applied

sciences
Article
Generating the Regular Axis from Irregular Column Grids
through Genetic Algorithm
Xi Wang 1, * , Cong Wu 1, * and Chengjun Bai 2

1 School of Architecture, Tianjin University, No. 92 Weijin Road, Nankai District, Tianjin 30072, China
2 School of Civil Engineering, Tianjin University, No. 92 Weijin Road, Nankai District, Tianjin 30072, China;
[email protected]
* Correspondence: [email protected] (X.W.); [email protected] (C.W.)

Featured Application: This study proposes an auto-generation method of the regular axis, ap-
plied in digital documentation of cultural heritage. The procedure is repeatable, the results show
the displacement of columns with visual expression and numerical analysis.

Abstract: Historic building information modeling (HBIM) provides an index frame for digital docu-
mentation of the cultural heritage, as a continuous process of reverse engineering. The index frame
should be a regular model, with a clear comprehension of each component; consequently, associated
knowledge could be live-updated with the investigation progress. Therefore, the method of data
registration stresses its importance. The axis is fundamental for Chinese traditional architecture as
the basis for positioning all components in wooden structures. However, displacement often happens.
To correct the displacement while modeling, the hypothetical axis should be determined first. This
paper thus proposes a method of generating the regular axis from irregular column grids and aims to
develop an automatic solution that is repeatable and transplantable. The finite element modeling



(FEM) abstracts the actual problem to enable numerical calculation. Starting from a candidate solu-


tion, the genetic algorithm calculates a solution closest to the expectation from the possible solutions
Citation: Wang, X.; Wu, C.; Bai, C.
in several minutes. The standard deviation is used to measure the amount of displacement based on
Generating the Regular Axis from
the hypothetical axis, which is expected to be minimum. This method is compatible with most kinds
Irregular Column Grids through
Genetic Algorithm. Appl. Sci. 2022,
of input data, e.g., point cloud, excel data, 2D drawing, mesh model, etc., and applied to a World
12, 2109. https://doi.org/10.3390/ Heritage Site in Qufu (Shandong, China). The results show the displacement of columns with visual
app12042109 expression and numerical analysis and prove that the proposed method is repeatable and traceable
and can easily be applied to different projects by changing input parameters.
Academic Editors: Zhihan Lv, Kai Xu
and Zhigeng Pan
Keywords: HBIM; reverse engineering; grid system; column displacement; regularized rebuild; FEM;
Received: 29 December 2021 algorithm generation; heritage documentation
Accepted: 14 February 2022
Published: 17 February 2022

Publisher’s Note: MDPI stays neutral

with regard to jurisdictional claims in 1. Introduction
published maps and institutional affil- 1.1. Review of Related Works
iations. During the digital documentation of cultural heritage, displacements are usually
found in wooden structures [1–4], resulting in difficulty of positioning while modeling.
Different from modern architecture, in Chinese traditional architecture, the axis is
designed for locating column heads, rather than column feet. Because eave columns are
Copyright: © 2022 by the authors.
inclined and raised in an arithmetic sequence, from center to corner, the plane projection of
Licensee MDPI, Basel, Switzerland.
centroids connection line of eave column heads is rectangular, while the eave column feet
This article is an open access article
are not.
distributed under the terms and
conditions of the Creative Commons
This manufacturing method is documented in “Ying Zao Fa Shi” [5–9] (which states
Attribution (CC BY) license (https://
building regulations of Song Dynasty, A.D. 1103), and it can be seen in many heritage
creativecommons.org/licenses/by/
sites in East Asia nowadays. However, this method is difficult to verify, as all the wooden
4.0/). structures have lost their original positions.

Appl. Sci. 2022, 12, 2109. https://doi.org/10.3390/app12042109 https://www.mdpi.com/journal/applsci

Appl. Sci. 2022, 12, 2109 2 of 15

As a continuous process of reverse engineering, i.e., regularized rebuild or ideal

reconstruction [10–13], an HBIM model should be an index frame [14,15], a prototype
library [16–18], or an ideal model [19–21], of which components could be comprehended
with architectural semantics, and knowledge could be associated through multidisciplinary
collaboration. The traditional method of digital documentation in China accumulates
a large number of 2D drawings, yet mistakes occur while modeling with reference to
these data.
As a result, the data acquired from 2D drawings have to be registered, e.g., a hypo-
thetical orthogonal grid system. Each axis is calculated through a certain algorithm so
that the average distance between each “as-found” column centroid and “as-designed”
column centroid is expected to be the smallest. Once an ideal grid system is established,
the numerical analysis could be applied.
In recent years, the method of algorithm modeling (or generative modeling) with visual
programming tools is developing increasingly [22–31]. This technique provides traceable
solutions which document the whole modeling procedure, including input parameters,
process algorithms, and resulting models. Based on this foundation, the research that
considers grids as a whole of the interrelated system cloud be further developed.
To create a whole column grid system, finite element modeling (FEM) with the utiliza-
tion of genetic algorithm is chosen as the method, as the multi-goals (result axes) are known
quantities while the genomes (input parameters) are unknown quantities. The FEM trans-
forms the problem of infinite degrees of freedom in a continuous domain into the problem
of finite degrees of freedom in a discrete domain, enabling numerical calculation [32–36].
Furthermore, compiling algorithms for mathematical representation, the genetic algorithm
is available to implement for optimizing FEM, with a Grasshopper plug-in, the Galapa-
gos generic solvers [37]. The Rhinoceros thus is chosen as the software platform for the
automatic solution through the genetic algorithm with FEM. The Galapagos as an evo-
lutionary solver in Grasshopper has been widely used for parametric design and result
optimization [38–46].

