Statistical and machine learning analysis of diesel engines fueled with Moringa oleifera biodiesel doped with 1-hexanol and Zr₂O₃ nanoparticles

Abstract

This experimental study based on DOE (Design of experiments) explores the performance and emission characteristics of Moringa oleifera-based biodiesel blends enhanced with zirconium oxide (ZrO₂) and 1-hexanol as boosting agents in a slow-speed diesel engine operating at 1500 rpm. The novelty lies in the synergistic use of these additives for improving fuel efficiency and reducing emissions, combined with advanced statistical and machine learning models for optimization and prediction. Four test blends were analyzed: 90D5MO5H + 25 ppm ZrO₂, 80D10MO10H + 50 ppm ZrO₂, 70D15MO15H + 75 ppm ZrO₂, and 100MO + 100 ppm ZrO₂. A comprehensive methodology involving experimental testing and statistical modelling using Gradient Boosting (GBoost), Extreme Learning Machine (ELM), and Response Surface Methodology (RSM) was employed. Key findings include a brake thermal efficiency (BTE) of 8.63% higher than diesel and a fuel consumption reduction of 46.13% (0.14 kg/kWh) for the 90D5MO5H + 25 ppm ZrO₂ blend. This blend also demonstrated superior combustion characteristics, including a peak cylinder pressure of 70 bar and a heat release rate (HRR) of 45 J/°CA. Emission analysis revealed significantly reduced hydrocarbon emissions (0.020%) for 100MO + 100 ppm ZrO₂ and the lowest carbon monoxide emissions (10.1%) for 90D5MO5H + 25 ppm ZrO₂. Among predictive models, ELM exhibited the highest accuracy with an R² value of 0.9604, outperforming other approaches. The findings suggest that optimized moringa oleifera blends with zirconium oxide and 1-hexanol offer a promising solution for sustainable and cleaner diesel engine operation, with potential applications in transportation and energy sectors aiming for reduced environmental impact.

Introduction

Petroleum and gas are easily stored, portable, and have a high chemical reaction perspective, so they are used as the primary means of energy in internal combustion engines¹. The significance of renewable gasoline, which can serve as a substitute for natural gas, is increasing steadily. This is driven by the dwindling availability of natural deposits in certain regions, their ongoing depletion, and the environmental issues they cause, including air pollution and harm to surrounding ecosystems². Minimizing the use of petroleum and diesel in engine combustion (ICE) and consequently lowering the emission of risky exhaust pollutants have been the focus of investigators and researchers for a considerable period^3,4. The demand and use for petroleum and coal are rapidly rising daily due to the growing number of vehicles on the road. In tandem with the decline of the supply and demand equilibrium, the costs of oil and gas also rise swiftly. Implementing innovative fuel options in internal combustion engines is a reasonable strategy to fulfil these increasing needs⁵. The term “alternative fuel” refers to any substance capable of replacing natural gas in engines equipped with internal combustion systems. The feasibility of using petroleum as a substitute for natural gas depends on its characteristics, such as cost-effectiveness, ethical acceptability, and availability⁶.

Moringa oleifera seeds, known for their high mono-unsaturated to saturated fatty acid (MUFA/SFA) ratio, as well as their rich content of sterols, tocopherols, and sulfur-containing amino acid peptides, are increasingly recognized as a valuable resource for a variety of food and non-food applications⁷. The swift proliferation of Moringa shrubs in tropical and tropical regions, even during extended periods of dryness, establishes the species just like a dependable asset to improve the nutritional well-being of local communities and, if improved farming methods are utilized, their financial situation, as it may produce biodiesel fuel from a source that does not compete alongside food for humans crops. There has been a single human pharmaceutical activity investigation into the seed as well as oil, despite its widespread use as conventional medicine. Certain positive testimony, nevertheless, warrants fresh attempts to get precise and conclusive facts regarding any benefits to people that accompany seedling consumption⁸. An extensive analysis of scientific information on the structure of oil derived from Moringa has initiated a systematic strategy for additional studies. These research investigations, utilizing the seeds as well as fuel, will concentrate on cultivating methods to enhance plant yield and will examine the health impacts on people who consume both the seeds and oil produced from them⁹.

A comparative analysis of the engine efficiency and pollution characteristics of diesel engines powered by moringa biodiesel, palm biodiesel as the jatropha plant biodiesel, plus diesel fuel is the objective of this work. In this work, a mere 20% of each biofuel (referred to as MB20, PB20, as well as JB is used to 20%, correspondingly) was evaluated in a diesel engine. This decision was based on the fact that the potential application of biodiesel as much as twenty percent in diesel engines requires adjustments. Furthermore, the chemical and physical properties of all the fuel samples are provided and compared to the requirements set by ASTM A6751. All biodiesel fuel types showed lower brake force (BP) and higher pedal-specific fuel consumption (BSFC) compared to diesel fuel. An experimental fuel blend including the RMA10 gasoline and FAME biofuel was used in the study. Specifically, the discharge of nitrogen dioxide in exhaust gases decreased by 24.1%, while the release of carbon dioxide in emission gasses diminished by 25.1%. Nevertheless, the economic effectiveness of diesel engine operation declined when biofuel was used, as the particular efficient fuel consumption rose by 9.7%. An optimal extent of sustainable development could be achieved by employing a fuel blend of 10 to 15 per cent renewables in an 80% diesel load. This scenario reduced nitrogen oxide emissions by 25.1%, and carbon dioxide oxide emissions by 19.81% more than conventional fuel¹⁰.

The Artificial neural networks were developed for different biodiesels to determine the performance parameters and also from the parameters such as injection timings and from the test results it is understood that the percentage of error concerning different exhaust gas parameters was very low compared to conventional experiments. Hence this study is useful for researchers willing to minimize emissions without any further engine modifications¹¹. From innovative experimentation¹², crude palm oil was investigated in diesel engines to evaluate performance parameters across different test trials, including blends of 20% crude palm oil + 80% diesel, 30% crude palm oil + 70% diesel, and 40% crude palm oil + 60% diesel, over a total operating time of 380 h. These tests revealed that the lowest emissions were achieved with 100% crude palm oil due to its higher oxidation capacity compared to the blended fuels.

The use of nickel nanoparticles in a diesel engine fueled with Neem biodiesel, blended with varying concentrations of methyl ester reactants, has been shown to enhance combustion parameters, resulting in improved performance and reduced emissions. Specifically, hydrocarbon emissions decreased by 2.2%, carbon monoxide emissions by 3.8%, and nitrogen oxide emissions by 11.3%. This improvement is attributed to the nickel nano catalyst’s effectiveness as a reactive agent in enhancing the physical properties of the fuel blends, as demonstrated in prior studies^13,14. Recent studies^15,16 have explored the properties of methanol-biodiesel blends for Azadirachta indica biodiesel-diesel mixtures, focusing on two formulations: 10% methanol + 90% biodiesel and 20% methanol + 80% biodiesel. The investigation included evaluating kinematic viscosity, flash point, fire point, and cetane number in accordance with marine protocol standards. The results revealed that the 20% methanol + 80% biodiesel blend performed best, offering a calorific value comparable to diesel. Specifically, the 20% blend showed a 7.2% increase in calorific value, a 40% increase in kinematic viscosity, and a 23% improvement in cetane index compared to pure diesel. These findings indicate that biodiesel blends with a 20% methanol content can enhance the fuel properties and performance of marine diesel engines. Studies^17,18 evaluated hone biodiesel blends under compression ratios of 17.5 to 18 and injection pressures of 180, 210, and 240 bar. The blend with 20% hone biodiesel exhibited the best performance at an injection pressure of 210 bar, achieving higher calorific values and reducing brake-specific fuel consumption (BSFC) by 0.05 kg/kWh compared to diesel. Similarly¹⁹, highlighted the superior performance of hone biodiesel across varying compression ratios and injection pressures, demonstrating its potential as an efficient and sustainable alternative fuel.

