A method for identifying inertial parameters of a human lower limb model based on an exoskeleton robot

Abstract

This invention discloses a method for identifying inertial parameters of a human lower limb model based on an exoskeleton robot. First, a dynamic model of a single-leg hip-knee double joint coupled with the human lower limb is established. Unknown parameters are linearly combined and written into the system's motion equations based on least squares identification. Then, an excitation trajectory is input, and a swinging experiment is performed only on the exoskeleton robot to construct an exoskeleton dataset. After completing the exoskeleton experiment, experiments are conducted on subjects with the same configuration to obtain a human-machine coupled dataset. Offline least squares methods are used to estimate the parameters of the exoskeleton and the human lower limb. This method differs from the general data of existing typical anthropometric models. The identified inertial parameters of the human lower limb model are closer to the characteristics of different users, effectively improving the accuracy of measurement and inertial characteristics of user lower limb segments. Furthermore, by correctly calculating the passive torque, the user's muscle torque is more realistically quantified, playing a crucial role in the research of active training modes for rehabilitation exoskeleton robots.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of rehabilitation exoskeleton robots, and particularly relates to a method for identifying inertial parameters of a human lower limb model based on an exoskeleton robot. BACKGROUND

[0002] According to the investigation of the World Health Organization (WHO), about 1 billion (15% of the total population) people in the world suffer from some kind of disability, and about 200 million people suffer from functional disorders. In addition, some of the work of people today requires stronger and more persistent muscle movement, but when people engage in high-intensity physical work for a long time, the human body will appear muscle fatigue. In order to solve the above problems, some scientific researchers have invented a wearable device, that is, a lower limb exoskeleton robot. The lower limb exoskeleton is used to reduce the burden of the human body in the civil field, and is used as an auxiliary rehabilitation device for patients to carry out rehabilitation training in the medical rehabilitation field. The lower limb exoskeleton integrates technologies such as sensing and control, and has the characteristics of bionics, robotics, information and control science, medicine and other multidisciplinary fields. Since the goal of rehabilitation training is to restore the movement ability of patients to the normal level, the study of human gait is the basis of the research of lower limb exoskeleton rehabilitation robots. However, most of the initially developed rehabilitation exoskeleton robots only provide passive mode training, that is, moving the lower limbs of the user along a predefined fixed trajectory. In recent years, many researchers believe that active mode can achieve more effective and optimal training. In this training mode, the quantification of the muscle strength of the user is very important for the adaptive behavior of the robot and for informing the user of their contribution to the training.

[0003] In rehabilitation robots, there are two widely used methods for quantifying the muscle strength of the user. One is to measure electromyography (EMG) by using surface electrodes attached to the skin of the user, and the other is to estimate muscle torque based on inverse dynamics analysis. By measuring the external torque applied at each joint of the exoskeleton robot and removing the inertia, Coriolis and gravity torques (referred to as passive torques) of the lower limbs of the user, the muscle torque of the human user can be estimated. Among them, the calculation of passive torques requires accurate estimation of the anthropometric and inertial characteristics of the lower limb segments, such as mass, center of mass position and moment of inertia, which are usually referred to as body segment inertial parameters (BSIPs). In the inverse dynamics analysis of human motion, BSIPs are usually estimated according to the human body measurement model. As can be seen, the separation of active muscle strength from joint torque is heavily dependent on the accuracy of the dynamic model of the lower limbs of the user, so the method for identifying the specific inertial parameters of the user is important. SUMMARY

[0004] To solve the above technical problems, the application provides a method for identifying inertia parameters of a human lower limb model based on an exoskeleton robot.

[0005] The technical scheme adopted by the application is as follows:

[0006] S1, a dynamic model of a single-leg hip-knee double-joint coupled by an exoskeleton robot and a human lower limb is established, and a linear combination of unknown parameters is written into a system motion equation based on least square identification;

[0007] S2, an excitation trajectory designed by particle swarm optimization (PSO) is input, a swing experiment is performed on the exoskeleton robot, i.e., the exoskeleton not worn by a human subject, and an exoskeleton data set is constructed;

[0008] S3, after the experiment of the exoskeleton is completed, torque of a double-joint of a lower limb of a healthy human body in a single-leg swing and real-time angle, angular velocity and angular acceleration information of the healthy human body in different body shapes are collected by sensors of the exoskeleton robot in the same configuration, so as to obtain a human-machine coupling data set;

[0009] S4, the data sets obtained in steps S2 and S3 and the motion equation in step S1 are used to estimate parameters of the exoskeleton and the human lower limb by using an offline least square method.

