Control of shape memory alloy actuators with a neuro-fuzzy feedforward model element

This paper describes development of a motion controller for Shape Memory Alloy (SMA) actuators using a dynamic model generated by a neuro-fuzzy inference system. Using SMA actuators, it would be possible to design miniature mechanisms for a variety of applications including miniature robots for micro manufacturing. Today SMA is used for valves, latches, and locks, which are automatically activated by heat. However it has not been used as a motion control device due to difficulty in the treatment of its highly nonlinear strain-stress hysteresis characteristic. In this paper, a dynamic model of a SMA actuator is developed using ANFIS, a neuro-fuzzy inference system provided in MATLAB environment. Using neuro-fuzzy logic, the system identification of the dynamic system is performed by observing the change of state variables (displacement and velocity) responding to a known input (input voltage to the current amplifier for the SMA actuator). Then, using the dynamic model, the estimated input voltage required to follow a desired trajectory is calculated in an open-loop manner. The actual input voltage supplied to the current amplifier is the sum of this open-loop input voltage and an input voltage calculated from an ordinary PD control scheme. This neuro-fuzzy logic-based control scheme is a very generalized scheme that can be used for a variety of SMA actuators. Experimental results are provided to demonstrate the potential for this type of controller to control the motion of the SMA actuator.