A Research On The SEMG-force Model Based On LVN In Fatigue State

Force estimation and fatigue analysis have extensive research and great value in the fields of sports,rehabilitation medicine and human-computer interaction.Existing force estimation models rarely achieve continuous force estimation under different fatigue stages,and generally avoid muscle fatigue through adequate rest.However,muscle fatigue is unavoidable,and will lead to dynamic changes in the nonlinear relationship between s EMG and force,which apparently reduces the estimation accuracy of the force estimation model.In view of the above problems,this paper actively considered muscle fatigue in force estimation research,analyzed the changes of s EMG characteristics with the increase of fatigue degree,and designed a continuous force estimation model adaptable to various fatigue levels.The main research contents are as follows:1.The fatigue features of s EMG signal were extracted and analyzed.Firstly,a constant-force pulse fatigue experiment of brachioradialis including different fatigue degrees was designed.Based on this experiment,19 s EMG features involving 11 in the time domain,6 in the frequency domain,and 2 non-Gaussian features were extracted.The change trend of 19 characteristics with fatigue was qualitatively analyzed,and the coefficient R~2,Pearson coefficient and nonlinear sensitivity SVR of each characteristic were calculated.Based on the results of qualitative and quantitative analysis,11 s EMG features with good effect on characterizing different fatigue levels were finally selected for the establishment of s EMG-force model.2.A novel s EMG-force model based on LVN(Laguerre Volterra Network)was designed and implemented.In this paper,LVN is introduced into the field of force estimation for the first time,and the classical polynomial fitting(POL),advanced fast orthogonal search(FOS),parallel cascade identification(PCI)and Laguerre(LET)models were implemented.The LVN and LET models were compared and analyzed,and the results showed that the accuracy of the LET decreased significantly under the influence of fatigue changes,while the LVN model was less affected by the muscle fatigue changes so that the dynamic changes of the s EMG-force relationship under different fatigue stages was better captured and higher accuracy of force estimation was achieved.3.A partly sparse LVN model(LVN?p S)fused with multiple s EMG fatigue features was designed and implemented.Since the problem of decreased force estimation accuracy based on the single-input LVN model during muscle fatigue,a novel dual-input LVN s EMG-force model was proposed.The model fuses 11 features as the second input and estimates force together with the mean standard deviations of s EMG signal.The continuous domain ant colony algorithm with global update is used to train model parameters,and the weights of each s EMG fatigue feature are trained at the same time.Additionally,a partly sparse penalty term was introduced to penalize the weights of 9features except MDF and MPF,so that the guiding weight of each fatigue feature to the model can be adaptively adjusted according to individual differences.The raising force fatigue experiment of brachioradialis and palmaris longus were designed,and LVN?p S proposed in this paper was compared with POL,FOS,PCI and LET models.The results showed that LVN?p S model can not only achieve higher-precision force estimation in different stages of in continuously increasing fatigue level,but also adapt to different muscles and different force patterns.The model proposed in this paper can capture the s EMG-force relationship of fatigue effects more acutely.and introduced a new method for the field of force estimation and provided a way to consider the effect of muscle fatigue.