【作者】 于繁华；

【导师】 刘寒冰；

【作者基本信息】 吉林大学 ， 道路与铁道工程， 2008， 博士

【摘要】 本文概述了桥梁结构损伤识别理论和计算智能技术等方面的内容,成功地将计算智能技术引入到桥梁结构的损伤识别中,并取了良好的效果。根据残余力向量的概念,把结构损伤识别问题转化成优化问题,并建立了适合结构损伤识别的多目标优化模型,为提高模型求解的准确性,利用灰色关联度能够很好地分析各非劣解与理想解之间的接近程度,并能掌握解空间全貌的特点,依据目标矢量序列和基准矢量序列之间的灰色关联度大小选取粒子群算法中的全局极值和个体极值,提出了适合高维多目标求解的灰色粒子群算法;利用支持向量机建立了一组待反演参数与各阶频率和一阶振型之间的函数关系,将所得到函数表达式与结构各阶实测频率和振型间的差异作为优化目标,建立了一个多目标优化模型,并用所提出的灰色粒子群算法进行求解;为提高神经网络的收敛速度,提出了基于局部学习的小波神经网络算法,并将该算法应用到桥梁结构损伤识别中,取得了较好的效果;为增强网络的泛化能力,提出了基于灰色聚类技术的神经网络集成方法,该方法改善了利用传统神经网络进行结构损伤识别时泛化能力低的情况。算例结果表明,上述方法提高了桥梁结构损伤识别的效果,得到了具有学术理论指导价值和实际应用参考价值的结果。更多还原

【Abstract】 With the development of society and civil engineering technology, many huge and novelty bridges were established recently. Meanwhile, some emergent brake accidents of some bridges occurred. It cause the publics pay more attention on these tragedies. And every government and science institutions were aware of the exigency of the research on detecting damage of bridge structure.The reasons that caused the function degenerating and structure braking of bridges were various. Some braking reasons were caused by artificiality, and some of them were caused by natural disasters. Some defects in inner structure usually existed in the latter. As this kind of defects were invisible by unaided eye, it is necessary to know how to analysis and judge the defects by some detector methods and how to know where they were. This problem became a top issue internal and oversea. In addition, bridges could not detect the satisfying degree of design objectives by destructive archetypal test just like for other mechanic manufacture, such as airplane, ship and vehicle. The reason is that bridges and buildings are belongs to single unit which could not be test by destructive archetypal test. So the nondestructive measuring technique on bridges and buildings is extraordinary concerned.How to choose the appropriate method to find the damage structure of bridge is a core problem for evaluation of structure safety. There are two methods according to the different solving approach when the dynamic characteristic is used to detect the damage structure. One is traditional mathematic method, another is computational intelligence method. Traditional mathematic method has some problems such as computing complicatedly and slowly, divergence, and obtaining local optimal solution. In order to solve this problem, the methods of the neural network, evolutionary computation and fuzzy system in computational intelligent technology are adapted to the field of detect damage of structure by many scholars home and aboard. And these methods have a good effect.This paper outlined technologies about detecting damage of structure and theories about computational intelligent technology. The first, the support vector machine and the concept of residual force is used to set a multi-objective optimization model of bridge for detecting damage of structure because detecting damage of structure is a back analysis problem. And the grey particle swarm algorithm is employed to solve it. This provided an effective approach for detecting damage of structure. The second, the wavelet neural network based on local studying and neural network integration based on grey clustering are presented respectively for detecting damage of structure. The results show that the proposed methods are effective.There are six chapters in this paper. The main content is shown as following:Chap1.The background, aim and meaning of this research are discussed in the beginning. Then, the researching and developing situation of the technology on detecting damage of bridge structure and computational intelligence, and detecting damage of structure based on computational intelligence are introduced.Chap2. In this chapter, the recognition of fingerprint analysis correlated with dynamic characterize of structure, damage detection method based on model updating theory and dynamic finite element model of damage structure are introduced. Then, the theory and basic methods of damage detecting of bridge structure and computational intelligence are introduced either. They are model of neural network, wavelet transform technology, the theory of integrated neural network, evolutionary computation and particle swarm algorithm.Chap3. The damage detecting of structure is transferred to optimization problem, and the multi-objective optimization model which is suited for damage detecting of structure is built. For this model belongs to higher-dimension optimization problem, the precision of traditional method is lower. For grey relevant degree can analysis approach degree between Pareto solutions and ideal solutions, and can master characteristic of solution space, a grey particle swarm algorithm for multi-objective is proposed to improve the precision of solutions. In this method global best and personal best in particle swarm algorithm are selected according to grey relevant degree between objective vector sequence and benchmark vector sequence. The proposed method can detect structure damage better through damage identification experiments of cantilever beam and truss structure. It provides an effective method for detect damage of bridge structure. In order to improve reliability and robust of reinforced concrete beam a multi-objective model is built for reliability robust optimization design by robust optimization design of reliability theory. The better effect is obtained for higher multi-objective optimization problem on robust optimization design of reliability by grey particle swarm algorithm.Chap4. This chapter describes the basic theory of Support Vector Machine (SVM). Then, the non-liner function is fit by SVM. After comparing with additional noise and BP neural network, the conclusion that SVM has strong robustness on liner fitting under noise is obtained. Meanwhile, a functional relationship from reversion parameters to frequencies and first order mode shape is established by the support vector machine. It replaces finite element calculation when the objective function is used to detect structure damage so that the computational velocity has been great improved. In order to improve the effect of detecting damage of bridge structure, a multi-objective optimization model is established by taking the functional expression obtained and difference between structure testing frequencies and modes as optimal objective. And it is solved by the proposed grey particle swarm algorithm. Through the detecting damage for one unit and multi-unite of simply supported beam, a satisfying result is obtained at damage identification and anti-noise capability when support vector machine and grey particle swarm algorithm is employed for less unites damage. So, support vector machine and grey particle swarm algorithm achieve the detecting damage of bridge structure. This method provides an effective way in this field.Chap5. In this chapter, the neural network technology is used for detecting damage of bridge structure. There are two parts in this chapter. In the first part, the wavelet neural network is employed to carry out the damage identification of concrete continuous bridges with three-spans. The wavelet neural network algorithm which is based on local studying is presented later. This method gets a better effect for the multi-units damage identification of concrete continuous bridges with three-spans. After analyzing this algorithm, the results that convergence of LCG algorithm is fast and anti-noise capability is better is obtained. In the second part, neural network integration is employed to detect the damage of truss structure and cable-stayed bridge. And a neural network integration based on the grey clustering technique is presents. The lower generalization of traditional neural network is improved by this method when it is employed for detecting damage structure. So, we can predict that neural network integration technology will have a vastitude foreground.Chap6. Conclusion and prospects. The research and conclusion of this dissertation is summarized and the application and the development foreground of detecting damage of structure based on computational Intelligence.更多还原