[0010] Further, the step S1 is specifically as follows:

[0011] The motion equation of the human lower limb model is expressed as:

[0012]

[0013] wherein, respectively represent joint angle, angular velocity and angular acceleration, represents a symmetric positive definite inertia matrix of the human lower limb, represents a centrifugal moment and Coriolis moment vector of the human lower limb, represents a gravity moment vector of the human lower limb, represents a human muscle moment, represents a moment from an external environment, represents a whole real number set.

[0014] The moment from the external environment can be expressed by subtracting a moment required for moving the exoskeleton from a moment generated by an exoskeleton robot actuator:

[0015]

[0016] wherein, represents the actuator torque vector applied at the joint, represents the exoskeleton friction torque vector, represents the symmetric positive definite inertia matrix of the exoskeleton, represents the centrifugal and Coriolis torque vector of the exoskeleton, represents the gravity torque vector of the exoskeleton. The motion equation of the human-exoskeleton system is represented as:

[0017]

[0018] wherein, M HR = M H + M R , C HR = C H + C R , G HR = G H + G R , K, C represent the friction coefficient.

[0019] Further, in the step S1, the double joint is specifically the hip joint and the knee joint.

[0020] Further, the step S3 is specifically as follows:

[0021] For a two-segment model of the human lower limb wearing an exoskeleton robot, each segment is defined by four parameters: length l, mass m, distance c of the joint mass center to the segment mass center position, and inertia J.

[0022] M HR , C HR , G HR are represented by the above four segment inertia parameters, and X1, X2, X3, X4 defined by linear combination of inertia parameters are written as unknown parameters identified by the least square method:

[0023]

[0024]

[0025] X3 = m2c2

[0026] X4 = m1c1 + m2l1

[0027] wherein m1 represents the mass of the thigh segment, m2 represents the mass of the shank segment, c1 represents the distance from the hip joint to the center of mass of the thigh segment, c2 represents the distance from the knee joint to the center of mass of the shank segment, l1 represents the length from the hip joint to the thigh segment, J1 represents the moment of inertia of the thigh segment, and J2 represents the moment of inertia of the shank segment.

[0028] The friction torque of the exoskeleton robot includes Coulomb dry friction and viscous friction, and X5, X6, X7 and X8 defined by friction coefficients K, C are written as unknown parameters to be identified by the least square method:

[0029] X5=K1

[0030] X6=C1

[0031] X7=K2

[0032] X8=C2

[0033] wherein K1 and C1 represent the Coulomb dry friction coefficient and the viscous friction coefficient of the thigh segment, respectively, and K2 and C2 represent the Coulomb dry friction coefficient and the viscous friction coefficient of the shank segment, respectively.

[0034] Further, the step S4 is specifically as follows:

[0035] The exoskeleton parameters X are estimated using the offline least square method R1 R2 R3 R4 R5 R6 R7 R8 and the parameters X of the lower limbs of the human body H1 H2 H3 H4 (the human body has no friction torque), and the estimation process is divided into two parts: (1) exoskeleton parameter identification; and (2) human lower limb parameter identification assuming that the exoskeleton parameters are known:

[0036] (1) exoskeleton swing experiment alone, the exoskeleton parameters are estimated, and the relationship between the torque τ and the joint angle θ is represented as:

[0037]

[0038]

[0039] wherein τ1 represents the hip joint torque of the exoskeleton robot, τ2 represents the knee joint torque of the exoskeleton robot, θ1, respectively represent the hip joint angle, angular velocity and angular acceleration of the exoskeleton robot, and θ2, ​​​​​​​​​​respectively represent the knee joint angle, angular velocity and angular acceleration of the exoskeleton robot, and g represents the gravity acceleration, which is 9.8.