1.2. Purpose and Significance

This paper aims to propose a method of automatically generating whole axes in
column grids at the same time with visible algorithm nodes and thousands of iterations,
which would be less possible by manual operating.
Starting with a pre-given candidate solution and point cloud, the Galapagos module
tries to move and rotate all of the axes with the expectation that the average distance
between each ideal axis intersection point and as-found column head centroid is the
smallest. When applying to different kinds of heritage sites, one only needs to change input
parameters by adjusting the numeric slider or removing non-participate columns, etc., and
then the procedure would finish the calculation within less than an hour.
The proposed method consists of several steps, as shown in the flowchart (Figure 1).
Appl. Sci. Sci.
Appl. 2022, 12, x
2022, 12,FOR
2109PEER REVIEW 3 of317
of 15

Figure
Figure 1. The
1. The Auto-generation
Auto-generation method.
method.

2. Materials
2. Materials andand Methods
Methods
2.1. Finite Element Modeling
2.1. Finite Element Modeling Using
Using Genetic
Genetic Algorithm
Algorithm
TakeTake
thethe Kuiwen
Kuiwen Pavilion
Pavilion as as
an an example.
example. It isItlocated
is located in the
in the Temple
Temple of Confucius,
of Confucius,
which has been part of the UNESCO World Heritage Site “Temple and Cemetery ofofCon-
which has been part of the UNESCO World Heritage Site “Temple and Cemetery Confu-
fucius and the Kong Family Mansion in Qufu” since 1994. The Kuiwen Pavilion was built in
cius and the Kong Family Mansion in Qufu” since 1994. The Kuiwen Pavilion was built
the Imperial style, which means that the structural members should ideally form column
in the Imperial style, which means that the structural members should ideally form col-
grids with rectangle spaces.
umn grids with rectangle spaces.
However, based on a plane section of column heads from the point cloud, an irregular
However, based on a plane section of column heads from the point cloud, an irregu-
grid system shows as-found column heads together with beams (Figure 2).
lar grid system shows as-found column heads together with beams (Figure 2).
Column heads serve as joints of beams, and together with beams, they form a spatial
framework, therefore it can be assumed that displacements are evenly distributed from
their original place, i.e., they meet Laplace-Gaussian distribution. To bring this problem to
numerical calculation, the column head could be simplified to one circle, which could be
deconstructed into two parameters: diameter and a centroid (point).
To define an axis, at least two centroids are required, yet this axis is certainly de-
termined by axiom. However, if there are three points or more, an axis is uncertainly
determined by least-squares fitting. Consequently, it can be assumed that grids are rectan-
gles, displacements of column heads are evenly distributed along each rebuild axis, and the
variance of distances between current centroids and rebuild centroids of column heads has
a minimum value. Therefore, Gauss sum instead of variance is expected to be minimum;
thus, it is used as the goal.
q r h i
σ= Var ( X ) = E ( X − µ)2
Appl.
Appl. Sci. 2022, 12,
Sci. 2022, 12, 2109
x FOR PEER REVIEW 4 of 15
17

Figure 2. A plane section of column heads from the point cloud shows an irregular grid system.
Figure 2. A plane section of column heads from the point cloud shows an irregular grid system.

Column
The headsofserve
variance as joints
a random of beams,
variable X is and together with
the expected valuebeams,
of the they formdeviation
squared a spatial
from the mean of X. That is, variance equals the average value of the Gauss sum. from
framework, therefore it can be assumed that displacements are evenly distributed The
their original
squared root ofplace, i.e., they meet
the variance is theLaplace-Gaussian
standard deviation, distribution. To bring
which is used as anthis problem
evaluating
to numerical
indicator calculation,
of the result. the column head could be simplified to one circle, which could
be deconstructed
Therefore, theinto two parameters:
problem is simplifieddiameter and acan
as such: how centroid (point).all axes at the same
one generate
To define an axis, at least two centroids
time while the minimum value of Gauss sum is reached? are required, yet this axis is certainly deter-
mined by axiom. However, if there
The methodology is shown as follows: are three points or more, an axis is uncertainly deter-
mined by least-squares fitting. Consequently, it can be assumed that grids are rectangles,
1. Use the Circle—FitPoints command to manually select related points from the point
displacements of column heads are evenly distributed along each rebuild axis, and the
cloud of a column head and then automatically draw a circle of the section profile.
variance
Repeatof distances
to go through between currentheads
all column centroids
(Figureand
3).rebuild centroids of column heads
2. Launch Grasshopper, use the Curve component of
has a minimum value. Therefore, Gauss sum instead to variance
collect alliscircles,
expectedusetoProject
be mini-
to
mum; thus, it is used as the goal.
obtain circles in the X-plane, use Area to obtain centroids of circles, use Deconstruct
to obtain X, Y, and Z values, 𝜎 and use𝑋 Sort List
𝑉𝑎𝑟 𝐸 𝑋to sort𝜇 centroids in a certain order
(Figure 4).
The Number
3. Use variance Slider
of a random variable
to manually giveXaisrange
the expected
for initialvalue of the squared
parameters includingdeviation
motions
fromand
the rotations
mean of of X. axes,
That then
is, variance equals the average value
connect them to the Genome of Galapagos of the Gauss sum. The
as a candidate
squared root for
solution of the genetic
variancealgorithm
is the standard
(Figure deviation,
5). which is used as an evaluating
indicator
4. of the
At this result. a candidate solution and resulting axes are shown; the average
moment,
Therefore,
distance the mm
is 144 problem
(Figureis simplified
6). as such: how can one generate all axes at the
sameThis
5. timesolution
while the minimum
could value of
be optimized as Gauss
long assum is reached?
a goal is given to the fitness of Galapagos,
The methodology
therefore Gauss sum is shown
is appliedas follows:
to calculate the applicability of axes, then the average
1. distance between current command
Use the Circle—FitPoints centroids and rebuild centroids
to manually is applied
select related pointsto evaluate
from the
the point
result
cloud (Figure 7). head and then automatically draw a circle of the section profile.
of a column
6. Open
Repeat to go throughstart
the Galapagos, the solver,
all column and(Figure
heads during3). the process, the result is displayed in
real-time (Figure 8).
2. Launch Grasshopper, use the Curve component to collect all circles, use Project to
7. When
obtain the result
circles seems
in the stableuse
X-plane, after
Area5 min of calculation,
to obtain centroids ofit could
circles,be
useoutput as the
Deconstruct
final
to obtain X, Y, and Z values, and use Sort List to sort centroids in a certain value
solution, if the average distance is acceptable. Or wait until the minimum order
is reached,
(Figure 4). though it may take longer (almost 1 h). The average distance is 68 mm
3. (Figure 9). Slider to manually give a range for initial parameters including motions
Use Number
and rotations of axes, then connect them to the Genome of Galapagos as a candidate
solution for the genetic algorithm (Figure 5).
 result (Figure 7).
6. Open the Galapagos, start the solver, and during the process, the result is displayed
in real-time (Figure 8).
7. When the result seems stable after 5 min of calculation, it could be output as the final
Appl. Sci. 2022, 12, 2109
solution, if the average distance is acceptable. Or wait until the minimum value is
5 of 15
reached, though it may take longer (almost 1 h). The average distance is 68 mm (Fig-
ure 9).

