<. Kanyilmaz et al.

Fig. 4 Parametric customization

of the nodes of the structure

full compression, full tension, and two asymmetric loadings material with penalization (SIMP) method (Bendsøe and
(Fig. 11). Sigmund 2004) is used.
The finite element model consisted of a solid discretized in The design domain is discretized so that only the material
a tetra-mesh generated by solidThinking Inspire. Its algo- densities equal to one remain and those equal to zero are
rithms use a combination of HyperMesh and Simlab for cancelled. The topology optimization is performed with the
meshing. The tetra-mesh configuration created in Optistruct/Hyperworks software (Altair Engineering 2017)
solidThinking Inspire presents first-order elements, which using the following inputs:
were fair and accurate for the initial set of analysis performed
in this study. More advanced analysis has been performed & Definitions of design and non-design spaces
using Hyperworks (Altair Engineering 2017). & Minimum compliance, as objective
& Material density distribution, as variable
2.2 Topology optimization of the tubular joint design & Volume fraction and maximum stress, as constraints
space
Our objective of topology optimization is to minimize the
Figure 12 shows our SDfAM process using a truss- compliance under volume and stress constraints (the joint
optimization analogy: the removal of material from a solid stiffness is maximized when the compliance is minimum).
body where it is not needed, based on stress analysis. This For this reason, as constraints, we used different volume frac-
approach is well-established in high-tech industries, such as tions of the design space (DS) that is described in Fig. 9. The
automotive and aerospace. In our case, the inner material dis- optimized solution was found by the SIMP method when the
tribution is considered as a design variable. The solid isotropic objective function and the constraints reached the

a Cross-section of the optimized joint b Free-form diagram and connection of the joint to the rest of the structure
geometry
Fig. 5 Hybrid tubular joint concept
Nature-inspired optimization of tubular joints for metal 3D printing

Fig. 6 Parametric joint builder tool developed in Grasshopper

convergence. Material density is cancelled in the zones where and “DS-L”, where the volume fractions of the design space
the stresses are low and placed for high stress concentration (DS) are, respectively, 50%, 40%, and 30%.
zones according to the loading path. Thus, the full solid core The entire structure is analyzed using SAP2000
of the joint is transformed into a sort of internal truss, resulting (Computer and Structures 2017) to estimate the design
in “internal natural stiffeners”. These joints with new shapes loads for the joints (Fig. 13). Ultimate limit state (ULS)
are then validated through finite element analysis. Moreover, and service limit state (SLS) combinations are consid-
some authors have demonstrated the possibility to also im- ered under code permanent and accidental loads for the
prove the fracture resistance at an increasing volume fraction structure considering Milan, Italy, as the location of the
during the topology optimization process (Da et al. 2018) and construction site. The joints were mainly loaded in dif-
to obtain a desired overall stiffness of the joints (Kang and ferent combinations of compression and tension actions.
Li 2017). Multiple combinations of axial loads observed in this
Table 3 shows the loading conditions considered for differ- analysis are considered during topology optimization.
ent shapes characterized by the degree of volume and stress The weighted compliance CW approach was considered as
constraints (σVon Mises < σyielding): full compression, full ten- the objective function. This global response is the weighted
sion, asymmetric. “Robust”, “Medium-weight”, and “Light- sum of the compliance of each subcase or load case (Altair
weight” solutions are called respectively “DS-R”, “DS-M”, University 2017), and it is defined by the following equation:

a Asymmetric joint (A)

b Symmetric joint (S)

Fig. 7 Typical connection model in Rhinoceros
 <. Kanyilmaz et al.

a Assymetric-unstiffened b Assymetric–benchmark c Symetric-unstiffened d Symetric - benchmark

Fig. 8 Unstiffened and benchmark models

number of control points, and from these, the new and

1
C W ¼ ∑i W i C i ¼ ∑ W i uTi f i smoother surfaces were built up with the original thicknesses.
2 i Finally, the stress values and distributions have been calcu-
where Wi is scale factor per each load case, uTi is displacement lated through finite element analysis. An “envelope load com-
vector per each load case, and fi: applied force per each load bination” is created by linear superposition of results to find the
case. maximum stress values. The different alternatives (DS-R, DS-
The shapes obtained from the topology optimization pro- M, DS-L, and a Benchmark) have been compared. The struc-
cess require a further post-processing. Before the 3D printing tural performance was monitored by a common reference point
stage, it is necessary to smoothen and clean the part surfaces to (RP) that is set at the intersection of structural profiles. This
achieve uniform shapes. The control of parameters, such as location showed the highest stress concentrations in the joint.
thickness and voids in the joints, is also essential to produce The densities are plot as a scale of colours in Optistruct
regular geometries. For this reason, a parametric post- (Hyperworks solver) to interpret the results. The range of values
processing tool is developed using Grasshopper (https:// is from 0 to 1 (Fig. 15) as an iso-contour plot. It represents the
www.grasshopper3d.com/) (Fig. 14). The geometry was absence or presence of material, respectively. An intermediate
intersected with cutting planes to generate the required value of 0.5 is selected because it shows defined shapes.