[0040] (2) The human body wears an exoskeleton to conduct experiments, estimate the human lower limb parameters, and the relationship between the moment τ and the joint angle θ is represented as:

[0041]

[0042]

[0043] wherein τ1 represents the human hip joint moment, τ2 represents the human knee joint moment, θ1, respectively represent the human hip joint angle, angular velocity and angular acceleration, θ2, respectively represent the human knee joint angle, angular velocity and angular acceleration. g represents the gravity acceleration, which is 9.8.

[0044] The above equation is written in linear form:

[0045]

[0046] wherein, represents the input moment, represents the regression vector as a function of geometry and θ, φ j ∈R 3 represents the vector of segment parameters, and T represents the transpose of the matrix.

[0047] The current estimated parameter vector is defined as The corresponding estimated moment vector is

[0048]

[0049] wherein, represents the regression vector matrix of the current estimation.

[0050] The mean square error (MSE) is used as the fitness function of parameter identification, that is, the moment estimation error is to judge the pros and cons of the current estimated parameter vector . Wherein, N represents the number of sampling points.

[0051] ​The method of the present application firstly establishes a single-leg hip-knee double-joint dynamic model of the exoskeleton human-robot coupled with the lower limbs of the human body, writes a linear combination of unknown parameters into a system motion equation based on least square identification, inputs an excitation trajectory, only performs a swing experiment on the exoskeleton robot, constructs an exoskeleton data set, and after completing the experiment on the exoskeleton, performs an experiment on the subject with the same configuration, acquires a human-machine coupling data set, and estimates the parameters of the exoskeleton and the lower limbs of the human body using an offline least square method. The method of the present application is different from the general data of the existing typical human body measurement model, the identified inertia parameters of the lower limbs of the human body are closer to the characteristics of different users, effectively improve the measurement and inertia characteristic accuracy of the lower limb segments of the user, and then correctly calculate the passive torque to more truly quantify the muscle torque of the user, which plays a key role in the active training mode research of the rehabilitation exoskeleton robot. BRIEF DESCRIPTION OF DRAWINGS

[0052] Figure 1 A flowchart of a method for identifying inertia parameters of a human lower limb model based on an exoskeleton robot according to the present application.

[0053] Figure 2 A human body model structure diagram in an embodiment of the present application. DETAILED DESCRIPTION

[0054] The method of the present application will be further described below in combination with the drawings and embodiments.

[0055] In this embodiment, the object of the data acquisition experiment is a healthy adult with different body types (height, weight, leg length, body fat rate, etc.), and during the experimental test process, the subject is in a standing position, and the whole leg is allowed to swing, and the knee is in a natural position. In each experiment, the subject is required to completely relax his leg, and the angles of the hip joint and the knee joint are applied by the exoskeleton to allow the human leg to be passively moved under the action of the torque applied by the exoskeleton. Each experiment is performed five times, and lasts for 40 seconds.

[0056] As shown in Figure 1 , a flowchart of a method for identifying inertia parameters of a human lower limb model based on an exoskeleton robot according to the present application, and the specific steps are as follows:

[0057] S1, a single-leg hip-knee double-joint dynamic model of the exoskeleton human-robot coupled with the lower limbs of the human body is established, and a linear combination of unknown parameters is written into a system motion equation based on least square identification;

[0058] S2, an excitation trajectory optimized by a particle swarm optimization (PSO) is inputted, and only the exoskeleton robot, i.e., the exoskeleton not worn by the human subject, is subjected to a swing experiment, and an exoskeleton data set is constructed;

[0059] S3, after completing the experiment of the exoskeleton, collecting the moment of force, real-time angle, angular velocity and angular acceleration information of the lower limbs of different body types of healthy people under single leg swing through the sensor of the exoskeleton robot in the same configuration, so as to obtain the human-machine coupling data set;

[0060] S4, according to the data set obtained in steps S2 and S3 and the motion equation in step S1, and using the off-line least square method to estimate the parameters of the exoskeleton and the lower limbs of the human body.