While numerous experiments have explored the application of Moringa-based biofuels in diesel engines, they typically demonstrate minor improvements in engine performance and slight reductions in emissions, often necessitating engine modifications to achieve these results. However, the potential of hexanol as a boosting fuel in biodiesel blends remains largely unexplored, with limited testing and minimal focus on its ability to enhance combustion and overall engine efficiency. Previous studies have also lacked comprehensive insights into the synergistic role of hexanol with catalysts like zirconium oxide (ZrO₂) in biodiesel applications and their optimization through advanced computational techniques. Motivated by the need for sustainable and environmentally friendly fuel alternatives, this work integrates experimental testing with machine learning models like Gradient Boosting (GBoost), Extreme Learning Machine (ELM), and Response Surface Methodology (RSM) to predict and optimize engine performance and emissions. The novelty lies in this dual approach, combining innovative fuel formulations with state-of-the-art predictive tools. The aim is to evaluate the performance, combustion, and emission characteristics of these biodiesel blends, offering a sustainable solution with real-world applications in the transportation and energy sectors while contributing to reduced environmental impact.

Materials and Methods

Materials

The materials used in this analysis are additives such as Zr₂O₃ and the ignition improver is 1-hexagonal. The materials are purchased from alridtch enterprises with a cost of RS 200 per kg and 1 hexagonal from the local supplier at Chennai Ambattur cost of RS 400 per litre. The raw biodiesel called moringa oleferia biodiesel is purchased from Andhra Pradesh per litre Rs 1200. The physical properties of the Zr₂O₃ Catalyst are mentioned in Table 1.

Table 1 Physical properties of the Zr₂O₃ catalyst.

Methods

The blends were proportionated by the following terms named 90D5MO5H + 25 ppm Zr₂O_3, 80D10MO10H + 50 ppm Zr₂O_{3 +} 70D15MO15H + 75 ppm Zr₂O₃ and 100 MO + 100 ppm Zr₂O_3. All the results obtained from the engine are compared with standard diesel. These blends’ physical and chemical properties are tested at the local testing Centre available at the EETA laboratory. A few tests such as kinematic viscosity and cetane numbers are tested at the Indian Institute of Madras Chennai. The ASTM standards are shown in the table to better understand the properties of the blends obtained during the tests.

Experimental Setup

The experimental apparatus set up used in this study is schematically shown in Fig. 1. The engine’s equipment comprised a thermocouple for measuring temperature, load cells, a tachometer, the Froude hydraulic device, and a fuel consumption measuring container with a precision of ± 0.5. A performance assessment of the combustion system was conducted using various fuel blends, ranging from 0–30% biodiesel by volume in diesel¹⁹. The combustion characteristics were analyzed, and the results were presented through graphs plotting key parameters such as pressure, temperature, and heat release rate over the engine cycle in Section “Combustion characteristics”. To investigate how different biodiesel blend ratios affect engine performance and emissions, and to determine whether higher biofuel concentrations can be used in diesel-powered vehicles without significant adverse effects, a higher percentage of biofuel was tested in this experiment. Initial trials were conducted to ensure a controlled water flow into the centrifuge and maintain the necessary torque for the pump. The engine was tested under torque conditions ranging from 20 to 100 Nm. Following every engine test, the speed, consumption of fuel (FC), brake power (BP), brake-specific fuel expenditure (BSFC), fuel equal power (FEP), and brake thermal efficiency (BTE) were measured and calculated for each torque condition. Following data collection, the entire process was repeated for all torques that were tested. The performance characteristics were calculated following the methodology employed. More and more studies are developing to test appropriate biodiesel blends in diesel engines to determine their performance and emissions quality, even though ASTM guidelines permit fuel mixes between 5 and 20%. Every single test was executed with a constant torque working condition ranging from 20 N m to 100 N m, with increments of 20 N m. The engine torque was manipulated by regulating the water flow being directed into the dynamometer. The transmission pumping water into the A dynamometer was employed to acquire various engine torques, which could be directly measured through the instrumentation board²⁰. Table 2 presents the engine specifications, while Table 3 provides the properties of the fuel blends.

Fig. 1

Experimental setup¹ (a) Actual set-up (b) Schematic diagram.

Table 2 Engine specifications.

Table 3 Properties of the fuel blends.

Uncertainty analysis

Defects and ambiguities may arise through several sources such as instrument design and measurement, fluctuating surroundings, assessments and observations, and so on. Broadly speaking, doubt can be categorized into two primary components, particularly specified mistakes and unplanned mistakes²¹. The former situation pertains to the aspect of consistency, whilst the second one encompasses the aspect of statistical measures. Unpredictability associated with the equipment employed for this field study is shown in Table 4, while the uncertainties in the observed values are provided in Table 4. The present study evaluates the uncertainty of the observed variable (DX) using a Gaussian distribution, as defined in Eq. (2), with a confidence limit of ± 2 s. The value 2 s represents the median range within which 95% of the measurements fall. Equation (1) is referenced from²².

$$\Delta {\text{R}} = \sqrt {\left( {\frac{{\partial {\text{P}}}}{{\partial {\text{p}}_{1} }}\Delta {\text{d}}_{1} } \right)^{2} + \left( {\frac{{\partial {\text{P}}}}{{\partial {\text{p}}_{2} }}\Delta {\text{d}}_{2} } \right)^{2} + \left( {\frac{{\partial {\text{P}}}}{{\partial {\text{p}}_{3} }}\Delta {\text{d}}_{3} } \right)^{2} + \cdots \left( {\frac{{\partial {\text{P}}}}{{\partial {\text{p}}_{{\text{n}}} }}\Delta {\text{d}}_{{\text{n}}} } \right)^{2} }$$

(1)

where “∂P/(∂P₁), ∂P/(∂p₂ ) and ∆d₁, ∆d₂, ∆d₃ represent the ratio of deviations to standard data and degree of uncertainties at repeated readings, respectively.

Table 4 Uncertainty analysis.

The actual readings and deviated readings due to the atmospheric and other defects can be found using Eq. (2) as done by²⁰

$$\sum P_{a} = \frac{{2\sigma_{p} }}{{\overline{x}_{np} }}*100$$

(2)

where P_a = Actual Measurements taken from the experimental set, σ_p = Result deviations from the experimental set and x_np = The intensity of the uncertainty.

Response surface methodology

A potent statistical tool for optimizing complex systems and processes is Response Surface Methodology (RSM). It analyses and enhances the links between multiple input variables and the output response using mathematical and statistical models. RSM’s main goal is to identify ideal conditions for desired responses by means of a sequence of experimental runs by means of interactions between variables. Usually using design of experiments ideas, RSM creates data to be used in empirical model construction. Often quadratic in character, these models offer a surface that reflects the response as a function of the input variables. Using surface analysis, researchers can pinpoint the parameters that either enhance or limit the response variable⁵.

RSM employs common designs such as Box-Behnken designs and central composite designs (CCD). For appropriate second-order models, CCD is especially useful; it also efficiently explores curvature in the response surface. Conversely, box-Behnken designs help to minimize the number of experimental runs while nonetheless offering a full picture of the interactions among the components³. RSM’s iterative character lets one improve experiments depending on the first results, increasing precision and efficiency. Applied extensively in many disciplines, including engineering, chemical processing, and product development, this approach helps to improve performance, lower costs, and raise quality in all spheres.

