Multi-index optimization-based multi-mode on-demand auxiliary intelligent control method for upper limb exoskeleton

Abstract

The invention discloses an upper limb exoskeleton on-demand auxiliary intelligent control method based on multi-index optimization and multi-mode, and the method employs an inner and outer ring force position mixed control structure based on an admittance model, an outer ring combines the three indexes of a task tracking error, patient muscle force and a rehabilitation training task period, the exercise ability of a patient is evaluated on the basis of a multi-index energy function, dynamic switching among a passive mode, an auxiliary mode and an anti-resistance mode of the upper limb exoskeleton co-fusion system is achieved, online self-adaptive adjustment of the auxiliary force level and the rehabilitation training track is achieved on the basis of fuzzy logic, and the method can adapt to rehabilitation patients with different exercise abilities. And the inner ring carries out trajectory tracking control on the ideal tracking trajectory position of the upper limb exoskeleton system obtained by the admittance model through a model-barrier-free Lyapunov control method based on quasi-time delay estimation. A simulation result verifies that the method can effectively improve the active participation degree of the patient in the rehabilitation training process, and shows good on-demand auxiliary ability and application prospects.

Description

Technical Field

[0001] The present invention belongs to the field of upper limb rehabilitation training exoskeleton systems, and in particular to an upper limb exoskeleton based on multi-index optimization and multi-mode on-demand auxiliary intelligent control method. Background Art

[0002] As an emerging technology, upper limb rehabilitation exoskeleton systems can provide patients with precise, stable, and safe movement assistance. However, patients with varying stroke severity may have different rehabilitation needs. Therefore, providing appropriate assistive force (resistance) to patients with varying motor abilities to enhance their participation in rehabilitation training has always been a key research issue in the field of rehabilitation exoskeletons.

[0003] To address the above issues, the paper (Pehlivan AU, Losey DP, O′Malley M K. Minimal assist-as-needed controller for upper limb robotic rehabilitation [J]. IEEE Transactions on Robotics, 2015, 32(1): 113-124.) proposed an on-demand assist control method based on task tracking error. By collecting the error between the patient's actual motion trajectory and the expected trajectory in real time, the stiffness coefficient of the impedance model is dynamically adjusted to achieve adaptive regulation of the assist force. However, this method only relies on the task error as the basis for adjusting the assist level, and fails to fully consider the changes in the patient's actual motor ability. In actual rehabilitation training, even if the patient has strong motor ability and high active force output, a large task tracking error may occur due to premature arrival at the target position or improper force output direction. The system may mistakenly judge the patient as having weak motor ability and blindly increase the assist level. This may not only reduce the personalized effect of training, but also weaken the patient's willingness to actively participate, affecting the efficiency of rehabilitation.

[0004] For example, the paper (Wu Q, Wang X, Chen B, et al. Development of aminimal-intervention-based admittance control strategy for upper extremity rehabilitation exoskeleton [J]. IEEE Transactions on Systems, Man, and Cybernetics: Systems, 2017, 48(6): 1005-1016.) proposes an on-demand assistance control method based on a multi-mode force field, which adaptively adjusts the assistance mode by monitoring the task tracking error in real time. However, the above control method uses a fixed force field boundary, which can easily cause patients to frequently switch between multiple modes when their motor ability is low, and can easily cause patients to slack off during rehabilitation training when their motor ability is strong. Summary of the Invention

[0005] The purpose of the present invention is to provide an upper limb exoskeleton based on multi-indicator optimization and multi-mode on-demand assistance intelligent control method, which evaluates the patient's motor ability by combining multiple motion indicators, and switches between assistance mode, resistance mode and passive mode according to the patient's motor ability, adaptively adjusting the assistance force level and rehabilitation training trajectory provided by the system.

[0006] The technical solutions for achieving the purpose of the present invention are:

[0007] An upper limb exoskeleton based on a multi-index optimized multi-mode on-demand assistance intelligent control method, using an inner and outer loop force-position hybrid control structure, can achieve switching between different assistance modes and adjust the assistance force;

[0008] The outer loop control structure is used to implement a multi-mode on-demand auxiliary control strategy: by capturing key information of the patient's movement in real time, including the task tracking error e t 、The patient's muscle strength h , rehabilitation training task cycle T, use the multi-index energy function EY to evaluate the patient's current motor ability, and realize the switching of the assist mode based on the current energy function value, and realize the assist force f based on fuzzy logic a , the frequency of rehabilitation training trajectory w d and radius r d Online adaptive adjustment;

[0009] The inner loop control structure is used to obtain the ideal tracking trajectory position P of the upper limb exoskeleton system using the admittance model. de Perform tracking control: Use the Q-TDE-based model-free BLF trajectory tracking control method to obtain the control law u(t) of the controller.

[0010] Compared with the prior art, the present invention has the following significant advantages:

[0011] (1) The present invention adopts a model-free BLF trajectory tracking control method based on Q-TDE technology. On the one hand, this method utilizes the data-driven extremely local model modeling principle to get rid of the dependence on precise mathematical modeling of the controlled object, and has good versatility and robustness; on the other hand, it adopts the obstacle Lyapunov function control strategy to effectively constrain the error range in the trajectory tracking process, and realize more accurate and stable trajectory tracking control.

[0012] (2) The present invention proposes a multi-mode force field method based on multi-index optimization, which can adaptively adjust the force field area according to the patient's motor performance during the rehabilitation process, thereby achieving personalized assistance for patients with different motor abilities and enhancing active participation.

[0013] (3) The present invention introduces a multi-metric fuzzy logic-based adaptive assist-as-needed control (MMFLA-AAN) method. By integrating multi-dimensional information such as task tracking error and patient muscle strength, fuzzy logic is used to adaptively adjust the size of the assist force and the rehabilitation trajectory, thereby achieving more personalized on-demand assistance, effectively improving the training effect, and preventing secondary injuries to patients during training. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] Figure 1 It is an overall flow chart of the control method of the present invention.

