Optimizing Performance and Emissions of VCR Diesel Engines Using Grey-Taguchi Methods and ANN with Biodiesel from Waste Cooking Soybean Oil

Growing environmental concerns and the search for sustainable energy alternatives have focused a lot of attention on improving IC engine performance and emissions. we used cooking oil to produce biodiesel, a highly promising alternative fuel due to its renewable characteristics and potential to mitigate emissions of greenhouse gases, among other attributes [1]. However, biodiesel derived from edible oils (such as canola, palm, rapeseed, olive, sunflower, maize, and soybean) is a significant contributor to global food insecurity and expensive prices [2, 3]. However, out of all edible feedstocks, waste cooking oil (WCO) is a viable and potentially cost-effective alternative for biodiesel production [4]. Due to the presence of numerous contaminants (PAHs, PCBs, and dioxin), WCO is readily reusable from residences, hotels, and restaurants that discard it after frying meals [5, 6]. These contaminants pose a threat to the welfare of both animals and humans. Hence, following a pre-treatment procedure, WCO emerges as a viable alternative for biodiesel production while simultaneously addressing the issue of water contamination [6]. Produced from soybean waste cooking oil, biodiesel serves as an environmentally beneficial, cost-effective, and renewable alternative to conventional biodiesel. It addresses environmental and food security concerns by reducing production costs, diverting waste from landfills, and minimizing reliance on virgin vegetable oils, all while substantially reducing greenhouse gas emissions. Soybean oil, a renewable resource, provides a sustainable and consistent feedstock for biodiesel production. In comparison to biodiesel produced from virgin vegetable oils, biodiesel derived from residual cooking oil is a more cost-effective alternative [7-9].

To fully harness the potential of biodiesel blends, this study investigates their performance and emission characteristics under various conditions. This study focuses on optimizing a diesel engine operating at varied input engine parameters and fuel blending with waste-cooking soybean biodiesel blends. The integration of Grey Taguchi methodology and artificial neural networks (ANN) aids in the optimization process, aiming to achieve two key objectives: improving IC engine performance and reducing emissions [10-13]. The Grey Taguchi methodology provides a strong framework for optimizing in the presence of uncertainty through its efficient approach to handling constrained and unreliable data [14, 15]. Compression ratio and injection time highly affect engine performance and emissions [16]. A study by Hirkude et al. [17] looked into how injection timing, injection pressure, and compression ratio (CR) affected the performance and emissions of diesel engines that used a mix of biodiesel from cooking waste and diesel fuel. According to the findings, Enhanced brake thermal efficiency (BTE) alongside reduced brake-specific fuel consumption (BSFC). Although a slight decrease in BTE was observed as BSFC increased when CR was further raised from 18 to 19, this trend aligns with the typical behavior of engine performance characteristics. The optimal conditions for BSFC and BTE were identified at an initial IT of 27° b TDC, 250 bar IP, and a CR of 18. For the fuel that was tested, EGT increased in response to an increase in IP, IT, and CR, while smoke opacity (OP) reduced. Shrivastava and Verma [18] studied the impact of fuel injection pressure and engine load on biodiesel and blends derived from Roselle oil. it was documented that carbon dioxide emissions increased by 1.6% at an injection pressure of 220 bar, whereas smoke and nitrogen oxide emissions decreased by 2.20% and 3.18% respectively, for the RB20 blend in comparison to diesel fuel. Heng Teoh et al. [19] optimized blends of diesel-coconut oil fuel using the grey-Taguchi method (GTM) concerning several parameters, including blend, load, and speed. The optimization result determined the most effective blend ratio, at a speed of 3850 rpm and the load of 25% about the CI engine's emissions, and performance. An optimal configuration of engine load, blend, and speed for a VCR diesel engine was investigated by Gul et al. [13] by utilizing the Taguchi design which operates on pure biodiesel (B100) derived from WCO and a 20% biodiesel and diesel. The primary aim was to attain the most substantial reduction in pollutants and NOx emissions feasible. They used Gray relational analysis to identify the optimal input parameters that would produce desirable output. These parameters include using B100 gasoline, operating at a speed of 2300 rpm, and maintaining a load of 100%. Moreover, the ANOVA approach revealed that the kind of fuel is the primary determining factor, accounting for a significant 44.28% influence on the output parameters. Both the experiment and the artificial neural network (ANN) simulation, which used the expected best combination, confirmed the big improvements in the response factor of the output. This proves the use of the Grey-Taguchi approach for lowering the amount of emissions while simultaneously enhancing combustion and performance. The use of an ANN modeling methodology is effective in determining the necessary output factors when sufficient experimental data is available [20]. Modern neural systems are statistical modeling tools that cater to non-linear input and output data. It is easy to optimize the relationship between input and output parameters. We can conduct optimization studies across all working circumstances by using neural networks for engine predictions [21-23]. Backward feed propagation, when applied to ANN, results in the best possible output. For forecasting outputs, we can make use of MATLAB software along with an artificial neural network [3]. Researchers have conducted several studies on applying the Grey Taguchi technique to optimize engine performance and emission characteristics. In the past, researchers have only looked at the Grey-Taguchi method with a few parameters [24-27]. For example, they looked at waste-cooking biodiesel using the Grey-Taguchi analysis with the BTE performance parameter and the NOx emission parameter. Gul et al. [13] utilized the L9 Taguchi design to optimize fuel type, engine speed, and load for a diesel engine, which alternately fuels with waste cooking oil biodiesel (B100) and a 20% blend of biodiesel with neat diesel (B20). Our research uses a more extensive set of parameters and shows the method’s effectiveness in optimizing both performance and emissions simultaneously. ANN was used to predict the responses of the optimized parameters. We compare the ANN-predicted results with the actual experimentation. This approach addresses the limited parameter consideration gap in earlier studies.

