Next Article in Journal
Enhancing Predictive Accuracy of Novel Creep Model for Stainless Steel 316 Using AI-Driven Optimization and Machine Learning Methods
Previous Article in Journal
Corrosion Behavior of Additively Manufactured Al Alloy in Alkaline Media
 
 
Search for Articles:
Title / Keyword
Author / Affiliation / Email
Journal
Article Type
 
 
Advanced Search
 
Section
Special Issue
Volume
Issue
Number
Page
 
Journals
Materials Proceedings
Volume 31
Issue 1
10.3390/materproc2026031029
materproc-logo
Submit to this Journal
► Article Menu

Article Menu

share Share announcement Help format_quote Cite question_answer Discuss in SciProfiles
first_page
settings
Order Article Reprints
Font Type:
Arial Georgia Verdana
Font Size:
Aa Aa Aa
Line Spacing:
Column Width:
Background:
Proceeding Paper

Machine Learning-Driven Optimization of Atomization Characteristics in Fuel Blends Using Nanomaterials: Meta Analysis †

by
Luke Ajuka
* and
Christopher Enweremadu
Department of Mechanical, Bioresources and Biomedical Engineering, University of South Africa, Private Bag X6, Johannesburg 1710, South Africa
*
Author to whom correspondence should be addressed.
†
Presented at the 4th International Conference on Applied Research and Engineering, Pretoria, South Africa, 21–23 November 2025.
Mater. Proc. 2026, 31(1), 29; https://doi.org/10.3390/materproc2026031029
Published: 23 April 2026
(This article belongs to the Proceedings of The 4th International Conference on Applied Research and Engineering)
Browse Figures
Versions Notes

Abstract

This study explores the integration of nanomaterials and machine learning (ML) in enhancing atomization and combustion behavior of nanofuels. Nanoparticles such as TiO2, Al2O3, and graphene derivatives improve fuel atomization, thermal conductivity, and emission reduction. A systematic review (2021–2025) and meta-analysis reveal short-term gains in brake thermal efficiency (+12.5%) and emission reduction (CO −12%, HC −25%, NOx −19%), though long-term stability remains limited by agglomeration and injector fouling. ML-models, including Bayesian-Ridge and Random-Forest, predict efficiency metrics effectively but underperform for emissions. The findings highlight the need for atomization descriptors and hybrid ML–CFD models for robust predictive combustion design.

1. Introduction

Efficient atomization and spray dynamics govern droplet size, evaporation, and air–fuel mixing, thereby influencing ignition delay, flame stability, and pollutant formation in engines [1,2,3]. Fuel properties such as viscosity, surface tension, volatility, and density critically affect breakup regimes, penetration, and combustion efficiency [4,5]. To enhance performance, nanomaterials including TiO2, Al2O3, and graphene derivatives have been incorporated into fuels, improving thermal conductivity, ignition delay, and emission reduction through catalytic effects [6,7], and other heat transfer mediums [8,9]. These advances highlight nano-fuels as promising candidates for sustainable combustion technologies, meanwhile, atomization processes, fuel properties, and nanomaterial enhancements collectively influence combustion performance of nanofuels, as highlighted in the next section.

1.1. Fuel Atomization, Combustion Performance, and Nanomaterial Innovations

Fuel atomization is the first critical step in combustion, where liquid fuel is disintegrated into fine droplets to maximize surface area and promote rapid evaporation and mixing with air (as shown in Figure 1 and Figure 2 for typical diesel break-up). Figure 1 shows the multiscale processes governing liquid fuel atomization, from nozzle flow and cavitation to final spray dispersion. The dashed (characteristic length scales) horizontal lines with arrows represent axial regions along the spray downstream of the nozzle to conceptually divide the spray evolution into dominant flow and breakup stages, while the faint dashed lines (along cavitation and turbulence region) indicate the continuity and spatial overlap of internal-flow-driven mechanisms, showing how cavitation inside the nozzle progressively evolves into turbulence-dominated flow downstream rather than ending abruptly at the nozzle exit, that is, cavitation and turbulence are not isolated regions, but part of a coupled flow development process influencing spray breakup. These stages collectively determine key spray descriptors, including the Sauter Mean Diameter (SMD), droplet size distribution, penetration length, and spray cone angle, which directly influence evaporation rate, air–fuel mixing, ignition delay, and pollutant formation. Figure 2 illustrates the classical Ohnesorge number (Oh) versus Reynolds number (Re) regime map, delineating dominant breakup mechanisms, including Rayleigh breakup, first wind-induced breakup, and atomization-dominated regimes. This regime is particularly sensitive to subtle changes in fuel physicochemical properties, and the inclusion of nanomaterials can shift the effective Oh–Re operating point by modifying surface tension, viscosity, and internal energy transport, thereby altering breakup transitions and spray morphology, a key limitation identified in the meta-analysis, and most machine-learning models, which are reported in the literature.
The resulting spray characteristics, such as droplet size distribution, penetration length, and dispersion directly determine ignition delay, flame propagation, and pollutant formation [10]. These behaviors are strongly influenced by fuel viscosity, volatility, and density, which govern breakup regimes and spray penetration [4,5]. Recent advances in nanomaterial additives further enhance atomization and oxidation processes, providing catalytic pathways for cleaner and more efficient combustion [6,7].
Materproc 31 00029 g001
Figure 1. Schematic of a fuel spray highlighting primary and secondary breakup stages. Adapted from Ref. [11].
Figure 1. Schematic of a fuel spray highlighting primary and secondary breakup stages. Adapted from Ref. [11].
Materproc 31 00029 g001
Materproc 31 00029 g002
Figure 2. Oh No (Classic Ohnesorge Number) vs. Re (Reynolds) diagram illustrating liquid jet breakup transitions. Adapted from Ref. [11]. The shaded region indicates a representative operating envelope for diesel direct-injection sprays under typical engine conditions; its boundaries are onlyindicative.
Figure 2. Oh No (Classic Ohnesorge Number) vs. Re (Reynolds) diagram illustrating liquid jet breakup transitions. Adapted from Ref. [11]. The shaded region indicates a representative operating envelope for diesel direct-injection sprays under typical engine conditions; its boundaries are onlyindicative.
Materproc 31 00029 g002

