Optimizing Thermal Efficiency in Diesel Engines: Predicting Performance with Ternary Blends, Variable Injection Pressures and EGR Using LSTM Machine Learning

Abstract

Modern society prioritizes Sustainable Development Goals (SDGs 7 and 13) to address the fuel requirements of transportation and agriculture, concentrating on clean energy and climate change mitigation. This study examines the combination of Simmondsia chinensis (jojoba) biodiesel and methyl acetate (MA) to improve combustion efficiency and decrease emissions in a Common Rail Direct Injection (CRDi) engine. The ternary test fuels comprised diesel, biodiesel (SCB), and MA additives, formulated as DB50 (50% diesel + 50% biodiesel), DBMA10 (50% diesel + 40% biodiesel + 10% MA), and DBMA20 (50% diesel + 30% biodiesel + 20% MA). Tests performed at 21ºCA for fuel injection time, with varied fuel injection pressures (FIP: 400, 500, 600 bar) and exhaust gas recirculation (EGR: 0, 10, 20%), demonstrated that DBMA20 enhanced brake thermal efficiency by 1.02% relative to DB50. NOx emissions decreased by 32.3% and 18.23% in DB50 relative to diesel at 400 bar fuel injection pressure and 20% exhaust gas recirculation. DBMA20 elevated smoke opacity and CO, HC emissions while decreasing FIP and augmenting EGR. Secondly, nonlinear test results and repetitive engine testing make improving IC engine performance with alternate fuels difficult. This challenge is solved using generalisable machine learning models and engine variable optimisation. Machine learning-based long-short-term memory (LSTM) models anticipate and optimise a CRDi engine that runs on ternary test fuel with various injection strategies for FIP and EGR experimental data as an input. This model accurately predicts thermal efficiency, fuel consumption, NOx, HC, CO, and smoke opacity. LSTM predicted R2 values of 0.91-0.991, with an MRE of 1%-5%. Best CRDi engine configuration: DBMA20 @ 600 bar FIP, 10% EGR. LSTM improves R2 and reduces MRE to enhance engine performance. An R2 value close to 1 is expected. It can conclude that the machine learning based forecasting method is an effective tool for assessing the in depth engine operation relation among input variables.