Statistical Method Of System Identification On Civil Engineering Structures

The system identification in civil engineering, focused on the health monitoring and the damage detection, is necessary indeed because the cost of the structures is always takes the key part in the national budget and the operating condition is always related to a number of lives. The data, which the identification based on, would be large for a long time monitoring, be uncertain for the intrinsic variety of the material and for the extrinsic random of the environment noises, be limited to learn for the computing ability restriction and for the requirement of the feedback in time. That is to say, there exist some features to be improved on in the traditional system identification methods: data inspection on the complex structures based on the simple group index, uncertainty system description based on the certainty knowledge, estimation on parameter quantitative changing based on the qualitative analyses, prediction on the limited swatch based on the infinity swatch learning theory, investigation the details based on the general coordinate. Then the identification based on the statistical method is significant for the theory studies and for the engineering applications.In this thesis the statistical identification and the indirect identification were proposed, and some novelty methods were investigated on the identification from the angle of the data features based on the research of the structure vibration, statistics, and system identification. The identification proposed were inspected into three groups: the data acquirement for the identification, the system identification on the structure response, and the statistical methods on the identification:1) The statistical method of system identification for civil engineering structures was defined: the system identification was performed by the features amplified by the data description and estimation, and a quantitative investigation was done on the acquirement, classification, analysis and interpretation of the structure response data.2) The dynamic response of the structures was acquired: the data were obtained based on the theoretical analysis, numerical calculation and experimental research. The dynamic response analysis was performed according to the data features of the frame structure and the operational bridge.3) The system identification was inspected on the structures: the features of the different parameters in the frame damage identification were explored. The bridge parameters were identified based on the model set by the Pattern Search algorithm, which could be applied to the multi-local-extremism and to discontinuous data. The local extremums and the convergence could be avoided in most of the conditions in the detection. The identification results were discussed with the ones which identified by Genetic Algorithm and Simulated Annealing. It is concluded that the Pattern Search shows more efficiency and precision in the identification.4) The calculation methods were inspected focussed on the statistical features of the excitation and the response: the random phase angle was calculated based on white-noise inputs, which were simulated in the normal distribution, the uniform distribution and on the congruence method respectively. The time-domain frequencies are depended on the subsection length, and the amplitude frequencies are depended on the distribution types. The road random excitation was simulated based on the trigonometric series model and on the auto-regressive model. The road surface profile amplitudes are depended on the roughness coefficients, and the curve shape on the random phase angle distribution. The road surface profile PSD curve amplitudes are approximately equal and the values are depended on the roughness coefficients. No dominant frequency emerges. It is concluded that the road excitation should be simulated on the trigonometric series model with the uniform distribution.5) The data inspection on the complex structure was grouped by the statistical classification: The roughness dynamic amplification factor is defined, and is calculated according to the classification on the response type, vehicle velocity, and road surface profile. The geometry significance was given to the general amplification factor in the 3-dimension space.6) The uncertainty system was investigated with the uncertainty knowledge with a confidence level, and the indirect identification was proposed: the indirect identification was proposed, in which the bridge parameters were identified based on the vehicle responses. The first frequency and the stiffness of the bridge were achieved by the point estimation and the interval estimation based on the semi-analytical, numerical, noised semi-analytical and numerical solution. The results were verified by the Genetic Algorithm. The residual and the error were calculated to assess the effect of the different model sets. It is concluded that the error of the point estimation and the distribution of the interval estimation are affected significantly by the model set type.7) The estimation on quantitive parameter changing was assessed on a control criterion with a confidence level quantitatively: the damage conditions were identified on the existence, location and identification with an upper control limit according to the Statistical Process Control theory. The results show a strong robustness even in the high noise level.8) The system response was identified on the statistical learning theory with limited swatch: the responses of the operational bridge with multi-uncertainty were decomposed to the data trend, the dada details and the noise details. The responses were identified with the data details and the data trend while the noises details excluding. The parameters of the kernel and the penalty were discussed in the algorithm.9) The characters of the structure were extracted with the coordinate transformations based on the Principal Component Analysis according to the data features: the damage qualitative identifications of the frame were explored with the hypothesis test and the Principal Component Analysis on the acceleration responses, in which the calculation cost was low and the results were illustrated in a time-save way. The damage existence, location and degree were identified on the modal curvatures with the coordinate transformations according to the Principal Component Analysis method, and the effect of the noises level were discussed, in which the identifications were more time-cost and were designed to a further detection for the important structures.