Analysis of results from the redesign of an electronic smell system for the benefit of cocoa quality Análisis de resultados a partir del rediseño un sistema de olfato electrónico para el beneficio de calidad del cacao

Abstract

The project focuses on the redesign of an electronic olfactory system used in the cocoa post-harvest process to determine the quality of the bean. An opportunity was identified to improve the design of the concentration and measurement chambers for volatile compounds, which involved reducing their size and the number of sensors. Initially, the system consists of a multisensory array of 9 MQ gas sensors, but in the redesign it is reduced to 6 sensors. This was because 3 sensors did not respond to the gas emissions emitted by the cocoa bean samples. The system consists of a PC with LabVIEW software to monitor and store data on volatile compounds in cocoa samples. It also includes an Arduino card connected to the multisensory matrix and actuator system, consisting of an air flow pump and three solenoid valves. Finally, the acquired data is processed using a Machine Learning algorithm to obtain the olfactory fingerprint that can discriminate samples of the fermented, overfermented and dry cocoa bean, the software implemented Python, the MinMaxScaler and Manhattan type of preprocessing, with Manhattan being the one that gave the best results in classification of samples with a total variance of 86%.