IOP Conference Series: Materials Science and Engineering

PAPER • OPEN ACCESS Related content

- Study on the electronic structure and
Optimization Study of Hydrogen Gas Adsorption hydrogen adsorption by transition metal
decorated single wall carbon nanotubes
on Zig-zag Single-walled Carbon Nanotubes: The P Modak, Brahmananda Chakraborty and
S Banerjee

Artificial Neural Network Analysis - Interaction of Methane with Single-Walled

Carbon Nanotubes: Role of Defects,
Curvature and Nanotubes Type
To cite this article: Nasruddin et al 2018 IOP Conf. Ser.: Mater. Sci. Eng. 333 012031 M. D. Ganji, M. Asghary and A. A. Najafi

- Hydrogen storage of capped single-walled

carbon nanotube via transition-metal
doping
Z. Y. Tian, Y. T. Liu, W. W. Wu et al.
View the article online for updates and enhancements.

This content was downloaded from IP address 152.118.231.228 on 22/04/2019 at 08:23

International Conference on Advanced Materials for Better Future 2017 IOP Publishing
IOP Conf. Series: Materials Science and Engineering 333 (2018) 012031 doi:10.1088/1757-899X/333/1/012031
1234567890‘’“”

Optimization Study of Hydrogen Gas Adsorption on Zig-zag

Single-walled Carbon Nanotubes: The Artificial Neural
Network Analysis

Nasruddin1*, M Lestari1, Supriyadi2, Sholahudin1

1
Department of Mechanical Engineering, Universitas Indonesia, Depok, Jawa Barat
16424, Indonesia
2
Department of Mechanical Engineering, Universitas Trisakti, Jakarta Barat 11440,
Indonesia

E-mail: nasruddin@eng.ui.ac.id

Abstract. The use of hydrogen gas in fuel cell technology has a huge opportunity to be applied
in upcoming vehicle technology. One of the most important problems in fuel cell technology is
the hydrogen storage. The adsorption of hydrogen in carbon-based materials attracts a lot of
attention because of its reliability. This study investigated the adsorption of hydrogen gas in
Single-walled Carbon Nano Tubes (SWCNT) with chilarity of (0, 12), (0, 15), and (0, 18) to
find the optimum chilarity. Artificial Neural Networks (ANN) can be used to predict the
hydrogen storage capacity at different pressure and temperature conditions appropriately, using
simulated series of data. The Artificial Neural Network is modeled as a predictor of the
hydrogen adsorption capacity which provides solutions to some deficiencies in molecular
dynamics (MD) simulations. In a previous study, ANN configurations have been developed for
77k, 233k, and 298k temperatures in hydrogen gas storage. To prepare this prediction, ANN is
modeled to find out the configurations that exist in the set of training and validation of
specified data selection, the distance between data, and the number of neurons that produce the
smallest error. This configuration is needed to make an accurate artificial neural network. The
configuration of neural network was then applied to this research. The neural network analysis
results show that the best configuration of artificial neural network in hydrogen storage is at
233K temperature i.e. on SWCNT with chilarity of (0.12).

1. Introduction
The most potential energy for future that environmental friendly with massive of quantity in this
universe is hydrogen [1]. However, the problem is about how to keep hydrogen gas on storage cover
up by single-walled carbon nano tube (SWCNT) as adsorbent [2, 3]. SWCNT is a structure of nano
tube with one carbon graphene builder, which is optimum in adsorbed hydrogen with inside and
outside of the nano tube as illustrated in Figure 1.

Content from this work may be used under the terms of the Creative Commons Attribution 3.0 licence. Any further distribution
of this work must maintain attribution to the author(s) and the title of the work, journal citation and DOI.
Published under licence by IOP Publishing Ltd 1
International Conference on Advanced Materials for Better Future 2017 IOP Publishing
IOP Conf. Series: Materials Science and Engineering 333 (2018) 012031 doi:10.1088/1757-899X/333/1/012031
1234567890‘’“”

Figure 1. The hydrogen adsorption in SWCNT with inside and

outside confirguration.