[0015] Figure 2 It is the overall structural diagram of the control method of the present invention.

[0016] Figure 3 This is a schematic diagram of the multi-mode force field planning of the present invention.

[0017] Figure 4 Schematic diagram of the membership function of input and output fuzzy variables.

[0018] Figure 5 Schematic diagram of the input-output relationship of the fuzzy system.

[0019] Figure 6 Schematic diagram of the human-machine coupling model of the upper limb exoskeleton system.

[0020] Figure 7 Schematic diagram of the rehabilitation training trajectory and tracking error of the trajectory tracking control method.

[0021] Figure 8Provides a schematic diagram of the changes in assistive force for the upper limb exoskeleton system in assistive mode.

[0022] Figure 9 Schematic diagram of the fuzzy system output, auxiliary force impedance stiffness, rehabilitation training trajectory frequency and radius changes in the auxiliary mode.

[0023] Figure 10 Schematic diagram of the ideal rehabilitation training trajectory and the patient's actual trajectory changes in the assisted mode.

[0024] Figure 11 Schematic diagram of the changes in the assistive force of the upper limb exoskeleton system under the multi-mode on-demand assistive control method.

[0025] Figure 12 Schematic diagram of the energy function, task tracking error and force field boundary changes under the multi-mode on-demand auxiliary control method.

[0026] Figure 13 Schematic diagram of the ideal rehabilitation training trajectory and the patient's actual trajectory changes under the multi-mode on-demand auxiliary control method.

[0027] Figure 14 This is a simulation diagram comparing the trajectory tracking error of the trajectory tracking control method proposed in the present invention and the trajectory tracking error of the TDE-iPD control method.

[0028] Figure 15 This is a simulation diagram comparing the auxiliary force changes of the MMFLA-AAN control method proposed in the present invention and other on-demand control methods.

[0029] Figure 16 This is a simulation diagram comparing the frequency and radius changes of the rehabilitation trajectory of the MMFLA-AAN control method proposed in the present invention and other on-demand control methods. DETAILED DESCRIPTION

[0030] The present invention will be further described below with reference to the accompanying drawings and specific embodiments.

[0031] In order to illustrate the technical solution and technical purpose of the present invention, the present invention is further introduced below with reference to the accompanying drawings and specific embodiments.

[0032] Combine Figure 1 The invention provides a multi-mode on-demand auxiliary control method based on multi-index optimization, which mainly includes the following steps:

[0033] Step 1: Establish a human-machine interaction dynamics model for the upper limb exoskeleton system. This includes the dynamics model of the upper limb exoskeleton and considers the coupling relationship between the upper limb exoskeleton system and the patient, ultimately resulting in a human-machine interaction dynamics model.

[0034] 1.1 Establishment of the joint space dynamics model of the upper limb exoskeleton system

[0035] Assuming that the motion of the upper limb exoskeleton system is rigid motion, the Lagrangian method is used to model the dynamics of the exoskeleton system, and the dynamic model of its joint space is established as follows:

[0036]

[0037] in, q is the angular acceleration, angular velocity and angle information of each joint of the upper limb exoskeleton system, M(q) is the inertia matrix of the upper limb exoskeleton system in the joint space, is the Coriolis force matrix of the upper limb exoskeleton system in the joint space, G(q) is the gravity matrix of the upper limb exoskeleton system in the joint space, and τ is the joint torque of each joint in the upper limb exoskeleton system.

[0038] 1.2 Establishment of the spatial dynamics model of the upper limb exoskeleton system

[0039] According to the principle of virtual work, the relationship between the joint torque and end force corresponding to the upper limb exoskeleton system is:

[0040] τ=J T f (2)

[0041] Where J is the Jacobian matrix corresponding to the terminal joint of the upper limb exoskeleton system, J T represents the transposed form of the Jacobian matrix; f represents the end force acting on the upper limb exoskeleton system.

[0042] Therefore, the dynamic equations of the joint space above are rewritten as follows:

[0043]

[0044] in, P are the terminal acceleration, velocity and position information of the upper limb exoskeleton system in the operation space; Represents the differential form of the Jacobian matrix with respect to time; Since the Jacobian matrix J in this upper limb exoskeleton system is not invertible (not a square matrix), the least squares method is used to solve the pseudo-inverse J + =J T (JJ T ) -1 To replace.

[0045] Arranging the above formula (3), we get:

[0046]

[0047] Among them, J +T That is the pseudo-inverse of the Jacobian matrix J+ The transposed form of .

[0048] By reorganizing the above formula (4) and combining some terms, we can get the following formula:

[0049]

[0050] Among them, M e (P)=J +T M(q)J + is the inertia matrix of the upper limb exoskeleton system in the operating space, is the centrifugal force and Coriolis force matrix in the upper limb exoskeleton system operation space, G e (P)=J +T G(q) is the gravity matrix in the operating space of the upper limb exoskeleton system.

[0051] 1.3 Establishment of a human-machine collaborative dynamics model for the upper limb exoskeleton system operating space

[0052] Since the end force f on the upper limb exoskeleton system in the above formula (5) includes the equivalent end input force f of the upper limb exoskeleton system c and human-computer interaction force int , so the above formula (5) can be rewritten as follows:

[0053]

[0054] The above formula is the human-computer interaction dynamics model of the upper limb exoskeleton system in the operating space.

[0055] Step 2: Establish an inner and outer loop force-position hybrid control structure of the upper limb exoskeleton integration system based on the admittance model, including an outer loop multi-mode on-demand auxiliary intelligent control method based on multi-index optimization and an inner loop model-free BLF control method based on Q-TDE technology.