Appl. Sci. 2022, 12, x FOR PEER REVIEW 6 of 17

Figure 3. Circles are generated from select points (partial display).
Figure 3. Circles are generated from select points (partial display).

Figure 4. Sort centroids by coordinates.

Figure 4. Sort centroids by coordinates.

As is shown in Figure 9, displacements of column heads including vectors (arrows)

and lengths (numbers) are calculated, while all of the axes are generated. The solution
process takes 3 h 16 min and stops automatically after reaching the minimum value.
This result shows that the manually rotated point cloud is very close to the orthogonal
direction as the initial angle is 0.00002 degrees. The winding corridor has an average
width of 3493 mm, and the four dimensions (3485 mm, 3488 mm, 3498 mm, 3500 mm)
have a standard deviation of 6 mm. Therefore, the winding corridor could be regarded as
Appl. Sci. 2022, 12, 2109 6 of 15

equidistantly offset from the second circle of columns. Similarly, the whole grid system
could be regarded as symmetric.
The
Figure 4. components of the
Sort centroids by entire node-visible algorithm are shown in Figure 10.
coordinates.

Figure 5. Use Number Slider to give a candidate solution to the Genome (red wires) of Galapagos
Appl. Sci. 2022, 12, x FOR PEER REVIEW 7 of 17for
Figure 5. Use Number Slider to give a candidate solution to the Genome (red wires) of Galapagos
for genetic algorithm solution.
genetic algorithm solution.

Figure 6. A candidate solution of generating axes and numerical analysis of this moment.
Figure 6. A candidate solution of generating axes and numerical analysis of this moment.

2.2. Hypothesis and Verification

Ideally, if two tangent circles of 400 mm diameter arrayed in a vertical line are given,
one axis should be generated; at this moment, each distance should be zero, the angle of
the axis should be zero, and the aligned dimension between two horizontal axes should
be also 400 mm. After the solution process, the result successfully verifies the hypothesis,
although it takes 33 min to reach the goal (Figure 11).
Appl. Sci. 2022, 12, 2109 7 of 15

Figure 6. A candidate solution of generating axes and numerical analysis of this moment.

Appl. Sci. 2022, 12, x FOR PEER REVIEW 8 of 17

Figure 7. Connect Gauss sum to fitness (blue wire) for genetic algorithm solution.
Appl. Sci. 2022, 12, x FOR PEER REVIEW 8 of 17
Figure 7. Connect Gauss sum to fitness (blue wire) for genetic algorithm solution.

Figure 8. The beginning of the solution process and numerical analysis of this moment.
Figure 8. The beginning of the solution process and numerical analysis of this moment.
Figure 8. The beginning of the solution process and numerical analysis of this moment.

Figure 9. The result of the solution process and numerical analysis of this moment.

FigureAs
9. The result of
isresult
shown the solution process and numerical analysis ofincluding
this moment.
Figure 9. The ofintheFigure 9, displacements
solution process andof column
numerical heads
analysis of thisvectors
moment.(arrows)
and lengths (numbers) are calculated, while all of the axes are generated. The solution
As istakes
process shown
3 h in
16 Figure
min and 9, stops
displacements of column
automatically heads including
after reaching the minimumvectors (arrows)
value.
and lengths (numbers) are calculated, while all of the axes are generated. The
This result shows that the manually rotated point cloud is very close to the orthogo- solution
process takes 3 h 16 min and stops automatically after reaching the minimum
nal direction as the initial angle is 0.00002 degrees. The winding corridor has an average value.
This
width of result shows
3493 mm, that
and thethe manually
four rotated
dimensions point
(3485 mm, cloud
3488ismm,
very3498
closemm,
to the orthogo-
3500 mm)
nal direction as the initial angle is 0.00002 degrees. The winding corridor
have a standard deviation of 6 mm. Therefore, the winding corridor could be regarded has an average
as
width of 3493 offset
equidistantly mm, and fromthe
thefour dimensions
second (3485 mm,
circle of columns. 3488 mm,
Similarly, the3498
wholemm,
grid3500 mm)
system
have a standard deviation of 6 mm. Therefore, the winding corridor could be regarded as
Appl. Sci. 2022, 12, 2109 8 of 15
Appl. Sci. 2022, 12, x FOR PEER REVIEW 9 of 17

Appl. Sci. 2022, 12, x FOR PEER REVIEW 10 of 17

Figure10.
Figure 10.The
TheGrasshopper
Grasshoppercomponents
componentsof
offinite
finite element
element modeling
modeling and genetic algorithm.
algorithm.

2.2. Hypothesis and Verification

Ideally, if two tangent circles of 400 mm diameter arrayed in a vertical line are given,
one axis should be generated; at this moment, each distance should be zero, the angle of
the axis should be zero, and the aligned dimension between two horizontal axes should
be also 400 mm. After the solution process, the result successfully verifies the hypothesis,
although it takes 33 min to reach the goal (Figure 11).

