IMPACT OF MACHINE LEARNING APPROACH USING ANN AND RSM TO EVALUATE THE ENGINE CHARACTERISTICS OF A DUAL-FUEL CI ENGINE

Abstract

The surge in fossil fuel consumption has severely impacted the environment, namely in terms of climate change, due to the influence of extensive pollution. The current study assesses and contrasts the ability of artificial neural networks (ANN), a machine learning technique, and a response surface methodology (RSM) derived model to predict important engine characteristics. The effect load (25%, 50%, 75%, and 100%), speed (1500 and 1800 RPM), compression ratio (17.5 and 18.5), and diesel-biodiesel blends (diesel, SM₂₀ , SM₄₀ , SM₆₀ , SM₈₀ , and SM₁₀₀) were investigated on a test engine (4-S single-cylinder DI diesel engine). Box-Behnken designs (BBDs) of RSM and a multi-layer perceptron (MLP) neural network with a topology of 4-10-6 were employed to study the principal engine performance (brake thermal efficiency 31.82% and 30.12%, brake-specific fuel consumption 0.2608 and 0.28 kg/kWh, and indicated mean effective pressure 5.85 and 5.35 bar) and emission (carbon dioxide 827 and 885 g/kwh, nitrogen oxides 1391 and 1247 ppm) parameters, respectively, for RSM and ANN. The projected outcomes showed below 10% error in almost all results when compared with experimental results. The outcomes of the present study reveal that RSM (with a regression coefficient of 0.997) and ANN (with training and test regression coefficients of 0.9967 and 0.984) can be employed to model processes that exhibit high predictability.