Study on Porosity of Electrospun Nanofiber. Membrane by Neural Network

Transcription

1 Applied Mathematical Sciences, Vol. 12, 2018, no. 22, HIKARI Ltd, Study on Porosity of Electrospun Nanofiber Membrane by Neural Network Ting Wang 1, Wenxia Dong 1, Ying Chen 2, 3, 4, Tiandi Pan 2 and Rudong Chen 1,* 1 Department of Mathematics, Tianjian Polytechnic University No. 399, Binshui Street, Xiqin District, Tianjian, , China * Corresponding author 2 Department of Textile, Tianjian Polytechnic University No. 399, Binshui Street, Xiqin District, Tianjian, , China 3 The Higher Occupation Education Department Tianjin University of Technology and Education, Tianjin, , China 4 Statistical Research Institute, Naikai University, Tianjin, , China Copyright 2018 Ting Wang et al. This article is distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited. Abstract In this paper, using artificial neural network (ANN) method to find the effects of electrospinning parameters including spinning distance (cm), applied voltage (kv), and volume flow rate (ml/ h) on the porosity of electrospun nanofiber membrane is mainly studied. The porosity of the nanofiber membrane was obtained through the Matlab software to calculate the pixel value. The study found that the applied voltage (kv) and the spinning distance (cm) on the nanofiber membrane porosity have greater impact. The correlation coefficient between the variables and ANN model (R 2 =0.996) shows splendid fitting with experimental data. The ANN model predicted the maximum porosity (59.88%) of electrospinning nanofiber membrane at the conditions of 19 cm of spinning distance, 26 kv of the applied voltage and 0.5 ml/h of volume flow rate.

2 1060 Ting Wang et al. Keywords: artificial neural network; Matlab 2012b; electrospun nanofiber; porosity 1 Introduction Electrospinning refers to a spinning technique in which a polymer solution or melt is spray-drawn under electrostatic action to obtain nanofibers. Electrospinning is a special fiber-making process in which polymer solutions or melts are spun in strong electric fields. Under the action of an electric field, the droplet at the needle will change from spherical to conical (ie, "Taylor cone") and extend from the tip of the cone to form a filament [1]. This method produces nano-diameter polymer filaments. Electrospinning process parameters affecting porosity can be divided into four categories: polymer properties, solute properties, solution properties and experimental parameters [2-4]. This paper focuses on the fourth category. The porosity of a fibrous membrane, expressed as a percentage, refers to the ratio of the pore volume of the fibrous membrane to the total volume of the fibrous membrane [5]. The porosity of nanofiber membranes has great application in filtration, tissue engineering and so on [5-7]. The measurement of the porosity of nanofibrous membrane includes density method, solution replacement method [8, 9]. Matlab and Photoshop are used to measure the porosity of electrospun nanofiber membrane. Artificial neural network is model that simulates the behavior characteristics of animal neural network and carries out distributed parallel information processing. The earliest proposed imitation of human brain function is the MP model established by American scientists Pitts and McCulloch in Artificial neural network is a multi-layer structure of the feedforward network, mainly by the input layer, the hidden layer and the output layer of three parts. Each node in the input layer corresponds to a predictor variable. The node of the output layer corresponds to the target variable, there may be more than one. Between the input layer and the output layer is hidden layer, hidden layer and the number of nodes in each layer determines the complexity of the neural network. In recent years, the use of neural networks on the porosity has attracted the interest of many scholars [1, 10-12]. In this paper, the influence of electrospinning process parameters on the porosity of nanofiber membranes is studied by means of neural network. A new mathematical method is provided to study the porosity of nanofiber membranes.

3 Study on porosity of electrospun nanofiber membrane Experimental 2.1 Materials Polyvinyl alcohol (PVA) with number-average molecule weight (Mn) of 84000~89000, alcoholysis degree of 86 ~ 89 mol% and the average degree of polymerization 1700 ~ 1800.was purchased from Changchun Petrochemical Co., Ltd.,Taiwan. 2.2 Preparation of electrospun nanofibers Polyvinyl alcohol was dissolved in distilled water to prepare a polyvinyl alcohol solution with a concentration of 12 wt %. The obtained mixed solution was stirred in a water bath at 80 C for 1 h until a homogeneous solution was formed, and then defibrated for standing. 2.3 Sample preparation At room temperature, pour the prepared spinning solution into four 5-ml syringes with a needle diameter of 1.2 mm. From the previous studies, the process parameters of electrospun nanofiber nonwoven technology parameters are five, namely: spinning time to take 90 minutes, solution concentration of 12 wt %; spinning distance (cm) take 11,13,15,17,19; applied voltage (kv) take 15,18,20,23,26; volume flow rate (ml/ h) take 0.5,0.7,1,1.2,1.5. In order to experiment the generality, the mathematical method of orthogonal experiment was used and five experiments was added on this basis as shown in TableⅠ and Table Ⅱ to design experiments. TABLE Ⅰ Orthogonal table

