ISSN (Print) : 2319-8613

ISSN (Online) : 0975-4024 Gabriel Awuku Dzukey et al. / International Journal of Engineering and Technology (IJET)

Process Parameter Optimization for

Selective Laser Melting of 316L Stainless
Steel Material using Taguchi’s Statistical
Design of Experiment Procedure
Gabriel Awuku Dzukey1, Ke Yang2
College of Mechanical and Electrical Engineering
Hohai University, Changzhou, Jiangsu Province, China
gadzukey@[Link], dzukeyg@[Link]; yangke_hhuc@[Link]
Abstract—Selective Laser Melting (SLM) is a type of additive manufacturing or 3D printing technology
that makes use of a high-power laser beam to completely melt powder material layer upon layer which
solidifies to form the desired part or component. In this experimental study, a careful selection of SLM
process parameters (laser power, scan speed, hatching distance, and layer height) was done for the selective
laser melting of 316L stainless steel powder material to produce samples with maximum density. A
statistical procedure was carried out on the experimental data using the Taguchi’s design of experimental
technique to optimize the carefully selected process parameters. The obtained results were analyzed using
analysis of variance (ANOVA) and signal-to-noise (S/N) ratio with the help of Minitab 18 statistical software
to determine the optimal parameters and a regression model was also established. The regression model
indicates a linear relationship between relative density and the process parameters. Density value higher
than 95% was obtained from this experimental study.
Keywords: Taguchi method, SLM, Optimization, 316L Stainless Steel
I. INTRODUCTION
Selective laser melting (SLM) is a type of Additive manufacturing (AM) or 3D Printing process that is currently
and rapidly being industrialized [1] because it is one of the most promising AM technologies allowing
manufacturing of metallic parts from a loose powder. By completely melting the metallic powder; the particles
are attached together to produce a sheet of molten powder layer-upon-layer which solidifies to form the part. This
manufacturing method is said to be best suited to produce complex and thin-walled components. Its characteristics
are really of great advantage to most industries since 3D printings give more freedom of design and the opportunity
to create optimal and lightweight components. Also, the SLM process and most of the 3D printing processes have
a common characteristic of being eco-friendly, if compared with conventional machining.
Grade 316L stainless steel is the low carbon version of the 316-standard molybdenum-bearing grade and is
immune from sensitization (grain boundary carbide precipitation). Thus, it is extensively used in heavy gauge
welded components (over about 6mm), a property suitable for SLM. It is widely used in various applications such
as Marine applications, Construction, Medical implants (including pins, screws and orthopedic implants like total
hip and knee replacements), Fasteners etc. This imposes some specific requirements on the steel material
including, wear resistance, hardness, high compressive strength, lightweight, economical material cost etc., [2].
Since SLM have some characteristics that meet the above specific requirements, there is the need find out the
quality of 316L stainless steel manufactured products using SLM.
Despite these several advantages that SLM provides, parts manufactured with this process usually present
porosity. Porosity is caused when non-appropriate parameters are used within the build-up process. Elevated levels
of porosity lead to a decrease in the relative density value of the SLM pieces, being the cause of poor mechanical
properties or even the fracture of it. The ability to produce complex parts is not enough, it must be complemented
with high performance and high mechanical properties. Therefore, improvement of the relative density which will
decrease the porosity of samples formed by SLM has become the main focus of many experts [3]-[5]. In SLM a
high-power laser is used to melt a powder material layer by layer to form a solid part. The build process is
controlled by changing the process parameters such as layer thickness, laser power, scanning speed and hatching
distance which all affect the build rate and properties of the finished part. For each SLM processed material, a
specific range of parameter combinations exists that results in sufficient quality, such as density and desired
surface roughness. There is, therefore, the need for an experiment to be conducted to determine a suitable
parameter set for each material (in this case SS316L) that will result to produce parts with a very high relative
density value.

