Identificação e previsão neural de atuadores com memória de forma na presença de deformação residual significativa

Abstract

This work aims to develop of artificial neural networks to identify the behavior of shape memory alloys applied as thermomechanical actuators, presenting considerable irreversible deformation. The accumulation of the residual deformation during the phase transformation of the shape memory actuators degrades the dimensional stability and modifies its hysteretic behavior, which difficults the design and control of these materials. For the learning of the neural model, nickel-titanium wires are exposed to two distinct conditions, training the shape memory material through twenty-five phase transformation cycles and the subsequent thermal cycling for one thousand five hundred cycles. The data of temperature and deformation of these two situations are used for identification and for validation of the two implemented models, the one-stepahead prediction and the multiple predictions of the evolution of the irreversible deformation. From the comparison between the experimental data and the estimates valus, it is observed that the one-step-ahead neural network adequately characterizes the elongation of the actuator, the decrease of the maximum transformation strain and the degradation of the actuator hysteresis. Further, although the performance of the multiple output neural architecture exhibits an average relative absolute error close to 3% for the studied conditions, a smaller accuracy in the identification of the shape memory material is observed, especially in the maximum and minimum values of the deformation in each cycle.