Study On The Output Based Modal Parameter Identification

Compared with the traditional methods, the output-only identification can reflect the system's natural properties, boundary conditions and environment loads more actually, represents the newest development in modal parameter identification. Since it is still in its seedtime, the output-only identification technique lie in some preliminary stages. As a result of the disturbance and un-sufficient ambient excitation, the properties of responses are complex and weak, which can fail the identification. For some modal testing on the large engineering structures, we have to expend more testing time and achieve large number of measurement data. The response signals may carry some properties, such as harmonic vibration, short time shock or nonstationarity, which is difficult to be expressed only in time or frequency domain. During the identification, the calculation errors and disturbance of environment excitation will result in the spurious modes. So the research should be put into several aspects: preprocessing response signals, knowing about the characteristics of systems, getting rid of the spurious modes, identification of weak modes.This dissertation is investigated both theoretically and experimentally, under the support of Natural Science Foundation of China(NSFC) for the project"Study on robust model order determination and blind modal parameter identification, under grant number 10302019". The main research work and conclusions are as follows:1. A large amount of literature related to output based modal parameter identification are reviewed. More focus are on the theory and algorithm, application background, significance and current research status, solved and open problems. Based on this summarization, we realized that it is worth devoting more efforts into the method of preprocessing abundant testing data, identification of weak modes and separation of physical and spurious modes.2. The theoretical background and algorithms of output-only subspace identification (SI) are discussed in this thesis. Subspace identification techniques are developed from the observability and controllability of linear time invariable (LTI) systems. Constructed by the state space models, SI is numerically robust and able to give high precision estimates. It is also the most advanced parameter identification techniques in time domain. Because of these virtues, SI is chosen to be the theoretical basis of this thesis. It is proved that, under stationary random excitation, the modal parameters can be estimated from an arbitrary row or column of the Hankel matrix constructed by auto-correlation functions of responses.3. The Gabor-expansion-based time frequency filtering is employed to process measured signals before conducting parameter identification. With Gabor expansion, the local characteristics of a signal can be shown in the time-frequency domain, according to which the properties of responses are acquired and the response signals can be filtered by elimination or interception. Two simulation cases validate that this technique can be applied to both stationary and nonstationary signals. Attribute to this technique, more information about the signals and systems can be gained, such as the distribution of natural frequencies, energy assemble functions, quantity of properties contents in each channel. The estimated results of a 7-DOF vibration system, which has great disparity in mode energy, have demonstrated that cross-correlation may weaken the signals of small energy, while auto-correlation may retain or enhance them.4. Compared with the data-driven stochastic subspace identification (SSI) method, the correlation-driven one can estimate parameters with equivalent precisions and less calculation time. This thesis discusses some alterations to traditional correlation-driven SSI for operational modal analysis. The numerical decomposition has been combined with the energy measurements of subspace method. The first change is the use of only auto-correlations in the Hankel matrix; Cross-correlations between responses are omitted in a first stage. The identifiability of weak characteristics and the robustness to noise contamination are enhanced. The second change is to use the so-called component energy index (CEI) as a pole stability criterion. In the stablization diagram, spurious modes are of low CEI and unstable, while the physical ones show opposite performance. The third one is that the identification procedure consists of two steps. The natural frequencies and damping ratios are estimated firstly. On the basis of identified frequencies, mode shapes are extracted from the signals obtained by filtering measured data with a series of band-pass filters.5. A vibration system of multiple-degree-of-freedom is used for the comparison of several commonly used output based techniques such as Frequency Domain Decomposition (FDD), Gabor Expansion, Stochastic Subspace Identification (Traditional and Improved) and Higher Order Statistics (HOS). According to the simulation results, which is attributed to the great disparity in mode energy, weak modes of the vibration system are in fact difficult to identify with common techniques. FDD is interfered by noise in natural frequency identification and gets inaccurate damping ratios. Although the time- frequency technique can estimate weak modes, the estimates are lack of consistency which results from the different adopted time domain identification methods, moreover the identified modes can not be closed too much. Since the higher order statistics will weaken the modes with small energy furthermore, the HOS method not only gets the values with large errors but also has heavy computation burden. In combination with CEI, the proposed improved SSI gives a better estimation of weak modes and a reliable separation of spurious and physical estimates.6. The proposed output based identification method is tested with a metallic frame structure subjected to wind excitation a backstay under wind and impact excitation, and a concrete-filled steel tubular arch bridge which was subjected to ambient excitation. Identified results show that the proposed method can give a reliable separation of spurious and physical modes as well as accurate estimates of weak modes only from response signals.