A mathematical model for mapping EMG signal to joint torque for the human elbow joint using nonlinear regression

Abstract

Numerous researchers have investigated the relationship between EMG and joint torque. Most of these studies use some conventional filtering (i.e. rectification followed by low pass filtering) to estimate the electromyogram (EMG) amplitude and then relate it to the joint torque. Currently some advanced pre-processing techniques (i.e. signal whitening) are also used to estimate the EMG amplitude and then relate it to joint torque. In this study we apply some pre-processing techniques like DC offset removal, noise filtering followed by rectification and then we calculate the moving average of the EMG signal. Thus we get a linear envelope (muscle activation) of the EMG signal and use that linear envelope to estimate the joint torque. To map the EMG to joint torque we propose a new mathematical model. This model has some unknown adj ustable parameters, and the values of these parameters are obtained using nonlinear regression. Five subjects took part in the experiments. They were asked to perform non-fatiguing and variable force maximal voluntary contractions (MVC) and submaximal voluntary contractions (SMVC), and the resulting elbow joint torque and EMG signals were recorded. This recorded data was entered to the model, to estimate best fit values for the unknown parameters. Once these values of the parameters were obtained they were put into the model and thus joint torque was estimated. Predictions made by our model are well correlated with experimental data in both MVC and SMVC, the correlation coefficient and mean square error obtained for experimental data during MVC are 0.998 and 0.056Nm respectively. The results of this new model were compared with other existing models and some new models and it was found that our model has greater correlation and least mean square error with experimental data. This model may be helpful in the control systems for recognition systems, robot manipulators, exoskeletons, EMG prosthesis and electric stimulators.

Forward estimation of joint torque from EMG signal through muscle synergy combinations

“…By following the estimation approaches [4], [8], the ankle joint torque is estimated via solving the pseudoinverse U pos vel † of activation-position-velocity matrix U pos vel as follows:…”

“…As the reflection of muscle activation, EMG has been found to be of relatively high correlation with movements of associated joints [1]. Researchers have investigated and addressed the identification of the relationship between joint movement and EMG (muscle activations) [2]- [8]. For example, Clancy et al proposed the method for estimating the elbow joint torque through EMG signals of multiple channels [4].…”

Competitive Learning in a Spiking Neural Network: Towards an Intelligent Pattern Classifier

“…ANN-based technics are used in a vast range of EMG classification tasks (see, e.g., [3][4][5]). In turn, regression allows one to estimate muscle effort strength by its EMG signal and, hence, can be used for proportional (gradual) control, e.g., for reconstruction of torque value of some joints [6]. In addition to other mathematical tools, ANNs were also successfully applied to the regression problem, including multichannel registration [7][8][9].…”

What ICA provides for ECG signal extraction from contaminated ECG observations without using differential amplifiers

“…It is also possible that the noise signal at different detection sites may be out of phase, and hence again they are not common mode so will still remain as a noise in the recorded ECG signal. There are some other sources of noise; power line interference, electrode contact noise, patient-electrode motion artifacts, Electromyogram (EMG) interference and baseline wandering [2]. Early researchers have used either time domain or spatial domain analysis to remove these artifacts [3], [4].…”