1.2. Fuel Complex Spray Behavior ML Modeling

Machine learning (ML) has become an increasingly valuable tool in modeling and predicting complex phenomena associated with spray dynamics and combustion. Traditional CFD models often require significant computational resources and may struggle to capture highly nonlinear interactions between spray parameters, fuel properties, and environmental conditions. In contrast, ML algorithms can efficiently learn and generalize such relationships from experimental or simulated datasets [12]. For example, Khan et al. [13], and Zvirblis et al. [14] applied deep neural networks and Gaussian process regression to predict droplet size distribution, atomization efficiency, and ignition delay with high accuracy, with submitted that it outperformed conventional correlations. However, the dependence of the technique on the availability and quality of training data, limited extrapolation capability outside the dataset range, and reduced interpretability of complex black-box models have been reported to constrain their standalone use in design and optimization [15,16]. Bottom of FormMoreover, hybrid approaches integrating CFD simulations with ML (CFD-ML) enable accelerated optimization of nozzle designs, fuel compositions, and operating parameters. These data-driven models can reduce computational overhead while maintaining high predictive accuracy, making them attractive for engine manufacture, rapid prototyping, and design evaluation [16]. For example, surrogate ML models trained on CFD datasets have been applied to optimize spray breakup length, droplet size distribution, and in-cylinder mixing under varying injection pressures, achieving comparable accuracy to full CFD at a fraction of the computational cost [17,18]. The limitations of this method include dependency on the fidelity of the underlying CFD training data, potential overfitting to narrow operating conditions, and reduced interpretability when extrapolated beyond calibrated design ranges, which may hinder robustness in real-world applications [15,19].
The interpretability of ML models remains a challenge, especially in safety-critical systems. Recent efforts focus on explainable AI (XAI) techniques that elucidate feature contributions, making ML outputs more transparent and trustworthy for spray and combustion diagnostics [20]. For example, SHAP (Shapley additive explanations) and LIME (Local Interpretable Model-agnostic Explanations) have been applied to identify the relative influence of injection pressure, droplet size, and ambient temperature on ignition delay and spray penetration, thereby providing physical insight into black-box predictions [21,22]. The limitations of this method include increased computational overhead, difficulty in scaling to deep learning architectures such as CNNs, and the risk of oversimplifying complex nonlinear interactions, which may restrict its reliability in high-dimensional spray–combustion systems [15,23].
Most experimental evaluations of nanofuels are short-term and conducted under laboratory-scale conditions. Long-term effects such as nanoparticle agglomeration, injector fouling, sedimentation, and degradation under thermal cycling are rarely addressed, yet they pose significant challenges for real-world implementation [17,24]. Hence, in-depth understanding on how these effects evolve over extended operational periods and how they impact injector performance, emissions, and maintenance may be vital. Therefore, this study aims to comprehensively evaluate the integration of nanomaterials and machine learning in fuel atomization and spray dynamics via meta-analysis induced by nanomaterials and their impact on combustion configurations, via a systematic synthesis of experimental and numerical studies (2021–2025) on the influence of nanomaterial additives on fuel atomization, combustion efficiency, and emissions in compression-ignition engines, quantify pooled performance and emission effects using meta-analytical techniques, evaluate the predictive capability of selected machine-learning models, and identify critical atomization, spray, and nanomaterial descriptors required to improve the robustness and generalizability of ML-based combustion optimization frameworks, while proposing future directions for the application of smart fuels and data-driven modeling in combustion systems.

2. Long-Term Effects of Nanofuels

Nanofuels are reported to consistently enhance efficiency and reduce emissions in the short term, as several studies have highlighted critical durability challenges under extended use [25]. Agglomeration and sedimentation reduce dispersion stability and impair atomization during storage and operation [26,27]. Injector fouling has been linked to oxide residue formation and spray deterioration in long-duration tests with metal-based nanofuels [28,29]. Moreover, thermal cycling degradation alters viscosity and nanoparticle structure, undermining combustion stability and efficiency [30,31]. Table 1 further establishes the extent of long-term reliability of nanofuels towards their performance and environmental benefits.
Although nanofuels demonstrate clear short-term improvements in brake thermal efficiency (BTE) and emission reduction, evidence from multiple studies highlights significant limitations under extended operation. Agglomeration and sedimentation remain major concerns, as nanoparticles tend to cluster during prolonged storage, reducing homogeneity and fuel atomization quality [26,27]. Similarly, injector fouling has been documented in long-duration tests with metal-oxide nanofuels, where deposits altered spray characteristics and increased fuel consumption [28,29]. Thermal cycling further exacerbates these issues, with repeated heating and cooling shown to destabilize nanofuel dispersions, leading to viscosity changes and restructuring of nanoparticles that compromise combustion stability [30,31]. These findings suggest that while nanofuels offer substantial performance gains initially, their long-term application may incur hidden penalties in terms of durability, engine wear, and operational reliability.

3. Methodology

To conduct this assessment, the steps in the literature identification comprised of systematic article extraction from Web of Science, ScienceDirect, ASME and IEEE, with focus on studies between 2021–2025. The search keywords utilised included biodiesel, nanofuel, nanoparticles, diesel engine, combustion, emission, performance, stability. Within the context of engineering systems research, a structured meta-analytic synthesis approach was employed, adapted from PRISMA principles for transparency in study identification, screening, eligibility, and inclusion. Due to heterogeneity in experimental designs, fuels, engines, and reported metrics (as illustrated in Figure 3), weighted effect sizes, subgroup comparisons, and random-effects correlation analyses were applied, which are consistent with established practices in combustion, and energy systems meta-research [39].

3.1. Subsection Inclusion and Exclusion Criteria

Articles comprising experimental and/or numerical studies of biodiesel blends with nanoparticles, with clear quantitative data on performance (BTE, BSFC, exergy efficiency) and emissions (CO, HC, NOx, smoke, CO2) were extracted. Reviews which included provided mechanistic or stability insights (for example, Hoang et al. [28], and Parveg & Ratner [31] were added, while retracted or integrity-compromised articles, for example, Pali et al. were excluded.