DOI: 10.21817/ijet/2019/v11i1/191101014 Vol 11 No 1 Feb-Mar 2019 6

ISSN (Print) : 2319-8613
ISSN (Online) : 0975-4024 Gabriel Awuku Dzukey et al. / International Journal of Engineering and Technology (IJET)

The Taguchi method is referred to in literature as one of the most powerful tools for the statistical design of
experiments, which considers multi-factors [6]-[9]. It makes use of some representative points to be investigated
from a comprehensive test in the form of orthogonal arrays which are highly fractionalized factorial layouts,
reducing the number of trials, shorten experiment time and reduce cost. In this paper, the influence of processing
parameters on the density of SLM has been studied. The Taguchi, Analysis of Variance (ANOVA) and regression
analysis methods, which are statistical experimental procedures, were used for the analysis and optimization of
the SLM parameters.
II. EXPERIMENT
A. Experimental Equipment
The experiments were carried out using the SLM 260 3D printer manufactured by the Jiangsu Jiuyi Machinery
Co., Ltd., China, which can be represented by the schematic diagram in Fig. 1. The main parameters of the SLM
260 printer are as follows: continuous Yb-fiber laser of 0.4KW maximum power, 260× 260×250mm (L×W×H)
maximum molding size, optical system: HurrySCANLAB 30 with dynamic focus and spot diameter between 80-
120um, construction speed: 50-1000mm3/min and of layer thickness between 0.02-0.1mm.

Fig. 1. Schematic diagram of SLM [4]

TABLE. I. Abbreviations

Abbreviations Meaning
k the relative density of the part under consideration (SS316L)
ma mass of the part in air
mfl mass of the part in fluid
ρfl= density of the fluid (density of water =0.9982g/cm3)
ρp the density of the part under consideration (density of SS316L=7.9g/cm3)
S/N signal-to-noise
n number of tests
yi measured characteristics (density) from experiments
ANOVA analysis of variance
Sm an average of the squares of the sums
SA the sum of squares correlated to the control factor A
ST the sum of squares of the variance
SE the sum of squares of the errors correlated to all controlled factors
N number of samples in each group
DF degree of freedom
SS sum of squares
MS mean square
B. Experimental Material
The material composition of 316L Stainless Steel powder material is shown in Table II below in percentage
(%). Carbon (C) content being 0.03 maximum, Manganese (Mn) 2.0 maximum, Phosphorus (P) 0.045 maximum,
Sulphur (S) 0.03 maximum, Silicon (S) 0.75 maximum, Chromium (Cr) in a range of 16.0-18.0, Nickel (Ni) 10.0-
14.0, Nitrogen constitutes up to 0.10 and the balance constitutes Iron (Fe).

DOI: 10.21817/ijet/2019/v11i1/191101014 Vol 11 No 1 Feb-Mar 2019 7

ISSN (Print) : 2319-8613
ISSN (Online) : 0975-4024 Gabriel Awuku Dzukey et al. / International Journal of Engineering and Technology (IJET)

TABLE II. Chemical Composition of 316LStainless Steel Material

C Mn P S Si Cr Ni Mo N Fe
≤0.030 ≤2.00 ≤0.045 ≤0.030 ≤0.75 16.00~18.00 10.00~14.00 2.00~3.00 ≤0.10 Balance

C. Taguchi Method
In accordance with the Taguchi method, the procedures are first to identify and select proper quality
characteristics according to experimental analysis. Secondly, the selection of the appropriate orthogonal array and
assigning of design parameters to conduct the experiment, analysis of the experimental results using S/N and
ANOVA analyses which are statistical procedures embedded in Minitab, a statistical software package. Finally,
the optimal process parameter can be obtained using Minitab [10].
D. Selection of Process Parameters
The process parameters were determined according to the Taguchi method using L9 orthogonal array with
four (4) factors (laser power, scanning speed, hatching distance, layer height) and three (3) levels as shown below
in Table III.
Table III. Factors and Levels

Parameters Level 1 Level 2 Level 3

Laser Power(P)/W 200 240 280
Scan speed(s)/mm/s 400 500 600
Hatching distance(t)/µm 50 60 70
Layer height (h)/ µm 40 50 60
E. Recording of Experimental Data
The samples with the cubic dimension of 10mm × 10mm × 10mm are printed on steel substrate of dimension
100mm × 100mm × 20mm using the parameters according to the L9 orthogonal array. The oxygen content in the
forming or working chamber of the SLM 3D printer was kept below 0.1%. The relative density measurements of
the 316L stainless steel samples were based on the Archimedes principle. The mass ma and mfl which represent
the masses of the SLM-formed 316L stainless steel samples in air and distilled water respectively are measured
with a high-precision electronic balance. The calculations for the samples under consideration follows the equation
(2) below:
𝜌 𝜌 Eq.1

