Intention Recognition of Elbow Joint based on sEMG Using Adaptive Fuzzy Neural Network

Abstract

In this paper, the adaptive fuzzy neural network (AFNN) based on the surface electromyography (sEMG) for estimating the elbow joint angle is established and investigated from the perspective of rapidity and accuracy. In addition, back propagation neural network (BPNN) and artificial neural network of radial basis function (RBFNN), as the classical method for data forecasting, have been applied to estimate the elbow joint angle for comparing with AFNN. Ultimately, the experimental simulation and result analysis demonstrate that the rapidity and accuracy of AFNN is superior to BPNN and RBFNN.

Design of Optimal Sliding Mode Control of Elbow Wearable Exoskeleton System Based on Whale Optimization Algorithm

“…The robot is worn by persons at the upper limb to help treatment of impaired upper limb functions due to stroke. Liu et al [22] have established adaptive fuzzy neural network (AFNN) for estimating the angle of elbow joint based on surface electromyography (sEMG). The AFNN showed better accuracy and rapidity as compared with other neural network structures, back-propagation-based NN and Radial basisbased NN.…”

“…( 18) and Eq. ( 20), the first-time derivative of sliding surface equation is given by: 𝑠̇= 𝑐 𝑒̇+ 𝑥̇2 − 𝑥̈1 𝑑 (22) where, c is defined as scalar design parameter. According to Eq.…”

A neural network-based model for lower limb continuous estimation against the disturbance of uncertainty

“…where y(t) represents the time-varying state variable, u(t) and E(t) are the input function and output function of a nonlinear dynamic model. Therefore, the error function is designed to be Ė(t) = −λE(t) and a set of n × n decoupled differential equations is equivalently generalized as ėi,j (t) = −λe i,j (t), (20) where i, j ∈ {1, 2, 3, ..., n}, λ > 0 is related to the rate of convergence. Theorem 1: When the error function e i,j (t) satisfies (20), the state variable y(t) of nonlinear dynamic system (19) globally and exponentially converges to y * (t), the control law can be generalized as…”

“…However, it is difficult to construct the HMM for practical application owing to some physiological parameters that cannot be directly measured. On the other hand, the regression model is constructed between the sEMG signals and joint motion information to predict human motor intention, among which neural network is the most commonly applied [19][20][21][22]. In [19], continuous wavelet transform integrated into back propagation neural network is proposed to estimate the continuous motion intention of the upper limb via the sEMG signals under human-robot interaction, but the correlation between the selected muscles and the recognized movements is not considered.…”

“…In [19], continuous wavelet transform integrated into back propagation neural network is proposed to estimate the continuous motion intention of the upper limb via the sEMG signals under human-robot interaction, but the correlation between the selected muscles and the recognized movements is not considered. In [20], an improved fuzzy neural network is applied to the elbow joint angle estimation, and the feasibility of this method is proved by comparative experiments. To estimate the posture of the upper limb in continuous motion, the SCA-LSTM algorithm is proposed, which reduces noise interference by custom maximum root mean square envelope preprocessing method [21].…”