You are on page 1of 25

See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/373238552

Buildability modeling of 3D-printed concrete including printing deviation: A


stochastic analysis

Article in Construction and Building Materials · August 2023

CITATIONS READS

0 98

3 authors, including:

Jinggao Zhu Miguel Cervera


Tongji University Universitat Politècnica de Catalunya
7 PUBLICATIONS 13 CITATIONS 421 PUBLICATIONS 9,645 CITATIONS

SEE PROFILE SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Computational Framework for Additive Manufacturing of Titanium Alloy Components (ADaMANT) View project

Strain localization in softening inelastic solids View project

All content following this page was uploaded by Jinggao Zhu on 20 August 2023.

The user has requested enhancement of the downloaded file.


1 Buildability modeling of 3D-printed concrete including printing deviation:
2 A stochastic analysis

3 Jinggao Zhua,b , Xiaodan Rena,∗, Miguel Cerverab,∗


4
a School of Civil Engineering, Tongji University, Shanghai, China

5
b International Centre for Numerical Methods in Engineering, Technical University of Catalonia, Barcelona, Spain

6 Abstract
Owing to the formwork-free construction, 3D concrete printing shows great prospects in construction industry. How-
ever, this manner also results in non-verified geometric precision, which is prone to generate uncontrolled printing
deviations between the actual and designed print paths. To include this in buildability modeling, a process-dependent
stochastic method is proposed based on the characteristics of the 3D printing process. Then, a stochastic analysis pro-
cedure is developed by Monte Carlo simulation and verified by the 3D-printed concrete (3DPC) cylinder test. Finally,
the verified method is applied to the stochastic analysis of 3DPC buildability modeling. By studying two structural
shapes and four correlation times, a clear understanding of the influence of printing deviation as well as guidances for
the design of 3D printing process are obtained.
7 Keywords: 3D–printed concrete, buildability, peridynamics, printing deviation, stochastic analysis

8 1. Introduction
9 With the development of additive manufacturing technology, 3D concrete printing comes into the construction
10 industry and attracts wide attention. In 3D concrete printing, the printable cementitious material is extruded from the
11 nozzle and printed layer-by-layer, fabricating the structures directly. As it no longer relies on formworks, it brings a
12 revolution to construction methods and allows more freedom in construction [1, 2, 3]. This freedom not only greatly
13 improves the productivity, but also enriches structural forms. In addition, 3D concrete printing is performed in an
14 automatic way by computer control, which is more intelligent and creative [4, 5]. With the above advantages, 3D
15 concrete printing is applied to more and more fields and faces great prospects in the construction industry.
16 However, the novelty of the technology also poses new requirements for 3D-printed concrete (3DPC) [6], includ-
17 ing flowability [7], extrudability [8] and buildability [9]. Compared to the former two capabilities, buildability is
18 more difficult to deal with as it is codetermined by two time-dependent processes. With the printing of new layers, the
19 upper gravity load on the freshly deposited lower 3DPC is continuously increasing. Without the formwork support,
20 this is dangerous although the printed concrete is gradually strengthening. If the strength of the lower 3DPC fails to
21 develop as required to resist the upper loading, the structure loses buildability before fully printed [10, 11]. Thus,
22 the buildability prediction of 3DPC structures plays an important role in the printing process design, as studied in the
23 present paper.
24 Various approaches have been proposed for the buildability analysis of 3DPC, including analytical models and
25 numerical methods. The analytical models take the main factors of the two time-dependent processes into consider-
26 ation. The rheology theory regards rheological properties of the fresh concrete as the main factors, as investigated
27 in references [12, 13]. Based on this, the lower bound analytical model [14, 15] and the empirical model [16] were
28 developed. The parameter model [17, 18] reduces the governing factors to 5 dimensionless failure parameters, namely
29 3 parameters defining elastic buckling and 2 parameters characterizing plastic collapse, which are used for the pre-
30 diction of the failure modes of wall structures. Rather than considering major factors, the numerical methods aim to

∗ Correspondingauthors.
Email address: rxdtj@tongji.edu.cn (Xiaodan Ren), miguel.cervera@upc.edu (Miguel Cervera)

Preprint submitted to Construction and Building Materials July 7, 2023


1 reproduce the 3D concrete printing process, which provides a more general way for buildability analysis. Studies have
2 been made from solid and fluid perspectives. The solid-like behavior of 3DPC was described by the Drucker-Prager
3 and Mohr-Coulomb models [10, 19] and simulated by the Finite Element (FE) method. Regarding the fluid perspec-
4 tive, Computational Fluid Dynamics (CFD) with generalized Newtonian and elastic-viscous-plastic fluid [20, 21] was
5 used for buildability modeling.
6 Actually, the solid- and fluid-like behaviors of 3DPC can only represent two limit states at the initial and final
7 printing stages and 3DPC experiences a fluid-to-solid transition process during 3D printing [22]. Thus, both behaviors
8 should be taken into consideration and the computational methods provide possibilities for this. Peridynamics (PD)
9 is a newly developed computational method reformulated from classical Continuum Mechanics [23]. In PD, the
10 mechanical behavior is modeled by bond interactions, which offers a flexible framework to treat both the fluid- and
11 solid-like behaviors. In addition, PD adopts an integral form of the governing equations and a meshfree discretization
12 scheme [24, 25]. This allows the modeling of large deformation and discontinuities, which enhances the capability to
13 capture the failure of 3DPC. Therefore, PD provides a well-suited numerical scheme for 3DPC buildability analysis.
14 The formwork-free construction results in non-verified geometric precision [26]. Without the restriction of form-
15 work, the printing system error is uncontrolled, and it is prone to generate printing deviations between the actual and
16 designed print paths. Although there is a growing awareness of the positioning errors [27] and some attempts on posi-
17 tioning accuracy optimization have been made [28], fully accurate 3D concrete printing is unrealistic and the printing
18 deviation is unavoidable [29]. Compared to the printing of metals and polymers, the precision of 3D concrete printing
19 is much lower and the influence of printing deviation is more obvious. As shown in reference [10], the deformation
20 and failure results of a 3DPC cylinder are different every time it is printed. Gambhir [30] pointed out that the critical
21 buckling load of a 3DPC shell could be several times lower due to the printing error. Also in the buckling prediction of
22 3DPC walls, the results are sensitive to the printing deviation [31], becoming more pronounced with concrete age and
23 structural dimensions. Therefore, in addition to the two time-dependent processes, the printing deviation also plays
24 an important role in 3DPC buildability, which should be incorporated into the buildability analysis.
25 In previous studies, most researchers focus on other types of deviations, such as the deviations caused by material
26 deformation of fresh 3DPC [32] and varying nozzle height [33], but neglect the position deviation generated by the
27 printing system. This kind of deviation mainly comes from the error of robot movements and can not be predicted
28 beforehand, which pose challenges for its consideration. Suiker and co-workers [34, 18] described it by periodically
29 shifting printed layers over a prescribed distance, the profile of which is idealized by a sinusoidal function and is char-
30 acterized by two lengthscale parameters, including the dimensionless amplitude and the dimensionless wavenumber.
31 This is introduced into the parameter model formulation and its influence on the buckling response is also investi-
32 gated. In the FE simulation [17, 31], the printing deviations are applied by the first buckling mode computed from a
33 bifurcation analysis and used for buckling simulation of 3DPC wall structures. Although the characteristic of position
34 movement is captured in the above attempts, the uncertainty feature is not considered.
35 Owing to the uncertainty feature of printing deviation, deterministic studies do not meet the requirements and
36 stochastic analysis should be conducted. The first attempt of stochastic analysis was performed by Chang et al. in
37 references [35, 36]. They used the stochastic method to account for the printing deviation and introduced it into
38 the discrete lattice model by randomly placing the lattice nodes within sub-cells through a pseudorandom number
39 generator. However, the consideration is somewhat simple and is independent of the printing process, the effectiveness
40 of which still lacks verification. In addition, although the printing deviation is described by the stochastic method,
41 the buildability modeling of 3DPC is still a deterministic simulation, and only one sample is considered. This not
42 only fails to cover all possible cases and leads to unconservative buildability results; but also fails to obtain the
43 statistical properties of the 3DPC system, which results in an unclear insight of the effect of printing deviation on
44 3DPC buildability.
45 Thus, a more accurate description for printing deviation and the corresponding stochastic analysis are required.
46 Considering the advantages of PD in buildability analysis, these are conducted in PD scheme in the present paper.
47 To describe this unpredictable position deviation, a process-dependent stochastic method is proposed, in which the
48 characteristics of the 3D printing process are considered. The stochastic processes are generated by the stochastic
49 harmonic method and mapped to the designed 3DPC structure to include the printing deviation. On this basis, the
50 Monte Carlo method is introduced and a stochastic analysis procedure is developed. As both the experimental ranges
51 of deformation curve and statistical properties of failure results are captured in the simulation of 3DPC cylinder
52 test, the effectiveness of the proposed method is verified. The method is finally applied to the stochastic analysis of
2
1 3DPC buildability modeling. Two different shapes of structures are considered and four different correlation times are
2 investigated, to gain a clear understanding of the influence of printing deviation on 3DPC buildability and guidances
3 for the 3D printing process design.

