Machine Learning Models for Predicting Piston Surface Deposit Formation, PM and Noise Emissions Using Ternary Blended Fuels in SI Engine

To ensure compatibility and evaluate the performance of renewable and non-carbon fuels for spark ignition engines over extended periods, long-term forecasting is essential. In order to lessen carbon deposition on SI engine parts and noise emissions, this study assessed the usage of ethanol and n-butanol as fuel additives. Three fuel samples were used in the current study, namely: gasoline 90% + 10% Ethanol, gasoline 90% + 10 n-Butanol, or gasoline 80% + 10% Ethanol + 10% n-Butanol, in addition to pure gasoline fuel for comparison. Carbon deposition on the engine piston crown and noise emission levels were experimentally measured using endurance testing for two hundred engine operating hours at a constant load and speed. Furthermore, artificial neural network (ANN) and long short-term memory (LSTM) models were developed and trained for emission analysis. The ANN model was used to predict emission and deposition parameters from the experimental dataset within the engine measurement period, while the LSTM model was employed to forecast the temporal evolution of carbon deposition and particulate emissions beyond the measurement period. Both models demonstrated high predictive accuracy and strong agreement with experimental observations. The carbon deposition formation rate on the top surface of the piston for ethanol/n-butanol blend baseline fuel was found to be lower than that of gasoline fuel, with a deposition rate of PF 100 is 71.85% and for 10% Ethanol is 51.76% added PF90E10. Alternatively, piston top surface deposition was reduced when n-butanol was added to gasoline fuel, with a deposition rate of PF90nBn10 is 34.47% for 10% n-Butanol added, and PF80E10nBn10 is 29.27% for 10% Ethanol and 10% n-Butanol added. Additionally, utilizing the four fuel samples, the study examined noise emissions during the whole testing hours. Comparing n-butanol/ethanol/gasoline PF80E10nBn10 ternary blend fuel, it demonstrated fewer noise emissions, suggesting that it can aid in lowering noise emissions from spark ignition engines. The ethanol/n-butanol ternary blend combination produced 15–20% fewer PM1, PM2.5, and PM7 emissions compared to pure petrol fuel, which is crucial for the fine/very fine particle air pollution problem. These findings demonstrate the suitability of ethanol/n-butanol blends and the effectiveness of the developed ML framework for long-term engine emission assessment.