[0056] After establishing the human-computer interaction dynamics model of the upper limb exoskeleton system, the present invention further designs the overall control architecture of the upper limb exoskeleton integration system, such as Figure 2 As shown, an inner and outer double-loop control structure is adopted, which can realize the switching of different assistance modes and the adjustment of assistance levels to meet the patient's personalized assistance needs at different rehabilitation stages.

[0057] The outer loop control system is mainly used to implement a multi-mode on-demand auxiliary control strategy, guided by the rehabilitation training task goal, by capturing key information of the patient's movement process in real time, including the task tracking error e t 、The patient's muscle strength h, rehabilitation training task cycle T and other indicators, use the multi-index energy function EY to evaluate the patient's current motor ability, and realize the switching of the assistance mode based on the current energy function value, and realize the assistance force f based on fuzzy logic. a , the frequency of rehabilitation training trajectory w d and radius r d Online adaptive adjustment.

[0058] The admittance model is a common compliance control model. The main idea is to convert the force signal into a position deviation signal, and then obtain an ideal reference trajectory that satisfies the compliance relationship. Its specific expression is as follows:

[0059]

[0060] Among them, f exo The magnitude of the multimodal force provided to the upper limb exoskeleton system, P are the terminal acceleration, velocity and position information of the upper limb exoskeleton system in the operating space, P de are the acceleration, velocity and position information of the ideal tracking trajectory of the upper limb exoskeleton system that satisfies the above multi-mode force interaction, M d 、B d , K d are the inertia coefficient, damping coefficient and stiffness coefficient of the admittance model respectively.

[0061] Through Laplace transform, the above formula (7) can be transformed into the following form:

[0062] P de =P+(M d s 2 +B d s+K d ) -1 f exo (8)

[0063] Here, s is the complex variable in the Laplace transform.

[0064] Therefore, the multi-mode force f provided by the upper limb exoskeleton system generated by the outer loop is obtained. exo After that, it is processed by the admittance model (8) and converted into the ideal tracking trajectory position P of the upper limb exoskeleton. de .

[0065] The inner loop control system is mainly used to obtain the ideal tracking trajectory position P of the upper limb exoskeleton system obtained by the admittance model. deFor tracking control, a model-free BLF trajectory tracking control method based on Q-TDE is used. On the one hand, it does not rely on the precise mathematical model of the upper limb exoskeleton system and can constrain the trajectory tracking error by constructing BLF. On the other hand, it combines Q-TDE technology to realize online correction and compensation of the unknown dynamic characteristics of the system.

[0066] Step 3: Establish the inner-loop trajectory tracking control method of the upper-limb exoskeleton fusion system. Based on the data-driven concept, the BLF trajectory tracking control law is derived through the extremely local model of the upper-limb exoskeleton system, and the unknown dynamic characteristics of the system are obtained using the Q-TDE technology;

[0067] 3.1 Establishment of state-space equations

[0068] According to the extremely local model, the following state space equation of the upper limb exoskeleton system can be constructed:

[0069]

[0070] Among them, x1(t) is the position vector of the exoskeleton end; Represents the velocity of the exoskeleton end; represents the acceleration measurement of the exoskeleton end; F(t) represents the dynamic characteristics of the system, which includes the unmodeled part of the upper limb exoskeleton system, nonlinear parts such as friction, and uncertain parts; α is the control input gain, which is a positive real number parameter; u(t) is the control law of the upper limb exoskeleton system.

[0071] Based on the above state-space equations, the expressions of the position-level error e1(t) and velocity-level error e2(t) of the upper limb exoskeleton system end are:

[0072]

[0073] Among them, P de is the ideal tracking trajectory position of the upper limb exoskeleton system obtained by the above admittance model, ξ(t) is the virtual control quantity, which is designed and given using the backstepping method in the following text.

[0074] Combining equations (9) and (10), the differential forms of the above position-level error e1(t) and velocity-level error e2(t) with respect to time can be expressed as follows:

[0075]

[0076] in, It is the differential form of the virtual control quantity ξ(t) with respect to time.

[0077] 3.2 Solution of BLF control law based on backstepping method

[0078] Select the logarithmic BLF function as shown below:

[0079]

[0080] Among them, V is the value of BLF function, k e is a positive real number parameter, which is the error bound.

[0081] The differential form of the above BLF function with respect to time is as follows:

[0082]

[0083] Substituting formula (11) into formula (13), we can obtain the following formula:

[0084]

[0085] Therefore, the form of the above-mentioned virtual control quantity ξ(t) is designed as follows:

[0086]

[0087] Wherein, k1 is a positive real number parameter.

[0088] Substituting equations (15) and (11) into equation (14), we can obtain:

[0089]

[0090] Therefore, the control law of the trajectory tracking controller designed based on the backstepping method can be expressed as follows:

[0091]

[0092] Wherein, k2 is a positive real number parameter.

[0093] 3.3 Q-TDE technology estimates the dynamic characteristics of the system

[0094] Because the term F(t) in the control law u(t) designed above contains the unknown dynamic characteristics of the system, it is difficult to directly obtain its accurate expression. To overcome this problem, the present invention uses the Q-TDE method to approximate the system dynamic characteristic term F(t) to obtain the control law u(t) of the control system.

[0095] From the above formula (9), we can see that the form of the system dynamic characteristics F(t) is as follows:

[0096]

[0097] Assuming that Δt is a very small time delay, the dynamic characteristics of the system in a short time can be ignored. Therefore, based on the above formula (18), the estimated value of F(t) is In the following form:

[0098]

[0099] Substituting the above equation (19) into equation (17), we can obtain the complete control law form of the model-free BLF controller based on Q-TDE, as shown in the following equation:

[0100]

[0101] Step 4: Establish a multi-mode assisted intelligent control method based on multi-index optimization for the outer loop of the upper limb exoskeleton integration system. Based on the patient's real-time motion performance such as task tracking error and output, the force field distribution is adaptively adjusted to select the appropriate assistance mode for the patient, including assistance mode, resistance mode, and passive mode.