4 2. Peridynamic theory

5 Peridynamics (PD) was firstly proposed by Silling [23] in 2000. In PD, the computational domain is discretized
6 into particles and the particles are connected by micro bonds. Only the particles in the same family H x characterized
7 by the horizon size δ, are neighbor particles and have bond connections. Taking the 2D case in Figure 1 as an example,
8 the family H x is a circular region in the reference configuration. Using the force density function f (xi , x) to describe
9 the bond interaction and summing all the bond interactions over the family leads to the governing equation of PD:
ˆ
ρ(x)ü(x, t) = f (xi , x)dVq + b(x, t) (1)
Hx

10 where ρ(x) is the density field, ü(x, t) is the acceleration field, and b(x, t) is the body force density field. As Eq.(1) is
11 formulated in an integral form based on particles, PD shows great advantages in modeling discontinuities and large
12 deformation, thus enhances the capabilities to capture the failure of 3D-printed concrete (3DPC) structures.

Figure 1: Schematics of Peridynamics

13 In Eq.(1), the major issue is the definition of the force density function f (xi , x). Different definitions of f (xi , x)
14 describe different mechanical behaviors, leading to the proposal of different PD models [37]. This forms a flexible
15 framework to treat both the fluid- and solid-like behaviors of 3DPC. The recently proposed fluid-solid integrated
16 PD model [38] is formulated based on this flexibility and shows good performance in capturing the fluid-to-solid
17 transition behavior of 3DPC. The model proposes a constitutive relation to describe the mechanical behavior of 3DPC
18 and incorporates the relationship into PD by bond-based correspondence.
19 In the fluid-solid integrated PD model [38], the fluid-like behavior of 3DPC at the initial printing stage is described
20 by the liquid bridge model. The sphere/sphere type liquid bridge model [39] is expressed as a non-negative force-
21 distance (F-H) function:
2πRγ cos θ
F= + 2πHγ sin α sin(θ + α) (2)
1 + H/2d sp/sp
22 As shown in Figure 2(a), R is the radius of particles; γ is the surface tension material property; θ is the contact angle
23 of the liquid-gas interface that is regarded as constant for simplicity. The immersion distance d sp/sp and “embracing
3
1 angle” α are related to the separation distance:
 r  v
t  r 
H 2V H 2V
= −1 + 1 +  , α = −1 + 1 +
   
d sp/sp (3)
πRH πRH 

2 2 R 2

2 where V is the volume of the liquid bridge. By converting the liquid bridge force F to stress and the separation
3 distance H to strain, the liquid bridge model can be written as a stress-strain relationship.

Figure 2: The schematics of fluid-solid integrated PD model (a) liquid bridge model (b) bi-scalar plasticity-damage model (c) affinity hydration
model

4 Regarding the solid-like behavior of 3DPC at the final printing stage, the bi-scalar concrete damage model [40, 41]
5 is applied. It introduces two damage variables d± to describe the tensile and compressive nonlinear behavior of
6 concrete separately [42]:
σ̂ = (1 − d+ )σ̄+ or σ̂ = (1 − d− )σ̄− (4)
7 where σ̂ is the stress and σ̄± are the effective stresses related to the plastic strains εp± , where the superscripts + and −
8 represent the variables under tensile and compressive states, respectively:

σ̄+ = E0 ε − εp+ or σ̄− = E0 ε − εp−


 
(5)

9 and E0 is the elastic modulus. As shown in Figure 2(b), with different evolutions of tensile and compressive damage
10 and plastic strains, different tensile and compressive strain-stress relationships of solid-like concrete can be captured.
11 The above two models are combined through the concrete hydration model to use the evolution of hydration for
12 fluid-to-solid transition modeling. In the affinity hydration model [43], the hydration reaction is driven by the chemical
13 affinity and the evolution function without temperature effect is given as [44, 45]:

ζ
! !
B2
ζ̇ = B1 + ζ (ζ∞ − ζ) exp −β (6)
ζ∞ ζ∞

14 where B1 , B2 , and β are material parameters, ζ∞ is the ultimate hydration degree that depends on the water/cement
15 ratio (w/c) of the mixture. As illustrated in Figure 2(c), the fluid-to-solid transition of 3DPC can be naturally driven
4
1 by the hydration degree ζ defined in Eq.(6) [38]:

Concrete = (1 − ζ) × Fluid + ζ × Solid (7)

2 Finally, the developed constitutive relation is incorporated into PD by bond-based correspondence [46], which
3 provides a direct way for strain calculation and stress mapping. The strain is calculated in bond-based PD as:

η+ξ − ξ
s= (8)
∥ξ∥
4 As shown in Figure 1, ξ and η + ξ are the undeformed (red arrow) and deformed bonds (blue arrow), respectively;
5 η = u(xi ) − u(x) is the relative displacement (green arrow). The stress mapping converts the bond stress to the force
6 density function:
f (η, ξ) = kσ̂n (9)
7 where k is the mapping function and n is the deformed bond direction given by:
η+ξ
n= (10)
η+ξ

8 The mapping function is defined by the strain energy density equivalence between PD and Continuum Mechanics. If
9 isotropic deformation is assumed, the mapping functions in different dimensions are as follows:




2/Aδ2 for 1D
 3
9/πBδ for 2D plane stress


k=

3
(11)



48/5πBδ for 2D plane strain

12/πδ4

 for 3D

10 where A is the cross-section area of the one-dimensional domain and B is the thickness of the two-dimensional domain.

11 3. Process-dependent stochastic method

12 In this section, the stochastic method for the description of printing deviation is developed. As 3D concrete
13 printing proceeds over time and generates different position deviations at different times, it is essentially a stochastic
14 process. The generated stochastic processes are then mapped to the designed 3DPC structure in a process-dependent
15 manner to include this unpredictable position deviation.

16 3.1. Stochastic process modeling


17 A stochastic process H(ω̄, t) can be regarded as a function of the random event ω̄ and the time t. To determine
18 the explicit expression of the function, two methodologies have been proposed, including the physical modeling
19 and mathematical expansion. Compared to the physical modeling that takes the physical mechanism of stochastic
20 excitation into account, the mathematical expansion is more general, as it transforms the stochastic process into a set
21 of random variables:
H(ω̄, t) ⇒ Ĥ(ω̄, t) = F χ, t

(12)
22 where Ĥ is the approximation of the stochastic process and χ = {χi , i = 1, 2, 3, · · · , n} is the vector collecting n
23 random variables. Representative approaches are the Spectral Representation (SR) method [47], Karhunen–Loève
24 (KL) expansion method [48] and Stochastic Harmonic Function (SHF) method [49], which can be written in a unified
25 form:
F χ, t = µ(t) + YMethod (t)

(13)
26 where µ(t) is the mean value at time t; YMethod (·) is the mathematical expansion of a zero-mean stochastic process.

