Abstract
Nanocatalysts play a significant role to improve the thermal and physical properties of biodiesel. In the present work, the multiwalled carbon nanotubes (MWCNTs) as an additive with the fraction of 30, 40, and 50 ppm are dispersed with the different biodiesel–diesel blends of 10%, 30%, and 50% of waste cooking oil (WCO)-based biodiesel (B10, B30, B50) for the prediction of four-stroke compression ignition (CI) engine emissions using multilayer neural network (MLNN) model. An MLNN model uses a backpropagation algorithm to map input and output parameters. The input parameters to MLNN are load, blends, and MWCNTs in ppm. On the other hand, the output parameters are HC, CO, and NOx. The results for the optimum topological structure of 3-10-3 denoted mean square error (MSE) equal to 0.095 that are capable of predicting the emissions for different operating conditions. Thereafter, the developed MLNN model is tested on an experimental setup consisting of a single-cylinder four-stroke CI engine and emission analyzer. The emission characteristics predicted by MLNN are called to be nearly experimental measurements with reasonable accuracy as it depicts the good “R” values as 0.95, 0.96, and 0.976 for HC, CO, and NOx, respectively, and also gives the reasonable average relative error values as 0.83%, 1.01%, and 1.05%, for HC, CO, and NOx, respectively. Further, the developed model is suitable for predicting emissions of CI engines, thus minimizing the cost, time, and labor effort.
1. Introduction
The research on alternative fuels has received attention due to growing demand and limited availability of petroleum; the cost is rising. Because of that, there is a requirement to search for alternate fuels for compression ignition (CI) engines [1]. In this connection, biodiesel from nonedible oils as feedstock becomes of greater interest. However, the cost associated with the production is high as compared with diesel. Due to the aforesaid reason, biodiesel is not appropriate for commercialization. In this context, there is a need for biodiesel whose production cost should be less compared with diesel. Further, waste cooking oil (WCO) is considered an economical biodiesel feedstock as its price is considerably lower than oil from other sources [2–5]. In the previous works, investigation of CI-engine parameters has been carried out via experiments [5–8]. Further, in recent advances, nanotechnology proved that nanomaterial-dispersed fuel can be used for better performance of engines due to its good mechanical and thermophysical properties than traditional materials [9–13]. The research work related to the experimentation of CeO2 [14] and ZnO2 [15] influence on CI engine operated with biodiesel is observed. Experimental discussion on the influence of carbon nanotubes over the diesel engine is reported by Tewari et al. [16] and by Basha and Anand [17], the result of Al2O3 and CeO2 added in biodiesel (Jatropha) is given in a study by Prabhu [18], and the influence of various concentrations of magnesium oxide in WCO-based biodiesel is appeared in a study by Ranjan et al. [19]. Even so, testing of engine for all operating conditions needs more money and time. As an alternative, the numerical/mathematical approach and neural network (NN) approach can be put to use for the investigation of engine variables associated with thermos-physical properties, performance, and emissions [20–25]. However, computational complexity limits the use of mathematical models. Alternatively, NN can be exploited for the same as it is advantageous over other techniques, viz., generalization capability, huge data-handling ability, mapping ability, etc. [20–25]. A NN model based on a backpropagation algorithm is utilized to determine the parameters/variables of the engine operated with blended biodiesel that has appeared in a study by Yusaf et al. [22]. Additionally, several studies [23–26] emphasized on the investigation of emission constituents and performance of engines operating on diesel, biodiesel, and nanoadditives biodiesel through the artificial NN. The metallic and carbon nanotubes (CNTs) become advantageous concerning emission reduction and have remarkable mechanical properties [16]. As far as the authors’ knowledge, limited work has highlighted the impact of multiwalled carbon nanotubes (MWCNTs) on WCO biodiesel in diesel engines. Further, the gap in the literature on engine emissions while adding CNTs to biodiesel fuel appears to be addressed. Furthermore, the ability of NN to predict emissions of engine runs on low-cost biodiesel (WCO) dispersed with MWCNTs has yet to be investigated in the previous works. Subsequently, this work has been implemented to fill the gap available in the literature. Keeping this concept in the mind, a multilayer NN has been constructed to predict the emissions, viz., HC, CO, and nitrogen oxides (NOx) from the CI engine run on WCO as biodiesel dispersed with MWCNTs.