[0061] In the embodiment, the step S1 is specifically as follows:

[0062] As shown in the formula (1), for the lower limbs of the human body wearing the exoskeleton robot, a two-segment model in the sagittal plane is considered. Figure 2

[0063] Wherein, the human leg is composed of rigid limb segments, and the lower leg and foot are regarded as a rigid segment, and each segment is connected through a fixed hinge. The human leg is rigidly connected with the exoskeleton robot, the two systems have the same kinematics, and the model only considers the motion in the sagittal plane. The model is composed of two rigid segments (thigh and lower leg) and two pin joints (hip joint and knee joint), and each segment is defined by five parameters: length l, mass m, distance c of joint center of mass to segment center of mass, and inertia moment J. θ represents the joint angle, and the counterclockwise is defined as positive. Segment 1 represents the thigh and hip joint, and segment 2 represents the lower leg and knee joint.

[0064] Wherein, s represents the distance from the binding belt between the exoskeleton robot and the human body to the joint center of mass. In the figure, the subscript h represents the human body, r represents the exoskeleton robot, 1 represents the thigh segment, and 2 represents the lower leg segment.

[0065] The motion equation of the human lower limb model is represented as:

[0066]

[0067] Wherein, θ, θ, and θ represent the joint angle, angular velocity and angular acceleration respectively, M represents the symmetric positive definite inertia matrix of the human lower limb, C represents the centrifugal moment and Coriolis moment vector of the human lower limb, G represents the gravity moment vector of the human lower limb, M represents the muscle moment of the human body, F represents the moment of force from the external environment, R represents the set of all real numbers.

[0068] The moment of force from the external environment can be represented by subtracting the moment required to move the exoskeleton from the moment generated by the exoskeleton robot actuator:

[0069] ​

[0070] in, This represents the actuator torque vector applied at the joint. This represents the exoskeleton friction torque vector. The symmetric positive definite inertia matrix representing the exoskeleton. This represents the centrifugal torque and Coriolis torque vector of the exoskeleton. Let represent the gravitational torque vector of the exoskeleton. Then the equation of motion for the human exoskeleton system can be expressed as:

[0071]

[0072] Among them, M HR =M H +M R C HR =C H +C R G HR =G H +G R , K and C represent the coefficients of friction.

[0073] In this embodiment, step S2 is specifically as follows:

[0074] Generally, the sampled data of the object to be identified contains a large amount of noise. Furthermore, due to the poor anti-interference capability of differential signals, offline differential analysis of joint angles to obtain joint angular velocity and angular acceleration may amplify these existing noise signals. To suppress the sensitivity of the identification results to noise signals and ensure the accuracy of the final identification results, an optimal excitation trajectory is designed for the sampling experiment before the data acquisition experiment.

[0075] Since Fourier series can theoretically fit any function, this property can be used to set weight parameters to optimize and design the excitation trajectory, i.e.:

[0076]

[0077]

[0078]

[0079] Where, θ d Indicates the angle of the excitation trajectory. and Let i = 1, 2, k = 0, ..., n, t ∈ [0, t] ... s ], t sdenotes the experimental duration, ω denotes the fundamental frequency, k denotes the sampling parameter, θ i,0 denotes the initial bias angle of the exoskeleton double joint, a i,k ,b i,k denotes the to-be-optimized parameter.

[0080] The condition number of the current estimated regression matrix is selected as the fitness function of the excitation trajectory, and PSO is used to optimize and design the excitation trajectory meeting the constraints.

[0081] In addition, in order to reduce the influence of noise, a zero-phase digital low-pass filter is designed to filter the sampling data (joint angle).

[0082] In the embodiment, the step S3 is specifically as follows:

[0083] M HR ,C HR ,G HR All are characterized by the four segment inertia parameters in the step S1, and X1, X2, X3 and X4 defined by linear combination of the inertia parameters are written as unknown parameters to be identified by the least square method:

[0084]

[0085]

[0086] X3=m2c2

[0087] X4=m1c1+m2l1

[0088] Wherein, m1 denotes the mass of the thigh segment, m2 denotes the mass of the lower leg segment, c1 denotes the distance from the hip joint to the mass center position of the thigh segment, c2 denotes the distance from the knee joint to the mass center position of the lower leg segment, l1 denotes the length from the hip joint to the thigh segment, J1 denotes the moment of inertia of the thigh segment, and J2 denotes the moment of inertia of the lower leg segment.

