<b>Adaptive Injection Phasing and Methanol </b><b>Blend Optimization for Low- Carbon </b><b>Engine Performance: A Meta- Analysis and </b><b>Machine-Learning Framework with CCS </b><b>Integration</b>

Abstract

Abstract

The transition to net-zero mobility, aligned with global decarbonization targets, has intensified interest in methanol as a clean combustion fuel and carbon-neutral energy carrier. When optimally blended, methanol’s high-octane rating, strong charge-cooling effect, and inherent oxygen content enhance brake thermal efficiency (BTE), improve premixed combustion, and suppress knock in both spark- and compression-ignition engines. This study integrates meta-analysis, physics-informed modelling, and machine learning to establish predictive relationships between methanol blend ratio, start-of-injection (SOI) phasing, and engine performance. Meta-analysis revealed a significant positive correlation between SOI advancement and BTE (r = 0.74, 95% CI: 0.60–0.85), with moderate heterogeneity (I² = 38–45%). Optimal performance was achieved at methanol blend ratios of 25–30% with SOI between −30° and −35° CA aTDC, balancing efficiency gains with emission constraints. Quantitative model evaluation demonstrated high predictive accuracy (R² ≈ 0.99, RMSE < 0.30), exceeding typical machine learning combustion models (R² ≈ 0.91–0.97). SHAP-based sensitivity analysis confirmed that methanol blend ratio (0.125 ± 0.015) and SOI (0.082 ± 0.010) are dominant parameters controlling system behaviour. Model comparison showed that XGBoost achieved the highest accuracy (MAE = 0.01 °CA, RMSE = 0.02 °CA, R² = 0.985), outperforming Random Forest (R² = 0.90) and the physics-informed correlation model (R² = 0.93), while linear regression exhibited inferior performance (R² = 0.90) due to its inability to capture nonlinear interactions. Residual analysis (±0.4 BTE) and validation with 95% confidence bands (±0.42 BTE) confirmed strong predictive reliability and generalisation. The integration of carbon capture and utilization (CCU) pathways with methanol synthesis further supports closed-loop CO₂ utilisation, reinforcing methanol’s potential in carbon-neutral and carbon-negative applications. Overall, the findings highlight the importance of adaptive, AI-driven SOI optimisation and integrated CCU strategies for advancing next-generation low-emission engine systems.