[0102] 4.1. Force Strategy of the Upper Limb Exoskeleton Co-integration System Based on Multi-Mode Force Field

[0103] This control method mainly focuses on the multi-mode on-demand auxiliary intelligent control method based on multi-index optimization. Therefore, a multi-mode force field is constructed around the task trajectory to achieve switching between auxiliary mode and resistance mode, such as Figure 3 As shown. The center of the circle is the ideal task trajectory position, and the resistance training area is the radius r s The circular area outside the resistance training area is the auxiliary training area. It should be noted here that the passive mode of the upper limb exoskeleton integration system does not involve the above-mentioned inner and outer loop control structure, and is a complete trajectory tracking control, that is, the above-mentioned Q-TDE-based model-free BLF trajectory tracking control method. However, based on the inner and outer loop control structure used in the present invention, the passive mode can also be achieved by the auxiliary mode. When the force required to complete the rehabilitation training task is completely provided by the upper limb exoskeleton integration system, the system is in passive mode at this time.

[0104] First, define the patient's task tracking error e during rehabilitation training. t As shown below:

[0105] e t =||P d -P h ||2 (21)

[0106] Among them, P d is the ideal rehabilitation training trajectory position, P h The trajectory position is trained for the patient's real task, and || ||2 represents the Euclidean norm of the vector.

[0107] Define the task tracking error auxiliary variable r a As shown below:

[0108]

[0109] Among them, δ>0 is a certain constant coefficient, is the task tracking error e t Differential form with respect to time.

[0110] Depend on Figure 3 It can be seen that in the resistance training area, the upper limb exoskeleton integration system provides resistance to the patient to prevent the patient from approaching the ideal position, thereby improving the patient's active participation and preventing the occurrence of training slackness. Based on the above task tracking error, the resistance f is designed r The specific design form is as follows:

[0111]

[0112] Among them, k r is a positive real number parameter of the resistance model; σ is a positive real number parameter of the controller, the main purpose of which is to limit f r The size of e t =||P d -P h ||2 is the task tracking error, A unit vector representing the direction of the task tracking error.

[0113] exist Figure 3 In the auxiliary training area, the upper limb exoskeleton integration system provides appropriate auxiliary force for the patient to help him complete the rehabilitation training trajectory with a smaller task tracking error and avoid secondary injuries during the rehabilitation training process. Based on the above task tracking error auxiliary variable, the auxiliary force f is designed. a The specific form is as follows:

[0114]

[0115] Among them, k a is the stiffness coefficient of the auxiliary force model, is the unit vector representing the direction of the task tracking error.

[0116] Therefore, the multi-mode force provided by the upper limb exoskeleton fusion system in the above multi-mode force field is as follows:

[0117]

[0118] Among them, f exo That is the magnitude of the multi-modal force provided by the upper limb exoskeleton integration system, represents the average task tracking error of the patient's i-th rehabilitation training task cycle.

[0119] 4.2 Adaptive force field region adjustment based on multi-index energy function

[0120] When patients have weak motor skills and are unable to overcome resistance to complete rehabilitation training, a larger resistance training area may cause them to frequently switch between resistance and assist modes during exercise, causing discomfort. On the other hand, when patients have stronger motor skills, a smaller resistance training area may cause them to slack off during training. Therefore, based on the aforementioned multi-mode force field planning method, this paper proposes an adaptive force field area adjustment method based on a multi-index energy function.

[0121] Therefore, based on the cycle T of the i-th rehabilitation training task i , the average task tracking error of patients in the i-th rehabilitation training task cycle Average patient muscle strength As an evaluation indicator, a multi-index energy function is established as follows:

[0122]

[0123] Among them, EY represents the value of the multi-index energy function, and ρ, β, and ε are all positive real number parameters.

[0124] From Equation (26), it can be seen that the multi-index energy function EY is proportional to the patient's motor ability. The stronger the patient's motor ability, the larger the value of the multi-index energy function. Based on the multi-index energy function, it can be concluded that when the multi-index energy function value is large, the resistance training area should be appropriately increased to increase the difficulty of the patient completing the rehabilitation training task and prevent the patient from slacking off during the training process. Conversely, when the multi-index energy function value is small, the resistance training area should be appropriately reduced to help the patient better complete the rehabilitation training task and prevent secondary injuries during the rehabilitation training process.

[0125] Based on the above energy function, the impedance training area radius r is designed s The adaptive update law of is as follows:

[0126] r s =r smin +(r smax -r smin )*sigmoid(EY) (27)

[0127] Where sigmoid(EY)=(1+e -EY ) -1, is a common activation function that can limit the function value to [0,1], thereby limiting r s ∈[r smin ,r smax ]. smin ,r smax The radius r s The minimum and maximum values of .

[0128] Step 5: Establish a multi-metric fuzzy logic-based adaptive assist-as-needed control (MMFLA-AAN) method for the upper limb exoskeleton integration system. Based on the assist mode in the above multi-modal force field, combined with multiple motion indicators, the assist force f provided by the upper limb exoskeleton system is adaptively adjusted based on fuzzy logic. a and the frequency and radius of rehabilitation training trajectories.

[0129] It can be seen from the above formula (24) that the stiffness coefficient k of the auxiliary force impedance model is a The larger it is, the greater the assistive force provided by the exoskeleton, and the higher the assistance level for the patient. Therefore, the on-demand assistive adjustment of the patient's assistance level is a process of adaptively adjusting the stiffness coefficient of the assistive force model.

