Characterizing cocoa refining by electronic nose using a Kernel distribution model

Abstract

Refining and conching are two important processes for chocolate manufacturing, as they help improve the flavor and texture of chocolates. However, methods such as GC-MS for assessing flavor changes are expensive. In this study, an e-nose was used to continuously monitor the volatile compounds of cocoa samples undergoing refining, for samples differing in sample mass and degree of roasting. The responses of the e-nose were characterized by three parameters (Rarea, Rpeak and Rwidth) that were able to detect the overall influence of roasting. These values along with sample mass were also used to train Kernel Distribution Models (KDM) which were implemented to better account for temperature and air flow fluctuations. For trained KDMs validated with data taken under the same conditions (sample mass, degree of roasting) the error was ∼1%. When validated under different conditions the error ranged from 2.9 to 9.3%. The degree of roasting had greater affect on the error than the sample mass. A trained KDM can be used to predict the overall volatile compounds at different refining stages, and detect whether the processing variables of sample mass and roasting degree varied from its training samples.