5
1 In the SHF method, both phase angles and wave frequencies in the trigonometric series are set as random variables,
2 which not only greatly reduces the random terms in the SR method, but also eliminates the limitation in complex cases
3 in the KL method. Thus, the SHF method is adopted here for stochastic modeling.
4 The SHF method expresses the stochastic process as the linear superposition of a series of stochastic harmonic
5 functions [49]:
XN
YSHF (t) = A(ω̃i ) cos (ω̃i t + ϕi ) (14)
i=1
6 where N is the truncated number of the expansion components; A(ω̃i ), ω̃i and ϕi are the amplitude, frequency and
7 phase angle of each component, respectively. The truncated number N divides the frequency domain [ωL , ωU ] into
8 small intervals [ωi−1 , ωi ], over which the frequencies ω̃i are randomly distributed. Here, ωL and ωU are the lower and
9 upper bounds of the frequency domain, respectively. For each frequency component ω̃i , there are a random phase
10 angle ϕi uniformly distributed in the range [0, 2π] and a random amplitude A(ω̃i ) related to the correlation function.
11 For the second type of SHF method in Figure 3, a uniform distribution of ω̃i is adopted and the amplitude is given as:
r
2
A(ω̃i ) = S (ω̃i )∆ωi (15)
π
12 where ∆ωi = ωi − ωi−1 is the length of the frequency interval; S is the power spectral density (PSD) function defined
13 as the Fourier transform of the correlation function R:
ˆ +∞
S (ω) = R(τ)e−iωτ dτ (16)
−∞

Figure 3: Schematics of the second type Stochastic Harmonic Function (SHF) method

14 The correlation function describes the temporal variability of the stochastic process. Here, the Gaussian distribu-
15 tion is used and the squared exponential autocorrelation function is adopted [50]:
τ
"  2 #
R(τ) = exp − (17)
c

16 where c = √tsπ is related to the correlation time t s . The correlation time t s represents the extent of temporal variability,
17 as shown in Figure 4. By defining different correlation times, three stochastic processes over 0 ∼ 13 s are drawn by
18 different color and lines in Figure 4(a). With the increase of the correlation time, the printing deviations are distributed
19 more uniformly along the printing process. To show this more clearly, the printing deviations are mapped to 3DPC
20 layers in Figure 4(b) (see Section 3.2 for additional details), where the color intensity represents the deviation value:
21 dark for larger and light for smaller. Similarly, as the correlation time develops, the color changes less frequently,
22 indicating a more uniform distribution of the printing deviation. As it determines the temporal variability, the choice
23 of the correlation time plays an important role in stochastic analysis, which will be further investigated in Section 5.
24 Substituting Eq.(17) into Eq.(16), the explicit expression of the PSD function is:
σc
"  2 #

S (ω) = √ exp − (18)
2 π 2
6
Figure 4: Printing deviations under different correlation times for 3D concrete printing (a) stochastic process (b) map to 3DPC layer

1 where σ is the standard deviation. According to the above equations, the stochastic process of printing deviations
2 can be generated based on the mean value µ, standard deviation σ and correlation time t s . The mean value represents
3 the amplitude of the printing deviation, which is comparable to the layer width. The standard deviation denotes the
4 deviation degree of the data and is typically described by the coefficient of variation (COV= σ/µ). The correlation
5 time represents the temporal variability, as shown in Figure 4. Thus, a relatively comprehensive description of the
6 uncertainty feature of printing deviation is established.

7 3.2. Process-dependent mapping


8 As the printing deviation mainly comes from the error of robot movements and finally leads to the deviation
9 between the designed and actual print path, it is essentially a positioning deviation. Thus, the generated stochastic
10 process is mapped to the designed 3DPC structure to include the printing deviation.
11 As mentioned in Section 2, PD is a meshfree method. In other words, the numerical model is based on particles,
12 so the mapping can be accomplished by the coordinate conversion in Figure 5:

Actual coordinate = Designed coordinate + Stochastic process (19)

13 The major issue is how to establish the correspondence between the stochastic process and the particles. Here, the
14 characteristics of the printing process are taken into consideration to better reproduce the actual situation, so it is
15 termed as process-dependent mapping.

Figure 5: Coordinate conversion for process-dependent mapping

7
1 3D concrete printing is a layer-by-layer process; in other words, the print head moves up and begins the printing
2 of a new layer after completing the previous layer. As each layer can be regarded as a new printing process, there
3 is little correlation of the printing deviation between layers. Thus, the whole 3D printing process can be simplified
4 into independent stochastic processes of each layer. For each layer, the correlation can not be neglected and the
5 closer points along the print path will show stronger correlation, which can be described by the correlation function in
6 Eq.(17). Therefore, the mapping consists of two stages, i.e. the interlayer dividing and intralayer mapping, as shown
7 in Figure 6.
8 In the interlayer dividing stage (stage I), owing to the layer-wise construction manner of 3D concrete printing, the
9 PD particles are divided into layer groups according to the z-coordinate. The z-dimension of each divided layer is
10 relatively small compared to the other two dimensions. The printing deviations in the z-direction are neglected and
11 those in the other two directions (x- and y-) are assumed to be independent from each other. Then in stage II, once the
12 time interval t1 for printing the current layer is identified, two independent stochastic processes of x- and y-directional
13 printing deviations can be generated according to Section 3.1. These two stochastic processes have different printing
14 deviation series but the same stochastic properties, including the mean value, the standard deviation and correlation
15 time, which are finally mapped to the corresponding layer group particles.

Figure 6: Schematics of process-dependent mapping

16 Thus, the main issue is the establishment of correspondence between stochastic process points and layer group
17 particles. The print path represents the position of the print head and is both related to the stochastic process modeling
18 and the particle generation, so it plays an important role in the correspondence establishment. For example, if the print
19 velocity is v, when the printing proceeds T time, the print head will move L = vT distance along the print path. In
20 other words, the point at t = T in the stochastic process corresponds to the particle at L = vT along the print path. As
21 illustrated in Figure 7, the stochastic process points (green circles) can be directly mapped to the particles along the
22 print path (red circles) via the print velocity. Regarding the particles along the width marked by blue squares, they are
23 located in the same position on the print path as the corresponding red particles along the print path. According to this
24 relation, they use the same deviation value as the particles located along the print path. Therefore, particles located
25 along the print path and its corresponding particles along the width need be identified and mapped by the same point
26 of stochastic process. To facilitate the search of particles, some special techniques are used during their generation.
27 For this, the particle generation follows a given order. Take a small segment along the print path in Figure 8
28 for example. By identifying particle number N along the print path (red line), the print path is divided and particles
29 along the print path are generated first. Then the particles along the width are generated by duplicating each particle
30 along the print path in the width direction (blue line) according to the layer width and the division number n1 . If
31 the height division of the current layer is greater than 1, the particles of the first division are duplicated division-by-
32 division (green line) according to the layer height and the division number n2 . In this way, the PD particles form a
33 tree structure in Figure 9. According to this structure, the particle along the print path and its corresponding particles
34 along the width and height directions are easily sought out by the particle number. For example, the i-th particle in
35 the print path direction corresponds to the particles N + (i − 1)n1 + 1 ∼ N + in1 in the width direction. If the layer has
8
Figure 7: Schematics of intralayer mapping

1 more than one height division, the particle number only need to plus (N + Nn1 )( j − 1) for j-th division.

Figure 8: Print path based particle generation method

2 4. Numerical implementation
3 With the process-dependent stochastic method in Section 3, numerical samples including printing deviations are
4 generated. To obtain a clear understanding of the influence of printing deviation on 3D-printed concrete (3DPC)
5 buildability, the stochastic analysis is required. This procedure is developed and verified in this Section.

