Mining building information modeling (BIM) event logs for improved project management

• Main Author: Pan, Yue
• Other Authors: Zhang Limao
• Format: Thesis-Doctor of Philosophy
• Language: English
• Published: Nanyang Technological University 2021
• Subjects: Engineering::Civil engineering
• Online Access: https://hdl.handle.net/10356/152484
• Institution: Nanyang Technological University

Currently, Building Information Modeling (BIM) serves as a project management tool to inform data-driven decisions in modeling, construction, operation, and maintenance. As BIM is progressively adopted in civil engineering, one kind of important BIM data called event log will be accumulated continuously to bring about some features of “big data”. To be more specific, BIM event logs keep detailed records of timestamp, activity, actor, and others in chronological order to track the evolution of the construction project. Noticeably, a lot of knowledge is hidden behind such an ever-growing data source, which deserves deep exploration. However, it is still a comparatively new development in BIM event log mining due to the difficulty in handling the disordered and non-intuitive log data in unstructured text content. Therefore, the motivation of this thesis is to employ artificial intelligence (AI)-related techniques in massive log data to better comprehend the construction project and shed light on data-driven decision-making. The contributions of this thesis lie in two major aspects. From the technical perspective, it provides an opportunity to fill a gap of data science talent in the Architecture, Engineering, Construction, and Operation (AECO) industry. From the application perspective, it is a significant step beyond existing performance assessment methods heavily relying on subjective judgment, enabling improvements in both the building design and construction process. In general, the proposed BIM event log framework contains three major steps: (1) Data preparation from massive event logs; (2) AI implementation for log data mining; (3) Knowledge discovery as a smart decision tool. The key findings are summarized as follows: (1) The deep learning-based approach can learn designers’ behavior to make a sequential prediction about the next possible design command class towards automation of the modeling process, and thus following the suggested command classes can potentially accelerate the design and prevent some unwanted mistakes. (2) The clustering- based approach can automatically generate several patterns on behalf of a person’s design behavior characteristics and distinguish design efficiency into the high, medium, and low level for design performance evaluation, and thus these extracted clusters provide concrete evidence for managers to strategically schedule work. (3) The social network-based approach can graphically understand the collaborative design from discovering potential communities of designers, identifying a designer’s role, predicting work transmission and collaboration evolution, which hold the promise of promoting design collaboration through better leadership and work arrangement. (4) The process mining-based approach can simulate and analyze activities of modeling a building with inherent conflicts and uncertainty, which is useful in making process improvement through detecting potential deviations, inefficiencies, and collaboration patterns. Moreover, a digital twin integrating BIM, Internet of Things (IoT), data mining, and process mining is developed for process simulation, bottleneck diagnosis, and performance prediction, which is proven useful in facilitating the better understanding and optimization of physical construction operations. In brief, the proposed BIM event log mining presents a unique opportunity to convert data into meaningful information to provide a variety of value-added services, which is bound to create long-lasting positive impacts on driving construction project management to go through constant innovations towards digitalization and intelligence.