| Windstorm as one of the most common natural disasters causes huge losses to china as well as the world every year.According to the statistics of wind damage,the destruction and collapse of low-rise buildings with opening is particularly common.In order to figure out the vulnerable areas of low-rise buildings with opening and then provide a reference for wind resistant design,it is necessary to investigate the wind characteristics such as the distribution of wind pressure and the flow of wind field around buildings of this kind.Numerical methods are indispensable tools for investigating the wind characteristics of buildings.However,the steady wind or the unsteady wind with simple characteristics is frequently assumed as the inlet in the existing numerical researches for wind characteristics.Obviously,the wind field obtained on the basis of the above assumption is different from the real wind field,which leads to the lack of reliability of results.Therefore,using unsteady wind field that meets the requirements as the inlet is necessary for wind resistance research.To get over the shortcomings of the current inlet flow generation methods,the Generative Adversarial Network（GAN）for artificial intelligence is applied,which generates data by learning the distribution characteristics of data This paper proposes a novel approach based on GAN to generate a reasonable wind inflow,and thereby the wind inflow is employed to investigate the wind characteristics of the low-rise buildings with opening.Additionally,the precursor method is modified to generate training data required for the GAN,because of the difficulty and lack of measured wind field data.The main contents and conclusions of this thesis are as follows:1.A modified approach is presented to avoid the shortcomings of the traditional precursor method,that is,the unsteady wind field generated through Discretizing and Synthesizing Random Flow Generation（DSRFG）is as the initial wind field.A reasonable criterion for modeling is determined by meshing analysis.The modified approach is verified by comparing the generated wind field with the target wind field from four characteristics including average wind profile,turbulence profile,turbulent integral scale and power spectrum.The results show that the modified precursor method has better performance,and the characteristics of the generated wind field are close to that of the target wind field.2.A numerical procedure based on GAN is proposed to generate the inlet flow.The main steps of the procedure are as follows:（1）Two GAN models(i.e.GAN?and GAN（35）?)are first established by the modified precursor method,hereinto the former is used to generate single-point phase spectrum while the latter is used to generate phase spectrum difference per unit distance;（2）The phase spectrum is obtained by Center Progressive Method based on the results of step（1）;（3）The wind field is generated by the phase spectrum and the amplitude spectrum calculated from the power spectrum.The results of GAN are evaluated qualitatively from the perspective of data distribution and quantitatively through 1-NN algorithm.Furthermore,the characteristics of the wind field generated by GAN is compared with that of the target wind field.The results show that a high-quality wind field can be efficiently generated by GAN,that is,the data distribution generated by GAN is agree with the target data distribution and the characteristics of the generated wind field are close to that of the target wind field.Additionally,in the quantitative evaluation,the average 1-NN accuracy of GAN（35）?is 0.467 and that of GAN?is 0.518.The ideal 1-NN accuracy is 0.5,which indicates that the data generated by GAN（35）?and GAN?possess good performance.3.Computational Fluid Dynamics（CFD）is used to investigate the wind field characteristics of low-rise houses with opening,where the inlet flow is determined as the unsteady wind field generated by GAN.The influence of inlet flow,house ventilation,opening size and opening position on various characteristics of wind field is investigated respectively.The characteristics of wind field include wind field flow,net wind pressure coefficient,external wind pressure coefficient and internal wind pressure coefficient.The main conclusions are drawn as follows:（1）Taking steady wind field as the inlet will greatly underestimate the wind pressure around buildings,such as the external wind pressure coefficient is underestimated by about30%,the net wind pressure coefficient is about 9.8%,and the extreme wind pressure coefficient is about 50%;（2）The overall force is the largest for the house with single-sided opening while that is the smallest for the house with no opening.The net wind pressure coefficient and extreme wind pressure coefficient of the former is 2.8 times and 3.1 times as much as the latter respectively;（3）The wind pressure coefficient,net wind pressure coefficient and extreme wind pressure coefficient increase with the increasing of the opening ratio of the windward surface;（4）The overall force on the downwind surface with opening is the smallest,and the net wind pressure coefficient of downwind surface,roof and leeward is only about 0.2.The innovative points of this thesis are reflected four aspects as follows:1.A modified approach is presented to avoid the shortcomings of the traditional precursor method,where the unsteady wind field generated through Discretizing and Synthesizing Random Flow Generation（DSRFG）is as the initial wind field.The quality of the generated wind field is improved;2.The empty calculation domain is introduced in the wind field characteristics evaluation to verify the wind field characteristic retention capability;3.A method based on GAN is proposed to generate the wind flow from the perspective of data characteristics,which has strong adaptability and good quality of data;4.The unsteady wind field is taken as the inlet in simulate wind field characteristics of low-rise buildings with opening,which makes the results are more credible;The influence of the wind inflow on low-rise buildings is investigated,which provides a reference for others researchers to choose the wind inflow... |
PDF Full Text Request