SWCNT rolled by three type rolling method as showed in Figure 2 [4]. It is result a different tube.
SWCNT that rolled by zig zag model was selected for this research because more effective to adsorb
hydrogen with it peaks form.

Figure 2. Step of SWCNT graphene sheet rolled into a tube [4]

With rolling zigzag method, SWCNT, as the adsorbent of hydrogen gas with various chilarity or
diameter of tube giving difference result on the adsorption capacity [5]. In the figure 1, the molecular
dynamic simulation of hydrogen adsorption by SWCNT is showed. Molecular dynamic simulation
(MDS) is a simulation method that used for simulating the adsorbed hydrogen by CNT on storage [6,
7]. Actually, the experiment and simulation has conducted an observation of the adsorption hydrogen
by SWCNT [3, 8, 9]. And there has shortcoming from both research methods. Then it is cover up by
artificial neural network (ANN). Neural network is a tool on MatLab software data apps that one of its
function is used for predict hydrogen storage capacity on different pressure condition precisely
without conducting simulation [10].

2
International Conference on Advanced Materials for Better Future 2017 IOP Publishing
IOP Conf. Series: Materials Science and Engineering 333 (2018) 012031 doi:10.1088/1757-899X/333/1/012031
1234567890‘’“”

In the previous research, ANN was modeled, as an effort to preparing ANN to be a smart
prediction tool that resulting good output. There is a configuration of ANN is needed [11]. So in the
previous research neural network configuration was observed [12]. The result is an ANN configuration
to predict hydrogen storage capacity called weight percent (wtp) without simulation was found. This
configuration is for three variances of hydrogen temperature storage which is 77k, 233k, and 298k
resulting difference configuration but with same optimum neuron that is neuron 10 [13]. The
configuration of ANN formed with 3 scale range variance for wtp data clarity, 2 type of training set
and validation set election variance, and 4 neuron election [14]. The best configuration that is chosen
from each temperature is the configuration that resulting wtp value closer to both the experiment and
simulation wtp resulted. The configuration that formed shown in Table 1 – Table 3.

Table 1. The error result with scale of 100-1000 and first election variant at temperature 77K.

(a). Error result of training output (b). Error result of validation output
RMS CV R2 RMS CV R2
N3 0.140146 2.414148 0.999436 0.338206 6.149633 0.996284
N5 0.140146 2.414148 0.999432 0.318123 5.784464 0.9967209
N7 0.112371 1.935687 0.999635 0.343534 6.246516 0.996686
N10 0.111936 1.928196 0.99964 0.332033 6.037397 0.996374

The focus of attention from Table 1 is the value of coefficient of determination parameter (R2) and the
best regression neuron that is neuron 10 i.e 0.999641. Then the graphs of target and validation data are
presented in Figure 3.

Figure 3. Regression result of training set and validation set at 77K[m3], (a). Training result with R2
value Test 0.9948, (b). Validation result with R2 value Test 0.97903.

The configuration for temperature of 233 K is shown in Table 2.

Table 2. The error result with scale of 100-10000 and first election variant at temperature 233K

(a). Error result of training output (b). Error result of validation output
RMS CV R2 RMS CV R2
N5 0.000957422 0.086474 0.998832 0.092853 9.040775 0.994739
N3 0.00375281 0.338953 0.995352 0.157781 15.37459 0.98192
N7 0.000830143 0.074978 0.998985 0.091785 9.814819 0.9949889
N10 0.000534122 0.048242 0.99935 0.190719 18.07104 0.974673

3
International Conference on Advanced Materials for Better Future 2017 IOP Publishing
IOP Conf. Series: Materials Science and Engineering 333 (2018) 012031 doi:10.1088/1757-899X/333/1/012031
1234567890‘’“”

It can be shown in Table 2 the same result as 77 K temperature, that the best regression neuron is
neuron 10 configuration. The graphs of target and validation data at temperature 233 K are presented
in Figure 4.

