Theoretical And Experimental Research On Dynamic System Identification In Time Domain

The research on dynamic system identification in time domain is a front subject in the field of current civil engineering. Combined with high-efficiency high-accuracy instruments for signal analysis, system identification technique based on the dynamic test is becoming an important means of damage detection of engineering structures. Because of the drawback of incomplete output response, uncertain environmental excitation and ill condition of inversion problem, it is very difficult to apply system identification directly to the dynamics detection of actual engineering structures, using the traditional theories and methods. According to the consideration mentioned above, this thesis focuses on the following aspects of research:1. Structure identification in time domain with incomplete input information is studied. Three compound inversion algorithms are constructed, whose additional correcting conditions in iterative process vary with the dynamic property of environmental excitation. Decomposition Algorithm and Updated Total Compensation Algorithm have the better calculating efficiency and stability than the original one. Periodic Statistical Average Algorithm, which is evolved from Statistical Average Algorithm, can identify the structural parameters and inverse the time history of periodic excitation of revolving machine. The influence of unknown inputs is discussed from the two levels of structure and element, and the parameters of part elements can't be identified rightly when there are more and intensive unknown inputs.2. Structure identification in time domain with incomplete output information is studied. The dynamic equations and parameter identification equations of substructure in general use are established, which make full use of the gathering laws in finite element method. With the reasonable definition of boundary node (element), internal node(element) and external node(element) of sub-structure, clear scientific rules are provided, which help to judge whether the response information is complete. When the sub-structures are divided, the distribution of unknown load and the overlap of boundaries must be considered. Sub-structure method not only overcomes the obstacle of information incompleteness, but also improves the calculating efficiency. Combining sub-structure method with compound inversion algorithm, the problem of structural parameter identification and load inversion under the condition of less measurement information is solved.3. Ill condition and regularization method of identification equation in timedomain is studied. Through condition number and numerical rank of the response matrix, the ill condition of identification equation is studied. It is illustrated that the type of structure and location of load are the main influence facts of ill condition property. In order to reduce the influence of ill condition, Tikhonov regularization method is evaluated in the thesis. It is concluded that regularization method with parameter constrains can obtain identification results of fairly good accuracy. Damp least squares solution is not a real Tikhonov regularization solution, which is still sensitive to measurement noise. But taking the regularized function in the form of Tikhonov as the objective function, damp least squares algorithm can solve the nonlinear ill-posed problem effectively.4. Signal denoising theories and methods are studied. A new signal denoising method based on virtual response information is proposed, which considers fully the response property of linear dynamic system in time domain and the property of frequency space partitioned by wavelet package analysis. Being a kind of realizable responses of structure under a certain excitation, virtual responses can be used to identify the structural dynamics system deservedly. Seismic response information is a batch of low signal-noise-ratio data, whose valid frequency zone is full of measurement noise. It is illustrated by numerical examples that the new method proposed in this thesis can eliminate efficiently the noise in Seismic responses, while threshold value denoising method based on wavelet analysis is of no effect.5. Hybrid genetic algorithm and its applications are studied. Hybrid genetic algorithm or multi-population hybrid genetic algorithm is constructed, which take BFGS algorithm or damp least squares algorithm as the local search operator according to the different characteristic of objective functions. Numerical results show that the hybrid algorithms can quickly converge to the global optimum with high probability in identifying nonlinear parameter system, estimating wavelet threshold values or searching optimal regularization solution.6. The model of reinforced concrete frame structure is identified based on the dynamic test. According to Seismic acceleration responses and the 1st order modal frequency, the stiffness parameters based on story model are identified, and then the elasticity modul values are obtained through the inverse operation of D-value method. The identification results of elasticity modul agree with the experimental results of material property, which testifies that the theory and method of system identification in time domain can apply to dynamics detection of engineering structures. The Seismic responses include much frequency information relating to seismic excitation, which isa kind of useful signal for the identification methods in time domain, while a kind of interferential signal for the identification methods in frequency domain.