The History of the Walls Update of indoor IFC models using point cloud geometry

Building Information Modelling (BIM) is a methodology that has changed many paradigms in the
construction world in the last decades, bringing more efficiency, reliability of data, and solving design
issues early on the building process. In the BIM world, IFC is a schema that allows vendor neutral
collaboration with open standards, and is therefore the most popular file format for exchanging
building models among stakeholders. Despite the large and ever increasing implementation of BIM
into new projects, much of the current building stock is represented only by paper drawings or 2D
CAD files. Furthermore, for facility management processes, it is important to have up to date data
from buildings, that is also reliable. The most popular methodology to capture geometry data from a
building site is by performing a laser scan and generating a point cloud of it. In the recent years,
methods have been developed to create BIM models from existing buildings. The process is often
manual and laborious, but current research focuses on using machine learning models that segment
building elements out of the point cloud, label them, and construct building data in BIM models based
on that geometry, aiming to automate this process. A limited amount of research however is focused
on the update of existing BIM models, and even more scarce research in the update of IFC files,
which approaches a more commercial implementation. A research gap was identified, as the need of
a procedure that allows the automated update of IFC files based on point cloud data. This was done
taking into account the importance of real-time reliable data, and the increased demand to use
buildings for a longer period of time and repurpose them, or renovate them, for sustainability reasons.
The process was identified as starting with an outdated IFC file, LiDAR data is collected, and based
on pre-determined segmentation standards the building elements are segmented from the point cloud.
The segmented elements, and indeed the point cloud, need to have their location aligned to the
coordinate system of the IFC file and the building geometry in IFC, a process whose automation is
covered in other research of the literature, just as the segmentation and labelling. This research
focused on the automation of the last steps of the updating process – the geometry extraction from
building elements segmented from the point cloud, the comparison and matching of that geometry
to that of IFC elements present in the outdated file, and the update based on the comparison. The
procedure generates new elements based on point cloud geometry, can remove elements depending
on their absence in the as-is state of the building, and can update an element, such as updating the
position of a column if a match is found but there is a deviation in position. The demonstration of the
process of IFC update based on point cloud geometry is implemented for Manhattan-World walls,
ceilings and columns, and is implemented in python using IfcOpenShell to access the IFC schema
and OpenCascade to visualize point clouds, building models and the update. The building element
type whose update is most extensively discussed and implemented is the IfcWallStandardCase, for
which new walls generated take a pre-existing IFC wall as template for semantics, looking for the
best fit based on thickness and area of the building, and algorithms are developed to improve the
geometry based on the context (e.g. wall connections). A concept named dynamic thresholds is also
used, in projects where walls of many thicknesses are used, to create flexibility with a large threshold
to match thick walls and a smaller threshold to match thinner walls. Another concept named Room
Mode was created to allow performing a check and update with a scan that represents only a part of
the model. The code is made publicly available.