Learning formal activities representation for automated quality control, alignment, and revision of construction schedules

Amer, Fouad

Permalink

https://hdl.handle.net/2142/114085 Copy

Description

Title

Learning formal activities representation for automated quality control, alignment, and revision of construction schedules

Author(s)

Amer, Fouad

Issue Date

2021-12-01

Director of Research (if dissertation) or Advisor (if thesis)

Golparvar-Fard, Mani

Doctoral Committee Chair(s)

Golparvar-Fard, Mani

Committee Member(s)

• Hockenmaier, Julia
• El-Rayes, Khaled
• Liu, Liang
• El-Gohary, Nora

Department of Study

Civil & Environmental Eng

Discipline

Civil Engineering

Degree Granting Institution

University of Illinois at Urbana-Champaign

Degree Name

Ph.D.

Degree Level

Dissertation

Keyword(s)

• Construction Planning
• Construction Scheduling
• Natural Language Processing
• Artificial Intelligence
• Machine Learning
• Data Mining

Abstract

Construction planning requires domain know-how, which is often ingrained with experienced construction planners through exposure to multiple projects over decades of work experience. Without explicit in-depth prior knowledge of the underlying processes, planning experts learn to form empirical associations between project problems and potential solutions. Almost in all situations, their experience is described through examples from previous projects as opposed to structured and principled knowledge. Given the lack of formal definitions and structured knowledge, transferring expert human planners' comprehensive knowledge to automated planning systems proves to be complex. And, while substantial research is conducted and many automated planning systems were developed, the complexities of actual construction projects on one hand, and the nuances between different construction approaches, on the other, have rendered the manual creation of explicit knowledge representations extremely tedious and, in many cases, entirely infeasible. Hence, most --if not all-- construction projects still rely on a single or a small group of experienced practitioners to manually perform their planning and scheduling activities. This dissertation presents a suite of machine learning (ML) and natural language processing (NLP) based methods and systems to learn formal construction planning and sequencing knowledge representations from existing construction records. The learned knowledge is leveraged to (1) generate lists of potential successor activities given a sequence of predecessors, (2) retrieve lists of predecessor activities required to enable the execution of a sequence of successors, (3) critique the quality of logic of an input construction schedule by identifying missing or wrong logical dependencies, (4) schedule a set of unordered activities, (5) align weekly work plans (WWPs) to master schedules following a semi-automated human-in-the-loop approach, (6) identify the implicit dependency constraints governing predecessor-successor relationships and their roles and flexibilities in an input construction schedule, and (7) revise lookahead plans to mitigate the effects of unavoidable activity delays. All the developed methods learn using free form activity descriptions and Work Breakdown Structure (WBS) tags only without any prior assumptions or requirements on the expressions used to represent construction activities. They are trained and tested on a dataset of 32 construction schedules collected from real projects from different industry segments and various construction companies. Quantitative and qualitative results are presented to show the validity and the potential of the proposed methods. Limitations and future work are discussed in detail.

Graduation Semester

2021-12

Type of Resource

Thesis

Permalink

http://hdl.handle.net/2142/114085

Copyright and License Information

Copyright 2021 Fouad Amer

Owning Collections