Functional electrical stimulation methods and systems

Abstract

This application relates to the fields of biomedical engineering and rehabilitation engineering technology, and discloses a functional electrical stimulation method and system based on physiological simulation and adaptive closed-loop feedback. The method includes: calculating an ideal muscle activation curve based on a target trajectory; performing spatiotemporal asynchronous co-coding to generate a biomimetic stimulation pulse sequence by simulating temporal asynchrony, spatial and intensity modulation, and multi-muscle synergistic timing; simultaneously applying stimulation while monitoring the actual kinematic state and muscle physiological state in real time using sensors; and dynamically adjusting the internal parameters of the spatiotemporal asynchronous co-coding (T-S-C coding) based on the evaluation results of trajectory error and fatigue index to form an adaptive closed loop. This application, by combining biomimetic coding with an intelligent closed loop, fundamentally solves the problems of motion stiffness, muscle fatigue, and poor control precision caused by traditional FES, achieving smooth, sustained, and precise personalized motor function rehabilitation.

Description

Technical Field

[0001] This application relates to the fields of biomedical engineering and rehabilitation engineering, and in particular to pulse sequence encoding and control technology for functional electrical stimulation (FES). Background Technology

[0002] Functional electrical stimulation (FES) is an important means of motor function rehabilitation for patients with neurological injuries. It applies electrical pulses to denervated muscles or nerves through electrodes, causing the muscles to contract and assisting patients in performing functional movements.

[0003] However, existing functional electrical stimulation systems face significant technical bottlenecks in practical applications.

[0004] First, the smoothness of movement is poor, resulting in stiff movements. Existing systems mostly employ stimulation strategies based on fixed parameters and synchronous triggering. This mode causes a large number of motor units within the muscle to be activated at the same time, producing a step-like force output. This results in a noticeable lack of smoothness in the joint movement trajectory, a lack of natural acceleration and deceleration transitions, and difficulty in completing tasks requiring fine control.

[0005] Secondly, muscles fatigue rapidly. Existing stimulation patterns often violate Henneman's size principle, preferentially activating high-threshold, easily fatigued fast-twitch muscle fibers rather than fatigue-resistant slow-twitch fibers. Simultaneously, this simultaneous activation deprives muscle fibers of rest and rotation opportunities, leading to fatigue occurring much faster than during natural exercise, severely limiting the duration and total training volume of a single rehabilitation session.

[0006] Third, the precision of multi-muscle coordinated control is insufficient. Many functional movements require precise timing coordination of agonist, synergist, and antagonist muscles. Existing systems often lack precise timing control of antagonist muscles, such as the timing of braking, which can easily lead to overshoot or swaying during the cessation of movement.

[0007] Furthermore, there is a lack of adaptive closed-loop adjustment capabilities. Most existing systems are open-loop, relying entirely on preset parameters and unable to sense and respond to the patient's dynamically changing physiological state (such as fatigue accumulation). Even if a few systems introduce simple closed-loop control, they can only adjust the overall "stimulus intensity" based on a single trajectory error signal, and cannot make targeted adjustments to anti-fatigue strategies, nor can they simultaneously ensure motion accuracy and meet anti-fatigue needs.

[0008] In summary, existing FES technology suffers from a series of technical problems, including stiff and unnatural movements, rapid muscle fatigue, and insufficient precision in multi-muscle coordinated control. Therefore, a novel functional electrical stimulation method is needed that can simulate the physiological characteristics of natural movement at the stimulation encoding level, while possessing real-time perception capabilities of the patient's movement and physiological states, and intelligently adjusting stimulation parameters based on the perception results. Summary of the Invention

[0009] The purpose of this application is to provide a functional electrical stimulation method and system to solve the problems mentioned in the background art.

[0010] This application discloses a functional electrical stimulation method, including:

[0011] (a) Plan the target motion task and obtain the target motion trajectory;

[0012] (b) Based on the target motion trajectory, calculate the ideal muscle activation curves of the multiple muscles required to achieve the target motion trajectory;

[0013] (c) Based on the ideal muscle activation curve, perform spatiotemporal asynchronous co-coding to generate a multi-channel stimulation pulse sequence, wherein the spatiotemporal asynchronous co-coding simulates at least one or more of temporal asynchrony, spatial and intensity modulation, or multi-muscle co-sequencing.

[0014] (d) Apply the multi-channel stimulation pulse sequence and simultaneously monitor the actual kinematic state and / or muscle physiological state in real time using sensors;

[0015] (e) Calculate the trajectory error based on the actual kinematic state and the target motion trajectory, and / or assess the muscle physiological state;

[0016] (f) Based on the assessment results of the trajectory error and / or the muscle physiological state, dynamically adjust the internal parameters of the spatiotemporal asynchronous co-coding to update the multichannel stimulation pulse sequence;

[0017] (g) Repeat steps (d) to (f) until the target motion trajectory is completed.

[0018] In a preferred embodiment, the time asynchronicity is achieved by setting a preset phase difference for different stimulation channels in the multi-channel stimulation pulse sequence. Or introduce random jitter or To achieve this, wherein the random jitter satisfies:

[0019]

[0020] Used to ensure asynchronous pulse delivery.

[0021] In a preferred embodiment, the channel The The pulse duration is determined by the following formula:

[0022]

[0023] in, Basic stimulation frequency, For channel The fixed phase difference This represents the time jitter.

[0024] In a preferred embodiment, the spatial and intensity modulation simulation of muscle fiber size principles includes:

[0025] When the ideal muscle activation curve When at low activation levels, low-intensity parameter combinations, including narrow pulse widths, should be used preferentially. ;

[0026] When the ideal muscle activation curve When at a high activation level, the pulse frequency is increased synergistically. Pulse width and / or amplitude ;

[0027] in, Indicates time A curve showing the ideal muscle activation level.

[0028] In a preferred example, when When using:

[0029]

[0030]

[0031] when When using:

[0032]

[0033]

[0034] in, The normalized ideal muscle activation curve has a value range of 0 to 1;

[0035] It has a narrow pulse width and is used to preferentially activate slow-twitch muscle fibers;

[0036] It features a wide pulse width, used to recruit fast-twitch muscle fibers;

[0037] Low stimulation frequency;

[0038] It is a high stimulation frequency.

[0039] In a preferred embodiment, the spatial and intensity modulation further includes: when the ideal muscle activation curve When the activation level is low, spatial rotation stimulation is performed between multiple electrode locations to spatially disperse muscle fatigue.

[0040] In a preferred embodiment, the multi-muscle synergistic timing includes: before the agonist in the multiple muscles ceases contraction, based on braking timing parameters. At any moment Pre-activation of antagonistic muscles provides braking torque and prevents overshoot, including:

[0041]

[0042] in, Indicates the activation moment of the antagonistic muscle. Indicates the moment of cessation of the agonist muscle. Indicates the braking advance time.

[0043] In a preferred embodiment, the braking advance time It takes 20 to 50 milliseconds.

[0044] In a preferred embodiment, the multichannel stimulation pulse sequence includes The pulse parameters of each stimulation channel, among which , For the number of muscles, each channel The pulse parameters are expressed as:

[0045]

[0046] in, The amplitude of the pulse. The pulse width. The pulse frequency, This represents the phase difference.

[0047] In a preferred embodiment, the actual kinematic state is monitored by an inertial measurement unit (IMU) deployed in the limb segments, the IMU comprising an accelerometer and a gyroscope.

[0048] In a preferred embodiment, the muscle physiological state is the muscle fatigue index. ;

[0049] The assessment of the muscle physiological state in step (e) includes calculating the muscle fatigue index by analyzing the frequency or time domain characteristics of the surface electromyography (sEMG) signal. The frequency domain features include the median frequency. or average power frequency The downward trend of the time-domain feature includes the change in the root mean square (RMS) value.

[0050] in, Indicates time The changing muscle fatigue index.

[0051] In a preferred embodiment, the monitoring of the surface electromyography (SEMG) signal is synchronized with the application of the multi-channel stimulation pulse sequence. The interference of stimulation artifacts on the SEMG signal is removed by performing SEMG sampling during the pulse intervals between stimulation pulses, or by using a blanking algorithm or a sample-and-hold algorithm.

[0052] In a preferred embodiment, the dynamic adjustment of step (f) specifically includes: when the muscle fatigue index is detected... Exceeding the fatigue threshold At the same time, the internal parameters of the spatiotemporal asynchronous co-coding are automatically adjusted to execute an anti-fatigue strategy, which includes: increasing the number of spatially rotated electrodes while keeping the total torque constant, and / or increasing the random jitter range of the temporal asynchrony. .

[0053] In a preferred embodiment, the dynamic adjustment of step (f) specifically includes: when the trajectory error is detected... Exceeding the error threshold At that time, the ideal muscle activation curve is automatically corrected. And / or adjust the parameters of the multi-muscle synergistic timing;

[0054] The formula for calculating trajectory error is:

[0055]

[0056] in, Indicates time The error in the changing trajectory. This represents the trajectory of the target. This represents the actual kinematic state.

[0057] In a preferred embodiment, the method further includes: after the target motion trajectory is executed, storing the actual kinematic state, muscle physiological state, and internal parameters adjusted in step (f) monitored in step (d); and iteratively optimizing the biomechanical model for step (b) and / or the spatiotemporal asynchronous cooperative coding parameters for step (c) based on the stored data.

[0058] In a preferred embodiment, the solution in step (b) is performed based on a biomechanical model through inverse dynamics calculation, wherein the biomechanical model is a Hill-type muscle model or a musculoskeletal model.

[0059] In a preferred embodiment, the pulse width in the multi-channel stimulation pulse sequence The range is from 50 μs to 500 μs, where The duration is 50 to 150 μs to preferentially activate slow-twitch muscle fibers. For longer than 250 μs, fast muscle fibers are recruited.

[0060] In a preferred embodiment, the stimulation frequency in the multi-channel stimulation pulse sequence The range is from 15Hz to 60Hz, with the lower frequencies of 20 to 35Hz used for fatigue resistance.

[0061] In a preferred embodiment, the random jitter in the time asynchronicity The range is 0 to 10 ms, where:

[0062]

[0063] Indicates synchronous stimulation;

[0064]

[0065] This indicates asynchronous stimulation and adjusts according to muscle fatigue levels. Size.

[0066] In a preferred embodiment, the fatigue threshold Median frequency The value decreased by 15% to 30% relative to the initial value.

[0067] In a preferred embodiment, the multiple muscles include agonist muscles, synergist muscles, and antagonist muscles.

[0068] This application also discloses a functional electrical stimulation system, comprising:

[0069] Multichannel stimulator: configured to apply a multichannel stimulation pulse sequence;

[0070] Sensor module: includes an inertial measurement unit and surface electromyography electrodes, configured to monitor the patient's actual kinematic state and muscle physiological state in real time;

[0071] Processor: Configured to execute any of the methods described above.

[0072] This application also discloses a functional electrical stimulation system, comprising:

[0073] The trajectory planning module is configured to plan the target motion task and obtain the target motion trajectory.

[0074] Activate the calculation module and configure it to calculate the ideal muscle activation curves of multiple muscles required to achieve the target motion trajectory based on the target motion trajectory;

[0075] The spatiotemporal asynchronous co-coding module is configured to perform spatiotemporal asynchronous co-coding to generate a multi-channel stimulation pulse sequence based on the ideal muscle activation curve, wherein the spatiotemporal asynchronous co-coding simulates at least one or more of time asynchrony, spatial and intensity modulation, or multi-muscle co-sequencing.