Figure 11. Verification of two columns (the disordered axes on the left, the result axes on the right).
Figure 11. Verification of two columns (the disordered axes on the left, the result axes on the right).

If there are four circles, equally distributed along the vertical axis, the result should
restore the movement and rotation, and the average distance should be 200 mm. However,
the result shows a wrong solution (Figure 12). The average distance is 182 mm, and the
 Appl. Sci. 2022, 12, 2109 9 of 15

Figure 11. Verification of two columns (the disordered axes on the left, the result axes on the right).

If there are four circles, equally distributed along the vertical axis, the result should
restore the movement and rotation, and the average distance
distance should
should be
be 200
200 mm.
mm.However,
However,
the result shows a wrong solution (Figure 12). The average distance is 182 mm, and the
initial angle is not zero. This is because four circles could form a rhombus, for which it is
impossible to determine the only axis.

Figure 12. Verification of four columns (the disordered axes on the left, the result axes on the right).
Figure 12. Verification of four columns (the disordered axes on the left, the result axes on the right).

Therefore, the hypothesis

Therefore, the hypothesis is
is changed
changed toto eight
eight circles, to form
circles, to form aa symmetric
symmetric grid
grid system
system
in which the horizontal dimension is 1600 mm and each vertical dimension
in which the horizontal dimension is 1600 mm and each vertical dimension is 800 is mm,
800 mm,
then
then
thesethese circles
circles are distributed
are distributed equally
equally (200 mm)
(200 mm) alongalong each vertical
each vertical axis. axis.
The generated
generated axes
axes successfully
successfully restore
restore the
themovement
movementand androtation
rotation(Figure
(Figure13),13),alt-
al-
hough
thoughthethesolution
solutionprocess
processtakes
takes 54
54min
mintotoreach
reachthe
theminimum
minimumvalue valuethen
thenstops
stopsautomat-
automat-
ically. However, when this grid system is asymmetric, the resulting axes have a smaller
Appl. Sci. 2022, 12, x FOR PEER REVIEW 11 of 17
average distance (183 mm) than the initial pre-given axes (200 mm) and a smaller smaller standard
standard
deviation (188 mm) than the initial pre-given axes (200 mm), but they fail to restore the
original grids (Figure 14).

Figure 13. Verification of eight symmetric columns (the disordered axes on the left, the result axes on
Figure 13. Verification of eight symmetric columns (the disordered axes on the left, the result axes
the right).
on the right).
Appl. Sci. 2022, 12, 2109 10 of 15
Figure 13. Verification of eight symmetric columns (the disordered axes on the left, the result axes
on the right).

Figure 14. Verification of eight asymmetric columns (the disordered axes on the left, the result axes
Figure 14. Verification of eight asymmetric columns (the disordered axes on the left, the result axes
on the right).
on the right).
Furthermore, both asymmetric and symmetric grid systems of six circles are tested.
Furthermore,
The both
only difference asymmetric
is that and
the top two symmetric
circles grid systems
are removed. of sixaxes
Both resulting circles
have are
a tested.
Thesmaller
only difference is that(177
average distance themm)
topthan
twothecircles
initialare removed.
pre-given Bothmm
axes (290 resulting
and 202 axes
mm), have a
smaller average
and both distance
resulting (177 amm)
axes have than
smaller the initial
standard pre-given
deviation axesthan
(188 mm) (290the
mm andpre-
initial 202 mm),
and given
both axes (312 mm
resulting andhave
axes 203 mm).
a The solver
smaller has reached
standard the minimum
deviation (188 valuethan
mm) and stopped
the initial pre-
Appl. Sci. 2022, 12, x FOR PEER REVIEW 12 of 17
Appl. Sci. 2022, 12, x FOR PEER REVIEW
automatically (Figures 12 of 17
given axes (312 mm and 15203and 16). The solver has reached the minimum value
mm). and stopped
automatically (Figures 15 and 16).

Figure 15. Verification

Figure ofofsix
15. Verification sixasymmetric columns
asymmetric columns (the(the disordered
disordered axes onaxes on the
the left, theresult
left, axes
the result
on axes on
Figure 15. Verification of six asymmetric columns (the disordered axes on the left, the result axes on
the right).
the right).
the right).

Figure 16. Verification of six symmetric columns (the disordered axes on the left, the result axes on
Figure 16. Verification of six symmetric columns (the disordered axes on the left, the result axes on
Figurethe16.
right).
Verification of six symmetric columns (the disordered axes on the left, the result axes on
the right).
the right).
Additionally,
Additionally, to to verify
verify that
that an
an odd
odd number
number of of columns
columns would
would have
have an
an inaccurate
inaccurate
result,
result, the hypothesis is changed to 10 circles. The generated axes successfully restore
the hypothesis is changed to 10 circles. The generated axes successfully restore the
the
rotation
rotation but
but fail
fail to
to restore
restore the
the movement.
movement. The
The solver
solver has
has reached
reached the
the minimum
minimum value
value and
and
stopped
stopped automatically
automatically (Figures
(Figures 17
17 and
and 18).
18).
 Appl. Sci. 2022, 12, 2109 11 of 15

Appl. Sci. 2022, 12, x FOR PEER REVIEW 13 of 17

Additionally, to verify that an odd number of columns would have an inaccurate
Appl. Sci. 2022, 12, x FOR PEER REVIEW 13 of 17
result, the hypothesis is changed to 10 circles. The generated axes successfully restore the
rotation but fail to restore the movement. The solver has reached the minimum value and
stopped automatically (Figures 17 and 18).

Figure 17.17.
Figure Verification
Verificationof
of10
10 asymmetric columns(the
asymmetric columns (the disordered
disordered axesaxes on left,
on the the left, the result
the result axes on
axes on
Figure
the 17. Verification of 10 asymmetric columns (the disordered axes on the left, the result axes on
right).
the right).
the right).

