基于神经网络的矩形钢管混凝土柱承载力分析及其可视化研究

As a main form of compostite structure, Concrete-filled Rectangular Steel Tube (CFRT) has been widely used in modern construction practice, and the carrying capacity calculation for CFRT column has become the key point for its engineering application. However, there are problems with both calculation accuracy and theoretical hypothesis of all the previous formulas, and the predictions of each calculation method show significant differences. Therefore,looking for a more precise carrying capacity prediction method for CFRT column and then displaying its loading process dynamically are the key problems that need to be solved in present civil engineering. Based on the analysis of previous experimental data, some new bearing capacity tests of CFRT columns will be designed and carried out in this project. Loading method,material property and cross-sectional dimention will be selected as the main parameters in the experimental study, and then lots of parametric numerical analysis will be performed. Based on the large quantity experimental and numerical simulation data, the artificial neural network, which is one of the intelligent computing techniques, will be adopted as a new method to establish an effective neural network model for predicting the CFRT column load-carrying capacity. Then, according to the detailed analysis on the data obtained from the whole process，the different failure mode will be investigated, the related deformation and evolution rules will be built, and the position changes of component node under different load will be calculated. Thus, the three-dimensional dynamic image of component deformation will be established and a fast and efficient way to obtain both the visual simulation of the whole process of specimen deformation and the final failure mode will be realized.In conclusion, the results of this project can expand the application of computer science into the civil engineering field. It not only provides a extensive applicability and high accuracy calculation method for the carrying capacity of CFRT column, but also has reference significance to the scientific computing and visualization for other engineering fields.

矩形钢管混凝土是组合结构的主要形式，应用越来越广，其承载力计算是工程应用的关键。已有计算公式具有精度和理论假定问题，各方法之间存在较大分歧，探索更高效准确的承载力分析方法并将其承载过程可视化是目前工程亟待解决的关键问题。本项目基于统计数据分析进行试验设计，补充关键受力形式和不同参数的构件承载试验及数值模拟，在得到大量数据源的基础上，突破常规手段，基于智能计算方法中的神经网络技术，训练出成熟的矩形钢管混凝土柱承载性能神经网络模型，进行参数化分析得到承载力高精度计算方法。结合参数化有限元全过程精细分析，剖析不同破坏模式，构建变形演化规则，计算不同荷载阶段的网格节点位置变化，建立构件变形的三维动态图像，快速高效获得构件变形全过程的可视化仿真及破坏模式。本研究成果可拓展计算机科学在土木工程领域的应用范围，不仅得到适用性广、准确度高的承载力计算方法，还对其它工程的科学计算及可视化有借鉴意义。

矩形钢管混凝土结构作为组合结构的主要形式之一，正在得到越来越广泛的应用。已有计算公式具有精度和理论假定问题，各方法之间存在较大分歧，探索更高效准确的承载力分析方法并将其承载过程可视化是目前工程亟待解决的关键问题。本项目在统计和分析了全面的钢管混凝土柱试验数据的基础上，对以往研究较少的矩形高强钢管混凝土柱进行了大量的轴压和偏压试验。对试验承载力、破坏模态及应力应变进行了深入分析以及有限元模拟，同时利用试验数据对现有规范进行了评价。建立了神经网络预测轴压和偏压承载力的模型（ANN），通过“Training-Validation-Test”过程得到了成熟可靠准确且具有良好泛化能力的模型，利用ANN模型对轴压和偏压钢管混凝土柱进行了参数化分析，得到了钢管混凝土承载力随不同参数的变化规律。同时，建立并训练了成熟的ANN模型对破坏模态以及钢管混凝土稳定承载力进行了预测。利用有限元软件分析了矩形钢管混凝土的受力机理以及破坏模态。使用人机交互方式，对构件进行参数化建模及三维可视化显示。基于前述的成果，软件集成包含计算与可视化两部分，利用神经网络以及推导的计算公式，可以在软件中计算出矩形钢管混凝土柱的承载力，同时根据有限元及试验的结果分析，显示出构件的变形形态，本项目成果为实际工程应用提供快捷高效、直观简便的设计方法。