6 4.1. Monte Carlo simulation


7 A statistical analysis method, i.e. the Monte Carlo method [51], is adopted for stochastic simulation. The Monte
8 Carlo method conducts the simulation of a large number of random samples and uses the response results of the
9
Figure 9: Tree structure of generated particles

1 samples for statistical analysis. For the present problem, the random samples are the numerical models with printing
2 deviations given by the process-dependent stochastic method in Section 3. The response results are the results obtained
3 from their buildability analysis. Therefore, the stochastic analysis procedure follows the scheme in Figure 10, which
4 is summarized as follows:
5 1) Develop the numerical model according to the designed 3DPC structure and follow Figure 8 to generate particles;
6 2) For each layer, use stochastic process modeling to generate printing deviations:
7 (a) Discretize the frequency domain [ωL , ωU ] into small intervals [ωi−1 , ωi ] and identify the mean value µ,
8 standard deviation σ, correlation time t s ;
9 (b) In each interval [ωi−1 , ωi ], get the uniformly distributed random wave frequency ω̃i . Correspondingly, get
10 the independent uniformly distributed random phase angle ϕi over [0, 2π] and the random amplitude A(ω̃i )
11 by Eq.(15);
12 (c) Use Eq.(14) to define the stochastic harmonic functions and generate the zero-mean stochastic process by
13 linear superposition of the functions. Plus the mean value to get the final stochastic process.
14 3) For each stochastic process, use the process-dependent mapping to map the printing deviations to the coordinates
15 of particles:
16 (a) Search the particle located along the print path and its corresponding particles located along the width
17 direction according to the tree structure in Figure 9;
18 (b) Follow Figure 7 to map the points of the stochastic process to the particles along the print path according
19 to the print velocity.
20 (c) When mapping each particle located along the print path, map the corresponding particles located along
21 the width direction with the same deviation value.
22 4) Construct the numerical samples including the printing deviations and perform the buildability analysis for each
23 sample.
24 5) Repeat steps (2) to (4) until the Monte Carlo simulation is completed.
25

10
Figure 10: Schematics of stochastic analysis procedure

11
1 4.2. Method verification
2 To validate the effectiveness of the above method, the hollow cylinder 3DPC test performed by Wolfs et al. [10]
3 is simulated here. In this test, the cylinder structure was printed five times. The statistical information of the test was
4 given by recording the results of each time, which provides a suitable example for method verification.
5 The dimensions of the hollow cylinder are presented in Figure 11(a). The inner radius of the cylinder is 230 mm,
6 the outer radius is 270 mm, and the layer height is 10 mm. Based on the above dimensions and a convergence study,
7 the numerical model of the designed 3DPC structure is developed in Figure 11(b), in which the particle spacing is
8 ∆x = 10 mm and the horizon size is δ = 3.015∆x.

Figure 11: (a) Dimensions of the hollow cylinder (unit:mm) and (b) developed numerical model

9 To introduce the printing deviation in the numerical model, the stochastic process modeling is conducted, which
10 requires the identification of the mean value µ, standard deviation σ and correlation time t s . As mentioned in Section
11 3.1, the mean value is comparable to the layer width. Here, it is chosen as 10% of the layer width µ = 0.1 × 40 = 4
12 mm. Based on the mean value and the measurement points of the first layer in the radial deformation plot, the standard
13 deviation is obtained as σ = 0.6 mm, leading to the coefficient of variation (COV) as 0.6/4 =15%. Regarding the
14 correlation time, it is assumed to be t s = 1s and a more detailed investigation on it is presented in Section 5.2.
15 Following the above approach, 200 samples are constructed, based on which the stochastic simulation of buildabil-
16 ity analysis is performed. Here, both the mechanical behavior of 3DPC and the time-dependent process are consid-
17 ered, which play an important role in buildability prediction. The fluid-solid integrated peridynamic (PD) model [38]
18 is used for the description of the mechanical behavior and the particles are activated layer-by-layer for the simulation
19 of the time-dependent process. The material parameters of the fluid-solid integrated PD model are calibrated by the
20 unconfined uniaxial compression test (UUCT) and they are as follows: ρ = 2070 kg/m3 , E0 (t[min]) = 1.14t + 81.73
21 kPa, fc+ (t[min]) = 0.015t + 0.594 kPa, fc− (t[min]) = 0.15t + 5.94 kPa, ε+c = 0.014, ε−c = 0.14, α+c = 0.2, α−c = 0.1 (for
22 more details see Appendix A). The layer-wise simulation is conducted with the printing speed v = 5000 mm/min.
23 The printing test [10] recorded the structure deformations after 23, 26 and 29 layers and the failure layer of each
24 print. The deformation shape is given in the form of radial displacement along the height direction. The comparison
25 between the simulation results and the experimental data is given in Figure 12. The cylinder shows a bulging outwards
26 mode, with maximum radial deformations located at mid-height of the structure; this is due to the vertical gravity
27 loading, Poisson’s ratio effect and the movement restriction at the bottom. The deformation curves of the 3DPC tests
28 show randomness and vary in a certain range (silver region). This can not be captured by deterministic simulation, and
29 requires the use of stochastic analysis. By stochastic simulation, a similar extent of randomness is reproduced by the
30 simulated deformation of the samples in blue dashed lines. Based on the sample results, the stochastic information is
31 obtained and drawn in Figure 12. It is observed that in all cases (23, 26, 29 layers), the mean curves in red solid lines
32 are almost within the range of the experimental scatter and the mean curves plus and minus 3 times standard variation
33 in green dotted lines basically cover the sample range. Thus, the effectiveness of the proposed method is verified.
12
(a) (b) (c)

Figure 12: Comparison of deformation results between simulation and experiment after (a) 23 layers (b) 26 layers and (c) 29 layers

1 The method is further verified by a quantitative comparison of the statistical information in Table 1, including
2 the mean value, standard deviation and COV. The number of the failure layer, maximum radial deformation and the
3 corresponding z position after 23, 26, 29 layers are compared for the simulation results and experimental data. The test
4 results exhibit variability, especially in the results of maximum radial deformation. The maximum of COV reaches
5 21%, which demonstrates the necessity of stochastic simulation. Through the stochastic analysis of 200 samples, the
6 variability in the failure and deformation results is well reproduced. There is only a small difference in the mean
7 values between the simulation and test results. This indicates that the proposed method can capture the statistical
8 properties of the 3DPC system.

Table 1: Comparison of stochastical information between simulation and experiment


µ σ COV µ σ COV
PD 9.8 1.6 17% PD 124.5 6.3 5%
23 layer 23 layer
Test 10.3 2.2 21% Test 114.7 4.5 4%
z position of
Max. radial PD 12.9 3.2 25% PD 127.9 3.8 3%
26 layer max. radial 26 layer
deformation(mm) Test 12.7 2.7 21% Test 116.5 2.8 2%
deformation(mm)
PD 16.4 3.1 19% PD 134.1 4.0 3%
29 layer 29 layer
Test 15.3 2.2 15% Test 114.7 2.6 2%
PD 29 2.0 7%
Number of failure layer
Test 29 2.4 8%

13
1 5. Stochastic analysis

2 Based on the procedure developed in Section 4, the stochastic analysis of buildability of 3D-printed concrete
3 (3DPC) including printing deviation is conducted in this Section. The lateral deformation and failure time characterize
4 the 3DPC buildability. Less deformation and longer failure time indicate that the 3DPC structure has better buildability
5 features. Thus, the stochastic analysis here focus on the deformation and failure behavior of 3DPC.

6 5.1. Target problem


7 Owing to the formwork-free feature, 3DPC structures have great freedom in structural shape. Thus, regular shaped
8 3DPC structures, such as the cylinder structure in Section 4.2, do not meet the actual demands. To take this into
9 consideration, other structural types are included in the analysis. Bester et al. [52] presented a 3DPC benchmark test
10 including different structural shapes, varying from ascending to descending structures, which offers a suitable problem
11 for stochastic analysis. As shown in Figure 13, the test keeps the layer width w = 40 mm and the layer height h = 10
12 mm but changes the radius R along the height direction to get different structural shapes. If the radius increases from
13 top to bottom Rn <R0 , it is an ascending structure; contrariwise, a descending structure (Rn >R0 ). Descending structures
14 have less support for the layer and are more prone to fail before full printing is achieved.