Figure
Figure
Figure18.
18.Verification
18. Verificationof
Verification of10
of 10 symmetric
10 columns(the
symmetriccolumns
symmetric columns (the
(the disordered
disordered
disordered axes
axes
axes onononthe
the theleft,
left, left,
thethetheresult
resultresult axes
axesaxes
on onon
the
theright).
right).
the right).

This
This
This resultshows
result
result showsthat
shows that for
that linear-arrayed
linear-arrayedcircles
for linear-arrayed circlesofof
circles a avertical
of averticalaxis,
vertical each
axis,
axis, column
each
each column shares
column shares
shares
equal
equal weight,
weight, so
so the
the three
three
equal weight, so the three circles circles
circles have
have more
more “attraction” than two circles, and
more “attraction” than two circles, and theaxis
“attraction” than two circles, the
and axis
the axis
deviates to their side. The average distance
deviatestototheir
deviates theirside.
side. The
The average distanceisis
average distance is192
192mm,
192 mm,
mm, which
which
which is is
even
isevenmore
even more“ideal”
more thanthan
“ideal”
“ideal” than
theoriginal
the originaldistribution
distribution of of 200
200 mm,
mm, despite
despitethe axis not being the same as the original one.
the original distribution of 200 mm, despite the
theaxis
axisnot
notbeing
beingthethesamesameasasthetheoriginal
originalone.
one.
The standard deviation is 196 mm, which is closer to 200 mm, compared to 188 mm, of six
Thestandard
The standarddeviation
deviation is is 196
196 mm,
mm, which
which is
is closer
closer to
to200
200 mm,
mm, compared
compared toto188
188 mm,
mm, ofof
sixsix
columns. After the solution process is complete, the resulting axes of 10 columns are more
columns.After
columns. After thesolution
solution process is complete, the resulting axes of 10 columns are more
accurate than thesix columns. process is complete, the resulting axes of 10 columns are more
accurate than six columns.
accurate than six columns.

3.3.Results
Results
The results of the hypothesis and verification are concluded in Table 1.
The results of the hypothesis and verification are concluded in Table 1.
Appl. Sci. 2022, 12, 2109 12 of 15

3. Results
The results of the hypothesis and verification are concluded in Table 1.

Table 1. The statistics of pre-given axes and generated axes.

Column Amount 2 4 6 6 8 8 10 10
Single Axis or
Single Axis Rectangle Grids
Initial State of Rectangle Grids
Pre-given Axes Symmetry Not Applicable Yes No Yes No Yes No
Distribution Not
200 mm
Distance Applicable
Restore the
Yes No No No Yes No No No
Movement
Restore the
Yes No Yes Yes Yes No Yes Yes
Rotation
Result of
Generation Average Distance 0 mm 182 mm 177 mm 177 mm 200 mm 183 mm 192 mm 192 mm
Standard Deviation 0 mm 372 mm 188 mm 188 mm 200 mm 188 mm 196 mm 196 mm
Duration 33 min 43 min 54 min 55 min 54 min 55 min 55 min 54 min
Outputs Evaluation Right Wrong Applicable Right Applicable

This method is proposed for the correction of an irregular grid system; the result could
accurately restore the original axes under certain conditions.
1. At least two rows.
2. Each row has four symmetrically displaced columns or has more than four symmetri-
cally displaced columns as long as the number is even (6, 8, 10, etc.).
Otherwise, the result could not accurately restore the original axes, but is still close to
the original axis and still has potential possibilities for application, as it is repeatable and
more convenient than manual calculation. The more columns a grid system has, the more
accurate the result is.

4. Conclusions
This method could either output geometric models or numeric data. After the solution
process, the resulting axes could be directly used for modeling. Alternatively, in a collabo-
rative workflow including many sessions, to make data exchange convenient, only grid
spacing and rotation angle are required. Both of them are numbers, and therefore easy to
store in a general format such as .txt, .xlsx, etc. The grid spacing is used to create a grid
system in BIM software such as Autodesk Revit and Bentley Openbuildings Designer. The
rotation angle is used to rotate the point cloud to fit the grid system; thereafter, the point
cloud is used as a reference during the modeling process.
Although this method aims to automate the solution process, regarding the diversity
of architectures there are still manual operations. For example, removing some columns
based on the standard column grids is a common practice, such as Kuiwen Pavilion. Not
all axis intersections have columns; therefore, the corresponding column has to be removed
from the algorithm.
Furthermore, this method could be applied to test whether these columns are designed
to be inclined or not. First, the axes of column bases are generated. The centroid of rebuild
axes on column bases could be recorded as C0, and the centroid of rebuild axes on column
heads could be recorded as C1. The vector from C0 to C1 on the XY plane would indicate
the movement between the column heads grid system and the column bases grid system.
If the column heads grid system is moved to make the X and Y coordinates of C1 and C0
consistent, the result would indicate whether these columns are designed to be inclined or
not (Figures 19 and 20).
Appl. Sci.
Appl. Sci. 2022,
Sci. 2022, 12,
12, x2109
2022, 12, x FOR
FOR PEER
PEER REVIEW
REVIEW 13 of
15 of 15
of 17
17
Appl. 15

Figure 19. The

Figure The inclination and
and the average
average distance.
Figure 19.
19. The inclination
inclination and the
the average distance.
distance.

