis is studied. Due to its low automation degree,the hierarchical clustering method is adopted to automatically pick up modal. Automatic modal analysis is realized with the help of the hierarchical clusteringmethod, classify the results into several categories according the similarity between theresults with eigenfrequencies, damping ratios, mode shapes and mode energy, somecategories will be selected if the number of its elements is large enough. In order toreduce the influence by spurious modes on modal selection, a criterion named modelsimilarity index is proposed which can effectively indicate the spurious modesobtained by the stochastic subspace identification. The energy of each mode iscalculated by the selection matrices C, the eigenvalues and eigenvectors of the statematrix A and the state output covariance matrix G, then the contribution of differentorder modes to the structural response can be understood, which in turn can help toconfirm the dominate mode in the structure. An improved stochastic subspaceidentification algorithm based on eigendecomposition is introduced to solve its lowefficiency problem. Compared with the traditional algorithm, the proposed algorithmneed much less cost of memory and computing time as it doesn’t have a process of theQR decomposition of a high-dimensional matrix and SVD of the projection matrix,and the proposed algorithm improved the identification computational efficiencywithout losing the quality, especially when analyzing a large amount of data, thecomputational efficiency can be improved significantly.②ERA(eigensystem realization algorithm) which is a commonly used method inmulti-component measurement analysis is studied. For the drawback that itsidentification precision is susceptible affected by the measurement noise, SVD isadopted to reduce measurement noise. In order to reduce the influence by spuriousmodes on modal selection, the energy matrix of each mode can be calculated by theselection matrices C, the eigenvalues and eigenvectors of the state matrix A and theinput distribution matrix B. The largest singular value of the energy matrix obtained bySVD is a measure for the energy contribution of each mode, which is named modalenergy level. Spurious modes resulting from noise or model redundancy are indicatedaccording their mode energy level. Take frequency, damping ratio, mode shape andmold the modal energy level as the clustering factor, automatic identification can berealized with the help of Hierarchical clustering method.③CWT (continuous wavelet transform)-based modal parameters identificationwhich is good at identifying low-frequency close modal is studied. Attempting toovercome its low efficiency problem, a rapid modal identification algorithm usingwavelet based on data reduction is introduced. The SVD is firstly used to reduce the covarience signals on the premise of keeping data information amount, to decrease thedata needed in calculation. And then SVD is also applied to positive power spectraldensity matrix to identify modal order and its corresponding frequency range, then thewavelet transform is took to identify different order modes from one frequency toanother frequency after data reduction. Compared to the original algorithm, thismethod improves the identification computational efficiency without losing the quality,and the computational efficiency can be improved significantly especially whenanalyzing a large number of data.④A modal analysis system with self-propery right is developed successfully inthis thesis. It contains data acquisition module, data process module, designs ofstructure model module, parameters identification module, modal validation module,and modal shaping module. The whole process of structural modal parametersidentification can be easily, quickly and efficiently completed by using this system.Each functional module is detailedly designed, and the key technologies of itsimplementation are also detailedly discussed. The system contains a variety of modalparameter identification methods, they can check each other to ensure the reliability ofthe identification results. At last, the system is used at some project applications andcomparative test, the results demonstrate that the modal analysis system designed inthis thesis has the value of engineering application.Finally, the works of this thesis were summaried, and the future prospect ofambient excitation based experimental modal parameter technique is also discussed.