Comparison of machine learning algorithms for predicting diesel/biodiesel/iso-pentanol blend engine performance and...

Area of Science:

• Combustion Engines
• Alternative Fuels
• Machine Learning

Background:

• Optimizing engine performance and reducing emissions are critical for sustainable transportation.
• Alternative fuels, such as iso-pentanol and biodiesel blends, offer potential solutions to mitigate environmental impact.
• Machine learning (ML) provides powerful tools for analyzing complex engine data and predicting performance metrics.

Purpose of the Study:

• To evaluate engine performance and exhaust emissions for various fuel blends using ML techniques.
• To identify optimal iso-pentanol and biodiesel ratios for enhanced engine efficiency and reduced emissions.
• To compare the predictive capabilities of Artificial Neural Network (ANN), Support Vector Machine (SVM), and Extreme Gradient Boosting (XGBoost) models.

Main Methods:

• Utilized ANN, SVM, and XGBoost models to analyze engine performance data (BTE, BSFC, exhaust gas temperature) and emissions (NOx, CO2, hydrocarbons).
• Focused on varying iso-pentanol ratios while keeping biodiesel ratios constant in diesel blends.
• Validated model performance using regression coefficient (R²), root mean square error, and mean absolute error.

Main Results:

• ANN demonstrated superior performance in predicting Brake Thermal Efficiency (BTE) (R²=0.984) and Brake-Specific Fuel Consumption (BSFC) (R²=0.958).
• SVM achieved the best performance in predicting exhaust gas temperature (R²=0.981).
• XGBoost excelled in predicting CO2 (R²=0.956) and hydrocarbon emissions (R²=0.973), while ANN showed strong NOx prediction (R²=0.959).

Conclusions:

• ANN models are highly effective for accurate engine performance prediction (BTE, BSFC).
• XGBoost models provide accurate predictions for key emission parameters (CO2, hydrocarbons).
• ML techniques offer a robust framework for optimizing alternative fuel blends in internal combustion engines.