A soft knee joint walking aid performance digital twin analysis method and system

Abstract

The application discloses a soft knee joint walking aid performance digital twin analysis method and system, relates to the technical field of digital twinning, and comprises the following steps: acquiring simulation input data under different driving air pressures and walking aid state data synchronously collected by a distributed force sensor, an angle sensor and an angular velocity sensor, and setting initial conditions of a digital twin model; adjusting contact boundary conditions and viscoelastic internal variables, constructing reference working conditions and control working conditions, and performing dynamic simulation; calculating auxiliary torques according to contact distribution forces and joint angles, calculating energy transmission efficiencies, energy balance residuals and state prediction residuals of each working condition; correcting model parameters based on the residuals and iterating to convergence, outputting auxiliary torques, auxiliary walking effects and wearing comfort evaluation results under different driving air pressures and joint angles, and screening suitable driving air pressures. The application realizes multi-factor dissipation decoupling and model closed-loop correction, and improves the accuracy and reliability of performance analysis.

Description

Technical Field

[0001] This invention relates to the field of digital twin technology, and in particular to a digital twin analysis method and system for the performance of a soft knee joint walking aid. Background Technology

[0002] In recent years, soft knee joint assistive devices have seen rapid development in the field of rehabilitation medicine due to their excellent human-machine wearability. To achieve accurate evaluation of flexible drive systems, digital twin technology has been widely adopted. Existing technologies mostly simulate joint movement by constructing virtual dynamic models and combine this with finite element analysis to reflect the nonlinear mechanical response of materials, providing fundamental support for the structural optimization and control strategy development of assistive devices.

[0003] However, existing digital twin methods have significant drawbacks. First, they confuse frictional dissipation caused by contact slippage with hysteretic dissipation due to material deformation accumulation, lacking a multi-physical decoupling mechanism and failing to accurately pinpoint the root cause of energy loss. Second, traditional model calibration relies solely on a single state prediction error, neglecting the underlying energy conservation constraint and lacking a residual sensitivity optimization mechanism. This approach is highly susceptible to numerical mismatch, leading to a sharp decrease in model accuracy under complex operating conditions, making it impossible to achieve high-fidelity adaptive parameter calibration and accurate assessment of real energy transfer performance. Summary of the Invention

[0004] In view of the aforementioned existing problems, the present invention is proposed.

[0005] Therefore, this invention provides a digital twin analysis method for the performance of soft knee joint walking aids, which solves the problem in the prior art that the contact friction and material hysteresis dissipation of soft walking aids are difficult to decouple and quantify accurately, resulting in the difficulty of high-precision calibration of digital twin model parameters and inaccurate evaluation of energy transfer performance.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution: In a first aspect, the present invention provides a digital twin analysis method for the performance of a soft knee joint walking aid, comprising: acquiring simulation input data and walking aid state data of the soft knee joint walking aid; and using the walking aid state data to set the initial conditions of the digital twin model. Based on the initial conditions, the contact boundary conditions and viscoelastic internal variables of the digital twin model are adjusted to form a baseline working condition and a control working condition. Dynamic simulations are then performed using the simulation input data to obtain simulation results. Based on the simulation results, the energy transfer efficiency and energy balance residuals of the baseline working condition and the control working condition are calculated respectively, and the state prediction residuals are calculated in combination with the walking aid state data. The parameter correction amount of the digital twin model is calculated based on the state prediction residual and the energy balance residual. The digital twin model is updated using the parameter correction amount and the dynamic simulation is iteratively executed until the preset convergence condition is met. The performance analysis results of the soft knee joint walking aid are then output. Wherein, the state prediction residual is the deviation between the simulation result of the baseline working condition and the state data of the walking aid, and the energy balance residual is the energy conservation difference between the baseline working condition and the control working condition in the dynamic simulation.

[0007] As a preferred embodiment of the digital twin analysis method for the performance of the soft knee joint walking aid described in this invention, the initial conditions of the digital twin model are set using the walking aid state data, including extracting contact force data and motion data from the walking aid state data. Within a preset time period, the increment of material damage state is calculated based on the digital twin model. When the increment is less than a preset threshold, the material damage state parameter of the digital twin model is set to a fixed value. The dissipated energy at the contact interface is calculated using the contact force data and the motion data, and the dissipated energy is decomposed into normal damping dissipated energy and tangential friction dissipated energy. The motion data, the material damage state parameters set to fixed values, and the normal damping dissipation energy are used together as the initial conditions for the digital twin model.

[0008] As a preferred embodiment of the digital twin analysis method for the performance of the soft knee joint walking aid described in this invention, the following is included: adjusting the contact boundary conditions and viscoelastic internal variables of the digital twin model to form a baseline working condition and a control working condition, wherein the original contact boundary conditions and viscoelastic internal variables are retained in the baseline working condition. The contact boundary conditions are individually adjusted to eliminate the energy dissipated by tangential friction, thus constructing a first-class control condition. By individually adjusting the viscoelastic internal variables to eliminate the memory effect caused by material deformation accumulation, a second type of control condition is constructed. By simultaneously adjusting the contact boundary conditions and the viscoelastic internal variables, and eliminating the energy dissipated by tangential friction and the memory effect caused by the accumulation of material deformation, a third type of control condition is constructed.

[0009] As a preferred embodiment of the digital twin analysis method for the performance of the soft knee joint walking aid described in this invention, the construction of the first type of control working condition includes: analyzing the contact boundary conditions and extracting the normal contact state and tangential contact force of the contact interface; When the normal contact state is a separation state, the tangential contact force is forcibly set to zero; When the normal contact state is closed, under the physical condition that the tangential friction dissipation energy is greater than or equal to zero, the tangential relative sliding velocity of all contact nodes in the contact interface is extracted, and the tangential constraint force to be solved is introduced at the contact node. Construct an objective function to minimize the sum of squares of the tangential relative slip velocities and simultaneously minimize the sum of squares of the tangential constraint forces; Calculate the second derivative matrix of the objective function, and when the positive definiteness of the second derivative matrix determines that the tangential constraint force has a unique solution, use the objective function to solve for the corresponding precise tangential constraint force. The precise tangential constraint force is added to the contact boundary conditions to suppress the relative sliding of the contact interface.

[0010] As a preferred embodiment of the digital twin analysis method for the performance of the soft knee joint walking aid described in this invention, the construction of the second type of control working condition includes: extracting an initial physical model containing the initial conditions and macroscopic stress state, and extracting the current total stress and the current instantaneous tangential stiffness from the initial physical model; Introduce the equivalent elastic prestress to be solved and the elastic tangential stiffness corresponding to the equivalent elastic prestress; Calculate the difference between the current total stress and the equivalent elastic prestress to generate the total stress difference; calculate the difference between the current instantaneous tangential stiffness and the elastic tangential stiffness to generate the instantaneous tangential stiffness difference. While maintaining the macroscopic stress state unchanged, calculate the equivalent elastic prestress in which the total stress difference and the instantaneous tangential stiffness difference simultaneously reach their minimum values; Using the precise equivalent elastic prestress, the updated value of the viscoelastic internal variable is calculated, and the viscoelastic internal variable is set as the updated value.

[0011] As a preferred embodiment of the digital twin analysis method for the performance of the soft knee joint walking aid described in this invention, the calculation of the energy transfer efficiency of the benchmark condition and the control condition includes extracting the joint torque in the drive-on state, the passive joint torque in the drive-off state, the joint angular velocity, and the total available energy in a single cycle from the dynamic simulation results of the benchmark condition and the control condition within a preset assistance time interval. By integrating the difference between the joint torque in the drive-on state and the passive joint torque in the drive-off state, based on the joint angular velocity, the effective assist mechanical work is obtained. The energy transfer efficiency is obtained by calculating the ratio of the effective mechanical work to the total available energy. The energy transfer efficiency under the benchmark operating condition is calibrated to the actual transfer performance. The first marginal contribution value is obtained by subtracting the actual transmission performance from the energy transfer efficiency of the first type of control condition. The second marginal contribution value is obtained by subtracting the actual transmission performance from the energy transfer efficiency of the second type of control condition. The third marginal contribution value is obtained by subtracting the actual transmission performance, the first marginal contribution value, and the second marginal contribution value from the energy transfer efficiency of the third type of control condition.