3.2. Data Extraction and Normalization

Key data (thermal efficiency, BSFC, CO, HC, NOx, smoke, CO2) were extracted and normalized relative to baseline diesel (D100) or B20 blends to ensure cross-study comparability. Nanoparticle concentration (ppm), type (CNT, GO, graphene, TiO2, metal oxides, carbon-based), and operating conditions (injection pressure, load) were recorded, and studies on statistical models such as RSM, Gaussian process regression, and LSBoost [27] were considered as supplementary predictive approaches. Over-weighting of individual studies was ensured by avoiding studies that with unavailable variance measures.

3.3. Data Analysis

Weighted mean effect sizes were calculated for performance and emissions, considering sample size and reported variability. The subgroup analysis was analysed based on Carbon nanoparticles (CNT, graphene, GO) against Metal nanoparticles (TiO2, others), concentration ≤ 100 ppm against >100 ppm, and engine condition modifiers (injection pressure, load), and stability insights from reviews [26,28] were integrated qualitatively, as indicated in the Table 2.

4. Result and Discussion

Model performance was evaluated using cross-validated coefficient of determination (R2) and root-mean-square error (RMSE), with mean absolute error (MAE) inferred as complementary to RMSE for distribution-sensitive emission metrics. Analysis of the metrics brake thermal efficiency (BTE), brake specific fuel consumption (BSFC), and emission (CO, HC, NOx) in the study are discussed in this section, and the indicators are summarized graphically in Figure 3, Figure 4 and Figure 5. Machine-learning models and performance evaluations were implemented using Python programming language (version 3.13; Python Software Foundation, Wilmington, DE, USA). The Random Forest regression model was developed using the scikit-learn library (version 1.8.0; scikit-learn Developers, Paris, France), while data handling and preprocessing were performed using NumPy (NumPy Developers, Austin, TX, USA) and pandas (pandas Development Team, New York, NY, USA). Model training, cross-validation, and error metric computation were carried out within this software environment, while the meta-analysis was conducted using Meta-essential (version 1.4; Erasmus Research Institute of Management, Netherland).

4.1. Metric Comparative Evaluation

The scatter plot in Figure 4 shows the cross-validated predictions versus observed values for BTE, BSFC, CO, HC, and NOx.
The plot (Figure 4) shows that most points of the BTE and BSFC lies close to the 1:1 line in the low-to-moderate range (=0–8% absolute change), whereas performance degrades for the emission metrics, particularly at the distribution tails. Model overpredicts negative extremes such as large NOx decrement and underpredicts large positive improvements, for example, high HC or CO reductions, indicating regression-to-the-mean behavior typical of tree ensembles trained on small samples. Points for CO and HC exhibit wider vertical spread at higher magnitudes (20–45%), evidencing increased variance and bias with effect size. NOx shows the strongest systematic bias with large observed decreases (<−40%) are predicted to be much less negative, while moderate increases are slightly overestimated. At higher magnitudes (20–45%), CO and HC exhibit wider vertical spread evidencing increased variance and bias with effect size, while NOx shows the strongest systematic bias, with large observed decreases (<−40%) are predicted much less negative, while moderate increases are slightly overestimated. Although these patterns imply limited sample size and non-explicit modelling of specific engine types, it shows the need for stronger priors or hierarchical structure for emission endpoints, and that incorporating mechanistic covariates (load/speed, injection/spray descriptors, baseline pollutant levels) may reduce bias at the extremes and improve calibration for large emission changes. The diagonal reference line in corresponds to the ideal condition where predicted values exactly match the actual experimental values.
Overall, the parity plot validates the feasibility of ML-assisted optimization for efficiency (BTE, BSFC), but exposes a feature gap for emissions precisely the gap atomization descriptors when quantified in reported studies.

Model Validation

An ML emission correlation was developed (Equation (1)) from the dataset, and compared with the literature study by Mahalingam & Ganesan [40] to establish the validity range of the results as shown in Figure 5, Figure 6 and Figure 7.
E M L = a L + b
where:
  • E M L = predicted emission (CO or UHC)
  • L = engine load (%)
  • a, b = ML-fitted coefficients inferred from Figure 4.
Materproc 31 00029 g006
Figure 6. Comparison of HC emissions predicted by ML correlation and experimental results of Mahalingam & Ganesan as a function of engine load [40].
Figure 6. Comparison of HC emissions predicted by ML correlation and experimental results of Mahalingam & Ganesan as a function of engine load [40].
Materproc 31 00029 g006
Materproc 31 00029 g007
Figure 7. Comparison of NOx emissions predicted by ML correlation and experimental results of Mahalingam & Ganesan as a function of engine load [40].
Figure 7. Comparison of NOx emissions predicted by ML correlation and experimental results of Mahalingam & Ganesan as a function of engine load [40].
Materproc 31 00029 g007
The Figure 5 validates the ML correlation against the experimental study by Mahalingam & Ganesan [40] for CO emissions across engine load. The absolute deviation remains within 0.002–0.004 CO%, corresponding to a relative error of 6.7%, indicating good predictive fidelity. The near-parallel slopes of the two curves imply a high correlation (expected R 2 > 0.9 for trend fitting), confirming that the ML model captures the dominant load-dependent combustion behavior.
The Figure 6 validates the ML correlation for hydrocarbon (HC) emissions against the experimental benchmark of Mahalingam & Ganesan [40], across engine load. Both datasets show a consistent increase in HC with load, with the ML model slightly overpredicting HC at low–medium loads and underpredicts at high load, reflecting regression smoothing.
Statistically, the absolute deviation between ML and experimental values gives a relative error of 6.9%. The near-linear and parallel trends suggest a high correlation (expected R2 > 0.9) for trend reproduction. Indicating a robust validation of the ML approach for capturing load-dependent UHC emission behavior.
The Figure 7 validates the ML correlation for NOx emissions against the experimental reference of Mahalingam & Ganesan [40], across engine load. Both datasets show an increase in NOx with load, confirming correct trend capture by the ML model. However, the ML curve exhibits systematic underprediction at higher loads, where thermal NOx formation intensifies.
Quantitatively, a relative error averaging 10.5% arose with loading, with highly correlated linear association (expected R2 > 0.9 for trend fitting). Overall, the figure provides trend-level validation of the ML approach for NOx prediction while clearly highlighting its physics-driven limitation at high engine loads, supporting the need for physics-informed feature integration.

