Damage Identification Of Inland River Frame-pier Wharf Based On Neural Network

Based on the case of an inland river pier wharf, it has been established frame model using ANSYS finite element analysis software to extract the inherent frequency and vibration model of the structure. Application of neural network to build a port component damage identification system, it was used to identify terminal damage location and damage degree of the component. The main research contents and conclusions of this paper are as follows:(1) According to the theory of damage mechanics, it has been established injury model of wharf structure and calculated the damage index. Through analyzing the change rule of the damage index in the condition that the damage degree is the same while the damage position is different, and the change rule of the damage index in the condition that the damage position is the same while the damage degree is different, it has gotten all the damage indexes which the damage identification needed. It can be used the damage index of NFRN to identify the primary location of the wharf structures' damage, the damage index of ND? to identify the specific location of the piles' damage, the damage index of X2 to identify the specific location of the pillars'and beams' damage, the damage index of RNF to identify the degree of the piles'damage, the damage index of FFC to identify the degree of the pillars' and beams'damage.(2) Applying the theory of neural network, it has been established the damage identification system of the frame-pier wharf. The recognition process can be divided into three steps. Firstly, the type of the damage member could be identified according to identify the primary location of the wharf structures' damage by PNN neural network. Secondly, in order to identify the specific position of the damage members, using BP neural network for further identify. Finally, it was identified the degree of the damage members using BP neural network. Through testing of the network, the result of the identification is effective.(3) According to analyze the neural network's anti-noise capability, we can draw two conclusions. When identifying the primary location of the damage members or identifying the specific position of the piles' damage members, due to the influence of the error is smaller, there's no need to train the neural network. When identifying the degree of the damage members or identifying the specific position of the pillars' and beams' damage members, as the bigger impact of the error, the training with noise must be carried out on the neural network to improve the anti-noise capability.(5) By comparing the anti-noise capability of each neural network, the effect of error on the frequency index is less than which on the modal shape index. So when measuring the mode shape, it should be taken necessary measures to de-noise.(6) The neural network identification effect was improved after training with noise. The improvement of the neural network effect, which is used to identify the damage member of pile and pillar, is better than that is used to identify the damage member of beam.