Towards Quantitative Integration of Reliability and Resiliency into Roadway Design: AI-Aided Road Vulnerability Assessment and Stochastic Uncertainty Modelling

加拿大拥有世界上最大的公路网之一。这些道路对加拿大人的经济繁荣和流动性至关重要。由于气候变化（如处理自然灾害后的大规模疏散）、道路使用者人口结构变化（如适应人口老龄化）以及新交通方式的引入（如自动驾驶汽车）带来的前所未有的挑战，对交通机构来说，评估和提高道路基础设施的可靠性和弹性是一个及时而关键的问题。为了确保道路能够应对并应对这些挑战，需要有效而准确的方法来评估现有道路的弹性。还需要将弹性定量集成到设计中的度量标准。我研究的长期目标是将设计和管理道路基础设施的过程转变为数据驱动和对不确定事件具有弹性的过程。我提出了未来5年解决关键知识差距的3个目标。首先，我将开发新的人工智能（AI）算法，用于从光探测和测距（LiDAR）和街道级图像中有效提取弹性关键道路特征。激光雷达是一种远程遥感技术，通过在道路上驾驶车载激光扫描仪来创建道路的虚拟3D模型（点云）。在我过去的工作基础上，我在这个项目中开发的算法将采用一种新的多尺度分割策略，自动提取弹性关键特征，用于前瞻性道路评估。其次，我将使用统计模拟来模拟不确定性，并评估使用人工智能算法提取的现有道路设计元素的可靠性。我将使用仿真结果提出新的弹性性能指标和设计图表，将可靠性定量整合到设计中。这项研究是开创性的，因为到目前为止，还没有办法在现有标准中定量地将可靠性和弹性集成到设计中。在最终目标下，我将使用空间统计来评估总体网络水平上的道路弹性。虽然已经对段级可靠性进行了评估，但网络级的评估是前所未有的。拟议的工作将为加拿大运输机构提供新的指标和新的算法，以主动评估道路的弹性和脆弱性。除了促进现有道路的弹性评估外，研究成果还将为比较新设计方案的弹性制定标准。这有助于设计更具弹性的道路，并在管理现有基础设施时优化有限公共资金的使用。在这个项目中，我将在机器学习、人工智能、统计建模和地理信息系统领域培养2名博士、2名硕士和5名本科生，这些都是工程市场上非常抢手的技能。

Canada is home to one of the world's largest road networks. These roads are essential to the economic prosperity and mobility of Canadians. Assessing and enhancing the reliability and resiliency of road infrastructure is a timely and critical issue to transportation agencies due to the unprecedented challenges resulting from climate change (e.g. handling mass evacuation after a natural disaster), changes in road user demographics (e.g. accommodating an aging population), and the introduction of new forms of mobility (e.g. self-driving cars). To ensure that roads are prepared for, and able to handle, such challenges, there is demand for efficient and accurate methods to assess resiliency of existing roads. There is also a need for metrics for quantitative integration of resiliency into design. The long-term objective of my research is to transform the process of designing and managing road infrastructure into one that is data driven and resilient against uncertain events. I propose 3 objectives over the next 5 years to address critical knowledge gaps. First, I will develop novel Artificial Intelligence (AI) algorithms for efficient extraction of resiliency-critical road features from Light Detection and Ranging (LiDAR) and street-level imagery. LiDAR is a form of remote sensing that creates virtual 3D models (point clouds) of roads by driving a truck-mounted laser scanner along a road. Building on my past work, which has been utilized by municipal and provincial agencies, algorithms I develop in this program will employ a novel multi-scale segmentation strategy to automatically extract resiliency-critical features for proactive road assessment. Second, I will use statistical simulation to model uncertainty and assess reliability of existing road design elements extracted using the AI algorithms. I will use simulation results to propose novel resiliency performance indicators and design charts for quantitative integration of reliability into design. This research is pioneering since, to date, there are no means of quantitatively integrating reliability and resiliency into design in existing standards. Under the final objective I will use spatial statistics to assess road resiliency on an aggregate network level. Although segment-level reliability has been assessed, a network-level assessment is unprecedented. The proposed work will present Canadian transport agencies with new metrics and novel algorithms for proactive assessment of road resiliency and vulnerability. Besides facilitating resiliency assessments of existing roads, research outputs will establish standards for comparing resiliency of new design alternatives. This helps design more resilient roads and optimize the use of limited public funds when managing existing infrastructure. In this program I will train 2 PhDs, 2 MScs, and 5 Undergraduates in areas of machine learning, AI, statistical modeling, and GIS, which are highly sought-after skills in the Engineering market.