Where:
𝑘 100 Eq.2

F. Experimental Calculation
1) Calculation of S/N: “Nominal is best”, “smaller is better” and “higher is better” are the three main categories
for performing S/N ratio analysis with regards to performance characteristics. In this study density is the criterion
and the desire is to obtain the highest relative density value. Therefore, the factors with the highest S/N ratio
values in this experiment gives the best performance. In view of this, the standard requires that among the three
categories,” higher is better” characteristic-form be used in this case [6]. Below is the formula in equation (3);
10 log ∑ Eq.3
2) ANOVA Calculation: The analysis of variance (ANOVA) is performed on the data to study the contribution
of all controllable factors and also to determine which factor has the most significant effect on the relative density
of the samples produced using SLM. The equations below were used to calculate some of the parameters used in
ANOVA [7].
∑
𝑆 Eq.4
∑
𝑆 𝑆 Eq.5
𝑆 ∑ 𝜂 𝑆 Eq.6
𝑆 𝑆 ∑𝑆 Eq.7
‘A’ is representation for the various factors; laser power, scanning speed, hatching distant and layer height.

DOI: 10.21817/ijet/2019/v11i1/191101014 Vol 11 No 1 Feb-Mar 2019 8

ISSN (Print) : 2319-8613
ISSN (Online) : 0975-4024 Gabriel Awuku Dzukey et al. / International Journal of Engineering and Technology (IJET)

III. RESULTS AND DISCUSSION

A. S/N Analysis
Table IV and Table V show the results of S/N ratio response analysis which indicates the influence of the
control parameters (laser power, scan speed, hatching distance, layer thickness) on the relative density values of
the SLM printed samples. The difference between the maximum and minimum values of the mean of S/N ratios,
also known as the range (delta) was used to determine parameters with the strongest influence [11]. In other words,
the larger this difference (delta) value, the higher the influence of the control parameter. According to the analysis
in Table V, among all other factors, Laser power was the most influential processing parameter and the strength
of influence of the factors decreasingly ranges from laser power, layer height, scan speed, and hatching distance.
Fig. 2 and 3 indicate the main effects plots for the S/N ratio and mean values of relative density respectively.
1) Influence of Laser Power on the Relative Density: From Fig.2, it can be seen that an increase in the Laser
power results in increasing relative density value at different speeds which are indicated by the rise in the S/N
ratio value. The increase is faster between laser powers of 240 - 280W than between 200 - 240W. The rising laser
power led to the right amount of energy (volumetric energy density) required to properly melt the powder bed,
hence the relative density happens to increase with laser power.
2) Effect of Scanning Speed on the Relative Density: The relative density rises with rising scanning speed
up to the maximum when the scanning speed is 500mm/s and after declines with increasing relative density. The
scanning speed and laser power have a very close relationship and just as the laser power affect the amount of
energy required in melting the powder bed, so it is with the scanning speed. Considering the same laser power,
lowering the scanning speed will cause an increase in the Volumetric Energy Density (VED), since the time spent
in a spot is reduced and vice versa. Less or too low VED values causes improper melting of the powder bed
leading to the presence of pores in the built samples and lower density values. Likewise, too high VED values
lead to evaporation of powder bed causing pores as well. In this case, the effect of the scanning speed of 400-
600mm/s among other factors has little influence on the relative density values. Between the speed of 400-500mm
/s, the relative density value increases from 88.09-8.47% and decrease by 1.06% when the speed is 500-600mm/s.
3) Effects of the Hatching Distance on the Relative Density: The density changed increasingly but with a
very small percentage of almost insignificant values with increasing hatching distance (a total change of 0.25%
relative density). From Table V, it can be seen that among other factors hatching distance has the least influence
on the relative density in this experiment. A very small scan distance or hatching distance can cause overlapping
of laser beam resulting in a very high concentrated laser energy at the overlapping area causing over burning of
the powder bed and this affects the density of the samples. Likewise, a wider hatching distance results in improper
melting of the powder between laser beam tracks, which also affects the density of the printed samples.
TABLE IV. Relative Density and S/N Ratio obtained using Taguchi’s L9 Orthogonal Array

