机场道面变形与开裂模式人工智能分析方法

发布时间：2018-06-22 05:44

本文选题：机场道面 + 基层变形　；参考：《哈尔滨工业大学》2014年博士论文

【摘要】：航空交通量的高速增长及飞机大型化对机场道面工作性能和安全提出了更高的要求，并且从目前国内外的机场营运状况来看，诸多机场道面在未达到设计使用年限前，就出现了不同程度变形或开裂破坏，且破坏原因复杂多样。因此，机场道面的变形机理与开裂模式研究愈加引起研究人员的关注。其中，以美国机场道面国家实验室进行的足尺道面试验最为著名，获得的沥青混凝土机场道面变形过程数据、水泥混凝土机场道面开裂过程数据、以及相应的试验现象记录，为道面破坏机理、全寿命分析等研究提供了重要的、基础性的参考资料。然而，目前基于有限元分析的沥青机场道面变形过程仍然不能完全反映试验呈现的状态，有限元进行的混凝土道面开裂模式预测不但计算繁杂，而且裂缝模式理想化，与真实状态差别较大。针对这两方面问题，本文应用神经网络和元胞自动机等人工智能技术，直接从试验数据出发，开展了以下研究：基于美国机场道面国家实验室的足尺沥青混凝土道面试验数据，建立模拟重复荷载作用下沥青混凝土机场道面不可见基层动态变形过程的径向基函数神经网络模型。此模型反映了沥青混凝土机场道面面层与地下各基层动态变形之间的对应关系，从而能够重现道面不可见层的动态变形过程。模型选择中，依据物理概念尝试了几种可行方案，最终对面层部分采用“续点法”，陆续取面层相邻三点作为输入，依次得到不可见层每一点的位移值，从而得到整层的位移曲线。模型训练精度判定选用试验数据，表明所建立的模型能较为精确地反映道面动态变形过程，并能模拟目前有限元尚未模拟出的最下部基层局部起鼓现象。为了验证不可见基层道面变形人工神经网络模型的正确性，自行设计一个多层层状体系变形试验，得到荷载作用下完全弹性层状体的横截面变形过程曲线，将试验结果与模拟结果对比，检验模拟方法的合理性与可用性。基于所建立的神经网络模型模拟出的道面动态变形状态，，定义反映机场道面面层与基层变形过程的特征参数。为定量反映基层变形对道面总变形的贡献，引入了基层变形积参数。分析了各基层即时平均厚度和初始厚度比值关系、各层即时平均变形幅值与平均厚度比值关系、各层即时最大变形幅值与平均厚度比值关系、任意两层即时平均厚度比值关系及任意两层最大变形比值关系，从参数分析观点直观地展示了道面各剖面层的变形发展过程的特征及联系。分析进一步鉴证了影响道面面层状态的基层层次，以及影响的动态情况。应用ANSYS进行有限元与神经网络数值模拟结果的对比分析。基于FAA机场沥青混凝土道面试验模型的结构特点，建立道面变形的“残余变形累积”计算方法，并利用该方法进行机场沥青混凝土道面变形的数值模拟，并同神经网络模型得到的结果以及试验结果进行对比。结果表明，“残余变形累积”算法是可行的，较现有考虑各种非线性因素影响的有限元算法节省计算耗时。提出一种基于标准水泥混凝土道面试验开裂模式，预测新道面开裂模式的二维元胞自动机方法。与传统方式不同，该元胞自动机方法以大飞机重载作用下足尺道面板试验开裂模式为基础，建立起描述基础道板与新道板构造特征的的元胞自动机数值模式，继而通过道板内类似区域匹配准则和裂纹投射准则，图构出不同尺寸新道板的开裂模式，进而得到新道面的开裂模式。该方法初步实现了直接基于试验道面的开裂模式，预测不同尺寸、不同边界条件的混凝土道面的开裂模式。进行了机场水泥混凝土道面移动荷载下最大应力分布的有限元模拟。分别对水泥混凝土道面板不同尺寸、边界约束条件进行移动荷载下最大应力分布的有限元分析，以模拟数据获得最大应力图，并据此判断不同条件下数值模拟的道面开裂模式。同元胞自动机预测的道面破坏模式的相应结果做对比，验证了两种计算结果的一致性，同时，元胞自动机方法较有限元方法更为简便和高效。
[Abstract]:The high speed increase of air traffic volume and the size of the aircraft have put forward higher requirements for the performance and safety of the airport road surface work. And from the current situation of the airport operation at home and abroad, there are different degrees of deformation or cracking failure before the design life of the airport, and the cause of the damage is complex and diverse. The research on the deformation mechanism and cracking mode of the field surface has aroused the attention of the researchers. Among them, the full scale road surface test, which is carried out by the National Airport Road Surface National Laboratory, is the most famous, the data of the deformation process of the asphalt concrete airport pavement, the data of the opening process of the cement concrete airport pavement, and the corresponding test phenomena are recorded. For the failure mechanism of the pavement, the whole life analysis and other research provide important and basic reference materials. However, the current deformation process of the asphalt Airport Pavement Based on the finite element analysis still can not fully reflect the state of the test. The finite element method for the prediction of the crack mode of the concrete pavement surface is not only complicated, but also the ideal crack mode. In view of these two aspects, this paper applies artificial intelligence technology such as neural network and cellular automata to carry out the following research directly from the experimental data.
Based on the full scale asphalt concrete pavement test data of the American airport road surface National Laboratory, a radial basis function neural network model is established for the dynamic deformation process of the unvisible base course of the asphalt concrete airport pavement under simulated repeated loads. This model reflects the dynamic deformation between the pavement surface layer of the asphalt concrete Airport and the subbase subbase. It can reproduce the dynamic deformation process of the invisible layer of the pavement. In the selection of the model, several feasible schemes are tried in accordance with the physical concept. Finally, the "continuation point method" is adopted in the opposite layer, and the three adjacent points of the surface layer are used as input, and the displacement values of each point in the invisible layer are obtained in turn, thus the displacement curve of the whole layer is obtained. The model training precision determines the test data, which shows that the model can accurately reflect the dynamic deformation process of the road surface, and can simulate the local drum phenomenon at the bottom of the lowest level which has not been simulated by the finite element. In order to verify the correctness of the artificial neural network model of the invisible pavement deformation, a multi layer layer is designed by itself. The deformation test of the shape system is used to obtain the cross section deformation process curve of the fully elastic layered body under the load, and the test results are compared with the simulation results, and the rationality and availability of the simulation method are tested.
