Professional Documents
Culture Documents
STP164420210127
STP164420210127
ABSTRACT
The additive manufacturing (AM) technology laser-based powder bed fusion of
metals (PBF-LB/M) has reached technological maturity and is on the brink
of industrialization. High utilization must be achieved to operate PBF-LB/M
machines with high investment costs economically. However, reaching high utili-
zation contradicts the realization of short throughput times, schedule reliability,
and low work in process. In PBF-LB/M production planning, these metrics are
influenced by the nesting and scheduling of build jobs. To determine these
metrics and optimize their tradeoff, knowledge of build times of build jobs is
essential. Therefore, an estimation method for PBF-LB/M manufacturing time
is required. Due to the characteristics of the PBF-LB/M scanning process, the
geometry complexity of the parts in a build job influences the manufacturing
time. This influence is not considered by approaches from the literature available
for AM production planning. A novel approach for the estimation of PBF-LB/M
printing time is proposed to address this research gap. It includes a regression
model for the estimation of the part scanning build rate dependent on the sur-
face-to-volume ratio of the parts in a build job. The accuracy of the proposed
Copyright V
C 2022 by ASTM International, 100 Barr Harbor Drive, PO Box C700, West Conshohocken, PA 19428-2959.
ASTM International is not responsible, as a body, for the statements and opinions expressed in this paper. ASTM International
does not endorse any products represented in this paper.
Keywords
additive manufacturing, build time estimation, printing time estimation,
laser-based powder bed fusion of metals, laser powder bed fusion, production
logistics, production planning, nesting, scheduling
Introduction
Today’s rapidly evolving competitive landscape leads to demand for faster innova-
tion cycles and shorter, steeper value chains.1 Due to the possibility of producing
parts with different geometries in one build job2 without the necessity of specific
tools for a production task,3 additive manufacturing (AM) technologies such as the
laser-based powder bed fusion of metals (PBF-LB/M) offer a solution to meet this
demand.
High utilization is essential for operating capital-intensive machines, such as
PBF-LB/M systems, economically. The objective high utilization typically contra-
dicts the realization of short throughput times, high schedule reliability, and low
level of work in process. This tradeoff of these metrics of production logistics is
known as the scheduling dilemma in production logistics.4,5 Unlike batching and
scheduling in production planning for traditional manufacturing technologies, nest-
ing and scheduling cannot be addressed individually in AM to optimize this trade-
off.6 Build jobs with high packing density lead to increased utilization and thereby
low cost, resulting in long build times that jeopardize the fulfillment of due dates.7
For a given set of different parts to be produced in a sequence of build jobs,
many nesting and scheduling options exist. Hence, it is necessary to evaluate the
metrics of production logistics for a multitude of nesting and scheduling options.
For this purpose, a method to estimate the time for manufacturing of different build
job compositions is required in PBF-LB/M production planning. To estimate the
build times of many different build job variations, a method with low computa-
tional cost per build job is required. Nevertheless, the estimation model must
achieve sufficient accuracy to ensure planning security. Furthermore, the estimation
model must be easy to build for different machine and material combinations with
given possibilities of a typical PBF-LB/M producer, or generally applicable. Other-
wise, PBF-LB/M producers would have to go to great expense to build models for
the different material and machine combinations used in their production systems.
In the following sections, relevant approaches from the literature for PBF-LB/M
build time calculations are reviewed considering the stated requirements of high esti-
mation accuracy, low effort to build, low effort in use (in terms of computational
cost), and broad applicability. The result is shown in figure 1, followed by a more
detailed discussion of the reviewed publications.
block (error of 1.07%), and a complete build job (error of 8.56%). Furthermore,
computation time is required to calculate the scanning time of each voxel, and the
material occupancy of each voxel must be known. Thus, the computational cost for
using the model is evaluated as medium. The model is applicable for one machine
material combination for different geometries in one build job.8
where:
tbuild(Pi) ¼ total share of the build time of the i-th part of the build job,
NL ¼ number of layers,
Vtot ¼ total volume of the build job,
Stot ¼ total surface area,
Ni ¼ number of parts in the build job, and
Ssupp ¼ total surface area of the support structures.
The authors did not justify the choice of the linear-regression model and the
independent variables. Furthermore, the accuracy of the build time estimation was
not evaluated. For building the regression model, values for the independent and
dependent variables of 24 past build jobs are required. These data are usually avail-
able for a PBF-LB/M producer. Thus, the effort for building the regression model is
considered low. Once the regression formula is obtained, it is possible to apply it
easily with low computational cost to many different build job variations for one
machine material combination.9
Coating Time
In this work, the coating time is defined as the time between the end of the scanning
process of one layer and the start of the scanning process of the next layer. During
this time, a new powder layer is applied. The powder supply platform is then moved
up to bring more powder material into the build chamber, while the build platform
is lowered by the thickness of the layer. After that, the powder layer is distributed
by a coater. In this work, the time for coating one layer is emulated using the
approach of Dirks and Schleifenbaum,10 but a measurement during an exemplary
build job would also be possible. To calculate the total time for coating of one build
job, the coating time per layer, tCoat,Lay, is multiplied by the number of layers,
nLayers. The number of layers equals the z-height, hz, divided by the layer height,
hLayer.7 The z-height and number of layers are known after the orientation of those
parts in the virtual build space.