Soft computing approaches

Gradient boosting

In the machine learning paradigm, Gradient Boosting (GBoost) regression is a method whereby a predictive model is iteratively created by aggregating the predictions of weak learners typically decision trees—to generate a strong learner. It is predicated on the boosting idea, in which each new model concentrates on fixing the errors created by the last one after the sequential training of models. GBoost fundamental concept is to use gradient descent to reduce the loss function of mean squared error. Every iteration a fresh weak learner usually a shallow decision tree is included in the model. Designed to roughly match the negative gradient of the loss function about the true output, this new learner fits the residuals, or errors, of the previous model. This method repeatedly improves the model by guiding the next learner to concentrate on the challenging-to-predict cases. Often the mean of the target values for regression issues, the boosting procedure starts with an initial forecast. The model then progressively adapts a weak learner to the residual errors, modulating its predictions to lower the total error. Following every stage, the model updates the forecast by aggregating the output of the present model with a weighted form of the new learner’s predictions. The learning rate parameter lets the model adjust its performance and avoid overfitting by controlling the amount each learner adds to the general prediction.

Although GBR supports several loss functions and is somewhat flexible, for regression the technique usually uses squared error loss. Furthermore, typically employed to improve generalization and lower overfitting are regularizing methods including shrinkage and subsampling. Gradient Boosting Regression essentially uses weak models to progressively increase performance while being computationally effective and scalable.

Gradient Boosting is an ensemble technique to improve the prediction accuracy by combining the output of multiple weak learners (typically decision trees) in a sequential manner. Each new learner focuses on the errors made by the previous ones, thereby reducing the model’s overall bias and variance.

Here’s how it can be applied to datasets (Fig. 2):

• 1. Pre-processing the dataset:

  • Input Features: The dataset is split into training, validation, and test sets.

  • Normalization/Standardization: Data is cleaned and normalized to ensure consistent scales across features.

• 2. Training the gradient boosting model:

  • Initial Prediction: A simple model (e.g., predicting the mean value of the target) is created as a baseline.

  • Error Computation: The residuals (errors) between the baseline prediction and actual target values are computed.

  • Weak Learner Addition: Decision trees are sequentially trained on these residuals. Each tree focuses on correcting the errors of the previous model.

  • Learning Rate: A learning rate is applied to control the contribution of each tree, preventing over fitting.

• 3. Hyper parameter tuning:

  • Parameters like the number of trees, tree depth, and learning rate are tuned using cross-validation for optimal performance.

• 4. Model evaluation:

  • The trained model is evaluated using metrics such as mean squared error, accuracy, or F1-score on validation/test datasets.

• 5. Final prediction:

  • The final model aggregates the predictions of all weak learners for robust results.

Fig. 2

Steps for gradient boosting and EML modelling.

Extreme learning machine

Extreme Learning Machine (ELM) is a fast and efficient machine learning algorithm primarily used for regression and classification tasks. Based on a single-hidden-layer feedforward neural network (SLFN), it presents a clear benefit over conventional neural networks by doing away with iterative training processes. In ELM, the weights between the input layer and hidden layer as well as the biases for the hidden neurons are randomly generated and remain constant all through the training phase. The weights between the hidden layer and the output layer alone are trained in the model. ELM is fundamentally based on the approximation theory, in which the input data is converted into a high-dimensional space and the hidden neurons can be seen as feature mapping units. Since it guarantees that the hidden neurons generate nonlinear transformations of the input data, hence producing more robust feature representations, the random initialization of input weights and biases is vital. Usually utilizing a least squares method, the output weights are analytically found once the hidden layer outputs are generated by minimizing the error between the predicted and actual values. ELM is quite quick relative to conventional neural networks because of this architecture since it avoids the requirement for back propagation and gradient descent. Nonetheless, as they can influence the generalization capacity and performance of the model, the number of hidden neurons and their matching weights must be carefully chosen. We know that various research has utilized ANOVA methodologies to examine the efficiency and pollution levels of biofuel vehicles. One study, for instance, looked into the effectiveness and pollution qualities of combinations of biodiesel composed of Moringa blends. By combining such oils with fuel oil, they were able to improve efficiency metrics while cutting pollutants by thirty per cent, according to their analysis of variance. Table 5 represents ANOVA results. Some research examines data on vehicle emissions using ANOVA. One research optimized a light-duty diesel engine burning and pollution using the analysis of variance method. Nitrogen oxides, or smoke, and particular energy use were all significantly reduced in the findings. Table 6 represents engine emissions data.

Table 5 ANOVA results for engine performance data.

Table 6 ANOVA results for engine emission data.

The Extreme Learning Machine is a fast, single-hidden-layer feed forward neural network. It randomly assigns input weights and biases and only trains the output weights, making it computationally efficient. Here’s its application (Fig. 2):

• 1. Pre-processing the dataset:

  • Input features are prepared similarly to Gradient Boosting, ensuring data consistency and feature scaling.

• 2. ELM training process:

  • Input Weights Initialization: Random weights and biases are assigned to the connections between the input and hidden layers.

  • Hidden Layer Activation: The input features are transformed using a nonlinear activation function (e.g., sigmoid or ReLU).

  • Output Weights Calculation: Using the transformed data, output weights are computed analytically (e.g., using a Moore–Penrose pseudoinverse) to minimize the error.

• 3. Hyper parameter selection:

  • Parameters such as the number of hidden neurons and the type of activation function are selected based on the dataset.

• 4. Model evaluation:

  • The ELM model is evaluated using the same performance metrics as Gradient Boosting.

• 5. Final prediction:

  • The trained model is used to make predictions on unseen test data.

Development of RSM models

Statistical interpretation of model data

To analyse the engine performance data, an Analysis of Variance (ANOVA) was conducted to evaluate the statistical significance of the results. The following Table 5 presents the ANOVA results for the engine performance parameters, which include the effects of different fuel blends and operating conditions on key performance indicators.

Similarly, an Analysis of Variance (ANOVA) was applied to assess the engine emission data. The following Table 6 presents the ANOVA results for engine emissions, highlighting the impact of various fuel blends and operational factors on emission levels.

$$\begin{gathered} {\text{BTE}} = 308.23 + 3.66*{\text{BP}} - 0.016*{\text{LHV}} + 0.00012 \hfill \\ \quad \quad \quad \;\;*{\text{BP}}*{\text{LHV}} - 0.78*{\text{BP}}^{2} + 0.00000022*{\text{LHV}}^{2} \hfill \\ \end{gathered}$$

(3)

$$\begin{gathered} {\text{BSFC}} = 6.23{-}0.43*{\text{BP}}{-}0.00022*{\text{LHV}} + 0.0000064 \hfill \\ \quad \quad \quad \;\;*{\text{BP}}*{\text{LHV}} + 0.0185*{\text{BP}}^{2} + 2.16228{\text{E}} - 009*{\text{LHV}}^{2} \hfill \\ \end{gathered}$$

(4)

$$\begin{gathered} {\text{CO}} = - 0.66{-}0.025*{\text{BP}} + 0.000037*{\text{LHV}} + 0.000003 \hfill \\ \quad \quad \quad \;\;*{\text{BP}}*{\text{LHV}} + 0.00099*{\text{BP}}^{2} - 4.83622{\text{E}} - 010*{\text{LHV}}^{2} \hfill \\ \end{gathered}$$

(5)

$$\begin{gathered} {\text{HC}} = - 2789.95{-}3.98*{\text{BP}} + 0.136*{\text{LHV}}{-}0.000084 \hfill \\ \quad \quad \quad *{\text{BP}}*{\text{LHV}} + 2.29*{\text{BP}}^{2} - 1.63246{\text{E}} - 00*{\text{LHV}}^{2} \hfill \\ \end{gathered}$$