Figure 13: Schematics of structure shape in 3DPC benchmark test

15 Here, both the less and more challenging structural shapes, i.e. ascending and descending structures, are taken
16 as target problem. As illustrated in Figure 14(a), the ascending structure is in the form of a half sphere dome with
17 diameter of 400 mm; the descending structure in Figure 14(b) is a 450 mm height cone with diameter decreasing from
18 300 mm to 150 mm. They are termed as dome 400 and cone 150, respectively. The corresponding numerical models
19 of both structures are shown in Figure 15, where the color represents the layer number: red for larger and blue for
20 smaller.
21 Before the stochastic analysis, a deterministic buildability simulation is conducted to verify the numerical model.
22 The fluid-solid integrated peridynamic (PD) model [38] is used for the description of the mechanical behavior; the
23 material parameters are calibrated by the direct shear test as follows: E0 (t[min]) = 1.58t + 112.26 kPa, fc+ (t[min]) =
24 0.02t + 1.042 kPa, fc− (t[min]) = 0.2t + 10.42 kPa, ε+c = 0.014, ε−c = 0.14, α+c = 0.2, α−c = 0.1 (for more details
25 see Appendix A). The time-dependent process is modeled by layer-wise particle activation with the printing speed
26 v = 3600 mm/min and the density ρ = 2100 kg/m3 given by the test.
27 The deformation and failure results of the simulation are compared to the experimental data in Table 2. The
28 failure layer, print duration as well as the radial deformation of both structures show good agreement between the PD
29 simulation and the test. The maximum relative deviation does not exceed 20%.
30 Next, the verified numerical model is used for stochastic analysis. As this test was only carried out once and
31 no stochastic information was provided, the stochastic parameters of the verification study in Section 4.2 are used.
32 Therefore, the mean value is 10% of the layer width µ = 0.1 × 40 = 4 mm and the coefficient of variation (COV) is
14
Figure 14: Schematics of the target problem (a) ascending structure and (b) descending structure (unit: mm)

Figure 15: Numerical model for target problems (a) dome 400 and (b) cone 150

Table 2: Comparison of failure-deformation results between PD simulation and test


Failure layer Print duration(s) Radial deformation(mm)
Structures
PD Test Deviation PD Test Deviation PD Test Deviation
Dome 400 15 16 6.2% 266.2 325 18% 29.9 29.9 0%
Cone 150 29 29 0% 294.7 315 6.4% 8.4 7.2 16.5%

15
1 15%, leading to the standard deviation σ = 0.15 × 4 = 0.6 mm. Regarding the correlation time, four correlation times
2 are used to investigate its influence. This is necessary as the accurate determination of the correlation time is still a
3 somewhat controversial issue in the field of uncertainty quantification [50, 53]. The correlation times: t s = 0.5s, 2.0s,
4 5.0s and 20.0s are chosen for investigation, which are around 1/40, 1/10, 1/4 and 1 times the time interval of printing
5 the base layer of dome 400, respectively.

6 5.2. Results and discussion


7 5.2.1. Dome 400
8 200 samples are produced and calculated for each correlation time. Figure 16 illustrates the typical numerical
9 samples under different correlation times. The color represents the magnitude of printing deviation: light for greater
10 and dark for smaller. The printing deviations generated by the process-dependent stochastic method are randomly
11 distributed, which is in accordance with its unpredictable nature. With the increase of the correlation time, the color
12 changes less frequently from Figure 16(a) to (d), indicating a more uniform distribution of printing deviation. When
13 the correlation time increases to the time interval of printing the base layer (Figure 16(d)), the color is uniform over
14 each layer. In this case, particles in the same layer have same printing deviation, leading to a layer-shifting mode.

Figure 16: Typical samples of dome 400 for (a) t s = 1/40t1 (b) t s = 1/10t1 (c) t s = 1/4t1 and (d) t s = 1t1 (unit: mm)

15 The deformation shapes of dome 400 at the time of failure are presented in Figure 17. The blue solid lines represent
16 the results of different samples, which are recorded by vertical position versus radial position along the red line (A1 -
17 A2 ) in Figure 14(a). The purple dotted lines denote the deterministic results without the consideration of printing
18 deviations and the red dashed lines are the original undeformed shape (A1 -A2 ). In addition to the deformation curves,
19 the failure points are also marked in Figure 17 by blue points. The deformation curves have a similar trend regardless
20 of the correlation time, deforming outwards at the base and changing to inward deformation with the increase of
21 the height. In this ascending structure, the radius of the structure decreases with the printing of the upper layers.
22 Correspondingly, the center of gravity moves towards the structure center; under the gravity loading, the settlement of
23 the structure changes from outward to inward, resulting in this deformation shape.
24 Similarly to the cylinder test, the deformation curves and failure points of the different samples show some dis-
25 persion and distribute in a certain band. Some samples even deform greater than the deterministic results, exhibiting a
26 more risky manner. This indicates that the printing deviations can pose adverse effects on 3DPC buildability and need
27 be included in buildability analysis.

16
(a) (b)

(c) (d)

Figure 17: Deformation shapes of dome 400 at the time of failure for (a) t s = 1/40t1 (b) t s = 1/10t1 (c) t s = 1/4t1 and (d) t s = 1t1

1 Comparing the results between different correlation times, it is found that the level of dispersion increases with
2 the correlation time. To investigate this trend, the mean and COV values of the failure time of dome 400 are compared
3 in Figure 18(a), which are drawn by different markers. The mean failure time of dome 400 shows an increase from
4 254 s to 262 s as the correlation time increases from 1/40 to 1 times the interval time of printing the base layer; the
5 corresponding COV grows from 5% to around 17%. This indicates that the increase of the correlation time leads to a
6 better buildability but greater variability.
7 To explain this phenomenon, the probabilistic information about the failure response is investigated in Figure
8 19(a), in the form of the probability density functions (PDFs) of the failure time under typical correlation times. It
9 is observed that the PDFs illustrate a bimodal distribution. With the increase of the correlation time, the peak of the
10 PDF becomes lower and the band becomes wider, leading to the increase of the mean and COV values in Figure 18(a).
11 Note that the second peak of PDF is concentrated around the failure time of the deterministic result given in Table 2.
12 Thus, the first peak is related to the introduction of printing deviation.
13 The first peak has a shorter failure time than the second peak. This is due to the non-uniform load distribution
14 caused by the printing deviation. With the introduction of printing deviation, the particle coordinates become random
15 so the load no longer distributes uniformly, leading to a more dangerous situation. Thus, the structure shows a poorer
16 buildability with shorter failure time, forming a new peak. This is consistent with the greater deformation in Figure
17 17 and further demonstrates the adverse effects of the printing deviation.
18 Interestingly, the PDF of the first peak diminishes with the increase of the correlation time, until becoming smaller
19 than the PDF of the second peak. This is related to the change of unevenness. When the correlation time is small,
20 the temporal variability is great, which results in a highly non-uniform load distribution. In this case, the printing
21 deviation plays a major role and its adverse effect is significant, so the structure is more likely to fail around the
22 first peak. When the correlation time increases, the temporal variability decreases and the structure tends to a layer-
23 shifting mode. As the radius of dome 400 increases from top to the bottom, each layer still has sufficient support when
24 layer-shifting happens. Thus, the influence of the printing deviation reduces with the correlation time, leading to the
25 probability moving from the first peak to the second peak.

