Biodiesel has evolved as a sustainable alternative to fossil diesel because it is renewable and yields lower exhaust emissions. The composition and properties of biodiesel vary significantly depending on the feedstock. Consequently, the engine’s behavioral response to biodiesel is closely linked to the specific feedstock source. The complex interdependence between biodiesel feedstock and engine characteristics demands careful selection of suitable feedstock to achieve better engine performance and lower exhaust emissions. The novelty of the current investigation lies in addressing this challenge by selecting optimal feedstocks using a machine-learning framework to produce biodiesel fuel with improved engine characteristics. The proposed framework involves predictive analysis, in which models to estimate engine characteristics are developed using engine load and biodiesel composition as inputs. The engine characteristics of interest include brake-specific fuel consumption (BSFC), oxides of nitrogen (NO x ), unburned hydrocarbons (HC), and carbon monoxide (CO) emissions. The predictive analysis confirms the applicability of artificial neural network (ANN) regression, Gaussian process regression (GPR), support vector machine regression (SVM), and random forest (RF) for building reliable models to estimate engine characteristics, with errors under 5%. Biodiesel has limited applications due to higher BSFC and NO x emissions than diesel; hence, the optimization target simultaneously minimizes BSFC and HC, CO, and NO x emissions. The optimized biodiesel markedly improves fuel economy, with BSFC 26% higher than diesel, while coconut biodiesel exhibits a 41% higher BSFC. NO x levels from mustard biodiesel are 97% higher than diesel, while the proposed biodiesel yields only a 35% increase in NO x . A blend of olive, coconut, and canola oils in volume proportions of 85%, 10%, and 5% (±1%) yields a biodiesel nearly identical to the proposed composition. Thus, an optimal feedstock for producing biodiesel with minimal BSFC and NO x emission penalty was determined using the machine-learning approach employed in this study.
Bukkarapu et al. (Tue,) studied this question.