Emprego de Sistemas de inferência difusa neuro-adaptativa, geradores de modelos de veículos militares representativos, para adestramento nas Forças Armadas

Abstract

This work addresses the representativeness of dynamic mathematical models built using Neuro-Adaptive Fuzzy Inference Systems (ANFIS), which can serve as inputs for training devices, whether virtual, constructive simulators or military systems, which involve the use of military vehicles. The study was carried out with three vehicles: one land, one sea and one air. For each vehicle, input parameters that involved greater human-machine interface were used, as well as important environmental parameters that created greater interference with the vehicles' behavior. ANFIS models were trained with real and theoretical data. The results showed that the models for land and sea vehicles presented a strong representation, with correlation coefficients of 0.98 and 0.99, respectively. For the aerial vehicle, the results were more variable. The model for the ground speed parameter showed strong representation, with a correlation coefficient of 0.99. However, for the parameters vertical speed, roll angle, pitch angle and yaw angle, the models showed moderate to poor representation. According to European Aviation Safety Agency (AESA) criteria, models for the roll angle, pitch angle and yaw angle parameters would be inappropriate for use in virtual simulation as they did not achieve the required accuracy of 100 %, considering an error tolerance of 20%. Overall, the results of the study showed that ANFIS models can be a promising tool for representing complex dynamic systems, such as those of military vehicles. However, it is necessary to carry out further studies to improve the representativeness of models for specific parameters, which require better data filtering, or which present discontinuities.