[0130] Generally speaking, the greater the patient's task tracking error, the weaker the patient's motor ability; conversely, the smaller the patient's task tracking error, the stronger the patient's motor ability. On the other hand, the greater the patient's exertion during training, the stronger the patient's motor ability; conversely, the smaller the patient's exertion during training, the weaker the patient's motor ability. Therefore, based on the idea of the stiffness coefficient of the above-mentioned adaptive auxiliary force model, when the patient's exertion is large and can complete the rehabilitation training movements with a smaller task tracking error, the auxiliary force for the patient is reduced, and the frequency and radius of the rehabilitation training trajectory are increased to encourage the patient to exert force more actively; conversely, when the patient's exertion is small and the task tracking error for completing the rehabilitation training movements is large, a larger auxiliary force is appropriately provided to the patient, and the frequency and radius of the rehabilitation training trajectory are reduced to help the patient complete the rehabilitation training movements better.

[0131] Therefore, the stiffness coefficient k of the auxiliary force model is designed a , the frequency of rehabilitation training trajectory w d and radius r d The adaptive update strategy is as follows:

[0132]

[0133] where γ k , γ w, γ r , k0, w0, r0 are all positive real number parameters; Δ represents the adaptive update rate, which is the output of the following fuzzy system.

[0134] Fuzzy systems deal with uncertainty and nonlinear problems through fuzzy rules. They do not require precise mathematical models, but instead transform human experience into update strategies through “if-then” rules. Referring to the adaptive update adjustment strategy in equation (28), the average task tracking error of the patient’s i-th rehabilitation training task cycle is and average patient muscle strength As the input of the fuzzy system, the above-mentioned adaptive update rate Δ is used as the output of the fuzzy system. A two-input and one-output fuzzy system is designed, and its core process includes three stages: fuzzification, fuzzy reasoning and defuzzification.

[0135] First, the fuzzification process is to transform the input clear variables into fuzzy variables and define the input of the fuzzy system. The domains after scaling are [0, 0.25] and [0, 15] respectively, and the domain of the output Δ of the fuzzy system after scaling is defined as [-1, 1]. The fuzzy subsets of input and output are defined as follows:

[0136]

[0137] In the above formula, VS, S, M, B, and VB represent fuzzy variables that increase from very small to very large; in {NVB, NB, NS, NVS, ZO, PVS, PS, PB, PVB}, Z0 represents a fuzzy variable of 0, NVB, NB, NS, and NVS represent fuzzy variables that increase from very small to very large in the negative range, and PVS, PS, PB, and PVB represent fuzzy variables that increase from very small to very large in the positive range.

[0138] All the above-mentioned fuzzy variable membership functions of input and output adopt Gaussian membership functions to ensure the smoothness of the fuzzy system output, such as Figure 4 shown.

[0139] The fuzzy reasoning process is to process and output the input fuzzy variables using the Mamdani reasoning method based on the fuzzy rule base in the fuzzy system. Based on the above-mentioned impedance stiffness adjustment strategy, the fuzzy rule base shown in Table 1 is designed.

[0140] Table 1 Fuzzy rules table

[0141]

[0142] The process of clarification is mainly to convert the output fuzzy variables obtained above into specific numerical variables. This paper uses the area center method to clarify the output fuzzy variables to obtain the above-mentioned adaptive update rate. The expression of the adaptive update rate Δ calculated using the area center method is as follows:

[0143]

[0144] in represents the membership degree of the fuzzy input d corresponding to the jth fuzzy rule, Indicates the weight value of the corresponding fuzzy rule.

[0145] Finally, the input-output correspondence surface of the fuzzy system constructed above is obtained as follows Figure 5 shown.

[0146] At this point, the overall design of the multi-mode on-demand assisted intelligent control method for the upper limb exoskeleton integration system based on multi-index optimization has been completed. In order to verify the effectiveness and feasibility of the above control method, the following combined with specific embodiments, through the establishment of joint simulation, the above control method is simulated and tested and its performance analyzed.

[0147] Example

[0148] Next, an overall simulation of the inner and outer ring control system of the upper limb exoskeleton is established. By performing typical rehabilitation task simulations, the effectiveness of the multi-mode on-demand auxiliary control method proposed above in terms of trajectory tracking accuracy, auxiliary force field adjustment, and changes in auxiliary force and training trajectory is verified. This demonstrates the effectiveness and advancement of the control method in improving rehabilitation training effects and the quality of human-computer interaction.

[0149] Step 1: Establish a virtual prototype of the upper limb exoskeleton human-computer interaction system. Construct a three-dimensional model of the upper limb exoskeleton system and the human body and assemble the two to establish a virtual prototype model of the human-computer interaction.

[0150] This simulation process uses a virtual prototype of the shoulder and elbow with two degrees of freedom, including the flexion / extension degrees of freedom of the shoulder joint and the flexion / extension degrees of freedom of the elbow joint. Based on the idea of co-simulation between SolidWorks and Matlab / Simulink, we first build a 3D model of the upper limb exoskeleton system and the human body in SolidWorks and assemble the two. Then, we import them into Matlab / Simulink through the Simmechanics toolbox to establish a human-machine coupling model. The specific diagram is shown below. Figure 6 shown.

[0151] Since the muscle force of the human upper limb can be approximated as a spring-damper model, and the patient's motor ability can be characterized by the stiffness coefficient of the spring-damper model, the muscle force at the patient's end in the above human-computer interaction system can be represented by the following model:

[0152]

[0153] where f h That is, the patient's muscle strength under simulation conditions, e t =||P d -P h ||2 represents the task tracking error, Represents the task tracking error e t The differential form with respect to time, || ||2 represents the Euclidean norm of the vector, P d and P h are the ideal rehabilitation training trajectory position of the patient and the actual task training trajectory position of the patient, K p , K d are the stiffness coefficient and damping coefficient of the patient's upper limb impedance model, respectively.

