Machine Learning-based Classification System for Building Information Models
Loading...
Journal Title
Journal ISSN
Volume Title
Perustieteiden korkeakoulu |
Master's thesis
Unless otherwise stated, all rights belong to the author. You may download, display and print this publication for Your own personal use. Commercial use is prohibited.
Author
Date
2020-03-16
Department
Major/Subject
Computer Science
Mcode
SCI3042
Degree programme
Master’s Programme in Computer, Communication and Information Sciences
Language
en
Pages
51
Series
Abstract
Building Information Modeling (BIM) provides information regarding a building in digital format for the comprehensive building life cycle, which comprises building design, construction, and operation. Compared to traditional manual processes, which are based mainly on 2D drawings, BIM has a huge potential to reduce time and costs, and increase efficiency. The concept of classifications in BIM is particularly crucial throughout the life cycle of BIM since it gives structure to the data and helps users analyze different parts of the building model. Despite its importance, only a few BIM tools offer accurate handling of classifications but still with great limitations. They require human expertise and effort for manually assigning the right categories or defining classification rules for assigning the categories. The main objective of this thesis is to improve upon the conventional and largely manual approach to element classifications by utilizing rich data in BIM. To achieve this, three different types of information, which are numeric sizes, textual descriptions, and 3D shape, are extracted from elements in the BIM model. To process this information, a classification model, which concatenates outputs from a convolutional neural network and a fully connected neural network, is trained. Furthermore, the study defines metrics for measuring the quality of the data and examines their relationship with system performance. Moreover, the prototype of the classification system is deployed to a cloud environment to investigate the impact of system configuration on the performance. The results of the experiments empirically prove that a data-driven approach using the rich information of BIM enables automatic classification using machine learning. Experiments on different system configurations show that additional computing resources improve processing time with a trade-off of cost. Analysis of the quality of data and time-accuracy trade-off offers an insight into the optimal data selection and system configuration with practical examples. Finally, this study illustrates the potential of machine learning for improving the BIM classification tools that can be integrated into existing BIM tools.Description
Supervisor
Truong, Hong-LinhThesis advisor
Kannala, MattiPreidel, Cornelius
Keywords
machine learning, object classification, BIM, IFC, 3D convolutional neural network
