Modeling of Cu Doped Cobalt Oxide Nanocrystal Gas Sensor for Methane Detection: ANFIS Approach

In this paper, Nano-sized copper-cobalt compound oxide powders have been prepared by sol-gel technique with different mole ratios of Cu/Co (from 0.00 to 0.15); Detection of the methane gas, the most chemically stable hydrocarbon, is done. The structural properties and morphology of powders were studied by X-ray diffraction (XRD), Fourier transform infrared spectroscopy (FTIR) and Transmission electron microscopy (TEM). XRD analysis confirms that Co3O4 and (CuO0.3CO0.7) Co2O4 phases have been formed and mean grain size were decreased with increasing dopant (from 28 to 24nm). According to TEM images it was found that the particles have cubic morphologies with nearly uniform distribution. Then an adaptive neuro-fuzzy inference system (ANFIS) models have been utilized for prediction of sensitivity values of the corresponded sensor. The results of ANFIS model show that the independent predicted Sensitivity values (S) compared to the measured target values have a good agreement. And also, high coefficient were found (R2 > 0.98) for the response.

Many Nano-sized pure and doped semiconducting oxides have been used for gas sensors [27][28][29][30]. Use of Nano-materials is interested because of much greater surface area to the bulk ratio rather than course materials. In addition, when the grain size reduce to nanometer, especially when the dimensions become less than two times of the spacecharge depth, a large fraction of atoms which will present in the surface and the surface properties become dominant. So, materials conduction type turns to surface conduction type [31][32][33]. The sensing mechanism depends on the changes in surface conductivity which is resulted from chemical reaction between target gases and the absorbed oxygen on the surface of the Nano-sized metal oxide. Consequently, it is important to enhance the rate of chemisorptions and surface reaction. Co 3 O 4 is most active catalyst for oxidation of the methane [34]. So, it is a good candidate for the methane gas sensor. Addition of second component as doping is becoming one of the most interesting methods for optimization of the gas sensing properties in this type of sensors. These additives can act as catalysts in surface sites for adsorption of oxygen and detected gas or as promoters of the gas sensing characteristics [35]. The copper oxide has good catalytic activity for oxidation of the methane [34], so it was used as dopant for the methane gas sensor. For this reason, in this research, Nano-sized Cu-doped Cobalt oxides with different concentrations of the dopant have been prepared by the solgel method for the methane sensor application.

Material and Methods
The present paper covers an ANFIS modeling and experimental study of the gas sensing of the copper-cobalt compound oxide Nano particles. Experimental measurements have been made and Nanosized powders have been prepared by the sol-gel technique with different mole ratios of the Cu/Co (from 0.05 to 0.15) for detection of the methane gas. The structural properties and morphology of the powders were studied by the X-ray diffraction (XRD), Fourier transform infrared spectroscopy (FTIR) and Transmission electron microscopy (TEM).

If x is A and y is B then z is f(x, y)
Where A and B are the fuzzy sets in the antecedents and z = f (x, y) is a crisp function in the consequent. f(x, y) is usually a polynomial for the input variables x and y.
Consider z= f (x, y) is a first-order Sugeno fuzzy inference system, which contains two rules.
Rule 1: If x is A 1 and y is B 1 , Then: f 1 =a 1 x +b 1 y +c 1 Rule 2: If x is A 2 and y is B 2 , Then: f 2 =a 2 x +b 2 y +c 2 ANFIS structure contains five layers excluding input layer.
Layer 0 is the input layer. It has n nodes where n is the number of inputs to the system. Layer 1 is the fuzzification layer in which each node represents a generalized bell curve membership function value of a linguistic term as follows: Each node in Layer 2 provides the strength of the rule by means of multiplication operator. It performs AND operation.
Layer 3 is the normalization layer which normalizes the strength of all the rules according to the following equation: Layer 4 is a layer of adaptive nodes. Every node in this layer computes a linear function where the function coefficients are adapted by using the error function of the multilayer feed-forward neural network.
Layer 5 is the output layer whose function is the summation of the net outputs of the nodes in Layer 4. The output f is computed as follows in Figure 1: The first layer, every node i in this layer is an adaptive node with node function in Equation 1: Where {x*} are σ premise parameters updated through hybrid learning algorithm and x is input variable. At least in the basic ANFIS method these parameters are not adjustable.
The second layer calculates the firing strength for each rule quantifying the extent which any input data belongs to that rule. The output of the layer is the algebraic product of the input signals as can be given as: [40][41][42] 2, Ai 1 Figure 2 shows the ANFIS structure of T, Q and Ratio Cu/Co values as the three input parameters and two linguistic values (healthy or bruised) as the output parameters. After training the ANFIS model, predicted performance was tested.