2. Experimental Procedure, Setup, and Investigations
2.1. Experimental Procedure
The study involves the preparation of biodiesel using WCO as feedstock, and then different blends (10%, 30%, and 50%) of WCO-based biodiesel (B10, B30, B50) and diesel have been prepared. Further, the dispersion of MWCNTs is carried out with the different dosing levels of 30, 40, and 50 ppm in WCO blends through ultrasonicator (Figure 1), which mixes the nanoparticles into the diesel+WCO biodiesel fuel, as it helps agglomerate particles’ restoration to nanometer range. The resulted MWCNTs-blended WCO biodiesel is placed in beaker under static condition to assure the stability. Further, the MWCNTs are obtained from “Platonic Nanotech Private Limited,” Mahagama, India. Further, some of the properties of MWCNTs are shown in Table 1. Thereafter, the scanning electron microscope (SEM) picture is shown in Figure 2 and transmission electron microscope (TEM) picture is shown in Figure 3. As shown in Figures 2 and 3, it is revealed that CNTs material are having tube-like structure and the width of 4–22 nm and length of about 60–150 nm. Figure 4 represents X-ray diffraction (XRD) pattern of MWCNTs, and they are not crystalline in nature.
2.2. Properties of Pure Diesel and WCO Biodiesel Blends with MWCNTs
Testing of the prepared samples dispersed with MWCNTs has been carried out to obtain the values of density in kg/m3, viscosity in cSt, calorific value in kJ/kg, and flash and fire point in °C. Further, the properties of pure diesel and WCO biodiesel blends with MWCNTs for 30, 40, and 50 ppm are shown in Tables 2–4, respectively.
2.3. Engine Setup
The direct injection (DI), four-stroke single cylinder CI engine is shown in Figure 5 and its instrumentation part is shown in Figure 6(a). It comprises of a bed, engine connected to eddy dynamometer. The dynamometer has been used to load the engine. To circulate the water to the calorimeter and dynamometer, pump is provided. A tank for fuel is placed into control panel accompanied by fuel unit for measurement. The test engine runs at steady speed of 1,500 rpm at rated power, which make use of B10, B30, and B50 blends. The emission characteristics for each test condition and blend are measured using emission analyzer, as shown in Figure 6(b). The fuel tank is made completely empty for every time, and then new WCO-blend-dispersed MWCNTs are poured into tank and engine emissions are measured. Further, the specification of engine and emission analyzer is shown in Tables 5 and 6, respectively.
(a)
(b)
2.4. Uncertainty Analysis
Uncertainty analysis of the various parameters has been carried out to overcome the errors involved in the instruments. The error and uncertainty of instruments depend upon the working and natural conditions and instrument’s accuracy [25]. In this, tests are performed many a time and mean values are utilized for further analysis (Table 7).
2.5. Experimental Investigations of Emissions
2.5.1. Hydrocarbons
Figures 7–9 show the effect of hydrocarbons (HC) with respect to load for B0, B10, B30, and B50 mixed with MWCNTs of 30, 40, and 50 ppm, respectively. It is distinguished from Figures 7–9 that HC emissions are increasing with the increase in the load for all the blends; this is because, for small loads, fuel consumption is less, resulting in less emissions, whereas for higher loads, fuel consumption is high as a result of higher emissions [11, 13, 18]. HC emissions are diminished by applying 30, 40, and 50 ppm of MWCNTs. Nanoparticles act as oxidation catalysts, which improve the oxidation and, as a result, lower the HC emissions [13, 18]. This is due to increase in the surface-to-volume ratio and good fuel–air-mixing rate.
2.5.2. Carbon Monoxide
Figures 10–12 show the effect of carbon monoxide (CO) with load B0, B10, B30, and B50 mixed with MWCNTs of 30, 40, and 50 ppm, respectively. It is noticed from Figures 10–12 that the CO emissions are decreasing with the increase in the load for all the blends, this is due to improved combustion; it is also seen that CO emission for the WCO biodiesel mixed with MWCNTs is marginally less in comparison with that of B0. This is because the increasing content of oxygen with MWCNTs in the blends and higher surface-to-volume ratio of nanoparticles [13, 18].
2.5.3. Nitrogen Oxides
Figures 13–15 show the effect of NOx with load B0, B10, B30, and B50 mixed with MWCNTs of 30, 40, and 50 ppm, respectively. It is noticed from Figures 13–15 that NOx emissions are increasing with the increment in the load for all the blends; this is because of improvement in combustible temperature, oxygen quantity in higher value, and fast reaction rate [27]. The addition of MWCNTs into fuel leads to complete combustion as it works as an oxygen-donating catalyst inside the combustion chamber [13, 18].
