This paper presents a new data-driven model of length-pressure hysteresis of pneumatic artificial muscles(PAMs) based on Gaussian mixture models(GMMs). By ignoring the high-order dynamics, the hysteresis of PAMs is mo...This paper presents a new data-driven model of length-pressure hysteresis of pneumatic artificial muscles(PAMs) based on Gaussian mixture models(GMMs). By ignoring the high-order dynamics, the hysteresis of PAMs is modeled as a first-order nonlinear dynamical system based on GMMs, and inversion of the model is subsequently derived. Several verification experiments are conducted. Firstly, parameters of the model are identified under low-frequency triangle-wave pressure excitations.Then, pressure signals with different amplitudes, shapes and frequencies are applied to the PAM to test the prediction performance of the model. The proposed model shows advantages in identification efficiency and prediction precision compared with a generalized Prandtl–Ishlinskii(GPI) model and a modified generalized Prandtl–Ishlinskii(MGPI) model. Finally, the effectiveness of the inverse model is demonstrated by implementing the feedforward hysteresis compensation in trajectory tracking experiments.展开更多
基金supported by the National Natural Science Foundation of China (Grant No. 91648104)Shanghai Rising-Star Program (Grant No. 17QA1401900)
文摘This paper presents a new data-driven model of length-pressure hysteresis of pneumatic artificial muscles(PAMs) based on Gaussian mixture models(GMMs). By ignoring the high-order dynamics, the hysteresis of PAMs is modeled as a first-order nonlinear dynamical system based on GMMs, and inversion of the model is subsequently derived. Several verification experiments are conducted. Firstly, parameters of the model are identified under low-frequency triangle-wave pressure excitations.Then, pressure signals with different amplitudes, shapes and frequencies are applied to the PAM to test the prediction performance of the model. The proposed model shows advantages in identification efficiency and prediction precision compared with a generalized Prandtl–Ishlinskii(GPI) model and a modified generalized Prandtl–Ishlinskii(MGPI) model. Finally, the effectiveness of the inverse model is demonstrated by implementing the feedforward hysteresis compensation in trajectory tracking experiments.
