Abstract: Traditional mechanism modeling method had large errors with the actual construction environment, whereas data-driven modeling mostly used black-box models, which was not conducive to knowledge discovery and understanding, therefore a knowledge mining-based shield machine cutterhead health assessment method was proposed. For the characteristics of shield excavation data with many dimensions, massive heterogeneity and strong noise interference, specific data pre-processing, feature screening and continuous feature discretization methods were proposed to establish a knowledge mining dataset combining the domain knowledge of shield excavation and machine learning algorithms. Then, the association rule mining algorithm was used to extract the mapping relationship among key features and different cutterhead health levels. The original rules were evaluated and ranked by using a comprehensive evaluation index that integrated reliability, completeness and simplicity to finally realize the shield machine cutterhead health assessment. The proposed method was validated based on the actual engineering data of one tunneling section of Xiamen Metro Line 3. The results show that the mined knowledge rules have a good agreement with the actual data distribution (average 93% or more), which verifies the effectiveness of the method.

Key words: shield machine cutterhead, health assessment, association rule, data mining, knowledge mining

摘要： 传统机理建模研究与实际施工环境误差较大，而数据驱动建模多采用黑箱模型，不利于知识发现与理解，为此提出一种基于知识挖掘的盾构机刀盘健康评估方法。针对盾构掘进数据维数众多、海量异构、强噪声干扰等特点，结合盾构掘进领域知识与机器学习算法提出针对性的数据预处理、特征筛选以及连续特征离散化方法，以此建立知识挖掘数据集。在此基础上，利用关联规则挖掘算法提取关键特征与不同刀盘健康状态之间的映射关系，采用融合可靠度、完整度与简洁度的综合评价指标适应度准则对原始规则进行评价排序，最终实现盾构机刀盘健康评估。基于厦门地铁3号线某掘进区间的实际工程数据对所提方法进行了应用验证。研究结果表明，所挖掘的知识规则与实际数据分布具有良好的吻合度（平均值93%以上），验证了该方法的有效性。

关键词: 盾构机刀盘, 健康评估, 关联规则, 数据挖掘, 知识挖掘