(6)

$$\begin{gathered} {\text{CO}}_{2} = 191.11 + 3.24*{\text{BP}}{-}0.0094*{\text{LHV}}{-}0.000058 \hfill \\ \quad \quad \quad *{\text{BP}}*{\text{LHV}} + 0.135*{\text{BP}}^{2} + 1.17720{\text{E}} - 007*{\text{LHV}}^{2} \hfill \\ \end{gathered}$$

(7)

$$\begin{gathered} {\text{NOx}} = 23425.89 + 95.19*{\text{BP}} - 1.165*{\text{LHV}} + 0.0125 \hfill \\ \quad \quad \quad *{\text{BP}}*{\text{LHV}} - 42.83*{\text{BP}}^{2} + 0.000015*{\text{LHV}}^{2} \hfill \\ \end{gathered}$$

(8)

BTE model

The BTE model was developed using ANOVA. It is shown in the mathematical form as Eq. (3). The ANOVA results are listed in Table 5. The surface diagram for BSFC model is depicted in Fig. 3a. Examining variance (ANOVA) for the response surface quadratic model provides noteworthy results for the response variable "BTE." With a very low probability (p < 0.0001), the model shows an F-value of 714.85, meaning that noise is statistically significant about this result. With a mean square value of 90.37, the model’s total of squares is 451.87, dispersed over 5 degrees of freedom. Important elements causing the model’s relevance are A-BP, B-LHV, A², and B². With an F-value of 2944.76 and a p-value of less than 0.0001, the factor A-BP shows particularly great individual contribution. Analogous to this, the factor B-LHV displays an F-value of 399.34 (p < 0.0001). With an F-value of 4.77, the interaction term AB has a p-value of 0.0538, therefore suggesting marginal significance. With F-values of 219.27 (p < 0.0001) and 7.66 (p = 0.0199), respectively, both quadratic terms, A² and B², are likewise significant. With a sum of squares of 1.26 and 10 degrees of freedom, the residual error—which reflects inexplicable variance is small. The good fit of the model to the data is underlined by its total sum of squares, 453.13. Important model terms imply that it might not be required to continue model refinement or reduction. The surface diagram shows that peak BTE is observed with low biodiesel blends and full engine load as described in experimental section.

Fig. 3

Surface diagram for (a) BTE (b) BSFC (c) CO (d) HC (e) CO₂ (f) NOx models.

BSFC model

The BSFC model was developed using ANOVA as shown in the form of algebraic expression in Eq. (4). The ANOVA results are listed in Table 5. The surface diagram for BSFC model is depicted in Fig. 3b. It is noted that lowest BSFC is attained at similar locations on surface diagram as was the case of peal BTE i.e. higher LHV and fuel engine loads. With an F-value of 125.14 and a p-value of less than 0.0001, the ANOVA for the response surface quadratic model connected to “BSFC” reveals that the model is very significant and indicates a very low possibility that such a result could arise owing of noise. With a mean square value of 0.026 the model’s sum of squares, distributed over 5 degrees of freedom, is 0.13. With F-values of 424.22 and 100.41 respectively and p-values less than 0.0001, A-BP and B-LHV are among the model terms with great significance. With an F-value of 8.57 (p = 0.0151) the interaction term AB is particularly noteworthy. Furthermore, whilst the term B² is not significant with a p-value of 0.5135, the quadratic term A² shows an F-value of 75.61 (p < 0.0001. Low unexplained variance is shown from the minimal residual sum of squares at 2.067E-003. With the major terms A, B, AB, and A² helping to define the model’s efficacy, the overall sum of squares for the model is 0.13, therefore demonstrating a solid fit to the data.

CO emission model

ANOVA was employed in the development of the CO emission model, presented as algebraic equation in Eq. (5). Figure 3c shows the surface diagram for the CO emission model. Lowest CO is observed at higher engine load where combustion improved. The ANOVA results are listed in Table 6. With an F-value of 190.52 and a p-value of less than 0.0001, the ANOVA for the response surface quadratic model for “CO” demonstrates the model is highly significant and indicates a very low possibility of this outcome arising owing of noise. With a mean square value of 3.085E-004 and a sum of squares of 1.543E-003 distributed over 5 degrees of freedom, the model has. With an F-value of 914.33 (p = 0.0001), A-BP has a major impact among the model factors; the quadratic term A² is also important and displays an F-value of 28.13 (p = 0.0003). With p-values of 0.5959, 0.1446, and 0.1177 respectively, other terms—B-LHV, AB, and B²—are not significant. With a sum of squares of 1.619E-005, the residual error is negligible and suggests low variance inexplicable by the model. With A-BP and A² as the most important factors and the non-significant components implying the possibility for model reduction to increase parsimony without compromising accuracy, the model fits the data generally rather well.

HC emission model

The HC emission model was developed using ANOVA; it is shown as algebraic equation in Eq. (6). The surface diagram for the HC emission model is shown in Fig. 3d. The ANOVA results are listed in Table 6. The lowest HC emission was observed when the load on engine was low and biodiesel blending was on higher side. The ANOVA results are listed in Table 3. With an F-value of 31.45 and a p-value of less than 0.0001, the ANOVA for the response surface quadratic model for “HC” shows a highly significant model indicating a minimum possibility that these results are due to noise. With a mean square of 573.48, the model’s total of squares is 2867.41, dispersed over 5 degrees of freedom. With F-values of 106.20 (p = 0.0001) and 35.18 (p = 0.0001 respectively, A-BP and B-LHV respectively emphasize their great impact and help to explain the relevance of the model. With an F-value of 13.18 (p = 0.0046), the quadratic term A² is likewise important. With p-values of 0.9000 and 0.1158 respectively, the interaction term AB and the quadratic term B² are not significant, either though. With a sum of squares of 182.34, suggesting a decent fit, the residual error is rather little. Given the little impact of AB and B², model reduction may help to simplify the model without sacrificing its explanatory ability.

CO₂ emission model

ANOVA was applied in development of the model utilizing CO₂ emission data. The CO₂ emission model is shown in Eq. (7) as algebraic equation. Surface diagram for the CO₂ emission model is shown in Fig. 3e. The ANOVA results are listed in Table 6. Lower LHV and full engine load exhibit more CO₂ emission level. The low CO₂ emission levels are observed at lower engine loads. The ANOVA for the response surface quadratic model for "CO₂" shows a highly significant model, with an F-value of 1679.69 and a p-value of less than 0.0001, indicating a negligible likelihood that the results are due to noise. The model’s sum of squares is 110.97, with a mean square of 22.19 across 5 degrees of freedom. The term A-BP is the most influential, with an exceptionally high F-value of 8195.78 (p < 0.0001), indicating its dominant effect on CO2 levels. B-LHV is also significant, with an F-value of 79.79 (p < 0.0001), along with the interaction term AB, which has an F-value of 10.99 (p = 0.0078). Both quadratic terms, A² and B², are significant, with F-values of 62.33 (p < 0.0001) and 21.28 (p = 0.0010), respectively. The residual sum of squares is minimal at 0.13, indicating a good fit of the model to the data. The significant contributions of A, B, AB, A², and B² highlight the complexity of the interactions and nonlinear effects in the model, while the low residual suggests that model reduction may not be necessary.