[0154] After obtaining the muscle force at the patient's end during the patient simulation based on the above model (31), the force acting on the patient's upper limb end can be obtained, which is converted into the joint space through Jacobi transformation, and finally the control input torque of the two joints of the patient's upper limb is obtained to realize the simulation control of the human body model.

[0155] Step 2: Establish a multi-mode on-demand assistive control method for the inner and outer loops of the upper-limb exoskeleton co-integration system. This includes an outer loop multi-mode on-demand assistive control method based on multi-metric optimization and an inner loop model-free BLF trajectory tracking control method based on Q-TDE technology. Since the detailed design process of this control method has been described above, it will not be repeated here. The following focuses on the parameter selection involved in this control method.

[0156] First, a model-free BLF trajectory tracking control method based on Q-TDE is constructed for the inner loop. The main parameter values are shown in Table 2 below.

[0157] Table 2 Inner loop control parameters

[0158]

[0159] Secondly, the outer ring is constructed based on the multi-mode force field of multi-index optimization, which mainly involves the multi-mode force f exo The parameter values in the energy function EY are shown in Table 3 below.

[0160] Table 3 Parameters involved in multi-mode force field

[0161]

[0162] Finally, an on-demand auxiliary intelligent control method based on multi-index fuzzy logic adaptation in the auxiliary mode of the outer loop is constructed. The main parameter values involved are shown in Table 4 below.

[0163] Table 4 MMFLA-AAN control parameters

[0164]

[0165] Step 3: Joint simulation verification and result analysis, mainly including the setting of simulation scenarios and analysis of simulation results.

[0166] 3.1. Simulation scenario setting and simulation results analysis of the inner loop trajectory tracking method

[0167] Select the coordinate origin O of the task space XY is the rotation center of the shoulder joint, the X-axis direction is the sagittal axis direction of the human body, the Y-axis direction is the coronal axis direction of the human body, the training trajectory is a circular trajectory with a radius of 0.15m, the training frequency is 1.57rad / s, and the initial position is [0.35m,0].

[0168] Firstly, the above-mentioned rehabilitation training trajectory is directly used as the ideal trajectory of the upper limb exoskeleton, and the control effect of the inner loop model-free BLF control method based on Q-TDE technology proposed above is studied.

[0169] The rehabilitation training trajectory and trajectory tracking error of the inner loop in the X and Y directions are given as follows: Figure 7 As shown. It can be seen that the overall control effect is good, and the mean root mean square deviation of the trajectory tracking error in the X and Y directions are [1.28*10 -7 ,2.5*10 -7 ], the error always converges within the error bound.

[0170] 3.2. Simulation scenario design of MMFLA-AAN control method in auxiliary mode and analysis of simulation results

[0171] The stiffness coefficient K of the patient's upper limb impedance model is obtained from the patient's spring damping model (31) p It can reflect the patient's motor ability, so the stiffness coefficient K of the patient's upper limb impedance model is changed during the simulation process. p Simulate patients with different exercise abilities.

[0172] In this simulation process, in order to ensure that the upper limb exoskeleton system is in auxiliary mode, r is set in this simulation process. s =0, set the initial stiffness coefficient K of the patient's upper limb impedance model p=850N / m to simulate patients with strong motor ability. The stiffness coefficient is set to gradually decrease within 0-100s during the simulation process, and is reduced to 50N / m at the end of the simulation to simulate patients with extremely weak motor ability.

[0173] Select the coordinate origin O of the task space XY is the rotation center of the shoulder joint, the X-axis direction is the sagittal axis direction of the human body, the Y-axis direction is the coronal axis direction of the human body, the initial training trajectory is selected as a circular trajectory with a radius of 0.15m, the initial training frequency is 1.57rad / s, and the initial position is [0.35m,0].

[0174] In the above simulation scenario, the auxiliary force f of the upper limb exoskeleton system is obtained a Changes such as Figure 8 As shown, the output Δ of the fuzzy system and the auxiliary force impedance stiffness k a , the radius of rehabilitation training trajectory and the changes in frequency are as follows Figure 9 As shown, the rehabilitation training task trajectory and the patient's actual trajectory are as follows Figure 10 shown.

[0175] The simulation results show that, due to the patient's initially strong motor ability, the fuzzy system output is negative, reducing the upper-limb exoskeleton's level of assistance while increasing the frequency and radius of the rehabilitation training trajectory, thereby enhancing the patient's active participation in the rehabilitation exercises. However, as the patient's motor ability gradually decreases during the simulation, the fuzzy system output gradually increases and transitions to a positive value. The upper-limb exoskeleton's level of assistance gradually increases, while the frequency and radius of the rehabilitation training trajectory decrease to ensure that the patient can successfully complete the rehabilitation exercises and prevent secondary injuries during the rehabilitation exercises.

[0176] 3.3. Simulation scenario design of multi-mode on-demand auxiliary intelligent control method and simulation results analysis

[0177] Similarly, by changing the stiffness coefficient K p Simulate the method of patients with different motor abilities and simulate the multi-mode on-demand assistive control method. In this simulation process, the stiffness coefficient K of the patient's upper limb impedance model is set within 0-30s. p =850N / m to simulate patients with strong athletic ability, and set the stiffness coefficient K of the patient's upper limb impedance model within 30-60s p =450N / m to simulate patients with weaker motor ability, and set the stiffness coefficient K of the patient's upper limb impedance model within 60-100s p =250N / m to simulate patients with extremely weak exercise ability.