[0089] The friction torque of the exoskeleton robot includes Coulomb dry friction and viscous friction, and X5, X6, X7 and X8 defined by the friction coefficients K and C are written as unknown parameters to be identified by the least square method:

[0090] X5=K1

[0091] X6=C1

[0092] X7=K2

[0093] X8=C2

[0094] where K1, C1 represent the Coulomb dry friction and viscous friction coefficients of the thigh segment, K2, C2 represent the Coulomb dry friction and viscous friction coefficients of the shank segment, the unit of angle is radian, and the unit of torque is Nm.

[0095] In the present embodiment, the step S4 is specifically as follows:

[0096] Estimating the exoskeleton parameters X using offline least squares method R1 ,X R2 ,X R3 ,X R4 ,X R5 ,X R6 ,X R7 ,X R8 and the parameters X of the human lower limbs H1 ,X H2 ,X H3 ,X H4 (the human body has no friction torque). The estimation process is divided into two parts: (1) exoskeleton parameter identification; (2) assuming that the exoskeleton parameters are known, the human lower limb parameter identification is carried out:

[0097] (1) Exoskeleton swing experiment alone, estimate the exoskeleton parameters, the relationship between torque τ and joint angle θ is expressed as:

[0098]

[0099]

[0100] where τ1 represents the hip joint torque of the exoskeleton robot, τ2 represents the knee joint torque of the exoskeleton robot. θ1, respectively represent the hip joint angle, angular velocity and angular acceleration of the exoskeleton robot, θ2, respectively represent the knee joint angle, angular velocity and angular acceleration of the exoskeleton robot, and g represents the acceleration of gravity, which is 9.8.

[0101] (2) The human body wears the exoskeleton to carry out the experiment, estimate the human lower limb parameters, the relationship between torque τ and joint angle θ is expressed as:

[0102]

[0103]

[0104] where τ1 represents the hip joint torque of the human body, τ2 represents the knee joint torque of the human body. θ1, respectively represent the hip joint angle, angular velocity and angular acceleration of the human body, θ2, respectively represent the knee joint angle, angular velocity and angular acceleration of the human body. g represents the acceleration of gravity, which is 9.8.

[0105] The above equation is written in linear form:

[0106]

[0107] where, represents the input torque, represents the regression vector as a function of geometry and θ, φ j ∈R 3 represents the vector of segment parameters, T represents the transpose of the matrix.

[0108] Let the current estimated parameter vector be the corresponding estimated torque vector is

[0109]

[0110] where, represents the regression vector matrix of the current estimate.

[0111] The mean square error (MSE) is used as the fitness function for parameter identification, i.e. the torque estimation error is to judge the pros and cons of the current estimated parameter vector . Where N represents the number of sampling points.

[0112] In summary, the parameters obtained from actual users through experiments are much more accurate than the widely used anthropometric model parameters, thereby effectively improving the accuracy of the user's active muscle torque quantification of the exoskeleton rehabilitation robot. The method of the present application is not limited to a 2-DOF lower limb exoskeleton, and can be extended to an exoskeleton with multiple degrees of freedom; but is only effective when the user moves in the air without ground contact.

[0113] Those skilled in the art will appreciate that the embodiments described herein are presented for the purpose of helping the reader to understand the principles of the present application, and should be understood as not limiting the scope of protection of the present application to such specific statements and embodiments. The present application can have various modifications and changes for those skilled in the art. Any modification, equivalent replacement, improvement, etc. made within the spirit and principles of the present application should be included in the scope of claims of the present application.

Claims

1. A method for identifying inertia parameters of a human lower limb model based on an exoskeleton robot, comprising the following specific steps: S1, establishing a single-leg hip-knee double-joint dynamic model coupled with the exoskeleton robot and human lower limb, and writing the linear combination of unknown parameters into the system motion equation based on least squares identification; S2, inputting the excitation trajectory designed by particle swarm optimization (PSO) optimization, and only performing swing experiments on the exoskeleton robot, i.e. the exoskeleton not worn by the human subject, to construct an exoskeleton dataset; S3, after completing the experiment of the exoskeleton, collecting the torque of the double joints of the lower limbs of healthy people of different body types under single-leg swing and real-time angle, angular velocity and angular acceleration information through the sensors of the exoskeleton robot with the same configuration, so as to obtain a human-robot coupling dataset; S4, according to the datasets obtained in steps S2 and S3 and the motion equation in step S1, and using off-line least squares method to estimate the parameters of the exoskeleton and human lower limbs.