4.2. Model Variation

The grouped bar chart compares RMSE_CV for BTE, BSFC, CO reduction, HC reduction, and NOx change using Bayesian Ridge and Random Forest is shown in Figure 8.
Figure 8 on metric errors, indicates that small errors dominates the efficiency metrics (BTE, BSFC) and substantially larger for emissions, especially NOx and HC. Between models, Random Forest slightly improves BSFC but is generally comparable or worse than Bayesian Ridge from the emission endpoints. The low RMSE for BTE/BSFC indicates that high-level inputs (blend, nanoparticle type/ppm, engine class) capture enough signal to support efficiency-oriented optimization. Larger RMSE observed in Figure 4 on CO/HC/NOx reflects unmodeled heterogeneity in the spray/evaporation pathway linking nanomaterials to combustion chemistry. The error inflation at NOx suggests the model cannot represent extremes (large decreases/increases) observed in certain blends/loads. It can be concluded that the parity analysis showing regression-to-mean for extremes shows that linear-Bayesian shrinkage (Bayesian Ridge) is competitive and preferable to Random Forest for noisy, small-N emissions reported in the data.

4.3. Coefficient of Determination (R2)

The grouped chart (BTE, BSFC, CO reduction, HC reduction, and NOx chang) is reported in R2_CV using Bayesian Ridge (BR) and Random Forest (RF), as illustrated in Figure 9.
Figure 9 show that the BR attained positive, moderate R2 on BTE (0.6) and BSFC (0.8) and low positive R2 on CO/HC (0.3–0.4) indicating that the blend composition, nanoparticle type/concentration, and engine class encode sufficient signal for energy-efficiency outcomes. Comparatively, RF performed lower: near-zero on BTE, moderate on BSFC (0.55), and negative on CO/HC (=−0.5 to −0.7). NOx exhibits negative R2 for both models, with BR markedly poor (=−2.3), indicating predictions worse than a mean baseline for this endpoint. The implication of emissions (CO/HC/NOx) with respect to nanomaterials to in-cylinder mixing and chemistry (SMD, cone angle, penetration, micro-explosion frequency, load/speed, injection parameters, baseline emissions, etc.) link shows limited atomization and evaporation mediators. On the model’s behaviour, BR’s shrinkage generalizes better on small, noisy samples (efficiency), whereas RF shows regression-to-mean and instability on heterogeneous emission data, consistent with the parity analysis.
Overall, Figure 9 shows that the R2_CV profile supports ML-assisted efficiency optimization of nanomaterial fuel blends but also underscores that incorporating atomization physics is essential to achieve more reliable and generalizable emission predictions.

4.4. Meta-Analysis of Correlation Across Subgroups

Figure 10 presents a forest plot of the correlation coefficients (r) obtained across multiple experimental studies investigating the influence of different nanomaterial additives on combustion and emission characteristics of CI engines.
The estimated correlation between the inclusion of a nanomaterial (moderator) and a specific response parameter (BTE, BSFC, or emissions of CO, HC, and NOx) are represented by the markers. Horizontal bars denote the associated confidence intervals, while the color codes distinguish the various nanomaterial subgroups (Graphene oxide (GO), multi-walled carbon nanotube (MWCNT), alumina (Al2O3), and MWCNT + Ferrocene). Across all studies, correlation magnitudes span approximately −0.20 ≤ r ≤ 0.45, indicating moderate yet consistent positive effects of nanomaterial incorporation on performance and emission outcomes. Li et al. [41] reported the strongest positive correlations for CO and HC reduction (r = 0.38–0.45) in both GO- and MWCNT-blended fuels, reflecting the enhanced oxidation kinetics promoted by carbonaceous nanoparticles. In contrast, their correlations for BTE and BSFC remained weak (<0.03), signifying that thermal efficiency improvements are indirect and secondary to emission mitigation.
Studies involving Al2O3, for example, Gad & Alenany [27], and Venu & Madhavan [42] revealed moderate positive correlations across all metrics (r = 0.05–0.16), demonstrating its thermal stability and catalytic contribution to uniform combustion and lowered pollutant formation. Conversely, Wang et al. [43] exhibited both positive and negative associations, particularly negative correlations for CO and HC (r = −0.10 to −0.20), likely arising from sub-optimal nanoparticle dispersion or excessive dosing at higher concentrations (25–100 ppm). Furthermore, the MWCNT + Ferrocene hybrid subgroup [44] produced uniformly positive correlations, with the highest observed for HC reduction (r = 0.20), attributable to combined metal–carbon synergism that promotes photo-assisted ignition and soot oxidation under xenon-lamp irradiation.
Overall, the pooled pattern of results confirms that carbon-based nanoparticles (GO, MWCNT) deliver the most pronounced emission benefits, while Al2O3 provides steadier but smaller performance gains. The few negative estimates highlight the sensitivity of nanofuel behavior to concentration, dispersion quality, and combustion environment. Despite moderate effect sizes, the uniformity of positive correlations across independent datasets underscores the reproducibility of nanomaterial-assisted combustion enhancement.

Random-Effects Meta-Analysis of Correlation Estimates

A random-effects meta-analysis encompassing 30 correlation estimates was conducted to quantify the overall relationship between nanomaterial additives and engine performance/emission responses. The pooled correlation coefficient from the random-effects meta-analysis encompassing 30 correlations estimated resulted was r = 0.182r (95% CI: 0.126–0.238), indicating a small-to-moderate but statistically significant positive effect. The test for overall effect yielded Z = 6.51, p < 0.001, confirming that the observed association is unlikely to be due to random chance. The between-study heterogeneity was moderate (I2 = 47.3%), suggesting that approximately half of the variance in observed correlations arises from differences in study design, nanoparticle type, or fuel formulation rather than sampling error alone. Despite this variability, the overall direction of effect remained consistent across subgroups, reinforcing that nanoparticle inclusion improves combustion efficiency and pollutant reduction under diverse engine operating conditions.

