Modeling and validation of the electromechanical behavior of highly stretchable cnt/polymer composites through network-based percolation model Seong-Won Jin1)**, Myeong-Seok Go**, Jae Hyuk Lim1)*, and Youngu Lee2) 1) Department of Mechanical Engineering, Jeonbuk National University, 567 Baekje-daero, Deokjin-gu, Jeonju-si, Jeollabuk-do 54896, Republic of Korea 2) Department of Energy Science & Engineering, Daegu Gyeongbuk Institute of Science and Technology (DGIST), Daegu 42988, Republic of Korea * Author to whom any correspondence should be addressed. **1st author equally contributed Abstract In this study, a program was developed to predict the electromechanical behavior of highly stretchable CNT-polymer composites in which the electrical properties change in response to uniaxial deformation. RVE was generated by randomly distributing CNTs in a polymer matrix using the Monte-Carlo simulation algorithm. In the case of tensile strain, the CNT and RVE configurations and network models were updated considering the tension and shrinkage by Poisson's ratio, and the electrical conductivity was calculated. Trade-off studies were conducted on RVE size, geometric periodicity of CNT fibers, and tensile strain to examine the appropriateness of the proposed methodology. In addition, the electromechanical behavior of CNT mixtures with different aspect ratios was confirmed and compared with the experimental data. When the size of RVE was more than three times larger than the length of CNT and CNT 1 with different aspect ratios were mixed, it was similar to the experimental data. This confirms that the proposed model adequately predicts the electrical conductivity of the uniformly dispersed CNT-polymer composite. Keywords: Carbon nanotubes, Electrical conductivity, Percolation threshold, Modified nodal analysis, Monte-Carlo simulations, Periodic conditions * Corresponding author: jaehyuklim@jbnu.ac.kr 2 1. Introduction Demand for flexible and stretchable strain sensors is increasing in various fields, such as structural health monitoring and soft robotics. One of the materials for highly stretchable strain sensors is carbon nanotube (CNT) based polymer composites [1]. Polymer materials are lightweight and have excellent mechanical properties but poor conductivity. However, the conductivity of polymer composites is greatly improved when carbon nanofillers such as carbon black, CNT, and graphite nanoplatelets with high conductivity are added [2-4]. The enhancement of electrical conductivity can explain by direct contact and the formation of a percolation network based on tunneling adjacent conductive fillers. Here, the tunneling effect is a quantum phenomenon that allows electrical conduction across isolating barriers between two highly conductive fillers [5]. When filler with a sufficient amount is evenly dispersed in the polymer, the overall conductive pathway begins to form. The amount of filler required to generate such a pathway is called the percolation threshold. The percolation threshold is affected by aspect ratio, tunneling mechanism, and physical contact. At the percolation threshold, the electrical conductivity of the nanocomposite rises rapidly, so research is being conducted to form it at a low fiber volume ratio [6]. In particular, low aspect ratio fillers such as carbon black cause percolation at a fiber volume fraction of 10 to 16% [7]. On the other hand, nanowires and CNTs have a high aspect ratio of 20 to 10,000 [8] and a one-dimensional conductive structure, which efficiently induces percolation even at a low fiber volume fraction of 0.2 to 1.5% [9-11]. In addition, it is possible to secure high electrical conductivity by maintaining percolation for a long time, even when a high strain rate is applied [5,6]. For developing these highly conductive materials, many studies have been conducted on analytical, numerical, and experimental approaches to calculate the equivalent electrical 3 conductivity of carbon nano-filler composites. [5, 6, 9, 12-16]. Since the experimental process requires a high cost and a lot of labor, many