Experimental
Nano-particles of copper-cobalt compound oxides were synthesized by sol-gel method. Systematically a proper amount of CoCl 2 .6H 2 O and CuCl 2 .2H 2 O dissolved in deionized water and ethanol. The initial solutions have been prepared with mole ratios of Cu/Co = 0.0, 0.05, 0.08, 0.10, 0.125 and 0.15. Then, citric acid (as complexing agent) was added in a mole ratio of C 6 H 8 O 7 /CoCl 2 .6H 2 O = 5, and primary sol was prepared by adding ethylene glycol (as polymerizing agent) in the mole ratio six time over cobalt cation and was stirred for 20 minutes. The clear mixed solution was refluxed at 120°C in oil bath for 6 hours. The final sol ensuing from reflux was kept at 80°C for 17 hours in oil bath to evaporate volatile compounds and a wet gel was formed. For removing the other residual volatile component the gel was heated again directly on hotplate at 180°C for 3 hours and then 220°C for 30 minutes. Finally the dry gel was annealed in 450°C for 1 hour to eliminate organic component and the copper-cobalt oxides Nano-powder was obtained. For gas sensing measurements, usually nanopowders screenprinted on the electrodes. So, using a component as binder, in many cases ethylene glycol, is necessary. Choi and Min [34] reported that sensitivity of a Co 3 O 4 sensor decreases with increasing of the ethylene glycol. Accordingly, sensors were fabricated by pressing followed by hot treatment. The same amount of each powders were pressed into pellets at a pressure of 7 ton and then heated in furnace for 2 hours at 800°C with heating rate of 2°C/min. Silver paste was used as electrodes on the two sides of one surface of pellets for recording of the resistant change.
Phase formation and crystallinity were studied by X-ray diffraction using a D8 Advanced Bruker with Cu kα radiation (λ=0.15405 nm) in a 2θ range from 15° to 85° with a step size of 0.05°. The atomic link x y x y x y investigation was performed using Fourier transform infrared (FTIR) spectroscopy on a Bomem MB-154 spectrometer. Transmission electron microscopy (TEM) was carried out on a LEO 912 AB transmission electron microscope for determination of the particle shape and size. The prepared sensors were located in the testing chamber for gas sensing characterization. All sensitivity measurements were done in constant temperature in the range of 200-300°C and fixed quantity of sample gas of methane (3000 to 6000ppm). Electrical resistance of the sensor was monitored by a digital multimeter.
The sensitivity was defined as (R gas -R air )/R air ×100(%) Where R air and R gas are the electrical resistance in clean air and in presence of the gas at the same temperature, respectively.

Experimental results
The XRD patterns of copper-cobalt oxides nanopowders with different mole ratio of copper have been shown in Major XRD peaks are corresponded to (3 1 1) plane at 36.9° and it intensity does not change evidently by adding copper impurity in samples but it has become slight wider. Beside the strongest peak, also other peaks corresponded to (2 2 0) and (4 4 0) planes have got wider. It's clearly identified that copper impurity decreases the crystallinity and enhances the amorphous state that lead to the decreasing of the crystalline size. No other phase has been created in samples with different Cu/Co mole ratios and it demonstrates that formed phase are stable and its formation is independent of copper quantity.
Mean particle's size in major peaks is calculated using Scherrer's equation.
Where k is constant, depend on crystal morphology in the range of 0.89-1.39 and it is 0.9 in this calculation. λ is the x-ray wavelength, β is the line broadening at half the maximum intensity (FWHM) in radians and θ is the Brrag angle of XRD peak. The particle size values are given in Table 1. It was varied between 24 and 28 and decrease as Cu/Co mole ratio increase to 0.15. Figure 4 shows FTIR spectra of copper-cobalt oxide Nano-particles with different mole ratios of Cu/Co = 0.0, 0.05, 0.10, 0.15 (S 1 , S 2 , S 4 and S 6 respectively).
The bands at 576 and 668 cm -1 are attributed to vibration of Co-O in the Co 3 O 4 . Any peak related to Cu-O could not be identified; it is probably Co-O peaks overlap Cu-O peak, and they have seen in very close wavelength. There is no peak that could be assigned to C-H which suggests that carbonate component removed completely from samples.
TEM images of cobalt oxide nanoparticles (S 1 ) and copper-cobalt oxide nanoparticles with Cu/Co = 0.15 (S 6 ) prepared by sol-gel method with different magnification are given in Figure 5.
It's found that particles have cubic morphologies with uniform distribution. The average particle size was 30 nm in S 1 and 15nm in S 6 which decreases by increasing of the dopant. It has a good agreement with result of Scherre's formula.

ANFIS comparative analyses and discussion
Two models of Sugeno, with an automatic extraction of data from FIS [GENFIS2] were used. The MATLAB software was adopted for comparison purposes. Moreover, the coverage threshold was fixes to 0.01. Table 2    the relationship between Temperature, Q, ratio Cu/Co and S are given in Figure 6: A complete graphical comparison of experimental data used for testing or validating the predictions of ANFIS models are shown in Figures 7&8 for the train and the test data respectively. The coefficient of determination values (R 2 ), which quantifies the degree of agreement between experimental observations and numerically calculated values were found greater than 0.96 for all output variables are shown in Figure 9&10. R 2 was defined as below:

Conclusion
Copper-cobalt oxide Nano-sized powders with different mole ratio of Cu/Co from 0.0 to 0.15 were prepared by sol-gel technique for gas sensor application. Cubic Co 3 O 4 and (CuO 0.3 CoO 0.70 ) Co 2 O 4 phases were firmly formed. The mean grain size varied from 28 to 24 by changing Cu/Co mole ratios from 0.0 to 0.15. The gas sensing performances of sensors were investigated in presence of methane gas. The results have shown that optimal operating temperature is 300°C in all sensors. Doping of metal oxide to Co 3 O 4 altered the sensing behavior by influence defect chemistry and decrease particle size. The performance  of the ANFIS Prediction and experimental results was measured using the average relative error (ARE), absolute average relative error (AARE), root mean square error (RMSE) and the Coefficients (R 2 ) values. The developed ANFIS model showed a good regression analysis with the R 2 greater than 0.98. As the regression coefficients indicate the ANFIS approach could be considered as an alternative and practical technique to evaluate the sensitivity (S) based on the Q, temperature and ratio of Cu/Co with a high degree of accuracy.