| Glass plate structures are widely used for engineering both in China and foreign countries.Generally,such structure is composed of glass plates and a support system.It is not complicated but used in a complex environment and has a rigorous requirement for the running condition.Under the long-term cyclic action of wind load,the support system is often subjected to significant wind-induced fatigue failure.Therefore,studying the wind-induced fatigue of such structure and its fatigue failure law is important for guaranteeing its running safety and reliability in its service life.Heliostat and glass curtain wall are two typical glass plate structures and thus were used in this study to survey the wind-induced fatigue of their support systems.A heliostat is a reflection unit in a solar power tower plant.It is an independent column structure arranged in a flat and wide region such as the open desert or Gobi desert generally.In the running process,a large area on two sides of the mirror board bears wind pressure for a long time,causing continuous accumulation of fatigue damage on the members of the support system.Gradually,fatigue failure would appear,posing a serious safety risk.In this study,the authors carried out a wind tunnel test on the support system of a h eliostat and simulated the fatigue of the support system to explore its wind-induced fatigue.Combined with a multivariate joint probability distribution model,an artificial neural network(ANN)was used to analyze the distribution law of the support system ’s fatigue life and predict the service life of its members.The main study results are provided below:(1)A wind tunnel test was conducted on a heliostat model u nder the designed working condition in a debugged wind field.Consequently,the time course of wind pressure on the mirror was obtained and used as the basic data of wind load in this study.The data analysis results indicate that the absolute value of the coefficient of net wind pressure on the mirror was 0-2.5.The coefficient of net wind pressure in the middle of the mirror was large and decreased outward gradually.(2)A finite element model of the heliostat structure was established.According to the wind pressure obtained from the wind tunnel test,the authors simulated the fatigue of the heliostat structure,conducted statics analysis,transient analysis,and fatigue analysis,and resolved the force on the model.Further,the fatigue damages at a dangerous point under different working conditions were calculated in the time-domain method,frequency method,and nCode fatigue analysis method respectively and compared to know the advantages,disadvantages,and precisions of the said methods.Compared with the rain-flow counting method,the nCode fatigue analysis method has a relative error of not more than 20% basically and thus can be used for fatigue calculation instead of the rain-flow counting method.(3)A multivariate joint probability distribution model suitable for analyzing the wind-induced fatigue of heliostat was created based on the movement of the sun and the joint distribution law of wind speed and direction.The typical working conditions were determined by mixed uniform design.In combination with the decision tree algorithm,a regression analysis was made on the finite element simulation results and the regression effect was verified.The results show that the regression value and the actual value had a mean deviation of 20% at most and a maximum deviation of 35% at most,which met the requirement for prediction.Furthermore,the authors also conducted a wind-induced probabilistic fatigue analysis,counted the working conditions of partial highly-probabilistic fatigue damages,and predicted the service life of the structural members.(4)The ANN model was optimized to conduct the regression analysis appropriately and predict the fatigue life distribution law of the heliostat structure,avoiding a large workload on the wind tunnel test and fini te element simulation.Besides,the regression effect of the optimized ANN was tested.The results reveal that most regression values deviated by less than 20% from the actual values,which ensured the reliability of the regression analysis.As discovered by summarizing the distribution laws and characteristics of the wind-induced fatigue life of the heliostat under different factors,the fatigue life of the selected member conformed to Gaussian distribution or Lorentz distribution roughly,while the critic al part vulnerable to fatigue damage of the column member was at the bottom.Data demonstrate that about 95% of fatigue damages at the dangerous points in the critical part were produced under 42 unfavorable working conditions.A glass curtain wall as a key part of a supertall is affected by wind greatly.For this reason,fatigue failure often appears in the structural connections of the curtain wall’s support system.Similar to the heliostat,the glass curtain wall also consists of glass plates and a support system.However,the curtain wall is mounted at a high position in a more complicated urban wind field.Hence,the wind-induced fatigue analyses of the two typical structures have both similarit ies and differences.In this study,wind tunnel test and fatigue simulation were conducted on a glass curtain wall for a supertall.Some mainstream regression algorithms in machine learning were adopted to calculate the probabilistic fatigue damages of different structural members and predict their service life.The main results are as follows:(1)A wind tunnel test and a fatigue simulation were implemented on a glass curtain wall to acquire the distribution of wind pressure on the structure.The shape coefficients in different areas as well as the standard values of wind loads in different floors and zones were calculated.Based on this data,the key area of the structure liable to wind-induced fatigue was determined.Additionally,the authors established a joint probability distribution model of the glass curtain walls and calculated the parameters of the model.The results suggest that the type-Ⅱ distribution of generalized extreme values accorded with the local wind speed and direction.Further,they also analyzed the force on the support system of the structur e and the fatigue in the part bearing large force.(2)Some mainstream regression algorithms in machine learning(including radial basis function neural network(RBF),fuzzy neural network(FNN),support vector machine(SVM),and random forest method)were used to make regression analysis on the simulated fatigue data of the structure ’s support system.Over a comparison between the advantages,disadvantages,and results of the algorithms,it is found that RBF took a short time and had a small deviation.RBF can be regarded as the best choice for calculating the fatigue life of the structural member.Based on the regression analysis results,the probabilistic fatigue damage on the member was predicted and its estimated distribution was verified by a Chi-squared test.The result implies that the member’s fatigue life was congruous with the kernel distribution.Furthermore,the parts with large and small probabilistic fatigue damages on the member were counted and the member’s service life was predicted for providing a reference and basis for the fatigue design of the glass curtain wall’s support system. |
PDF Full Text Request