analytical/numerical methods have been developed and used as preliminary studies for material development. For example, Kim et al. [16] generated a finite element (FE) model of the carbon black composite, added an interface to consider the tunneling effect and electrical network, and calculated the equivalent electrical conductivity. Alian et al. [10] generated a FE model of CNT composites and applied an axial load. Then, the electrical conductivity of the nanocomposite was calculated by updating the equivalent electrical model with the displacement of the CNT in the modified FE model. This method can accurately predict the equivalent electrical conductivity for mechanical strain and residual stress. However, when the shape of the filler is complex or the number is large, it isn't easy to create an FE model, and a lot of calculation time is required. Accordingly, it is impossible to process using multiple samples due to the predictive characteristics of electrical conductivity in which several samples are analyzed, and the average is processed. Jung et al. [3] developed a network model for a curved CNT-polymer highly stretchable composite and studied its electrical properties under large deformation. Using Poisson's ratio definition of the polymer material, the geometry of CNT and representative volume element (RVE) for the strain was updated, and the electrical conductivity was calculated. Haghgoo et al. [15] studied the effect of temperature and thermal expansion on the resistance of CNTpolymer composites. The numerical analysis method using such an electrical network has the effect of reducing the calculation time because it does not require the creation of an FE model. In this study, a program for predicting the electromechanical behavior of highly stretchable CNT-polymer composites in response to uniaxial strain is developed and verified. Trade-off studies on RVE size, geometric periodicity of CNT fibers, and tensile strain were conducted to review the appropriateness of the methodology. In addition, verification of the electromechanical behavior of CNT mixed states with different aspect ratios is also conducted. 4 To this end, Section 2 describes generating RVEs by randomly distributing CNTs using the RSA (Random Sequential Adsorption) algorithm. And considering the contact and tunneling of the filler, the CNT network formation and the calculation process of updating the position of the CNT before and after the deformation of the RVE according to the tensile load are described. In addition, the method of calculating the CNT network's electrical conductivity by calculating the CNT's specific and tunneling resistance is explained. Section 3 compares the electrical conductivity calculation results according to the RVE size, the geometric periodicity of the CNT fiber, and the tensile strain. The program is finally verified by comparing them with the experimental results. Finally, Section 4 concludes the thesis with a conclusion. 5 2. Simulation procedure In this study, the model generation process and electrical conductivity calculation are shown in Figure 1. First, input parameter values such as CNT diameter and CNT length are determined. Next, CNTs are randomly distributed using the RSA algorithm, and the CNTs are updated according to whether periodicity is applied. Next, a network model is calculated by applying contact and tunneling resistance between neighboring CNTs considering the distance between CNTs. Finally, when there is a tensile strain, the CNT and RVE configurations and network models are updated considering the tension and shrinkage due to Poisson's ratio, and their electrical conductivity is calculated. START Before applying uniaxial strain Generate random RVE Generate network model considering tunneling effect Calculate electrical conductivity After applying uniaxial strain Applying strain in uniaxial direction Update CNT and RVE configurations and network models Calculate electrical conductivity END Figure 1. Flow chart of calculate electrical conductivity 6 2.1 RVE modeling The RSA algorithm was