Fig. 9 CAD process of the typical symmetric joint (dimensions are in mm)
Nature-inspired optimization of tubular joints for metal 3D printing

900

800

700

600

500
Stress [MPa]

400

300

200

100

00
0,000 0,050 0,100 0,150 0,200 0,250 0,300 0,350
Strain

Ramberg-Osgood Stress-Strain Curve Fty Ftu True stress-strain

a Benchmark Steel b Stainless Steel 316L-1.4404

Fig. 10 Material properties

Consequently, the values above 0.5 are displayed, and the other traditional (and costly) way of preventing this situation is to
ones below 0.5 are filtered from the iso-contour plot. use welded stiffeners inside the geometry to decrease the
Each one of the TO solutions reached the convergence after stress concentrations in the critical zones. On the other hand,
several iterations (in the range of 25 to 28 iterations) to obtain the new solutions provide much better performance with their
the minimum compliance and the lowest percentage of violat- “naturally” stiffened shapes thanks to the material distribution,
ed constraints. The optimization resulted in stiffening the hol- which was optimized according to the different load paths.
low section joint. A higher percentage of DS results in more The yielding stress was not exceeded in the proposed so-
internal stiffeners. For the DS-R case, two sets of stiffeners are lutions. Figure 19 compares the structural performance in
formed: two vertical and two horizontal ones, creating an in- terms of maximum stresses and weight information among
ternal void in the joint. For the DS-M case, just two horizontal different topologies. The DS-R solution, with a slightly higher
stiffeners are generated. This also occurs to the DS-L case, weight, significantly drops the maximum stresses in the joint.
differing only on the thickness, because the two resulting hor- The light and benchmark joints perform similarly; however,
izontal stiffeners are thinner compared to those of the DS-M the fabrication of the benchmark joint would require many
case. The trend of results can be observed in both symmetric cuts and welding operations.
and asymmetric joint cases (Figs. 16, and 17). Although the In the asymmetric joint case (A), a similar trend of
new solutions (DS-R, DS-M) slightly weigh more than the results is obtained (Fig. 20). The unstiffened joint pre-
benchmark, they are expected to have a significantly better sented the highest stress concentrations along the mem-
structural performance with easier fabrication. bers’ intersection zone. As in the previous cases, the
The first estimation of stresses with a lower computational topology optimized joints with the “natural stiffener”
time was obtained by linear analysis. For this initial part, two formations represent a vast drop of the stress values
solutions were analyzed considering the weight: the robust and an improved distribution (Fig. 21).
case DS-R (50% DS) and the light case DS-L (30% DS) for A comparison of the linear analysis results of symmetric
the symmetric and asymmetric joints. Regarding the (S) and asymmetric (A) cases is presented in Table 4. Their
unstiffened symmetric joint cases (cross-section views are optimized shapes presented a similar stress decrease against
shown in Fig. 18), the most stressed regions correspond to their respective unstiffened cases. Both DS-R solutions
the intersection zones where the structural profiles converge. achieved a large percentage of reduction with respect to the
In that case, the material reached yield stresses because of the corresponding unstiffened cases: 81% and 82% less, respec-
high concentration of stresses. In current practice, the tively, for the symmetric and asymmetric cases. Both DS-L

Table 2 Material properties

CASE Material Young Poisson coef. Density ρ Yield strength Ultimate strength Strain at rupture Th. coef. Exp.
modulus E (MPa) υ (−) (t/m3) Fty (MPa) Ftu (MPa) εmax (%) (1/k)

Untiffened 200 Steel 0.290 7.870 380 615 25 14.700

E + 03 E − 09 E − 06
Benchmark Steel 200 0.290 7.870 380 615 25 14.700
E + 03 E − 09 E − 06
Solid (to be optimized SS 316 L 180 0.290 7.900 380 485 30 17.300
and printed) E + 03 E − 09 E − 06
 <. Kanyilmaz et al.

Fig. 11 Loading conditions used

in the simulations

a Full compression b Full tension c Converted load pattern 1 d Converted load pattern 2

cases reached an important stress decrease compared with the 3 Tests by non-linear failure analysis
unstiffened cases: 76% and 78% less, respectively, for
the symmetric and asymmetric. Despite the similarities The new joint proposals have been tested using non-linear fail-
in terms of shape and weight, the DS-L solutions pre- ure analysis. Their behaviour has been compared among them
sented an improved linear behaviour with respect to the and with the benchmark joint. Plastic deformations were in-
benchmark joint: For example, the unstiffened joint duced in the joints under different loading conditions by load-
stress reduction reaches 76%, with the symmetric sym- ing them until failure. Hence, the collapse situation was predict-
metric (S) Vol = 30% DS case and 69% with the ed by determining which zones would be the first to reach the
benchmark symmetric (S) case. ultimate strain. From the four load cases used in the previous

Simply supported beam Tubular joint

Initial geometry
and boundary
conditions

Design domain
discretization

Densities

0< ≤ ≤ 1

= 0

Final material
distribution

Fig. 12 Topology optimization scheme process following the (SIMP) method