2. The method for identifying inertia parameters of a lower extremity model of a human body based on an exoskeleton robot according to claim 1, characterized in that, The step S1 is specifically as follows: The motion equation of the human lower limb model is expressed as: ； wherein, respectively represent joint angles, angular velocities and angular accelerations, denotes a symmetric positive definite inertia matrix of the lower limbs of the human body, denotes a centrifugal and Coriolis force moment vector of the lower limbs of the human body, denotes a gravity moment vector of the lower limbs of the human body, denotes a human muscle moment, denotes a moment from the external environment, denotes the set of all real numbers; The torque of the external environment can be expressed by subtracting the torque required for moving the exoskeleton from the torque generated by the exoskeleton robot actuator: ； wherein, represents an actuator torque vector applied at the joint, represents an exoskeleton friction torque vector, represents a symmetric positive definite inertia matrix of the exoskeleton, represents a centrifugal and Coriolis torque vector of the exoskeleton, represents a gravity torque vector of the exoskeleton; then the motion equation of the human-exoskeleton system is represented as: ； wherein , represents the coefficient of friction.

3. The method of claim 1, wherein, In the step S1, the double joints are specifically hip joint and knee joint.

4. The method of claim 2, wherein, The step S3 is specifically as follows: For the two-segment model of the human lower limb wearing the exoskeleton robot, each segment is defined by four parameters: length l, mass m, distance c of joint center of mass to segment center of mass position, and inertia moment J; All are characterized by the above four segment inertia parameters, write the linear combination defined by the inertia parameters Unknown parameters as the least square method identification: ； ； ； ； wherein, represents the mass of the thigh segment, represents the mass of the shank segment, represents the distance of the hip joint to the center of mass position of the thigh segment, represents the distance of the knee joint to the center of mass position of the shank segment, represents the length of the hip joint to the thigh segment, represents the moment of inertia of the thigh segment, represents the moment of inertia of the shank segment; The friction torque of the exoskeleton robot includes Coulomb dry friction and viscous friction, and the friction coefficient defined unknown parameters identified as the least square method: ； ； ； ； wherein, Cp and Cp represent the Coulomb dry friction and viscous friction coefficients of the thigh segment, respectively, Cp and Cp represent the Coulomb dry friction and viscous friction coefficients of the calf segment, respectively.

5. The method for identifying inertia parameters of a lower extremity model of a human body based on an exoskeleton robot according to claim 4, characterized in that, The step S4 is specifically as follows: Estimating exoskeleton parameters using offline least squares and parameters of the lower limbs of the human body Without the friction torque of the human body, the estimation process is divided into two parts: (1) exoskeleton parameter identification; (2) assuming that the exoskeleton parameters are known, the parameters of the lower limbs of the human body are identified: (1) Experiments of the exoskeleton swinging alone to estimate the exoskeleton parameters, the moment of force and the relationship between the joint angle is expressed as: ； ； wherein, represents a hip joint torque of the exoskeleton robot, represents a knee joint torque of the exoskeleton robot; respectively represent a hip joint angle, angular velocity and angular acceleration of the exoskeleton robot, respectively represent a knee joint angle, angular velocity and angular acceleration of the exoskeleton robot, represents a gravitational acceleration, and takes a value of 9.8; (2) The human body wears an exoskeleton to conduct experiments, estimates the parameters of the lower limbs of the human body, and the relationship between the moment and joint angle is expressed as: ； ； wherein, represents the moment of the hip joint of the human body, represents the moment of the knee joint of the human body; respectively represent the angle, angular velocity and angular acceleration of the hip joint of the human body, respectively represent the angle, angular velocity and angular acceleration of the knee joint of the human body; represents the acceleration of gravity, and takes the value of 9.8; The above equation is written in linear form: ； wherein, represents an input torque, represents a regression vector as a function of geometry and function, represents a vector of segment parameters, represents the transpose of a matrix; Let the current estimate of the parameter vector be The corresponding estimate of the moment vector is ； wherein, represents the currently estimated regression vector matrix; Adopting mean square error (MSE) as the fitness function of parameter identification, i.e. the moment estimation error is , to judge the good or bad degree of the current estimated parameter vector ; wherein N represents the number of sampling points.

Images