used as a method for generating CNTs. The RSA algorithm randomly distributes fibers through Monte Carlo simulation [17, 18]. The shape of the CNT is a cylinder with a long aspect ratio, as shown in Figure 2. First, the coordinate ( x0 , y0 , z0 ) of the starting point of the CNT is randomly determined inside the cube of size Lx Ly Lz . Determine the coordinate ( x1 , y1 , z1 ) of the endpoint of the CNT as shown in Eq. (1) below. x1 x0 l cos cos y1 y0 l cos sin z z l sin 0 1 (1) Here, l is the length of the CNT, is the elevation angle, is the azimuth angle, and and are expressed as Eq. (2, 3). (1 2 rand ), 2 2 2 2 rand , 0 2 (2) (3) Where, rand represents an arbitrary number in the interval [0, 1]. Figure 2. Configuration of the CNT The above process is repeated as often as the number of CNTs with a target fiber volume ratio ( V f ) to produce CNTs. When generating CNTs, overlapping of CNTs was allowed [14]. 7 In addition, in this study, two cases of RVE were classified to compare the electrical conductivity according to the application of the periodic boundary condition. In the nonperiodic case, which is the first case, as shown in Figure 3, when CNTs are placed across the boundary of the window during the RVE creation process, the part that crosses the edge is cut off, and only the region within the window is finally placed. On the other hand, in the case of periodic, the second case, geometric periodicity is satisfied by arranging the part that crosses the window boundary of the CNT on the opposite side of the window, as shown in Figure 4. Next, regions outside the edges of the RVE were deleted. However, periodic boundary conditions were given, so the flow of electricity continued. It is known that using this method, the equivalent electrical conductivity can be effectively and accurately predicted even if the RVE size is small and the number of CNTs is small compared to a model that does not implement periodic fiber arrangement. This will be further discussed in the numerical analysis results in Section 3.Ошибка! Источник ссылки не найден. (a) (b) Figure 3. Configuration of the RVE with the geometric non-periodicity (a) ISO view, (b) Front view 8 (a) (b) Figure 4. Configuration of the RVE with the geometric periodicity (a) ISO view, (b) Front view For the RVE generated through the RSA method to have randomness, it is necessary to ensure that the distribution of the azimuth and elevation angles of the cylindrical filler is uniform. This is known to be related to the number of fillers and the relative length ratio of fillers to RVE size [19-21]. We review this in this section. To verify the randomness of the generated RVE, the azimuth and elevation angles of the cylindrical fillers were measured, and their distribution was checked and compared with the utterly random case. If the cylindrical filler has a directionality, the distribution of angles will appear biased to one side. The directional distribution integral of the filler in 3D space is calculated as in Eq. (4) [21]. 2 0 2 ( p)dp s ( , ) cos d d (4) Here, ( p ) is the directional distribution function of the filler, and when the cylindrical filler is evenly spread and randomly distributed, the same value as Eq. (5) can be obtained. ( p) 1 2 (5) Therefore, ( ) is derived as in Eq. (6). 1 cos( ) cos( )d 0 2 2 ( ) (6) 9 Where the elevation angle distribution f ( ) of the filler when the section of [0, ] is divided into 18 is as shown in Eq. (7). f ( ) cos( ) cos( ) 2 18 36 (7) By repeating the above procedure, the azimuth distribution f ( ) of the filler can be obtained as in Eq. (8), and the result of the angle distribution of the pillar is shown in Figure 5. f ( ) 2 1 1 1 cos( )d 18 18 18 2 2 (8) In this study, CNTs with a length of 5μm were generated for RVE size L . Currently, CNTs are approximately 380, 3,000, and 10,300, respectively. Currently, the number of samples of each RVE used is 300. At this time, the error of the CNT shape regarding the ideal azimuth and elevation angles were defined as Eq. (9) and (10) and shown in Figure 5. R = 1 18 1 ( f (i ) 18 )2 , i [1 ,2 , 18 i 1 R = cos(i ) 2 1 18 ( f (i ) ) , i [1 , 2 , 18 i 1 36 ,18 ] [5,15, ,175] , 18 ] [85, 75, (9) ,85] (10) The error of the sample derived through this process is ( R , R ) = (8.04%, 10.75%) in the case of L / l 1, and it was confirmed that the error decreases as L / l increases. Therefore, in this study, analysis was performed in consideration of L / l 3 . 