4 1062 Ting Wang et al. TABLE Ⅰ (Continued): Orthogonal table TABLE Ⅱ Orthogonal experimental table No Spinning distance (cm) Applied voltage (kv) Volume flow rate (ml/h) No Spinning distance (cm) Applied voltage (kv) Volume flow rate (ml/h) Morphological characterization The surface of polyvinyl alcohol nanofiber membrane with different parameters under vacuum condition was sprayed with gold. The morphology after treatment was observed by scanning electron microscope (SEM,TM-3030, Japan) coating with magnification of Figure 1 shows the SEM micrograph of electrospun nanofiber mat.

5 Study on porosity of electrospun nanofiber membrane Threshold determination Figure 1 The SEM of electrospun nanofiber mat In Photoshop, the threshold command turns a gray scale or color image into a high-contrast black-and-white image. We can specify a color scale as a threshold, all pixels that are brighter than the threshold to white, and all pixels that are darker than the threshold to black. In this paper, the scanning electron micrographs of the nanofiber membranes are imported into Photoshop. The SEM is transformed into gray scale images, the thresholds of the images are adjusted, and a suitable threshold is found by observing changes of the images. 2.6Porosity measurement The pixel values corresponding to each position in the scanning electron micrograph (SME) are calculated by Matlab software. Then, using the threshold values obtained above, the part larger than the threshold value is a white pixel value, that is, a non-porous part. Matlab procedures are as follows: filename='1.jpg'; ddata=imread(num2str(filename)); figure;imshow(ddata); gdata=rgb2gray(ddata); figure;imshow(gdata); xlswritecopy('1.xls',gdata) Thus the porosity can be calculated as follows:

6 1064 Ting Wang et al. N n p(%) 100 N Where n is the pixel value of the white portion, N is the pixel value of the entire image, and N is the porosity of the nanofiber membrane. 2.7 Experimental design In this paper, only three process parameters (spinning distance, applied voltage, volume flow rate) and one output parameter (porosity) are studied, so a three-input neuron and one neural network model can be established, as shown in Figure 2. For Hidden layer neurons, according to "hidden neurons try to minimize the number of convergence as fast as possible, approaching the error as small as possible" principle, we have been trained to select 12 hidden neurons. Transfer function selection as 'transig', 'transig' such a transfer function, select 'trainlm' as a training function. (1) Figure 2 Architecture of a three-layer neural network with one hidden layer 3 Results and discussion 3.1 Porosity measurement results First of all, the threshold was determined using Photoshop software. Secondly, pixel values for the SEM image at various positions were obtained by Matlab software. Then the porosity of the nanofiber membrane can be calculated, the results of 30 groups of nanofiber membrane porosity experiments are shown in Table Ⅲ. TABLE Ⅲ Electrospun nanofiber membrane porosity results No Threshold Magnification Porosity/% No Threshold Magnification Porosity/%

7 Study on porosity of electrospun nanofiber membrane 1065 TABLE Ⅲ (Continued): Electrospun nanofiber membrane porosity results Artificial neural networks results After observing the sample images, we can see that the 6 # and 18 # samples have problems. Figure 3 Samples of 6# and 18# As we can see, 6 #, 18 # samples are almost no fiber, so two samples are removed. Neural network model is trained and simulated using Matlab 2012a.The training of the ANN was stopped after 23 because the targeted MSE value was reached, as shown in Figure 4. Then the functional relationship between each parameter and porosity was obtained: f(x1,x2,x3)= 36.01/(exp(19.59/(exp(1.283*x1-1.3*x *x3) ) + 1.0) /(exp(1.252*x *x2) *x3-13.0) + 1.0) /(exp( *x *x3-1.29*x1) + 1.0) /(exp(4.591*x *x *x ) + 1.0) /(exp( *x *x *x1) + 1.0) /(exp(0.9808*x *x2

