Cerium Oxide Blended Biodiesel: Machine Learning and Response Surface Methodology for Prediction of Responses

Abstract

Because of the rising demand for diesel, researchers are looking into finding a new alternative fuel. Biodiesel is an excellent alternative to neat diesel due to its renewable, biodegradable, and non-toxic nature. However, its response characteristics, such as brake thermal efficiency (BTE) and brake-specific fuel consumption (BSFC) should be predicted correctly. Thus, the present work includes the production of biodiesel from cottonseed oil with two-step transesterification and the investigation of response characteristics for a single-cylinder diesel engine fueled with Cerium oxide, i.e. nanoparticle additive (NA), which is blended cottonseed oil biodiesel. Compression ratio (CR) and NA levels have varied from 16 to 18 and 50 to 100 ppm, respectively. Input parameters, namely CR and NA levels are considered for the present investigation. The present study presents a novel method that uses deep learning-based surrogate modelling, a machine learning (ML) technique to forecast the responses. The optimum operating conditions are a CR of 18 and an NA level of 83.877 ppm. The study results demonstrate that the deep learning model provides a convincing substitute for classical regression models such as Random Forest, Decision Tree, Support Vector Machines, Gradient Boosting, and K-Nearest Neighbor Regressor. Further, multi-objective optimization of input parameters is performed using the desirability function approach. The optimized parameters were attained at a composite desirability of 0.847. Lastly, confirmation experiments are performed to validate the results of non-linear regression models and found satisfactory with an error percentage of less than five.