3.1 Taguchi optimization

To address the issues, we employed the Taguchi technique in addition to grey relational analysis, which incorporates multiple responses. The S/N ratio proved to be the most effective approach for attaining optimal outcomes. The Taguchi technique is the most preferred method for finding the optimum solution with minimal trials. The Taguchi technique measures the standard characteristics using the signal-to-noise quantitative relation (S/N) [36]. The Taguchi approach enhances product quality by prioritizing a mean performance characteristic value that is close to the designated target value rather than one that falls within the specified limits. Figures 2 and 3 present the outcomes of the Taguchi study for the analysis of performance and emission characteristics of engines that include BSFC, BMEP, BTE, BP, NOx, CO, HC, and CO₂. We employ the Taguchi technique for optimization to determine the optimal input parameters. Figures 2(a), 2(b), 2(c), and 2(d) illustrate the analysis of the signal-to-noise ratio for the optimization of BTE, BP, BMEP, and SFC, respectively. The main effects plot shows that B30WCO and B40WCO biodiesel blends have higher SN ratios than other blends across all performance metrics (BTE, BP, and BMEP). These findings suggest they could boost engine efficiency and power. Higher compression ratios also tend to have a positive effect on BTE, BP, and BMEP. On the other hand, for specific fuel consumption (SFC), a lower SN ratio is better. The plot shows that some blends and higher compression ratios may help lower SFC. The study found that the B20WCO blend improved BTE, BP, and BMEP. It achieved the highest BTE when the engine was running at 18 CR, 30 MPa of pressure, and 16°b TDC. The same fuel mixture produced the largest brake pressure when the engine operated at a CR of 16, an IP of 60 MPa, and an IT of 16°b top dead center. To achieve the greatest brake mean effective pressure, the engine should operate on a blend of B20WCO fuel at a CR of 16, with an IP of 60 MPa and an IT of 16 °b TDC. Figures 3(a), 3(b), 3(c), and 3(d) individually exhibit the analysis of the S/N ratio for the optimization of NOx, CO, HC, and CO₂. We apply the "larger is better" criteria to BTE, BP, and BMEP, while we use the "smaller is better" criteria for SFC, NOx, CO, HC, and CO₂. The main effects plot study shows that higher biodiesel blends, particularly B30WCO and B40WCO, tend to have lower SN ratios for NOx, CO, and HC emissions, which could mean less of these pollutants. Also, higher compression ratios tend to lower SN ratios for both NOx and CO, which means they help lower these emissions. However, it's not as clear how the compression ratio changes HC emissions. The opposite is also true: higher biodiesel blends cause higher SN ratios for CO₂ emissions. This is because biodiesel naturally has more carbon content than diesel fuel. Intending to reduce emissions, the blend B30WCO showed the lowest NOx at 18 CR, 50 MPa, and 25°b TDC, while the blend B10WCO revealed the lowest value of CO at 17 CR, 60 MPa, and 25°b TDC. The lowest value of HC is shown by Blend B30WCO at 15 CR, 60 MPa, and 25°b TDC. Blend B30WCO showed the lowest value of CO₂ at 18 CR, 50 MPa, and 25°b TDC. The not-optimal in-cylinder pressure and temperature conditions caused by this compression ratio are responsible for the discrepancy in the signal-to-noise ratios observed at the 17:1 compression ratio. These conditions were not favorable to complete combustion of the biodiesel blend, resulting in reduced performance and emissions signals. This discovery emphasizes the susceptibility of biodiesel combustion to accurate engine configurations. A high CR can enhance BTE by obtaining additional energy from the fuel during combustion. When injected earlier with a high CR, the fuel undergoes more complete combustion, leading to a reduction in HC and CO emissions. ANOVA results indicate that injection pressure is the primary contributing factor. At a high injection pressure of 60 MPa, BTE is higher as a result of better atomization from the high injection pressure, which leads to more complete combustion. The increased oxygen content in WCO biodiesel can result in higher in-cylinder temperatures, which can facilitate the formation of NOx. Figure 3(a) also shows that the higher blends have higher NOx.