8 1066 Ting Wang et al *x ) + 1.0) /(exp(0.2119*x *x *x ) + 1.0) /(exp(0.6171*x *x *x ) + 1.0) /(exp(0.901*x *x *x ) + 1.0) /(exp( *x *x *x1) + 1.0) /(exp(1.585*x *x *x ) + 1.0) /(exp(1.149*x *x *x ) + 1.0) ) + 1.0) where x1, x2, x3, respectively, that spinning distance, applied voltage and volume flow rate. Figure 4 The graphic of error variation depending on iteration of ANN TABLE Ⅳ The Matlab and ANN predicted electrospun nanofiber porosity porosity porosity No Matlab/% ANN/% error/% No Matlab/% ANN/% error/%

9 Study on porosity of electrospun nanofiber membrane 1067 TABLE Ⅳ (Continued): The Matlab and ANN predicted electrospun nanofiber porosity R 2 =0.996 Mean absolute error (%)=0.187 Mean square error (%)= Result analysis According to the function relation f(x1, x2, x3), the effects of all the single and dual factors on the porosity of the nanofiber membrane were studied. The effect plots, which describe the effects of single factor and dual factors on the porosity of the nanofiber membrane, are shown Figure 5 and Figure 6 respectively. When x1 = 15, x2 = 20, the porosity changes with x3, as shown in Figure 5 (a); When x1 = 15, x3 = 1, the porosity changes with x2, as shown in Figure 5 (b); When x2 = 20, x3 = 1, the porosity changes with x1, as shown in Figure 5 (c). When x1 = 16, the porosity changes with x2, x3, as shown in Figure 6(a); When x2 = 21, the porosity changes with x1, x3, as shown in Figure 6 (b); When x3 = 16, the porosity changes with x1, x2, as shown in Figure 6 (c).

10 Porosity (%) Porosity (%) Porosity (%) 1068 Ting Wang et al. Volume flow rate (ml/h) Applied voltage (kv) Spinning distance (cm) Figure 5 Fixed two variables, the changes of porosity with the other variables

11 Study on porosity of electrospun nanofiber membrane 1069 Figure 6 Fixed one variable, the changes of porosity with the remaining two variables We can also use the ANN weight matrix to evaluate the relative importance (RI) of different input parameters on the porosity of electrospinning nanofiber membrane to the output parameters, and one based on the weight of the connection weight is proposed[13,14]: RI j N h (( I { i mj m 1 k 1 Ni N h N i (( I / mk k 1 m 1 k 1 N / IW ) L ) mk mk mn I ) L )} Where RI j is the relative importance of the output parameters of different input parameters; N i, N h, respectively, the number of input neurons and hidden neurons; I, L, respectively, for the input layer to the hidden layer of the weight matrix and the hidden layer to the output layer weight matrix ; Subscript n is the output parameter. In the paper, j = 1, 2, 3, Ni = 3, Nh = 12, n = 1, I,L are given in Table Ⅴ. mn (2)

12 1070 Ting Wang et al. TABLE Ⅴ Weights and bias in training Layer Weight Bias Hidden layer I 11 I 12 I I 21 I 22 I I 31 I 32 I I 41 I 42 I I 51 I 52 I I 61 I 62 I I 71 I 72 I I 81 I 82 I I 91 I 92 I I 10,1 I 10,2 I 10, I 11,1 I 11,2 I 11, I 12,1 I 12,2 I 12, b b b b b b b b b b 10, b 11, b 12, Output layer L 1 L 2 L 3 L 4 L 5 L L 7 L 8 L 9 L 10 L 11 L b The relative importance of the nanofiber membrane porosity is calculated as shown in Figure 7. The three process parameters have a strong influence on the porosity of the nanofiber membrane. In the research, the influence of these three parameters on the porosity of the nanofiber membrane should not be neglected.

13 Study on porosity of electrospun nanofiber membrane 1071 However, the influence of voltage and distance on the porosity of nanofiber membranes is even more significant relative importance ( % ) 26.0 distance voltage rate Figure7 The relative importance of the parameters of the porosity of the nanofibrous membrane 3.4 Optimization In this paper, the goal is to find the maximum porosity of nanofiber membrane. Optimization finds a set of conditions that meet the maximum porosity. In Figure 5, the optimum conditions in the given range for maximum porosity of electrospinning nanofiber membrane were 19 cm of spinning distance, 26 kv of the applied voltage and 0.5 ml/h of volume flow rate. In order to insure the predictive ability of the ANN model, more electrospinning experiments were carried out. The result and experiment conditions was shown in Table Ⅵ.