[0076] A multichannel stimulator configured to apply the multichannel stimulation pulse sequence;

[0077] The sensor module is configured to monitor the actual kinematic state and / or muscle physiological state in real time while the multi-channel stimulation pulse sequence is applied;

[0078] The state assessment module is configured to calculate the trajectory error based on the actual kinematic state and the target motion trajectory, and / or assess the muscle physiological state;

[0079] An adaptive adjustment module is configured to dynamically adjust the internal parameters of the spatiotemporal asynchronous co-coding module based on the assessment results of the trajectory error and / or the muscle physiological state, in order to update the multi-channel stimulation pulse sequence; and

[0080] The control module is configured to control the multi-channel stimulator, the sensor module, the state assessment module, and the adaptive adjustment module to repeatedly perform stimulation, monitoring, assessment, and adjustment until the target motion trajectory is completed.

[0081] The functional electrical stimulation method of this application systematically solves the core problems of traditional functional electrical stimulation technology, such as motor stiffness, rapid muscle fatigue, and low control precision, by organically combining spatiotemporal asynchronous collaborative coding, multi-source sensing monitoring, and adaptive closed-loop feedback, and achieves high-quality, low-fatigue, and high-precision motor function rehabilitation.

[0082] First, this application fundamentally improves the smoothness and naturalness of movement through spatiotemporal asynchronous co-coding. Traditional functional electrical stimulation systems employ synchronous stimulation, where multiple electrodes simultaneously deliver stimulation pulses to the same muscle, activating all motor units within that muscle at the same moment. This results in an "all-or-nothing" step force output, leading to stiff and discontinuous joint movement trajectories. The temporal asynchronicity of this application is achieved by setting preset phase differences for different stimulation channels or introducing random jitter, staggering the pulse delivery times of multiple stimulation channels. This asynchronous delivery mechanism ensures that some motor units are always in a contracted state within any minute time window, while on a macroscopic time scale, the total muscle force output exhibits a smooth and continuous change. Compared to traditional synchronous stimulation, the method in this application significantly reduces the abruptness of joint movement trajectories, resulting in a smoother and more natural movement perceived by the patient, closely resembling normal voluntary movement patterns. By setting fixed phase differences for the channels and using randomly distributed jitter, the system achieves a balance between microscopic asynchronicity and macroscopic smoothness while maintaining controllable overall stimulation intensity.

[0083] Secondly, this application significantly delays the onset of muscle fatigue through spatial and intensity modulation. Traditional functional electrical stimulation often uses fixed high-intensity stimulation parameters. This mode violates the Hunemann size principle, tending to preferentially activate high-threshold, easily fatigued fast-twitch muscle fibers, while low-threshold, fatigue-resistant slow-twitch muscle fibers are not fully utilized, leading to significant muscle fatigue in patients within a short period. The spatial and intensity modulation of this application simulates the muscle fiber size principle. When the ideal muscle activation curve is at a low activation level, the system preferentially uses low-intensity parameter combinations (such as narrow pulse width and low frequency). These parameters have high spatial selectivity and can preferentially activate superficial, low-threshold slow-twitch muscle fibers. When the activation level is high, the system synergistically increases the pulse frequency, pulse width, and amplitude to gradually recruit deep fast-twitch muscle fibers. This graded recruitment strategy makes the muscle force output conform to physiological laws, significantly reducing energy consumption and fatigue accumulation. In addition, at low activation levels, the system performs spatial rotation stimulation between multiple electrode locations, allowing sufficient rest time for the muscle area under each electrode, effectively dispersing fatigue spatially. Compared to traditional methods, the method described in this application can significantly extend the time it takes for muscles to reach their fatigue threshold while producing the same torque output. This means that patients can undergo rehabilitation training for longer periods, and the effectiveness of each training session and its cumulative effect are greatly improved.

[0084] Third, this application achieves high-precision execution of complex movements through multi-muscle synergistic timing control. Many functional movements require not only the contraction of agonist muscles but also the timely braking or stabilizing effect of antagonist muscles. Traditional functional electrical stimulation systems typically only stimulate agonist muscles, neglecting the synergistic effect of antagonist muscles, resulting in a lack of braking torque during the stopping phase of the movement, which can easily lead to overshoot, oscillation, or instability. This application precisely controls the stimulation initiation time, duration, and activation sequence of agonist, synergist, and antagonist muscles through multi-muscle synergistic timing control. In particular, the system activates the antagonist muscle in advance according to braking timing parameters before the agonist muscle is about to stop contracting, so that the torque generated by the antagonist muscle matches the gradually weakening torque of the agonist muscle, achieving smooth deceleration and precise positioning of the joint. The braking advance time can be flexibly set according to the movement speed and inertial characteristics, ensuring that the movement stops accurately at the target position without overshoot. This precise timing synergy between agonist and antagonist muscles not only improves the accuracy of individual movements but also provides a technical basis for complex movements that require fine coordination of multiple muscles (such as finger grasping and wrist pronation and supination).

[0085] More importantly, this application enables the aforementioned spatiotemporal asynchronous co-coding strategy to be dynamically optimized and always maintain its optimal state through real-time sensing monitoring and adaptive closed-loop feedback. Traditional functional electrical stimulation systems are either open-loop, relying entirely on preset parameters and unable to cope with changes in the patient's physiological state, or they only use simple proportional-integral-derivative controllers to adjust a single feedback signal, resulting in limited regulatory capabilities. This application monitors the actual kinematic state in real time through an inertial measurement unit and monitors the muscle physiological state, especially fatigue state, in real time through surface electromyography, providing the system with comprehensive state perception capabilities. The system compares the actual motion trajectory with the target trajectory to calculate the trajectory error and assesses the muscle fatigue index; these two types of information are used to dynamically adjust the internal parameters of the spatiotemporal asynchronous co-coding.

[0086] When the detected trajectory error exceeds a preset threshold, the system automatically corrects the ideal muscle activation curve or adjusts parameters of the multi-muscle synergy timing (such as braking timing) to reduce the error in the next cycle or at the next moment, achieving precise trajectory tracking. This iterative correction mechanism based on error feedback can compensate for the deviation between the biomechanical model and the patient's actual physiological characteristics, enabling the system to achieve high-precision control through closed-loop learning even when the model is not completely accurate. Compared to open-loop functional electrical stimulation, the trajectory tracking error of the method in this application is significantly reduced.

[0087] When the muscle fatigue index exceeds the fatigue threshold, the system immediately activates an anti-fatigue strategy. While maintaining a relatively constant total torque output, it increases the number of spatially rotated electrodes and the range of temporally asynchronous random jitter, further dispersing the fatigue burden spatially and temporally. This dynamic adjustment mechanism based on fatigue feedback ensures that even during prolonged training, the system can intelligently adjust its stimulation strategy to combat fatigue accumulation and maintain exercise quality. Traditional open-loop bionic coding may produce good results in the initial stages, but once the patient's physiological state (such as fatigue level) changes, the preset coding parameters become ineffective. This application fundamentally solves the robustness problem of bionic coding in practical applications through closed-loop adaptation.

[0088] Furthermore, the three dimensions of spatiotemporal asynchronous co-coding (temporal asynchronicity, spatial and intensity modulation, and multi-muscle co-coordination timing) do not act in isolation, but rather work synergistically to produce a cumulative effect. Temporal asynchronicity ensures smoothness at the microscopic level, spatial and intensity modulation optimizes fatigue distribution at the mesoscopic level through size principles and spatial rotation, and multi-muscle co-coordination timing achieves precise coordination of multiple joints and muscles at the macroscopic level. These three dimensions work together in the generation process of the stimulus pulse sequence, resulting in optimal smoothness, fatigue resistance, and accuracy in the final stimulus output. Adaptive closed-loop feedback then adjusts the parameters of these three dimensions, adjusting the temporal jitter range, spatial rotation strategy, intensity modulation parameters, and co-coordination timing parameters based on real-time monitored motion errors and fatigue states, achieving "structured intelligent adjustment" rather than the "black-box adjustment" of traditional controllers. This structured adjustment method not only has better regulatory effects but also better interpretability and clinical acceptability.

[0089] This method also possesses long-term learning and personalized optimization capabilities. By storing the actual kinematic state, muscle physiological state, and adjusted internal parameters during each training session, the system can iteratively optimize the biomechanical model and spatiotemporal asynchronous co-coding parameters based on this historical data. As patients undergo multiple training sessions, the system gains a deeper understanding of the patient's specific physiological characteristics, and the stimulation strategy becomes increasingly precise, ultimately achieving truly personalized rehabilitation. This long-term learning capability enables this method not only to address short-term physiological state changes during a single training session through real-time closed-loop processing, but also to adapt to the recovery and changes in the patient's motor abilities during long-term rehabilitation through model iteration. Since there are significant differences in muscle strength, fatigue characteristics, and neuromuscular response characteristics among different patients, the personalized mechanism of this application ensures that each patient receives the stimulation program most suitable for their individual characteristics, avoiding the poor effects or side effects of a "one-size-fits-all" approach.

[0090] In summary, the method of this application systematically improves functional electrical stimulation technology from three levels: stimulation coding strategy (spatiotemporal asynchronous collaborative coding), state perception capability (multi-source sensor monitoring), and intelligent adjustment mechanism (adaptive closed-loop feedback and iterative optimization). This significantly enhances its motor smoothness, fatigue resistance, and control precision, while also providing personalization and adaptive capabilities. This offers a more effective, comfortable, and precise means of motor function rehabilitation for patients with neurological injuries such as stroke and spinal cord injury.

[0091] The specification of this application contains numerous technical features distributed across various technical solutions. Listing all possible combinations of these technical features (i.e., technical solutions) would make the specification excessively lengthy. To avoid this problem, the various technical features disclosed in the above-described invention, the various technical features disclosed in the following embodiments and examples, and the various technical features disclosed in the accompanying drawings can be freely combined to form various new technical solutions (all of which are considered to have been described in this specification), unless such a combination of technical features is technically infeasible. For example, one example discloses feature A+B+C, and another example discloses feature A+B+D+E. Features C and D are equivalent technical means that serve the same function, and technically only one needs to be used; they cannot be used simultaneously. Feature E can technically be combined with feature C. Therefore, the solution A+B+C+D should not be considered as described because it is technically infeasible, while the solution A+B+C+E should be considered as described. Attached Figure Description

[0092] Figure 1 This is a schematic flowchart of a functional electrical stimulation method according to an embodiment of this application.

[0093] Figure 2 This is a schematic diagram of the structure of a functional electrical stimulation system according to an embodiment of this application. Detailed Implementation

[0094] In the following description, many technical details are presented to help the reader better understand this application. However, those skilled in the art will understand that the technical solutions claimed in this application can be implemented even without these technical details and various variations and modifications based on the following embodiments.

[0095] Explanation of some concepts:

[0096] Functional Electrical Stimulation (FES) is a rehabilitation technique that uses surface electrodes or implanted electrodes to apply specific patterns of electrical pulses to stimulate denervated muscles or nerves, thereby assisting or replacing patients in performing specific functional movements. FES is primarily used for motor function rehabilitation in patients with neurological injuries such as stroke, spinal cord injury, and cerebral palsy.

