ANN-driven prediction of optimal machine learning models for engine performance in a dual-fuel mode powered by biogas and fish oil biodiesel

• ANN models accurately predict engine performance under various conditions, enhancing energy efficiency through precise emissions and vibration monitoring. • Emission parameters predicted by the FFBP ANN model (0.967 correlation) demonstrate the effectiveness of knowledge-based systems in optimizing complex engine behaviors for better energy management. • Vibration characteristics prediction (0.94 correlation) ensures reliable system performance monitoring, reducing energy losses through timely maintenance. • The B25 biodiesel blend, identified as optimal through ANN models, improves combustion efficiency, achieving 97% accuracy with Random Forest and 95% with Bayes Net, contributing to fuel savings and reduced emissions. • Developed ANN models effectively align predictions with real-world performance, reducing the need for extensive experimental testing and optimizing engine efficiency. Global climate change is increasingly driven by carbon dioxide emissions from fossil fuels, intensifying the greenhouse effect and global warming. Biodiesel, particularly fish oil biodiesel provides a sustainable alternative that reduces greenhouse gas (GHG) emissions. In this study, the engine was tested with various blends of fish oil biodiesel and conventional diesel, using volume ratios from 20 % to 40 % in 5 % increments. Subsequently, it was operated in dual-fuel mode with two mixtures: (a) Pure Methane (CH4) and (b) Methane (CH4) + Carbon dioxide (CO2), injected through the intake manifold at flow rates of 4, 8, and 12 LPM alongside fresh air and the biodiesel blends. Performance analysis included emissions and combustion characteristics, such as nitrogen oxides (NOx), smoke, hydrocarbons (HC), carbon dioxide (CO 2 ), carbon monoxide (CO), brake thermal efficiency (BTE), peak pressure (PP), maximum pressure rise rate (MPRR) and vibrational characteristics under varying biogas flow rates and engine loads. Engine performance was monitored using vibrational data analysed by machine learning (ML) models based on Bayes net and random forest algorithms. B25 with M7.2C4.8 (i.e. Methane–7.2 LPM & CO 2 -4.8 LPM) stands out as the top performer, achieving the highest classification accuracy at 97 %. The combustion and emission parameters for optimal blend were predicted using a feedforward backpropagation artificial neural network (ANN) with a 3-12-8 neuron architecture. Additionally, vibrational characteristics were analysed with an ANN configured as 2-4-4-5. The results showed that these ANN models effectively predicted engine parameters under different load conditions, with average R-values of 0.97 and 0.98, respectively. The B25 blend significantly reduced emissions and enhanced combustion efficiency, highlighting its potential in mitigating GHG emissions and promoting sustainable alternative fuels.