РЕЗИМЕ Предмет овог рада је моделирање понашања конструкција и околног тла на основу поступака идентификације система, односно експерименталних података. У овом раду су објашњене све процедуре идентификације система и њихова примена у деформационој анализи конструкција. Позанто је да се понашање конструкције и тла моделује на основу геомеханичких и статичких прорачуна. Геомаханички прорачуни се базирају на основу узетих узорака тла. Ови узорци не могу да репрезентују понашање целог тла на коме

more »

... е гради објекат. Статички прорачун се базира на правилничким одредбама. Међутим, приликом изградње објекта долази до грешака (квалитет уграђеног материјала и пропуста приликом градње) и необезбеђивања пројектоване геометрије објекта. Сврха моделирања на основу експерименталних података је: провера прорачуна, спречавање људских и материјалних катастрофа као и побољшање стандарда изградње. Познато је да се мониторингом конструкција и тла баве: грађевинска, геолошка и геодетска струка. Све ове струке морају да сарађују и користе исту терминологију да би се успешно извршио мониторинг конструкција. Кључне речи: Деформациона анализа, Идентификација система, модел, временска серија, функција преноса, потпростор стања, вештачка неуронска мрежа, детрендовање Научна област: Геодезија Ужа научна област: Инжењерска геодезија Abstract Milovanović Đ. Branko ABSTRACT The subject of this dissertation is modeling the behavior of structures and surrounding ground using system identification procedures, respectively based on experimental data. This paper explains all the procedures of identification systems and their application in deformation analysis of structures. It is known that structural and surrounding ground behavior is modeled on the basis of geomehanical and static calculations. Geomehanical calculations are based on samples taken on the ground. These samples may not represent the behavior of the whole ground on which build an structure. Statically analysis is based on rule book regulations. However, during construction of the building leads to errors (quality materials and omission during the building) and failure to ensure the projected geometry. The purpose of modeling based on experimental data is checking the project, preventing human and material disasters and improve the standard of construction. It is known that the monitoring of structures and ground deal in: civil engineering, geology, and geodetic professions. All these professions must work together and use the same terminology in order to successfully performed construction monitoring. By geodetic Abstract method we measures only the structural responses (output signals) and it is only profession which determines the movement and deformation in the absolute system. Practical research is meant: analysis of time series of geodetic monitoring in time and frequency domain and possibility of using these data to define the system order and order of models, analyzing the complexity of experimental modeling of deformation processes using the system identification procedure, the possibility of approximation of nonlinear dynamical processes by linear models, the proposed procedure determining the sampling of observations based on experimentally determining model (system identification). This research was derived on the simulated time series of settlement construction and the measured data of displacements of the pylon of the bridge, "Sloboda" in Novi Sad during test load of structure. The first analyzes the importance of ensuring stationary or weak stationary of the process. Stationary of the process provides by detrending data, trend removed from process. This is a very hardly job and it was shown that it is not necessary to ensure stationary of the process. The differences in the estimated parameters of model for the stationary and no stationary process are negligible. For data analysis in time domain used coefficient of autocorrelation, partial autocorrelation and cross correlation, in the frequency domain used Schuster periodogram. It has been shown that graphical representations of data can be used to defined order of system, which can be for the structure first (subsidence and temperature effects) or second order (the loads). The correlation coefficient provides valuable information about the order of the model, but for a definite model must test the significance of parameters. For linear modeling were applied transfer function models (regression models) and state space models -subspace method (Euler method and Tustin approximation), for a non-linear artificial neural network (backpropagation neural network) and the pseudolinear regression model for nonlinear processes. Also, it was used so-called commonly used transfer functions, which are used in management systems, for modeling the settlement of construction. Parameters of the model are estimated using least squares, nonlinear least squares method and the prediction error method. On the basis of the criteria approximation for simulation and prediction, it was concluded that the autoregressive model with external input signals, in the case of linear or nonlinear Abstract process, the most suitable for modeling the behavior of structures. Subspace method (Euler method) can be used to predict one step ahead, a supоsed form of transfer function (the most commonly used transfer function) is very suitable for modeling settling of the structure (the advantage is that it requires a small number of epochs of measurement for determining the model parameters). Backpropagation neural network (artificial neural network) is adequate for simulation of the structural behavior. Percentage of approximation would be better that there are many data in time epoch, which is rare in the geodetic deformation analysis. Analysis of sampling period showed that the epoch of measurement should be defined by the Sampling theorem, which determines the sampling period based on the transition state of the system. Transition state of the system is determined on the basis of projected structural behavior at the start of building, and after based experimental data.