Figure 20.
Figure 20. The
20. The column
The column height.
column height.
height.
Figure
If they are designed to be inclined, the outer eaves columns are inclined inward in the
If they
they are
are designed
designed to to be
be inclined,
inclined, the
the outer eaves
eaves columns are are inclined
inclined inward
inward inin
frontIfand back eaves, with 10‰ of the columnouter height; 8‰columns
of the column height in the two
the
the front
front and
andandback
back eaves,
eaves, with
with 10‰
10‰ ofof the column height; 8‰ of the column height in the
sides eaves, the corner columns arethe columninheight;
inclined 8‰ of theatcolumn
both directions the same height
time.in the
two sides
two sides eaves,
eaves, and
and thethe corner
corner columns
columns are are inclined
inclined in in both
both directions
directions at at the
the same
same time.
time.
After the manual check, the result shows that the inclination of the columns almost
After the
After the manual
manual check,
check, thethe result
result shows
shows that
that the
the inclination
inclination of of the
the columns
columns almost
almost
conforms to a designed inclination, as the average distance is 50 mm, the column height is
conforms
conforms to a designed inclination, as the average distance is 50 mm, the column height
5027 mm, to and a the
designed
ratio isinclination,
almost 10‰. asHowever,
the average thisdistance is 50
result still needsmm,further
the column height
investigation
is 5027 mm,
is mm, andand the
the ratio
ratio isis almost
almost 10‰.
10‰. However,
However, this this result
result still
still needs further
further investi-
investi-
to 5027
confirm, as all of the columns have inclinations, rather than justneeds
eaves columns. This
gation
gation to
to confirm,
confirm, as
as all
all of
of the
the columns
columns have
have inclinations,
inclinations, rather
rather than
than
practice has not been found in historical archives or architectural books. The cause of this just
just eaves
eaves columns.
columns.
This practice
This
specialpractice has not
has
inclination not beenfurther
been
needs found in
found in historical
historical
research archives
to archives
verify or architectural
or
whetherarchitectural books. The
books.
this construction The cause of
cause
method of
this
this
columnspecial
special inclination
inclination
inclination needs further
needs further
is common research
research
in other to verify whether
to verifyarchitectures.
Imperial-style this construction
whether this construction method method
of column
of column inclination
inclination is is common
common in in other
other Imperial-style
Imperial-style architectures.
architectures.
Author Contributions: Conceptualization, X.W., C.W. and C.B.; Funding acquisition, C.W.; Investi-
AuthorX.W.
gation,
Author Contributions:
Contributions: Conceptualization,
and C.W.; Methodology,
Conceptualization, X.W.,
X.W., X.W., C.W.
C.W. and
C.W. andProject
C.B.;
and C.B.; Funding
C.B.; Funding acquisition,
administration, C.W.;
C.W.;
acquisition, Inves-
Resources,
C.W.; Inves-
tigation, X.W.
C.W.; Software,
tigation, and C.W.;
X.W.;C.W.;
X.W. and Methodology,
Validation, X.W., C.W.
X.W.; Visualization,
Methodology, and
X.W., C.W.X.W.; C.B.; Project
and Writing—originaladministration, C.W.;
draft preparation,
C.B.; Project administration, Re-
C.W.;X.W.;
Re-
sources,
sources, C.W.;
Writing—review Software, X.W.;
and editing,
C.W.; Software, Validation,
X.W.; X.W. X.W.;
and C.W.
Validation, Visualization,
AllVisualization,
X.W.; X.W.;
authors haveX.W.; Writing—original
read Writing—original draft
and agreed to thedraft prep-
published
prep-
aration,of
aration,
version X.W.;
X.W.; Writing—review and
the Writing—review
manuscript. and editing,
editing, X.W.
X.W. and
and C.W.
C.W. All
All authors
authors have
have read
read and
and agreed
agreed toto the
the
published version
published version of
of the
the manuscript.
manuscript.
Funding: This research was funded by Ministry of Science and Technology of the People’s Republic
Funding:
Funding:
of This research
This
China (projectresearch was funded
was funded by
No. 2020YFC1522400, by Ministry
Ministry of Science
of Science and
No. 2020YFC1522405). and Technology
Technology ofof the
the People’s
People’s Republic
Republic
of China (project NO. 2020YFC1522400, NO. 2020YFC1522405).
of China (project NO. 2020YFC1522400, NO. 2020YFC1522405).
Institutional Review Board Statement: Not applicable.
Institutional Review
Institutional Review Board
Board Statement:
Statement: Not
Not applicable.
applicable.
Informed Consent Statement: Not applicable.
Informed Consent
Informed Consent Statement:
Statement: Not
Not applicable.
applicable.
Data Availability Statement: The source code of the node-visible algorithm is not openly available
Data Availability Statement:
Data Statement: The
The source
source code of the node-visible algorithm is not openly available
due toAvailability
the consideration of data safety. code of the node-visible algorithm is not openly available
due to the consideration of data safety.
due to the consideration of data safety.
Appl. Sci. 2022, 12, 2109 14 of 15

Acknowledgments: This paper was partially supported by the following institutions: The Qufu Ad-
ministrative Bureau of Historical Relics, Qufu; The Qufu San Kong Heritage Architecture Engineering
Management Office, Qufu; The Palace Museum, Beijing; Jiayuguan Academy of Silk Road (the Great
Wall) Culture, Jiayuguan; Cultural Heritage Administration of Jizhou District, Tianjin.
Conflicts of Interest: The authors declare no conflict of interest.