5. Conclusions

This study highlights the importance of atomization-focused descriptors for data-driven optimization with actionable insights into fuel formulation and engine design, by identifying carbon-based nanoparticles as more reliable emission reducers. Furthermore, in studies where emission objectives dominate acquisition, for constrained optimization of atomization features, a multi-objective Bayesian optimization with penalized uncertainty may be considered to explore high-impact nanomaterial regimes safely. Therefore, for enhanced results, nanomaterial and engine-mediated descriptors such as Primary/secondary breakup (Sauter Mean Diameter: SMD), size distribution width, ligament/bag-to-shear breakup regime flags), Spray structure (cone angle, penetration length, liquid length, vapor penetration, droplet number density), Thermo-evaporation (micro-explosion frequency, superheat/latent heat proxies, surface temperature, evaporation time, optical/diagnostics (minimum ignition energy shifts, flame lift-off length), nanomaterial descriptors (particle size, shapes including sphere/rod/sheet, effective thermal conductivity, surface energy, zeta potential, surfactant type, and stability index) are essential for robustness of result generalization.
This study is constrained by heterogeneous experimental conditions, limited long-term datasets, and incomplete reporting of spray descriptors across the literature, which restricts emission-model generalization.
Future research directions on emission optimization should integrate inputs with spray/atomization diagnostics descriptors, long-duration engine testing and operating conditions, and CFD–ML frameworks should be adopted along with hierarchical/multi-task learning (partial pooling by engine type and fuel class) and uncertainty-aware/quantile losses to handle tail behavior of nanofuels.

Author Contributions

Conceptualization, L.A. and C.E.; methodology, L.A.; software, L.A. and C.E.; validation, L.A., and C.E.; formal analysis, L.A.; investigation, L.A.; data curation, L.A.; writing—original draft preparation, L.A.; writing—review and editing, C.E. and L.A.; visualization, L.A.; supervision, C.E. All authors have read and agreed to the published version of the manuscript.

Funding

This research received no external funding.

Institutional Review Board Statement

Not Applicable.

Informed Consent Statement

Not Applicable.

Data Availability Statement

The data utilized can be found at Scopus, ScienceDirect, ASME and IEEE databases (Open access) or made available to serious researchers.

Acknowledgments

Authors appreciate the University of South Africa for the resource used in this study.

Conflicts of Interest

The authors declare no conflict of interest.

