Machine learning predictions on the output parameters of CRDI engines fueled with ternary blend

Abstract

This study aims to employ a machine learning algorithm (MLA) to predict CRDI engine emissions and performance using alternative feedstock. This study started with a diesel-SCOME- Methyl Acetate ternary mix. The engine was tested with fuel injection time (FIT) of 23?, 21?, and 19? bTDC with exhaust gas recirculation (EGR) levels of 10%, 15%, and 20% at estimated power productivity. Retard injection time and increasing EGR rates reduced in-cylinder peak pressure. Operating conditions with the maximum BTE were 21? bTDC and 10% EGR. Adjusting injection time and EGR reduced nitrogen oxide relative to the baseline. Smoke opacity was 1% lower at 21? bTDC and 10% exhaust gas recirculation than in conventional diesel operation. Retard injection time and exhaust gas recirculation increased HC and CO emissions. However, MLAs predict CI engine operation and discharge properties. The long short-term memory (LSTM) Model predicts engine output characteristics with a squared correlation (R2) of 0.92 to 0.961. At the same time, mean relative error (MRE) values ranged from 1.74 to 4.68%. These results show that the LSTM models provide superior predictive capabilities in this investigation, particularly when considering numerous variables to analyse engine responses.