The Application On Fatigue Analysis Of Structure Based On ANN

Fatigue is one of the most common failures in mechanical structure, and the fatigue reliability index is also an important technical item for mechanical products. So, how to estimate the fatigue life and reliability accurately is of significant practical and theory value for improving the quality of the product and eliminating the hazard of accident.However, the fatigue of material is a very complex problem because it is affected by many random factors and can't be expressed by explicit mathematic formula. Therefore, the much accurate method to determine the reliability of material under given fatigue life usually uses an experimental technique which is done under the real load condition, then, the fatigue life can be obtained by analyzing the statistical experimental results, so does the reliability under given fatigue life. In reality, constrained by the conditions, it is difficult to carry out plenty fatigue experiments, so it is hardly to obtain satisfactory results by statistical analysis.Artificial neural networks (ANN) is a rapid developed front subject to meet the requirement of actual engineering and technical fields. ANN is a technical system that simulates the structure and function of human brain neural network. In recent years, it developed rapidly, and obtained wide spread application. ANN is a kind of large-scale parallel non- dynamic system, with good non-linear mapping ability, the formidable solution counter-question ability, the real-time computation ability as well as the characteristic of simulation processing and the digital processing coexistence. Therefore, as an outstanding non-parameter method, ANN can be used in analysis of the reliability of the structure.Among so many types of artificial neural networks, the feed-forward networks, BP and RBF networks are both suitable for nonlinear function approaching. However, BP network has the limitation of local minimum, while RBF network has biology background and based on function approaching theory, the output of network has linear relation to the connection weights, so it is suitable for nonlinear function approaching of multiple variables for its arbitrary none-linear approaching ability. RBF network especially suitable for the fields of multiple variables none-linear approaching, mode identification as well as adapt filtering. The RBF network has the best approximation performance, and its structure has the characters of linearity relations between outputs and weights, fast training, without local optimum, thus in this paper the RBF network is introduced to reliability analysis of structure.In this paper, the histories and present situations of reliability and optimization are summarized, and the artificial neural network is introduced. Based on the traditional structure reliability analysis and optimization design method, a new approach is put forward, that is the response surface method and optimization design based on RBF artificial neural network. This method can be used to simulate the performance function that can't be expressed explicitly, and with the simulated performance function, the reliability can be calculated, meanwhile, the detailed computation steps are provided. With finite element method and statistical analysis, the ANN topology can be established for analysis of the reliability. The concrete reliability index can be obtained to the implicit function or the performance function can't be explicitly expressed.To illustrate the utility of the proposed method, some examples are taken and the analysis results show that this method has much accuracy and has practical value.The main contents and achievement of this paper are as follows: Firstly, the significance of the fatigue analysis and optimization design is presented, and the histories and present situations which are summarized, and the limitations of traditional structure reliability analysis methods are discussed.Secondly, the artificial neural network is introduced.0n the basis of analysis of advantage and disadvantage of some neural network, Radial Basis Function (RBF) neural network is chosen to establish the reliability analysis and optimized design model.Thirdly, based on the traditional structure reliability analysis and optimization design method, the response surface method and optimization design based on RBF artificial neural network is put forward. With finite element method and statistical analysis, the ANN topology can be established for analysis of the reliability. The concrete reliability index can be obtained to the implicit function or the performance function can't be explicitly expressed. Some examples are taken and the analysis results show that this method has much accuracy and has practical value.This paper is to the question of fatigue life reliability analysis and optimization design of the mechanical structure, the response surface method and optimization design based on RBF artificial neural network is put forward, it is a beneficial research for further consummated and development of structure fatigue life reliability analysis and optimization design. With further development of ANN theory and the neural computer technology, the ANN based fatigue life reliability analysis and optimization design method research will have a more important theoretical and the practical significance.