Figure 4. Regression result of training set and validation set at 233 K[m3], (a). Training result with
R2 value Test 0.99808, (b). Validation result with R2 value Test 0.99819.

The result for temperature 298 K is presented in Table 3 and Figure 5. The result of the best
configuration is also neuron 10.

Table 3. The error result with scale of 0-100 and second election variant at temperature 298K.

(a). Error result of training output (b). Error result of validation output
RMS CV R2 RMS CV R2
N3 0.055917 6.441731 0.996896 0.077435 8.337538 0.995263
N5 0.036758 4.230856 0.998657 0.066016 6.81065 0.996264
N7 0.030488 3.512508 0.999078 0.036559 3.904285 0.998822
N10 0.026574 3.060571 0.9993 0.120471 13.18335 0.984872

As the previous result from other temperature, on this temperature, neuron 10 showing it best
regression where the R square closer to 1 than the other neuron.

Figure 5. Regression result of training and validation at 298 K[m3], (a). Training result with R2
value of 0.99792, (b). Validation result with R2 value Test 0.99672.

4
International Conference on Advanced Materials for Better Future 2017 IOP Publishing
IOP Conf. Series: Materials Science and Engineering 333 (2018) 012031 doi:10.1088/1757-899X/333/1/012031
1234567890‘’“”

In this research the configuration has been tested to predict wtp for ideal pressure. Hydrogen
storage ideal pressure is the pressure that used in hydrogen storage in gas form. The pressure that
conditioned to keeping hydrogen gas on storage has range 0-120 bar according to the previous
research on hydrogen storage. More than 120 bar is limit closer to compressed tank pressure. It is
unsuitable to implement the compressed tank on this purpose vehicle that is light weight vehicle, such
as vehicle. Less than 0 bar or closer to -100 bar is the pressure on liquid gas storage. Hydrogen liquid
storage absolutely no effective implemented for land transportation like vehicle.
(0,12), (0,15), and (0,18) chirality of SWCNTs were observed to find the optimum chirality to
adsorb the hydrogen gas [15]. This chirality has used on previous research finding neural network
configuration. And now this configuration is implemented to find the optimum chirality on optimum
temperature in hydrogen adsorption.

2. Experimental
2.1. Computation method
With each best neural network configuration for each temperature, the testing method is used. On the
three variances of temperature and chirality that used in the previous research that is temperature 77K,
233K, and 298K with (0,12), (0,15), and (0,18) chirality of SWCNT[16]. The required pressure is
inputted with range 0-120 bar, the approachment of wtp resulted is analogically [15]. Data is grouped
according with neural network characteristic.

2.2. Training ANN on MatLab

After the data has grouped, the procees is training on MatLab with neural network tools. It is applied
for each chilarity on each temperature. For the final step is analyzing the optimum chirality on the
optimum temperature consider from wtp resulted from the pressures inputted equally for dependent
variables.

3. Result and Discussion

On chirality (0,12) that showing by Figure 6 (a), conditioning storage with temperature 77K wtp value
going to be constant after pressure given more than 50 bar and increasing on the pressure more than
140 bar. Meanwhile, on temperature of 233K, wtp value going steadily increases according to the
increasing of pressure value. And on 298K, the wtp is going to be constant after given pressure more
than 10 bar.

Figure 6. Weight percent (wtp) value with SWCNT; (a). (0,12) chirality, (b). (0, 18) chirality.