hz
tCoat ¼ tCoat,Layer nLayers ¼ tCoat,Lay (3)
hLayer
VSupport
tScan,Support ¼ (4)
V_ Support
The support volume is known from the CAD file after the automatic generation
of the support structures. The support build rate is retrieved from the emulated
data, but it is also possible to obtain it by measuring. Therefore, a build job with
only support structure is executed, and the printing time is measured. With exclu-
sively support structures in the build job, the part scanning time is zero. Thus, the
support scanning time is calculated by subtracting the measured coating time from
the printing time (cf. equation (2)). Then, the support scanning build rate is the
support scanning time divided by the support volume.
The scanning speed for the contour is usually lower than it is for the inner area.
Therefore, the average time for fusing material is assumed to be higher for parts
with a high surface-to-volume ratio. Furthermore, it is reasonable to assume that
more delays and jumps are necessary if the geometry is more complex. Those
assumptions are illustrated in figure 3.
The cross-section of a geometry with low complexity is shown on the left,
and the cross-section of a more complex geometry is illustrated on the right. The
ratio of the length of slow contour to fast hatch vectors is higher for the complex
geometry. Additionally, more jumps and delays are necessary to scan the more
complex geometry. As a consequence of those observations, parts with a high
surface-to-volume ratio, an operationalization of the geometry complexity, have
a lower overall scanning build rate respectively longer part scanning time per
scanning build rate is shown on the right. The geometry complexity, CPart, repre-
sented by the part surface, APart, to volume, VPart, ratio is used as the independent
variable. The part scanning build rate, V_ Scan , is defined as the dependent variable.
where:
APart
CPart ¼ :
VPart
The part volume, VPart, is divided by the part scanning build rate, V_ Scan;est: , that
is estimated using a regression function, freg(CPart), to calculate the scanning time as
part of the total build time estimation.
VPart
tScan,Part ¼ (7)
V_ Scan, est:
Thus, adding the time components results in the following formula for the esti-
mated printing time, tPrint,est.
VPart VSupport hz
tPrint,est: ¼ þ þ tCoat,Layer (8)
V_ Scan,est: V_ Support hLayer
Representative values for the dependent and independent variables are required
to build the regression model. The chosen 35 part geometries span a big surface-to-
volume ratio from 0.288 1/mm to 3.742 1/mm. The parameter sets for the two
materials IN718 and AlSi10Mg are considered to demonstrate the possibility of
building the regression model for different materials. To limit the effort for building
the regression model, data preparation is carried out for build jobs with one part
each. It is assumed that the resulting estimation bias is negligible because the jump
time of the laser spot from one geometry to another in build jobs with multiple
geometries is minimal compared with the overall printing time. This results in
70 build jobs, a sample size, respectively, of 35 for each material.
Part surface and volume are extracted from the CAD models of the parts. In this
work, the approach of Dirks and Schleifenbaum10 is used to emulate the part scan-
ning time. The values of the dependent variable part scanning build rate are calcu-
lated by dividing the respective part volume by the emulated part scanning time.
Alternatively, it is possible to obtain a data set of values for the dependent and inde-
pendent variables by measuring during build jobs in a PBF-LB/M production. How-
ever, the part scanning time must be obtained indirectly because measuring it directly
is not applicable. Therefore, the coating time and support scanning time are obtained
and subtracted from the printing time to calculate the part scanning time.
To consider possible nonlinear dependencies of the scanning build rate from
the geometry complexity, the MATLAB curve-fitting toolbox is used to define the
regression function. The fit of different regression functions is evaluated using the
coefficient of determination, R2. The function with the best fit, quantified by the
highest R2 value, is further used in the overall PBF-LB/M printing time estimation
model.
REGRESSION MODEL
Figure 5 shows the regression diagrams for IN718 and AlSi10Mg. The best curve fit
with a high R2 value is realized using a fractional rational regression function with
the following formula:
p
V_ Scan,est: ¼ freg ðCPart Þ ¼ (9)
CPart þ q
where:
p and q ¼ regression parameters.
For IN718 the R2 value is 0.851, and for AlSi10Mg it is 0.914. As expected, the
scanning build rate is higher for build jobs with low geometry complexity. This
result complies with the assumption of higher scanning times for parts with a com-
plex geometry because of a higher proportion of slow contour scanning, idle jump
times, and delay times. The regression curve shows a convex shape with a lower
sensitivity of the part scanning build rate for higher geometry complexity values.