References

  1. Bayvel, L.; Orzechowski, Z. Liquid Atomization; Taylor & Francis: New York, NY, USA, 2019; pp. 1–476. [Google Scholar] [CrossRef]
  2. Feng, Z.; Zhang, J.; Gu, J.; Leng, X.; He, Z.; Nishida, K. Improving biodiesel atomization performance in CI engines: A review of spray behavior, droplet impingement, and advanced techniques. Processes 2025, 13, 2527. [Google Scholar] [CrossRef]
  3. Gallego, A.; Cacua, K.; Gamboa, D.; Rentería, J.; Herrera, B. Ignition Delay and Burning Rate Analysis of Diesel–Carbon Nanotube Blends Stabilized by a Surfactant: A Droplet-Scale Study. Energies 2023, 16, 7740. [Google Scholar] [CrossRef]
  4. Guo, T.; Li, G.; Li, H.; Huo, H.; Zhang, X.; Wang, Z.; Bai, H. Experimental study on the effects of fuel properties on spray macroscopic characteristics, particle size distribution, and velocity field in a constant volume chamber. Exp. Therm. Fluid Sci. 2026, 171, 111602. [Google Scholar] [CrossRef]
  5. Yadav, M.; Yadav, A.M.; Ahmad, A. Enhancing combustion and emission characteristics of CI engines through atomization and fuel–air mixing using non-circular orifices: A path towards sustainable biodiesel utilization. Green. Technol. Sustain. 2025, 3, 100161. [Google Scholar] [CrossRef]
  6. Ghazaly, N.; Abdulhameed, A. Performance and emissions of nanoadditives in diesel engines: A review. J. Ilm. Tek. Elektro Komput. Inform. 2023, 9, 997–1008. [Google Scholar] [CrossRef]
  7. Ma, S.; Guo, Q.; Wei, J.; Yin, Z.; Zhuang, Y.; Zhang, Y.; Dai, Q.; Qian, Y. Analyzing the effect of carbon nanoparticles on the combustion performance and emissions of a DI diesel engine fueled with a diesel–methanol blend. Energy 2024, 300, 131616. [Google Scholar] [CrossRef]
  8. Shi, W.; Li, F.; Lin, Q.; Fang, G.; Chen, L.; Zhang, L. Effects of Nanoparticle Additives on Spray Characteristics of Liquid Jets in Gaseous Crossflow. Energies 2020, 13, 1574. [Google Scholar] [CrossRef]
  9. Ajuka, L.O.; Enweremadu, C.C. Performance Investigation on the Heat Transfer of Composite rGO-SiO2 Nanofluid in a Radiator. Future Cities Environ. 2025, 11, 1–16. [Google Scholar] [CrossRef]
  10. Khuong, L.S.; Hashimoto, N.; Fujita, O. Spray, droplet evaporation, combustion, and emission characteristics of future transport fuels for compression-ignition engines: A review. J. Traffic Transp. Eng. (Engl. Ed.) 2024, 11, 575–613. [Google Scholar] [CrossRef]
  11. Baumgarten, C. Mixture Formation in Internal Combustion Engines; Springer: Berlin, Germany, 2006; pp. 1–294. [Google Scholar]
  12. Zhou, L.; Song, Y.; Ji, W.; Wei, H. Machine learning for combustion. Energy AI 2021, 7, 100128. [Google Scholar] [CrossRef]
  13. Khan, S.; Masood, M.; Medina, M.; Alzahrani, F. Advancing fuel spray characterization: A machine learning approach for directly injected gasoline fuel sprays. Fuel 2024, 371, 131980. [Google Scholar] [CrossRef]
  14. Zvirblis, T.; Ciziuniene, K.; Matijosius, J. Application of machine learning for fuel consumption and emission prediction in a marine diesel engine using diesel and waste cooking oil. J. Mar. Sci. Eng. 2025, 13, 1328. [Google Scholar] [CrossRef]
  15. Ihme, M.; Chung, W.T. Artificial intelligence as a catalyst for combustion science and engineering. Proc. Combust. Inst. 2024, 40, 105730. [Google Scholar] [CrossRef]
  16. Wilk-Jakubowski, J.L.; Pawlik, L.; Frej, D.; Wilk-Jakubowski, G. Data-driven computational methods in fuel combustion: A review of applications. Appl. Sci. 2025, 15, 7204. [Google Scholar] [CrossRef]
  17. Tao, Z.; Sun, H.; Deng, B.; Wang, J.; Calautit, J. Integrating computational fluid dynamics (CFD) and machine learning to improve urban green infrastructure for heat mitigation and air quality: A systematic review. Build. Environ. 2025, 284, 113516. [Google Scholar] [CrossRef]
  18. Enes, T.; Vedat, M. An analysis for car fuel estimation with regression methods. JAIDA 2023, 3, 107–115. [Google Scholar]
  19. Salehi, F.; Beheshti, A.; Eftekharian, E.; Chen, L.; Hardalupas, Y. Data-driven modelling of spray flows: Current status and future direction. J. Energy Inst. 2025, 119, 101991. [Google Scholar] [CrossRef]
  20. Soudagar, M.E.M.; Shelare, S.; Marghade, D.; Belkhode, P.; Nur-E-Alam, M.; Kiong, T.S.; Ramesh, S.; Rajabi, A.; Venu, H.; Khan, T.M.Y.; et al. Optimizing IC engine efficiency: A comprehensive review on biodiesel, nanofluid, and the role of artificial intelligence and machine learning. Energy Convers. Manag. 2024, 307, 118337. [Google Scholar] [CrossRef]
  21. Hasan, M.M. Understanding model predictions: A comparative analysis of SHAP and LIME on various ML algorithms. J. Sci. Technol. Res. 2024, 5, 17–26. [Google Scholar] [CrossRef]
  22. Ricci, F.; Avana, M.; Mariani, F. A deep learning method for the prediction of pollutant emissions from internal combustion engines. Appl. Sci. 2024, 14, 9707. [Google Scholar] [CrossRef]
  23. Antoniadi, A.M.; Du, Y.; Guendouz, Y.; Wei, L.; Mazo, C.; Becker, B.A.; Mooney, C. Current challenges and future opportunities for XAI in machine-learning-based clinical decision support systems: A systematic review. Appl. Sci. 2021, 11, 5088. [Google Scholar] [CrossRef]
  24. Vaz, M.G.J.; Karathanassis, I.; Gavaises, M.; Mouokue, G. Data-driven prediction of spray macroscopic characteristics for marine injectors using neural networks. Fuel 2026, 405, 136736. [Google Scholar] [CrossRef]
  25. Fayad, M.A.; Mahdi, R.R.; Radhi, A.A.; Mohammed, S.A.; Alawee, W.H.; Chaichan, M.T.; Khashan, K.S.; Dhahad, H.A. Impact of adding nanoparticles into oxygenated fuel blends on engine performance and emissions characteristics of NOx and PM in diesel engines: A comprehensive review. Results Chem. 2025, 18, 102699. [Google Scholar] [CrossRef]
  26. Chen, Y.; Zhang, J.; Zhang, Z.; Zhang, B.; Hu, J.; Zhong, W.; Ye, Y. A comprehensive review of stability enhancement strategies for metal nanoparticle additions to diesel/biodiesel and their methods of reducing pollutants. Process Saf. Environ. Prot. 2024, 183, 1258–1282. [Google Scholar] [CrossRef]
  27. Gad, M.S.; Alenany, A. Forecasting the performance and emissions of a diesel engine powered by waste cooking biodiesel with carbon nanoadditives using tree-based, least square boost and Gaussian regression models. Fuel 2025, 385, 134073. [Google Scholar] [CrossRef]
  28. Hoang, A.T.; Le, M.X.; Nižetić, S.; Huang, Z.; Ağbulut, Ü.; Veza, I.; Said, Z.; Le, A.T.; Tran, V.D.; Nguyen, X.P. Understanding behaviors of compression ignition engines running on metal nanoparticle additive-included fuels: A control comparison between biodiesel and diesel fuel. Fuel 2022, 326, 124981. [Google Scholar] [CrossRef]
  29. Risberg, P.A.; Alfredsson, S. The Effect of Zinc and Other Metal Carboxylates on Nozzle Fouling. In Proceedings of the SAE 2016 World Congress and Exhibition, Detroit, MI, USA, 12–14 April 2016. [Google Scholar] [CrossRef]
  30. Jain, A.; Bora, B.J.; Kumar, R.; Barik, D.; Kumar, N.; Paramasivam, P.; Chohan, J.S.; Majdi, H.S.; Alsubih, M.; Islam, S.; et al. Assessment of Nahar biodiesel with TiO2 nanoparticles at increased fuel injection pressure of a diesel engine: Exergy, energy, statistical analysis. Case Stud. Therm. Eng. 2025, 71, 106125. [Google Scholar] [CrossRef]