10 (a) (b) (c) Figure 5. Azimuth and elevation distribution of cylindrical fillers with V f 3% : (a) L / l 1 (b) L / l 2 (c) L / l 3 for l 5 2.2 Percolation modeling 11 The total electrical resistance of the CNT-polymer composite can be expressed as the sum of the CNT resistivity and the tunneling resistance. The specific resistance RCNT in the longitudinal direction of the CNT can be calculated as in Eq. (11). RCNT 4l CNT D 2 (11) Here, l is the length of the CNT, D is the diameter of the CNT, and CNT is the intrinsic electrical conductivity of the CNT. The interconnect features between CNTs are not in direct contact and are several nanometers apart due to a thin layer of insulating polymer preventing direct contact between CNTs [14]. Tunneling resistance considers the resistance between CNTs that are several nanometers apart. Tunneling resistance Rtunnel between two CNTs is as Eq. (12) [9]. Rtunnel h 1 2e 2 M (12) Where h is Planck’s constant, e is electron charge, h 2 e2 12.9504 kΩ, M is the number of conduction channels, and T is the transmission probability of electron charges between contacting surfaces [13]. T can be estimated by solving the Schrödinger equation for a rectangular potential barrier, as shown in Eq. (13) [9, 14, 15]. d vdW 0 d D d vdW exp d tunnel (13) T d D exp D d vdW d D d cutoff dtunnel Here, d vdW is the Van der Waals distance, d is the minimum distance between two CNT, d cutoff is the cutoff distance, and dtunnel represents the characteristic tunneling length as shown in Eq. (14). 12 dtunnel Where (14) 8me ΔE is the Dirac constant, me is the electron mass, and ΔE is the height of the barrier. D RCNT l d Tunneling Rtunnel Figure 6. Example of CNT resistance and tunneling The effective resistance can be obtained as Eq. (15) using the modified nodal analysis (MNA) method [10]. I external K 1 Vnodal V 1 0 I source source (15) Here, I external is the external current; the value is 0, Vsource is the voltage between the two opposite sides of RVE, Vnodal is the node voltage, I source is the current passing through the circuit due to Vsource , and K is the global coefficient matrix that can be obtained as Eq. (16) m K [ Kije ] (16) e 1 Where m is the total number of resistors in the percolating network, i and j are the end nodes of the resistor, and K ije is the element conductance. Electrical conductivity can be calculated by obtaining the equivalent resistance Reffective as shown in Eq. (17) using Ohm’s law. Reffective Vsource I source (17) 2.3 Realization of the electromechanical behaviour 13 The electrical resistance changes according to the strain rate when using nanocomposite as a strain sensor. To this end, when a significant strain is applied in the uniaxial direction, the shape and position of the CNTs are updated, and the network connection shape is also changed. To simplify the calculation process, uniaxial strain is applied in the z-axis direction, and at this time, the x, y, and z-axis strains are expressed as Eq. (18) below. d x dx dy dz , d y , d z x y z (18) Here, x , y , and z are strains in the x, y, and z axes. Integrating Eq. (18) and applying Poisson's ratio definition of the matrix as in Eq. (19) below is expressed in Eq. (20) below. This assumption is valid for CNT composites with V f within 5%. d d trans d x y d axial d z d z L L L L L dx L L dy dz L L z x y (19) (20) Where is Poisson’s ratio, L is the RVE size, L is the length of the RVE increses. By solving this and exponentiating it, the relationship between the strains in the x, y, and z axes can be defined as in Eq. (21) below [3]. 