[0012] As a preferred embodiment of the digital twin analysis method for the performance of the soft knee joint walking aid described in this invention, the method of updating the digital twin model using the parameter correction amount and iteratively executing the dynamic simulation includes correcting the time step parameter and mesh parameter of the simulation calculation when the energy balance residual exceeds a preset numerical error threshold. Extract the parameters to be corrected from the digital twin model, the parameters to be corrected including the contact boundary conditions and the viscoelastic internal variables; When the state prediction residual is greater than a preset residual threshold, the partial derivative of the state prediction residual with respect to the parameter to be corrected is calculated. Arrange the partial derivatives according to the parameter dimension to construct a residual sensitivity matrix; When it is determined that the parameter to be corrected has a unique solution, a regularized optimization equation is constructed using the residual sensitivity matrix, and the parameter correction amount is obtained by solving the regularized optimization equation. When the parameter correction amount changes the initial conditions of the digital twin model, the digital twin model is updated using the parameter correction amount, and the initial conditions are reset. When the parameter correction amount does not change the initial conditions of the digital twin model, the digital twin model is updated using the parameter correction amount, and the dynamic simulation is performed again.

[0013] Secondly, the present invention provides a digital twin analysis system for the performance of a soft knee joint walking aid, comprising: a data unit for acquiring simulation input data and walking aid status data of the soft knee joint walking aid, and using the walking aid status data to set the initial conditions of the digital twin model; The simulation unit, based on the initial conditions, adjusts the contact boundary conditions and viscoelastic internal variables of the digital twin model to form a baseline working condition and a control working condition, and performs dynamic simulations using the simulation input data to obtain simulation results. The evaluation unit calculates the energy transfer efficiency and energy balance residuals of the baseline working condition and the control working condition respectively based on the simulation results, and calculates the state prediction residuals in combination with the walking aid state data. The correction unit calculates the parameter correction amount of the digital twin model based on the state prediction residual and the energy balance residual, updates the digital twin model using the parameter correction amount, and iteratively executes the dynamic simulation until the preset convergence condition is met, and outputs the performance analysis results of the soft knee joint walking aid.

[0014] Thirdly, the present invention provides a computer device including a memory and a processor, wherein the memory stores a computer program, wherein: when the computer program is executed by the processor, it implements any step of the digital twin analysis method for the performance of a soft knee joint walking aid as described in the first aspect of the present invention.

[0015] Fourthly, the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program is executed by a processor, it implements any step of the digital twin analysis method for the performance of a soft knee joint walking aid as described in the first aspect of the present invention.

[0016] The beneficial effects of this invention are as follows: By constructing multiple control conditions through separate adjustments to contact boundary conditions and viscoelastic internal variables, the tangential frictional dissipation and material deformation accumulation effect of the soft walking aid are precisely decoupled, quantifying the marginal contribution of each variable to energy transfer efficiency and overcoming the technical bottleneck of independently assessing soft drive losses. A dual evaluation mechanism based on state prediction residuals and energy balance residuals is established, using the residual sensitivity matrix and regularized optimization equations to solve for parameter corrections. This ensures high-fidelity mapping of kinematic states while strictly adhering to the underlying physical law of energy conservation, achieving adaptive high-precision calibration of the digital twin model and significantly improving the accuracy and reliability of performance analysis. Attached Figure Description

[0017] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 A flowchart for the digital twin analysis method of soft knee joint walking aid performance. Detailed Implementation

[0019] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0020] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0021] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that is mutually exclusive with other embodiments.

[0022] Reference Figure 1 As one embodiment of the present invention, this embodiment provides a digital twin analysis method for the performance of a soft knee joint walking aid, comprising the following steps: S1: Obtain simulation input data and walker status data of the soft knee joint walker, and use the walker status data to set the initial conditions of the digital twin model.

[0023] Considering the complex nonlinear human-machine interaction behavior of soft knee-joint walking aids during actual wear, their physical state is far from an ideal zero-stress initial value state, but rather includes complex interface prestress and material history effects. Therefore, it is necessary to preemptively eliminate the slowly evolving fatigue damage interference in the early stages of simulation and separate the work component directly related to slippage from the mixed interface energy. This scheme aims to freeze irrelevant slowly varying parameters and clarify the initial energy boundaries in the normal and tangential directions, thus serving as an absolute benchmark for subsequent independent adjustment of contact boundary conditions and calculation of marginal contribution values.

[0024] The simulation input data includes drive activation control commands, drive deactivation control commands, drive input energy, drive air pressure parameters, preset assist time intervals, external load parameters, and environmental boundary parameters. The drive air pressure parameters include multiple drive air pressures set according to a preset air pressure gradient, each corresponding to simulation input data and walker status data for a complete gait cycle. The walker status data includes contact force data, joint angles, joint angular velocities, structural displacement, and structural deformation data. The baseline condition and various control conditions use the same simulation input data and time sampling reference.

[0025] Contact force data and motion data are extracted from the walker's state data. The motion data is used to extract nodal displacements, nodal velocities, joint angles, and joint angular velocities at the start of the analysis window when initial conditions are set. Motion data at other times within the analysis window are compared with the baseline simulation state to calculate the state prediction residual.

[0026] The motion data includes the three-dimensional spatial trajectory, joint angles, and joint angular velocities of the soft knee joint walker during the current gait cycle. The contact force data is collected through a sensor array attached to the human-machine interface. The sensor array includes distributed force sensors, angle sensors, and angular velocity sensors.

[0027] The distributed force sensors are deployed in three zones: a fixed thigh region, a knee joint transition region, and a fixed lower leg region. Flexible strip-type sensor arrays are laid along the circumference of the limb in the fixed thigh and lower leg regions to collect the contact force distribution generated by the soft knee joint assistive device at a fixed position. Flexible surface arrays are laid along the knee joint flexion and extension direction on the inner and outer sides of the knee joint in the knee joint transition region to collect the spatial transfer of contact force during assisted movement. The sensing units of each distributed force sensor are arranged at a preset interval, and the position of each sensing unit in the human-machine interface coordinate system is recorded. At least some of the sensing units in the distributed force sensors employ a combined normal and tangential force measurement structure to simultaneously collect normal pressure and tangential shear force; the remaining sensing units are used to collect normal pressure.

[0028] To minimize the impact of the sensor array on knee flexion and extension movements and wearing comfort, the sensing units with rigid encapsulation structures are not installed at the patellar prominence and the main flexion position of the popliteal fossa. When it is necessary to collect contact force data at the corresponding positions, thin, flexible sensing units are used, and the surface of the thin, flexible sensing units is flush with the inner lining surface of the soft knee joint walker. Each sensing unit is led out along the non-primary load-bearing area of ​​the soft knee joint walker via flexible wires. The flexible wires have a reserved follow-up length when crossing the knee flexion area to avoid restricting knee joint movement or causing additional interference to the contact force distribution.

[0029] The angle sensor is disposed on the outer side of the knee joint, and its rotation axis is aligned with the flexion-extension axis of the knee joint. The two connecting ends of the angle sensor are respectively fixed to the thigh fixation area and the lower leg fixation area, for acquiring the relative angle between the thigh and the lower leg. The angular velocity sensor includes a first angular velocity sensor and a second angular velocity sensor respectively fixed to the thigh fixation area and the lower leg fixation area. The first angular velocity sensor and the second angular velocity sensor use the same coordinate axis direction, and the joint angular velocity is determined based on the difference in angular velocity between the two sensors along the flexion-extension axis of the knee joint.

[0030] Before collecting the status data of the walking aid, each of the distributed force sensors is zeroed under no-load conditions and calibrated with a known load. The angle sensor is zero-position calibrated, and the first and second angular velocity sensors are calibrated in coordinate direction. The distributed force sensors, angle sensors, and angular velocity sensors are synchronously collected using a unified sampling frequency and timestamp. A sensor position mapping relationship is formed based on the number, arrangement position, and force direction of each sensor unit for subsequent calculation of contact distributed force, auxiliary torque, and state prediction residual.

[0031] The motion data represents the three-dimensional spatial trajectory, joint angles, and joint angular velocities of the walking aid during the current gait cycle. The contact force data is collected by distributed force sensors arranged on the contact surface between the soft knee joint walking aid and the human body; the joint angles are collected by angle sensors located at the knee joint movement position; and the joint angular velocities are collected by angular velocity sensors.

[0032] The soft knee joint walking aid is controlled sequentially according to multiple driving air pressures to perform assistive actions. Under each driving air pressure, the contact distribution force, joint angle and joint angular velocity within a complete gait cycle are collected synchronously. The contact distribution force, joint angle and joint angular velocity are time-aligned according to a unified timestamp to obtain the walking aid status data corresponding to each driving air pressure.

[0033] The status data of the walking aid corresponding to each of the aforementioned driving air pressures are stored separately and formed into independent analysis data groups. Data from different driving air pressures are not mixed in the same analysis window. When comparing the performance corresponding to each driving air pressure, the same sensor position mapping relationship, sampling frequency, time alignment reference, and gait cycle division standard are used to ensure that the contact force distribution, joint angle, and joint angular velocity under different driving air pressures have the same data comparison benchmark.