[0178] Similarly, select the coordinate origin O of the task space XYis the rotation center of the shoulder joint, the X-axis direction is the sagittal axis direction of the human body, the Y-axis direction is the coronal axis direction of the human body, the initial training trajectory is selected as a circular trajectory with a radius of 0.15m, the initial training frequency is 1.57rad / s, and the initial position is [0.35m,0].

[0179] In the above simulation scenario, the multi-mode force changes provided by the upper limb exoskeleton system are obtained as follows: Figure 11 As shown, the multi-index energy function value EY changes and the period average task tracking error and the radius of the resistance region r s The relationship as Figure 12 As shown, the rehabilitation training task trajectory and the patient's actual trajectory change as shown in Figure 13 shown.

[0180] It can be seen from the simulation results that when the patient's motor ability is strong, the upper limb exoskeleton system provides resistance for the patient. Figure 12 It can be seen that when the patient's motor ability weakens, the upper limb exoskeleton system switches to the assistive mode, and the MMFLA-AAN intelligent control method adjusts the assistive force and rehabilitation training trajectory. Figure 11 and changes in rehabilitation training trajectories Figure 13 It can be seen that in the assist mode, the smaller the patient's motor ability, the greater the assist force provided by the upper limb exoskeleton system, and the smaller the radius and frequency of the rehabilitation training trajectory, which can achieve a better on-demand assistive control effect and mobilize the patient's enthusiasm for rehabilitation training.

[0181] Step 4: Comparative simulation analysis, mainly including comparison between the above trajectory tracking control method and other trajectory tracking control methods, and comparison between the above on-demand auxiliary control method and other on-demand auxiliary control methods, to verify the advanced nature of the above on-demand auxiliary control method.

[0182] 4.1 Comparison of the above trajectory tracking control method with other trajectory tracking control methods

[0183] To demonstrate the effectiveness of the proposed Q-TDE-based model-free BLF trajectory tracking control method (QTDE-BLF-MFC), this paper introduces the Q-TDE-based intelligent PD control (Quasi-time delay estimation-based PD control, QTDE-iPD), which is expressed as follows:

[0184]

[0185] Among them, e1 is the position level error of the end of the exoskeleton system, K p1 and K d1are proportional gain and differential gain respectively, α1 is the control input gain, The dynamic characteristics of the system obtained by Q-TDE technology are expressed as follows:

[0186]

[0187] Here, Δt represents the micro-delay.

[0188] During the joint simulation process, the control parameters of the intelligent PD controller based on Q-TDE technology are shown in Table 5 below.

[0189] Table 5 Control parameter values of QTDE-iPD

[0190]

[0191] Select the coordinate origin O of the task space XY is the rotation center of the shoulder joint, the X-axis direction is the sagittal axis direction of the human body, and the Y-axis direction is the coronal axis direction of the human body. The rehabilitation training trajectory is a circular trajectory with a radius of 0.15m, the training frequency is 1.57rad / s, and the initial position is [0.35m,0].

[0192] The final trajectory tracking error is Figure 14 As shown in the figure, it can be seen that the trajectory tracking control method proposed by the present invention and the comparative method both have some oscillations in the initial situation due to the delay estimation error. However, the trajectory tracking control method proposed by the present invention has a smaller oscillation amplitude than the comparative method, and the trajectory tracking error converges faster than the comparative method. After stabilization, the trajectory tracking error is also smaller than that of the comparative method.

[0193] 4.2 Comparison of the above-mentioned on-demand auxiliary control method with other on-demand auxiliary control methods

[0194] In order to verify that the above-mentioned multi-index fuzzy logic adaptive on-demand auxiliary intelligent control method (MMFLA-AAN) is more advanced, other on-demand auxiliary control methods are introduced for comparative analysis.

[0195] On the one hand, the on-demand auxiliary control method based on Horf oscillator (HORF-AAN) proposed in the literature (Nanjing University of Science and Technology. Seven-degree-of-freedom upper limb exoskeleton and its on-demand auxiliary control method: 202411372433.X[P]. 2025-01-17.) is introduced for analysis;

[0196] On the other hand, the adaptive central pattern generator (ACPG) proposed in the literature (Sharifi M, Mehr JK, Mushahwar VK, et al. Autonomous locomotion trajectory shaping and nonlinear control for lower limb exoskeletons [J]. Ieee / ASME Transactions on Mechatronics, 2022, 27(2): 645-655.) is introduced, and the adaptive on-demand assisted control method (ACPG-AAN) combined with the energy function for task trajectory adjustment is introduced, that is, the adaptive on-demand assisted control method in the literature (Tian Y, Guo Y, Wang H, et al. Training task planning-based adaptive assist-as-needed control for upper limb exoskeleton using neural network state observer [J]. Neural Computing and Applications, 2024, 36(26): 16037-16055.) is compared and analyzed.

[0197] In this simulation process, in order to ensure that the upper limb exoskeleton system is in auxiliary mode, r is set in this simulation process. s =0, set the initial stiffness coefficient K of the patient's upper limb impedance model p =850N / m to simulate patients with strong motor ability. The stiffness coefficient is set to gradually decrease within 0-100s during the simulation process, and is reduced to 50N / m at the end of the simulation to simulate patients with extremely weak motor ability.

[0198] Select the coordinate origin O of the task space XY is the rotation center of the shoulder joint, the X-axis direction is the sagittal axis direction of the human body, the Y-axis direction is the coronal axis direction of the human body, the initial training trajectory is selected as a circular trajectory with a radius of 0.15m, the initial training frequency is 1.57rad / s, and the initial position is [0.35m,0].

[0199] In addition, the control parameter values involved in the above-mentioned HORF-AAN control method are shown in Table 6 below, and some control parameters of the above-mentioned ACPG-AAN control method are shown in Table 7 below.