17
(a) (b)

Figure 18: Mean and COV values of failure time under different correlation times for (a) dome 400 and (b) cone 150

(a)
(b)

(c)

Figure 19: Probability information of the failure time for dome 400 (a) PDFs under typical correlation times (b) contour to PDF surface (c) PDF
evolution surface against correlation time

18
1 This trend is more intuitively observed by the PDF evolution surface in Figure 19(c), which shows the PDFs for
2 different correlation times. The color represents the probability of failure time: red for larger probability and blue
3 for smaller probability. With the increase of the correlation time, the probability of the first peak shows an abrupt
4 decrease initially and a slower decrease afterwards. Also, the growth of the second peak shows a significant change.
5 This indicates that the influence of the printing deviation has a sensitive region of correlation time. To discover the
6 sensitive region, the contour of PDF surface is investigated in Figure 19(b). The flow of PDF against correlation time
7 is more evident in this figure. The two bands represent two peaks of PDF, the color of which becomes more blue
8 and red, respectively, indicating an decrease and increase in probability. The color change is more dramatic when
9 correlation time is smaller than 0.1 times the interval time of printing the base layer, which gives the sensitive region
10 of correlation time for the present problem.
11 Additionally, the discrepancy between the mean value, COV as well as PDF all shows an obvious change for
12 different correlation times. It is concluded that the correlation time of the printing deviation has a significant impact
13 on the stochastic properties of the 3DPC system and needs be carefully identified in buildability modeling, especially
14 when the correlation time is smaller than 0.1 times the interval time of printing the base layer.

15 5.2.2. Cone 150


16 Regarding cone 150, the typical samples for different correlation times are given in Figure 20. Similar to dome 400,
17 the printing deviations are also stochastically distributed, the uniformity of which also increases with the correlation
18 time. When the correlation time is t s = 1t1 (Figure 20(d)), it can cover the time interval of printing each layer. At this
19 time, particles in the same layer show the same printing deviation, so each layer undergoes an overall shifting.

Figure 20: Typical samples of cone 150 for (a) t s = 1/40t1 (b) t s = 1/10t1 (c) t s = 1/4t1 and (d) t s = 1t1 (unit: mm)

20 The corresponding results are presented in Figures 18(b), 21 and 22. Compared to dome 400, cone 150 shows
21 a different deformation shape in Figure 21, which is recorded by the vertical position versus radial position along
22 the red line (B1 -B2 ) in Figure 14(b). The maximum radial deformations mainly locate in the middle parts of the
23 structure, in a bulging outwards mode. As the radius of structure grows with the increase of the height, the center of
24 gravity moves away from the structure center. Thus, for this descending structure, the structure settlement is always
25 outward. Similarly, cone 150 also shows randomly distributed deformation curves and failure points, some of which

19
1 also exceed the deterministic results and illustrate a more risky manner. This further demonstrates the necessity of
2 including printing deviations for buildability modeling.

(a) (b)

(c) (d)

Figure 21: Deformation shapes of cone 150 at the time of failure for (a) t s = 1/40t1 (b) t s = 1/10t1 (c) t s = 1/4t1 and (d) t s = 1t1

3 With the increase of the correlation time, the distribution range becomes wider and the randomness grows, simi-
4 larly to the former case. Cone 150 shows greater variability than dome 400 under all correlation times. This maybe
5 due to its challenging structure shape. In cone 150, the radius is increasing with the printing of upper layers, so lower
6 layers can not provide sufficient support to upper layers. In this case, the printing deviation is prone to aggravate this
7 manner, leading to increase in variability. It can be concluded that the printing deviation can pose more adverse effects
8 on challenging shaped structures.
9 The effect of the correlation time on the stochastic properties is also investigated in this case, including the mean
10 value, COV and PDF. As shown in Figure 18(b), the COVs of failure time grow from 12% to around 30% when the
11 correlation time increases from 1/40 to 1 times the interval time of printing the base layer. This increase trend in
12 variability is similar to dome 400. However, the corresponding mean values illustrate a decrease trend, indicating that
13 the growth of the correlation time undermines the buildability. This is quite different from the former case, in which
14 the increase of the correlation time improves the buildability.
15 This can be explained by the probability distribution in Figure 22(a). As we can see, the PDFs under typical
16 correlation time also show a bimodal distribution. The second peak is also concentrated around the failure time of the
17 deterministic result given in Table 2 and its failure time is also longer than the first peak, similarly to dome 400. This
18 indicates that the printing deviation again induces non-uniform load distribution and reduces the 3DPC buildability
19 in the case of cone 150. However, in this case, the probability of the first peak does not move to the second peak
20 (longer failure time) with the increase of the correlation time. On the contrary, it moves in an opposite direction and
21 the failure time decreases. As discussed in the former case, as the correlation time increases, the structure approaches
22 a layer-shifting mode. This reduces the influence of the printing deviation for the ascending structure (dome 400) as
23 each layer still has sufficient support. But for descending structure cone 150, this is more dangerous as more area loses
24 its lower support, which undermines the buildability. As a result, probability moves to shorter failure time, leading to
25 a decline in mean failure time.

20
1 More detailed probability information is included in Figures 22(b) and (c) to further investigate the effect of the
2 correlation time. As shown in Figure 22(c), the probability of the first peak shows an abrupt decline at the beginning,
3 which is the sensitive region of correlation time discovered in the case of dome 400. Afterwards, the probability of
4 the first peak still has a value much larger than the second peak. This demonstrates that the adverse effect of printing
5 deviation is more significant on challenging shaped structures, which is consistent with the deformation results in
6 Figure 21. Regarding the probability change of the first peak, the contour of the PDF surface in Figure 22(b) indicates
7 that the band of the first peak moves in the direction of decreasing failure time, some probabilities of which even form
8 a third band with shorter failure time.
9 The opposite trends with correlation time in these two cases give a clear understanding of the influence of printing
10 deviation on 3DPC buildability. The printing deviation changes the load distribution thus localizes the failure zone,
11 leading to a more dangerous situation. This adverse effect undermines the buildability and forms another peak in PDF
12 with shorter failure time than the deterministic result. The variability of printing deviation decreases with the corre-
13 lation time and forms a layer-shifting mode when the correlation time covers the time interval of printing each layer.
14 This mode reduces the adverse effect for ascending structures as they have sufficient lower support but aggravates the
15 insufficient support of descending structures.

(a)
(b)

(c)

Figure 22: Probability information of the failure time for cone 150 (a) PDFs under typical correlation times (b) contour to PDF surface (c) PDF
evolution surface against correlation time

16 According to the above analysis, some guidances for 3D printing process design are obtained. The printing devia-

21
1 tions need be controlled in 3D concrete printing to reduce its adverse effect on buildability, especially for challenging
2 shaped structures. The correlation time has a significant influence on the adverse effect, which may provide insight
3 for printing deviation control.

4 6. Concluding remarks

5 This paper presents a process-dependent stochastic method to include the printing deviations for buildability mod-
6 eling of 3D-printed concrete (3DPC). Stochastic process modeling and process-dependent correspondence are pro-
7 posed to reproduce this unpredictable position deviation. On this basis, a stochastic analysis procedure is developed
8 by Monte Carlo simulation and verified by the simulation of a 3DPC cylinder test. Also, the stochastic analysis of two
9 different types of structures is conducted to understand the influence of the printing deviation on 3DPC buildability.
10 According to the obtained results, concluding remarks are drawn as follows:
11 1) The printing deviation yields buildability variability, especially for challenging shaped structures. This indicates
12 that the printing deviation has an adverse effect on 3DPC buildability, which requires consideration in buildability
13 modeling.
14 2) With the introduction of printing deviation, the probability density functions (PDFs) of failure time illustrate a
15 bimodal distribution. The second peak is concentrated around the deterministic result and the corresponding
16 failure time is longer than the first peak. The formulation of the first peak demonstrates that the printing devi-
17 ation has a significant effect on 3DPC buildability and its shorter failure time value indicates that this effect is
18 detrimental.
19 3) With the increase of the correlation time, the variability of printing deviation decreases and forms a layer-shifting
20 mode when the correlation time covers the time interval of printing each layer. This mode reduces the adverse
21 effect for ascending structures as they have sufficient lower support but aggravates the insufficient support of
22 descending structures, which improves and undermines the mean buildability, respectively.
23 4) Guidances on 3D printing process design are drawn. The printing deviation need be controlled and reduced in 3D
24 concrete printing, especially for challenging shaped structures. The correlation time can significantly affect the
25 influence of printing deviation on 3DPC buildability, which may provide insight for printing deviation control.
26 Studies of robot control [54, 55] offer effective ways for the reduction of printing deviation. Combining the adap-
27 tive Monte Carlo localization method [56] with scan matching is a popular approach for achieving high precision.
28 Also, the development of new algorithms, such as the Iterative Closest Point [57] and the Normal Distribution Trans-
29 form [58], can enhance the improvement of precision. Regarding the correlation time, it is related to various factors
30 in the printing process, such as the localization algorithm and the scanner feedback frequency. Thus, control of the
31 correlation time is more involved and will require further investigation.