[0034] The contact distribution force includes the normal contact force and tangential contact force collected by each sensing unit, and retains the arrangement position of each sensing unit on the human-machine contact surface, which is used to subsequently convert the contact distribution force into an auxiliary torque generated by the soft knee joint walker on the knee joint.

[0035] After obtaining the basic state, considering that the soft elastomer material will undergo irreversible microscopic damage under long-term cyclic loading, in order to prevent the slow-varying damage parameters from introducing numerical calculation drift in a single short-cycle analysis, it is necessary to determine its evolution state.

[0036] Specifically, within a preset time period, i.e., the analysis window, the increment of material damage state is calculated based on the digital twin model. When the increment is less than a preset threshold, the material damage state parameter of the digital twin model is set to a fixed value. The fixed value is the current material damage state parameter value determined at the start of the analysis window, not a reset of the material damage state parameter to zero. The same fixed value is used for the baseline condition and various control conditions within the same analysis window.

[0037] When the increment of the material damage state is not less than the preset threshold, the analysis window containing multiple complete gait cycles is reduced to a smaller number of complete gait cycles without truncating individual gait cycles. When the analysis window already contains only one complete gait cycle, another complete steady-state gait cycle is selected for damage increment assessment. If the reassessment still does not meet the preset threshold, the marginal contribution calculation within the current analysis window is paused.

[0038] This determination process relies on the damage mechanics equations of continuum, quantifying short-time evolution by integrating the equivalent stress over time. The specific calculation formula is as follows: in, This represents the increment of the material damage state calculated within a preset time period; Indicates the start time of the analysis window; Indicates the end time of the analysis window; This represents the fatigue evolution constant pre-calibrated by the physical properties of the soft material; Indicates time Material equivalent stress; This represents the current material damage state parameter value, which is determined at the start time of the analysis window and remains unchanged within that analysis window. It represents the fatigue evolution constant index, which is pre-calibrated based on the physical properties of the soft material; This represents a time integral infinitesimal.

[0039] The initial material damage benchmark value is greater than or equal to zero and less than one to avoid the denominator being zero in the material damage evolution calculation; the fatigue evolution constant and its exponent are both positive values.

[0040] The dissipated energy at the contact interface is calculated using the contact force data and the motion data, and then decomposed into normal damping dissipated energy and tangential friction dissipated energy. Since the wearable device experiences both compression and micro-peristalsis on the human body surface, the total dissipated energy needs to be strictly separated by inner product integration along the normal and tangential directions of the contact surface. The formula for solving its components is as follows: in, This represents the measured normal damping dissipation energy calculated based on the walking aid's status data; Indicates the start time of the analysis window; Indicates the end time of the analysis window; Indicates time The damping component in the normal contact force; Indicates time The relative compression velocity vector; This represents a time integral infinitesimal.

[0041] in, This represents the measured tangential frictional dissipation energy calculated based on the walking aid's status data. Indicates the start time of the analysis window; Indicates the end time of the analysis window; Indicates time The actual tangential frictional force; Indicates time The tangential relative slip velocity vector; This represents a time integral infinitesimal.

[0042] Here, dissipated energy is uniformly defined as a non-negative quantity, and the contact force vector adopts the direction of the actual contact force acting on the relatively moving body.

[0043] This embodiment performs offline replay analysis on gait cycles for which data acquisition has been completed. The start time of the analysis window is used as the initial time for the dynamic simulation. The initial conditions include two parts: initial dynamic state values ​​and energy statistical boundaries. Motion data and material damage state parameters serve as the initial dynamic state values, while normal damping dissipation energy serves as the energy statistical boundary common to all working conditions. The normal damping dissipation energy is not directly substituted into the dynamic control equations as displacement, velocity, or stress state quantities. Instead, it is written into the digital twin model as the reference value of the normal dissipation accumulator or its corresponding normal damping parameter.

[0044] The motion data and material damage state parameters are constructed as initial dynamic states, and the normal damping dissipation energy is set as the energy statistical boundary associated with the initial dynamic states. Together, they constitute the initial conditions of the digital twin model as referred to in this application. That is, each term is constructed as an initial state vector containing nodal displacements, nodal velocities, joint angles, joint angular velocities, element damage degree, and normal energy boundaries, and forcibly assigned and injected into the starting storage module of the finite element dynamics solver, making it the unified starting point for subsequent baseline conditions and all control conditions.

[0045] By pre-locking slowly varying material damage parameters and precisely stripping tangential friction energy, cross-interference terms affecting the calculation of energy transfer in soft knee joints are eliminated from the physical level. This not only defines stringent computational boundaries for the digital twin model, preventing the simulation solver from diverging and collapsing under long-time-step integration, but also makes subsequent individual adjustment of boundary conditions to eliminate specific memory effects a quantifiable engineering operation, ensuring that the multi-dimensional residual iterative correction architecture has a unique and high-fidelity convergence benchmark.

[0046] S2: Based on the initial conditions, adjust the contact boundary conditions and viscoelastic internal variables of the digital twin model to form a baseline working condition and a control working condition, and use the simulation input data to perform dynamic simulations to obtain simulation results.

[0047] Considering that in real-world wearable interactions with soft knee-joint walking aids, the dissipation of frictional friction at the human-machine interface and the hysteresis dissipation caused by viscoelastic deformation within the material are often highly coupled, making it difficult to directly measure the dissipation in physical entity testing, this approach leverages the strong intervention characteristics of digital twin space to artificially "cut off" specific physical dissipation channels in the simulation environment using a controlled variable method. Since this scheme requires extremely precise evaluation of the independent marginal contributions of friction and material hysteresis to the final energy transfer efficiency, it is necessary to determine absolute initial conditions and establish a parallel control matrix including the original baseline and multiple extreme non-destructive control groups. This provides a complete and orthogonal data source for subsequent residual extraction and calculation of the independent and interactive effects of the two factors.

[0048] In the baseline condition, the original contact boundary conditions and viscoelastic internal variables are retained. That is, no physical resistance is intervened in the simulation, serving as a comparative reference surface for mapping to the real physical world. Simultaneously, to achieve cross-decoupling, the contact boundary conditions are adjusted individually to eliminate the tangential frictional energy dissipation, constructing a first-type control condition; the viscoelastic internal variables are adjusted individually to eliminate the memory effect caused by material deformation accumulation, constructing a second-type control condition; and the contact boundary conditions and viscoelastic internal variables are adjusted simultaneously, eliminating both the tangential frictional energy dissipation and the memory effect caused by material deformation accumulation, constructing a third-type control condition.

[0049] The baseline, first-type, second-type, and third-type control conditions are all established from the same initial state copy and use the same driving commands, driving input energy, joint motion boundaries, analysis windows, and environmental parameters. The joint motion boundaries refer to the external connection constraints or driving end motion conditions commonly used in all conditions; they do not mean that all measured joint angles and joint angular velocities within the analysis window are used as forced inputs. For state variables included in the state prediction residuals, only their values ​​at the start of the analysis window are used to set the initial conditions; the measured values ​​at other times within the analysis window are only used as comparison data for the simulation results. State variables already used as forced motion boundary inputs are no longer included in the state prediction residuals.

[0050] Except for the contact boundary conditions or viscoelastic internal variables that are explicitly adjusted in various control cases, all other model parameters and boundary conditions remain consistent. The third type of control case performs both types of adjustments synchronously from the same initial state copy, and does not use the simulation final state of the first or second type of control case as the initial state.

[0051] For the independent operation of friction removal, a first type of control case is constructed. Specifically, the contact boundary conditions are analyzed to extract the normal contact state and tangential contact force of the contact interface. The contact boundary conditions include the correspondence between contact nodes on both sides of the contact interface, the normal direction of the contact surface, the normal gap between nodes, the normal contact force, and the tangential contact force. After each discrete time step or each nonlinear contact iteration, the contact state identifier corresponding to each contact node is read. When the dynamic solver of the digital twin model does not directly output the contact state identifier, the normal contact state is determined based on the normal gap between nodes and the normal contact force. When the normal gap between nodes does not exceed the preset contact tolerance and the normal contact force is greater than zero, the corresponding contact node is determined to be in a closed state; when the normal gap between nodes exceeds the preset contact tolerance, or the normal contact force is equal to zero, the corresponding contact node is determined to be in a separated state. For the contact node in the closed state, the tangential contact force output by the dynamic solver is read. When the dynamic solver only outputs the resultant contact force, the resultant contact force is decomposed into a normal component along the normal direction of the contact surface and a tangential component located in the tangential plane of the contact interface, based on the normal direction of the contact surface. The tangential component is then determined as the tangential contact force of the corresponding contact node. All contact nodes in the closed state are grouped into a set of active contact nodes, and this set of active contact nodes is used for solving the subsequent tangential constraint forces.