1 z 1 x 1 y (21) In conclusion, the lengths Lx , L y , and Lz of RVE for deformation in the z-axis direction are changed as shown in Eq. (22-24). Lx (1 x ) Lx (1 z ) Lx (22) Ly (1 y ) Ly (1 z ) Ly (23) Lz (1 z ) Lz (24) 14 In addition, the elevation angle, azimuth angle, and center point ( X c , Yc , Z c ) of CNT are expressed as in Eq. (25-27). (25) tan 1[(1 z )1 tan ] (26) ( X c , Yc, Z c ) [(1 z ) X c , (1 z ) Yc , (1 z ) Z c ] (27) The shapes of CNT and RVE updated by Z-axis deformation are shown in Figure 7 and Figure 8. Figure 7. Configuration of CNTs under tensile strain Tensile strain Figure 8. Configuration of RVE under tensile strain 15 3. Results and discussions In this study, the filler was assumed to be multi-walled CNTs (MWCNTs), and the RVE size, geometric periodicity of CNT fibers, electrical conductivity concerning tensile strain, and resistance to deformation were predicted. Table 1 shows the geometrical and physical properties of CNTs and polymers. Table 1 Instrinsic and phsycal properties of CNT and polymer [9] Property Value 5, 7, 8, 9 CNT length l CNT diameter D 50, 230 1.4 Cutoff distance d cutoff Unit μm nm nm Conduction channels M Intrinsic conductivity of CNT CNT 460 10,000 S/m Polymer Poisson’s ratio Energy barrier height E 0.33 1 eV 16 3.1 Periodicity Figure 9 shows the predicted electrical conductivity and the experimental results depending on whether the periodic boundary condition was applied. The length of CNT is 5μm, the diameter of CNT is 50nm, the size of RVE is 15 15 15 m3 , and the number of CNT is about 3,438 per V f 1% . Each point on the graph represents the average value of 300 samples. As a result of the prediction, the percolation threshold value was around 0.625% [refer]. The predicted electrical conductivity values in the vicinity of the percolation threshold differ somewhat depending on whether the CNT arrangement is periodic. Still, after that, it was confirmed that the electrical conductivity monotonically increased to a similar value regardless of whether or not the periodic boundary condition was applied. (a) (b) Figure 9. Prediction results according to (a) periodic boundary conditions, (b) RVE size 3.2 RVE size effect As mentioned earlier in subsection 2.3, the size of RVE compared to the length of CNTs is crucial to realize the ideal random CNT distribution, which also affects the effective electrical conductivity. To check such effect, a set of evaluations on the effective electrical conductivity was conducted by increasing the RVE size from 5 μm to 15 μm by 2.5 μm increments with the length of CNTs fixed at 5 μm. The number of CNTs for each RVE size is 127, 430, 1019, 1989, 17 and 3438, respectively. As shown in Fig. 10, the prediction results show that the slope of the curve became steeper as the size of the RVE increased around the percolation threshold. Figure 10. Prediction results according to RVE size 3.3 Tensile strain effect In addition, the variation in the electrical resistance was investigated according to the tensile strain. For simulation the test results and data set for the simulation are given in Table 2. Reportedly, the length of the CNTs was 5~9 μm in the test samples; for simulation, the average length of 7 μm of the CNTs was chosen. The diameter of CNT is 230nm, the size of RVE is 20 20 20 m3 , the V f of Test Sample#1 is 2.6%, and the V f of Test Sample#2 is 4.2%. When the RVE size was 20 20 20 m3 , the resistance curve was bent as the tensile strain increased, which was determined to be due to the insufficient number of CNTs. Therefore, the number of CNTs was increased by increasing the size of the RVE to 30 30 30 m3 , and as a result, the resistance curve did not bend. Compared to the test results, it was confirmed that Test Samples #1 and #2 had similar resistance curves at 2.6% and 5% fiber volume ratios, respectively. Table 2 Test results and data set for the simulation Case CNT length l RVE size L A-3 A-4 5~9 μm 5~9 μm 7 μm 7 μm 7 μm 7 μm 7 μm A-5 A-6 7 μm 7 μm A-7 7 μm Test Sample #1 [3] Test Sample #2 [3] A-1 A-2 20 μm 20 μm 20 μm 20 μm 20 μm 30 μm 30 μm 35 μm 30 μm 30 μm 35 μm 30 μm 18 Vf Number of CNT 2.6% 4.2% 2.5% 3% 5% 2.5% 2.6% 2.6% 2.7% 3% 3% 4% 688 825 1,375 