No. Laser Scan Hatching Layer Relative S/N Ratio

power(W) speed(mm/s) distance(µm) thickness(µm) Density (%)
1 200 400 50 40 83.52 38.43581
2 200 500 60 50 84.02 38.48765
3 200 600 70 60 80.95 38.16434
4 240 400 60 60 85.58 38.64745
5 240 500 70 40 88.11 38.9005
6 240 600 50 50 86.69 38.75938
7 280 400 70 50 95.16 39.56909
8 280 500 50 60 93.27 39.39484
9 280 600 60 40 94.58 39.51599

DOI: 10.21817/ijet/2019/v11i1/191101014 Vol 11 No 1 Feb-Mar 2019 9

ISSN (Print) : 2319-8613
ISSN (Online) : 0975-4024 Gabriel Awuku Dzukey et al. / International Journal of Engineering and Technology (IJET)

TABLE V. Response Table for Signal to Noise Ratios- Larger is better

Level Laser Power (W) Scan Speed (mm/s) Scan Distance (µm) Layer Thickness(µm)
1 38.36 38.88 38.86 38.95
2 38.77 38.93 38.88 38.94
3 39.49 38.81 38.88 38.74
Delta 1.13 0.11 0.02 0.22
Rank 1 3 4 2

Fig.2. Main effect plots for S/N ratio

Main Effects Plot for Means

Data Means
Laser power Scan speed Scan distance Layer thickness

94

92
Mean of Means

90

88

86

84

82
200 240 280 400 500 600 50 60 70 40 50 60

Fig.3. Main effects plots for means of Relative Density

B. ANOVA Analysis
The ANOVA analysis was done to find the process parameters significantly influencing the relative density
value of the SLM printed samples. This was done for a confidence level of 95%, which is for a significance level
of α=0.05. Table VI and Table VII show the analysis of variance results using Minitab Statistical Software package,
which makes use of the formulas in Equations (4)-(7).

DOI: 10.21817/ijet/2019/v11i1/191101014 Vol 11 No 1 Feb-Mar 2019 10

ISSN (Print) : 2319-8613
ISSN (Online) : 0975-4024 Gabriel Awuku Dzukey et al. / International Journal of Engineering and Technology (IJET)

As it can be seen in Table VI, the test of significance cannot be performed because P-values could not be
calculated and this is because the error term has a zero (0) Degree of Freedom (DOF or DF). According to statistics,
in cases such as this, pooling of insignificant factors should be considered, this adds them to the error term thereby
making the DOF of the error term non-zero, but a factor is pooled when it fails the test of significance.
Unfortunately, the test of significance can only be done when the error term has nonzero DOF. As a rule, even if
a test of significance is not possible, pooling should be considered by starting with pooling a factor if its influence
(using S to compare instead of P-value) is 10 % or lower than the most influential factor [12]. This rule was
applied to recalculate the ANOVA terms in Table VII with the help of the Minitab Software. The P values for
Laser power and layer height were less than 0.05 and are therefore the significant factors influencing the response
(relative density). The scanning speed has little effect.
TABLE VI. Analysis of Variance before pooling

Factors DF Seq SS Contribution Adj SS Adj MS F-Value P-Value

Laser power 2 205.013 95.12% 205.013 102.507 * *
Scan speed 2 1.730 0.80% 1.730 0.865 * *
Scan distance 2 0.115 0.05% 0.115 0.058 * *
Layer thickness 2 8.672 4.02% 8.672 4.336 * *
Error 0 * * * *
Total 8 215.531 100.00%
TABLE VII. Analysis of Variance after pooling