5
International Conference on Advanced Materials for Better Future 2017 IOP Publishing
IOP Conf. Series: Materials Science and Engineering 333 (2018) 012031 doi:10.1088/1757-899X/333/1/012031
1234567890‘’“”

Figure 6 (b) shows graphic formed by chirality (0,15), conditioning storage with temperature 77K
giving the wtp value highest than given by chirality (0,12). The weight percent (wtp) value chirality
(0,15) on the pressure more than 30 bar has getting wtp more than 6, while the wtp on chirality (0,12)
getting wtp value more than 6 on the pressure more than 70 bar. Then on temperature 233K wtp value
is going constant on the pressure more than 100 bar. And on 298K, the wtp is going increase with a bit
difference.
The last is graphic resulted by chirality (0,18) on Figure 8 showing that conditioning storage with
temperature 77K, wtp 6 has got on pressure just more than 20 bar, it is easier to got then on chirality
(0,15). Then on temperature 233K wtp value is going constant on the pressure more than 60 bar. And
on 298K, same graphic on chirality (0,15) is resulted on chirality (0,18) too, the wtp is going to
increase with a bit difference.
The results shown in Figure 6-8 are obtained from molecular dynamics simulations using the
LAMMPS program developed by the Department of Energy (DoE) United States [7, 17, 18]. For
setting the molecule in initial condition used packmol program [19].

4. Conclusion
Conditioning storage on 77K resulting wtp value that higher than other temperature. From Figure 6,
Figure 7, and Figure 8 showing that conditioning storage with temperature 77K as increasing the
chirality so the wtp value is easy to get with low pressure. In conclusion the recommended
temperature and chirality that optimum to adsorb hydrogen from this observation is use SWCNT
chirality 18 with storage temperature on 77K. Regression result of training set and validation set at
298 K[N3], (a). Training result with R2 value Test 0.99792, (b). Validation result with R2 value Test
0.99672.

References
[1] Srinivasa M S and Kumar E A 2014 J. App Thermal Eng. 72(2) 176-189
[2] Zhao L, Cheng J R and Huang D C 2004 Chinese J of Chem. Phys. 17(5) 572-576
[3] Barghi S H, Tsotsis T T, and Sahimi M 2014 Int. J. Hyd. En. 39(3) 1390-1397
[4] Dai H 2002 J. Surf. Sci. 500(1) 218-241
[5] Nasruddin, Kosasih E A, Budhi K, supriyadi, Zulkarnain I A 2016. IJTech 7(2) 264-273
[6] Peng L 2011 Atomistic Modeling of Hydrogen Storage in Nanostructured Carbons (PhD Diss-
University of Tennessee)
[7] Plimpton, S 1995 J. comp. phys. 117(1) 1-19
[8] Tarasov BP, Maehlen JP, Lototsky MV, 2003 J. alloys comp. 356 510-514
[9] Klechikov AG, Mercier G, Merino P, Blanco S, Cesar M and Alexandr VT 2015 J. Micr.
Mesop. Mat. 210 46-51
[10] Sholahudin S and Han H 2016 Energy. 115 1672-1678
[11] Malinova T and Guo Z 2004 Mat. Sci. Engi. 365(1) 219-227
[12] Yao H M, Vuthaluru H B, Tade M O and Djukanovic D 2005 Fuel. 84(12) 1535-1542
[13] Yıldız, Z. and H. Uzun 2015 J. Micr. Mesop. Mat. 208 50-54
[14] Tian q, Zhang Y, Wu Y, Tan Z 2009 Int. J. Hyd. En. 34(4) 1931-1936
[15] Supriyadi 2017 Optimization of Doping Elements on Carbon-based Nanostructures as
Hydrogen Storage Sedia: Molecular Thermodynamics Analysis, (PhD Diss-Universitas
Indonesia)
[16] Nasruddin, Kosasih E A, Supriyadi, Zulkarnain I A 2016 J. Mech. Eng. Mechatron. 1(01)
[17] Banerjee S and Puri I K 2008 Nanotechnology. 19(15) 155702
[18] Pei Q, Zhang Y, and Shenoy V 2010 Carbon. 48(3) 898-904
[19] Martínez L, Andrade R, Birgin E G AND Martínez J M. 2009 J. comp. chem. 30(13) 2157-
2164