Behavioral modeling and experimental verification of a smart servomotor used in a thermal control louver of a satellite using dynamic neural network-based NARX

Louvers are powerful devices for the thermal management of satellites. Nevertheless, the high mass and power consumption and the low reliability of servomotors serving as the actuators of louvers, make the space applications of these technologies very restricted. To tackle this problem, this paper utilizes a Shape Memory Alloy (SMA) to build a smart servomotor for using in a laboratory louver. The major bottleneck of the use of thermal SMAs is the existence of complex nonlinear hysteretic characteristics in the behavior of these materials. In this paper, a nonlinear autoregressive exogenous model is proposed to predict the nonlinear hysteric behavior of an SMA. This model is based on a dynamic neural network that its fine function is achieved by a suitable selection of the architecture and the transfer functions of the output and hidden layers. The proposed model is first trained with a batch of test data at the frequency of 0.01 Hz and then validated with another batch of data at the frequency of 0.008 Hz. The training and validation data are obtained from a laboratory louver equipped with an SMA spring as the opening actuator of blades. The mean square error of the proposed model for the training and validation data is 1.0325 and 1.0835 degrees, respectively.