Automatic Identification Of Modal Parameters Under Ambient Excitation

Modal parametric identification of engineering structures is one of the key problems in the field of structural health monitoring due to its application in structural model updating,damage identification or optimization design and so on.Meanwhile,owing to the requirements of online monitoring and real-time warning of structural safety,a structural health monitoring system is in need of automatically identifying the true modes of the structure.On this background,this research is focused on the topic of automatic identification of modal parameters by combining with the newly developed deep learning technology and modal parameter uncertainty theory to automate and speed up the whole process of modal parameter identification.In chapter 1,the existing researches of automatic modal parameter identification are reviewed and summarized to find out the drawbacks of this research field.Then the research framework and research contents of this study are clarified.In chapter 2,the theories of some common modal parameter identification methods are introduced.The Matlab codes of the data-driven stochastic subspace identification method、the covariance-driven stochastic subspace method、the PolyMAX method and the enhanced frequency domain decomposition method are programmed.Then the case of ASCE four layer frameworks is used to point out the human interference and uncertainty factors existing in these methods,thus providing the basis for the automatic identification and uncertainty research.Chapter 3 presents an iterative algorithm for automatic analysis of stabilization diagrams based on genetic algorithms and fuzzy clustering methods.This methodology can automatically remove the false modes from the stabilization diagram by setting an arbitrary initial clustering center.In this way,the real modes of the structure could be automatically recognized without any humman interference.At last,this method is validated by a mass-spring numerical example and a frame model.Chapter 4 presents a method for automatically analyzing of the stabilization diagram based on convolutional neural networks.The concept of single modal stability diagram is proposed and used as a training sample of convolutional neural network.By the skill of translation and changing the label of the stable poles,the samples of convolutional neural networks are expanded.Then these pre-processed training samples are substituted into the initial convolutional neural network for training.Nextly,the convolutional neural network could be obtained which can automatically determining the false mode in the stabilization diagram.At last,the method is validated by a mass-spring numerical example and the ASCE frame model.Chapter 5 deduces the formula of fast uncertainty calculation of modal parameters based on covariance driven stochastic subspace identification method.On the basis of the uncertainty calculation,the concept of the uncertainty diagram is proposed to distinguish the false mode and improve the computational efficiency of modal identification.In addition,the convolutional neural network is also used for automatically identifying the uncertainty diagrams so as to realize the automation of modal parametric identification.Then,the method is also validated by a mass-spring numerical example and the ASCE framework model.Chapter 6 applys the three kinds of automatic modal parameter identification methods to real engineering structures such as Swiss Z24 bridge,Guangzhou TV Tower,Canadian HCT concrete frame building,Hong Kong Ting nine bridge and Australian S101 bridge.The automatic modal parameter identification of these cases further verified the reliability of the proposed algorithm and engineering applicability.In addition,the advantages and disadvantages of the three methods are compared,and the problem of dividing close modes,exploring the influence of the order of acceleration data on the identification results and sketching the three-dimensional mode shapes of these real engineering examples under the condition of lacking the accelerometers are discussed.Chapter 7 summarizes the research work and the research results of this paper.The shortcomings and the direction of the subject which shoud be further studied are also pointed out.