  31. Parveg, A.S.M.S.; Ratner, A. A comprehensive review of liquid fuel droplet evaporation and combustion behavior with carbon-based nanoparticles. Prog. Energy Combust. Sci. 2025, 106, 101198. [Google Scholar] [CrossRef]
  32. Onojowho, E.E.; Betiku, E.; Asere, A.A. Performance and CO2 emission of a single cylinder compression ignition engine powered by Khaya senegalensis non-edible seed fuel blends. Heliyon 2024, 10, e28380. [Google Scholar] [CrossRef]
  33. Shelare, S.D.; Belkhode, P.N.; Nikam, K.C.; Jathar, L.D.; Shahapurkar, K.; Soudagar, M.E.M.; Veza, I.; Khan, T.M.Y.; Kalam, M.A.; Nizami, A.S.; et al. Biofuels for a sustainable future: Examining the role of nano-additives, economics, policy, internet of things, artificial intelligence and machine learning technology in biodiesel production. Energy 2023, 282, 128874. [Google Scholar] [CrossRef]
  34. Liu, G.; Yu, R.; Deng, S.; Yan, W.; Liu, D. Temperature measurement of nanofluid fuel flames dispersed with multiple particles using emission spectroscopy. Int. J. Therm. Sci. 2025, 211, 109713. [Google Scholar] [CrossRef]
  35. Garugubilli, R.; Prasad, V.V.S.; Sagari, J. Experimental investigation of copper oxide nanoparticles dispersed in an Azadirachta indica oil–biodiesel blend: Stability and physicochemical properties approach. Nanosci. Technol. 2024, 15, 39–50. [Google Scholar] [CrossRef]
  36. Li, J.; Yue, Z.; Wang, J.; Yang, R.; Liu, H.; Yao, M. Spray and combustion characteristics of Al2O3 nano-diesel under engine-relevant conditions. Appl. Therm. Eng. 2025, 280, 128415. [Google Scholar] [CrossRef]
  37. Jathar, L.D.; Meherunnesa; Shelare, S.; Shahapurkar, K.; Abedin, T.; Farade, R.A.; Rajabi, A.; Khan, T.M.Y.; Singh, R.; Cuce, E.; et al. A Short Review on Nano-Additives to Enhance Biodiesel Performance in Diesel Engines. J. Therm. Sci. 2025, 34, 1450–1473. [Google Scholar] [CrossRef]
  38. Karchkhadze, K.; Khetsuriani, N. Long-term storage and quality stability of biodiesel fuels. Int. J. Low-Carbon Technol. 2024, 19, 1198–1203. [Google Scholar] [CrossRef]
  39. Berretta, M.; Furgeson, J.; Wu, Y.N.; Zamawe, C.; Hamilton, I.; Eyers, J. Residential energy efficiency interventions: A meta-analysis of effectiveness studies. Campbell Syst. Rev. 2021, 17, e1206. [Google Scholar] [CrossRef]
  40. Mahalingam, S.; Ganesan, S. Effect of Nano-fuel additive on performance and emission characteristics of diesel engine using biodiesel blends with diesel fuel. Int. J. Ambient Energy 2018, 41, 316–321. [Google Scholar] [CrossRef]
  41. Li, J.; Wei, J.; Chen, H.; Xu, Y.; Liu, Y.; Dai, Q. Study on the combustion and emission characteristics of a compression ignition engine using diesel/ethanol blend with carbon nanoadditives. Renew. Energy 2025, 246, 122941. [Google Scholar] [CrossRef]
  42. Venu, H.; Madhavan, V. Effect of Al2O3 nanoparticles in biodiesel–diesel–ethanol blends at various injection strategies: Performance, combustion and emission characteristics. Fuel 2016, 186, 176–189. [Google Scholar] [CrossRef]
  43. Wang, X.; Zhang, Y.; Karthikeyan, C.; Boomadevi, P.; Marousek, J.; Nasif, O.; Alharbi, S.A.; Xia, C. Role of injection pressure on fuel atomization and spray penetration in Thevetia peruviana and Jatropha curcas biodiesel blends with nanoparticles. Fuel 2022, 324, 124527. [Google Scholar] [CrossRef]
  44. Primiceri, P.; de Fazio, R.; Strafella, L.; Carlucci, A.P.; Visconti, P. Photo-induced ignition phenomenon of carbon nanotubes by Xenon pulsed light: Ignition tests analysis, automotive and new potential applications, future developments. J. Appl. Res. Technol. 2017, 15, 609–623. [Google Scholar] [CrossRef]
Materproc 31 00029 g003
Figure 3. Prisma Flow Chart for Selection of Studies.
Figure 3. Prisma Flow Chart for Selection of Studies.
Materproc 31 00029 g003
Materproc 31 00029 g004
Figure 4. Parity plot for response metrics (Random Forest). The solid diagonal line represents the line of perfect agreement (Predicted = Actual). Data points closer to this line indicate better predictive performance.
Figure 4. Parity plot for response metrics (Random Forest). The solid diagonal line represents the line of perfect agreement (Predicted = Actual). Data points closer to this line indicate better predictive performance.
Materproc 31 00029 g004
Materproc 31 00029 g005
Figure 5. Comparison of CO emissions predicted by ML correlation and experimental results of Mahalingam & Ganesan as a function of engine load [40].
Figure 5. Comparison of CO emissions predicted by ML correlation and experimental results of Mahalingam & Ganesan as a function of engine load [40].
Materproc 31 00029 g005
Materproc 31 00029 g008
Figure 8. RMSE (Bayesian Ridge and Random Forest) across all metrics.
Figure 8. RMSE (Bayesian Ridge and Random Forest) across all metrics.
Materproc 31 00029 g008
Materproc 31 00029 g009
Figure 9. Cross-validated (R2) for Bayesian Ridge and Random Forest across all metrics.
Figure 9. Cross-validated (R2) for Bayesian Ridge and Random Forest across all metrics.
Materproc 31 00029 g009
Materproc 31 00029 g010
Figure 10. Forest plot of correlation coefficients across studies showing the relationship between nanoparticle additives and combustion/emission performance indicators. Sources: [27,41,42,43,44].
Figure 10. Forest plot of correlation coefficients across studies showing the relationship between nanoparticle additives and combustion/emission performance indicators. Sources: [27,41,42,43,44].
Materproc 31 00029 g010
Table 1. Appraisal of long-term effects of nanofuels.
Table 1. Appraisal of long-term effects of nanofuels.
Title of Study/AuthorMethodologyRemarksBias
Forecasting biodiesel performance and emissions with nano-additives [27]. Study was experimental, using B20 blends with CNT, graphene, GO at 25–100 ppm; coupled with machine learning prediction (GPR, LSBoost).Short-term gains in BTE and reduced emissions were reported, but ML projections suggested sedimentation and performance decay beyond 500 h unless stabilizers are included.Study focused on predictive modelling, not validated with actual long-term field tests.
Assessment of Nahar biodiesel with TiO2 nanoparticles at increased injection pressure [30].Engine experiments were conducted at 200–280 bar injection pressure and variable loads with 150 ppm TiO2 nanofuel; RSM optimization used.Reported initial stability of TiO2 dispersion; however, thermal cycling degradation observed, altering viscosity and slightly reducing atomization efficiency after repeated tests.Limited test duration (~100 h) under lab cycling; lacks storage/field validation.
Performance and CO2 emission of a single cylinder compression ignition engine powered by Khaya senegalensis non-edible seeds fuel blends [32]. Khaya senegalensis biodiesel blends + ANSYS CFD.Reported blend instability over cycles; injector clogging risk noted.CFD heavy; less real long-term validation.
