Towards fully integrated and automated construction planning on earthworks projects

Abstract: Cost effective sensors are increasingly adopted to enable a smart construction site that integrates sensory information and algorithms to improve the performance of construction operations from various perspectives including cost, schedule, risk, safety and so on. Imaging sensors and location sensors are increasingly applied in construction to monitor the site and tracking the movements of equipment and materials. On the other hand, various analytical methods and simulation-based approaches have been proposed to analyze and optimize construction operations. Such analyses are usually performed based on a predefined or presumed project execution plan. Generating a project execution plan from real-time sensory information becomes a bottleneck for the seamless integration of the information system and further analysis. This thesis addresses the challenge of linking the sensory information and the analysis algorithms with automated planning, with a focus on earthwork projects. The first challenge is that a valid plan requires sufficient understanding of the project from heterogeneous information coming from the design and the site. This includes the design of the structure, the actual progress and construction site condition, and the surrounding environment that may introduce additional constraints or may be affected by the construction operations. This thesis proposes an information management system based on Keyhole Markup Language (KML) and Google Earth to manage and visualize heterogeneous site information, especially 3D models, aerial and ground images, panoramas and Geographic Information System (GIS) data of the site environment. The system is the first to be demonstrated on integrated management of 3D models, images, and GIS data. For this problem, the heterogeneity of the data and the efficiency and effectiveness requirements for information retrieval and visualization raise a particular challenge. Through the combined use of photogrammetry and advanced visualization technologies such as virtual reality(VR) and augmented reality (AR), the system provides a cost-effective approach for identifying potential problems and finding practical constraints on the construction site. For earthwork projects, it provides quantitative progress updates with image-based 3D reconstruction and qualitative constraints obtained through interaction with the user. Another critical challenge is to formulate the earthwork planning problem and generate an optimized executable plan automatically given the identified constraints. Though earthwork allocation optimization has been heavily investigated in existing research, producing a valid execution plan, i.e., the work breakdown structure and the project network has rarely been touched. The involvement of sequence in planning raises great challenges for optimized planning. Another critical issue in planning earthmoving operations is temporal-spatial conflicts. Because modeling such a conflict requires prior knowledge of individual activities, it introduces a chicken-and-egg loophole for activity definition. Towards this end, I split the automated earthmoving planning problem into the optimization problem and the automated planning problem. In which, the optimization module models earthmoving operations using flow network model and produces an optimized solution pool without introducing temporal-spatial conflicts. Then, an automated planner based on the classical planning model in automated planning research is proposed to derive a temporal-spatial conflict free plan from the optimized solution pool using heuristics based approach.