Research On Self-sensing SMA Wire Actuator Based On Resistance And Recurrent Neural Network

Shape memory alloy(SMA)wires are increasingly used to manufacture various linear actuators due to their excellent material properties.SMA wires are used in compact devices such as camera modules due to its small size,but additional sensors will squeeze a lot of effective space.The resistance of SMA wire will be affected by its own resistivity,length and cross-sectional area changes when the electrical excitation signal is applied.Therefore,the motion state of SMA wire actuator can also be reflected by detecting the resistance value in the phase transformation process,which is also known as the self-sensing characteristics of SMA wire.The displacement of the actuator can be predicted by the resistance of the SMA wire when the load is constant,and objects of different stiffness can also be distinguished by the resistance of the SMA wire when the displacement is constant.These studies help monitor the working state of SMA wires without external sensors,save equipment space,and help simplify SMA wires in engineering applications.The main research contents are as follows:1.A simple temperature stress model under constant length thermal cycling condition is proposed by combining the simplified phase transformation dynamics equation and the constitutive equation.The theoretical model of the voltagetemperature-stress relationship is also established by combining the thermodynamic equations,and the accuracy of the model is verified through experiments.Under the condition of constant displacement,the effective method of estimating the stress of the SMA wire actuator based on the free response frequency of the SMA wire vibration is proposed and verified.2.A new kind of phase transformation detection method is proposed for resistance changes under constant load conditions.This method improves the existing detection methods,increases the avoidance of abnormal values,and can support the phase transformation detection under various of electrical excitation signal.These increase the robustness of the phase transformation detection method to a certain extent.This method can well reflect the phase transformation state of the SMA wire through the resistance and resistance change trend,detect the start and end points of the phase transformation,and can well distinguish the thermal equilibrium point and the end point of the phase transformation.This turns on the thermal protection of the SMA wire selfsensing actuator and reduces its power consumption.3.A voltage-displacement prediction model is established by the weighted superposition of the hysteresis operator based on PI model,but this method is limited by the form of electric excitation signal.A piecewise linear resistance-displacement model is proposed based on the resistance change equation and strain equation,but this model has problems such as large amount of calculation and complicated operation.Based on the above shortcomings,the displacement prediction models based on BP neural network and LSTM neural network are constructed respectively.The results show that the best prediction effect of displacement output is achieved by using LSTM neural network in the form of voltage and resistance as input signals.The motion state of the actuator can be estimated without additional displacement sensors,that is,the displacement prediction function of SMA wire self-sensing actuator can be realized.4.Under constant displacement condition,objects with different stiffness are classified by BP neural network and LSTM neural network respectively with resistance as input signal.The results show that LSTM neural network can distinguish objects with different stiffness well,which means that the stiffness can be predicted by the trend of resistance change,which means that the SMA wire self-sensing actuator can be used to classify and detect the test parts without additional sensors.