Research On Frame Structure Damage Identification Based On The Curvature Mode Analysis Of Wavelet-Neural Network Analysis

Frame structure is widely adopted in engineering. However, there are often initial damages or cumulative damages due to the increasing of durable years and frequent external loads during operation. All these factors may lead to a reduction of the structural resistance,, even the destruction of the whole structure. Therefore, it has an important engineering and practical meaning to study damage diagnoses’ methods and monitor the working situation of frame structures.The wavelet transform has a great capacity to analyze the partial characteristics of signals in both time and frequency domain. And the neural network is characterized by strong self-learning and nonlinear mapping ability. So the thesis focuses on studying the damage identification of frame structures through combining the wavelet transform and the neural network. The main theory is as follows:firstly, build a finite element model of damaged frame structures and analyze its dynamic characteristics to gain a curvature modal; then the structures’wavelet coefficient can be gained by a continuous wavelet transform of the curvature modal; finally, the damaged structures can be located by the maxima modulus of wavelet coefficients. The neural network gets constructed with the natural frequencies of damaged structures as its input parameters. Thus the damage degree of frame structures can be identified.The thesis regards frame structure as the object of research and establishes the finite element model of frame structures (frame structures with a layer of a cross or a layer and two cross). Based on the principle of wavelet analysis and the Lanczos algorithm, the curvature modal of frame structures can be obtained. And the wavelet coefficients are derived from a continuous wavelet transform of the curvature modal. Then the damaged structures can be located by the maxima modulus of wavelet coefficients. The neural network gets constructed with the natural frequencies of damaged structures as its input parameters. Thus the degree of frame structures’damage can be identified.The above method is applied to more complex frame structures with two layers of a cross or two layers of two cross on the basis of simple frame structure in this paper. Finite element models of frame structures are established. The damaged structures can be traced by the maxima modulus of wavelet coefficients and the degree of structural damage can be identified by the neural network. The method is verified effective. The Method in the paper also can be applied to structural damage diagnoses for the reference.