Font Size: a A A

Neural Network-Based Structural Reliability Analysis And Optimal Design

Posted on:2005-12-13Degree:DoctorType:Dissertation
Country:ChinaCandidate:L ZhangFull Text:PDF
GTID:1118360125450011Subject:Solid mechanics
Abstract/Summary:PDF Full Text Request
The safety of engineering structures is one of the major objectives of structure design. In the practical engineering structures, uncertain information is existed universally, such as loading and structural parameters, and which lead to the appearance of random structure system with random parameters. The reliability study of random structure system is very important for design purposes, because which could help the designer to establish acceptable tolerances on structures and govern the fluctuations of the system parameters for safe operations. In the end, the reasonable structure could be designed with enough reliability and little cost.Neural network (NN) is a complex network system consists of many nerve centers with different frame. The form and function of each nerve center is simple, but the network system could realize different complex functions. Because NN is a complex non-linear dynamics system with many eximious characteristics, such as self-organizing, self-adapting and self-learning abilities, and so NN has been used successfully in different studies.The author focuses on the research of structural reliability based on NN in the paper, and the NN with Back Propagation (BP) algorithm is adopted, because which is the most adult and be used widely in practical engineering. In the present study, the use of NN is motivated by the approximation concepts inherent in reliability analysis. The applying of NN brings a new method to structural reliability, and then some difficult problems could be resolved effectively by NN. In this work, the main productions obtained are following as:1. A neural network method is advanced to simulate the distribution functions and their inverse functions merely based on the sample of random variables, and then the explicit formula of the distribution functions and theirinverse functions are given, which avoids the iterative process and the complex numerical integral in the supposition test method. At the same time, which is ensured that the distribution function obtained by NN is a monotone nondecreasing and bounded function. Moreover, the direct sampling method could be applied expediently in the sampling of random variables.2. Two methods are investigated in structural reliability optimal design. In the first one, a numerical approach method is used in the reliability optimal design with arbitrary distribution parameters. By this method, the probabilistic constraints could be transformed into deterministic constraints, and the reliability optimal design parameters could be obtained accurately and quickly. In the second one, MCS-NN method is applied in reliability structural optimal design with many failure modes, and then the explicit expression between the design parameters and the system reliability is given rightly. Therefore, the process of reliability optimal design can be implemented expediently.3. In the study of the local stress concentration, more attention has been focused on the confirmation of stress concentration factors at present, and seldom has been done on the corresponding reliability problem in the world. In the practical engineering structure, the local stress concentration is an important matter leads to the structural failure, and so it is necessary to research the structural reliability under the state of local stress concentration. In this study, NN is used to confirm the stress concentration factors firstly, and then the corresponding reliability analysis and optimal design methods are given for the structure with local stress concentration.4. For the reliability analysis of complex structures, SFEM (or PFEM) and RSM are the two main methods at present. SFEM method has been applied widely in practical engineering, which could solve normal distribution parameters and low non-linear problems effectively, but it is difficult to deal with non-normal distribution parameters and high non-linear problems. For this case, RSM method has been advanced recently, which could effectively solve the problem exits in SFEM method. For...
Keywords/Search Tags:random parameters, distribution function, structural reliability, local stress concentration, FEM, neural network, optimal design.
PDF Full Text Request
Related items