Estimation Accuracy
With a low MAPE of 2.09%, the accuracy of the proposed PBF-LB/M printing time
estimation model is high. The achieved R2 is higher than for the scanning build rate
regression (0.851 for IN718 and 0.914 for AlSi10Mg). An explanation for this is a
balancing effect of positive and negative deviation of single-part scanning times in
one build job. As shown in figure 6, the deviation of the estimated and emulated
printing times is low across all 20 different build jobs. However, further research for
the proposed data-retrieval method of measuring the input variables (cf. fig. 2)
instead of emulation is required.
In this work, the printing time, without warm-up and cool-down times, is esti-
mated. If the model included the measurable and fix warm-up and cool-down
times, the absolute error is assumed to remain the same. Thus, the MAPE value of
the whole build time would be even lower than 2.09%.
Broad Applicability
A one-time effort is required to build the estimation model for one machine and
material combination, as in Fera et al.,7 Baumers et al.,8 and Rickenbacher et al.9
Therefore, further research should quantify the influence of contour and hatch
process parameters (scanning speed, hatch distance, layer thickness) on the
estimated part scanning build rate. If systematic influence is found and quantified,
the estimation model can be built for one material and adapted to others. Thus,
applicability for other materials would be realized. The attempt of Hällgren,
Pejryd, and Ekengren12 to project the build time to other materials by using the
ratio of maximum print speeds from material data sheets as a multiplier does not
deliver accurate results. Thus, further research should investigate an adaption of
their approach by considering material-specific scanning parameter sets for con-
tour and hatches.
Furthermore, the applicability of the built estimation model for different scan-
ning strategies is not investigated in this work and should be subject of further
research. Mugwagwa et al.13 find only minor variance in scanning times when using
different island scanning strategies for parts with small x-y-dimensions, while
the difference increases with larger x-y-dimensions. Masoomi, Thompson, and
Shamsaei14 report a 23% reduction of build time when using continuous scanning
strategy compared to sky-writing. To systematically investigate the influence of
scanning strategies, different part complexities, x-y part dimensions and scanning
strategies should be the subject of further research.
Conclusion
In this work, a model for estimating PBF-LB/M printing times was proposed. The
knowledge of accurately estimated printing times of differently nested build jobs is
necessary in PBF-LB/M production planning. Without this information, it is not
possible to achieve schedule reliability and optimize throughput times, machine uti-
lization, and work in process.
Relevant approaches from the literature were reviewed to identify suitable
approaches for adaption according to the following requirements: high estimation
accuracy, low effort for building the model, low effort for using the model, and
broad applicability. The reviewed approaches with availability for PBF-LB/M pro-
ducers do not consider the influence of the complexity of part geometries on the
part scanning build rate.
To address this research gap in the presented approach, the geometry complex-
ity of the parts in a build job is considered in a regression model for the estimation
of part scanning build rates. The geometry complexity is operationalized by the sur-
face-to-volume ratio of the parts in a build job. Hence, the surface-to-volume ratio
is used as the independent variable in the regression model to estimate the depen-
dent variable part scanning build rate. The regression model is part of the presented
model for estimating PBF-LB/M printing times for different build job variations.
The proposed approach achieves high estimation accuracy for PBF-LB/M
printing times with a MAPE of 2.09%. Further research should focus on improving
the adaptability of the proposed PBF-LB/M printing time estimation model so it
only needs to be built once per machine and material combination with respective
scanning parameter sets. Furthermore, the influence of different scanning strategies
ACKNOWLEDGMENTS
The authors would like to thank the Federal Ministry of Education and Research of
Germany for funding the research in the project Industrialization and Digitalization
of Additive Manufacturing for automobile series production (IDAM, grant number
13N15084).
References
11. J. Huang, Q. Qin, J. Wang, and H. Fang, “Two Dimensional Laser Galvanometer Scanning
Technology for Additive Manufacturing,” International Journal of Materials, Mechanics and
Manufacturing 6, no. 5 (2018): 332–336, https://doi.org/10.18178/ijmmm.2018.6.5.402
12. S. Hällgren, L. Pejryd, and J. Ekengren, “Additive Manufacturing and High Speed Machin-
ing—Cost Comparison of Short Lead Time Manufacturing Methods,” Procedia CIRP 50
(2016): 384–389, https://doi.org/10.1016/j.procir.2016.05.049
13. L. Mugwagwa, D. Dimitrov, S. Matope, and I. Yadroitsev, “Evaluation of the Impact of
Scanning Strategies on Residual Stresses in Selective Laser Melting,” nos. 5–8, The Inter-
national Journal of Advanced Manufacturing Technology 102 (2019): 2441–2450, https://
doi.org/10.1007/s00170-019-03396-9
14. M. Masoomi, S. M. Thompson, and N. Shamsaei, “Quality Part Production via Multi-Laser
Additive Manufacturing,” Manufacturing Letters 13 (2017): 15–20, https://doi.org/10.1016
/j.mfglet.2017.05.003