Hybrid deep learning and RSM modeling of diesel engine performance using TiO

This study investigates the performance and emission features of a single-cylinder diesel engine fuelled with waste plastic oil (WPO) blended with diesel, 1-butanol, and TiO2 nanoparticles. Novelty lies in the synergistic doping of TiO2 nanoparticles with 1-butanol, which enhances durability, fuel efficiency, and emission reduction compared to conventional diesel. Test blends include 90D7PO3B + 25ppmTiO2, 85D10PO5B + 50ppmTiO2, 80D13PO7B + 75ppmTiO2, 100PO + 100ppmTiO2, and neat diesel (100D). The blend 80D13PO7B + 75ppmTiO2 achieved the highest brake thermal efficiency (37.3%) and the lowest brake-specific fuel consumption (0.22 kg/kWh), while 85D10PO5B + 50ppmTiO2 recorded the maximum cylinder pressure (72 bar). Hydrocarbon and CO emissions were minimized with 100PO + 100ppmTiO2 (50 ppm, 0.035% respectively), whereas CO2 and NOx emissions were lowest for 80D13PO7B + 75ppmTiO2 (8% and 1800 ppm). To further explore and optimize multi-factor effects, Response Surface Methodology (RSM) was combined with deep machine learning models. RSM constructed quadratic surrogates, validated using ANOVA, revealing nonlinear curvature effects across responses. Multi-objective optimization identified an operating point at IMEP 7.7 bar and LCV 42,597 kJ/kg, yielding BTE 35.26%, BSFC 0.22 kg/kWh, CO 0.018%, HC 40.54 ppm, CO2 5.89%, and NOx 1345 ppm, confirming the classical efficiency NOx trade-off. Predictive modelling using Bayesian neural networks (BNN) and linear regression demonstrated that BNN consistently outperformed LR, achieving higher R2, lower MSE, and tighter residual alignment across all responses. The integration of TiO2–1-butanol doping with advanced predictive optimization demonstrates a viable alternative fuel pathway, delivering improved performance, reduced emissions, and statistically validated machine learning–RSM frameworks for future engine research.