Study On Dynamic Reliability Of Random Structures Based On Machine Learning

The thin-walled circular tube is one of commonly used in aviation and aerospace fields.It mainly experiences the thermal load generated by sunlight in space,and the thermal deformation,thermal stress and thermal fatigue of the structure under the thermal load have gradually become a research hotspot for this kind of structures.Therefore,it is of great significance to analyze its reliability under sunlight in the practical engineering.The temperature in the structure will change and cause thermal load due to sunlight,and subsequently the structure will deform due to the thermal load,and the deformation will then affect the temperature distribution of the structure again.Finally,the structural thermal vibration occurs,at this time,the fluid inside the structure will also vibrate,and the interaction between heat,structure and fluid is called the thermal-structural-fluid three field coupling phenomenon.In this work,the dynamic response of thin-walled circular tube under three field coupling is mainly addressed and the reliability of thin-walled circular tube is analyzed by machine learning as well.The main content of this work is as follows.(1)The development of structure reliability,machine learning and intelligent algorithm are introduced,and the research status of thin-walled tube structure under thermal load is summarized.The application status of machine learning and intelligent algorithms in the field of structural reliability is then discussed,and the key theories used in this work are analyzed.The main research content and specific research methods are also determined.(2)Based on the theory of heat transfer,the dynamic finite element equation of thin-walled circular tube structure under the thermal-structural-fluid coupling is established,and the procedures to solve the equation and obtain dynamic responses by Wilson-? method are presented.The structural performance function is constructed by using S-N curve based on the load-life interference.Then,a surrogate model for solving the dynamic response is constructed with the artificial neural network(ANN)so as to improve calculation efficiency,and the structural failure rate is also studied.In numerical example,the dynamic response of the component is solved,and the failure rate is analyzed with the presented surrogate model established by ANN.The efficiency of ANN method is verified by comparing the results obtained from the model presented with those from Monte Carlo method.The results show that the structural failure rate will increase with the increasing number of cycles of high cycle fatigue(HCF)stress.(3)In order to improve the efficiency,an alternative model to solve dynamic response is established by using support vector regression(SVR),and the parameters of SVR model are optimized with genetic algorithm,in which the problem of initial selection of SVR parameters is dealt with well.A distributed collaborative extremum support vector regression method is proposed by combining the SVR,DC strategy and extreme response.Considering the interaction of high cycle fatigue stress and low cycle fatigue stress together and the strength degradation of components,the performance function is constructed based on residual strength.Finally,the structural reliability is computed by using DC-ESVR proposed.The feasibility of DC-ESVR is verified by using numerical examples.Moreover,the efficiency of DC-ESVR is verified by comparing results of Monte Carlo method with those of DC-ESVR.The calculation results indicate that the reliability decreases with the increasing HCF cycles and the strength degradation coefficient.