32 Appendix A. Calibration for material parameters of fluid-solid integrated peridynamic model

33 The unconfined uniaxial compression test (UUCT) provides the experimental data of elastic modulus, compressive
34 strength and even the compressive strain-stress curves at different concrete ages, which is typically used for the
35 calibration of the “solid phase” of 3D-printed concrete [10, 19]. This requires seven parameters (E0 , fc± , ε±c , α±c ) in the
36 fluid-solid integrated peridynamic (PD) model [38].
37 To accomplish this, a PD model of the UUCT is developed and the concrete damage model is adopted for sim-
38 ulation. First, the compressive parameters are directly calibrated from UUCT. For example, the compressive time-
39 dependent parameters (E0 , fc− ) are calibrated from the stiffness and compressive strength data at different ages. If
40 compressive strain-stress curves are available, the compressive damage parameters (ε−c , α−c ) can be obtained from the
41 corresponding strain of the peak stress and the slope of descending segment. Then the tensile parameters ( fc+ , ε+c , α+c )
42 can be determined by a typical relation used for the description of different tensile and compressive nonlinearities of
43 concrete: fc+ = 0.1 fc− , ε+c = 0.1ε−c , α+c = 2α−c . More details are given in [38].
44 In terms of the direct shear test, the shear strength can be related to the compressive strength. Then a similar way
45 can be followed for the material parameters calibration.

22
1 Acknowledgment

2 Financial supports from the Spanish Ministry of Economy and Competitiveness, through the Severo Ochoa Pro-
3 gramme for Centres of Excellence in R&D (Grant No. CEX2018-000797-S), the National Natural Science Foundation
4 of China (Grant No. 52078361) and the Innovation Program of Shanghai Municipal Education Commission (Grant
5 No. 2017-01-07-00-07-E00006) are greatly appreciated.

6 References
7 [1] J. G. Sanjayan, B. Nematollahi, 3d concrete printing for construction applications, in: 3D concrete printing technology, 2019, pp. 1–11.
8 [2] M. Batikha, R. Jotangia, M. Y. Baaj, I. Mousleh, 3d concrete printing for sustainable and economical construction: A comparative study,
9 Automation in Construction 134 (2022) 104087.
10 [3] S. Marconi, M. Carraturo, G. Alaimo, S. Morganti, G. Scalet, M. Conti, A. Reali, F. Auricchio, Additive manufacturing: Challenges and
11 opportunities for structural mechanics, 50+ Years of AIMETA: A Journey Through Theoretical and Applied Mechanics in Italy (2022)
12 437–451.
13 [4] K. Tan, The framework of combining artificial intelligence and construction 3d printing in civil engineering, in: MATEC web of conferences,
14 volume 206, EDP Sciences, 2018, p. 01008.
15 [5] E. Forcael, J. Pérez, Á. Vásquez, R. Garcı́a-Alvarado, F. Orozco, J. Sepúlveda, Development of communication protocols between bim
16 elements and 3d concrete printing, Applied Sciences 11 (2021) 7226.
17 [6] S. Hou, Z. Duan, J. Xiao, J. Ye, A review of 3d printed concrete: Performance requirements, testing measurements and mix design,
18 Construction and Building Materials 273 (2021) 121745.
19 [7] Y. Chen, S. Chaves Figueiredo, Ç. Yalçinkaya, O. Çopuroğlu, F. Veer, E. Schlangen, The effect of viscosity-modifying admixture on the
20 extrudability of limestone and calcined clay-based cementitious material for extrusion-based 3d concrete printing, Materials 12 (2019) 1374.
21 [8] C. Zhang, Z. Hou, C. Chen, Y. Zhang, V. Mechtcherine, Z. Sun, Design of 3d printable concrete based on the relationship between flowability
22 of cement paste and optimum aggregate content, Cement and Concrete Composites 104 (2019) 103406.
23 [9] L. Casagrande, L. Esposito, C. Menna, D. Asprone, F. Auricchio, Effect of testing procedures on buildability properties of 3d-printable
24 concrete, Construction and Building Materials 245 (2020) 118286.
25 [10] R. Wolfs, F. Bos, T. Salet, Early age mechanical behaviour of 3d printed concrete: Numerical modelling and experimental testing, Cement
26 and Concrete Research 106 (2018) 103–116.
27 [11] A. P. Rubin, L. C. Quintanilha, W. L. Repette, Influence of structuration rate, with hydration accelerating admixture, on the physical and
28 mechanical properties of concrete for 3d printing, Construction and Building Materials 363 (2023) 129826.
29 [12] N. Roussel, Rheological requirements for printable concretes, Cement and Concrete Research 112 (2018) 76–85.
30 [13] R. Jayathilakage, J. Sanjayan, P. Rajeev, Direct shear test for the assessment of rheological parameters of concrete for 3d printing applications,
31 Materials and Structures 52 (2019) 1–13.
32 [14] J. Kruger, S. Zeranka, G. van Zijl, 3d concrete printing: A lower bound analytical model for buildability performance quantification,
33 Automation in Construction 106 (2019) 102904.
34 [15] J. Kruger, S. Zeranka, G. v. Zijl, Quantifying constructability performance of 3d concrete printing via rheology-based analytical models, in:
35 Rheology and Processing of Construction Materials, 2019, pp. 400–408.
36 [16] A. Perrot, D. Rangeard, A. Pierre, Structural built-up of cement-based materials used for 3d-printing extrusion techniques, Materials and
37 Structures 49 (2016) 1213–1220.
38 [17] R. Wolfs, A. Suiker, Structural failure during extrusion-based 3d printing processes, The International Journal of Advanced Manufacturing
39 Technology 104 (2019) 565–584.
40 [18] A. S. Suiker, R. J. Wolfs, S. M. Lucas, T. A. Salet, Elastic buckling and plastic collapse during 3d concrete printing, Cement and Concrete
41 Research 135 (2020) 106016.
42 [19] H. Liu, T. Ding, J. Xiao, V. Mechtcherine, Buildability prediction of 3d–printed concrete at early-ages: A numerical study with drucker–prager
43 model, Additive Manufacturing 55 (2022) 102821.
44 [20] R. Comminal, W. R. L. da Silva, T. J. Andersen, H. Stang, J. Spangenberg, Modelling of 3d concrete printing based on computational fluid
45 dynamics, Cement and Concrete Research 138 (2020) 106256.
46 [21] M. T. Mollah, R. Comminal, M. P. Serdeczny, D. B. Pedersen, J. Spangenberg, Stability and deformations of deposited layers in material
47 extrusion additive manufacturing, Additive Manufacturing 46 (2021) 102193.
48 [22] V. Mechtcherine, F. P. Bos, A. Perrot, W. L. da Silva, V. Nerella, S. Fataei, R. J. Wolfs, M. Sonebi, N. Roussel, Extrusion-based additive
49 manufacturing with cement-based materials–production steps, processes, and their underlying physics: a review, Cement and Concrete
50 Research 132 (2020) 106037.
51 [23] S. A. Silling, Reformulation of elasticity theory for discontinuities and long-range forces, Journal of the Mechanics & Physics of Solids 48
52 (2000) 175–209.
53 [24] S. A. Silling, E. Askari, A meshfree method based on the peridynamic model of solid mechanics, Computers & structures 83 (2005)
54 1526–1535.
55 [25] P. Hartmann, C. Weißenfels, P. Wriggers, A curing model for the numerical simulation within additive manufacturing of soft polymers using
56 peridynamics, Computational Particle Mechanics 8 (2021) 369–388.
57 [26] G. Placzek, L. Brohmann, K. Mawas, P. Schwerdtner, N. Hack, M. Maboudi, M. Gerke, A lean-based production approach for shotcrete
58 3d printed concrete components, in: ISARC. Proceedings of the International Symposium on Automation and Robotics in Construction,
59 volume 38, IAARC Publications, 2021, pp. 811–818.