[0097] Spatio-Temporal Asynchronous Coordination (TSC) is a stimulus pulse sequence encoding strategy proposed in this application, which includes three dimensions: temporal asynchrony (T), spatial and intensity modulation (S), and multi-muscle coordination timing (C). This encoding strategy generates a more natural and less fatigue-inducing stimulus pattern by simulating the physiological characteristics of descending control signals in the central nervous system.

[0098] Motor Unit: A motor unit (MU) is the basic functional unit of the neuromuscular system, consisting of an α motor neuron and all the muscle fibers it innervates. When a motor neuron is excited, all the muscle fibers it innervates contract simultaneously. The force output of a muscle is generated by the coordinated activation of multiple motor units.

[0099] Henneman's Size Principle: A fundamental principle in neurophysiology, Henneman's Size Principle states that in natural movement, the recruitment order of motor units follows a progression from smallest to largest. When less muscle force is required, the central nervous system preferentially activates low-threshold, slow-contraction motor units. These units generate little force but are extremely fatigue-resistant. As the required force increases, high-threshold, fast-contraction motor units are gradually recruited. These units generate greater force but are more prone to fatigue.

[0100] Agonistous, antagonist, and synergist muscles: In a specific movement, the agonist muscle is the muscle primarily responsible for generating the torque required for that movement; the antagonist muscle is the muscle that generates torque in the opposite direction to the agonist muscle, playing a braking and stabilizing role in motor control; and the synergist muscle is the muscle that works in conjunction with the agonist muscle to generate torque in the same direction. For example, in elbow flexion, the biceps brachii is the agonist muscle, the triceps brachii is the antagonist muscle, and the brachialis is the synergist muscle.

[0101] Ideal muscle activation curve: The ideal muscle activation curve refers to the ideal activation level of each muscle over time, calculated through inverse dynamics, to achieve a specific target movement trajectory. This activation level is a normalized value ranging from 0 (completely inactive) to 1 (fully activated), describing the degree of activation the muscle needs to achieve at each moment.

[0102] Target movement trajectory: The target movement trajectory refers to the ideal movement path that the patient's limb joints should follow, set according to the needs of the rehabilitation task. It is usually expressed as a function of the joint angle changing over time. The planning of the target movement trajectory needs to consider the smoothness of the movement, physiological rationality, and rehabilitation goals.

[0103] Trajectory error: Trajectory error refers to the deviation between the patient's actual joint angle and the target joint angle during functional electrical stimulation. Trajectory error reflects the precision of stimulation control and is an important basis for adaptive adjustment.

[0104] Muscle Fatigue Index (FI): The muscle fatigue index (FI) is an indicator used to quantify the degree of muscle fatigue. This application calculates the muscle fatigue index by analyzing the frequency domain characteristics (such as the decreasing trend of median frequency or average power frequency) or time domain characteristics (such as changes in root mean square value) of surface electromyography signals. The higher the index, the more severe the muscle fatigue.

[0105] Surface electromyography (sEMG) is the acquisition of electrical signals generated during muscle contraction using electrodes attached to the skin surface. sEMG signals contain rich information about muscle activity and can be used to assess physiological parameters such as muscle activation and fatigue levels.

[0106] Median frequency: The median frequency is a frequency domain characteristic parameter of surface electromyography (EMG) signals, referring to the frequency value that divides the signal power spectral density into two parts with equal energy. Studies have shown that as muscle fatigue deepens, the spectrum of surface EMG signals shifts towards lower frequencies, and the median frequency shows a decreasing trend. Therefore, the relative decrease in median frequency can be used as a quantitative indicator of muscle fatigue.

[0107] Inertial Measurement Unit (IMU): An inertial measurement unit (IMU) is a sensor module that integrates an accelerometer and a gyroscope, capable of measuring the linear acceleration and angular velocity of an object. In this application, the IMU is deployed in the patient's limb segments to measure joint angles and angular velocities in real time, thereby acquiring the actual kinematic state.

[0108] Inverse dynamics calculation: Inverse dynamics calculation refers to the process of calculating the muscle torque and activation level required for a given desired motion trajectory by using a biomechanical model and dynamic equations to solve in reverse. This is in contrast to forward dynamics calculation (which calculates the motion output given muscle activation).

[0109] Biomechanical models: Biomechanical models are mathematical models that describe the mechanical properties of the human musculoskeletal system, including the force-length relationship and force-velocity relationship of muscles, the elastic properties of tendons, and the geometric constraints of joints. Commonly used biomechanical models include Hill-type muscle models and complex musculoskeletal models built on software platforms (such as OpenSim).

[0110] Stimulation pulse sequence: A stimulation pulse sequence refers to a train of electrical pulses output by a functional electrical stimulation system, organized according to a specific timing and parameters. The parameters of each pulse include pulse amplitude, pulse width, pulse frequency, and phase difference. This application generates a multi-channel stimulation pulse sequence using a spatiotemporal asynchronous cooperative coding strategy.

[0111] Pulse width, frequency, amplitude, and phase difference: Pulse width refers to the duration of a single electrical stimulation pulse, usually measured in microseconds. Pulse width affects the spatial selectivity of stimulation and the type of muscle fibers activated. Pulse frequency refers to the number of pulses fired per unit time, usually measured in Hertz. Frequency affects the degree of fusion of muscle contraction and force output. Pulse amplitude refers to the current intensity of the pulse, usually measured in milliamperes. Amplitude affects the effectiveness of stimulation and the activation range. Phase difference refers to the relative offset of pulse firing times between different stimulation channels, used to achieve temporal asynchrony.

[0112] Slow-twitch and fast-twitch muscle fibers: Muscle fibers can be classified into different types based on their contraction speed and fatigue characteristics. Slow-twitch muscle fibers (Type I, S type) have slow contraction speed and low strength but are extremely fatigue-resistant, mainly used for maintaining posture and low-intensity endurance activities; fast-twitch muscle fibers (Type II) have fast contraction speed and high strength but are easily fatigued, with Type IIa (FR type) having moderate fatigue resistance and Type IIb (FF type) generating the greatest strength but being the most easily fatigued. This application delays fatigue by preferentially activating slow-twitch muscle fibers through adjusting stimulation parameters.

[0113] Stimulation artifacts: Stimulation artifacts refer to large-amplitude interference signals generated by the stimulating current on the surface electromyography (EMG) acquisition electrodes during functional electrical stimulation (FES). Stimulation artifacts can mask the true EMG signals, affecting the accurate assessment of muscle physiological state. This application removes the interference of stimulation artifacts through a synchronous sampling strategy or signal processing algorithm.

[0114] Braking timing parameters: In motion control, braking timing parameters refer to the amount of time required to activate the antagonist muscle in advance to achieve a smooth stop. Setting this parameter requires consideration of motion speed, inertial characteristics, and the strength characteristics of the agonist and antagonist muscles to ensure that the movement stops accurately at the target position without overshoot or undershoot.

[0115] The following is a brief summary of some of the innovative aspects of this application:

[0116] In summary, the fundamental inventiveness of this application lies in constructing a non-obvious dual-simulation closed-loop architecture. This architecture achieves an organic unity of deep-level biomimetic simulation and real-time dynamic optimization of the motor control mechanism of the central nervous system through a structured parameter feedback path formed between the front-end spatiotemporal asynchronous co-coding module and the back-end adaptive adjustment module.

[0117] Specifically, in upper or lower limb rehabilitation training scenarios for stroke or spinal cord injury patients, traditional technical approaches face a fundamental dilemma in their technical design: While an open-loop biomimetic coding strategy (i.e., implementing only one or more of time asynchrony (T), spatial and intensity modulation (S), or multi-muscle synergistic temporal modulation (C), without closed-loop adjustment) can produce a superior initial stimulation effect under ideal conditions where the patient's physiological state is stable and perfectly matches the model assumptions, this strategy quickly fails due to the lack of a self-correction mechanism once the patient's actual neuromuscular response characteristics deviate from the preset model parameters, or when muscle fatigue dynamically evolves during training. This leads to discrepancies in the actual movement trajectory. (Collected via sensors) and the target's motion trajectory Unacceptable deviations occur between (generated by the planning module) and the muscle fatigue index, while... (Calculated based on the frequency or time domain characteristics of surface electromyography (sEMG) signals) Accelerated accumulation; if a traditional closed-loop feedback control strategy (e.g., a proportional-integral-derivative PID controller) is used, although it can be based on trajectory error... (Formula 1) makes some degree of error correction, but this correction is achieved by adjusting a uniform, unstructured "total stimulus" or "total activation level". This "black box" adjustment cannot specifically target the physiological mechanisms of muscle recruitment (size principle), the temporal distribution characteristics of motor unit release (asynchronousness), and the synergistic temporal relationship between multiple muscles (synergy). Therefore, it cannot fundamentally solve the problems of motion stiffness and rapid fatigue caused by traditional synchronous stimulation, nor can it achieve high-precision complex movement control that requires precise temporal coordination between agonist and antagonist muscles.

[0118] This application overcomes the aforementioned difficulties through an unconventional technical concept: its core lies in the deep coupling of the "structured characteristics of biomimetic coding" and the "adaptive characteristics of closed-loop feedback." Specifically, the spatiotemporal asynchronous co-encoder (also referred to as the "encoder" or "spatiotemporal asynchronous co-encoder module") is based on an ideal muscle activation curve. (Calculate the target trajectory through inverse dynamics) (Calculated to generate a multi-channel stimulation pulse sequence) The pulse sequence is transmitted through phase difference in the time dimension. and random jitter Asynchronous delivery is achieved by modulating activation-level dependent parameters in both spatial and intensity dimensions (narrow pulse widths are preferred for low activation). and low frequency It selectively activates slow-twitch muscle fibers, and synergistically increases activity during periods of high activation. , and By recruiting fast-twitch muscle fibers and simultaneously performing inter-electrode spatial rotation during periods of low activation, hierarchical recruitment conforming to the size principle is achieved, and braking timing parameters are used in the synergistic dimension. Controlling antagonistic muscles Early activation enables smooth braking; and the adaptive adjustment module does not simply adjust a single control quantity based on a single error signal, but simultaneously receives dual-source feedback signals from the sensor module, namely kinematic feedback. and physiological feedback Based on these two types of heterogeneous feedback information, and through a pre-defined rule base or learning algorithm, the internal three-dimensional parameter space of the spatiotemporal asynchronous co-encoder is structurally and specifically adjusted: when the trajectory error... Exceeding the threshold In this case, the adjustment module does not simply increase a general "stimulus intensity," but rather precisely corrects the ideal activation curve. Values ​​at specific times and / or adjustments to co-sequencing parameters This allows for directional correction of trajectory errors while maintaining the overall framework of the biomimetic coding strategy (TSC); when the fatigue index Exceeding the threshold At the same time, the adjustment module does not simply reduce the stimulation intensity (which would lead to insufficient torque and motion failure), but rather, under the constraint of maintaining the total torque output, it increases the number of spatially rotated electrodes (S-dimensional adjustment) and expands the temporal jitter range. (T-dimensional adjustment) further disperses the fatigue burden in space and time. This "fatigue redistribution under torque constraints" strategy is something that traditional controllers cannot achieve.