2,321 2,414 3,833 2,507 2,785 4,423 3,713 7 μm A-8 30 μm 35 μm 5% 5% 4,642 7,371 (a) (b) Figure 11. The relative resistance change curve (a) , (b) 3.4 Mixed CNT effect As shown in Figure 12, the change in resistance according to the tensile strain was confirmed by mixing the CNT length. Table 3 shows the network model for resistance prediction and the information on Test Sample [3]Table 2. RVE was set to the same size as 30 30 30 m3 for all of them. At the same fiber volume ratio, it was confirmed that the increase in resistance decreased as the length of the CNT compared to the size of the RVE increased. In addition, by mixing the same number of CNTs with 5, 7, and 9 μm lengths, the test sample could be accurately simulated. It was confirmed that Test Samples #1 and #2 had similar resistance curves to the network model at 2.5% and 5% fiber volume fractions, respectively. Table 3 Test and network model data Case CNT length l Test Sample #1 [3] Test Sample #2 [3] B-1 B-2 B-3 B-4 B-5 5~9 μm 5~9 μm 7 μm 8 μm 9 μm 5,7,9 μm 5,7,9 μm RVE size L Vf Number of CNT 20 μm 20 μm 30 μm 30 μm 30 μm 30 μm 30 μm 2.6% 4.2% 2.5% 2.5% 2.5% 2.5% 3% 2,321 2,031 1,805 2,321 2,785 19 B-6 B-7 B-8 B-9 5,7,9 μm 5,7,9 μm 5,7,9 μm 5,7,9 μm 30 μm 30 μm 30 μm 30 μm 3.5% 4% 4.5% 5% 3,249 3,714 4,178 4,642 (a) (b) Figure 12. The relative resistance change curve 최종 정리 Table TEST Sample 1 (Reference) Tensiel Strain 7μm, 30μm, 2.5% w/ CNT Mixed effect 5,7,9μm, 30μm, 2.5% TEST Sample 2 (Reference) Tensiel Strain 7μm, 30μm, 5% w/ CNT Mixed effect 5,7,9μm, 30μm, 5% 10% 0. 6084 20% 2.7376 30% 6.5779 40% 12.6236 50% 24.7909 60% - 0.9626 (58.23%) 0.8346 (37.18%) 3.1318 (14.40%) 2.5101 (-8.31%) 10.0629 (52.98%) 6.3194 (-3.93%) 26.8728 (112.9%) 12.2384 (-3.05%) 69.9114 (182.0%) 29.5224 (19.09%) - 10% 0.0760 20% 0.2281 30% 0.6084 40% 1.0266 50% 1.6730 60% 2.6996 0.3351 (340.7%) 0.3380 (344.5%) 0.7322 (221.0%) 0.7231 (216.9%) 1.2130 (99.39%) 1.1847 (94.73%) 1.8047 (75.79%) 1.7281 (68.33%) 2.4775 (48.08%) 2.3722 (41.79%) 3.2645 (20.93%) 3.1290 (15.90%) - 4. Conclusions In this study, a program was developed and verified to predict the electromechanical behavior of highly stretchable CNT-polymer composites in which the electrical properties 20 change in response to uniaxial deformation. Monte-Carlo simulation method was used to generate RVEs with randomly distributed CNTs. Monte-Carlo simulation method was used to generate RVEs with randomly distributed CNTs. In the case of tensile strain, the CNT and RVE shapes and network models were updated by considering the shrinkage by Poisson's ratio. To examine the appropriateness of the proposed methodology, trade-off studies were conducted on RVE size, geometric periodicity of CNT fibers, and tensile strain. When the RVE size was more than three times the aspect ratio of CNT, the CNT showed a uniform distribution and accurately expressed the increase in resistance according to tensile strain. Next, the electromechanical behavior of CNT mixtures with different aspect ratios was confirmed and compared with the experimental data. When the aspect ratio was mixed rather than one, the analysis result accurately expressed the experimental data. In conclusion, the proposed model accurately predicts the percolation of CNT-polymer composites and helps predict electrical properties under extensive strain conditions. Also, from the expected results, it was confirmed that the RVE size, the geometric periodicity of the CNT fiber, and the degree of CNT dispersion affect the resistance. As a future study, we plan to conduct a study to predict the electrical conductivity of composite materials containing complex-shaped fillers. Acknowledgements This study was conducted with the support of the National Research Foundation of Korea, funded by the Ministry of Science and ICT. (NRF-2020R1F1A1075588). References [1] Li X, Hu H, Hua T, Xu B and Jiang S 2018 Wearable strain sensing textile based on onedimensional stretchable and weavable yarn sensors Nano Research 11 5799-811 21 [2] Chanda A, Sinha SK and Datla NV 2021 Electrical conductivity of