[0052] Subsequently, physical judgment and numerical intervention are performed. When the normal contact state is in a separated state, the tangential contact force is forcibly set to zero; when the normal contact state is in a closed state, under the physical condition that the tangential frictional dissipation energy is greater than or equal to zero, the tangential relative sliding velocity of all contact nodes in the contact interface is extracted, and the tangential constraint force to be solved is introduced at the contact node. Here, "all contact nodes" refers to all contact nodes in the closed state of the normal contact state. For contact nodes in the separated state of the normal contact state, only the tangential contact force is set to zero; the corresponding tangential relative sliding velocity is not extracted, nor is the tangential constraint force introduced or solved for the corresponding contact node. For contact nodes in the closed state of the normal contact state, the corresponding contact node is included in the set of active contact nodes, and the tangential constraint force is solved based on the corresponding tangential relative sliding velocity.

[0053] The tangential constraint force is also limited by the contact closure state and the upper limit of the tangential force that the contact interface can withstand. For contact nodes in the closed state, the direction of the candidate tangential constraint force is set to be opposite to the tangential slip tendency of the node when no candidate tangential constraint force is applied. The amplitude of the candidate tangential constraint force must not exceed the upper limit of frictional bearing capacity determined by the product of the friction parameter and the current normal contact force, and must not exceed the maximum tangential bearing capacity pre-calibrated according to human-machine interface material tests; the smaller of the two values ​​is taken as the upper limit of the tangential force of the contact node. For contact nodes in the separated state, the candidate tangential constraint force is fixed at zero. Any candidate solution should satisfy the condition that the tangential constraint force does not continuously input net mechanical energy into the relative slip process.

[0054] In each discrete time step or each nonlinear contact iteration, when the current set of active contact nodes remains unchanged, the feasible region of the tangential constraint is a convex set, and the Hessian matrix corresponding to the objective function remains positive throughout the entire feasible region, it is determined that the objective function has a unique minimum solution within the feasible region; a positive Hessian matrix at only a single iteration point is not sufficient as a basis for determining a globally unique solution.

[0055] If there is no tangential constraint force that satisfies the physical constraints, or if the tangential frictional dissipation energy after the solution is still higher than the preset elimination threshold, then the first marginal contribution value will not be calculated using this first type of control working condition.

[0056] To reduce the tangential frictional dissipation energy in the first type of control condition to below a preset elimination threshold, the tangential relative sliding velocities of all contact nodes in the contact interface are extracted, and tangential constraint forces to be solved are introduced at the contact nodes. An objective function is constructed, which minimizes the sum of squares of the tangential relative sliding velocities and simultaneously minimizes the sum of squares of the tangential constraint forces. The specific mathematical expression of this optimization process is as follows: in, This represents the scalar value of the objective function to be optimized. This represents the virtually introduced tangential constraint force vector to be solved; This indicates the pre-set speed penalty weight; 1 represents the total number of all contact nodes in the contact interface; l represents the serial number of the contact node in the contact interface. This represents the tangential relative slip velocity at the l-th contact node; This indicates the pre-set force penalty weight; This represents the tangential constraint force vector to be solved at the l-th contact node.

[0057] The initial value of the tangential relative sliding velocity is obtained from the motion data, but it is not a fixed value during the objective function solution process. For each set of candidate tangential constraint forces, the candidate tangential constraint forces are applied to the corresponding contact nodes, and the tangential relative sliding velocity of each contact node is recalculated using the contact dynamics equations. Therefore, the tangential relative sliding velocity is the response quantity of the tangential constraint force, and the objective function uses the tangential constraint force as the optimization variable.

[0058] The second derivative matrix of the objective function is the Hessian matrix of the objective function with respect to the tangential constraint force. Both the velocity penalty weight and the force penalty weight are positive values. The velocity penalty weight is set according to the reciprocal of the square of the tangential characteristic velocity, and the force penalty weight is set according to the reciprocal of the square of the tangential characteristic force, to eliminate the dimensional difference between the two objective terms. Under the premise of ensuring numerical stability, the velocity term has a higher priority than the constraint force term, ensuring that the tangential frictional dissipation energy first meets the preset elimination threshold, and then the solution with the smaller constraint force amplitude is selected from the candidate solutions that meet this threshold.

[0059] To obtain the optimal constraint force, the second derivative matrix of the objective function is calculated. Based on the positive definiteness of the second derivative matrix, if a unique solution for the tangential constraint force exists, the corresponding precise tangential constraint force is obtained using the objective function. Here, the second derivative matrix is ​​the Hessian matrix of the objective function, and its positive definiteness is one of the criteria for determining whether the objective function has a unique minimum. After obtaining the minimum solution, the precise tangential constraint force is added to the contact boundary conditions to suppress the relative sliding of the contact interface. By applying a reverse constraint force, the slip is forcibly approximated to zero at the finite element level, thereby reducing the tangential friction dissipation energy in the first type of control condition to below a preset elimination threshold, and is considered to have been eliminated.

[0060] To eliminate the non-destructive viscoelastic memory effect caused by material deformation accumulation before the start of the analysis window, a second type of control condition is constructed. Specifically, an initial physical model containing the initial conditions and macroscopic stress state is extracted, and the current total stress and current instantaneous tangential stiffness are extracted from the initial physical model. To eliminate the stress relaxation (memory effect) caused by historical deformation, it must be replaced by an ideal elastic state without damage. Therefore, the equivalent elastic prestress to be solved and the elastic tangential stiffness corresponding to the equivalent elastic prestress are introduced. A multi-dimensional physical state approximation equation is constructed, that is, the calculated difference between the current total stress and the equivalent elastic prestress is calculated to generate the total stress difference, and the calculated difference between the current instantaneous tangential stiffness and the elastic tangential stiffness is calculated to generate the instantaneous tangential stiffness difference. Nonlinear solution is performed, and under the premise of maintaining the macroscopic stress state unchanged, the equivalent elastic prestress that simultaneously minimizes the total stress difference and the instantaneous tangential stiffness difference is calculated. The specific calculation formula for the state approximation is as follows: in, This represents the constructed scalar residual value used to measure the degree of deviation between the virtual elastic state and the current time-dependent viscoelastic state; This represents the equivalent elastic prestress tensor to be solved; This represents the current total stress tensor extracted from the initial physical model; This represents the dimensional compatibility weighting coefficient between the stress and stiffness terms. This represents the current instantaneous tangent stiffness matrix extracted from the initial physical model; This represents the elastic tangent stiffness matrix corresponding to the equivalent elastic prestress tensor. Denotes the Frobenius norm of a tensor or matrix; Indicates a predetermined reference scale for normal stress; It represents a pre-determined positive stiffness reference scale.

[0061] Using the precise equivalent elastic prestress, the updated values ​​of the viscoelastic internal variables are calculated, and these updated values ​​are set to the viscoelastic internal variables. Taking the generalized Maxwell viscoelastic constitutive model as an example, the equivalent elastic prestress is used to absorb the total stress contribution generated by each viscoelastic branch at the current moment. For models using branch overstress as viscoelastic internal variables, the overstress of each viscoelastic branch is updated to zero; for models using viscous strain as viscoelastic internal variables, the viscous strain is updated to the equilibrium reference value corresponding to the current total strain, making the elastic strain difference of each viscoelastic branch zero. Thus, the current macroscopic stress state is maintained by the equivalent elastic prestress, while the non-damage-related internal variables used to record the historical loading process no longer retain their original historical memory. The update does not change the already fixed irreversible material damage state, structural geometry, boundary loads, and contact state.

[0062] After updating the viscoelastic internal variables, the recoverable material energy difference at the start of the analysis window before and after the update is also calculated. This recoverable material energy difference is not involved in solving the objective function of the equivalent elastic prestress, but is used to verify the initial energy boundary after projection. When the recoverable material energy difference does not exceed the preset projection energy tolerance, the initial energy boundary before and after the update is considered consistent; when it exceeds the projection energy tolerance, it is included in the energy balance as an initial energy correction term for the second and third type control conditions, or the corresponding control condition is marked as invalid. The additional initial energy generated by the projection operation must not be directly included in the effective auxiliary mechanical work.

[0063] When the total stress difference and instantaneous tangential stiffness difference obtained by the solution do not exceed the corresponding preset projection tolerance, and the updated initial physical model satisfies macroscopic force equilibrium, the second type of control case is determined to be valid; otherwise, the second type of control case is marked as invalid, and the second marginal contribution value and the third marginal contribution value are not used to calculate the case.