Source DF Seq SS Contribution Adj SS Adj MS F-Value P-Value

Laser power 2 205.013 95.12% 205.013 102.507 1775.52 0.001
Scan speed 2 1.730 0.80% 1.730 0.865 14.99 0.063
Layer thickness 2 8.672 4.02% 8.672 4.336 75.10 0.013
Error 2 0.115 0.05% 0.115 0.058
Total 8 215.531 100.00%
C. Regression Analysis
A multiple linear regression equation, which models a relationship between the process parameters and the
response was also developed using Minitab 18 statistical software package. Table VIII indicates the results of
ANOVA regression analysis, it can be seen that regression coefficient (R-sq) was 95.65% which is an indication
of the change in the value of independent variables (process parameters) corresponding to the unit change in the
dependent variable (relative density). In other words, 95.65% of the variability can be explained by the model.
The positive coefficient associated with laser power in the regression equation below suggests that the relative
density increases with increasing laser power likewise the negative coefficient of the scan speed and layer
thickness suggest increasing scanning speed and layer thickness decreases the relative density of the SLM printed
samples.
The regression equation is as follows;
𝑅𝑒𝑙𝑎𝑡𝑖𝑣𝑒 𝐷𝑒𝑛𝑠𝑖𝑡𝑦 60.51 0.1438𝐿𝑎𝑠𝑒𝑟 𝑝𝑜𝑤𝑒𝑟 0.00340𝑆𝑐𝑎𝑛 𝑠𝑝𝑒𝑒𝑑 0.1068𝐿𝑎𝑦𝑒𝑟 𝑡ℎ𝑖𝑐𝑘𝑛𝑒𝑠𝑠 Eq.8
TABLE VIII. ANOVA Regression Analysis

Source DF Adj SS Adj MS F-Value P-Value

Regression 3 206.147 68.716 36.61 0.001
(Significant)
Error 5 9.385 1.877
Total 8 215.531
R-sq=95.65% R-sq(adj)=93.03%

DOI: 10.21817/ijet/2019/v11i1/191101014 Vol 11 No 1 Feb-Mar 2019 11

ISSN (Print) : 2319-8613
ISSN (Online) : 0975-4024 Gabriel Awuku Dzukey et al. / International Journal of Engineering and Technology (IJET)

D. Optimization
An optimization study was performed to find out the optimal processing conditions at which the desirable
(maximum) relative density values of the SLM printed samples can be accomplished [13]. The aim is to obtain
the maximum relative density without any limitation on the process parameters [9]. Minitab 18 was used to
analyze the experimental data. As shown in Table IX, each process parameter and the response (relative density)
have been given a specific goal. By using Minitab, it is required that the value of the response should be as large
as possible (usually taking values impossible to achieve) to obtain the optimized parameters and or optimum
solutions based on the optimized parameters. The optimum solutions obtained using the Minitab software are
displayed in Table X, which indicates that the optimized parameter set of the printed SLM samples are: 280W
laser power, 500mm/s scan speed and 40 µm layer thickness.
TABLE IX. Criteria for Numerical Optimization

Name Goal Lower Limit Upper Limit

Laser Power (W) in range 200 280
Scan speed(mm/s) in range 400 600
Hatching distance (µm) in range 50 70
Layer thickness (µm) in range 40 60
Relative Density (%) or response maximize 80.95 target=100
TABLE X. Optimal Solution as obtained by Minitab 18 for Relative Density

Laser Scan Layer Relative Composite

Solution power speed thickness Density Fit Desirability
1 280 500 40 95.5667 1.00000 (selected)

2 280 500 50 95.4533 1.00000

3 280 400 40 95.1867 1.00000

4 280 400 50 95.0733 0.99390

5 280 600 40 94.5067 0.95402

IV. CONCLUSION
This experimental study made use of Taguchi’s Design of Experiment method to estimate the optimal
processing parameters in Selective laser melting (SLM) of 316L stainless steel material with an emphasis on
obtaining highly dense formed products. The analysis of the experimental data was performed by varying the
process parameters up to three levels. The significant parameters contributing to the relative density of formed
SLM samples were determined using ANOVA and S/N ratio analysis, a mathematical model which can be used
to predict relative density values was also proposed by the help of Minitab 18 statistical software package. Within
the scope of this research work, the following conclusion among other factors can be drawn:
1. The optimum levels of process parameters for maximizing the relative density were determined. The optimal
condition for printing highly dense (above 95% relative density) SLM product was found out to be Laser power
=280W, Scan speed= 500mm/s, Layer thickness (layer height) = 40µm and a Hatching distance around 40-70
µm or using an acceptable working range.
2. In accordance with the results obtained from the statistical analysis, it was found out that, the Laser power was
the most significant parameter or factor affecting the density with a percentage contribution of 95.12 %.
3. The regression analysis performed showed that the linear multiple regression model established demonstrated
a very good relationship between process parameters and the response (relative density) with a high regression
coefficient (95.65%).
4. Minitab Statistical software package can be used for optimizing the SLM process parameters and determining
the corresponding optimum processing conditions.
ACKNOWLEDGMENT
The contribution of the Jiangsu Jiuyi Machinery Co., Ltd, Kunshan-China, where the SLM samples were
printed for this experimental study, is greatly appreciated.

