Bayesian-Optimized Transesterification and EVA-Assisted Cold-Flow Control for Non-Edible Renewable Biodiesel in Heavy-Duty Fleets

Abstract

Biodiesel from non-edible feedstocks remains a viable route to a low-carbon freight fuel because it can be blended with petroleum diesel in existing compression-ignition fleets. Jatropha oil and waste cooking oil provide a suitable feedstock combination. This work develops an integrated experimental-modeling pipeline for Jatropha/WCO biodiesel intended for B20 heavy-duty transport use. A 50-run Central Composite Design result was analysed, and ANN, XGBoost, and ensemble models were trained with Bayesian optimization and SHAP-based feature ranking; fuel properties, pour-point depressants, and fleet economics were then evaluated. Cold-flow behavior was screened for EVA, PMA, and alkyl-naphthalene additives, and a Monte Carlo techno-economic model was applied to a 100-vehicle Class 8 fleet. The ensemble model identifies a high-yield operating window, and the optimized conditions yielded 97.4% predicted FAME and 97.1 ± 0.6% measured FAME. EVA at 1.0 wt% delivered a CFPP of −12 °C for B20, and the fleet model yielded a 10-year NPV of USD 2.91 million with an 18.0% CO₂-equivalent reduction. These outcomes establish a linked route from process chemistry to winter operability and fleet economics. Future work should extend the framework to include oxidation stability, broader feedstock variability, and pilot-scale validation of the renewable fleet.