Mechanical, Electrical and Plumbing (MEP) systems are critical assets in buildings. A series of systematic specifications have been developed and extensive experiences have been accumulated as human knowledge to guide the design and maintenance of MEP systems. However, most of the MEP-related knowledge is represented in the form of unstructured texts and heterogeneously dispersed in the design documents and Internet. It is therefore difficult for managing, querying and utilizing them. To address this challenge, the research study described in this paper constructed a knowledge graph by automatic collecting and storing of MEP knowledge from unstructured data. Specifically, the MEP documents were first acquired from the Internet, and multiple Natural Language Processing (NLP) techniques were then adopted to extract entity and discover relationship from the information documented in these documents. Finally, the knowledge graph was established and presented in a vivid form. The constructed knowledge graph is expected to contribute to the promotion of AI technology in the Architecture, Engineering and Construction (AEC) industry.
Recommended citation: Leng, S., Hu, Z.Z.*, Luo, Z., Zhang, J.P., and Lin, J.R. (2019). "Automatic MEP Knowledge Acquisition Based on Documents and Natural Language Processing" in Proceedings of the 36th CIB W78 2019 Conference. 800-809. Newcastle, United Kingdom.