23
1 [27] R. Buswell, J. Xu, D. De Becker, J. Dobrzanski, J. Provis, J. T. Kolawole, P. Kinnell, Geometric quality assurance for 3d concrete printing
2 and hybrid construction manufacturing using a standardised test part for benchmarking capability, Cement and Concrete Research 156 (2022)
3 106773.
4 [28] V. Mechtcherine, V. N. Nerella, F. Will, M. Näther, J. Otto, M. Krause, Large-scale digital concrete construction–conprint3d concept for
5 on-site, monolithic 3d-printing, Automation in construction 107 (2019) 102933.
6 [29] R. J. Wolfs, F. P. Bos, E. C. Van Strien, T. A. Salet, A real-time height measurement and feedback system for 3d concrete printing, in: High
7 Tech Concrete: Where Technology and Engineering Meet: Proceedings of the 2017 fib Symposium, held in Maastricht, The Netherlands,
8 June 12-14, 2017, Springer, 2018, pp. 2474–2483.
9 [30] M. L. Gambhir, Stability analysis and design of structures, Springer Science & Business Media, 2004.
10 [31] R. Wolfs, F. Bos, T. Salet, Triaxial compression testing on early age concrete for numerical analysis of 3d concrete printing, Cement and
11 Concrete Composites 104 (2019) 103344.
12 [32] L. Lachmayer, R. Dörrie, H. Kloft, A. Raatz, Process control for additive manufacturing of concrete components, in: Third RILEM
13 International Conference on Concrete and Digital Fabrication: Digital Concrete 2022, Springer, 2022, pp. 351–356.
14 [33] R. Comminal, W. R. L. da Silva, T. J. Andersen, H. Stang, J. Spangenberg, Influence of processing parameters on the layer geometry in 3d
15 concrete printing: experiments and modelling, in: Second RILEM International Conference on Concrete and Digital Fabrication: Digital
16 Concrete 2020 2, Springer, 2020, pp. 852–862.
17 [34] A. Suiker, Mechanical performance of wall structures in 3d printing processes: Theory, design tools and experiments, International Journal
18 of Mechanical Sciences 137 (2018) 145–170.
19 [35] Z. Chang, Y. Xu, Y. Chen, Y. Gan, E. Schlangen, B. Šavija, A discrete lattice model for assessment of buildability performance of 3d-printed
20 concrete, Computer-Aided Civil and Infrastructure Engineering 36 (2021) 638–655.
21 [36] Z. Chang, H. Zhang, M. Liang, E. Schlangen, B. Šavija, Numerical simulation of elastic buckling in 3d concrete printing using the lattice
22 model with geometric nonlinearity, Automation in Construction 142 (2022) 104485.
23 [37] S. A. Silling, M. Epton, O. Weckner, J. Xu, E. Askari, Peridynamic states and constitutive modeling, Journal of Elasticity 88 (2007) 151–184.
24 [38] J. Zhu, X. Ren, M. Cervera, Peridynamic buildability analysis of 3d-printed concrete including damage, plastic flow and collapse, Additive
25 Manufacturing (2023) 103683.
26 [39] Y. I. Rabinovich, M. S. Esayanur, B. M. Moudgil, Capillary forces between two spheres with a fixed volume liquid bridge: theory and
27 experiment, Langmuir 21 (2005) 10992–10997.
28 [40] J. Y. Wu, J. Li, R. Faria, An energy release rate-based plastic-damage model for concrete, International journal of Solids and Structures 43
29 (2006) 583–612.
30 [41] X. Ren, S. Zeng, J. Li, A rate-dependent stochastic damage–plasticity model for quasi-brittle materials, Computational Mechanics 55 (2015)
31 267–285.
32 [42] M. Cervera, J. Oliver, R. Faria, Seismic evaluation of concrete dams via continuum damage models, Earthquake engineering & structural
33 dynamics 24 (1995) 1225–1245.
34 [43] F.-J. Ulm, O. Coussy, Modeling of thermochemomechanical couplings of concrete at early ages, Journal of engineering mechanics 121
35 (1995) 785–794.
36 [44] M. Cervera, J. Oliver, T. Prato, Thermo-chemo-mechanical model for concrete. i: Hydration and aging, Journal of engineering mechanics
37 125 (1999) 1018–1027.
38 [45] L. Jendele, V. Šmilauer, J. Červenka, Multiscale hydro-thermo-mechanical model for early-age and mature concrete structures, Advances in
39 Engineering Software 72 (2014) 134–146.
40 [46] J. Zhu, X. Ren, Failure modeling of concrete: A peri-dynamical approach with bond-based correspondence to bi-scalar damage model,
41 Engineering Fracture Mechanics 268 (2022) 108470.
42 [47] M. Shinozuka, G. Deodatis, Simulation of stochastic processes by spectral representation, Applied Mechanics Reviews 44 (1991) 191–204.
43 [48] S. Huang, S. Quek, K. Phoon, Convergence study of the truncated karhunen–loeve expansion for simulation of stochastic processes, Interna-
44 tional journal for numerical methods in engineering 52 (2001) 1029–1043.
45 [49] J. Chen, W. Sun, J. Li, J. Xu, Stochastic harmonic function representation of stochastic processes, Journal of Applied Mechanics 80 (2013).
46 [50] D.-C. Feng, Y.-P. Liang, X. Ren, J. Li, Random fields representation over manifolds via isometric feature mapping-based dimension reduction,
47 Computer-Aided Civil and Infrastructure Engineering 37 (2022) 593–611.
48 [51] J. Hurtado, A. Barbat, Monte carlo techniques in computational stochastic mechanics, Archives of Computational Methods in Engineering 5
49 (1998) 3–29.
50 [52] F. A. Bester, M. van den Heever, P. Kruger, S. Zeranka, G. Van Zijl, Benchmark structures for 3d concrete printing, in: International
51 Federation for Structural Concrete Proceedings of the 16th fib International Symposium, 2019.
52 [53] Y.-P. Liang, X. Ren, D.-C. Feng, Wind-induced probabilistic failure analysis of super large cooling tower with random field properties,
53 Journal of Wind Engineering and Industrial Aerodynamics (2023) 105274.
54 [54] J. Röwekämper, C. Sprunk, G. D. Tipaldi, C. Stachniss, P. Pfaff, W. Burgard, On the position accuracy of mobile robot localization based on
55 particle filters combined with scan matching, in: 2012 IEEE/RSJ International Conference on Intelligent Robots and Systems, IEEE, 2012,
56 pp. 3158–3164.
57 [55] K. Subrin, T. Bressac, S. Garnier, A. Ambiehl, E. Paquet, B. Furet, Improvement of the mobile robot location dedicated for habitable house
58 construction by 3d printing, IFAC-PapersOnLine 51 (2018) 716–721.
59 [56] N. Akai, Reliable monte carlo localization for mobile robots, Journal of Field Robotics 40 (2023) 595–613.
60 [57] C. Zeng, X. Chen, Y. Zhang, K. Gao, A structure-based iterative closest point using anderson acceleration for point clouds with low overlap,
61 Sensors 23 (2023) 2049.
62 [58] P.-C. Kung, C.-C. Wang, W.-C. Lin, A normal distribution transform-based radar odometry designed for scanning and automotive radars, in:
63 2021 IEEE International Conference on Robotics and Automation (ICRA), IEEE, 2021, pp. 14417–14423.

24

View publication stats

You might also like