Based on the dynamic deformation state of the road surface simulated by the established neural network model, the characteristic parameters reflecting the plane surface layer and the deformation process of the base course are defined. In order to quantitatively reflect the contribution of the base deformation to the total deformation of the pavement, the base deformation product parameters are introduced, and the ratio relationship between the immediate average thickness and the initial thickness of the base course is analyzed. The relationship between the average deformation amplitude and the average thickness, the ratio of the immediate maximum deformation amplitude to the average thickness of each layer, the ratio relation of the instantaneous average thickness of two layers and the ratio of the maximum deformation of any two layers, the characteristics and relations of the deformation development process of each section of the road surface are intuitively demonstrated from the viewpoint of parameter analysis. The basic level of pavement surface condition and the dynamic state of influence are verified.
ANSYS is used to compare the numerical simulation results of finite element and neural network. Based on the structural characteristics of the asphalt concrete pavement test model of FAA airport, the calculation method of "residual deformation accumulation" of the pavement deformation is established, and the numerical simulation of the deformation of the pavement of the airport asphalt concrete is simulated with this method, and the neural network model is obtained. The results are compared with the experimental results. The results show that the "residual deformation accumulation" algorithm is feasible, and the finite element algorithm, which has the influence of various nonlinear factors, saves time.
A two-dimensional cellular automaton method based on the standard cement concrete pavement test cracking model to predict the crack mode of the new pavement is proposed. Different from the traditional method, the cellular automaton method is based on the full scale plane test cracking mode of the full scale road surface under heavy load of the large aircraft, and establishes the elements to describe the structural characteristics of the basic slab and the new channel plate. The numerical model of cellular automata, then through the similar region matching criterion and the crack projection criterion in the channel plate, is used to figure out the cracking mode of the new path plate of different sizes, and then get the crack mode of the new pavement. This method has preliminarily realized the crack mode based on the test surface and predicted the concrete pavement with different sizes and different boundary conditions. Cracking mode.
The finite element simulation of the maximum stress distribution under the moving load of the airport cement concrete pavement is carried out. The finite element analysis of the maximum stress distribution under the moving load on the different sizes and boundary conditions of the cement concrete pavement is carried out respectively. The maximum stress of the simulated data is obtained by the simulated data, and the pavement surface of the numerical simulation under different conditions is judged. The comparison of the corresponding results of the channel failure model predicted by the cellular automata verified the consistency of the two results. At the same time, the cellular automata method is more convenient and efficient than the finite element method.
【学位授予单位】：哈尔滨工业大学
【学位级别】：博士
【学位授予年份】：2014
【分类号】：U416.21;V351;TP18

【参考文献】