NOx emission model

The NOx emission data was used to develop the model by using ANOVA. The NOx emission model is depicted in the form of algebraic equation in Eq. (8). Figure 3f exhibits the surface diagram for the NOx emission model. Full engine load and high LHV shows a higher NOx emission level. The ANOVA results are listed in Table 6. This is attributed to high combustion chamber temperatures. With an F-value of 589.50 and a p-value of less than 0.0001, the ANOVA findings for the NOx emission model show a highly significant model demonstrating the model is a powerful predictor and unlikely to have occurred by chance. Comprising a mean square of 6.814E + 005 over 5 degrees of freedom, the model’s sum of squares is 3.407E + 006. With an extraordinarily high F-value of 2864.54 (p < 0.0001), A-BP is the most important element among the model components in terms of NOx emissions. Though their effects are less than those of A-BP, B-LHV with an F-value of 6.26 (p = 0.0314) and the interaction term AB with an F-value of 5.76 (p = 0.0273) are also statistically significant. With an F-value of 72.51 (p < 0.0001), the quadratic term A² is significant and emphasizes non-linear effects of A-BP on NOx emissions. With a p-value of 0.0868, which denotes a smaller non-linear effect of B-LHV, the term B² is not significant, though. With a mean square of 1155.91 across 10 degrees of freedom, the low residual sum of squares of 11,559.05 imply the model catches most of the variability in NOx emissions. The model offers a strong explanation of the NOx emissions considering the important terms (A, B, AB, and A²) and the minimum residual error; still, there is space for model improvement by eliminating the non-significant B² term.

Desirability approach for optimization

Response Surface Methodology (RSM) makes extensive use of the desirability method as a commonly utilized optimization tool to identify the ideal mix of several answers. Based on predefined goals like maximizing, minimizing, or reaching a target value, it turns every response into a desirability function ranging from 0 (undesirable) to 1 (completely desirable). Usually using geometric mean, this distinct desirability are then aggregated into an overall desirability value. The optimization process balances the trade-offs of contradictory reactions to optimize this general desirability. Applications of this approach abound in industrial processes, developing goods, and multi-objective optimization projects. The desirability bar plots are depicted in Fig. 4. The optimized levels of each parameter are listed in Table 7.

Fig. 4

Desirability levels.

Table 7 Optimization results.

ML modelling

Correlation analysis

The correlation matrix given in Table 8, shows relationships between many performance and emission parameters including brake power (BP), lower heating value (LHV), brake thermal efficiency (BTE), brake specific fuel consumption (BSFC), and emissions such CO, HC, CO₂, and NOx. These interactions and trade-offs between engine performance and emissions assist one to better grasp them (Fig. 5). Beginning with braking power (BP), CO₂ (0.99) and NOx (0.98) show a significant positive association suggesting that emissions of CO₂ and NOx climb as brake power increases. This implies that more complete combustion brought about by higher power output generates more CO₂ and NOx. With BTE (0.91), a comparable strong positive association indicates that braking power and efficiency increase concurrently, hence increasing engine efficiency at higher power output. On the other hand, BP shows substantial negative associations with CO (-0.98) and BSFC (-0.83), therefore as power rises, CO emissions and fuel consumption per kW output drop, indicating more effective combustion.

Table 8 Correlation matrix.

Fig. 5

Heat map of correlational values.

With a modest positive association between lower heating value (LHV) and hydrocarbon (HC) emissions (0.46), fuels with higher energy content usually generate more unburned hydrocarbons. LHV does not affect general combustion efficiency or emissions trends, though, as there is no appreciable association between LHV and other factors including braking power or emissions including CO₂, CO, and NOx. Positive correlations with BP (0.91), NOx (0.95), and CO₂ (0.84) show that as BTE improves emissions of CO₂ and NOx rise. Higher engine efficiency at higher combustion temperatures matches this to more complete oxidation (CO₂) and greater NOx generation. Conversely, BTE displays substantial negative associations with BSFC (-0.97) and CO (-0.93), implying that as efficiency increases the engine consumes less fuel and generates less carbon monoxide emissions, hence stressing the need for effective combustion.

Where lower BSFC equates to greater thermal efficiency, as less fuel is needed to generate a given amount of power, brake-specific fuel consumption (BSFC) has a very significant negative association with BTE (-0.97). Decreased fuel usage per unit of power is thus connected to decreased emissions of NOx (-0.88) and CO (-0.88), which have negative relationships as well. Higher power and efficiency lead to lower CO emissions, which is expected as more complete combustion produces less incomplete combustion byproducts like CO. CO emissions show negative relationships with BP (-0.98) and BTE (-0.93). A strong inverse connection with NOx (-0.99) emphasizes that CO and NOx emissions have an inverse relationship, where more complete combustion lowers CO but raises NOx due to greater combustion temperatures favouring NOx development. With LHV (0.46), hydrocarbon (HC) emissions exhibit a modest positive association suggesting that some fuels with higher energy density may produce more unburned hydrocarbons, perhaps due to incomplete combustion at lower engine efficiency. Though to a lesser degree than other factors, HC also correlates somewhat with CO₂ (0.76) and NOx (0.75), implying that greater HC emissions are related to increases in CO₂ and NOx. Higher brake power and efficiency lead to more CO₂ emissions due to more complete combustion converting more carbon into CO₂, so strongly correlated CO₂ emissions are with BP (0.99), BTE (0.84), and NOx (0.95). Higher NOx levels also accompany high CO₂ emissions, suggesting that efficient combustion and higher temperatures support both CO₂ and NOx generation.

Positive correlations between NOx emissions and BP (0.98), BTE (0.95), and CO₂ (0.95) indicate that rising power and efficiency likely result in greater combustion temperatures that enable NOx generation. The significant inverse link between NOx and CO (-0.99) highlights, even more, the usual trade-off in combustion processes: lower CO emissions are usually connected with greater NOx emissions since more complete combustion reduces CO while increasing NOx generation. This study shows generally that although higher brake power and thermal efficiency usually translate into lower fuel consumption and reduced CO emissions, they also result in increased NOx and CO₂ emissions, so highlighting the difficulties in maximizing engine performance while minimizing pollutants, particularly.

Model development and prediction evaluation

BTE model

The data collected through lab-based engine testing was used for the development of the BTE model. The comparative analysis is depicted in Fig. 6a and b for GBoost and ELM, respectively. With almost negligible training MSE and R² values of 1, indicating an ideal fit, the Brake Thermal Efficiency (BTE) model results for both Gradient Boosting (Gboost) and Extreme Learning Machine (ELM) exhibit great performance on the training data. On the test set, their behaviour does, however, show clear contrasts. With a lower test MSE (0.863) than Gboost (1.124), ELM suggests improved predicting accuracy on unseen data (Table 9). The test R² numbers also help to further confirm ELM explaining somewhat more variation (0.9604) than Gboost (0.9484). Mirroring this, the Mean Absolute Percentage Error (MAPE) shows ELM shows greater generalization with a lower test MAPE (3.92%) than Gboost (4.55%). Both models had 0% error during training. This reflects ELM’s better generalizing capacity since it implies that it is more efficient in reducing mistakes on the test set.

Fig. 6

Actual and predicted BTE model (a) GBoost (b) ELM.

Table 9 Statistical values of the developed model.

