Research On Damage Detection And Optimal Sensor Placement For Piezoelectric Smart Structures

The research on realizing the self-detecting damage function is one of the main research contents of smart materials and structures. There are two important problems that are related to the self-detecting damage function. One is the method of damage detection; the other is the problem of optimal sensor placement. It has been of an important theoretical meaning and a great practical value for applications of smart materials and structures to research on these two problems. Based on the finite element numerical simulation, the piezoelectric smart composite laminated plates is simulated, and the transient responsive signals of piezoelectric sensors and their characteristics are obtained under the low-velocity impact load. On the basis of the above, the methods of damage detection and optimal sensor placement for smart materials and structures are researched in this dissertation.Support Vector Machine (SVM) based on VC Theory and Structural Risk Minimization Principle is progressing rapidly in recent years, and has become a very young and useful component of Statistical Learning Theory. Nowadays SVM become an ideal network model for pattern recognition and nonlinear regression. In this dissertation, the regression theory of Least Square Support Vector Machine (LS-SVM) is applied to detect the impact damage locations for the piezoelectric smart composite laminated plates, and compared with the improved BP neural network. The results state clearly that, LS-SVM possesses the advantages such as the faster speed, better dissemination ability etc. And LS-SVM is not sensitive to the order transform of network imput vectors, especially meets the requirements of establishing the objective function based on damage detection for the problem of optimal sensor placement.The method of genetic algorithm integrated neural network is proposed to optimize sensor placement based on damage detection for smart structures. In this method, LS-SVM is adopted as a kind of neural networks to establish the performance function based on damage detection, and genetic algorithm is applied to optimize the performance function. Considered the cost-effective factor roundly, the optimal number of sensors can be determined through the method. The implementation process and feasibility of the method of genetic algorithm integrated neural network are analyzed in this dissertation. The results show that the method is feasible, and can be applied to realize the optimal sensor placement corresponding to its primal sensor placement. For the more sensors primal placement, the number of sensors can be reduced effectively through the method. Moreover, this method is analyzed through the test based on the active monitoring scheme. In the test, the characteristics of sensors'responsive signals are extracted based on the method of Power Spectrum Density Maximum.In this dissertation, based on the finite element method, the wing box specimen of a plane is simulated, and its piezoelectric responsive signals are obtained under the impact load firstly. Then the method of genetic algorithm integrated neural network is applied to determine the optimum piezoelectric sensor placement for the wing box specimen of a plane based on damage detection. The result of optimizing sensor placement corresponding to its primal sensor placement is obtained, and can give a certain of guidance for the practical piezoelectric sensor placement for the wing box specimen of a plane.The research is partially supported by the grant from National Natural Science Foundation of China(90205031).