Study On Damage Recognition Of Intelligent Materials And Structures Based On FBG Sensor And Wavelet Packet Analysis

Fiber Bragg grating sensors, which provide the accurate condition of collection signal for Structure Health Monitoring, have been obtained extensive application in most engineering field because of its unique characteristics. Wavelet analysis, which has been greatly developed in recent years, is a new signal-processing tool and the foundation of realizing structural damage recognition because its time-frequency localization characteristic may simultaneously analyze and process signal in time and frequency domain. In this thesis, the exploratory research works on the fabrication of fiber grating (FBG) sensor, the demodulation of FBG signal, the analysis of structural dynamical response characteristics have been done. The extracting method of structural damage eigenvalue, the foundation model of coupled neural network, and the realization of structural damage recognition have been studied.MEMS technology has been used to improve the sensitivity of fiber grating. Changes of photoproduced index and its effects on the property of fiber grating have been investigated under loaded and unloaded hydrogen. Two kinds of new fiber grating strain sensors, embedded and pasted, have been developed and the technology for embedding and pasting FBG sensors has been studied. The static and dynamic characteristics of pasted fiber grating strain sensor have been emphatically studiedBased on the analysis and comparison of the popular FBG demodulated technologies, a new high-speed FBG demodulated technology based on brink-filter has been proposed and used as the demodulated means. The demodulated theory of the new demodulated technology has been presented and the key technology of the demodulated method has been investigated. The possibility of the demodulated technology has been analyzed and validated by the experiments.The relationship between the stress and strain of intelligent composite materials upholded plank under the static force has been deduced. The analytical solution of the four-side upholded plank's first order inherent frequency and the other order inherent frequencies were obtained by analyzing the deformation of intelligent composite materials plank under the dynamical load. The finite-element model of four-side upholded plank was suggested by employing the finite-element analysis software. The inherent frequencies and the freely vibrating models of four-side upholded plank before and after damage have been simulated.The basic principles and methods of multiscale analysis principle and wavelet package analysis technology have been presented. The selected principle of wavelet base has been analyzed and discussed. N of wavelet base function DbN and j ofwavelet package decomposed scale have been ascertained. The structural damage recognition has been analyzed with mode identification method and the method for extracting the wavelet package signal energy eigenvalue has been analyzed. The wavelet package signal component energy eigenvector was used as the damage index property of structural damage recognition according to the multiresolution trait of wavelet package.The learning train process of BP arithmetic has been deduced in detail. Then the coupled neural network model has been constructed according to the data collecting and processing of the multisensor. It has been proposed by employing the coupled neural network method to realize the damage recognition of multisensors. The character parameters to sensor structural signals in the different regions have been drawn with the wavelet package transform technology. Based on the BP arithmetic, three coupled neural network models have been constructed to realize the affirmance, orientation and quantification of structural damage, respectively. Taking the structure model of the intelligent materials four-side upholded plank as an example, the measured data of four-side upholded plank structure under 32 kinds of damage conditions were used for wavelet package transform. The first-degree coupled neural network has been trained by using the damage index property as the input samples.