[0119] The non-obviousness of this "dual simulation" architecture lies in the fact that the front-end spatiotemporal asynchronous co-coding simulates the physiological process of "motor planning and neural coding" in the central nervous system, while the back-end adaptive adjustment module simulates the physiological process of "state perception and real-time reflex regulation" in the sensory nervous system. The closed loop formed by the two through the parameter feedback pathway is not a simple linear mapping of "perception-response," but a higher-order mapping of "perception-interpretation-structured reconfiguration." That is, the adjustment module needs to interpret the two types of heterogeneous information—motor error and fatigue state—into differentiated adjustment instructions for the three coding dimensions of T, S, and C. This interpretation process itself implies a profound understanding of the complex dynamic characteristics of the neuromuscular system, which is far beyond what those skilled in the art can easily obtain using conventional technical means. Furthermore, the three dimensions of T, S, and C are not independent of each other, but form an implicit coupling by working together on the two system-level goals of final torque output and fatigue accumulation: the enhancement of temporal asynchrony T (increasing...) While providing anti-fatigue effects, it may reduce the instantaneous controllability of torque. The optimization of spatial and intensity modulation S (spatial rotation and size principle) needs to coordinate with the asynchronous strategy of T dimension to ensure torque smoothness while delaying fatigue. The accuracy of multi-muscle synergistic timing C depends on the accuracy of the TS encoding parameters of the agonist and antagonist muscles respectively. Therefore, the dynamic adjustment of the adjustment module to these three dimensions of parameters must be synergistic and coupled, rather than isolated and linearly superimposed. This synergistic optimization problem in multi-dimensional parameter space has high nonlinearity and complexity, and its technical implementation path is by no means obvious to ordinary people skilled in the art.

[0120] It is through the triple deep coupling of the aforementioned "bionic coding structure" and "intelligent closed-loop mechanism" at the parameter, logic, and functional levels that this application achieves simultaneous optimization of three mutually constraining technical indicators: motion smoothness (reducing abruptness through T), fatigue resistance (following the size principle and spatial dispersion through S), and control accuracy (achieving multi-muscle coordination through C and eliminating errors through closed-loop). The effect of this multi-objective simultaneous optimization is unattainable by any single means in the prior art (whether it is open-loop bionics or closed-loop PID), thus producing outstanding and unexpected technical effects.

[0121] To make the objectives, technical solutions, and advantages of this application clearer, the embodiments of this application will be described in further detail below with reference to the accompanying drawings.

[0122] The embodiments of this application can be applied to the rehabilitation of motor function in patients with various neurological injuries (such as stroke, spinal cord injury, etc.).

[0123] Taking the rehabilitation of upper limb motor function in stroke patients as a typical application scenario, stroke-induced hemiplegia often causes patients to lose active control over the affected upper limb, making it impossible to perform basic daily movements such as elbow flexion to pick up objects, finger grasping, and wrist rotation. FES technology attempts to replace or assist these missing nerve signals through electrical stimulation, driving related muscles such as the biceps and triceps to contract in a predetermined pattern to complete functional movements.

[0124] After extensive and in-depth research, the inventors of this application discovered that the general problems mentioned in the background art are particularly prominent in this specific scenario. For example, traditional FES systems, represented by commercial systems, apply synchronous pulse trains with fixed pulse widths and frequencies to multiple electrode locations on the biceps brachii simultaneously when assisting patients in performing elbow flexion movements. This "all-or-nothing" activation method results in a very stiff sensation of movement felt by the patient. Furthermore, in tasks requiring precise finger gripping or wrist rotation, this rigid motion output leads to an extremely low task success rate.

[0125] At the same time, the control precision problem is also very serious in this scenario. For example, in the stopping phase of the elbow flexion movement, due to the lack of early braking and precise timing coordination of the antagonist muscle (such as the triceps brachii), if only the stimulation of the agonist muscle (biceps brachii) is stopped, the forearm will continue to move due to inertia, resulting in overshoot and inability to accurately position.

[0126] The following embodiments of this application (such as the first embodiment) will focus on the specific application scenario of upper limb rehabilitation, demonstrating how the method of this application systematically solves the specific technical problems of motion stiffness, muscle fatigue and poor control precision in the above application scenario through spatiotemporal asynchronous collaborative coding based on physiological simulation and adaptive closed-loop feedback based on multi-source sensors (such as IMU and sEMG).

[0127] The first embodiment of this application relates to a functional electrical stimulation method based on physiological simulation and adaptive closed-loop feedback. This embodiment takes elbow flexion movement in upper limb functional rehabilitation after stroke as an example to illustrate the various steps of the method and its technical implementation details.

[0128] The system hardware configuration of this embodiment will be described below.

[0129] The functional electrical stimulation system in this embodiment preferably includes the following hardware components:

[0130] Multi-channel constant current stimulator: It has 8 independently controllable stimulation channels, each of which can independently set the pulse amplitude, pulse width and pulse frequency, and has millisecond-level timing control accuracy.

[0131] Electrode array: includes multiple surface electrodes placed at the movement points of target muscles such as the biceps brachii (for example, as the agonist muscle) and triceps brachii (for example, as the antagonist muscle).

[0132] Sensor module: Includes an inertial measurement unit (IMU) and surface electromyography (sEMG) electrodes. Specifically, the IMU includes a triaxial accelerometer and a triaxial gyroscope, which, in a preferred configuration, are fixed to the patient's forearm to measure elbow joint angle and angular velocity; the sEMG electrodes can be placed in parallel with or reused with the stimulation electrodes to acquire muscle electrical activity signals.

[0133] Central processing unit: including microcontroller unit (MCU) and digital signal processor (DSP), for executing the various steps of the method of this application, including inverse dynamics calculation, spatiotemporal asynchronous cooperative coding, real-time signal processing and adaptive parameter adjustment.

[0134] In addition, the entire system can achieve data communication between modules via wired or wireless means to ensure real-time synchronization of stimulus output, sensor acquisition, and parameter adjustment.

[0135] The process in this embodiment is as follows: Figure 1 As shown, the method includes the following steps:

[0136] Step S100: System Initialization and Model Loading

[0137] Before rehabilitation training begins, the system first executes initialization step S100. This step specifically loads the patient's personalized biomechanical model. In one case, if the patient is using the system for the first time, a general Hill-type muscle model based on anthropometric parameters (such as limb length and weight) is loaded as the initial model. It should be noted that the Hill-type model is a well-known model that models muscles as three elastic elements connected in series and parallel, capable of describing muscle force-length relationships and force-velocity relationships.

[0138] In another preferred scenario, for patients who have undergone multiple training sessions, the system retrieves historical data accumulated during previous training from the database. This data includes the actual kinematic trajectories collected in previous training tasks, muscle fatigue index change curves, and optimized model parameters. In this way, the system can utilize the patient's specific physiological characteristics to make subsequent activation curve calculations more accurate.

[0139] Simultaneously, the system initializes each sensor module, performs zero-point calibration on the IMU to eliminate the effects of gravity and installation deviations, and performs impedance testing on the sEMG electrodes to ensure good signal quality. The system also sets initial values ​​for various threshold parameters, such as the trajectory error threshold. Set to 5 degrees, fatigue threshold Set to median frequency The values ​​are 20% lower than the initial values. It should be understood that these thresholds will be dynamically optimized based on the patient's real-time performance during subsequent adaptive adjustments.

[0140] Step S200: Target Trajectory Planning

[0141] In step S200, the target motion task is planned and the target motion trajectory is obtained. Specifically, in this embodiment, the therapist sets an elbow flexion task through the system interface, requiring the patient's elbow joint to flex from an extended position (0 degrees) to 30 degrees, with the entire movement lasting 2 seconds. Based on this task description, the system automatically generates a smooth target joint angle trajectory. .

[0142] The trajectory is preferably planned using the Minimum Jerk principle. It should be understood that this is a trajectory generation method that simulates the natural movement of the human body, which can ensure the smoothness of the movement.

[0143] More specifically, the target trajectory Within the time interval [0, 2 seconds], the trajectory smoothly transitions from 0 degrees to 30 degrees, exhibiting a bell-shaped distribution of acceleration followed by deceleration. The maximum angular velocity occurs at the midpoint of the motion (i.e., at 1 second), approximately 45 degrees / second. It should be understood that this trajectory design aligns with the motion control strategies of the human nervous system, avoiding the unnaturalness of step-like or constant-speed movements. The system stores this target trajectory as discrete time-series data, with a sampling frequency of 100Hz, meaning a target angle value is recorded every 10 milliseconds. This data will be used as a reference for error calculation and trajectory tracking control in subsequent steps.

[0144] It should be noted that target trajectory planning includes not only joint angles but also implicitly the corresponding angular velocity and angular acceleration information. This information can be calculated from the angular trajectory through numerical differentiation. The angular velocity and angular acceleration information is crucial for subsequent inverse dynamics calculations because the torque generated by the muscles needs to overcome not only the limb's gravity but also provide the inertial torque required for acceleration or deceleration.

[0145] Step S300: Ideal activation solution

[0146] In step S300, the ideal muscle activation curves for the multiple muscles required to achieve the target motion trajectory are calculated based on the target motion trajectory. This can be considered the core front-end step of the method in this application, which converts the desired motion trajectory into muscle activation instructions through inverse dynamics calculation.

[0147] The system first establishes a positive mapping relationship from muscle activation to joint torque and then to joint motion based on the biomechanical model loaded in step S100. In the elbow flexion movement of this embodiment, the main muscles involved include the agonist biceps brachii, the synergist brachialis, and the antagonist triceps brachii. For each muscle, the Hill-type model describes the muscle activation level. (Values ​​range from 0 to 1, where 0 represents complete relaxation and 1 represents maximum voluntary contraction) and the force generated by the muscle. The relationship between muscle force and joint torque is further converted into joint torque via the moment arm. The net torque of the joint is obtained by algebraically summing the torques generated by multiple muscles. This net torque then drives limb movement according to Newton's second law, for example, satisfying the following relationship:

[0148]

[0149] in, For the moment of inertia of limb rotation, Joint angular acceleration, The damping coefficient is... The joint angular velocity, This is the gravitational torque.

[0150] Therefore, step S300 performs the inverse operation of the aforementioned forward process. The system determines the target trajectory based on the target trajectory. And its derivatives (angular velocity and angular acceleration), first calculate the net joint torque required to achieve this motion. In the elbow flexion task of this embodiment, The activation value is positive in the initial stage of movement (0 to 1 second) to provide elbow flexion torque, and gradually decreases or even turns negative in the later stage of movement (1 to 2 seconds) to achieve deceleration and stopping. Next, the system uses an optimization algorithm to solve for the activation curves of each muscle. This ensures that the resultant torque generated by these muscles is as close as possible to the desired torque. .

[0151] For example, in this embodiment, the system calculates three ideal muscle activation curves:

[0152] Activation curve of the biceps brachii The intensity gradually increases from 0 to 0.6 during the period from 0 to 1 second (indicating moderate intensity activation), and then gradually decreases to 0.1 during the period from 1 to 2 seconds to maintain joint position;

[0153] Activation curve of the synergist brachialis muscle Similar to the biceps brachii but with a slightly lower amplitude, the peak value is approximately 0.4.

