CN109885159B - State space electromyography model construction method based on forward dynamics and Hill model - Google Patents

Abstract

The invention relates to a state space electromyography model construction method based on forward dynamics and a Hill model, which comprises the steps of firstly collecting electromyography signals of relevant muscles of a joint in a continuous motion state, carrying out band-pass filtering processing on the electromyography signals, then solving relevant muscle activation through nerve activation, substituting the relevant muscle activation into a Hill muscle model, then simplifying and substituting parameters of the Hill muscle model, combining the substituted simplified model with joint forward dynamics to obtain a prediction model in a discrete time state, finally carrying out feature extraction on root mean square and wavelet coefficients on the collected relevant electromyography signals to form a measurement equation as state feedback, and fitting the fitting equation with joint motion to obtain a final state space electromyography model. Compared with the traditional angle estimation method, the model has obvious improvement on the aspects of prediction precision, real-time performance and the like.

Description

State space electromyography model construction method based on forward dynamics and Hill model

Technical Field

The invention belongs to the field of electromyographic signal processing, relates to construction of a state space electromyographic model, and particularly relates to a state space electromyographic model based on forward dynamics and a Hill muscle model.

Background

Surface Electromyography (sEMG) is an input signal source of the modern leading edge scientific and technical man-machine interaction of the comparative hot, is a non-stable weak signal, is a group of action potential sequences generated by muscle excitation and related movement units together, has obvious characteristic distinction, abundant contained information and simple and non-invasive acquisition, and becomes the research field of the hot in the current man-machine interaction technology. The study on surface electromyographic signals has mainly focused on two processes, feature extraction and pattern recognition. The corresponding research results are also mature, and a plurality of discrete action categories can be identified. However, in the field of rehabilitation medical robots and the like, prediction of continuous motion variables of patients is more often required to realize smooth and flexible control of the rehabilitation robots.

The traditional joint continuous motion estimation method comprises the steps of extracting electromyographic features and then establishing a regression model of sEMG and continuous motion through a neural network. A physiological muscle model is also a way to estimate continuous joint motion. Buchanan et al propose a forward dynamics model based on electromyographic signals, which consists of a Hill Muscle Model (HMM), muscle activation dynamics, and joint forward dynamics. The model involves a plurality of physiological parameters, is difficult to calculate and has limited practical application. HMMs are the most commonly used muscle models to estimate continuous joint motion, but there are two problems: firstly, the HMM involves many complex physiological parameters which are difficult to identify, and the calculation load is also large; the second is that the HMM can calculate the joint moments directly from sEMG signals, but if continuous joint motion estimation is required, the motion states also need to be calculated from the moments. This typically results in cumulative errors that reduce the prediction accuracy.

While the above problem can be effectively solved by a method combining HMM with joint forward dynamics and simplified substitution of model parameters, which does not require calculation of joint moments but can calculate joint movements directly from sEMG signals. Meanwhile, the electromyographic features are used for forming a measurement equation to serve as feedback, and a closed-loop prediction algorithm is used, so that the continuous motion of the joint can be accurately estimated.

Disclosure of Invention

The invention relates to a state space electromyography model for joint angle estimation, which comprises the steps of firstly collecting electromyography signals of relevant muscles of a joint in a continuous motion state, carrying out band-pass filtering treatment on the electromyography signals, then obtaining relevant muscle activation a (k) through nerve activation u (k), substituting the relevant muscle activation a (k) into a Hill muscle model, then simplifying and substituting parameters of the Hill muscle model, then combining the simplified model after substitution with joint forward dynamics to obtain a prediction model in a discrete time state, and finally carrying out prediction on the collected relevant electromyography signalsLine root mean square X_rmsSum wavelet coefficient alpha_j,kAnd (3) extracting the characteristics, forming a measurement equation as state feedback, and fitting the measurement equation with joint motion through a fitting equation to obtain a final state space electromyographic model. Compared with the traditional angle estimation method, the model has obvious improvement on the aspects of prediction precision, real-time performance and the like.

In order to achieve the above object, the method of the present invention mainly comprises the following steps:

step one, constructing a simplified Hill muscle model.

1) Muscle activation was first determined from nerve activation:

a (k) is muscle activation, u (k) is nerve activation, k is time, and A is a nonlinear shape factor defining the curvature of the function.

2) Parameter substitution of the hill muscle model:

F^mtis tendon force, f_A(l) Is a normalized main power-length relationship, f_V(v) Is a normalized force-velocity relationship, f_P(l) Is a normalized passive force-length relationship,

is the maximum isometric force and phi is the pinnate angle.

3) The joint moment expression is as follows:

τ＝F^mt·r

τ is the moment and r is the moment arm.

And step two, constructing a Hill muscle model and a forward dynamics combination model.

1) Knee angular acceleration is calculated as follows:

I_eis the moment of inertia of the lower leg, τ_egIs the additional moment and the gravitational moment, tau_egmIs its maximum value.

2) The acceleration after the parameter substitution and model combination is calculated as follows:

s_iare instead parameters, all of which are constants.

3) The discrete-time prediction model is as follows:

T_sis the time of the sampling, and,

is the angular velocity of the joint, θ_kIs the joint position.

Thirdly, extracting the features of the electromyographic signals after the filtering processing to extract root mean square X_rmsSum wavelet coefficient alpha_j,kThe composition measurement equation is used as state feedback. The electromyographic features are then fitted to the joint movement as follows.