3.2 Grey analysis

Table 6 presents the data for 16 experiments (Exp. No. 1 to 16). Each experiment lists the S/N ratio values for each parameter. These values seem to have been normalized, as they all fall within the range of 0 to 1. Table 7 depicts the normalized values, and Table 8 represents the deviation sequence. The deviation sequence calculated after normalization helps determine the degree of similarity between data sequences. Table 9 presents the grey relational grade and ranking. Table 10 probably summarizes the impact of different factors on the average Grey Relational Grade (GRG).

3.3 Anova-based analysis

Furthermore, ANOVA was performed to identify which parameter has a considerable effect on the output of the engine as well as the significance of individual input parameters. By illustrating the interrelationships among these variables, analysis of variance (ANOVA) facilitates the finding of the optimal set of input parameters that optimize engine efficiency and reduce emissions. ANOVA is carried out considering the grey relational grade and the values are presented in Table 11. ANOVA results revealed that the Injection Pressure (IP) is the significant parameter affecting multiresponse with a minimum P value and a contribution of 52.20%. It was found that for the B30WCO blend, BTE increases while SFC decreases with increasing Injection Pressure. This is because atomization works better at higher IP, exposing more of the fuel droplet's surface to the hot air and causing it to completely burn. Figure 2 also reveals that as IP increases, CO decreases, which is the result of complete combustion due to a good-quality fuel mixture. Increasing the injection pressure from 50 to 60 MPa results in an increase in NOx and CO₂. This is because higher injection pressures can improve atomization and mixing, which can improve combustion efficiency. However, higher injection pressures can also result in higher peak temperatures, which can contribute to NOx formation. The amount of oxygen present during combustion influences NOx formation. Increasing injection pressure may alter oxygen availability and affect NOx emissions. The reason for the increase of CO₂ is complete combustion due to high IP which also implies a higher consumption of oxygen, leading to an increase in CO₂ production.

2.png

Figure 2. S/N ratio plots for a) BTE, b) BP, c) BMEP, d) SFC

3.png

Figure 3. S/N ratio plots for a) NOx, b) CO, c) HC, d) CO₂

Table 6. Signal to noise ratio (S/N ratio)

Table 7. Normalize data

Table 8. Deviation sequence

3.4 Result validates using ANN

Artificial neural networks (ANNs) enhance the experimental optimization process by predictively optimizing engine performance and emissions. We train artificial neural networks (ANNs), known for their ability to represent complex nonlinear relationships, using experimental data to predict engine behavior under various working conditions. This study employs a feed-forward back propagation network with three neural layers. Input, hidden, and output layers contain four (LOGSIG), ten (LOGSIG), and eight (PURLINE), respectively, neurons. Using eighteen experimental results, the hidden layer number of neurons is determined through trial and continued Continue iterating until the mean squared error between the real experimental data and the expected output data is reduced to its minimum value. The ANN model is trained using the 'Trainlm' function, which adjusts the weight and bias values by Levenberg-Marquardt optimization. We used the ANN model to predict engine performance and emissions due to its ability to efficiently capture complex, nonlinear relationships between input parameters (CR, IP, IT, and load) and output variables. The model's flexibility and predictive accuracy make it ideal for engine setting optimization and emissions improvement in real-world applications.

