考虑不确定性的结构损伤区间反演分析

发布时间：2018-03-28 16:33

本文选题：损伤识别　切入点：不确定性　出处：《南昌大学》2017年硕士论文

【摘要】：结构在服役过程中,受到自然环境和人为等因素的影响,会出现不同程度的损伤。随着损伤的积累,结构的安全性、适用性和耐久性功能将不断降低,容易引发安全性事故。所以研究有效的结构损伤识别方法,及时监测现役结构的工作健康状态,对于损伤诊断和保障结构的安全运行有着重要的意义和价值。基于确定性的结构损伤识别方法为结构健康状态诊断和安全评估奠定了坚实基础。由于施工多样性、量测环境的局限性、荷载的长期性以及其它不可预见的情况,会导致实际工程中存在诸多的不确定性。基于确定性的损伤识别方法不直接考虑不确定性因素的影响,从理论上讲不具备求解不确定问题的能力。不确定性因素的存在会使得确定性结构损伤方法在应用于实际结构时效果不佳。因此,需要发展不确定性的损伤识别方法,有效地对结构损伤进行诊断。本文围绕观测不确定性和模型不确定性,提出两种适用于不确定性的结构损伤识别方法。具体研究内容如下:(1)将观测不确定性处理为含高斯型噪声的观测信息,模型不确定性处理为用区间数描述的结构参数,提出结合扩展卡尔曼滤波和遗传算法的区间反演分析方法。扩展卡尔曼滤波中状态向量的最优估计值作为目标函数,在已知结构参数不确定性范围内,通过遗传算法进行寻优以获得状态向量的上下界值,从而实现结构参数的区间反演。(2)将观测不确定性处理为含未知但有界(UBB)噪声的观测信息,提出基于集员滤波的结构损伤识别方法。首先,结构的状态向量用椭球集合描述。然后,在结构参数反演过程中,对结构非线性系统进行线性化处理,将线性化误差、系统误差近似为系统噪声,采用区间分析方法计算噪声边界,并用椭球集合外包。接着,通过观测信息不断地更新状态向量椭球,使得逐渐逼近真实解。最后,取收敛后的状态向量椭球的边界值作为结构参数的识别区间值,中心值作为最优估计值,用于指示损伤,从而实现不确定性结构的损伤识别。(3)对于反问题求解的不适定性问题,采用正则化技术处理。在集员滤波最优估计中引入的正则化可以转化为参数不等式约束,从而更新状态向量椭球。同时结合先验信息约束,将其作为超平面体,进一步更新状态向量椭球。反演算法的两种改进之处均能提高结构损伤识别精度。
[Abstract]:In the course of service, the structure will be affected by natural environment and man-made factors, which will lead to different degrees of damage. With the accumulation of damage, the safety, applicability and durability of the structure will be reduced. It is easy to cause safety accidents. Therefore, effective structural damage identification methods are studied to monitor the working health of active structures in a timely manner. It has important significance and value for damage diagnosis and safe operation of structure. The method of structural damage identification based on certainty lays a solid foundation for structural health condition diagnosis and safety assessment. The limitation of the measuring environment, the long-term load and other unforeseen conditions will lead to many uncertainties in the actual engineering. The damage identification method based on certainty does not directly consider the influence of uncertain factors. Theoretically speaking, there is no ability to solve uncertain problems. The existence of uncertain factors will make the effectiveness of deterministic structural damage methods poor when applied to practical structures. Therefore, it is necessary to develop uncertain damage identification methods. The structural damage is effectively diagnosed. This paper focuses on the observation uncertainty and the model uncertainty. Two structural damage identification methods suitable for uncertainty are proposed. The specific research contents are as follows: (1) the observation uncertainty is treated as the observation information with Gao Si type noise, and the model uncertainty is treated as the structural parameter described by interval number. An interval inversion analysis method based on extended Kalman filter and genetic algorithm is proposed. The optimal estimation value of state vector in extended Kalman filter is taken as the objective function, which is within the uncertainty range of known structural parameters. In order to obtain the upper and lower bounds of the state vector by genetic algorithm (GA), the interval inversion of structural parameters is realized. The observation uncertainty is treated as the observation information with unknown but bounded UBB-noise. A structural damage identification method based on set member filter is proposed. Firstly, the state vector of the structure is described by ellipsoid set. Then, in the process of structural parameter inversion, the nonlinear structural system is linearized and the linearization error is transformed. The system error is approximated to the system noise, the noise boundary is calculated by interval analysis method, and the ellipsoidal set is outsourced. Then, the state vector ellipsoid is updated continuously through the observation information, so that the real solution is gradually approximated. Finally, The boundary value of the convergent state vector ellipsoid is taken as the identification interval value of structural parameters, and the center value is taken as the optimal estimate value, which is used to indicate damage, thus realizing damage identification of uncertain structures. The regularization technique is adopted. The regularization introduced in the optimal estimation of set member filtering can be transformed into parameter inequality constraints, and the state vector ellipsoid can be updated. At the same time, it is regarded as a hyperplane body with prior information constraints. The state vector ellipsoid is further updated. The two improvements of the inversion algorithm can improve the accuracy of structural damage detection.
【学位授予单位】：南昌大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TU317

【参考文献】