[0154] Activation curve of the antagonist muscle, triceps brachii Maintain a near-zero level in the early stages of the movement, but at 1.85 seconds (i.e., from the moment of stopping)... The braking torque (0.15 seconds ahead) is briefly activated to around 0.2 seconds and lasts for about 0.2 seconds to provide braking torque to prevent overshoot.

[0155] It should be noted that this pre-activation of antagonistic muscles fully reflects the characteristics of multi-muscle synergistic temporal control in this application.

[0156] It should be emphasized that the solution obtained in step S300 It is a continuous, idealized activation curve that reflects the activation commands the central nervous system "should" send to the muscles to achieve smooth movement. However, in practical FES applications, continuous activation signals cannot be directly applied; they must be converted into discrete electrical pulse sequences. This is precisely the task that the next step, S400, must accomplish.

[0157] Step S400: Spatiotemporal asynchronous cooperative coding

[0158] Step S400 is the core innovation of this application. Based on the ideal muscle activation curve obtained in step S300, this step performs spatiotemporal asynchronous cooperative coding (TSC coding) to generate a multi-channel stimulation pulse sequence. TSC coding includes three dimensions of biomimetic simulation: temporal asynchrony (T), spatial and intensity modulation (S), and multi-muscle cooperative temporality (C).

[0159] Implementation of time asynchronicity (T):

[0160] Timing asynchrony is achieved by setting a phase difference for different stimulation channels or introducing random jitter. Specifically, in this embodiment, for the biceps brachii, the system preferably uses three stimulation channels (channels 1, 2, and 3), with electrodes for these three channels respectively positioned at different movement points of the biceps brachii. The system sets a basic stimulation frequency. The frequency is 30Hz, which means that ideally each channel emits 30 pulses per second. However, to achieve time asynchrony, the pulse emission times of the three channels are not completely synchronized.

[0161] aisle The The pulse duration can be determined by the following formula:

[0162]

[0163] in, The basic stimulation frequency (30Hz in this embodiment) is used. For channel The fixed phase difference This represents the time jitter.

[0164] More specifically, the phase difference of channel 1 Set to 0 milliseconds, channel 2 Set to 11 milliseconds (approximately 1 / 3 of the period of 33.3 milliseconds), channel 3 It is set to 22 milliseconds (approximately 2 / 3 of the cycle). This setting ensures that the pulse firing times of the three channels are staggered at any given moment.

[0165] In addition, the system preferably introduces random jitter. (This is also referred to as "time jitter" in this article). In the early stages of exercise, when muscle fatigue is low, The pulses can be randomly drawn from a Gaussian distribution with a mean of 0 and a standard deviation of 2 milliseconds. This causes the actual firing time of each pulse to fluctuate randomly within ±4 milliseconds of the theoretical time. This randomness further enhances the asynchronous characteristics. According to the adaptive adjustment mechanism in step S700, when increased muscle fatigue is detected, The standard deviation will automatically increase to 5 milliseconds, that is, the jitter range will expand to ±10 milliseconds, thereby making the activation of motion units more dispersed in time and delaying the accumulation of fatigue.

[0166] A key technical effect of temporal asynchrony is that, although the contraction of a single motor unit exhibits pulsatile characteristics (contraction-relaxation cycle), due to the staggered firing of pulses from multiple channels, at the macroscopic level (e.g., within a 50-millisecond time window), some motor units are always in a contracted state, thus keeping the total force generated by the muscle relatively smooth. This simulates the mechanism by which the central nervous system achieves smooth force output by asynchronously recruiting motor units under natural conditions.

[0167] Implementation of spatial and intensity modulation (S):

[0168] Furthermore, spatial and intensity modulation simulates Henneman's size principle of muscle fibers and incorporates a spatial rotation strategy to combat fatigue. The encoding rules for this dimension are based on the ideal muscle activation curve. The activation level is dynamically adjusted to adjust the stimulation parameters.

[0169] when When at a low activation level (e.g., defined as...) The system prioritizes using low-intensity parameter combinations. Specifically, the pulse width is set to... :

[0170]

[0171] In this embodiment, The value is set to 100 microseconds. Narrower pulse widths are believed to preferentially activate slow-twitch muscle fibers (Type I) with lower activation thresholds; these fibers, while generating less force, possess extremely strong anti-fatigue properties. Simultaneously, the pulse frequency is set to... :

[0172]

[0173] The value is set to 20Hz, which is lower than the frequency required to produce a full tetanic contraction, thereby reducing the energy consumption of the muscle.

[0174] More importantly, at low activation levels, the system also employs a spatial rotation stimulation strategy. Although three channels are available, the system does not activate all channels simultaneously. Instead, it preferably uses a "rotational" approach: for example, only channels 1 and 2 pulse during the first 500 milliseconds, while channel 3 rests; during the next 500 milliseconds, channels 2 and 3 activate, while channel 1 rests; during the following 500 milliseconds, channels 1 and 3 activate, while channel 2 rests; and so on. This spatial rotation strategy disperses fatigue spatially, allowing muscle tissue under individual electrodes to have brief rest periods, significantly delaying the accumulation of overall fatigue.

[0175] when When at a high activation level (e.g., defined as...) The system needs to recruit more motor units and increase their firing frequency to generate greater torque. At this point, the system collaboratively increases several stimulation parameters: the pulse width increases to... :

[0176]

[0177] In this embodiment With a pulse width of 300 microseconds, it is believed that a wider pulse can activate fast-twitch muscle fibers (Type IIb) with a higher activation threshold. The pulse frequency was also increased synchronously. :

[0178]

[0179] A value of 40Hz is used; higher frequencies allow for more cohesive muscle contractions, generating greater steady-state force. Simultaneously, all three channels are activated without spatial switching, maximizing force output. Pulse amplitude. The stimulation intensity may also be increased as needed (e.g., from 15 mA to 25 mA) to further enhance the stimulation intensity.

[0180] For moderate activation levels ( The system can use linear interpolation or nonlinear mapping of parameters to smoothly transition pulse width, frequency and spatial activation strategy, avoiding abrupt changes in parameters.

[0181] The beneficial effects of spatial and intensity modulation are as follows: by simulating the size principle, the system preferentially utilizes fatigue-resistant slow-twitch muscle fibers when the torque demand is low, and only recruits fatigue-prone fast-twitch muscle fibers when the torque demand is high; through spatial rotation, the system distributes the fatigue load among multiple electrode positions, thereby reducing the stimulation dose at any position per unit time, which greatly delays the onset of overall muscle fatigue.

[0182] Implementation of multi-muscle synergistic timing (C):

[0183] Furthermore, multi-muscle synergistic temporal control involves the precise temporal coordination of multiple muscles (agonists, synergists, and antagonists). In the elbow flexion task of this embodiment, the biceps brachii and brachialis muscles, as agonists and synergists respectively, are encoded according to the aforementioned T and S rules, based on their respective... The curve generates a corresponding pulse sequence.

[0184] The encoding of the antagonist muscle, the triceps brachii, reflects the key characteristics of synergistic timing. Based on the solution results of step S300, the activation curve of the triceps brachii... A brief activation is required at the end of the motion to provide braking. The system adjusts based on braking timing parameters. At the moment of agonist muscle cessation The antagonistic muscle is activated at a specific point in the preceding time. This point can be determined by the following formula:

[0185]

[0186] in, Indicates the activation moment of the antagonistic muscle. This indicates the moment when the agonist muscle stops (2 seconds in this example). This indicates the braking advance time. In this embodiment, The initial setting was 30 milliseconds (i.e., 0.03 seconds), therefore the antagonist muscle in It is activated at the very first second.

[0187] Specifically, at 1.97 seconds, the system sends a series of low-intensity pulses to the stimulation channels (channels 7 and 8) located in the triceps brachii. For example, the pulse width is 150 microseconds, the frequency is 25 Hz, and the duration is approximately 150 milliseconds. This brief activation of the antagonist muscle generates a torque in the opposite direction to that of the agonist muscle, acting as a "brake" to smoothly stop the flexion movement of the elbow joint at the target position of approximately 30 degrees, avoiding overshoot caused by inertia.

[0188] It should be understood that braking timing parameters The choice is crucial. If If the time interval is too long (e.g., 100 milliseconds), the antagonist muscle will intervene prematurely, preventing the agonist muscle from completing the flexion movement; if If the value is too small (e.g., 10 milliseconds), the braking effect will not have enough time to take effect, and overshoot will still occur. In this embodiment, The initial value is set empirically to be in the range of 20 to 50 milliseconds, and will be optimized according to the actual trajectory error in the adaptive adjustment in step S700.

[0189] The beneficial technical effect of multi-muscle synergistic timing is that by precisely controlling the activation timing of agonist and antagonist muscles, the system achieves high-precision control of complex movements (especially movements that require precise positioning and smooth stopping), solving the problems of motion overshoot and oscillation commonly found in traditional FES.

[0190] Output of the pulse parameter set:

[0191] In summary, the final output of step S400 is A sequence of pulse parameters for each stimulation channel. In this embodiment, (3 channels for the biceps brachii, 2 channels for the brachialis, 2 channels for the triceps brachii, and 1 spare channel). Each channel The pulse parameters at each time step It can be represented as a set of parameters:

[0192]

[0193] in, Pulse amplitude (unit: milliampere). Pulse width (unit: microseconds) The pulse frequency (unit: Hertz). The phase difference is expressed in milliseconds. These parameters change dynamically throughout the motion according to the TSC encoding rules. The system preferably recalculates and updates these parameters at a refresh rate of 100 Hz to ensure that the stimulus output accurately follows the changes in the ideal activation curve.

[0194] Step S500: Stimulus Execution and Real-time Sensing

[0195] In step S500, a multi-channel stimulation pulse sequence is applied while the patient's actual kinematic state and / or muscle physiological state are monitored in real time via sensors. This step is a concrete manifestation of innovation point 2 of this application, providing the necessary sensing capabilities for closed-loop control.

[0196] Stimulus execution:

[0197] The multichannel stimulator generates a set of pulse parameters based on step S400. Electrical stimulation is applied to the target muscle through corresponding electrode channels. The constant current output circuit of each channel ensures that the current flowing through the muscle remains constant even if the skin-electrode impedance changes, thus guaranteeing the stability of the stimulation effect. Preferably, the stimulation pulse is a biphasic pulse, i.e., each pulse contains a positive phase and a negative phase, to balance the charge and avoid electrode polarization and tissue damage.

[0198] In this embodiment, the stimulus comes from The moment begins and continues for 2 seconds until the exercise task is completed. During the 0 to 1 second period, pulses are alternately delivered to channels 1, 2, and 3 of the biceps brachii according to the TSC coding rules, with pulse parameters varying accordingly. The intensity gradually increases with the increase of the activation curve. During the period of 1 to 2 seconds, the pulse parameters decrease as the activation curve decreases. At 1.97 seconds, antagonistic muscle channels 7 and 8 are briefly activated to provide braking.

[0199] Real-time kinematic monitoring:

[0200] Simultaneously, the inertial measurement unit (IMU) begins acquiring actual motion data of the elbow joint. The IMU is fixed to the distal end of the patient's forearm, measuring the linear acceleration of the forearm using a three-axis accelerometer and the angular velocity using a three-axis gyroscope. The system can then calculate the actual angle of the elbow joint from this raw data using sensor fusion algorithms (such as complementary filters or Kalman filters). and actual angular velocity .

