Automated As-Built Building Information Modeling: Human Computation: an Interactive Refinement Approach

(Warning: this essay is in a very rough form, I am still trying to make it more understandable, if you find the problem description paragraph is vague, please let know, I will try to improve it based on your comments, if you know any references related to this type of study, please let me know, I will be happy to share my reference library with you, thank you very much!)

The potential of Building Information Modeling (BIM) technology for various engineering applications has been recognized by many people from both industry and academia. These applications include: facility management, building performance simulation, construction scheduling, and many more. One critical issue to make these digital models of facilities is to make the information in it accurate and up-to-date. Since BIM comes from the domain of building design, it is very good at helping people to represent their design intent, and most BIM tools (e.g. Revit, ArchiCAD) provide automated supports for engineers to design new buildings and navigate through their virtual design models. However, these design models usually live in a perfect "virtual world", and might not necessarily reflect the actual status of "imperfect" buildings in the real world. Hence, according to what is actually constructed on the site, it is necessary to update the BIM design models to reflect their "as-built" status on the site.

A natural approach for capturing as-built conditions of constructed facilities is to utilize 3D imaging technology to capture detailed 3D geometry of these facilities, and reconstructe 3D as-built polygonal models of these facilities. Based on these 3D polygonal models, it is possible to use them as a starting point for reconstructing an as-built BIM. Generally, the process is: collect 3D data --> reconstruct 3D polygonal model in a CAD package such as AutoCAD --> export reconstructed CAD model into a BIM package such as Revit to create Object-Oriented BIM model. Many companies in the industry has adopted this workflow for reconstruct as-built BIMs or updating an as-designed BIM.

The issue associated with above mentioned workflow is that some part of this process requires human operators to conduct time-consuming and reprtitive manual operations. One such bottleneck lies in the process of reconstructing as-built BIM from CAD model: usually, engineers need to manually insert BIM objects to the model based on the geometric information in the as-built CAD model. First, inserting large number of objects manually is time-consuming. Second, inserting these objects at the precise location, and modifying their parameters (e.g. height and width of a window) to make the inserted object's geometry fit well with the as-built data are all challenging tasks. It happens that in order to speed up the modeling process, engineers tend to make decisions such as "windows on the same floor are on the same elevation", "walls on first floor and the wall on the second floor are coplanar", to simply copy windows, or extrude walls during BIM modeling. These assumptions might influence the fidelity of the reconstructed as-built BIM, since wall may not be coplanar, windows might not be well aligned. Making these assumptions actually causes loss of detailed information captured in 3D imaging data. However, it is very difficult to require engineers to manually confirm the fidelity of each objects they model, that will make such modeling process even more time-consuming.

One approach to combine the power of engineer and computer is an automated-refinement approach. Using their assumptions, engineers insert object at their rough locations, with approximately correct sizes. Even though these first guesses about object locations and dimensions might not be very accurate, it is helpful for an algorithm to identify 3D data close to these inserted objects, and modify the location and dimensions of these objects according to the 3D data points.

I call this process as a rough-modeling and automatic refinement process. Using this approach, engineers just need to insert objects at rough locations, and computer will do the rest to make the rough BIM fidel to the as-built 3D data. Without losing fidelity of the model, this approach should be able to substantially improve the productivity of as-built BIM reconstruction process, and make 3D imaging data more useful for stakeholders.

Critical issues need to be addressed is: 1) which assumptions can accelarate the modeling process while still keep the model good enough for a refinement algrithm? 2) which refinements should be conducted by refinement algorithms and how? location refinement? dimension refinement? what else? parameteric models should be adopted in this approach, but some issues might need to be further investigated.