BSFC model

The BSFC model was developed using data gathered by lab-based engine testing. Figure 7a and b respectively for GBoost and ELM show the comparison study. Strong training performance with low error is obtained by the Brake Specific Fuel Consumption (BSFC) model findings for Extreme Learning Machine (ELM) and Gradient Boosting (Gboost). With a very low train MSE (1.55E-09) and a R² of 1. Gboost gets a practically ideal fit. ELM does, however, closely follow with a somewhat higher train MSE (5.47E-07) and an R² of 0.9999, thereby demonstrating an outstanding fit as well. ELM somewhat beats Gboost on the test set with a lower test MSE (0.0005 vs. 0.0006) and a higher R² value (0.9368 vs. 0.9232), so indicating ELM has better predictive accuracy on unseen data. For Gboost at 0.0s1%, the Train MAPE displays almost faultless performance; ELM exhibits a rather higher inaccuracy of 0.15%. With a lower Test MAPE (5.95%) ELM once more shows superior generalization error than Gboost (6.659%), suggesting that ELM is more efficient in lowering prediction error on the test set. With a little trade-off in training error, ELM offers generally improved test performance.

Fig. 7

Actual and Predicted BSFC model (a) GBoost (b) ELM.

CO emission model

The CO emission model was developed using the lab-based engine testing data. Figure 8a and b respectively for GBoost and ELM show the comparison study. On the training data, both Extreme Learning Machine (ELM) and Gradient Boosting (Gboost) show good performance according to the CO emission model results. Gboost achieves a nearly perfect train MSE (2.73E-09) and R² of 0.9999; ELM shows a somewhat higher train MSE (6.36E-09) and R² of 0.993. ELM shows a better balance on the test set with a lower test MSE (1.34E-06) than Gboost (2.09E-06), and a test R² of 0.9873, somewhat less than Gboost’s 0.98 but still good. Concerning generalization error, ELM shows better predictive accuracy on unknown data by having a lower test MAPE (4.129%) than Gboost (5.506%). Reflecting its closer fit on the training set, Gboost has a substantially lower train MAPE (0.1208%), than ELM (2.47%). With lower error in CO emission projections, ELM offers generally better test performance.

Fig. 8

Actual and predicted CO model (a) GBoost (b) ELM.

HC emission model

The HC emission model was developed using the lab-based engine testing data. Figure 9a and b for GBoost and ELM respectively show the relative analysis. Gboost and ELM performance shows clear variations according to the Hydrocarbon (HC) emission model results. Both models have great training performance; ELM achieves a perfect fit (train MSE = 1.47E-06, train R² = 1.0) and zero training error (train MAPE = 0.00%). Although quite accurate, Gboost has a somewhat higher training MSE (0.0045) and a train R² of 0.9999 together with a minimum train MAPE of 0.11%. With a far lower test MSE (2.62) than Gboost’s considerably higher test MSE (8.456), ELM shows better generalization ability when assessed on the test data. This smaller error suggests that on unknown data ELM is more successful in forecasting HC emissions. With a higher test R² of 0.9737 than Gboost, which has a test R² of 0.915, ELM also explains more of the variation in the test data.

Fig. 9

Actual and Predicted HC model (a) GBoost (b) ELM.

Furthermore, ELM beats Gboost in terms of Mean Absolute Percentage Error (MAPE) on the test set; whereas Gboost shows a far larger test MAPE of 8.42%, ELM shows a lot smaller test MAPE of 4.17%. This underlines even more how precise and consistent ELM provides for HC emissions, hence reducing the forecast errors and providing better generalizing across fresh data. Given the test performance in particular, ELM is the more efficient model overall for this use data.

CO₂ model

The CO₂ emission model was developed using the data gathered by lab-based engine testing. Figure 10a and b respectively for GBoost and ELM show the comparison study. Strong performance for both Gradient Boosting (Gboost) and Extreme Learning Machine (ELM) is shown by the CO₂ emission model results; ELM exhibits a little edge in generalizing. Reflecting their capacity to exactly capture the variance of the training set and minimize error, both models attain flawless accuracy with identical results—train MSEs near zero (Gboost: 4.50E-07, ELM: 1.44E-06), train R² values of 1, and minimum train MAPE (0.01%). When used on the test data, ELM shows better predicting ability than Gboost (0.1245), hence lowering the test MSE (0.0713). Given its higher test R² result of 0.984, which somewhat beats Gboost’s test R² of 0.9815, this implies ELM is more successful in estimating CO₂ emissions for unseen data. While Gboost has a higher test MAPE of 6.16%, ELM further separates itself with a lower test MAPE of 4.55%, indicating smaller mistakes in CO₂ emission predictions, on the test set. These findings show ELM’s stronger overall performance in precisely estimating CO₂ emissions and better generalizing capability.

Fig. 10

Actual and predicted CO₂ model (a) GBoost (b) ELM.

NOx model

The NOx emission model was developed employing the obtained data from lab-based engine testing. Figure 11a and b for GBoost and ELM respectively show the relative analysis. The NOx emission model results show that both Gradient Boosting (Gboost) and Extreme Learning Machine (ELM) perform exceptionally well, though ELM demonstrates a slight advantage in generalization. Both models perform nearly perfectly on the training data; Gboost has a train MSE of 37.82 and an R² of 0.9998, while ELM displays an even lower train MSE (2.57E-06) and a somewhat higher R² of 0.9999. With ELM scoring 0.00% train MAPE and Gboost at 0.25%, both models also reduce training mistakes. With a lower test MSE of 756.33 than Gboost’s 1068.4, ELM beats Gboost when assessed on the test set, therefore showing that ELM generates more accurate NOx emission forecasts on unseen data. With a higher test R² of 0.9956 ELM also shows its capacity to explain more variance in the test data than Gboost’s test R² of 0.9938.

Fig. 11

Actual and predicted NOx model (a) GBoost (b) ELM.

Regarding Mean Absolute Percentage Error (MAPE) on the test set, ELM once more exhibits higher performance with a test MAPE of 3.31%, below that of Gboost’s 4.30%. These findings imply that ELM is the better model for this work since it can more successfully reduce prediction errors for NOx emissions and has more generalizing ability.

Results and discussion

Brake thermal efficiency

The Brake thermal efficiency is a key parameter that defines the amount of chemical energy released by the fuel in relation to the energy supplied. In other words, it measures the ratio of the brake power output of the engine to the heat supplied by the fuel²³.

It also measures how efficiently an engine converts the heat from fuel into productive work. The heat generated from the chemical energy of the fuel depends on the air–fuel ratios and the calorific values of the fuel²⁴. From Fig. 12, it is understood that Brake thermal efficiency is good for the blend 90D5MO5H + 25 ppm Zr₂O₃ subjected to the brake power attainment of 5.2 kw at peak load conditions. The presence of 25 ppm Zr₂O₃ with 5% of Hexanol improves the calorific values of the fuel to perform the brake thermal efficiency by 34%, but for diesel, the brake thermal efficiency is found to be 31.3%, The improved in BTE is found for the blend 90D5MO5H + 25 ppm Zr₂O₃ by 8.63% than diesel. This is because of higher the calorific value for the blend 90D5MO5H + 25 ppm Zr₂O₃ causes to achieve very good BTE than other blends. The presence of 25 ppm Zr₂O₃ catalyst promotes to accelerate the combustion very faster to achieve this brake thermal efficiency than other blends²⁵.

Fig. 12

Brake power vs brake thermal efficiency.

BSFC

Brake specific fuel consumption (BSFC) refers to the amount of fuel required to achieve a specific brake thermal efficiency. In other words, it is the inverse of brake thermal efficiency (BTE)²⁶. As shown in Fig. 13, the fuel consumption for the blend 90D5MO5H + 25 ppm Zr2O3 is the lowest at peak loads, at 0.14 kg/kWh. In comparison, diesel fuel consumption is 0.26 kg/kWh. This reduction in fuel consumption is attributed to the higher oxygen supply, which enhances the calorific value, facilitated by the uniform mixing of the Zr2O3 catalyst with hexanal and moringa diesel blends²⁷.The Zr₂O₃ catalyst promotes the combustion with an equal supply of oxygen for achieving quicker combustion leading to achieve lowest fuel consumption. Hence the blend 90D5MO5H + 25 ppm Zr₂O₃ has the very lowest consumption followed by 46.13% lower than diesel. It is also noticed that gradual increment of Zr₂O₃ catalyst from 25 to 100 ppm causes to increase in densities of the blends results in higher oxygen supply leads to falls lower calorific value resulting in higher fuel consumption than 90D5MO5H + 25 ppm Zr₂O₃ and diesel²⁸.