[0201] The IMU's sampling frequency is 100Hz, consistent with the target trajectory's sampling frequency, ensuring time synchronization. In this embodiment, the system records data in real time. Changes: Under ideal conditions, We should follow closely However, in practice, due to factors such as individual patient differences, muscle fatigue, and imperfections in stimulation parameters, and There will always be a certain trajectory error. .

[0202] Real-time physiological monitoring:

[0203] Similarly, surface electromyography (sEMG) electrodes continuously acquire electromyographic signals from the biceps brachii to assess the physiological state of the muscle, particularly the degree of fatigue. In this embodiment, the sEMG electrodes and stimulation electrodes are arranged adjacent to each other, or a dedicated sEMG acquisition electrode is placed in the gap between the stimulation channels.

[0204] To ensure the accurate implementation of the technical solution, the core calculation formulas involved in this application are summarized and explained as follows:

[0205] 1. Pulse firing time calculation (time asynchronicity):

[0206] No. The first channel The timing of pulse emission Defined as:

[0207]

[0208] in, Based on the base frequency, For channel Phase offset, This represents the amount of random jitter that varies over time.

[0209] 2. Intensity modulation strategy (Henneman magnitude principle):

[0210] Based on ideal muscle activation level Adjusting pulse width and frequency :

[0211] when Time (prioritizes activation of slow-twitch muscles):

[0212]

[0213] when Time (recruiting fast muscle groups):

[0214]

[0215] when When the parameters are linearly interpolated within the aforementioned interval, the parameters are interpolated.

[0216] 3. Calculation of Muscle Fatigue Index (FI):

[0217] Based on the median frequency of surface electromyography signals calculate:

[0218]

[0219] in The median frequency of the initial state.

[0220] 4. Trajectory error correction (iterative learning control):

[0221] Based on trajectory error Update the activation curve for the next cycle:

[0222]

[0223] in This is the learning gain coefficient.

[0224] Those skilled in the art will understand that sEMG signal acquisition faces the problem of stimulation artifacts. Each stimulation pulse not only activates the motor units in the muscle but also generates significant potential changes in the tissue. These changes superimpose on the real electromyographic signal, forming artifacts, often with amplitudes several orders of magnitude larger than the actual electromyographic signal. To address this issue, this embodiment preferably employs a time-domain synchronous sampling strategy: the sEMG acquisition module and the stimulator's pulse firing sequence are precisely synchronized. Immediately after each stimulation pulse, the system triggers a brief "blanking window," during which sEMG sampling is paused or the sampled data is discarded (e.g., within 2 milliseconds after the pulse ends), thus avoiding the primary influence of artifacts. During the inter-pulse interval, the system resumes normal sampling to capture real muscle electrical activity.

[0225] It should be noted that, since the stimulation frequency is 20 to 40 Hz and the pulse interval is 25 to 50 milliseconds, there is still a sufficient time window (approximately 20 to 45 milliseconds) for effective sEMG sampling within each interval. The system acquires sEMG signals at a high sampling rate of 1000 Hz, and even after deducting the blanking window, approximately 20 to 45 effective sample points can still be acquired within each interval, which is sufficient for subsequent frequency and time domain analysis.

[0226] After bandpass filtering (e.g., 20 to 500 Hz) to remove motion artifacts and power frequency interference, the sEMG signal can be used to calculate its power spectral density via short-time Fourier transform (STFT). The median frequency is then extracted from the power spectrum. This refers to the frequency point that divides the power spectrum in two. Typically, medical research indicates that during muscle fatigue, the conduction velocity of muscle fibers decreases, causing the frequency components of the sEMG signal to shift to lower frequencies. It will gradually decrease as fatigue levels increase. The system calculates in real time. and the initial values ​​at the start of training. By comparing these factors, a muscle fatigue index can be defined. for:

[0227]

[0228] For example, if the initial median frequency The frequency is 80Hz, at the current moment. If the frequency is 72Hz, then:

[0229]

[0230] This indicates a muscle fatigue level of 10%.

[0231] A key technical advantage of step S500 is that, through dual-modal sensing (kinematic + physiological), the system can not only sense whether the action is correct (via...) and (Comparison), and can also sense "muscle fatigue" (through) (Assessment). This multi-source sensing provides a comprehensive information foundation for subsequent intelligent decision-making and adaptive adjustment, and is a prerequisite for achieving true closed-loop control.

[0232] Step S600: Feedback Interpretation and Status Assessment

[0233] In step S600, the trajectory error is calculated based on the actual kinematic state and the target motion trajectory, and the muscle physiological state is evaluated. This step transforms the raw sensor data acquired in step S500 into meaningful control signals.

[0234] Trajectory error calculation:

[0235] Specifically, the system calculates the current trajectory error at each sampling time (e.g., every 10 milliseconds). For example, it can be defined as the difference between the target angle and the actual angle:

[0236]

[0237] in, Ideal joint angles (unit: degrees) representing the target's motion trajectory. This indicates the actual monitored joint angle (unit: degrees). Positive. The value indicates that the actual angle lags behind the target angle (i.e., insufficient elbow flexion), and a negative value indicates that the actual angle lags behind the target angle. The value indicates that the actual angle is ahead of the target angle (i.e., excessive elbow flexion or overshoot).

[0238] Furthermore, in this embodiment, the system not only calculates the instantaneous error Furthermore, it is preferable to calculate the statistical characteristics of the error. For example, within a 200-millisecond sliding time window, the root mean square (RMSE) value of the system error is calculated as an indicator of tracking accuracy, and the maximum absolute value of the error is calculated as an indicator of the maximum deviation. These statistics provide a more robust basis for the adaptive adjustment in step S700, avoiding misjudgments caused by transient noise.

[0239] The system also calculates the error relative to an error threshold. The relationship. If (In this embodiment) For example, if the degree is set to 5 degrees, the system determines that the current trajectory tracking accuracy is insufficient and needs to trigger the trajectory error correction strategy in step S700.

[0240] Physiological status assessment:

[0241] Similarly, the muscle fatigue index calculated in the system evaluation step S500 In this embodiment, the fatigue threshold For example, setting it to 20%, which is the median frequency. The value has decreased by 20% relative to the initial value. The system continues to monitor. The trend of change: in the early stages of motion (e.g., the first 500 milliseconds). It typically remains at a low level, below 5%; as stimulation continues, Gradually rising; if Exceed The system determines that the muscles have entered a state of significant fatigue and needs to trigger an anti-fatigue strategy in step S700.

[0242] In addition to threshold determination, the system can also evaluate The rate of change (i.e., the rate of fatigue accumulation). If An excessively large time derivative indicates that muscle fatigue is accumulating rapidly, even at present. Even if the threshold has not been exceeded, the system will take preventative measures, such as enhancing the spatial rotation strategy in advance or appropriately reducing the movement speed of the target trajectory.

[0243] Optionally, step S600 can also generate a comprehensive status assessment report, including trajectory tracking quality level (e.g., "Excellent", "Good", "Needs Improvement") and fatigue risk level (e.g., "Low", "Medium", "High"). These high-level assessment results, along with the original... and The data is then passed to step S700 as a decision input.

[0244] The beneficial technical effect of step S600 is that by converting multimodal sensing data into structured error and state information, the system can accurately identify the problems existing in the current FES execution (whether it is an accuracy problem or a fatigue problem), providing clear optimization goals and directions for subsequent adaptive adjustments.

[0245] Step S700: Dynamic adaptive adjustment of parameters

[0246] In step S700, based on the trajectory error and physiological state assessment results from step S600, the internal parameters of the spatiotemporal asynchronous co-coding are dynamically adjusted to update the multi-channel stimulation pulse sequence. Essentially, step S700 is a real-time decision-making module that transforms the system from "open-loop execution" to "closed-loop optimization."

[0247] Adjustment strategy based on trajectory error (Rule A):

[0248] When the system detects a trajectory error Exceeding the error threshold At that time, the trajectory error correction strategy is triggered. For example, in this embodiment, it is assumed that at a certain moment during the motion process (e.g. (seconds), system detected The degree exceeded If the error is positive, it indicates that the elbow joint is not flexed sufficiently, and the actual angle lags behind the target angle.

[0249] The system first analyzes the cause of the error. By comparing the current pulse parameters with the ideal activation curve, the system may find that the current stimulation intensity of the agonist muscle is insufficient to generate the required torque. Therefore, the system can take the following adjustment measures:

[0250] Measure 1: Correct the ideal muscle activation curve The system can use the Iterative Learning Control (ILC) algorithm, based on the error of the current iteration. The system compensates for the activation curve in the next cycle (or a subsequent period of the current cycle). Specifically, the system will... exist Add an increment within a short period of time after the second. This increment can be proportional to the error, for example:

[0251]

[0252] like Spend, Degree, then:

[0253]

[0254] Therefore, originally Revised to (and limited to the range [0,1]).

[0255] Measure 2: Fine-tune the timing parameters of multi-muscle synergy. Specifically, the system checks whether the braking timing of the antagonistic muscle is appropriate. If error analysis indicates overshoot (negative error) at the end of the movement, the system will reduce... (This allows antagonistic muscles to intervene earlier); if undershoot (positive error) occurs at the end of the movement, it increases... (To delay the intervention of the antagonist muscle) or reduce the stimulation intensity of the antagonist muscle. In this embodiment, if a positive error is detected near the stopping time, the system will... The time was increased from 30 milliseconds to 40 milliseconds, allowing the antagonist to intervene slightly later, thus avoiding premature braking that could prevent the agonist from reaching the target angle.

[0256] These adjusted parameters are immediately fed back to the TSC encoder in step S400. The encoder adjusts according to the new parameters. and Recalculate pulse parameters This generates an updated stimulus sequence. Since the adjustment is performed in real time, the new stimulus sequence takes effect from the next control period (e.g., after 10 milliseconds), thus enabling a rapid response to errors.

[0257] Adjustment strategy based on fatigue state (Rule B):

[0258] Similarly, when the system detects the muscle fatigue index Exceeding the fatigue threshold At that time, an anti-fatigue strategy is triggered. In this embodiment, it is assumed that at a certain moment during the motion process (e.g., (seconds), system detected Exceeded This indicates that the biceps brachii has entered a state of significant fatigue.

[0259] The system immediately took the following anti-fatigue measures:

[0260] Measure 1: Strengthen the spatial rotation strategy (S-dimensional adjustment). The system checks the current electrode activation mode. If it finds that certain channels have been under high load for an extended period (e.g., channel 1 has been working almost continuously for the past second), the system will increase the rotation frequency or expand the rotation range. Specifically, the system will shorten the rotation cycle from 500 milliseconds to 300 milliseconds, reducing the working time of each channel and increasing its rest opportunities. Simultaneously, the system may activate channels that were originally in standby mode (e.g., channel 4), distributing the load to more spatial locations.

[0261] Furthermore, while maintaining the total torque output unchanged, the system reduces the stimulation intensity of individual electrodes by increasing the number of activated electrodes. For example, if channels 1 and 2 originally output 15 mA each, now channels 1, 2, and 4 each output 10 mA, with the total current still at approximately 30 mA, but the stimulation density at a single location is significantly reduced, thereby alleviating local fatigue.

