盾构渣土流塑性检测与改良状态评估方法研究

柳 洋

摘 要

随着我国城市化的快速发展，地铁工程成为了城市建设的新方向。在城市地铁工程建设的各类工法中，土压平衡盾构法凭借低成本、效率高、适用广等优势得到了广泛应用。在土压平衡盾构机掘进过程中，盾构机穿越的地层往往复杂多变，施工人员需要通过渣土改良工法使渣土保持塑性流动状态，从而实现盾构机土仓的稳定“进出土”和“保压”。目前，施工人员一般通过人工旁站或查看监控方式评价出渣口渣土形态，并结合盾构机掘进参数，综合评估当前渣土改良状态，该方式受工人技术水平和主观工作状态的影响较大，若不能及时反馈渣土情况，将导致盾构施工安全和质量问题。

针对土压平衡盾构掘进过程中存在的改良渣土流塑性“识别难”和改良状态“评估难”问题，本文提出了基于目标检测的改良渣土流塑性检测方法和基于FCM的渣土改良状态评估方法。首先，总结了改良渣土流塑性相关研究及评估标准，按照改良渣土形态进行流塑性分类，建立改良渣土形态数据集，构建基于YOLOv5的改良渣土流塑性检测模型，实现了改良渣土流塑性的自动检测。其次，系统分析渣土改良状态的影响因素，建立渣土改良状态评估指标体系，确定了渣土改良状态评估等级的划分标准与相应改良措施，将改良流塑性检测模型识别的改良渣土流塑性数据与百环掘进数据相结合进行聚类分析，划分了不同环号的渣土改良状态等级，实现了渣土改良状态的客观评估。最后，集成研究成果，搭建盾构排土流塑性检测与决策支持系统。

研究表明，采用本文提出的改良渣土流塑性检测方法和渣土改良状态评估方法，能够准确识别出渣口改良渣土流塑性和客观评估渣土改良状态。在改良渣土流塑性检测方面，针对出渣口三类改良渣土的平均检测mAP达0.781，FPS为47张/秒，能够较好地满足实时检测的精度和速度要求；在渣土改良状态评估方面，针对三类渣土改良状态的平均正确率达92.32%，与实际工程情况较为符合，有助于实现渣土改良状态的准确、高效、客观评估，为盾构机的实时参数调控提供了有力支持。

关键词： 土压平衡盾构；渣土改良；流塑性；目标检测；模糊聚类

Abstract

With the rapid development of urbanization in China, urban space is becoming more and more tense. In order to alleviate the urban pressure caused by population expansion, the development of underground space mainly based on subway construction has become a new direction of urban construction. Among various construction methods of urban subway construction, EPB is widely used because of its advantages of low cost, high efficiency and wide application. In the process of EPB shield tunneling, the stratum traversed by the shield machine is often complex and changeable. It is necessary to improve the conditioned soil to make the conditioned soil flow in a plastic state, so as to realize the stable "in and out" and "pressure maintaining" of the shield engine room. At present, constructors generally evaluate the form of the conditioned soil at the soil outlet through manual side station or observation image, and comprehensively evaluate the soil conditioning status in combination with the tunneling parameters of shield machine. This method is greatly affected by the technical level and subjective working state of workers. If the soil conditioning status is not fed back in time, it will lead to the safety and quality problems of shield construction.

Aiming at the problems of "difficult identification" of conditioned soil plasticity and "difficult evaluation" of soil conditioning status in the process of EPB shield tunneling, this paper puts forward the conditioned soil plasticity detection method based on target detection and the soil conditioning state evaluation method based on FCM. Firstly, this paper summarizes the relevant research and evaluation standards of conditioned soil plasticity, classifies the plasticity according to the conditioned soil form, establishes the conditioned soil plasticity data set, constructs the conditioned soil plasticity detection model based on YOLOv5, and realizes the automatic detection of conditioned soil plasticity. Secondly, this paper systematically analyzes the influencing factors of soil conditioning status, establishes the evaluation index system of soil conditioning status, determines the division standard and corresponding improvement measures of soil conditioning status evaluation grade, combines the conditioned soil plastic data identified based on the plastic detection model with the first hundred excavation data for cluster analysis, divides the soil conditioning status grades of different ring numbers, and realizes the objective evaluation of soil conditioning status. Finally, integrating the research results, a conditioned soil plasticity detection and decision support system based on EPB is established.

The results show that the conditioned soil plasticity detection method and the soil conditioning state evaluation method proposed in this paper can accurately identify the conditioned soil plasticity at the soil outlet and objectively evaluate the soil conditioning state. In terms of conditioned soil plasticity detection, the average mAP of three types of conditioned soil is 0.781 and the FPS is 47 sheets / s, which can better meet the requirements of real-time detection accuracy and speed; In the evaluation of soil conditioning state, the average accuracy of three types of soil conditioning state is 92.32%, which is more consistent with the actual engineering situation, which is helpful to realize the accurate, efficient and objective evaluation of soil conditioning state, and provides strong support for the real-time parameter regulation of shield machine.

Key words: EPB, soil conditioning, plasticity, target detection, fuzzy clustering