FRACTIONAL ORDER PIλDμ FOR TRACKING CONTROL OF A NOVEL REHABILITATION ROBOT BASED ON IIMO-BP NEURAL NETWORK ALGORITHM

Abstract

In this study, we develop a novel multi-posture lower limb rehabilitation robot with three postures, which can provide different amplitudes and frequencies of rehabilitation training for hip, knee and ankle joints, respectively. The kinematic and dynamic analyses of the robot are carried out to solve the kinematic forward and backward solutions and the Lagrangian dynamics equations of the lower limbs. The angle, angular velocity and angular acceleration ideal trajectory curves of the rehabilitation motion are derived by using a quintic polynomial trajectory planning scheme. An improved ions motion optimization (IIMO) algorithm is proposed and applied to optimize the initial weight of back propagation (BP) neural network, and algorithm is used to adjust five parameters of fractional order [Formula: see text] ([Formula: see text]) control in controller design. The passive training experiment results of prototype show that the designed controller has the largest average error of angle and angular velocity of hip, knee and ankle joints in high amplitude and high frequency movement mode, which are 1.091∘, 0.716∘, 0.412∘, 1.551∘/s, 1.394∘/s, 1.498∘/s, respectively. At low amplitude and low frequency, the maximum average errors are the smallest, which are 0.351∘, 0.341∘, 0.167∘; 0.833∘/s, 0.842∘/s, 0.398∘/s, respectively. The actual trajectory curve fits well with the designed one. The highest accuracy of angle and angular velocity can reach 99.165% and 99.116% through comprehensive comparison of all motion modes. Therefore, the overall error is small. The stability of rehabilitation training process is ensured, and the rationality and effectiveness of trajectory planning and control design are verified.