STABILITY ASSESSMENT FOR SUSTAINABLE AND RESILIENT TUNNELLING USING AI

在过去的30年里，人口密集的城市之间对快速通勤的需求有所增加。随着技术的进步和汽车的环境优势以及所遇到的社会效益，对这种基础设施的需求也在增加。后者意味着，由于地形限制，连接（偏远）地区的铁路和公路隧道的数量也有所增加。对于现有隧道，尤其是在英国，其中一个关键的考虑因素是它们的稳定状态。更具体地说，维多利亚时代的隧道位于国家铁路网上，在过去的150年里，它们对促进全国铁路服务的可靠发展至关重要。然而，其中一些隧道已经开始挑战现有的网络。这些隧道最初是用几圈砖砌体作为衬里建造的。多年后，这些砖衬里开始退化，砖从周围的环中脱落，落在轨道上，使运输服务恶化。在这些隧道最初建造的隧道环境和条件方面存在科学和技术知识的差距，这加剧了它们目前的稳定性评估，在大多数情况下，这些隧道已经超过了其建议的使用寿命（Atkinson et al . 2020）。目前，所有铁路网络隧道的状况评估，都是由评估人员在轨道上进行手工、视觉和触觉调查，并在有需要时使用脚手架或移动高架工作平台进行评估。由于在铁路上工作涉及健康和安全风险，人工和主观评估过程，以及干扰通道，包括不允许火车运行的线路块，这导致了对旅行公众的干扰，因此我们一致努力使这些隧道的状况评估自动化。当前状况评估方法的另一个主要局限性是，隧道的检查报告是通过人工将可视化记录的缺陷输入到Microsoft Excel电子表格中生成的，该电子表格提供了检查结果的示意图记录和隧道状况标记指数，该指数对隧道状况进行评分。在比较不同时间或不同评估员进行的考试记录时，考试的视觉性质和对定性的、往往是主观的记录的依赖可能导致沟通不周。这种评估的主观性带来了安全问题，因为在进行定性评估时，由于隧道内的环境和条件过于严格，往往会遗漏缺陷。该研究项目的主要目的是研究人工智能（AI），机器学习/深度学习和信号/图像处理技术的适用性，以快速识别由于这些维多利亚隧道的生命周期中可能发生的各种地质和地质力学现象而在隧道中观察到的常见缺陷。关注长期机制(Paraskevopoulou, 2016；2018年，Paraskevopoulou等人2017年，Paraskevopoulou和Diederichs， 2018年)对当前状态做出了贡献，旨在扩大我们的行业合作伙伴Bedi咨询有限公司（Bedi）和国家铁路公司使用新兴技术（如3D激光扫描，摄影测量和无人机调查）捕获的数据的现有知识和实践，以调查自动化/学习算法的应用，客观快速地识别长期隧道性能。后者将有助于将这些与围岩的地质力学行为（Bedi et al 2017）联系起来，为可持续和弹性隧道工程提出缓解措施和设计指南（Bedi和Orr 2014）。

The demand for fast commuting between densely populated cities has increased over the last 30 years. The need of such infrastructures has risen along with the technological advancement and the environmental advantage to cars and social benefits that encounters. The latter implies that the number of railway and road tunnels connecting (remote) areas faster due to topographical limitations has also risen. One of the key considerations on existing tunnels especially in the UK is their stability state. More specifically, the Victorian tunnels that lie on the nation's railway network have been vital to facilitating a reliable rail service across the country for the past 150 years. However, some of these tunnels have started to challenge the existing network. These tunnels were originally constructed using several rings of brick masonry as a lining. After many years, these brick linings have started to degrade and bricks are becoming dislocated from their surrounding rings, falling on the tracks deteriorating the transport service. There is a gap of scientific and technical knowledge in the tunneling environment and conditions at which these tunnels were initially constructed which exacerbates their current stability assessment where in most cases these tunnels have exceeded their proposed lifetime (Atkinson et al 2020). Currently, condition assessment of all Network Rail tunnels is conducted by assessors through manual, visual, and tactile survey from track level and where necessary through the use of scaffolds or Mobile Elevated Working Platforms. Due to the health and safety risks associated with working on the railway, the manual and subjective assessment process as well as disruptive access that includes line blocks in which trains are not allowed to run, which leads to disruption to the travelling public, there is a concerted effort to automate the condition assessment of these tunnels. Another main limitation of the current method of condition assessment is that examination reports for tunnels are produced using manual entry of visually logged defects in to a Microsoft Excel spreadsheet that provides a schematic record of the examination findings and a Tunnel Condition Marking Index that scores the tunnel condition. The visual nature of examinations and the reliance on qualitative and often subjective records can lead to miscommunication when comparing the records of examinations carried out at different times or by different assessors. This subjectivity in assessment poses a safety concern in that defects are often missed due to the didcult environment and conditions within the tunnel under which qualitative assessments are visually made. The main aim of this research project is to investigate the applicability of articial intelligence (AI), machine learning/deep learning and signal/image processing technologies to rapidly recognise common defects that are observed in tunnels due to various geological and geomechanical phenomena that can take place over the lifetime of these Victorian tunnels. Focusing on the long-term mechanisms (Paraskevopoulou, 2016; 2018, Paraskevopoulou et al. 2017, Paraskevopoulou and Diederichs, 2018) that have contributed to the current state aiming at expanding the current knowledge and practice performed by our industry partners, Bedi Consulting Ltd (BEDI), and National Rail from data captured using emerging technologies such as 3D laser scanning, photogrammetry and drone based survey to investigate the application of automation/learning algorithms to objectively and quickly recognise long-term tunnel performance. The latter will contribute to relating these to the geomechanical behaviour (Bedi et al 2017) of the surrounding rock masses proposing mitigation measures and design guidelines (Bedi and Orr 2014) for sustainable and resilient tunnelling projects.