Biofuels for a sustainable future: Examining the role of nano-additives, economics, policy, internet of things, artificial intelligence and machine learning technology in biodiesel production [33]. Review on nano-additives, IoT, AI in biodiesel fuel.Highlighted storage stability issues and sedimentation in WCO biodiesel.Study conducted general review, with little primary data.
Temperature measurement of nanofluid fuel flames dispersed with multiple particles using emission spectroscopy [34]. Study focused more on spectroscopy of nanofluid flames with multiple particles.Report shows that nanoparticle deposits impair thermocouple function; indicates sedimentation and fouling.Report were on Lab flame only; and not long-term engine.
Experimental evaluation of stability and performance of biodiesel blends with carbon nanoparticles [32]. Experimental biodiesel blends with carbon nanoparticles.Reported nano-agglomeration at >0.2% wt. concentration during storage.Short-term study; <30 days stability.
A critical review on nano-additives for biodiesel-fueled compression ignition engines [28].Review of nano-additives in CI engines.Study highlighted degradation under thermal cycling, especially for TiO2 and Al2O3.Review; data heterogeneity across studies.
The effect of zinc and other metal carboxylates on nozzle fouling [29].Nozzle fouling experiments on a single-cylinder engine in three parts: (1) zinc neodecanoate concentration effect; (2) neodecanoates of Zn, Na, Ca, Cu, Fe on fuel flow loss; (3) effect of RME concentration in Zn-neodecanoate contaminated petroleum diesel. Nozzles cut open; deposits analyzed by SEM and EDX.Single-cylinder tests show metal carboxylates (Zn, Na, Ca, Cu, Fe neodecanoates), not just zinc or biodiesel can rapidly foul injector nozzle holes; higher cation charge greater fuel-flow loss. SEM/EDX confirms the contaminant metal in deposits, typically higher at the inlet than the outlet, forming 2–4 μm granulates (near common-rail fuel-filter pore sizes)Single-cylinder test environment and contaminant dosing may not fully represent all real-world multi-cylinder duty cycles and fuel system variability (e.g., filtration, temperature history, additive packages).
Stability and physicochemical properties of CuO nanoparticles dispersed in biodiesel blend [35]. Prepared CuO-biodiesel blend (B20) at 75 ppm, used surfactants (Triton X / QPAN 80 / Tween 80) + ultrasonication; stability assessed via UV-Vis absorbance/transmittance over 3 weeks.Surfactants improved stability but stability declined by week 3, highlighting time-dependent aggregation/sedimentation risk, which is an important ‘long-term’ constraint for storage and deployment (supply chain, transport, tank residence time).Only weeks-scale; UV-Vis is a proxy measure (not full rheology/particle sizing through time); lacks thermal cycling, vibration, real tank/engine circulation, so long-term real-world stability may deviate.
Spray and combustion characteristics of Al2O3 nano-diesel under engine-relevant conditions [36].Optical constant-volume vessel experiments (engine-relevant pressures/temps) using diffuse backlight imaging, high-speed imaging, and two-colour method; compared neat diesel vs Al2O3 nano-diesel (100 mg/L) with surfactant (CTAB) preparation steps for stable dispersion.Nano-diesel showed shorter liquid penetration, slightly higher spray cone angle, longer ignition delay, shorter flame lift-off length, and notably reduced soot indicator (average KL factor −59.76% at 3.5 MPa). Mechanistically, suggests better mixing + catalytic oxidation can reduce soot formation, which is relevant to long-term deposit/DPF burden even though not a durability trial.Not an endurance/field study, using constant-volume vessel simplifies in-cylinder flow/thermal transients, ‘long-term effects’ inferred from soot/temperature trends rather than measured injector deposits, wear, or multi-hour engine durability.
A short review on nano-additives to enhance biodiesel performance in diesel engines [37]. Review synthesis of experimental studies on nano-additives’ effects on performance, combustion, emissions, and wear/durability, highlighting evidence patterns and research gaps.Concludes that nano-additives frequently improve performance/emissions but explicitly stresses that long-term engine durability and component compatibility evidence remains limited and is a priority for future work.Review bias from heterogeneity across studies (different nanoparticles, doses, dispersion methods, test cycles). Long-term conclusions remain constrained because many included studies are short duration and not field-validated endurance tests.
Long-term storage and quality stability of nano-biodiesel fuels [38]. Developed a nano-composition biodiesel and tracked multi-year storage stability (2018–2021) using FTIR + gas chromatography, plus standard physical/chemical property checks against specification limits.Reported fuel properties remained stable over ~4 years, supporting feasibility for long-term storage without loss of key quality indicators—directly addressing one of the biggest real-world barriers to nanofuels (aging/sedimentation/oxidation).Storage-study bias: results may depend on specific nano-composition and storage conditions; does not directly test engine endurance, injector fouling, or aftertreatment compatibility over long use cycles.
Table 2. Major Performance and Emission Indicators.
Table 2. Major Performance and Emission Indicators.
MetricWeighted Mean Effect Size (%)Range Across Studies (%)Key References
Brake Thermal Efficiency (BTE)12.57.5–19.0[27]
Brake Specific Fuel Consumption (BSFC)13.07.0–20.0[27]
Exergy Efficiency22.015.0–40.0[31]
CO Reduction12.46.5–20.0[27,28]
HC Reduction25.515.0–36.0[27]
NOx Reduction18.85.0–35.0[27,28]
Smoke Reduction13.34.0–21.0[27]
CO2 Reduction20.76.0–47.0[32]
Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to people or property resulting from any ideas, methods, instructions or products referred to in the content.

Share and Cite

MDPI and ACS Style

Ajuka, L.; Enweremadu, C. Machine Learning-Driven Optimization of Atomization Characteristics in Fuel Blends Using Nanomaterials: Meta Analysis. Mater. Proc. 2026, 31, 29. https://doi.org/10.3390/materproc2026031029

AMA Style

Ajuka L, Enweremadu C. Machine Learning-Driven Optimization of Atomization Characteristics in Fuel Blends Using Nanomaterials: Meta Analysis. Materials Proceedings. 2026; 31(1):29. https://doi.org/10.3390/materproc2026031029

Chicago/Turabian Style

Ajuka, Luke, and Christopher Enweremadu. 2026. "Machine Learning-Driven Optimization of Atomization Characteristics in Fuel Blends Using Nanomaterials: Meta Analysis" Materials Proceedings 31, no. 1: 29. https://doi.org/10.3390/materproc2026031029

APA Style

Ajuka, L., & Enweremadu, C. (2026). Machine Learning-Driven Optimization of Atomization Characteristics in Fuel Blends Using Nanomaterials: Meta Analysis. Materials Proceedings, 31(1), 29. https://doi.org/10.3390/materproc2026031029

Article Metrics

Back to TopTop