Dynamic Model Based Neural Implicit Embedded Tracking Control Approach for Dielectric Elastomer Actuators With Rate-Dependent Viscoelasticity

In this work, we propose a dynamic model neural implicit embedded controller (called NEC) for high-precision tracking control of dielectric elastomer actuators (DEAs) by eliminating the rate-dependent viscoelasticity and mechanical vibration. To this end, we first establish a lumped parameter model for DEAs that can fully characterize the complex dynamic responses with creep, rate-dependent hysteresis, and mechanical resonance (4.21 and 6.85 Hz). Then, a neural network with the dynamic model is designed and trained to obtain feedforward control voltage for removing the nonlinearity of DEAs, which mainly consists of i) An encoder for extracting the features of the desired displacement and velocity; ii) A temporal decoder for calculating voltage based on those features. Finally, to further remove the model uncertainty and random errors, a feedback control strategy is adopted. The experimental results of tracking different complex trajectories (including sinusoidal, triangle, changing frequency, and amplitude) demonstrate that the NEC successfully eliminates the rate-dependent viscoelasticity and mechanical vibration of DEAs within a frequency range of 0.2 to 5 Hz. The maximum errors and the root-mean-square tracking errors are reduced to 6.98% and 2.72%, respectively, validating the effectiveness of our control approach. The dynamic model based neural network control strategy can self-adapt to different trajectories with changing frequency and amplitude, accelerating their applications in high-precision tracking control of DEAs.