Virtual and visual production management system for proactive project controls

Team-based planning and real-time communication of risk and progress are two principal mechanisms of a proactive project controls program; a program that maintains a smooth flow of production on construction projects. Recent empirical observations from construction projects suggest that successful implementation of such control mechanisms requires dedicated facilitators and engages practitioners in a relatively in-depth learning process. Sustaining this level of commitment for the duration of all projects is difficult, and in its absence, many project teams revert to traditional practices. To address these knowledge gaps, this thesis presents the theoretical foundation for a project controls system that a) improves understanding of how construction performance can be captured, communicated, and analyzed in form of a production system; b) predicts the reliability of the weekly work plan and look-ahead schedules, supports root-cause assessment on plan failure at both project and task-levels; c) facilitates information flows; and d) decentralizes decision-making. Specifically, a new model-driven computer vision technique is presented that takes advantage of the growth in visual data available on jobsites (i.e., ground and aerial images and videos) to map the current state of production in 3D and over project timeline to expose waste at both project and task-levels. To automate measurement of construction progress, these production maps are first integrated with BIM and project schedules. The resulting visual production models are used within a new end-to-end system that reasons about geometry and appearance and reports the state of work in progress per BIM element and schedule task. The method uses a convolutional neural network (CNN) model to assess both geometry and appearance of the site using production maps. This model is trained using new synthetic and real-world material datasets and also uses a multi-view inference procedure to improve the accuracy of progress monitoring. By comparing these production maps against expected progress reported via 4D BIM, and also leveraging new statistical models, the reliability in the future state of production is forecasted to highlight potential issues in a location-driven scheme. A new client-server architecture interface is also introduced that supports collaborative decision making that eliminates root causes of waste; and provides visual interfaces between people and information that enable effective pull flow, decentralizes work tracking, and facilitates in-process quality control and hand-overs among contractors. Experimental results from several real-world projects are shown to validate every algorithm and the visual and virtual project controls system as whole. The benefits and the limitations of the proposed method are also discussed using real-world Key Performance Indicators (KPIs).

Name thesis:degree_name

Ph.D.

Level thesis:degree_level

Dissertation

Discipline thesis:degree_discipline

Civil Engineering

Grantor

University of Illinois at Urbana-Champaign

Year dc:date

2020

Author dc:creator

• Lin, Jacob Je-Chian

Contributors dc:contributor

• Golparvar-Fard, Mani
• El-Rayes, Khaled
• Liu, Liang
• El-Gohary, Nora
• Hoiem, Derek

• project controls
• point cloud
• BIM
• lean construction
• progress monitoring
• deep learning

Statement dc:rights

• Copyright 2020 JACOB JE-CHIAN LIN

Language dc:language

en

Handle dc:identifier

http://hdl.handle.net/2142/108709

OAI identifier oai:identifier

oai:www.ideals.illinois.edu:2142/108709