random and aligned nanocomposites: Theoretical models and experimental validation Composites Part A: Applied Science and Manufacturing 149 106543 [3] Jung S et al. 2019 Modeling Electrical Percolation to optimize the Electromechanical Properties of CNT/Polymer Composites in Highly Stretchable Fiber Strain Sensors Scientific Reports 9 20376 [4] Alameri I and Oltulu M 2020 Mechanical properties of polymer composites reinforced by silicabased materials of various sizes Applied Nanoscience 10 4087-102 [5] Lu X, Liu Y, Pichon L, He D, Dubrunfaut O and Bai J, editors. Effective electrical conductivity of CNT/polymer nanocomposites. 2020 International Symposium on Electromagnetic Compatibility EMC EUROPE; 2020 23-25 Sept. 2020. [6] Bao W, Meguid S, Zhu Z and Weng G 2012 Tunneling resistance and its effect on the electrical conductivity of carbon nanotube nanocomposites Journal of Applied Physics 111 093726 [7] Brunella V, Rossatto BG, Mastropasqua C, Cesano F and Scarano D 2021 Thermal/electrical properties and texture of carbon black PC polymer composites near the electrical percolation threshold Journal of Composites Science 5 212 [8] Batiston E, Gleize PJP, Mezzomo P, Pelisser F and Matos PRd 2021 Effect of Carbon Nanotubes (CNTs) aspect ratio on the rheology, thermal conductivity and mechanical performance of Portland cement paste Revista IBRACON de Estruturas e Materiais 14 [9] Bao W, Meguid S, Zhu Z and Meguid M 2011 Modeling electrical conductivities of nanocomposites with aligned carbon nanotubes Nanotechnology 22 485704 [10] Alian A and Meguid S 2019 Multiscale modeling of the coupled electromechanical behavior of multifunctional nanocomposites Composite Structures 208 826-35 [11] Yahya N, Sikiru S, Rostami A, Soleimani H, Alqasem B, Qureishi S and Ganeson M 2020 Percolation threshold of multiwall carbon nanotube-PVDF composite for electromagnetic wave propagation Nano Express 1 010060 [12] Song W, Krishnaswamy V and Pucha RV 2016 Computational homogenization in RVE models with material periodic conditions for CNT polymer composites Composite Structures 137 9-17 [13] Zeng X, Xu X, Shenai PM, Kovalev E, Baudot C, Mathews N and Zhao Y 2011 Characteristics of the electrical percolation in carbon nanotubes/polymer nanocomposites The Journal of Physical Chemistry C 115 21685-90 [14] Haghgoo M, Ansari R, Hassanzadeh-Aghdam MK and Nankali M 2022 A novel temperaturedependent percolation model for the electrical conductivity and piezoresistive sensitivity of carbon nanotube-filled nanocomposites Acta Materialia 230 117870 [15] Haghgoo M, Ansari R, Hassanzadeh-Aghdam MK, Tian L and Nankali M 2022 Analytical formulation of the piezoresistive behavior of carbon nanotube polymer nanocomposites: The effect of temperature on strain sensing performance Composites Part A: Applied Science and Manufacturing 163 107244 [16] Kim D-W, Lim JH and Yu J 2019 Efficient prediction of the electrical conductivity and percolation threshold of nanocomposite containing spherical particles with three-dimensional random 22 representative volume elements by random filler removal Composites Part B: Engineering 168 38797 [17] Meakin P and Jullien R 1992 Random sequential adsorption of spheres of different sizes Physica A: Statistical Mechanics and its Applications 187 475-88 [18] Torquato S, Uche OU and Stillinger FH 2006 Random sequential addition of hard spheres in high Euclidean dimensions Physical Review E 74 061308 [19] Tian W, Qi L, Su C, Liu J and Zhou J 2016 Effect of fiber transverse isotropy on effective thermal conductivity of metal matrix composites reinforced by randomly distributed fibers Composite Structures 152 637-44 [20] Tian W, Qi L, Zhou J, Liang J and Ma Y 2015 Representative volume element for composites reinforced by spatially randomly distributed discontinuous fibers and its applications Composite Structures 131 366-73 [21] Qi L, Tian W and Zhou J 2015 Numerical evaluation of effective elastic properties of composites reinforced by spatially randomly distributed short fibers with certain aspect ratio Composite Structures 131 843-51 23