[0262] Measure 2: Enhance time asynchronicity (adjustment of T dimension). Add random jitter to the system. The scope. As mentioned earlier, The standard deviation was increased from the initial 2 milliseconds to 5 milliseconds, making the timing of pulse firing more random and the activation of motor units more dispersed. This adjustment allows muscle fibers more micro-intervals in time, which helps to clear metabolic products (such as lactic acid) and replenish energy substances.

[0263] Measure 3: (If necessary) Adjust the stimulation frequency (S-dimensional adjustment). If fatigue accumulation is still too rapid, the system may appropriately reduce the stimulation frequency, for example, from the current 30Hz to 25Hz, to reduce the total activation load on the muscles. However, a lower frequency will lead to a decrease in force output, so the system will simultaneously slightly increase the pulse width or amplitude to partially compensate for the loss of force, ensuring that the target task can still be completed.

[0264] Measure 4: (Extreme Case) Adjust the target trajectory. If A sustained and rapid increase, even exceeding 30%, indicates that the muscles are nearing their fatigue limit. Continuing to force the original task may lead to muscle damage or loss of control. At this point, the system can automatically reduce the difficulty of the target trajectory, for example, slowing the elbow flexion speed by 50% (extending the completion time from 2 seconds to 4 seconds), or temporarily reducing the target angle (from 30 degrees to 20 degrees), or even prompting the patient and therapist to pause training and rest. In this embodiment, the system is equipped with a "safety restriction" module. When this happens, the current task will be automatically interrupted and an alarm will be issued.

[0265] The adjustment of the anti-fatigue strategy is also fed back to the encoder in step S400 in real time. The new TSC encoding parameters (including more aggressive spatial rotation, greater temporal jitter, adjusted frequency and pulse width) take effect immediately, thereby dynamically combating fatigue accumulation during motion execution.

[0266] Coordination and balance of adjustments:

[0267] It should be noted that the key to step S700 lies in balancing two objectives: trajectory tracking accuracy and fatigue management. In some cases, these two objectives may conflict. For example, increasing the stimulus intensity can improve trajectory tracking, but it can accelerate fatigue; performing spatial rotation can combat fatigue, but it may cause slight fluctuations in force output due to changes in activation position, affecting accuracy.

[0268] Therefore, step S700 preferably resolves this contradiction through a multi-objective optimization strategy. The system can assign weights to the two objectives: in the early stages of exercise or when fatigue levels are low, trajectory accuracy has a higher weight (e.g., 0.7), with the system prioritizing the accuracy of the movement; as fatigue intensifies, the weight of fatigue management gradually increases (e.g., from 0.3 to 0.6), with the system focusing more on protecting muscle function. The system can use a fuzzy controller or a rule-based decision tree, based on the current... and By dynamically adjusting various parameters, the optimal balance between accuracy and fatigue can be found.

[0269] The core technical effect of step S700 lies in the fact that, through real-time, structured adaptive adjustment, the system can maintain stable control performance in dynamically changing environments (muscle fatigue, fluctuations in patient condition). Compared with traditional open-loop FES or simple PID closed-loop FES, the adaptive mechanism of this application does not crudely increase or decrease the "total stimulus," but intelligently adjusts the internal strategy of TSC encoding (temporal distribution, spatial distribution, multi-muscle synergy), thereby achieving the ideal state of "precision without fatigue."

[0270] Step S800: Loop execution and task termination

[0271] In step S800, steps (d) to (f) are repeated until the target motion trajectory is completed. Specifically, in this embodiment, the system executes the closed-loop cycle of steps S500 to S700 at a control frequency of 100Hz. Every 10 milliseconds, the system completes a full "perception-evaluation-adjustment-execution" cycle: the sensor collects new data... and Data (step S500), calculate the latest The status is evaluated (step S600), the TSC coding parameters are updated based on the evaluation results (step S700), and the stimulator outputs the updated pulse sequence (return to step S500).

[0272] This high-frequency closed-loop system enables the system to respond quickly to changes in the patient's condition. For example, if the patient's muscles stiffen due to tension during exercise, causing the actual angle to suddenly deviate from the target, the system can detect the abnormality within 100 milliseconds (10 control cycles) and apply corrective adjustments to keep the deviation within an acceptable range.

[0273] Therefore, in the 2-second elbow flexion task of this embodiment, the system executes a total of 200 closed-loop cycles (2 seconds × 100Hz). Ideally, the trajectory error... Throughout the process, the temperature remained within ±3 degrees Celsius, and the fatigue index was maintained within ±3 degrees Celsius. At the end of the task, the percentage was approximately 15%, which did not exceed the threshold. The task was in... The time was completed in seconds, and the system detected it. degrees and angular velocity Once the goal has been achieved, stop stimulating output.

[0274] If an abnormality occurs during task execution (such as sensor failure, continuous error exceeding limits, or excessive fatigue), the system's safety monitoring module will immediately interrupt the task and display fault information to ensure patient safety.

[0275] Step S900: Iterative Model Optimization

[0276] In step S900, the personalized physiological-mechanical model iterative optimization is performed after the target motion trajectory is executed. This step is crucial for the long-term performance improvement of the system.

[0277] Specifically, after the 2-second elbow flexion task in this embodiment is completed, the system automatically stores all training data into the patient record database. The stored data includes: the complete time-series data collected in step S500 (…). sequence, Sequence), the pulse parameter sequence generated in step S400 ( The sequence), the parameter adjustment record executed in step S700 (when and what kind of adjustment was triggered), and the error statistics (RMSE, maximum deviation, etc.) calculated in step S600.

[0278] The system then initiates an offline optimization algorithm. This algorithm analyzes the data from this task and identifies weaknesses in the system's performance. For example, if the trajectory error is consistently large over certain time periods, it indicates that the biomechanical model in step S300 is inaccurate in estimating the muscle torque-activation relationship for this patient. The system can use the data collected this time ( , Data pairs are used to update key parameters in the biomechanical model, such as the maximum isometric force of muscles, lever arm curves, and shape parameters of force-length relationships, through parameter identification algorithms (such as least squares or nonlinear optimization).

[0279] The updated model is automatically loaded in step S100 of the next training session (e.g., the second day of rehabilitation therapy). Using the optimized model, the ideal activation curve is calculated in step S300. This will better match the patient's actual physiological characteristics, thereby reducing initial errors and improving trajectory tracking accuracy.

[0280] Furthermore, the system can learn the optimal TSC encoding parameters. This is achieved by analyzing historical data (specific TSC parameters → generated trajectory error). and fatigue The system may detect that for this patient, phase difference is appropriate. Milliseconds (instead of the initial 11 milliseconds) can result in smoother power delivery, or better braking timing. Milliseconds (instead of the initial 30 milliseconds) are more effective at preventing overshoot. The system encodes these "experiences" into a personalized set of parameters, which are used as initial values ​​in subsequent training, thereby reducing the amount of real-time adjustments required in step S700 and improving the system's response speed and stability.

[0281] In a preferred embodiment, if the patient undergoes multiple training sessions (e.g., more than 10 times), the system accumulates sufficient data to train a more advanced learning model. For example, using a reinforcement learning algorithm, with a weighted sum of trajectory error and fatigue index as the "reward function," a model can be learned from the current state (…). , , To the optimal adjustment action ( , , , Alternatively, a strategy function (e.g., Gaussian process regression) can be used to establish a nonlinear mapping from patient state to optimal stimulus parameters. These machine learning models can be integrated into the adaptive adjustment module of step S700 to make the system's decision-making more intelligent.

[0282] A significant technical advantage of step S900 lies in the system's ability to "learn" and "evolve." For each patient, the system's performance continuously improves with increased usage. Initially, the system relies on a general model and default parameters, which may result in mediocre performance; after several training sessions, the system has been deeply optimized for that specific patient, leading to more accurate trajectory tracking, more effective fatigue management, and truly personalized, precise rehabilitation.

[0283] The second embodiment of this application relates to a functional electrical stimulation system, the structure of which is as follows: Figure 2 As shown, the functional electrical stimulation system includes:

[0284] The trajectory planning module is configured to plan the target motion task and obtain the target motion trajectory.

[0285] Activate the calculation module and configure it to calculate the ideal muscle activation curves of multiple muscles required to achieve the target motion trajectory based on the target motion trajectory;

[0286] The spatiotemporal asynchronous co-coding module is configured to perform spatiotemporal asynchronous co-coding to generate a multi-channel stimulation pulse sequence based on the ideal muscle activation curve, wherein the spatiotemporal asynchronous co-coding simulates at least one or more of time asynchrony, spatial and intensity modulation, or multi-muscle co-sequencing.

[0287] A multichannel stimulator configured to apply the multichannel stimulation pulse sequence;

[0288] The sensor module is configured to monitor the actual kinematic state and / or muscle physiological state in real time while the multi-channel stimulation pulse sequence is applied;

[0289] The state assessment module is configured to calculate the trajectory error based on the actual kinematic state and the target motion trajectory, and / or assess the muscle physiological state;

[0290] An adaptive adjustment module is configured to dynamically adjust the internal parameters of the spatiotemporal asynchronous co-coding module based on the assessment results of the trajectory error and / or the muscle physiological state, in order to update the multi-channel stimulation pulse sequence; and

[0291] The control module is configured to control the multi-channel stimulator, the sensor module, the state assessment module, and the adaptive adjustment module to repeatedly perform stimulation, monitoring, assessment, and adjustment until the target motion trajectory is completed.

[0292] The first embodiment is a method embodiment corresponding to this embodiment. The technical details in the first embodiment can be applied to this embodiment, and the technical details in this embodiment can also be applied to the first embodiment.

[0293] The above embodiments have the following technical effects:

[0294] The functional electrical stimulation method described above systematically solves the core problems of traditional functional electrical stimulation techniques, such as motor stiffness, rapid muscle fatigue, and low control precision, by organically combining spatiotemporal asynchronous collaborative coding, multi-source sensing monitoring, and adaptive closed-loop feedback. This achieves high-quality, low-fatigue, and high-precision motor function rehabilitation.

[0295] First, the above embodiments fundamentally improve the smoothness and naturalness of movement through spatiotemporal asynchronous co-coding. Traditional functional electrical stimulation systems employ synchronous stimulation, where multiple electrodes simultaneously deliver stimulation pulses to the same muscle, causing all motor units within that muscle to be activated at the same moment, resulting in an "all or nothing" step force output. Consequently, joint movement trajectories exhibit stiff and discontinuous characteristics. The temporal asynchronicity of the above embodiments, achieved by setting preset phase differences for different stimulation channels or introducing random jitter, ensures that the pulse delivery times of multiple stimulation channels are staggered. This asynchronous delivery mechanism ensures that within any tiny time window, some motor units are always in a contracted state, while on a macroscopic time scale, the total force output of the muscle exhibits a smooth and continuous change. Compared to traditional synchronous stimulation, the joint movement trajectories generated by the method in the above embodiments show significantly reduced abruptness, and the movement perceived by the patient is smoother and more natural, approaching a normal voluntary movement pattern. By setting fixed phase differences for the channels and using jitter based on random distribution, the system achieves a balance between microscopic asynchronicity and macroscopic smoothness while ensuring controllable overall stimulation intensity.