Fig. 13

Brake power vs brake-specific fuel consumption.

Engine speed

Engine speed is a key factor in determining the performance characteristics of fuel and the required brake power at peak loads²⁹. The engine speed depends on the type and nature of the load, as well as the properties of the test fuel. Figure 14 illustrates the engine speeds achieved with various test blends at different revolutions. It is observed that all blends exhibit favorable speed characteristics under different loads and brake power conditions. The blend 90D5MO5H + 25 ppm Zr₂O₃ offers the lowest speed of 1480 rpm for attaining maximum brake power subjected to peak loads, this is due to the turbulence property offered by the blend that promotes the better surface tensions for the blend 90D5MO5H + 25 ppm Zr₂O₃ than diesel and other blends³⁰.

Fig. 14

Brake power vs engine speed.

Combustion characteristics

This section discusses the combustion parameters of the engine, focusing on the cylinder pressures generated during crank angle rotations. Additionally, important parameters such as heat release rates at various crank angles are examined in detail.

Inline cylinder pressures

The in-cylinder pressure of a diesel-powered engine is primarily influenced by the combustion characteristics and the volume of the prepared fuel mixture^6,31. The pressure within the cylinder begins to rise at the onset of ignition and peaks shortly after combustion processes initiate. This phenomenon, known as delayed ignition, occurs due to the lag between spark initiation and the start of combustion. During full combustion, the rapid sequence of uncontrolled events leads to significant pressure buildup within the cylinder, typically resulting in higher heat release rates^32,33. Figure 15 illustrates the information on inline cylinder pressures acquired for various blends. The horizontal axis of the graph reflects the fluctuations in the crank position degrees, while the vertical axis represents the pressures in the cylinder. This largely reflects the relationship between the changing of crank positions and the increase in cylinder tensions with diesel and different blends³⁴. Experiments in Fig. 14 revealed that the 90D5MO5H + 25 ppm Zr₂O₃ blend achieved the highest cylinder pressure of 70 bar. The 80D10MO10H + 50 ppm Zr₂O₃ blend reached 62.1 bar, the integrated 70D15MO15H + 75 ppm Zr₂O₃ reached 61.2 bar, the mix 100 MO + 100 ppm Zr₂O₃ reached 60.15 bar, and the blend fossil diesel fuel reached 59.17 bar. An additional factor contributing to the variation in for-cylinder pressures is the increased deferral phase seen in the 90D5MO5H + 25 ppm Zr₂O₃ blend, which has a lower percentage of biodiesel compared to alternative blend sectors³⁵.

Fig. 15

Crank angle vs cylinder pressures.

Heat release rate

The Heat release rate is the quantity of thermal energy liberated during the process of cremation. It serves as a guide to ascertain the prevailing features of the feedstock throughout burning³⁶. The accumulated rate of heat released serves as a parameter utilized to determine the amount of unburnt weight consumed by the embers while ignition.

The combination of HRR characteristics is shown in Fig. 16. This characterizes the total rate at which heat is released from different fuel mixtures. The Heat Release Rate (HRR) for the combustibility 70D15MO15H + 75 ppm Zr₂O₃ was 38 J/deg CA, which is comparable to 100 D. Furthermore, the fuel 90D5MO5H + 25 ppm Zr₂O₃ exhibits an exceptionally high HRR followed by 45 J/deg CA. That can be attributed to the superior atomization and high caloric content of the fuel 90D5MO5H + 25 ppm Zr₂O₃ in comparison to other blends. The Heat Release Rate (HRR) for the fuels 80D10MO10H + 50 ppm Zr₂O₃, 70D15MO15H + 75 ppm Zr₂O₃, and 100 MO + 100 ppm Zr2O3 were 39 J/deg CA, 37 J/ deg CA, and 30 J/deg CA, respectively. This can be attributed to worse atomization, shorter ignition delay, and lower calorific value of, 80D10MO10H + 50 ppm Zr₂O₃, and 100 MO + 100 ppm Zr₂O₃ compared to 100 D³⁷.

Fig. 16

Crank angle vs heat release rate.

Emissions

CO

The Production of carbon monoxide into the atmosphere is a signal of a hazardous gas that might have severe environmental adverse effects. The colourless, odourless, as well as unpleasant characteristics of the gas were used to illustrate its emission³⁸. It can also be described as the gradual expulsion of gases at a relatively sluggish response rate, which might result in faulty cremation and the production of an amalgam of diverse formations, ultimately expelled in the way of emissions³⁹. Figure 17 depicts the atmospheric production of the chemical carbon monoxide. The results from the experiments showed the greatest levels of CO emissions were recorded under entire load conditions for pure diesel at 0.025%. The levels of carbon monoxide emissions ranged from 0.020% for 100 MO + 100 ppm Zr₂O₃ pollutants. This phenomenon can be attributed to the elevated oxygen levels and insufficient provision of chemical fuel⁴⁰.

Fig.17

Brake power vs CO emissions.

CO₂

Carbon dioxide is chemically inert and harmless, and it is emitted as a result of inadequate ignition or a decreased level of oxygen emission in the presence of air. The substance has the potential to cause environmental damage and result in respiratory issues⁴¹. The process associated with carbon bond breaking and subsequent formation of new bonds during the reaction of dissolved oxygen with the atmosphere to liberate CO₂ vitality throughout the environment is often referred to the carbon dioxide breakdown. Carbon dioxide (CO₂) emissions released into the atmosphere are determined by the concentration of the accessible carbon during fuel burning. The dioxide emission is depicted in Fig. 18. The findings from the experiments indicate that 100 MO + 100 PPM Zr₂O₃ generates a greater quantity of CO₂ emissions at 11.5%, while 90D5MO5H + 25 PPM Zr₂O₃ produces a lower amount at 10.1%. The primary reason for this is the superior burning efficiency observed in diesel and pure Moringa biodiesel with 100 ppm concentrations of Zr₂O₃ catalysts^42,43.

Fig. 18

Brake power vs CO₂ emissions.

HC

In the most difficult part of the method of evaporation, when fuel is burned, permanent HC emission levels, which are mostly made up of naturally occurring molecules, are discharged towards the surrounding environment⁴⁴. Alternatively, it can be characterized as the inadequate burning of fuels caused by the insufficient air-fueled combination, inadequate conditions, and reduced spark latency throughout fuel combustion⁴⁵. Chemical pollutants refer to both unaltered and portions of burned hydrocarbon particles. Observations from Fig. 19 indicate that diesel generates greater hydrocarbon emissions of 65 ppm under 100% load conditions, 90D5MO5H + 25 ppm Zr₂O₃ creates 68 ppm, 80D10MO10H + 50 ppm Zr₂O₃ produces 70 ppm, and 70D15MO15H + 75 ppm Zr₂O₃ produces 69 ppm. The fewest chemical pollutants are created by 100 MO + 100 ppm Zr₂O₃ at 40 ppm. One of the primary factors contributing to the reduced emissions of 100 MO + 100 ppm Zr₂O₃ is its lesser petrochemical content, which, in turn, results in improved combustion due to the addition of more oxygen^46,47.