[0200] Table 6 Control parameter values of HORF-AAN

[0201]

[0202] Table 7 Control parameter values of ACPG-AAN

[0203]

[0204] The final auxiliary force comparison of the upper limb exoskeleton system is as follows: Figure 15 As shown in the figure, the changes in the upper limb exoskeleton rehabilitation training trajectory are compared. Figure 16 shown.

[0205] Depend on Figure 15 It can be seen that compared with the above-mentioned HORF-AAN control method, the MMFLA-AAN control method proposed in this article has a lower assistance level when the patient's motor ability is stronger, and provides greater assistance force when the patient's motor ability is weaker. It can be seen that the on-demand assistance control method proposed in the present invention is more sensitive to the patient's motor ability and can achieve better on-demand assistance control effect; compared with the ACPG-AAN control method, the MMFLA-AAN control method proposed in the present invention always provides smaller assistance force throughout the entire rehabilitation training simulation process, and as the patient's motor ability decreases, the assistance force provided by the on-demand assistance control method proposed in the present invention increases faster. It can be seen that the on-demand assistance control method proposed in the present invention can better mobilize the patient's active participation in the rehabilitation training process and can achieve better on-demand assistance control effect.

[0206] Depend on Figure 16 It can be seen that compared with the above-mentioned HORF-AAN control method, the ACPG-AAN control method and the MMFLA-AAN control method proposed in this paper will first increase the radius and frequency of the rehabilitation training task trajectory to increase the difficulty of the rehabilitation training task when the patient's motor ability is strong, so as to prevent the patient from slacking off in the rehabilitation training task; compared with the ACPG-AAN control method, the frequency and radius of the rehabilitation training trajectory of the MMFLA-AAN control method proposed in this paper will be larger during the entire rehabilitation training process, and Figure 15 It can be seen that the upper limb exoskeleton system provides less assistive force under the MMFLA-AAN control method. Therefore, the on-demand assistive control method proposed in this paper can better mobilize patients to complete more difficult rehabilitation training tasks with less assistive force, and thus can better mobilize patients' active participation in the rehabilitation training process.

Claims

1. An upper limb exoskeleton based on multi-index optimization multi-mode on-demand auxiliary intelligent control method, characterized in that: The use of an inner and outer loop force-position hybrid control structure enables switching between different assist modes and adjustment of assist force; The outer loop control is used to realize the multi-mode on-demand auxiliary control strategy: by capturing the key information of the patient's movement in real time, including the task tracking error e t 、The patient's muscle strength h , rehabilitation training task cycle T, use the multi-index energy function EY to evaluate the patient's current motor ability, and realize the switching of the assist mode based on the current energy function value, and realize the assist force f based on fuzzy logic a , the frequency of rehabilitation training trajectory w d and radius r d Online adaptive adjustment; The inner loop control is used to obtain the ideal tracking trajectory position P of the upper limb exoskeleton system using the admittance model. de Perform tracking control: Use the model-free obstacle Lyapunov function trajectory tracking control method based on quasi-delay estimation to obtain the control law u(t) of the controller.

2. The upper limb exoskeleton according to claim 1 is characterized by: The control law u(t) of the controller is: Where e1(t) and e2(t) are the position level error and velocity level error respectively, x2(t) represents the velocity of the exoskeleton end, α is the control input gain, and k e is the error bound, k1 and k2 are both positive real number parameters, Δt is the time delay, and ξ(t) is the virtual control variable.

3. The upper limb exoskeleton according to claim 1 or 2 is characterized in that: The ideal tracking trajectory position P of the upper limb exoskeleton system obtained by the admittance model de for: P de =P+(M d s 2 +B d s+K d ) -1 f exo P is the position information of the upper limb exoskeleton system in the operation space, M d 、B d , K d are the inertia coefficient, damping coefficient and stiffness coefficient of the admittance model respectively, s is the complex variable in Laplace transform, f exo The magnitude of multimodal forces provided to an upper limb exoskeleton system.

4. The upper limb exoskeleton based on multi-index optimization multi-mode on-demand auxiliary intelligent control method according to claim 1 is characterized in that: The switching form of auxiliary mode is: Among them, f exo That is, the magnitude of the multi-mode force provided by the upper limb exoskeleton integration system, e t represents the task tracking error, represents the differential form of task tracking error with respect to time, represents the average task tracking error of the patient's i-th rehabilitation training task cycle, k r is the positive real number parameter of the resistance model, σ is the positive real number parameter of the controller, The unit vector representing the direction of the task tracking error, P d is the ideal rehabilitation training trajectory position, P h The trajectory position is trained for the patient's real task, k a is the stiffness coefficient of the auxiliary force model, r a is the auxiliary variable of task tracking error, r s is the radius of the impedance training area.

5. The upper limb exoskeleton according to claim 4 is characterized in that: The radius of the impedance training area r s : r s =r smin +(r smax -r smin )*sigmoid(EY) Among them, sigmoid is the activation function, r smin ,r smax For r s The minimum and maximum values of EY are the values of the multi-index energy function, and ρ, β, and ε are all positive real number parameters. is the average patient muscle strength in the i-th rehabilitation training task cycle, T i is the cycle of the i-th rehabilitation training task.

6. The upper limb exoskeleton based on multi-index optimization multi-mode on-demand auxiliary intelligent control method according to claim 1 is characterized in that: Assisting force f a , the frequency of rehabilitation training trajectory w d and radius r d The online adaptive adjustment of is achieved through the following adaptive update strategy: where k a is the stiffness coefficient of the auxiliary force model, γ k , γ w , γ r , k0, w0, r0 are all positive real number parameters; Δ represents the adaptive update rate, r a is the task tracking error auxiliary variable, The unit vector representing the direction of the task tracking error, P d is the ideal rehabilitation training trajectory position, P h Train trajectory positions for patient-realistic tasks.

Images