Scholarlink

<p>Globally, the growing concern regarding climate change has necessitated a transition towards cleaner fuels and advanced combustion strategies in compression ignition (CI) engines, driven significantly by the need to mitigate engine emissions. This shift is particularly critical given the substantial contribution of conventional engine technologies to atmospheric pollution and global warming. Therefore, this paper presents a comprehensive investigation into the performance characteristics and emission profiles of a single-cylinder CI engine fuelled by specifically formulated blends incorporating diesel, hydrogen (H₂), light gas oil (LGO) biofuel, and water injection. Thirteen fuel modes are simulated using Simulink<sup>®</sup> for the experimental data. The collected experimental data were subjected to comprehensive statistical and trend-based analysis. Initial normality assessment, conducted using the Shapiro–Wilk test, revealed that all dataset variables exhibited significant deviations from normal distribution (<i>p</i> &lt; 0.05). Given this non-parametric characteristic of the data, Spearman’s rank correlation coefficient was deliberately employed as the most appropriate method to evaluate the strength and direction of monotonic relationships between the variables. In addition, LOWESS trendlines confirmed non-linear but monotonic relationships, supporting the choice of the proposed model. The study used a hybrid deep learning framework for predictive analysis, incorporating deep feature extraction via a 50-layer Residual Network and optimization through Chaos Vortex Optimization. This was integrated into a deep reinforcement learning (DRL) architecture with an artificial neural network (ANN) for accurate prediction of engine responses. The study found LGO25H10 to be a highly effective fuel blend for optimum running of the engine among all tested blends. The R<sup>2</sup> of the DRL-ANN prediction technique measured 95.978%, 95.978%, 94.789%, 96.875%, 94.978%, and 96.990% for BSFC, BTE, HC, CO, NOx, and smoke emission, correspondingly projecting the model effectiveness. The state-of-the-art techniques confirm that the DRL-ANN approach efficiently assesses CI engine characteristics.</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

An Optimal Neural Network Model for Projecting of CI Engine Operational Efficiency and Emission in Hydrogen Blend Using Simulink

  • 期刊论文

  • 发表日期：

    2025年11月10日

查看全文
  • Atanu Roy,
  • Sabyasachi Pramanik,
  • Kalyan Mitra

摘要

Globally, the growing concern regarding climate change has necessitated a transition towards cleaner fuels and advanced combustion strategies in compression ignition (CI) engines, driven significantly by the need to mitigate engine emissions. This shift is particularly critical given the substantial contribution of conventional engine technologies to atmospheric pollution and global warming. Therefore, this paper presents a comprehensive investigation into the performance characteristics and emission profiles of a single-cylinder CI engine fuelled by specifically formulated blends incorporating diesel, hydrogen (H₂), light gas oil (LGO) biofuel, and water injection. Thirteen fuel modes are simulated using Simulink® for the experimental data. The collected experimental data were subjected to comprehensive statistical and trend-based analysis. Initial normality assessment, conducted using the Shapiro–Wilk test, revealed that all dataset variables exhibited significant deviations from normal distribution (p < 0.05). Given this non-parametric characteristic of the data, Spearman’s rank correlation coefficient was deliberately employed as the most appropriate method to evaluate the strength and direction of monotonic relationships between the variables. In addition, LOWESS trendlines confirmed non-linear but monotonic relationships, supporting the choice of the proposed model. The study used a hybrid deep learning framework for predictive analysis, incorporating deep feature extraction via a 50-layer Residual Network and optimization through Chaos Vortex Optimization. This was integrated into a deep reinforcement learning (DRL) architecture with an artificial neural network (ANN) for accurate prediction of engine responses. The study found LGO25H10 to be a highly effective fuel blend for optimum running of the engine among all tested blends. The R2 of the DRL-ANN prediction technique measured 95.978%, 95.978%, 94.789%, 96.875%, 94.978%, and 96.990% for BSFC, BTE, HC, CO, NOx, and smoke emission, correspondingly projecting the model effectiveness. The state-of-the-art techniques confirm that the DRL-ANN approach efficiently assesses CI engine characteristics.