Fig. 19

Brake power vs HC emissions.

NO_X

Nitrogen oxide (NOx) emissions are highly toxic and might seem brown. A significant phenomenon observed during the release of nitrogen emissions was identified using Zeldovich’s mechanism⁴⁸. This mechanism involves the expansion of chemical bonding between nitrogen molecules in situations with restricted oxygen supply at high temperatures. Potential consequences of this condition include severe allergic reactions including pneumonia in humans⁴⁹. Figure 20 depicts the atmospheric nitrogen oxide production during engine activity. The primary cause for the growth of nitrogen dioxide is the increased supply of oxygen resulting from a greater quantity of air molecules. This process expedites the production of nitric oxide, reducing the released gases of hydrocarbons and carbon monoxide. The sustainability of the environment and the use of eco-friendly fuels have become popular due to the increasing strictness of pollution standards. The chemical reactions involved in this process are followed by⁵⁰

$${\text{O}}_{2} + {\text{N}} = {\text{O}} + {\text{NO}}$$

(9)

$${\text{O}} + {\text{N}}_{2} = {\text{NO}} + {\text{N}}$$

(10)

$${\text{N}} + {\text{O}}_{2} = {\text{O}} + {\text{NO}}$$

(11)

Fig. 20

Brake power vs NO_X emissions.

An analysis of Fig. 20 reveals that 100 D exhibited the highest nitrogen oxide production at 1900 ppm, but 70D15MO15H + 75 ppm Zr₂O₃ created the fewest pollutants at 1608 ppm. The main influence contributing to generating most pollutants like NOx is the elevated quantity of oxygen present in biodiesel fuel. In addition to the breathable air in the combustion area, the nitrogen in the ambient air at higher temperatures also contributes to creating a greater quantity of Nitrogen⁵¹.

Oxygen content

Oxygen is toxic and possesses a striking black colour, which can lead to the development of chronic headaches. Figure 21 illustrates, the absorption level of oxygen at the finish of the pure combustion process, O₂ is taken during combustion whereby the nitrogen in the air will ultimately be absorbed⁵². The observations revealed that under 100% load circumstances, 100 MO + 100 ppm Zr₂O₃ had the maximum oxygen level compared to 100 D, 90D5MO5H + 25 ppm Zr₂O₃, 80D10MO10H + 50 ppm Zr₂O₃, and 70D15MO15H + 75 ppm Zr₂O₃. Primarily, this is due to the provision of oxygen-enriched fuel. Biodiesel is essentially an oil-based fuel that is enhanced with oxygen. This is the main factor contributing towards the greater level of oxygen for 100 MO + 100 ppm Zr₂O₃ compared to every other mix under fully loaded circumstances⁵³.

Fig. 21

Brake power vs oxygen content.

Conclusion

The study and investigation of the Moringa biodiesel blend proportionated with the hexanol catalysts on marine diesel engines has been analyzed. The concentration levels of Zr₂O₃ catalyst varying from 25 ppm, 50 ppm, 75 ppm and 100 ppm give some interesting results compared to diesel. The physical properties of the test blends are significantly analysed at the testing centre to determine their properties to liberate the combustion process. The addition of this Zr₂O₃ catalyst to moringa biodiesel improves the properties for enhancing the combustion rates with reduced emissions. The following results were achieved during the Experimentation process of the Marine diesel engine.

• Among various blends, 90D5MO5H + 25 ppm Zr₂O₃ possesses a very good brake thermal efficiency than diesel. Because of the good calorific value offered by this blend than diesel, by 8.63% than diesel.

• 90D5MO5H + 25 ppm Zr₂O₃ attains the very lowest fuel consumption at peak loads followed by 0.14 kg/ kWh, But for diesel the fuel consumption is 0.26 kg/kWh. Hence the blend 90D5MO5H + 25 ppm Zr₂O₃ has very lowest consumption followed by 46.13% lower than diesel.

• The blend 90D5MO5H + 25 ppm Zr₂O₃ offers the lowest speed of 1480 rpm for attaining maximum brake power subjected to peak loads, This is due to the turbulence property offered by the blend that promotes the better surface tensions for the blend 90D5MO5H + 25 ppm Zr₂O₃ than diesel and other blends.

• 90D5MO5H + 25 ppm Zr₂O₃ blend achieved the highest cylinder pressure of 70 bar. The reason behind the increase in cylinder pressures for this blend is the higher surface tensions offered by this blend resulting in fine spray throughout the combustion and the higher cetane number provided by the blend 90D5MO5H + 25 ppm Zr₂O_3.

• The Heat Release Rate (HRR) for the combustibility 70D15MO15H + 75 ppm Zr₂O₃ was 38 J/ deg CA, which is comparable to 100 D. Furthermore, the fuel 90D5MO5H + 25 PPM Zr₂O₃ exhibits an exceptionally high HRR followed by 45 J/deg CA. That can be attributed to the superior atomisation and high caloric content of the fuel 90D5MO5H + 25 PPM Zr₂O₃ in comparison to other blends.

• The greatest levels of CO emissions were recorded under entire load conditions for pure diesel at 0.025%. The levels of carbon monoxide emissions ranged from 0.020% for 100 MO + 100 ppm Zr₂O₃ pollutants. This phenomenon can be attributed to the elevated oxygen levels and insufficient provision of chemical fuel.

• 100 MO + 100 PPM Zr₂O₃ generates a greater quantity of CO₂ emissions at 11.5%, while 90D5MO5H + 25 PPM Zr₂O₃ produces a lower amount at 10.1%. The primary reason for this is the superior burning efficiency observed in diesel and pure Moringa biodiesel with 100 ppm concentrations of Zr₂O₃ catalysts.

• One of the primary factors contributing to the reduced HC emissions of 100 MO + 100 ppm Zr₂O₃ followed by 42 ppm is its lesser petrochemical content, which, in turn, results in improved combustion due to the addition of more oxygen.

• 70D15MO15H + 75 ppm Zr₂O₃ created the fewest pollutants concerned to NO_X at 1608 ppm. The main influence contributing to generating most pollutants like NOx is the elevated quantity of oxygen present in biodiesel fuel. In addition to the breathable air in the combustion area, the nitrogen in the ambient air at higher temperatures also contributes to creating a greater quantity of Nitrogen.

• 100 MO + 100 ppm Zr₂O₃ had the maximum oxygen level compared to 100 D, 90D5MO5H + 25 ppm Zr₂O₃, 80D10MO10H + 50 ppm Zr₂O₃, and 70D15MO15H + 75 ppm Zr₂O₃. Primarily, this is due to the provision of oxygen-enriched fuel. Biodiesel is essentially an oil-based fuel that is enhanced with oxygen. This is the main factor contributing towards the greater level of oxygen for 100 MO + 100 ppm Zr₂O₃ compared to every other mix under fully loaded circumstances. Among various blends 90D5MO5H + 25 ppm Zr₂O₃ possesses a lower oxygen content rate of 4%; due to complete combustion taking place at the combustion phase.

Future research could expand on this study by investigating the long-term performance and durability of moringa oleifera-based biodiesel blends in a variety of engine types and operational conditions to ensure consistent efficiency and emission reduction. Further optimization of the concentrations of zirconium oxide and 1-hexanol, along with exploring the impact of other additives, could enhance the blends’ performance. Additionally, integrating these biodiesel blends with other renewable fuels and conducting comprehensive life cycle and economic assessments would provide valuable insights into their sustainability and feasibility for large-scale applications. Finally, expanding the use of machine learning models to predict fuel performance and emissions across different engine configurations would strengthen the potential for widespread adoption in the transportation and energy sectors.