A constitutive artificial neural networks-based mechanical model of the pneumatic artificial muscles

Abstract Pneumatic artificial muscles (PAMs), recognized as typical smart material actuators, have perennially presented a formidable challenge in the realm of precise mechanical modeling due to the hyperelasticity and nonlinearity. In order to construct the mechanical model of the PAM, we propose a constitutive artificial neural network-based mechanical model. Utilizing the constitutive artificial neural network (CANN), we have constructed a strain energy function for PAMs that satisfies symmetry, objectivity, and polyconvexity. Furthermore, by employing the principle of virtual work and considering the hyper-elastic material, the geometric constraints, and the deformation of the internal air chamber, we have derived the mechanical model of PAMs. To verify the accuracy of the proposed model, the finite element simulation is used to demonstrate the modeling accuracy under different load conditions for PAMs with different geometries and constitutive model conditions. Finally, the accuracy and generalization of the proposed model is validated through experiments on a PAM experimental platform.