Research On The Method Of Multi-stage Structure Damage Detection Based On Neural Network For Frame Structure

Civil engineering structures such as tall buildings, roads and bridges, tunnels and rock, and marine platforms will inevitably produce different degrees of damage due to the influence of natural factors such as earthquakes, fires, typhoons, and other natural disasters or fatigue, corrosion, and other structural reasons caused by long-term load effect. Damage accumulated over a long period of time but did not receive timely attention and proper treatment, will consequently lead to reduction in structural performance and even damage which can cause huge losses of life and property. If the on-line monitoring system can be built timely which can be applied to detect and assess damage within the structure to predict changes in structural performance and residual life, then to maintain a corresponding repair treatment and to evacuate residents reasonably according to the results, will not only improve the operational efficiency of engineering structures, but also ensures security of lives and property. After three decades of reform and opening up economic construction and urban development, China is gradually transforming from the period of a massive construction projects into a new period of research structural damage. Structural damage detection technology as a frontier area will certainly become a hot research topic of civil engineering.Structural damage may cause changes in dynamic characteristics, if the nonlinear mapping relation between structural damage and dynamic characteristics is established, the structural damage can be identified. Artificial neural network as a research focusc in recent years with its ability to approximate any nonlinear mapping, and with the features such as distributed parallel processing, adaptive learning and robust fault-tolerantance is widely applied in many other fields. In this thesis, the advantage of the hot research topics significance is combined, the research exploration is from frame level component, making use of a variety of neural network and the parameters to identify and analyze the damage. The research work is divided into the following aspects:①Systematic introduction to related concepts of damage detection and current development status between home and abroad is made. The exploration of the current existing problems of identification of the damage and development trends, and combined with intelligence theory to discuss and demonstrate the method and relevant concepts of multi-stage structure damage detection based on neural network. ②The method of multi-stage structure damage detection based on neural network for frame structure is introduced, which is capable to figure out occurrence of structural damage, distribution of damage type, determination of the damage location, the detection of the degrees of damage of multi-stage neural network detection technology, is used to avoid the problems such as long training time, combined blast and so on which caused by the structural damage index disposable entering into the network.③To make use of the before and after structural damage modal parameters (modal frequency, mode shape), combined with theoretical formula to derive a series of damage indicators, and to research the relationship between their location and degree of damage. And to provide the basis of the selection of input parameters in different stage of the subsequent damage detection research based on neural network.④To make use of finite element analysis software ANSYS, to establish a five-story frame structure theoretical model of component damage, and to complete finite element analysis and to extract modal parameters by making use of parametric design language APDL to write user procedures, combined with MATLAB software platform programming for frame structural damage identification to apply in the joint network multi-stage detection analysis;⑤A numeral simulation analysis is presented, used for structural damage at all stage structures of different modal parameters indicators, to construct the identity of signature for damage-sensitive, and as character parameters to the corresponding neural network (BP, PNN and RBF) in the structured alarming detection, initial position, the specific location and extent of assessment, to achieve efficient identification of the joint network.