14 1072 Ting Wang et al. TABLE Ⅵ Validation of ANN using different levels of applied voltage, spinning distance and volume flow rate Actual values of the variables Nanofiber membrane porosity Spinning distance Applied voltage Volume flow rate No (cm) (kv) (ml/h) Threshold Matlab ANN A B C D Conclusions In this study the impact of three electrospinning parameters, namely applied voltage (kv), spinning distance (cm), and volume flow rate (ml/h), on the porosity of electrospun nanofiber mats was determined by ANN. The porosity of the nanofiber membrane measured by the Matlab software is close to the porosity of the nanofiber membrane simulated by the ANN, indicating that the performance of ANN model for predicting was good. The results also showed that the applied voltage and spinning were the two most critical parameters affecting the porosity of the nanofiber mats. Increasing the spinning distance resulted in less nanofiber mat porosity whereas increasing the applied voltage caused an increase in nanofiber mat porosity. Acknowledgements. This research was supported by The Science and Technology Plans of Tianjin (No. 15PTSYJC00230), NSFC Grant No References [1] G.E. Wnek, M.E. Carr D.G. Simpson, Electrospinning of nanofiber fibrinogen structures, Nano Letters, 3 (2003), no. 2,

15 Study on porosity of electrospun nanofiber membrane 1073 [2] H. Brooks, N. Tucker, Electrospinning predictions using artificial neural networks, Polymer, 58 (2015), [3] A. Doustgni, Effect of electrospinning process parameters of polycaprolatone and nanohydroxyapatite nanocomposite nanofiber, Textile Research Journal, 85 (2015), no. 14, [4] Tan Xiaohong, Nanotechnology and Nano Textiles, Shanghai: Donghua University Press, [5] S.S. Sreedhara, N.R. Tata, A novel method for measurement of porosity in nanofiber mat using pycnometer in filtration, Engineered and Fabric, 81 (2013), no. 4, [6] L. Ghasnmi-Mobaraheh, D. Semnani, M. Morshed, A novel method for porosity measurement of various surface layers of nanofibers mat using image analysis for tissue engineering applications, Applied Polymer Science, 106 (2007), no. 4, [7] S. Zhang, W.S. Shim, I. Kim, Design of ultra-fine nonwovens via electrospinning of Nylon 6: spinning parameters and filtration efficiency, Materials and Design, 30 (2009), no. 9, [8] Wang Yanlian, Miao Qinxian, Fiber porosity measurement and density measurement of cellulose content, Heilongjiang Paper Mill, 37 (2009), no. 1, [9] Y.Y. Hsu, J.D. Grasser, D.J. Trantolo, C. Lyons, P. J. Gangadharam, D. Wise, Effect of polymer foam morphology and density on kinetics of in vitro controlled released of isoniazid from compressed foam matrices, Biomedical Materials Research Part A, 35 (1997), no. 1, o;2-g [10] M.A. Karimi, P. Pourhakkak, M. Adabi, S. Firoozi, M. Adabi, M. Naghibzadeh, Using an artificial neural network for the evaluation of the parameters controlling PVA/chitosan electrospun nanofibers diameter, E-Polymers, 15 (2015), no. 2, [11] S. Firoozi, A. Amani, M.A. Derakhshan, H. Ghanbari, Artificial neural

16 1074 Ting Wang et al. networks modeling of electrospun polyurethane nanofibers from chloroform /Methanol solution, Nano Research, 41 (2016), [12] E. Vatankhah, D. Smrani, M.P. Prabhakaran, M. Tadayon, S. Razavi, S. Ramakrishna, Artifical neural network for modeling the elastic modulus of electrospun polycaprolactone/gelation scaffolds, Acta Biomacterialia, 10 (2014), no. 2, [13] K. Nasouri, H. Bahrambeygi, A. Rabbi, A. Shoushtari, A. Kaflou, Modeling and optimization of electrospun PAN nanofiber diameter using response surface methodology and artificial neural networks, Applied Polymer Science, 126 (2012), no. 1, [14] M.B. Kasiri, H. Aleboyeh, A. Aleboyeh, Modeling and optimization of heterogeneous Photo-Fenton process with response surface methodology and artificial neural networks, Environmental Science and Technology, 42 (2008), no. 21, Received: March 31, 2018; Published: August 30, 2018