[0064] The update only applies to internal variables that record the history of non-damaging viscoelastic effects and does not change the fixed irreversible material damage state. The update does not disable the viscoelastic constitutive model, nor does it set the material as an ideal elastic material with no dissipation throughout the process. After the simulation of the second control case begins, each viscoelastic branch still re-evolves according to the same constitutive relationship with the current loading, and can generate hysteretic dissipation caused by the new loading process within the analysis window. Therefore, the second marginal contribution value is used to characterize the influence of the existing viscoelastic history memory at the start of the analysis window, and does not indicate that all material hysteretic dissipation has been eliminated.

[0065] The internal state variables used to record historical dissipation in the elastic state inversion model (such as the generalized Maxwell viscoelastic constitutive model) are forcibly overwritten. The third type of control case synchronously performs tangential slip restraint constraints and viscoelastic internal variable overwriting from the same initial state copy. The third type of control case is considered valid only if the following conditions are met simultaneously: the tangential constraint solution corresponding to the first type of control case is valid; the equivalent elastic state projection corresponding to the second type of control case is valid; the tangential friction dissipation energy recalculated by the third type of control case does not exceed a preset elimination threshold; the total stress difference, instantaneous tangential stiffness difference, and initial energy storage difference of the third type of control case meet the corresponding projection tolerances; and the energy balance residual of the third type of control case itself meets the numerical error requirements. The validity of the third type of control case cannot be directly determined solely based on the individual calculation results of the first and second type of control cases.

[0066] By reconstructing the boundary constraints and material constitutive state variables at the underlying level in a virtual simulation field, the originally complex black box of physical losses is cleverly decomposed into a set of gray boxes that can be independently observed and numerically quantified. This not only avoids the technical dead end of not being able to separate a single dissipation channel in actual physical experiments, but also endows the digital twin model with a multi-dimensional reference benchmark with orthogonal and independent characteristics, successfully opening up an executable path from macroscopic mixed dissipation to the reverse analysis of underlying decoupled parameters.

[0067] S3: Based on the simulation results, calculate the energy transfer efficiency and energy balance residuals of the baseline working condition and the control working condition respectively, and calculate the state prediction residuals in combination with the walking aid state data.

[0068] To map discrete torques and kinematic characteristics into a unified work efficiency index based on strict physical dimension transformation rules, and further to isolate energy loss boundaries caused by specific physical factors through differential operations between various operating conditions, this not only completes the dimensionality upgrade from physical-level mapping data to system-level performance parameters, but also serves as a target guide and computational benchmark for subsequent construction of a sensitivity matrix using multidimensional residuals and regularized iterative optimization.

[0069] From the dynamic simulation results of the baseline and control conditions, the joint torque in the drive-on state, the passive joint torque in the drive-off state, the joint angular velocity, and the total available energy in a single cycle are extracted within the preset assist time interval.

[0070] The passive joint torque in the drive-off state is obtained through passive benchmark simulation in the same state. For any working condition, except for the drive input being turned off, the structural geometry, joint motion trajectory, contact boundary conditions, material damage state, and viscoelastic state corresponding to that working condition are all consistent with those in the drive-on state in the drive-off state.

[0071] For each of the aforementioned driving air pressures corresponding to contact distribution forces, the position of each sensing unit relative to the knee joint rotation center and the direction of force on each sensing unit are determined. Based on the joint angle, the positions of each sensing unit and the contact distribution forces are transformed to a unified knee joint coordinate system. The torque generated by each contact distribution force about the knee joint rotation center is determined, and the component of the torque along the knee joint flexion-extension axis is extracted. The torque components corresponding to all sensing units are summed to obtain the measured auxiliary torque generated by the soft knee joint walker on the knee joint at the current moment.

[0072] Based on a unified timestamp, the measured auxiliary torque at each moment is correlated with the joint angle, joint angular velocity, and driving air pressure at the corresponding moment, and a correspondence between the joint angle and the measured auxiliary torque is established according to the driving air pressure. For multiple complete gait cycles under the same driving air pressure, under the condition that the joint movement direction and knee joint movement phase are the same, the measured auxiliary torque within the same preset joint angle range is statistically analyzed to obtain the auxiliary torque under different driving air pressures and different joint angles.

[0073] The current direction of knee joint movement is determined based on the joint angular velocity. When the direction of the measured auxiliary torque is consistent with the direction of knee joint movement within the preset assistance phase, the measured auxiliary torque is determined as a positive auxiliary torque; when the two directions are opposite, the measured auxiliary torque is determined as a negative torque. The measured auxiliary torque corresponding to each driving air pressure is calculated in chronological order to obtain the auxiliary torque variation curve under different driving air pressures. The measured auxiliary torque is used to verify the joint torque output by the digital twin model in the driving activation state and to evaluate the assisted walking effect of the soft knee joint walking aid.

[0074] Under the same driving air pressure, joint movement stage, and time sampling point, the measured auxiliary torque is compared with the joint torque output by the digital twin model in the driving-on state to obtain the joint torque deviation. The joint torque deviation is then dimensionless and used as the joint torque component in the state prediction residual for subsequent calculation of the parameter correction. Therefore, the measured auxiliary torque obtained from the contact distribution force conversion is not only used to output the assisted walking effect but also to verify and correct the digital twin model.

[0075] To ensure a strict conversion between dynamic parameters and energy dimensions, and to avoid the loss of physical meaning caused by simple scalar addition, it is necessary to introduce a time factor and dynamic integration of kinematic states. Therefore, by integrating the difference between the joint torque in the drive-on state and the passive joint torque in the drive-off state, combined with the joint angular velocity, the effective assist mechanical work is obtained.

[0076] The specific mathematical expression of this physical integration process is as follows: in, Indicates the first The effective assist mechanical work calculated under various working conditions; This represents the operating condition index value corresponding to the current operation; Indicates the start time of the preset assist time interval; Indicates the end time of the preset assist time interval; Indicates at time The extracted first Joint torque under various operating conditions with the drive in the open state; Indicates at time The extracted first Passive joint torque under the drive-off state in this operating condition; Indicates the current moment in the continuous data extraction; Indicates at time Extracted joint angular velocities; This represents a time integral infinitesimal.

[0077] The total available energy refers to the energy actually input into the drive system within the same cycle and capable of being used for power transfer. This includes the mechanical energy output by the driver or the equivalent input energy converted from the drive input, but excludes the active work done by the human body, changes in gravitational potential energy, and energy stored on the non-drive side before the start of the analysis window. If there is a difference between the elastic energy stored on the drive side at the start and end of the cycle, this difference is used to correct the drive input energy. When the elastic energy stored on the drive side increases at the end of the analysis window relative to the start time, the increase is deducted from the drive input energy; when the elastic energy stored on the drive side decreases, the decrease is included in the total available energy. The corrected total available energy should be greater than zero; otherwise, the corresponding operating condition is marked as invalid. The same energy statistical boundary is used for the baseline operating condition and various control operating conditions. For any operating condition, the ratio of the effective power transfer mechanical work to the total available energy is determined as the energy transfer efficiency of that operating condition.

[0078] Based on the joint angle, the knee joint movement stage corresponding to the preset assistance time interval is determined. Within the preset assistance time interval, the peak value, average value, and duration percentage of the positive assistance torque corresponding to each driving air pressure are calculated. The peak value, average value, and duration percentage of the positive assistance torque are compared with preset assistance index ranges. When all indicators are within the corresponding preset assistance index range, it is determined that the corresponding driving air pressure can provide effective walking assistance; when any indicator is not within the corresponding preset assistance index range, the corresponding result of insufficient assistance, assistance timing deviation, or reverse action is output.

[0079] The contact pressure distribution of the human-machine interface is determined based on the contact force distribution and effective sensing area of ​​each sensing unit. The maximum contact pressure, the dispersion of the contact pressure distribution, and the percentage of contact areas exceeding a preset pressure threshold are extracted from this distribution. The maximum contact pressure, the dispersion of the contact pressure distribution, and the percentage of contact areas are compared with corresponding preset comfort thresholds. When all indicators do not exceed the corresponding preset comfort threshold, the wearing comfort under the corresponding driving air pressure is deemed to meet the requirements. When any indicator exceeds the corresponding preset comfort threshold, the location of the sensing unit exceeding the preset comfort threshold is determined, and the corresponding local pressure location of the human-machine interface and the evaluation result indicating that the wearing comfort does not meet the requirements are output.

[0080] Based on the combined evaluation results of the walking assistance effect and wearing comfort corresponding to each driving air pressure, the driving air pressure that simultaneously meets the preset assistance index range and the preset comfort threshold is selected as the suitable driving air pressure for the soft knee joint walking aid.

