Digital Twin supports precise renovation planning


The renovation of existing buildings remains a difficult task for the construction industry, as the work of planners is often hampered by missing, incomplete or outdated building plans. In recent years, digital twins capable of describing both the current state and the future stages of a building have become a reliable source of information during the planning phase. This article describes the difficult workflow of deriving an initial digital twin for a health clinic comprising approximately 100,000 m² in support of subsequent building changes.

This digital twin project was carried out in a clinic on a medical university campus in Germany. The clinic, which has been in operation since 1987, is now due to undergo a major basic renovation of the eight-story building. The renovation project is expected to last at least 17 years, including the planning phase, and the total budget amounts to 590 million euros. In order to avoid cost overruns, meet the construction schedule and maintain quality assurance, Building Information Modeling (BIM) was chosen as one of the planning strategies.

The challenges of this project are manifold, ranging from the sheer size of the building to the high precision requirements due to the complex nature of the technical equipment inside the building – especially in the 17 operating theatres. The scope of this project is to provide data to be used in the planning phase in the field of BIM during the complete renovation process. In terms of geometric quality and level of accuracy, the client requires LOA30, ie an accuracy of 5 mm at short distance and 15 mm throughout the building.

Figure 1: Planes detected in a single scan that were used for registration. The colors indicate the orientation of the planes in the local coordinate system of a scan.

Data acquisition

The basis for this project was established by a total station network consisting of 103 well-distributed ground control points which were determined in a higher coordinate system. Data was recorded using a GeoMax Zoom90 total station with an angular accuracy of one second, while observations were optimized in a geodetic network adjustment. Besides determining the coordinates of the points, the accuracies of the individual points were calculated on the basis of the propagation of the errors which serve as weights in the following processing steps. The network coordinates had an overall accuracy better than 3 mm.

Due to high precision requirements, static laser scanning was chosen for 3D data acquisition instead of kinematic laser scanning systems. To ensure fast data acquisition, four identical static laser scanners were used, namely the Zoller + Fröhlich IMAGER 5010 X series. Deviation from individual calibrations is a critical issue when using multiple systems in a project . Therefore, all systems were regularly checked using a plan-based calibration routine (Rietdorf et al. 2004) that assesses the quality of the scanner axis configuration.

The initial state of the building was captured before any renovation measures were undertaken. This involved six weeks of digitization and establishment of the total station network. In close coordination with the clinic management, the scans were captured mainly during the day by four survey teams in parallel. A total of 12,932 laser scans were captured, including a raw data volume of 540 GB (*.zfs format). A major challenge in projects of this size is how to organize all the scans. In this case, a naming convention was applied based on the 14 architectural entities of the building.

Data security

According to German law, as in many other countries, healthcare facilities fall under the category of critical infrastructure. Thus, to avoid the risk of lawsuits, data security was a key factor in this project; cloud-based or external data storage and processing solutions were strictly prohibited. Instead, all calculations had to be done on local computers and data was stored on a protected internal server.

Saving point clouds

Independent software Scantra 3.0 was used to save the approximately 13,000 scans. Since Zoller + Fröhlich grants third-party developers direct access to its proprietary *.zfs file format, there was no need to convert the original scans before processing. This saved considerable time and disk space. The chosen software follows a sequential data substitution process that supports projects of this size (Wujanz et al. 2018). In general, the following procedure was deployed:

  1. Import of static laser scan data including inclinometer readings and ground control points
  2. Detection of aircraft and local targets
  3. Pairwise registration between scan pairs and point matching
  4. Network adjustment/quality assurance
  5. Exporting saved point clouds

In this project, pairwise records were calculated based on identical planes between scan pairs. Figure 1 illustrates planes detected in a single scan that replace the original point cloud and therefore significantly reduce memory requirements. In addition to the design parameters, each design receives covariances which serve as weights when registering in pairs.

Figure 2: Bird’s eye view of the network graph, with circles representing single scans and arrows representing paired records.

Network optimization

In practice, a common method is to divide a project into individual groups linked together by ground control points. A notable drawback of this “Frankensteinian” course of action is that transition zones are unavoidable between individual groups. In order to avoid this, all scans have been optimized in a common adjustment. Key to this hugely ambitious task was the aforementioned data substitution strategy.

The data volume of the entire project has been reduced from 540 GB of raw data volume to 4.6 GB of geometric information and 10.4 GB of visual information in the form of images for each scan. Only geometric information is needed for block adjustment, which minimizes discrepancies between pairwise recordings, inclinometer readings and ground control points. Optimizing the final network with 24,706 pairwise records and 408 point identities at ground checkpoints took just over three minutes on a regular Dell laptop with 64 GB of RAM and 2.3 GHz with eight processors. . A maximum of 1.4 GB of RAM was to be provided by the computer. Figure 2 shows a bird’s eye view of the network graph.

Quality assurance and deliverables

A typical measure to check the quality of a network in the 3D mapping community is to provide residuals (or statistics thereof) to ground control points. Since these measures can be easily manipulated in its favor (Wujanz 2022), individual geodetic quality measures have been provided for each scan based on error propagation. These measurements are also called “positioning error” and quantify how statistically “shaky” a single sweep is relative to ground control points. Note that these metrics vary greatly depending on the given survey setup. Figure 3 illustrates these variations in a complex exterior portion of the project that consists of 338 scans. It can be clearly seen that the vast majority of stations are positioned in the 2.5 to 4.9 mm (1σ) range, but this drops significantly in some parts. This can be explained by the facade, which has mainly glass elements and was heavily vegetated in some parts with very limited views for the tacheometry.

Figure 3: Variation of individual positioning errors in an exterior part of the project. Triangles indicate ground control points, while arrows represent pairwise recordings between scans and circles between individual scans. The color filling of the circles highlights their individual positioning error.

Identifying errors is a time-consuming process in laser scanning. Therefore, a procedure was applied to identify outliers based on the result of the network adjustments. This produced a shortlist of statistically suspicious sightings. As a result, only a few sightings had to be inspected rather than all of them, saving considerable time. Recording this massive dataset took a total of three weeks. After final quality assurance was completed, the saved scans were voxelized and filtered by range to minimize data volume for the client. The resulting *.E57 files occupied 2.3 TB of memory. In addition to the point cloud itself, the customer also received a network overview that helps identify the analyzes required in the planning process.


This article describes a 3D mapping project of a large area clinic that was captured by static laser scanning. The resulting digital twin will serve as the basis for BIM-based planning during the 17-year renovation period and had to meet LOA30. Challenges for this project included the building’s size and complex architecture, which resulted in an extensive static laser scanning array. As a result, a huge dataset consisting of nearly 13,000 scans had to be processed through plan-based registration and network adjustment. After saving, the project point cloud was exported and submitted to the client as *.E57 files.

Further reading

Rietdorf, A., Gielsdorf, F. and Gruendig, L. (2004). A concept for the calibration of terrestrial laser scanners. In Proceedings of INGEO 2004 and the Central and Eastern Europe Regional Conference of FIG Surveying Engineering. Bratislava, Slovakia (Vol. 11, p.13).

Wujanz, D., Schaller, S., Gielsdorf, F. and Gründig, L. (2018). Planar recording of several thousand laser scans on standard hardware. International Archive of Photogrammetry, Remote Sensing and Spatial Information Science, 42(2).

Wujanz, D. (2022). Taming mistakes… pt. 10: What residues at geodetic control points DO NOT TELL YOU…. (Accessed: 05.09.2022)


Comments are closed.