Artificial neural networks (Fuzzy ARTMAP) analysis of the data obtained with an electronic tongue applied to a ham-curing process with different salt formulations

This paper describes the determination of optimum values of the parameters of a Simplified Fuzzy ARTMAP neural network for monitoring dry-cured ham processing with different salt formulations to be implemented in a microcontroller device. The employed network must be set to the limited microcontroller memory but, at the same time, should achieve optimal performance to classify the samples obtained from this application. Hams salted with different salt formulations (100% NaCl; 50% NaCl+50% KCl and 55% NaCl+25% KCl+15% CaCl2+5% MgCl2) were checked at four processing times, from post-salting to the end of their processing (2, 4, 8 and 12 months). Measurements were taken with a potentiometric electronic tongue system formed by metal electrodes of different materials that worked as nonspecific sensors. This study aimed to discriminate ham samples according to two parameters: processing time and salt formulation. The results were analyzed with an artificial neural network of the Simplified Fuzzy ARTMAP (SFAM) type. During the training and validation process of the neural network, optimum values of the control parameters of the neural network were determined for easy implementation in a microcontroller, and to simultaneously achieve maximum sample discrimination. The test process was run in a PIC18F450 microcontroller, where the SFAM algorithm was implemented with the optimal parameters. A data analysis with the optimized neural network was achieved, and samples were perfectly discriminated according to processing time (100%). It is more difficult to discriminate all samples according to salt formulation type, but it is easy to achieve salt type discrimination within each processing block time. Thus, we conclude that the processing time effect dominates salt formulation effects. (C) 2015 Elsevier B.V. All rights reserved.