Real-time monitoring of fat crystallization using pulsed acoustic spectroscopy and supervised machine learning

Abstract

Enhancing the control and yield of lipid crystallization is fundamental in several industrial areas, including pharmaceutical, cosmetic and food manufacturing. However, the multi-component nature of fats and oils poses a challenge in the understanding and control of the final product properties. While the crystallization of lipid has been extensively studied with offline techniques, online monitoring of the process would be highly advantageous, especially in large-scale sheared vessels. In this work, a novel method to calculate the solid fat content (SFC%) of crystallizing lipids under shear, based on an acoustic probe and supervised-machine learning, is presented. The temperature, composition and ultrasonic velocity of the samples, and the SFC(%) measured with nuclear magnetic resonance were used to develop a predictive model to calculate the SFC(%) during crystallization. Gaussian models showed the highest accuracy compared to linear and regression tree models (RMSE = 0.03 vs 0.7 and 0.25, respectively). • A novel method to calculate SFC% of lipids in shear crystallization was presented. • The method used an in-situ, non-destructive acoustic probe to sound velocity. • The acoustic data was coupled with pNMR and machine learning to predict SFC%. • A Gaussian regression model showed the highest accuracy in predicting SFC%.