Figure 4 illustrates a main effects plot for SN ratios. Blend, CR, IP, and IT are the various factors that influence the mean signal-to-noise (SN) ratio. This plot serves as a visual representation of these relationships. SN Ratio: A measure of signal quality relative to noise. A larger SN ratio generally indicates better performance or quality. The factor "Blend" appears to have a significant impact on the SN ratio. B30WCO appears to yield the highest mean SN ratio. Compared to Blend, the effects of CR, IP, and IT on the SN ratio appear less pronounced.

Table 9. Grey relational grade and rank

Table 10. Main effect on mean GRG

4.png

Figure 4. Main effects plot for GRG using S/N ratio

Table 11. Results on ANOVA on grey relational grade

5.png

Figure 5. Neural network structure

6.png

Figure 6. Regression analysis of ANN

Optimizing engine performance and emissions through the interplay of engine operating parameters and the inherent characteristics of biodiesel blends. Injection pressure was found to be the most important parameter, affecting the atomization and mixing of fuel with air. Increased injection pressures facilitate smaller fuel droplets, enhancing a uniform air-fuel mixture and optimizing combustion efficiency. This mechanism is especially evident in the improved Brake Thermal Efficiency (BTE) and lower unburned emissions (CO and HC) observed at elevated injection pressures. The Grey-Taguchi optimization framework effectively addressed the trade-offs between performance metrics and emissions. The fuel properties of biodiesel blends, including viscosity, density, and calorific value, justify the observed trends. The slightly elevated viscosity of biodiesel blends (e.g., 2.43 cSt for B30WCO versus 2.09 cSt for diesel) influences spray formation and penetration, demanding optimal injection pressures to achieve effective atomization. The lower calorific value of biodiesel blends (e.g., 9995 cal/g for B30WCO versus 10235 cal/g for diesel) leads to a higher BSFC, needing additional fuel to achieve comparable power output. Although this, the B30WCO blend exhibited considerable emission reductions of 33.3% in NOx, 22.5% in HC, and 14.06% in CO₂ attributable to the oxygenated characteristics of biodiesel, which optimize the combustion process and reduce incomplete combustion. The analysis underlines the significance of density, which increases with elevated biodiesel content (e.g., 841 kg/m³ for B40WCO versus 816 kg/m³ for diesel). Higher density fuels enhance spray depth but demand precise adjustment of injection timing to prevent excess combustion delays. The flash and fire points of biodiesel blends, which are considerably elevated compared to diesel, enhance safety and stability during operation, rendering biodiesel an acceptable substitute in practical applications.

These findings confirm the optimization process that utilizes the Grey-Taguchi method to determine the optimal parameter combinations. The application of the B30WCO blend in a VCR engine exhibits this balance, showcasing how optimized injection pressures and timing can reduce the limitation of biodiesel's properties while obtaining environmental and performance advantages. Furthermore, the ANN model validated the effectiveness of these optimizations by precisely forecasting the intricate interconnections between engine parameters and results.

Figure 5 depicts this study's architecture of the Artificial Neural Network (ANN) model. The network consists of an input layer with four neurons representing the input parameters: CR, IP, IT, and load. The hidden layer contains 10 neurons, which process the inputs using weighted sums and biases, followed by a non-linear activation function. The output layer has eight neurons corresponding to the predicted engine performance and emission characteristics, including BTE, BP, BMEP, SFC, NOx, CO, HC, and CO₂. The ANN was trained to model the non-linear relationships between input parameters and output metrics, enabling accurate prediction of engine behavior under different operating conditions.

Figure 6 contains a visual representation of the performance of an artificial neural network (ANN) model in predicting a target variable. It contains plots for the training, validation, and testing datasets, as well as critical metrics such as R-squared values and output comparisons. The model's high R² values for the training, validation, and testing datasets (0.97949, 0.99971, and 0.98291, respectively) indicate that it has a strong ability for generalization and fits the data well. In all cases, the 'Fit' lines closely follow the 'Data' points, indicating a strong correlation between the predicted and actual values. Figure 7 shows ANN model regression performance in predicting engine performance and emissions. The model inputs are CR, IP, IT, and Load. The outputs are BTE, BP, BMEP, SFC, NOx, CO, HC, and CO₂ emissions. Regression plots show predicted vs. actual results for training (top left), validation (top right), testing (bottom left), and all data combined (bottom right).