DOI: 10.21817/ijet/2019/v11i1/191101014 Vol 11 No 1 Feb-Mar 2019 12

ISSN (Print) : 2319-8613
ISSN (Online) : 0975-4024 Gabriel Awuku Dzukey et al. / International Journal of Engineering and Technology (IJET)

REFERENCES
[1] Spears T.G. and Gold S.A., (2016). ‘In-Process Sensing in Selective Laser Melting (SLM) Additive Manufacturing’. Integrating
Materials and Manufacturing Innovation, a Springer Open Journal. DOI 10.1186/s40192-016-0045-4. Cited 10 February 2018 from
[Link]
[2] (2004) AZO Materials website ‘Stainless Steel - Grade 316L - Properties, Fabrication and Applications (UNS S31603)’, [Online].
Available: https
[3] Gu H., Gong H., Pal D., Rafi K., Starr T., Stucker B.,(2013), Influences of Energy Density on Porosity and Microstructure of Selective
Laser Melted 17-4PH Stainless Steel Cited 17 February 2018 from [Link]
[Link]
[4] Sun J., Yang Y., Wang D., Parametric optimization of selective laser melting for forming Ti6Al4V samples by Taguchi method. Optics
& Laser Technology 49 (2013) 118–124
[5] Laakso, P.; Riipinen, T.; Laukkanen, A.; Andersson, T.; Jokinen, A.; Revuelta, A.; Ruusuvuori, K. Optimization and simulation of SLM
process for high-density H13 tool steel parts. Phys. Procedia 2016, 83, 26–35.
[6] Canel T, Kaya A.U, Celik B. Parameter optimization of nanosecond laser for micro drilling on PVC by Taguchi method. Optics & Laser
Technology 2012; 44:2347–53.
[7] Yang DX, Li XY, He DY, Nie ZR, Huang H. Optimization of weld bead geometry in laser welding with filler wire process using
Taguchi’s approach. Optics Laser Technology 2012; 44:2020–5.
[8] Li CH, Tsai MJ, Yang CD. Study of optimal laser parameters for cutting QFN packages by Taguchi’s matrix method. Optics & Laser
Technology 2007; 39:786–95.
[9] Benyounis K.Y., Olabi AG, Hashmi MSJ. Multi-response optimization of CO2 laser-welding process of austenitic stainless steel. Optics
& Laser Technology 2008; 40:76–87.
[10] Yang W H, Tarng YS. Design optimization of cutting parameters for turning operations based on the Taguchi method. Journal of
Materials Processing Technology 1998; 64:122–9.
[11] Kumar S, Choudhary A. K. S., Anwar J., and Sharma V., Optimization of Process parameters in Direct Metal Deposition Technique
using Taguchi method. International Journal of Mechanical Engineering and Technology, 7(3), 2016, pp. 225–
[Link]://[Link]/[Link]?JType=IJMET&VType=7&IType=3
[12] Roy R. K., 2001, Design of Experiments Using the Taguchi Approach: 16 Steps to Product and Process Improvement, John Wiley &
Sons, Inc. (US) - ISBN 0-471-36101-1. (211).
[13] Eltawahni HA, Olabi AG, Benyounis K.Y. Effect of process parameters and optimization of CO2 laser cutting of ultra-high-performance
polyethylene. Materials & Design 2010; 31:4029–38.

AUTHOR PROFILE
Gabriel Awuku Dzukey is currently studying for an MSc. in Mechanical Engineering, at the College of
Mechanical and Electrical Engineering, Hohai University, Changzhou Campus, China. He studied for his BSc. in
Agricultural Engineering from the Kwame Nkrumah University of Science and Technology, Kumasi-Ghana
(2015). His current research interest is in Laser Additive Manufacturing.
Ke Yang is currently a Professor at the College of Mechanical and Electrical Engineering, Hohai University,
Changzhou, China. He holds a Ph.D. from the University of Science and Technology, Wuhan, China (2009). His
area of research includes Metal Material, Welding and Metallurgy, and Surface Engineering.

DOI: 10.21817/ijet/2019/v11i1/191101014 Vol 11 No 1 Feb-Mar 2019 13