3D Intelligent Spatial Modeling for Infrastructure Digital Twins

近年来，数字孪生技术作为一种新兴的基础设施数据数字框架被引入，它创建了物理基础设施资产及其周围环境、流程和系统的数字副本。在此框架中，物理和虚拟基础设施模型（物理孪生模型）使用来自物联网（IoT）传感器（数据孪生模型）的数据绑定在一起，并使用高级数据分析技术（如人工智能，机器学习和预测分析）进行增强（智能孪生模型）。这种建模、分析和模拟将为预测性资产维护提供动力，支持规划决策，并实现国家现有和未来基础设施的性能优化。在这个发现提案中，我们提出了一种全面的方法来开发一个新的3D框架，该框架具有基础设施数字孪生模型、城市可持续性和智慧城市规划的潜在应用。通过推进我们正在进行的增强城市空间建模研究，我们将开发创新的深度学习技术，从视觉感知数据中自动重建基础设施资产的物理孪生模型，这将用于新兴的数字孪生技术的应用。该计划的目标包括：1）开发一种新的基于图的深度学习方法，以增强场景分割的上下文关系的结构化表示; 2）开发一种新的深度学习方法，用于重建具有组合基元集的建筑模型的广义形状;以及3）开发一种新的深度学习方法，用于树的逆过程建模。拟议研究的成果有可能显着推进和革命性的基础设施管理技术。该研究计划开发的系统和方法将导致高精度3D建模的自主生产和复杂基础设施资产的详细语义识别，能够与当前以人为中心的观察方法竞争。这将作为一个数字孪生平台，集成了传感器网络、数据分析和仿真，以提高基础设施的效率。在该计划中接受培训的HQP将能够将他们的知识和技能转移到加拿大的各个行业，并通过现代地理空间，数字孪生和人工智能技术为精密测绘领域做出贡献。 该研究项目的最终目标是开发合理的理论，通过感知和控制过程，在数字孪生框架内重建我们的基础设施和环境的通用和可互操作的3D模型，为创建机器智能和演示大规模工程应用中的工作系统做出贡献，以支持基础设施的可持续性和智慧城市。

In recent years, digital twin technology has been introduced as an emerging digital framework for infrastructure data, which creates a digital duplicate of physical infrastructure assets and their surrounding environment, processes, and systems. In this framework, the physical and virtual infrastructure models (physical twins) are tied together using data from Internet of Things (IoT) sensors (data twins), and enhanced using advanced data analytics technologies, such as artificial intelligence, machine learning, and predictive analytics (intelligence twins). This modelling, analysis, and simulation will power predictive asset maintenance, support planning decisions, and enable performance optimization of the nation's existing and future infrastructure. In this Discovery proposal, we propose a comprehensive approach to developing a novel 3D framework with potential applications towards infrastructure digital twins, urban sustainability, and smart city planning. By advancing our on-going research on augmented urban space modeling, we will develop innovative deep learning techniques to automatically reconstruct physical twins of infrastructure assets from visual sensory data, which will be used for application of emerging digital twin technologies. The objectives of this proposed program include: 1) developing a new graph-based deep learning method to enhance structured representation of contextual relations for scene segmentation; 2) developing a new deep learning method for reconstructing a generalized shape of building models with a combinatorial set of primitives; and 3) developing a new deep learning method for the inverse procedural modeling of trees. The outcomes of the proposed research have the potential to significantly advance and revolutionize infrastructure management techniques. The systems and approaches developed in this research program will result in the autonomous production of high-accuracy 3D modeling and the detailed semantic recognition of complex infrastructure assets, able to compete with the current human-centric methods of observation. This will be presented as a digital twin platform with the integration of sensor networks, data analytics, and simulation to improve infrastructure efficiency. The HQPs trained within this program will be able to transfer their knowledge and skills to various Canadian industries and contribute to the field of precision mapping via modern geo-spatial, digital twin and AI technologies. Our ultimate goal for this research program is to develop sound theories, which enable the reconstruction of generalized and interoperable 3D models of our infrastructure and environment within digital twin frameworks through perceptual and control processes, contributing to the creation of machine intelligence and the demonstration of working systems in large-scale engineering applications to support infrastructure sustainability and smart city.