Search for content and authors

Micropositioning using shape memory alloy actuators

Estibaliz Asua ,  Alfredo García-Arribas ,  Victor Etxebarria 

Departamento de Electricidad y Electrónica (UPV/EHU), Apartado 644, Bilbao 48080, Spain

Actuators based on Shape Memory Alloys SMA benefit from a high power to weight ratio and have found applications in many areas. They have traditionally been used as “on-off” electromechanical actuators due to the non-linear and hysteretic nature of the martensite-austenite transformation. To accurately command the actuator to an intermediate position within its range of movement it is necessary to use feedback control. The ideal solution is to model the hysteresis mathematically so it can be compensated in the control loop. However, useful models of phase transitions are difficult to obtain. Our approach to the problem is to model empirically the behavior of the wire. For this purpose we have trained a neural network.
The experimental set-up uses a commercial Nitinol wire, 75 mm long and 0.15 mm thick, suspended from a fixed support with its lower end fastened to the core of a LVDT (linear variable differential transformer) position sensor. An additional weight is hanged to the wire to make it recover its original length. The wire is joule-heated by a controlled current. The resolution attainable by the position sensor is half a micron, approximately.
The neural network is trained by applying currents to the wire with different amplitudes and frequencies. After the learning process, the neural network performs reasonably well as shown in Figure 1.

When used in a controller, the neural network nicely compensates the non-linear and hysteretic behavior of the wire, and any simple lineal controller works adequately. We have used a proportional-integral with anti-windup control, obtaining excellent results. Figure 2 shows the performance of the controller for set-points at 0.5 and 2 mm. The inset shows the magnitude of the error.

In conclusion, we demonstrate that shape memory alloys can be used as actuators with micrometer accuracy, provided that a suitable control strategy is implemented. A neural network is very effective for hysteresis compensation.

hysteresis_1.gif step_1.gif
Figure 1 Figure 2

Legal notice
  • Legal notice:

Related papers

Presentation: Oral at E-MRS Fall Meeting 2007, Symposium E, by Alfredo García-Arribas
See On-line Journal of E-MRS Fall Meeting 2007

Submitted: 2007-03-06 13:21
Revised:   2009-06-07 00:44