RBF based adaptive neuro-fuzzy inference system to torque estimation from EMG signal
- Publication Type:
- Conference Proceeding
- Citation:
- 2017 IEEE Symposium Series on Computational Intelligence, SSCI 2017 - Proceedings, 2018, 2018-January pp. 1 - 8
- Issue Date:
- 2018-02-02
Closed Access
| Filename | Description | Size | |||
|---|---|---|---|---|---|
| 08285440.pdf | Published version | 750.39 kB |
Copyright Clearance Process
- Recently Added
- In Progress
- Closed Access
This item is closed access and not available.
© 2017 IEEE. There are several knee joint dynamics parameters such as knee joint torque, joint angle, joint damping that are associated with muscle EMG (Electromyography) signal. But EMG signal being dynamic in time and space, it is on lots of occasion difficult to be mapped to the estimation of knee joint dynamics. EMG signal is a function of velocity, angle and muscle activation level. Under such circumstances, it is a challenging task to map EMG to knee joint torque unless the data is collected following a very restricted and very specific protocol. But unfortunately such ideal condition of fixed protocol does exist practically, when the lower limb has to encounter and undergo different experiences in terms of various dynamic conditions. There are three different RBF Networks available with Radial Basis Function inside the neurons at the hidden layers. In Adaptive Neuro Fuzzy Inference System type non parametric model where Fuzzy membership function with rules based inference system of the first two layer decide about which RBF Network at the hidden layer to be used to estimate Knee Joint Torque. An ANFIS-RBF model has been proposed to be able to choose RBF Network adaptively each trained with one specific type of EMG-Torque profile at the Hidden Layer to address the EMG signal whose relationship to knee joint torque is changing dynamically with respect to velocity, position. This model can be imagined as two layer controller where Fuzzy decision making ability is in the higher layer and Neural Network learning ability of muscle in the lower layer. The high level Fuzzy will decide about the type of Neural Network model of lower level based on some predetermined knowledge of Muscle EMG-Force relationship at different muscle activation level. In this knowledge based (or Rule Based Connectivity) network Three RBF network has been embedded separately into the network to address low, moderate and high amplitude EMG signals respectively. High level Fuzzy controller will help the ANFIS-RBF architecture switch between the RBFs based on rules to estimate the desired torque. The three RBF will have three different sets of Centroids and standard deviations. The ANFIS-RBF is able to exhibit a promising result in estimating the torque with an MSE (Mean Square Error) as low as 181.8823.
Please use this identifier to cite or link to this item: