Offline ANN-PID Controller Tuning on a Multi-Joints Lower Limb Exoskeleton for Gait Rehabilitation

Abstract

This paper presents Artificial Neural Network (ANN) as an optimization tool in tuning Proportional-Integral-Derivative (PID) controller’s gain of a multi-joints Lower Limb Exoskeleton (LLE) for gait rehabilitation. The interest in wearable post-stroke and spinal cord injury rehabilitation devices such as LLE has been increasing due to the demand for assistive technologies for paralyze patients and to meet the concerns in the increasing number of ageing society. The dynamic of three degree of freedom LLE was determined using Euler-Lagrange equation, and PID parameters were initially tuned using the Ziegler-Nichols (ZN) method. The paper compares different ANN-based algorithms in tuning PID controller’s gain for LLE applications. The method compared and evaluated with other methods and dynamic systems in the literature. ANN-based algorithms, Gradient Descent, Levenberg-Marquardt, and Scaled Conjugate Gradient, are utilized for PID tuning of each joint in the LLE model. The result shows faster convergence and improves step response characteristics for each controlled joint model. The overshoot values found to be 0.3126%, 0.6335%, and 0.2619% compared to the ZN method with 10.5582%, 15.1643%, and 11.8511% for hip, knee, and ankle joints, respectively. It can be ascertained that the PID controlled of LLE has been optimally tuned significantly by different ANN methods, which reduced its steady-state errors.

Keywords

• Proportional-integral-derivative
• Lower Limb Exoskeleton
• Artificial Neural Network
• Euler-Lagrange