3.5 Comparison of results under optimal conditions

The most optimum combination of inputs was determined by the Taguchi-grey method (B30WCO, 18 CR, and IP 60 bar IP and IT 25 °b TDC) was used to compare the diesel and biodiesel response parameters, as illustrated in Figure 8. These results indicate that certain responses to biodiesel were positive, while others were not as favorable when compared to standard diesel. As a result, we observe significant improvements in response outcomes, including emission characteristics. In the meantime, we observe only minor degradation in the responses to engine performance parameters.

7.png

Figure 7. Experimental and ANN predicted results

8.png

Figure 8. Comparison of engine output parameters

3.6 Experimental and ANN predicted results

We compared the experimental results to the Artificial Neural Network's (ANN) predicted results. The results demonstrate that the ANN methodology can effectively be used to estimate the performance and characteristics of I.C engines. This can be achieved with a reduced number of trials, eliminating the need for a detailed experimental study. As a result, both engineering effort and costs are minimized.

3.7 Uncertainty analysis

In experimental research, statistical analysis of uncertainty is essential because instruments used to measure various parameters are prone to random errors caused by factors such as equipment conditions, laboratory calibration, environmental variables, and reading measurements. Therefore, we use a mathematical expression known as the propagation of errors to minimize the probability of errors occurring. During experiments, equipment errors can help determine the uncertainties in parameter values estimated by instruments. Eq. (9) is utilized to determine the overall percentage uncertainty. Each parameter in the equation contributes to the engine's overall performance and emissions profile. Each parameter's uncertainty reflects the potential variability or measurement error, thereby influencing the reliability of the experimental results.

$\begin{gathered}\text{Total percentage uncertainties}=\sqrt{\left[\begin{array}{c}(\mathrm{CO})^2+(\mathrm{NoX})^2+(\mathrm{HC})^2+\left(\mathrm{CO}_2\right)^2 \\+(\mathrm{BP})^2+(\mathrm{SFC})^2 \\+(\text { Pressure })^2+(\text { Torque })^2 \\+(\text { Temp })^2+(\text { Fluid flow })^2\end{array}\right]} \\=\sqrt{\left[\begin{array}{c}\left.(0.1)^2+(0.01)^2+(0.47)^2+(0.07)^2+(0.1)^2\right] \\+(3.1)^2+(0.1)^2+(0.2)^2+(0.1)^2+(0.12)^2\end{array}\right]} = \pm 3.15\end{gathered}$           (9)

It is essential to consider the physical significance of these results to interpret the reliability and accuracy of the experimental results. For example, uncertainties in emissions measurements, influenced by factors such as sensor precision and environmental conditions, indicate the variability in detecting the precise levels of pollutants (CO, NOx, HC, and CO₂). Similarly, uncertainties reflect the sensitivity of engine performance metrics (BP and SFC) to operational conditions, such as fuel quality and engine temperature. The total uncertainty is 3.15%. When making broader generalizations, it is necessary to consider a quantifiable margin of error. The observed trends are statistically significant and robust, as this level of uncertainty confirms. However, it also demonstrates the importance of careful consideration in experimental design and measurement technique precision.

3.8 Comparison of engine output parameters for B30WCO and diesel

B30WCO demonstrated enhanced fuel efficiency in comparison to B00 diesel, as evidenced by its reduced BSFC. In addition, the biodiesel blend exhibited a substantial decrease in hydrocarbon (HC) and carbon monoxide (CO) emissions, suggesting a beneficial environmental effect. Nevertheless, B30WCO exhibited a slight decrease in BMEP and BP, which implies a potential decrease in engine power output. Although the biodiesel blend demonstrated a higher brake thermal efficiency (BTE), indicating improved energy conversion, it also resulted in slightly increased carbon dioxide (CO₂) emissions as a result of its higher carbon content. The substantial increase in nitrogen oxide (NOx) emissions from B30WCO was a critical concern, requiring additional research to develop mitigation strategies.