References
1. Massafra, A.; Prati, D.; Predari, G.; Gulli, R. Wooden Truss Analysis, Preservation Strategies, and Digital Documentation through
Parametric 3D Modeling and HBIM Workflow. Sustainability 2020, 12, 4975. [CrossRef]
2. Ba, Z.; Jiao, P.; Zhang, L.; Liu, B.; Zhang, X. Damaged Deformation Status and Seismic Performance Analysis of Guanghua
Building in Jiayuguan. Technol. Earthq. Disaster Prev. 2020, 15, 293–304.
3. Yang, H.W.; An, D.W. Column-leaning Displacement of Korean Traditional Wooden Architecture: 3D Scanning Data of Columns
of Sungryeoljeon in Namhansanseong. J. Asian Archit. Build. Eng. 2017, 16, 279–286. [CrossRef]
4. Xu, M.G.; Ding, Z.W. Study of Ultimate Displacement of Timber Structures in Chinese Ancient Buildings. In Proceedings of the
4th International Conference on Civil Engineering, Architecture and Building Materials (CEABM), Haikou, China, 24–25 May
2014; pp. 1759–1762.
5. Qiao, X. On the Form of Gongxian and Liaoli in Yingzao Fashi. Stud. Hist. Nat. Sci. 2007, 26, 523–536.
6. Park, Z. A Design of the Bracketing Set of Japanease Zen style seen from the viewpoint of an Immanent Oblique Line. J. Archit.
Inst. Korea Plan. Des. 2002, 18, 101–108.
7. Chang, H. The Study of the Techniques of the Xia-ang Structure in China during the Tang and Song Dynasty. J. Archit. Hist. 2011,
20, 115–130.
8. Guo, Q. Yingzao Fashi: Twelfth-Century Chinese Building Manual. Archit. Hist. 1998, 41, 1–13. [CrossRef]
9. Wang, X.; Wu, C.; Que, R.; Bai, C. Algorithm Modeling of Ancient Architecture for Heritage Documentation. ISPRS—Int. Arch.
Photogramm. Remote Sens. Spat. Inf. Sci. 2019, XLII-2-W15, 1209–1213. [CrossRef]
10. Wang, X.; Wu, C.; Que, R. Regularized Rebuild Workflow of HBIM for Built Heritage Documentation. ISPRS Ann. Photogramm.
Remote Sens. Spat. Inf. Sci. 2021, VIII-M-1-2021, 193–200. [CrossRef]
11. Yang, J.; Wang, X.; Wu, C.; Bai, C. Regularized Reconstruction of Grid System for Traditional Chinese Timber Structure Building
in Hbim. ISPRS—Int. Arch. Photogramm. Remote Sens. Spat. Inf. Sci. 2019, XLII-2/W15, 1229–1233. [CrossRef]
12. Soto-Martin, O.; Fuentes-Porto, A.; Martin-Gutierrez, J. A Digital Reconstruction of a Historical Building and Virtual Reintegration
of Mural Paintings to Create an Interactive and Immersive Experience in Virtual Reality. Appl. Sci. 2020, 10, 597. [CrossRef]
13. Franco, P.A.C.; Rueda Márquez De La Plata, A.; Cruz Franco, J. From the Point Cloud to BIM Methodology for the Ideal
Reconstruction of a Lost Bastion of the Cáceres Wall. Appl. Sci. 2020, 10, 6609. [CrossRef]
14. Li, D.; Wang, X.; Bai, C.; Wu, C. Discussion on the Problem of Regularized Reconstruction in Hbim. ISPRS—Int. Arch. Photogramm.
Remote Sens. Spat. Inf. Sci. 2019, XLII-2/W15, 657–662. [CrossRef]
15. Wu, C.; Li, K.; Li, S.; Zhang, L. From digitization to informatization: The application of information technology in the field of
Architectural Heritage. China Cult. Herit. 2016, 2, 18–24.
16. Murphy, M.; McGovern, E.; Pavia, S. Historic Building Information Modelling—Adding intelligence to laser and image based
surveys of European classical architecture. ISPRS J. Photogramm. Remote Sens. 2013, 76, 89–102. [CrossRef]
17. Murphy, M.; McGovern, E.; Pavía, S. Parametric Vector Modelling of Laser and Image Surveys of 17th Century Classical
Architecture in Dublin. In Proceedings of the 8th International Symposium on Virtual Reality, Archaeology and Cultural Heritage
VAST, Brighton, UK, 26–30 November 2007.
18. Murphy, M.; McGovern, E.; Pavia, S. Historic building information modelling (HBIM). Struct. Surv. 2009, 27, 311–327. [CrossRef]
19. Martina, A. Informative Models for Architectural Heritage. Heritage 2019, 2, 2067–2089.
20. Capone, M.; Lanzara, E. SCAN-TO-BIM vs. 3D Ideal Model Hbim: Parametric Tools to Study Domes Geometry. ISPRS—Int. Arch.
Photogramm. Remote Sens. Spat. Inf. Sci. 2019, XLII-2/W9, 219–226. [CrossRef]
21. Han, S.; Wu, C.; Li, D.; Li, J.; Liu, Y.; Feng, K.; Di, Y. Workflows for Condition Inspection Documentation Ofarchitectural Heritage
Based on HBIM: Taking Three Duty Rooms Inforbidden City of Beijing as an Example. ISPRS—Int. Ann. Photogramm. Remote
Sens. Spat. Inf. Sci. 2017, 4, 123. [CrossRef]
22. Collao, J.; Lozano-Galant, F.; Lozano-Galant, J.A.; Turmo, J. BIM Visual Programming Tools Applications in Infrastructure Projects:
A State-of-the-Art Review. Appl. Sci. 2021, 11, 8343. [CrossRef]
23. Gigliarelli, E.; Calcerano, F.; Cessari, L. Heritage Bim, Numerical Simulation and Decision Support Systems: An Integrated
Approach for Historical Buildings Retrofit. Energy Procedia 2017, 133, 135–144. [CrossRef]
24. Diara, F.; Rinaudo, F. Building Archaeology Documentation and Analysis through Open Source HBIM Solutions via Nurbs
Modelling. ISPRS—Int. Arch. Photogramm. Remote Sens. Spat. Inf. Sci. 2020, 43, 1381–1388. [CrossRef]
25. Diara, F.; Rinaudo, F. From Reality to Parametric Models of Cultural Heritage Assets for HBIM. ISPRS—Int. Arch. Photogramm.
Remote Sens. Spat. Inf. Sci. 2019, XLII-2/W15, 413–419. [CrossRef]
26. Bagnolo, V.; Argiolas, R.; Cuccu, A. HBIM for Archaeological Sites: From SFM Based Survey to Algorithmic Modeling. ISPRS—Int.
Arch. Photogramm. Remote Sens. Spat. Inf. Sci. 2019, XLII-2/W9, 57–63. [CrossRef]
Appl. Sci. 2022, 12, 2109 15 of 15

