Automatic Classification of Design Conflicts Using Rule-based Reasoning and Machine Learning—An Example of Structural Clashes Against the MEP Model

Abstract

With the emergence of 3D technologies in a recent decade, BIM software makes it easy to detect those conflicts in the early stage of a project. Clash detection in BIM software is now a common task. Among those conflicts found by BIM software, however, a relatively high percentage belongs to 'pseudo conflicts'-which are permissible or tolerable, but BIM software does not reveal this information. Thus, currently BIM managers have to manually inspect every detected conflict to classify the type of conflict. Some researchers urged an automated process to facilitate this laborious process. This study implemented both a rule-based reasoning system and machine learning classifiers to help classify those BIM-detected conflicts. Preliminary testing results indicate that machine learning algorithms can achieve comparable results to a traditional rule-based system, but with much less costs and energy in developing.

A BIM-Based Approach for Optimizing HVAC Design and Air Distribution System Layouts in Panelized Houses

“…For example, if the MEP system and the structural system collide, it is considered a hard clash; in such a case, the MEP system must re-route to solve this conflict. According to [67], machine learning can be used to classify the design conflicts in BIM automatically. Furthermore, it is concluded that the current method of overlaying 2-D drawings to identify interferences of building MEP systems and structural framing is "prone to errors, causing ad-hoc rework and reduced headroom and maintainability issues" [67].…”

Investigation of Machine Learning for Clash Resolution Automation

“…Machine Learning is a branch of computer science that focuses on imparting knowledge to the machines to self-perform tasks (Huang & Lin, 2019). The process of imparting knowledge or knowledge modeling in Machine Learning can be divided into two main steps.…”

“…Application of rule-based automation for clash resolution has been tested. However, developing rule-based automation systems are time-consuming, and their effectiveness depends on the number of rules defined (Huang & Lin, 2019). One alternative solution to rule-based automation is the implementation of Machine Learning.…”

“…Machine Learning is a branch of computer science that relates to how machines can learn to selfperform tasks similar to humans without explicit coding (Huang & Lin, 2019). Based on how knowledge is transferred to the machine, machine learning can be divided into supervised learning, unsupervised learning, and reinforcement learning (Theobald, 2017).…”

“…Hsu, et.al, 2020 used supervised learning to automate the clash resolution of MEP for the basement of student residence (Hsu et al, 2020). Huang and Lin, 2019 worked on comparing the accuracy of rule-based automation and supervised learning in automating the clash classification task (Huang & Lin, 2019).…”