[0081] When only one of the driving air pressures simultaneously meets both the preset assist index range and the preset comfort threshold, that driving air pressure is determined as the suitable driving air pressure. When multiple driving air pressures simultaneously meet both the preset assist index range and the preset comfort threshold, the driving air pressure with the highest energy transfer efficiency among those meeting the wearing comfort requirements is selected as the suitable driving air pressure; when the energy transfer efficiency difference between multiple driving air pressures does not exceed a preset efficiency difference, the driving air pressure with the lower value is selected as the suitable driving air pressure. When no driving air pressure simultaneously meets both the preset assist index range and the preset comfort threshold, a result indicating that no suitable driving air pressure has been obtained is output, and the assist index or comfort index that was not met for each driving air pressure is marked respectively.

[0082] After obtaining the energy transfer efficiency for all operating conditions, a differential calculation stage with multi-factor decoupling is performed to accurately quantify the specific proportion of each physical loss mechanism. The energy transfer efficiency of the benchmark operating condition is calibrated as the true energy transfer performance. To isolate and quantify the energy absorption effect of tangential friction, the energy transfer efficiency of the first type of control operating condition is subtracted from the true energy transfer performance to obtain a first marginal contribution value. To quantify the influence of the material's viscoelastic memory effect, the energy transfer efficiency of the second type of control operating condition is subtracted from the true energy transfer performance to obtain a second marginal contribution value. To capture and quantify the nonlinear interference coupling effect between friction and hysteresis, the energy transfer efficiency of the third type of control operating condition is used to subtract the true energy transfer performance, the first marginal contribution value, and the second marginal contribution value to obtain a third marginal contribution value.

[0083] The third marginal contribution value is determined using the two-factor difference principle. The change in energy transfer efficiency of the third control condition relative to the baseline condition represents the total efficiency change resulting from simultaneously eliminating tangential friction dissipation and material deformation cumulative memory effect; the first and second marginal contribution values ​​represent the efficiency changes resulting from individually eliminating tangential friction dissipation and individually eliminating material deformation cumulative memory effect, respectively. When the two factors are independent, the total efficiency change is equal to the sum of the first and second marginal contribution values; when the two factors are coupled, there is a difference between the total efficiency change and the sum of the two individual marginal contribution values. Therefore, subtracting the first and second marginal contribution values ​​from the total efficiency change of the third control condition relative to the baseline condition yields a residual amount that characterizes the non-additive interaction effect produced by the combined action of tangential friction dissipation and material deformation cumulative memory effect, and this residual amount is determined as the third marginal contribution value.

[0084] The difference between the first marginal contribution value, the second marginal contribution value, and the third marginal contribution value can be expressed by the following algebraic relationship: in, This represents the calculated first marginal contribution value; This indicates the energy transfer efficiency under the first type of control condition; It represents the actual transmission performance under the reference operating condition.

[0085] in, This represents the calculated second marginal contribution value; This indicates the energy transfer efficiency under the second control condition; It represents the actual transmission performance under the reference operating condition.

[0086] in, This represents the calculated third marginal contribution value; This indicates the energy transfer efficiency under the third control condition; Indicates the actual transmission performance under the reference operating condition; This represents the calculated first marginal contribution value; This represents the calculated second marginal contribution value.

[0087] When the first type of control condition is invalid, the first and third marginal contribution values ​​are not output; when the second type of control condition is invalid, the second and third marginal contribution values ​​are not output; when the third type of control condition is invalid, the third marginal contribution value is not output. Actual performance and other marginal contribution values ​​unaffected by invalid conditions continue to be output.

[0088] Energy balance residuals were calculated for the baseline and various control conditions. For any given condition, within the same analysis window, the following parameters were statistically analyzed: driving input energy, joint boundary mechanical work, system kinetic energy change, system potential energy change, soft structure and driving side energy storage change, normal damping dissipation energy, tangential friction dissipation energy, and material hysteresis dissipation energy. Using a unified notation convention where energy inflow to the system is positive and work done by the system and energy dissipation are negative, the energy terms were algebraically summed, and the resulting unclosed difference was used as the energy balance residual for that condition.

[0089] For the first and third control conditions, the net mechanical work generated by the tangential constraint force within the analysis window should also be calculated and included in the joint boundary mechanical work according to the actual energy flow direction. When the net mechanical work generated by the tangential constraint force exceeds the preset constraint work tolerance, the corresponding control condition will be marked as invalid and will not be used to calculate the marginal contribution value.

[0090] The increase in system kinetic energy, system potential energy, soft structure energy storage, and drive-side energy storage at the end of the analysis window relative to the start time is subtracted from the net input energy as energy retained within the system; when the corresponding state energy decreases, it is included in the energy balance with the opposite sign. All energy values ​​are algebraically summed according to the same analysis window and a uniform energy flow direction.

[0091] The material hysteresis dissipation energy is determined based on the dissipation stress and corresponding strain rate of each viscoelastic branch. Within the analysis window, the instantaneous dissipation power generated at each integration point of the soft material is integrated over time and accumulated across the material volume to obtain the material hysteresis dissipation energy for that operating condition. When the viscoelastic constitutive model can directly output the dissipation energy of each viscoelastic branch, the dissipation energy output by each branch is accumulated. The normal damping dissipation energy, tangential friction dissipation energy, and material hysteresis dissipation energy are all recalculated based on the dynamic simulation results of each operating condition, rather than directly using the initial dissipation determined by measured state data.

[0092] The state prediction residual is obtained by subtracting the corresponding walker state data from the simulated state variables such as joint angles, joint angular velocities, contact forces, and structural displacements under the baseline working condition, using a uniform time sampling point. State variables of different dimensions are dimensionless before being incorporated into the state prediction residual vector. The state prediction residual and the energy balance residual are used together to evaluate the state mapping accuracy and numerical energy closure of the digital twin model.

[0093] In the subsequent threshold determination, the L2 norm of the dimensionless state prediction residual vector is used to represent the magnitude of the state prediction residual, the absolute value of the energy balance residual is used to represent the degree of energy non-closure, and the L2 norm of the dimensionless parameter correction vector is used to represent the parameter correction magnitude.

[0094] S4: Calculate the parameter correction amount of the digital twin model based on the state prediction residual and the energy balance residual, update the digital twin model using the parameter correction amount and iteratively execute the dynamic simulation until the preset convergence condition is met, and output the performance analysis results of the soft knee joint walking aid.

[0095] Considering the simulation and differential calculations performed in the preceding steps, although the marginal contributions of boundary friction and material memory effect under various control conditions have been quantified, the parameters of the digital twin model on which it is based may still deviate from the actual physical system, leading to the accumulation of errors in long-term iterative calculations. Since this scheme includes strongly nonlinear contact slippage and large deformation hysteresis in soft materials, a single state feedback is prone to getting trapped in local minima and causing numerical oscillations. Therefore, it is necessary to introduce underlying energy conservation as a mandatory pre-constraint for correction. By establishing a hierarchical judgment between numerical stability and parameter optimization, a high-fidelity correction matrix is ​​solved, thereby continuously approximating the true contact stiffness and viscoelastic constitutive model in multiple iterations. This serves as the final core piece of the closed-loop calibration puzzle for the final output of reliable marginal contributions and performance evaluation results.

[0096] When the energy balance residual exceeds a preset numerical error threshold, the time step parameters and mesh parameters of the simulation calculation are corrected. After each correction of the time step parameters or mesh parameters, the dynamic simulation of the corresponding operating condition is re-executed, and the energy balance residual of that operating condition is recalculated. If, after a preset number of time step updates and mesh parameter corrections, the energy balance residual of the corresponding operating condition still does not meet the numerical error threshold, if the corresponding operating condition is a baseline operating condition, the current performance analysis is paused; if the corresponding operating condition is a control operating condition, the control operating condition is marked as invalid, and the marginal contribution value dependent on the control operating condition is masked. The physical parameter correction process only begins when the energy balance residuals of the baseline operating condition and all valid control operating conditions do not exceed the numerical error threshold.

[0097] The energy balance residual is used to determine whether the dynamic simulation results have numerical credibility, and the state prediction residual is used to solve the physical parameter corrections for the contact boundary conditions and viscoelastic internal variables. The two types of residuals jointly constrain the update of the digital twin model in the order of numerical energy closure first and then physical parameter calibration.

[0098] When the overall system energy exhibits significant non-conservation, any optimization of physical parameters will lead to directional errors due to distortion of the underlying data. Therefore, it is essential to prioritize reducing the time step or refining the mesh in the contact region. Assuming the energy base of the dynamic solver remains stable, a physical-level calibration phase is initiated, extracting the parameters to be corrected from the digital twin model. These parameters include the contact boundary conditions and the viscoelastic internal variables. The contact boundary conditions to be corrected refer to the contact parameters describing the actual human-machine contact relationship in the baseline conditions, excluding the tangential constraint force, tangential dissipation elimination threshold, and the effectiveness judgment threshold for the control conditions introduced to suppress slippage in the first and third control conditions.

