Neuro-fuzzy operational performance of a coffee harvester machine Marcelo de Carvalho Alves, Fábio Moreira da Silva, Tomás de Aquino Ferreira, Flávio Castro Silva Neuro-fuzzy operational performance of a coffee harvester machine Marcelo de Carvalho Alves *1 Corresponding author , Fábio Moreira da Silva 2 , Tomás de Aquino Ferreira 3 , Flávio Castro Silva 4 *1 Federal University of Mato Grosso, Faculty of Agronomy and Veterinary Medicine, Department of Soil and Rural Engineering, Av. Fernando Correa da Costa S/N, Coxipo District, 78060-900, Cuiabá-MT, Brazil 2,3,4 Federal University of Lavras, Engineering Department, P.O.Box 3037, 37200-000, Lavras - MG, Brazil mdecalves@ufmt.br, famsilva@ufla.br; tomasaf@ufla.br; flavioufla@globo.com doi: 10.4156/jcit.vol4.issue2.alves Abstract The objective of this work was to develop and to evaluate neuro-fuzzy systems as a methodology to describe coffee harvester machine operational performance when compared to multiple regression models. It was considered as input variables fruit maturation index, in the levels of 75.70, 87.00, 98.70%, operational speed, in the levels of 0.16, 0.26, 0.57m.s-1 and rods vibration, in the frequencies of 13.33, 15.00, 16.66, 18.33Hz. Coffee fruit harvest efficiency and plant leaf fall were considered as output variables. Hybrid neural network training was applied to input and output data in order to optimize fuzzy systems parameters for coffee fruit harvest efficiency and plant leaf fall prediction. Neuro-fuzzy models presented better performance when compared to multiple regression models. Based on developed neuro-fuzzy systems control maps, levels of speed and vibration could be recommended according to fruit maturation stage in the field. Keywords Coffea arabica L., Anfis, coffee lateral harvester. 1. Introduction Coffee fruit harvest machines can be used to reduce operational costs, improve yield quality and efficiency in plane and moderately sloping lands [8,10]. Therefore, there is a tendency for expansion of coffee crop mechanization due to human’s valorization and optimization of coffee harvest process [9]. However, machine speed variations, harvester fiberglass rods vibration penetrating sidelong the plants and fruit maturation stage are decisive factors for a successful coffee mechanical harvest operational performance [4,10]. Santinato et al. [7] compared statistical averages to evaluate Jacto K3 ® harvest machine performance in ‘Mundo Novo’ and ‘Catuaí’ cultivars under speed of 0.33 m.s -1 and vibration of 10.00, 13.33 and 16.67Hz. Operational performance improvement was observed under 3 times machine pass and vibrations of 13.33 and 16.67Hz. However, Silva et al. [10] studying operational performance of a lateral coffee harvester machine using techniques of basic statistics, concluded that machine harvest efficiency improvement occurred under speed of 0.16m.s -1 and rods vibration of 16.67Hz. In spite of that, fruit maturation, harvester speed and rods vibration interaction were not considered in data analysis. Considering that coffee harvest could be influenced by many large scale factors with built-in uncertainties, Soft Computing, which is an integrated approach based on Artificial Intelligence techniques, such as Neural Networks, Fuzzy Logic and Genetic Algorithms [3], could be used to construct generally satisfactory solutions for harvester machines operational performance pattern characterization. Those techniques have been applied to characterize physical, chemical and biological processes [5,16] as well as in automotive engineering [15] and industrial process control [13], with higher accuracy and precision when compared to classic statistics. Fuzzy modeling advantages were also related to flexibility of incorporation of additional variables, representation of human expertise knowledge using fuzzy if-then rules (computing with words), modeling complex nonlinear functions, dealing with perception, pattern recognition and classification problems, learning at unknown or changing environment and easiness integration with sensors and soft computing methods, such as neural networks and genetic algorithms [5,14,17]. 52