The value of u is 1 and 2,

is a fixed parameter that is identified off-line,

and

are the root mean square and wavelet coefficients of time k.

Finally obtaining a nonlinear expression of the state space electromyography model:

wherein

a_k＝a(k)，ω_kIs process noise, upsilon_kIs the measurement noise.

The state space electromyography model based on the forward dynamics and the Hill muscle model has the following characteristics:

the myoelectric state space model for joint angle estimation, which is established by the invention, combines forward dynamics with a Hill muscle model, simplifies parameters, enables the model to be directly used for knee joint motion estimation, and reduces accumulated errors. Meanwhile, electromyographic features such as root mean square and wavelet coefficients are extracted, a measurement equation is established, system errors and external interference are reduced, and the joint prediction precision is improved.

Drawings

FIG. 1 is a flow chart of an embodiment of the present invention;

FIG. 2 is a graph of electromyographic signal characteristics and real-time angles collected by the present invention;

Detailed Description

As shown in fig. 1, the present embodiment includes the following steps:

step one, establishing a simplified Hill muscle model.

1) Muscle activation was first determined from nerve activation:

a (k) is muscle activation, u (k) is nerve activation, k is time, and A is a nonlinear shape factor defining the curvature of the function.

2) Parameter substitution of the hill muscle model:

F^mtis tendon force, f_A(l) Is a normalized main power-length relationship, f_V(v) Is a normalized force-velocity relationship, f_P(l) Is a normalized passive force-length relationship, F₀ ^mIs the maximum isometric force and phi is the pinnate angle.

3) The joint moment expression is as follows:

τ＝F^mt·r

τ is the moment and r is the moment arm.

And step two, combining the forward dynamics with the Hill muscle model.

1) Knee angular acceleration is calculated as follows:

I_eis the moment of inertia of the lower leg, τ_egIs the additional moment and the gravitational moment, tau_egmIs its maximum value.

2) The acceleration after the parameter substitution and model combination is calculated as follows:

s_iare instead parameters, all of which are constants.

3) The discrete-time prediction model is as follows:

T_sis the time of the sampling, and,

is the angular velocity of the joint, θ_kIs the joint position.

And step three, extracting the characteristics of the filtered electromyographic signals, and extracting a root mean square coefficient and a wavelet coefficient to form a measurement equation as state feedback.

1) The root mean square is calculated as follows:

X_rmsis the root mean square value, x, of the data₁，x₂And x_nIs the processed data, and n is 16.

2) We extract the j-level wavelet coefficients using discrete wavelet transform:

f(t)＝A_J+∑_j≤JD_j

D_jare detail values:

D_j＝∑_k∈Zα_j,kφ_j,t(t)

wherein Z is a set of positive integers, α_j,kIs a wavelet coefficient, phi_j,t(t) is the mother wavelet function.

A_JAre approximations:

A_J＝∑_j＞JD_j

the features and the real angles of the collected electromyographic signals are shown in fig. 2.

3) The fitting equation relating electromyographic features to joint motion is as follows:

the value of u is 1 and 2,

is a fixed parameter that is identified off-line,

and

are the root mean square and wavelet coefficients of time k.

The parameter identification is shown in table 1:

TABLE 1 parameters with and without load

The nonlinear expression of the state space electromyography model obtained finally is as follows:

wherein

a_k＝a(k)，ω_kIs process noise, upsilon_kIs the measurement noise.

Claims (1)

1. The state space electromyography model construction method based on forward dynamics and a Hill model is characterized by comprising the following steps of:

step one, constructing a simplified Hill muscle model;

1) muscle activation was first determined from nerve activation:

a (k) is muscle activation, u (k) is nerve activation, k is time, a is a nonlinear shape factor defining a curvature of the function;

2) parameter substitution of the hill muscle model:

F^mtis tendon force, f_A(l) Is a normalized primary power-tendon length relationship, f_V(v) Is a normalized force-velocity relationship, f_P(l) Is a normalized passive force-tendon length relationship,

is the maximum isometric force, f is the pinnate angle;

3) the joint moment expression is as follows:

τ＝F^mt·r

τ is the moment, r is the moment arm;

step two, constructing a Hill muscle model and a forward dynamics combination model;

1) knee angular acceleration is calculated as follows:

τ_eg＝τ_egm·sin(θ)

I_eis the moment of inertia of the lower leg, τ_egIs the sum of the additional moment and the gravitational moment, τ_egmIs τ_egθ is the knee angle;

2) acceleration after parametric substitution and model combination

The calculation is as follows:

s_iare substitution parameters, all constants;

3) the discrete-time prediction model is as follows:

T_sis the time of the sampling, and,

is the angular velocity of the joint at time k, θ_kIs the joint position at time k;

thirdly, extracting the characteristics of the filtered electromyographic signals, and extracting a measurement equation consisting of a root mean square coefficient and a wavelet coefficient to serve as state feedback; then fitting the electromyographic characteristics with joint movement according to the following formula;

the value of u is 1 and 2,

is a fixed parameter that is identified off-line,

and

is the root mean square and wavelet coefficients of time k;

finally obtaining a nonlinear expression of the state space electromyography model:

wherein

a_k＝a(k)，ω_kIs process noise, upsilon_kIs the measurement noise.

Images