Machine learning based prognostics and statistical optimization of the performance of biogas-biodiesel blends powered engine

Abstract

In this study, waste biomass-derived biogas was employed as the main fuel while the biodiesel-diesel blend was used as pilot fuel. This paper describes the development of a Decision Tree and Response Surface methodology-based statistical framework for prediction modeling and optimization. The compression ratio, fuel injection time, fuel injection pressure, and biogas flow rate were employed as controllable inputs, and brake thermal efficiency, peak combustion pressure, and exhaust emission were selected as responses. The experimental data for model development was gathered for the development of prediction models and optimization. The decision tree-based models were robust with almost negligible mean squared errors and R2 values of more than 0.9487 for all models. Response surface methodology-based optimized engine parameters were validated with the following results compression ratio was 17.9, fuel injection pressure was 225 bar, fuel injection timing was 26.3-degree crank angle after top dead center, and the biogas flow rate was 0.85 kg/h. Validation results were within 5 % of the model-optimized results. The prognostic models for all control factors were developed with decision tree-based machine learning with high predictive efficiency and low errors.

Keywords

• Machine learning
• Renewable energy
• Prognostics
• Efficiency
• Exhaust emissions
• Biomass gasification