基于检测数据智能分析的力学建模与盾构载荷地质适应性研究

大型盾构掘进装备的载荷是直接关系施工安全与效率的关键力学量，随着基础设施建设快速发展，对于我国多样性地质条件与长隧道建设工程实践，实现载荷智能控制的需求日益凸显。目前载荷建模理论难以突破单一地质本构的局限，迫切需要具有较好地质适应性的载荷研究新方法。本项目提出将智能机器学习与力学量纲分析结合的工程检测数据分析新思路，并应用于掘进载荷地质适应性建模。针对高维、多源的载荷影响因素以及变量间耦合等特点，采用主成分分析、正交试验、聚类等分析方法组合，进行数据降维，对众多影响因素分类归纳、筛选判别主要特征，通过量纲分析揭示较为普适的载荷与地质类及操作类主控参量间内在作用规律，建立包含剪切破坏和张拉破坏两大类地质适应性特征的载荷模型构架，结合具体装备构型与地质工况定量识别系数，给出适用多样地质的较为准确的载荷力学预测模型，为智能控制提供理论基础，并扩展实验力学数据分析在交叉领域中的应用。

The loads acting on the shield tunneling machine are crucial mechanical parameters, which are closely related to safe and efficiency tunneling process. Since the rapid development of infrastructure construction, there has an increasing demand for the intelligent control of the loads, especially in terms of the complex geological conditions in long distance construction. At present, the theory of modeling on shield loads is still confined to single geological constitutive equation. Therefore, it is urgent to have a new and proper method with better geological adaptability. This project proposes a new approach on the analysis of on-site detection data through the combination of machine learning and dimension analysis. This method will be applied to load-modeling and discuss its geological adaptability. For high-dimension, multi-source, and multi-coupling characteristics, algorithms such as principal component analysis, clustering and orthogonal test will be used for dimensionality reduction of data. Moreover, with the classification for the various factors and the selection for main features, it is feasible to reveal common rules among the loads and the geological/operational parameters by dimensional analysis, as well as to establish a novel model which contains both two main kinds of geological characteristics. They are the shear failure of soil and the tensile failure of rock. The identification of coefficients in model is performed to estimate the effect of the structural characteristics and the specific geological parameters for accurately load predicting under various geology. The work will provide theoretical basis for intelligent control and expand the application of experimental mechanics data analysis into crossover fields.