AbstractsEngineering

In the Architecture, Engineering and Construction (AEC) industry, inadequate collaboration between project stakeholders and disciplines often leads to conflicts and interoperability issues. Research has been conducted in knowledge formalization to bridge the knowledge gaps and information silos. Formalizing construction knowledge is challenging to formalize because most construction knowledge implicitly resides in the minds of construction experts, which is difficult to represent in a formal and explicit manner. The proposed study is built upon previous research findings, and attempts to formalize tacit knowledge in Mechanical, Electrical and Plumbing (MEP) design coordination by capturing necessary information with a model-based information capture system and reasoning about the captured data with data mining techniques. The vision of this research is that the formalized knowledge can be used to provide guidance for early design review incorporating construction considerations, facilitate structured learning from past experience, as well as train novice engineers. In summary, this research has three main contributions. First, this research presents a formalized knowledge representation schema to capture process knowledge in design coordination, which was successfully implemented in a model-based knowledge capture system developed by the author. Second, a model-based knowledge capture system was developed to store clash information in the form of categorized features and link such categorized information directly to the relevant model elements, which can also facilitate organization and management of clashes and supports searching and grouping functions. A prototype system was developed as a plugin to a widely used BIM-based design coordination application and was demonstrated with project data gathered from three new construction projects in the United States. Third, this research applied data mining techniques for knowledge discovery and reuse in MEP design coordination. Classification models were developed to provide predicted solutions for identified clashes based on historical data. The classification algorithms that produced the best results were selected, which reached precision rates of over 70%. The effectiveness of the classification models was tested in a novice experiment.