27. Bagnolo, V.; Argiolas, R.; Cuccu, A. Digital Survey and Algorithmic Modeling in HBIM. Towards a Library of Complex
Construction Elements. ISPRS—Int. Arch. Photogramm. Remote Sens. Spat. Inf. Sci. 2019, XLII-4/W12, 25–31. [CrossRef]
28. Argiolas, R.; Cazzani, A.; Reccia, E.; Bagnolo, V. From Lidar Data towards HBIM for Structural Evaluation. ISPRS—Int. Arch.
Photogramm. Remote Sens. Spat. Inf. Sci. 2019, XLII-2/W15, 125–132. [CrossRef]
29. Cubukcuoglu, C.; Ekici, B.; Tasgetiren, M.F.; Sariyildiz, S. Optimus: Self-Adaptive Differential Evolution with Ensemble of
Mutation Strategies for Grasshopper Algorithmic Modeling. Algorithms 2019, 12, 141. [CrossRef]
30. Chen, X.; Wang, X. Parameterization of Yingzao Fashi: Generating the Palace Style Structural Carpentry System Based on
Algorithmic Language. Huazhong Archit. 2016, 34, 41–43.
31. Brumana, R.; Della Torre, S.; Previtali, M.; Barazzetti, L.; Cantini, L.; Oreni, D.; Banfi, F. Generative HBIM modelling to embody
complexity (LOD, LOG, LOA, LOI): Surveying, preservation, site intervention—The Basilica di Collemaggio (L’Aquila). Appl.
Geomat. 2018, 10, 545–567. [CrossRef]
32. Wang, J.; He, J.X.; Yang, Q.S. Finite Element Analysis on Mechanical Behavior of Column Foot Joint in Traditional Chinese
Timber Structures. In Proceedings of the 14th International Symposium on Structural Engineering (ISSE-14), Beijing, China,
12–15 October 2016; pp. 1396–1402.
33. Abbate, E.; Invernizzi, S.; Spanò, A. HBIM parametric modelling from clouds to perform structural analyses based on finite
elements: A case study on a parabolic concrete vault. Appl. Geomat. 2020. [CrossRef]
34. Pepe, M.; Costantino, D.; Garofalo, A.R. An Efficient Pipeline to Obtain 3D Model for HBIM and Structural Analysis Purposes
from 3D Point Clouds. Appl. Sci. 2020, 10, 1235. [CrossRef]
35. Brumana, R.; Banfi, F.; Cantini, L.; Previtali, M.; Torre, S.D. HBIM Level of Detail-Geometry-Accuracy and Survey Analysis for
Architectural Preservation. ISPRS—Int. Arch. Photogramm. Remote Sens. Spat. Inf. Sci. 2019, XLII-2/W11, 293–299. [CrossRef]
36. Angjeliu, G.; Cardani, G.; Coronelli, D. Digital Modelling and Analysis of Masonry Vaults. ISPRS—Int. Arch. Photogramm. Remote
Sens. Spat. Inf. Sci. 2019, XLII-2-W11, 83–89. [CrossRef]
37. Rutten, D. Galapagos: On the Logic and Limitations of Generic Solvers. Archit. Des. 2013, 83, 132–135. [CrossRef]
38. Vighnesh, R.; Anand, K.B.; Aadhish, B.K.; Divakar, S.; Kumar, C.M. Use of parametric tools in material system selection to
minimize building cooling load. In Proceedings of the 2nd International Conference on Aspects of Materials Science and
Engineering (ICAMSE), Chandigarh, India, 5–6 March 2021; pp. 4451–4457.
39. González, J.; Fiorito, F. Daylight Design of Office Buildings: Optimisation of External Solar Shadings by Using Combined
Simulation Methods. Buildings 2015, 5, 560–580. [CrossRef]
40. Nadiri, P.; Mahdavinejad, M.; Pilechiha, P. Optimization of Building Facade to Control Daylight Excessiveness and View to
Outside. J. Appl. Eng. Sci. 2019, 9, 161–168. [CrossRef]
41. Girotti, C.; Marins, K.R.d.C.; Lara, A.H. Análise da morfologia urbana para maximização de geração de energia fotovoltaica no
Belenzinho, em São Paulo. Ambiente Construído 2019, 19, 7–22. [CrossRef]
42. Li, J.; Huang, Y.; Zhang, L.; Zhang, M. Climate Adaptability Optimization of Large Space Building form in Cold Area: Take
Liaodong Bay Cruise Terminal Passenger Transport Center as an Example. Ind. Constr. 2021, 51, 6–11.
43. Vlaun, N.; van Waart, A.; Tenpierik, M.; Turrin, M. A Sound Working Environment: Optimizing the Acoustic Properties of Open
Plan Workspaces Using Parametric Models. In Proceedings of the 7th Annual Symposium on Simulation for Architecture and
Urban Design (SimAUD), London, UK, 16–18 May 2016; pp. 239–246.
44. Manni, M.; Lobaccaro, G.; Lolli, N.; Bohne, R.A. Parametric Design to Maximize Solar Irradiation and Minimize the Embodied
GHG Emissions for a ZEB in Nordic and Mediterranean Climate Zones. Energies 2020, 13, 4981. [CrossRef]
45. Lobaccaro, G.; Wiberg, A.H.; Ceci, G.; Manni, M.; Lolli, N.; Berardi, U. Parametric design to minimize the embodied GHG
emissions in a ZEB. Energy Build. 2018, 167, 106–123. [CrossRef]
46. Lee, S.; Lee, K.S. Optimization of Apartment-Complex Layout Planning for Daylight Accessibility in a High-Density City with a
Temperate Climate. Energies 2020, 13, 4172. [CrossRef]