3. Multilayer Neural Network
In this work, the feed-forward backpropagation multilayer neural network (MLNN), as shown in Figure 16, is applied to map each parameter because it is called as universal approximator and highly appropriate for nonlinear plants [24, 25]. The superiority of MLNN compared with other approaches is that it could be utilized to model the nonlinear plant and can come up with input–output/target mapping, adaptive, and fault tolerance [28, 29]. MLNN is used to model the variables and the output/target is depicted by Equation (1).where is the target of MLNN. The tangent hyperbolic (tanh) function and linear function are utilized for unseen (hidden) and output layers, using Equations (2) and (3), respectively [28, 29].
In Equations (1)–(3), is the input to the input neurons, is the weight vector, to be calculated over the entire training process, and are the input variables to the network.
The engine load, WCO biodiesel blends, and MWCNTs ppm are the inputs to the MLNN. One can calculate the weights by minimizing the cost function defined by Equation (4).where and the weight decay matrix is designated as .
To acquire the outline, data points on the network structure are segregated into first subset 70%, second subset 15%, and third subset 15% of data. The first subset data are utilized for training process, whereas second and third subsets are used for validation and cross-validation (testing), respectively. The best epoch can be selected by observing early stopping technique (Figure 17). Several MLNN architectures/topologies are trained by varying neuron numbers from 2 to 12 in hidden layer. For this, MATLAB® 2014(b) is used. The architecture/topology that describes the minimal mean square error (MSE) corresponds to validation data is opted as the optimum network to predict the target parameters. The validation data MSE for various topologies are given in Table 8. The optimum architectures seem to be suitable for predicting emissions over any test conditions of engine.
4. Results and Discussion
In this manuscript, the results acquired with MLNN predict the emissions of four-stroke CI engine operated on biodiesel produced from WCO added with MWCNTs are blended with neat diesel that is discussed in this section. In the present research, biodiesel produced from WCO added with a different fraction of MWCNTs. Further, the model’s accuracy is evaluated via regression analysis. The correlation coefficient (R) criterion is employed for the measurement of model’s accuracy. The “R” indicates the correlation between predicted and experimental values. The experimental and predicted emissions are compared for all test conditions. The experimental and MLNN-model-predicted emission characteristics are shown in Figures 18–20. These figures indicate the MLNN model accurately as its “R” values are 0.96, 0.95, and 0.98 for HC, CO, and NOx, respectively. It further shows that the MLNN model is efficient in predicting emissions of tested engine with reasonable accuracy. Further, the relative error between experimental and predicted values is described using Equation (5).where and are experimental and predicted values, respectively.
Thereafter, the values of RE are average of all test conditions. The prediction ability of the model for all test conditions of the engine is measured using averaged RE and R between experimental and predicted HC, CO, and NOx values for all test conditions (Table 9).
5. Conclusions
In this manuscript, the use of the MLNN model for the prediction of HC, CO, and NOx from a four-stroke single-cylinder CI engine runs on WCO biodiesel added with MWCNTs has been investigated. The MLNN model developed in this work uses a backpropagation algorithm. The optimum structure 3-10-3 is chosen for emission prediction at any given test conditions of engine as it denotes the MSE equal to 0.095. Further, the prediction ability of the model for all test conditions of the engine is measured using averaged RE and correlation coefficient (R). The average RE between experimental and predicted HC, CO, and NOx values is found to be 0.83%, 1.01%, and 1.05%, respectively. The “R” between experimental and predicted HC, CO, and NOx are also given in Figures 18–20, and are found to be 0.96, 0.95, and 0.976, respectively. The results indicate that values predicted from MLNN and the experiments are close to one another. Furthermore, the comparison of experimental results and MLNN predicts shows that CI/diesel engines run on WCO biodiesel dispersed with MWCNTs can be correctly modeled through MLNN. Therefore, the use of MLNN is appropriate for predicting the emissions of the CI engine, minimizing the cost, time, and labor effort. Additionally, the model can also be fit into the controller.
Data Availability
All the data used to support the findings of this study are included within the article.
Conflicts of Interest
The authors declare that they have no conflicts of interest.
Authors’ Contributions
All authors contributed equally in the preparation of the manuscript.
Funding
The work has been financially supported by Vision Group on Science and Technology (VGST), Bengaluru 560001, Karnataka, India through Research Grant for Scientist/Faculty (RGS/F) Scheme (Letter No. KSTePS/VGST/RGS-F/GRDNo. 980/2020-21/104 dated 26/08/2021).