Scholarlink

<p>This study presents an investigation of the performance, combustion, and emission characteristics of lemongrass oil–diesel blends (LGO20) doped with titanium dioxide and graphene oxide nanoparticles at concentrations of 50, 75, and 100&#xa0;p.p.m in a single-cylinder four-stroke water-cooled Kirloskar TV1 direct-injection diesel engine. A total of 8 fuel blends tested at 6 load conditions from no load to full load generated 48 experimental observations. According to the experimental results, the LGO20GO75 blend gave the highest brake thermal efficiency (BTE) (31.05%), which is an improvement of 7.07% over LGO20 and by 3.50% over baseline diesel; it had the lowest Brake-Specific Fuel Consumption (BSFC) (0.3071&#xa0;kgk<sup>−1</sup>Wh<sup>−1</sup>), which is a reduction of about 13.83% compared to LGO20, and gave maximum reductions in carbon monoxide (CO) (29.84%), hydrocarbons (HC) (16.39%), and smoke opacity (12.33%) when compared with neat diesel. The oxides of nitrogen (NO<sub>x</sub>) increased by about 20% due to increased in-cylinder temperatures that are associated with enhanced combustion. As a second analytical step, an ensemble machine learning framework (Random Forest, Gradient Boosting, XGBoost, and ANN as a benchmark comparator) was used to identify trends in all combustion and emissions throughout the experimental matrix. This predictive performance (0.9972 for BTE using gradient boosting; 0.9917 using XGBoost) is limited to the experimental operating regime, which means results should not be extrapolated to an untested domain of operation. Machine learning models tend to overfit with limited data, and as the ML analysis is based on only 48 experimental points, the resulting models cannot be easily generalised to other scenarios. Even with a 70:30 train–test split or using fivefold cross-validation, overfitting is still a risk. This observed predictive performance should only be interpreted as a trend-level fit for the tested blend, load and nanoparticle concentration range that was considered. Overall findings position LGO20GO75 as a potential lower emission substitute for diesel and demonstrate utility, though with some limitations of ensemble ML methods for predicting combustion parameters.</p> Graphical abstract <p></p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Combustion performance and emission prediction of TiO2 and graphene oxide nanoparticle-enhanced lemongrass oil–diesel blends using ensemble machine learning

  • 期刊论文

  • 发表日期：

    2026年7月2日

查看全文
  • P. V. Elumalai,
  • Jagan Padmanaban,
  • N. S. Senthur,
  • Wasurat Bunpheng,
  • Choon Kit Chan,
  • Mohammed Al Awadh,
  • Fayaz Hussain

摘要

This study presents an investigation of the performance, combustion, and emission characteristics of lemongrass oil–diesel blends (LGO20) doped with titanium dioxide and graphene oxide nanoparticles at concentrations of 50, 75, and 100 p.p.m in a single-cylinder four-stroke water-cooled Kirloskar TV1 direct-injection diesel engine. A total of 8 fuel blends tested at 6 load conditions from no load to full load generated 48 experimental observations. According to the experimental results, the LGO20GO75 blend gave the highest brake thermal efficiency (BTE) (31.05%), which is an improvement of 7.07% over LGO20 and by 3.50% over baseline diesel; it had the lowest Brake-Specific Fuel Consumption (BSFC) (0.3071 kgk−1Wh−1), which is a reduction of about 13.83% compared to LGO20, and gave maximum reductions in carbon monoxide (CO) (29.84%), hydrocarbons (HC) (16.39%), and smoke opacity (12.33%) when compared with neat diesel. The oxides of nitrogen (NOx) increased by about 20% due to increased in-cylinder temperatures that are associated with enhanced combustion. As a second analytical step, an ensemble machine learning framework (Random Forest, Gradient Boosting, XGBoost, and ANN as a benchmark comparator) was used to identify trends in all combustion and emissions throughout the experimental matrix. This predictive performance (0.9972 for BTE using gradient boosting; 0.9917 using XGBoost) is limited to the experimental operating regime, which means results should not be extrapolated to an untested domain of operation. Machine learning models tend to overfit with limited data, and as the ML analysis is based on only 48 experimental points, the resulting models cannot be easily generalised to other scenarios. Even with a 70:30 train–test split or using fivefold cross-validation, overfitting is still a risk. This observed predictive performance should only be interpreted as a trend-level fit for the tested blend, load and nanoparticle concentration range that was considered. Overall findings position LGO20GO75 as a potential lower emission substitute for diesel and demonstrate utility, though with some limitations of ensemble ML methods for predicting combustion parameters.

Graphical abstract