[0099] The contact boundary conditions are parameterized in finite dimensions using at least one of the following: normal contact stiffness, tangential contact stiffness, friction parameters, and contact anchor point positions. The viscoelastic internal variables are arranged in finite dimensions using the internal stress, internal strain, or state parameters corresponding to the relaxation process of each viscoelastic branch, thereby forming a sequence of parameters to be corrected.

[0100] It is necessary to quantify the specific mapping weight of these underlying parameter deviations to the macroscopic observation state. Therefore, when the state prediction residual is greater than the preset residual threshold, the partial derivative of the state prediction residual with respect to the parameter to be corrected is calculated.

[0101] Apply a small positive or bidirectional perturbation of a preset amplitude to any parameter to be corrected, and re-execute the dynamic simulation of the reference working condition while keeping the other parameters to be corrected unchanged. Determine the corresponding partial derivative based on the ratio of the change in the state prediction residual before and after the perturbation to the parameter perturbation amount.

[0102] To integrate the multi-dimensional scattered derivatives into a system-level evolution direction guidance matrix, the partial derivatives are arranged according to the parameter dimension to construct a residual sensitivity matrix. Its internal arrangement and analytical calculation expressions for the elements are as follows: in, The first element in the constructed residual sensitivity matrix represents the... Line 1 The numerical values ​​of the column elements; Indicates the index of the state response component in the state prediction residual vector; The first element in the state prediction residual vector represents the... Each state response component; Indicates the first parameter in the sequence to be corrected. One physical variable; This indicates the index of the model parameter in the parameter sequence to be corrected.

[0103] By exhaustively differentiating all state residuals along the parameter dimension, a residual sensitivity matrix is ​​formed that represents the overall picture of the "state deviation-parameter perturbation" mapping of the digital twin model. The residual sensitivity matrix is ​​then checked for matrix rank, condition number, and column correlation. If the residual sensitivity matrix has full column rank for the parameter to be corrected, the condition number does not exceed a preset condition number threshold, and the absolute value of the correlation coefficient between any two columns does not exceed a preset correlation threshold, then the parameter to be corrected is determined to be uniquely identifiable; otherwise, an unidentifiable parameter result is output, and the current round of physical parameter updates is paused.

[0104] After obtaining the residual sensitivity matrix, when it is determined that the parameter to be corrected has a unique solution, a regularized optimization equation is constructed using the residual sensitivity matrix. By solving the regularized optimization equation, the parameter correction amount is obtained.

[0105] Assuming the parameter to be corrected is uniquely identifiable, to suppress the excessively large condition number of the sensitivity matrix and the amplified effect of measurement noise on the parameter solution, a regularization optimization strategy with additional damping is adopted. The corresponding equation for solving the parameter correction is as follows: in, This represents the column vector of parameter corrections containing multidimensional variables obtained from the solution; This represents the complete residual sensitivity matrix formed by the arrangement and combination of elements of the residual sensitivity matrix; This represents the transpose of the residual sensitivity matrix; This represents the pre-defined damping regularization coefficient; Representation and matrix A unit diagonal matrix with consistent dimensions; This represents the complete state prediction residual vector.

[0106] The state prediction residual is uniformly calculated by subtracting the corresponding walker state data from the baseline simulation state quantity. Therefore, the negative sign in the parameter correction expression indicates that the parameter to be corrected is updated in the direction of reducing the state prediction residual.

[0107] When the parameter correction amount changes the initial conditions of the digital twin model, the digital twin model is updated using the parameter correction amount, and the initial conditions are reset; when the parameter correction amount does not change the initial conditions of the digital twin model, the digital twin model is updated using the parameter correction amount. After completing the parameter update, using the updated baseline condition as the common physical baseline, the tangential constraint force of the first type of control condition is resolved, the equivalent elastic prestress and viscoelastic internal variable update values ​​of the second type of control condition are recalculated, and the third type of control condition is simultaneously reconstructed. Subsequently, the dynamic simulation of each condition is re-executed. The tangential constraint force, viscoelastic internal variable update values, or simulation final state of the previous round of control conditions must not be directly used.

[0108] The preset convergence conditions include: the energy balance residuals of the baseline operating condition and various effective control operating conditions do not exceed the numerical error threshold; the state prediction residual of the baseline operating condition does not exceed the residual threshold; and the parameter correction amounts obtained in two consecutive iterations do not exceed the preset parameter convergence threshold. When all the above conditions are met simultaneously, the iteration stops.

[0109] The performance analysis results include at least the actual transmission performance under the baseline condition, the first marginal contribution value, the second marginal contribution value, the third marginal contribution value, the energy balance residuals for each condition, the state prediction residuals under the baseline condition, parameter correction records, the measured auxiliary torque under different driving air pressures, the effect of assisted walking, the evaluation results of wearing comfort, and the appropriate driving air pressure.

[0110] The measured auxiliary torque under different driving air pressures includes at least the results of the auxiliary torque changing over time for each driving air pressure and the results of the auxiliary torque changing with joint angle; the assisted walking effect includes at least the peak value of the positive auxiliary torque, the average value of the positive auxiliary torque, the percentage of the duration of the positive auxiliary torque, and the judgment results of insufficient assistance, assistance timing deviation, or reverse action; the wearing comfort evaluation results include at least the maximum contact pressure, the dispersion of the contact pressure distribution, the percentage of the contact area exceeding the preset pressure threshold, and the local pressure position of the human-machine interface.

[0111] That is, the updated contact stiffness and other evolution parameters are directly applied under the existing initial values, and the closed-loop cycle continues until the residual state of the entire system is lower than the threshold. Finally, the performance analysis results such as energy transfer efficiency are output using the digital twin model that meets the preset convergence conditions.

[0112] By pre-setting energy conservation checks, non-physical property divergences caused by mesh distortion or excessively large integration steps are effectively filtered out, providing a clean mechanical data source for subsequent parameter optimization. Simultaneously, by using regularized optimization equations with damping factors combined with the sensitivity matrix to solve for the inverse correction, the ill-conditioned matrix inversion problem caused by the strong coupling of multiple parameters of friction and material hysteresis is overcome, ensuring that the iterative updates of parameters possess both stability and convergence. This fundamentally endows the digital twin model with the ability for dynamic adaptive and precise calibration, eliminating blind trial and error, and truly enabling it to output system-level performance analysis results that accurately reflect the physical realities of the operating conditions and the marginal energy loss mechanism.

[0113] This embodiment also provides a digital twin analysis system for the performance of soft knee joint walking aids, including: The data unit acquires simulation input data and walker status data of the soft knee joint walker, and uses the walker status data to set the initial conditions of the digital twin model.

[0114] The simulation unit, based on the initial conditions, adjusts the contact boundary conditions and viscoelastic internal variables of the digital twin model to form a baseline working condition and a control working condition, and performs dynamic simulations using the simulation input data to obtain simulation results.

[0115] The evaluation unit calculates the energy transfer efficiency and energy balance residuals of the baseline and control conditions based on the simulation results, and calculates the state prediction residuals by combining the walking aid state data.

[0116] The correction unit calculates the parameter correction amount of the digital twin model based on the state prediction residual and the energy balance residual, updates the digital twin model using the parameter correction amount, and iteratively executes the dynamic simulation until the preset convergence condition is met, and outputs the performance analysis results of the soft knee joint walking aid.

[0117] This embodiment also provides a computer device applicable to the digital twin analysis method for the performance of soft knee joint walking aids, including: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to implement the digital twin analysis method for the performance of soft knee joint walking aids as proposed in the above embodiment.

[0118] The computer device can be a terminal, comprising a processor, memory, communication interface, display screen, and input devices connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, carrier networks, NFC (Near Field Communication), or other technologies. The display screen can be an LCD screen or an e-ink screen. The input devices can be a touch layer covering the display screen, buttons, a trackball, or a touchpad on the computer device's casing, or an external keyboard, touchpad, or mouse.

[0119] This embodiment also provides a storage medium storing a computer program that, when executed by a processor, implements the digital twin analysis method for the performance of a soft knee joint walking aid as proposed in the above embodiments. The storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as Static Random Access Memory (SRAM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Erasable Programmable Read Only Memory (EPROM), Programmable Read-Only Memory (PROM), Read-Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.

