[P002] Poster

1:00 PM - 1:50 PM JST(4:00 AM - 4:50 AM UTC)

[P01-128] Machine learning based prediction model for free fatty acid removal during the esterification of mixed fat in biodiesel production

○Pitipat Phaprawet¹, Suphitchayanee Namboonlue¹, Kitisak Ngowsakul¹, Pakorn Uttayopas², Tunyaboon Laemthong¹ (1. Department of Chemical Engineering, Faculty of Engineering, Thammasat School of Engineering, Thammasat University (Thailand), 2. Department of Mechanical Engineering, Faculty of Engineering, Thammasat School of Engineering, Thammasat University, Pathum Thani 12120 (Thailand))

Keywords:

Mixed fat,Free fatty acids,Esterification,Machine learning

Mixed fat derived from the poultry processing industry represents a promising low-cost feedstock for biodiesel production due to its continuous availability and potential for waste valorization. Its composition, primarily triglycerides containing both saturated and unsaturated fatty acids, is suitable for conversion via transesterification. However, mixed fat typically contains a high level of free fatty acids (FFA), which leads to saponification during alkaline transesterification, catalyst consumption, emulsion formation, and reduced process efficiency. Consequently, an acid-catalyzed esterification step is required to reduce FFA before the main reaction. The efficiency of FFA reduction depends on multiple interacting process variables, including alcohol-to-oil ratio, catalyst loading, reaction temperature, and reaction time. These factors exhibit nonlinear relationships and complex interactions, making the identification of optimal operating conditions challenging using conventional experimental approaches. To address this complexity, we used Extreme Gradient Boosting (XGBoost) to model and predict FFA reduction under various esterification conditions. As an ensemble learning algorithm based on gradient boosting, XGBoost effectively captures nonlinear dependencies and multivariable interactions while incorporating regularization mechanisms to mitigate overfitting. Its flexibility in hyperparameter tuning and suitability for limited experimental datasets make it well-suited for modeling complex chemical reaction systems such as acid-catalyzed esterification of high-FFA feedstocks.