[0120] In summary, this invention achieves precise decoupling and quantification of contact slippage and material hysteresis effects in soft knee joint walkers, as well as adaptive high-precision calibration of the model based on dual constraints of motion state and energy conservation. This solves the problems of ambiguity in loss mechanisms and numerical mismatch in traditional methods, significantly improving the fidelity and reliability of performance analysis. It also addresses the issues of fuzzy loss mechanisms and numerical mismatch in traditional methods by: acquiring simulation input and state data to set initial conditions for the digital twin model; adjusting contact boundary conditions and viscoelastic internal variables to construct benchmark and control conditions and performing dynamic simulations; calculating energy transfer efficiency and energy balance residuals based on simulation results; and extracting state prediction residuals based on state data; and calculating parameter corrections based on the dual residuals to drive the closed-loop iterative update of the digital twin model until convergence.

[0121] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A digital twin analysis method for the performance of a soft knee joint walking aid, characterized in that: This includes acquiring simulation input data and walker status data of a soft knee joint walker, and using the walker status data to set the initial conditions of the digital twin model; Based on the initial conditions, the contact boundary conditions and viscoelastic internal variables of the digital twin model are adjusted to form a baseline working condition and a control working condition. Dynamic simulations are then performed using the simulation input data to obtain simulation results. Adjusting the contact boundary conditions and viscoelastic internal variables of the digital twin model to form a baseline and control conditions includes retaining the original contact boundary conditions and viscoelastic internal variables in the baseline condition. The contact boundary conditions are individually adjusted to eliminate tangential frictional energy dissipation, thus constructing a first-class control condition. By individually adjusting the viscoelastic internal variables to eliminate the memory effect caused by material deformation accumulation, a second type of control condition is constructed. Simultaneously adjust the contact boundary conditions and the viscoelastic internal variables, while eliminating the energy dissipated by tangential friction and the memory effect caused by the accumulation of material deformation, and construct a third type of control condition; The construction of the first type of comparative working condition includes analyzing the contact boundary conditions and extracting the normal contact state and tangential contact force of the contact interface; When the normal contact state is a separation state, the tangential contact force is forcibly set to zero; When the normal contact state is closed, under the physical condition that the tangential friction dissipation energy is greater than or equal to zero, the tangential relative sliding velocity of all contact nodes in the contact interface is extracted, and the tangential constraint force to be solved is introduced at the contact node. Construct an objective function to minimize the sum of squares of the tangential relative slip velocities and simultaneously minimize the sum of squares of the tangential constraint forces; Calculate the second derivative matrix of the objective function, and when the positive definiteness of the second derivative matrix determines that the tangential constraint force has a unique solution, use the objective function to solve for the corresponding precise tangential constraint force. The precise tangential constraint force is added to the contact boundary conditions to suppress the relative sliding of the contact interface; The construction of the second type of control condition includes extracting an initial physical model containing the initial conditions and macroscopic stress state, and extracting the current total stress and the current instantaneous tangential stiffness from the initial physical model; Introduce the equivalent elastic prestress to be solved and the elastic tangential stiffness corresponding to the equivalent elastic prestress; Calculate the difference between the current total stress and the equivalent elastic prestress to generate the total stress difference; calculate the difference between the current instantaneous tangential stiffness and the elastic tangential stiffness to generate the instantaneous tangential stiffness difference. While maintaining the macroscopic stress state unchanged, calculate the equivalent elastic prestress in which the total stress difference and the instantaneous tangential stiffness difference simultaneously reach their minimum values; Using the precise equivalent elastic prestress, the updated value of the viscoelastic internal variable is calculated, and the viscoelastic internal variable is set as the updated value; Based on the simulation results, the energy transfer efficiency and energy balance residuals of the baseline working condition and the control working condition are calculated respectively, and the state prediction residuals are calculated in combination with the walking aid state data. The parameter correction amount of the digital twin model is calculated based on the state prediction residual and the energy balance residual. The digital twin model is updated using the parameter correction amount and the dynamic simulation is iteratively executed until the preset convergence condition is met. The performance analysis results of the soft knee joint walking aid are then output. Wherein, the state prediction residual is the deviation between the simulation result of the baseline working condition and the state data of the walking aid, and the energy balance residual is the energy conservation difference between the baseline working condition and the control working condition in the dynamic simulation.

2. The digital twin analysis method for the performance of soft knee joint walking aids as described in claim 1, characterized in that: Setting the initial conditions for the digital twin model using the walking aid status data includes extracting contact force data and motion data from the walking aid status data; Within a preset time period, the increment of material damage state is calculated based on the digital twin model. When the increment is less than a preset threshold, the material damage state parameter of the digital twin model is set to a fixed value. The dissipated energy at the contact interface is calculated using the contact force data and the motion data, and the dissipated energy is decomposed into normal damping dissipated energy and tangential friction dissipated energy. The motion data, the material damage state parameters set to fixed values, and the normal damping dissipation energy are used together as the initial conditions for the digital twin model.

3. The digital twin analysis method for the performance of soft knee joint walking aids as described in claim 2, characterized in that: Based on the simulation results, the energy transfer efficiency and energy balance residuals of the baseline working condition and the control working condition are calculated respectively. The state prediction residuals are calculated in combination with the walking aid state data. This includes extracting the joint torque in the drive-on state, the passive joint torque in the drive-off state, the joint angular velocity, and the total available energy in a single cycle from the dynamic simulation results of the baseline working condition and the control working condition within a preset assist time interval. By integrating the difference between the joint torque in the drive-on state and the passive joint torque in the drive-off state, based on the joint angular velocity, the effective assist mechanical work is obtained. The energy transfer efficiency is obtained by calculating the ratio of the effective mechanical work to the total available energy. The energy transfer efficiency under the benchmark operating condition is calibrated to the actual transfer performance. The first marginal contribution value is obtained by subtracting the actual transmission performance from the energy transfer efficiency of the first type of control condition. The second marginal contribution value is obtained by subtracting the actual transmission performance from the energy transfer efficiency of the second type of control condition. The third marginal contribution value is obtained by subtracting the actual transmission performance, the first marginal contribution value, and the second marginal contribution value from the energy transfer efficiency of the third type of control condition.

4. The digital twin analysis method for the performance of soft knee joint walking aids as described in claim 3, characterized in that: The parameter correction amount of the digital twin model is calculated based on the state prediction residual and the energy balance residual. The digital twin model is updated using the parameter correction amount and the dynamic simulation is iteratively executed until the preset convergence condition is met. The performance analysis results of the soft knee joint walking aid are output, including correcting the time step parameter and mesh parameter of the simulation calculation when the energy balance residual exceeds the preset numerical error threshold. Extract the parameters to be corrected from the digital twin model, the parameters to be corrected including the contact boundary conditions and the viscoelastic internal variables; When the state prediction residual is greater than a preset residual threshold, the partial derivative of the state prediction residual with respect to the parameter to be corrected is calculated. Arrange the partial derivatives according to the parameter dimension to construct a residual sensitivity matrix; When it is determined that the parameter to be corrected has a unique solution, a regularized optimization equation is constructed using the residual sensitivity matrix, and the parameter correction amount is obtained by solving the regularized optimization equation. When the parameter correction amount changes the initial conditions of the digital twin model, the digital twin model is updated using the parameter correction amount, and the initial conditions are reset. When the parameter correction amount does not change the initial conditions of the digital twin model, the digital twin model is updated using the parameter correction amount, and the dynamic simulation is performed again.

5. A digital twin analysis system for the performance of a soft knee joint walking aid, based on the digital twin analysis method for the performance of a soft knee joint walking aid as described in any one of claims 1 to 4, characterized in that: Includes a data unit for acquiring simulation input data and walker status data of a soft knee joint walker, and using the walker status data to set the initial conditions of the digital twin model; The simulation unit, based on the initial conditions, adjusts the contact boundary conditions and viscoelastic internal variables of the digital twin model to form a baseline working condition and a control working condition, and performs dynamic simulations using the simulation input data to obtain simulation results. The evaluation unit calculates the energy transfer efficiency and energy balance residuals of the baseline working condition and the control working condition respectively based on the simulation results, and calculates the state prediction residuals in combination with the walking aid state data. The correction unit calculates the parameter correction amount of the digital twin model based on the state prediction residual and the energy balance residual, updates the digital twin model using the parameter correction amount, and iteratively executes the dynamic simulation until the preset convergence condition is met, and outputs the performance analysis results of the soft knee joint walking aid.

6. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that: When the processor executes the computer program, it implements the steps of the digital twin analysis method for the performance of the soft knee joint walking aid as described in any one of claims 1 to 4.

7. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by the processor, it implements the steps of the digital twin analysis method for the performance of the soft knee joint walking aid as described in any one of claims 1 to 4.

Images