[0296] Secondly, the above-described embodiments significantly delay the onset of muscle fatigue through spatial and intensity modulation. Traditional functional electrical stimulation often employs fixed high-intensity stimulation parameters. This approach violates the Hernemann size principle, preferentially activating high-threshold, easily fatigued fast-twitch muscle fibers, while low-threshold, fatigue-resistant slow-twitch muscle fibers are not fully utilized, leading to significant muscle fatigue in patients within a short period. The spatial and intensity modulation of the above-described embodiments simulates the muscle fiber size principle. When the ideal muscle activation curve is at a low activation level, the system preferentially uses low-intensity parameter combinations (such as narrow pulse width and low frequency). These parameters have high spatial selectivity, preferentially activating superficial, low-threshold slow-twitch muscle fibers. When the activation level is high, the system synergistically increases the pulse frequency, pulse width, and amplitude to gradually recruit deeper fast-twitch muscle fibers. This graded recruitment strategy ensures that muscle force output conforms to physiological laws, significantly reducing energy consumption and fatigue accumulation. Furthermore, at low activation levels, the system performs spatial rotation stimulation across multiple electrode locations, allowing sufficient rest time for the muscle regions beneath each electrode, effectively dispersing fatigue spatially. Compared to traditional methods, the methods described above can significantly extend the time it takes for muscles to reach the fatigue threshold while producing the same torque output. This means that patients can undergo rehabilitation training for longer periods, and the effectiveness of individual training sessions and cumulative effects are greatly improved.

[0297] Third, the above embodiments achieve high-precision execution of complex movements through multi-muscle synergistic timing control. Many functional movements require not only the contraction of agonist muscles but also the timely braking or stabilizing effect of antagonist muscles. Traditional functional electrical stimulation systems typically only stimulate agonist muscles, neglecting the synergistic effect of antagonist muscles, resulting in a lack of braking torque during the stopping phase of the movement, which can easily lead to overshoot, oscillation, or instability. The multi-muscle synergistic timing control of the above embodiments precisely controls the stimulation initiation time, duration, and activation sequence of agonist, synergist, and antagonist muscles. In particular, the system activates the antagonist muscle in advance according to the braking timing parameters before the agonist muscle is about to stop contracting, so that the torque generated by the antagonist muscle matches the gradually weakening torque of the agonist muscle, achieving smooth deceleration and precise positioning of the joint. The braking advance time can be flexibly set according to the movement speed and inertial characteristics to ensure that the movement stops accurately at the target position without overshoot. This precise timing synergy between agonist and antagonist muscles not only improves the accuracy of individual movements but also provides a technical basis for complex movements that require fine coordination of multiple muscles (such as finger grasping and wrist pronation and supination).

[0298] More importantly, the above embodiments, through real-time sensing monitoring and adaptive closed-loop feedback, enable the spatiotemporal asynchronous cooperative coding strategy to be dynamically optimized and always maintain its optimal state. Traditional functional electrical stimulation systems are either open-loop, relying entirely on preset parameters and unable to cope with changes in the patient's physiological state; or they only use simple proportional-integral-derivative controllers to adjust a single feedback signal, with limited adjustment capabilities. The above embodiments, through inertial measurement units to monitor the actual kinematic state in real time and through surface electromyography to monitor the muscle physiological state, especially fatigue state, in real time, provide the system with comprehensive state perception capabilities. The system compares the actual motion trajectory with the target trajectory to calculate the trajectory error and assesses the muscle fatigue index; these two types of information are used to dynamically adjust the internal parameters of the spatiotemporal asynchronous cooperative coding.

[0299] When the detected trajectory error exceeds a preset threshold, the system automatically corrects the ideal muscle activation curve or adjusts parameters of the multi-muscle synergy timing (such as braking timing) to reduce the error in the next cycle or at the next moment, achieving precise trajectory tracking. This iterative correction mechanism based on error feedback can compensate for the deviation between the biomechanical model and the patient's actual physiological characteristics, enabling the system to achieve high-precision control through closed-loop learning even when the model is not completely accurate. Compared to open-loop functional electrical stimulation, the trajectory tracking error of the method described in the above embodiments is significantly reduced.

[0300] When the muscle fatigue index exceeds the fatigue threshold, the system immediately activates an anti-fatigue strategy. While maintaining a relatively constant total torque output, it increases the number of spatially rotated electrodes and the range of temporally asynchronous random jitter, further distributing the fatigue burden spatially and temporally. This dynamic adjustment mechanism based on fatigue feedback ensures that even during prolonged training, the system can intelligently adjust its stimulation strategy to combat fatigue accumulation and maintain exercise quality. While traditional open-loop bionic coding may produce good results in the initial stages, the preset coding parameters become ineffective once the patient's physiological state (such as fatigue level) changes. The above embodiment, through closed-loop adaptation, fundamentally solves the robustness problem of bionic coding in practical applications.

[0301] Furthermore, the three dimensions of spatiotemporal asynchronous co-coding (temporal asynchronicity, spatial and intensity modulation, and multi-muscle co-coordination timing) do not act in isolation, but rather work synergistically to produce a cumulative effect. Temporal asynchronicity ensures smoothness at the microscopic level, spatial and intensity modulation optimizes fatigue distribution at the mesoscopic level through size principles and spatial rotation, and multi-muscle co-coordination timing achieves precise coordination of multiple joints and muscles at the macroscopic level. These three dimensions work together in the generation process of the stimulus pulse sequence, resulting in optimal smoothness, fatigue resistance, and accuracy in the final stimulus output. Adaptive closed-loop feedback then adjusts the parameters of these three dimensions, adjusting the temporal jitter range, spatial rotation strategy, intensity modulation parameters, and co-coordination timing parameters based on real-time monitored motion errors and fatigue states, achieving "structured intelligent adjustment" rather than the "black-box adjustment" of traditional controllers. This structured adjustment method not only has better regulatory effects but also better interpretability and clinical acceptability.

[0302] The methods described above also possess long-term learning and personalized optimization capabilities. By storing the actual kinematic state, muscle physiological state, and adjusted internal parameters during each training session, the system can iteratively optimize the biomechanical model and spatiotemporal asynchronous co-coding parameters based on this historical data. As patients undergo multiple training sessions, the system gains a deeper understanding of the patient's specific physiological characteristics, and the stimulation strategy becomes increasingly precise, ultimately achieving truly personalized rehabilitation. This long-term learning capability enables the methods described above not only to address short-term physiological state changes during a single training session through real-time closed-loop mechanisms, but also to adapt to the recovery and changes in the patient's motor abilities during long-term rehabilitation through model iteration. Since different patients exhibit significant differences in muscle strength, fatigue characteristics, and neuromuscular response characteristics, the personalized mechanism of the above embodiments ensures that each patient receives the stimulation program most suitable for their individual characteristics, avoiding the poor effects or side effects of a "one-size-fits-all" approach.

[0303] In summary, the methods described above systematically improve functional electrical stimulation (FES) technology from three levels: stimulation coding strategy (spatiotemporal asynchronous collaborative coding), state perception capability (multi-source sensor monitoring), and intelligent adjustment mechanism (adaptive closed-loop feedback and iterative optimization). This significantly enhances FES's smoothness of movement, fatigue resistance, and control precision, while also providing personalized and adaptive capabilities. This offers more effective, comfortable, and precise motor function rehabilitation for patients with neurological injuries such as stroke and spinal cord injury.

[0304] It should be noted that those skilled in the art should understand that the functions of each module shown in the above-described embodiments of the functional electrical stimulation system can be understood with reference to the relevant descriptions of the aforementioned functional electrical stimulation methods. The functions of each module shown in the above-described embodiments of the functional electrical stimulation system can be implemented by a program (executable instructions) running on a processor, or by specific logic circuits. If the functional electrical stimulation system described in this application is implemented as a software functional module and sold or used as an independent product, it can also be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application embodiment, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer system (which may be a personal computer, server, or network system, etc.) to execute all or part of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as a USB flash drive, mobile hard drive, read-only memory (ROM), magnetic disk, or optical disk. Thus, the embodiments of this application are not limited to any specific hardware and software combination.

[0305] Accordingly, this application also provides a computer storage medium storing computer-executable instructions, which, when executed by a processor, implement the various method implementations of this application.

[0306] Furthermore, this application also provides a functional electrical stimulation system, including a memory for storing computer-executable instructions and a processor; the processor is used to implement the steps in the above-described method embodiments when executing the computer-executable instructions in the memory. The processor may be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), etc. The aforementioned memory may be read-only memory (ROM), random access memory (RAM), flash memory, hard disk, or solid-state drive, etc. The steps of the methods disclosed in the embodiments of this application can be directly implemented by a hardware processor, or implemented by a combination of hardware and software modules in the processor.

[0307] It should be noted that in this patent application, relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or system that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or system. Without further limitations, an element defined by the phrase "comprising one" does not exclude the presence of other identical elements in the process, method, article, or system that includes said element. In this patent application, if it refers to performing an action according to an element, it means performing the action at least according to that element, including two cases: performing the action only according to that element, and performing the action according to that element and other elements. Expressions such as "multiple," "repeatedly," and "various" include two, two times, two kinds, and more than two, more than two times, and more than two kinds.

[0308] All documents mentioned in this application are considered to be incorporated in their entirety into the disclosure of this application so that they can serve as a basis for modifications if necessary. Furthermore, it should be understood that after reading the foregoing disclosure of this application, those skilled in the art can make various alterations or modifications to this application, and these equivalent forms also fall within the scope of protection claimed in this application.

Claims

1. A functional electrical stimulation system, characterized by Comprise: a trajectory planning module configured to plan a target motion task and obtain a target motion trajectory; an activation solving module configured to solve ideal muscle activation profiles of multiple muscles required to achieve the target motion trajectory based on the target motion trajectory; a spatiotemporal asynchronous synergistic encoding module configured to perform spatiotemporal asynchronous synergistic encoding to generate a multi-channel stimulation pulse sequence based on the ideal muscle activation profiles, the spatiotemporal asynchronous synergistic encoding including temporal asynchrony, spatial and intensity modulation, and multi-muscle simultaneous timing; wherein the temporal asynchronicity is achieved by setting a preset phase difference for different stimulation channels in the multi-channel stimulation pulse sequence or by introducing a random jitter wherein the random jitter satisfies for ensuring the non-synchronicity of the pulse firing; wherein the spatial and intensity modulation mimics muscle fiber size principle, including: when the ideal muscle activation curve is at low activation level, preferentially using low intensity parameter combination, including narrow pulse width and low frequency to preferentially activate slow muscle fibers; when the ideal muscle activation curve is at high activation level, synergistically increasing pulse frequency , pulse width and / or amplitude to recruit fast muscle fibers; and, when the ideal muscle activation curve is at low activation level, performing spatial rotation stimulation between multiple electrode locations to spatially disperse muscle fatigue; Wherein, the multi-muscle coordination timing comprises: before the agonist in the multi-muscle stops contraction, according to the braking timing parameter , at time , the antagonist is activated in advance to provide braking torque and prevent movement overshoot, wherein , represents the activation time of the antagonist, represents the stop time of the agonist, represents the braking advance time; a multi-channel stimulator configured to apply the multi-channel stimulation pulse sequence; a sensor module configured to monitor actual kinematic state and / or muscle physiological state in real time while the multi-channel stimulation pulse sequence is applied; a state evaluation module configured to calculate trajectory error based on the actual kinematic state and the target motion trajectory, and / or evaluate the muscle physiological state; an adaptive adjustment module configured to dynamically adjust internal parameters of the spatiotemporal asynchronous synergistic encoding module based on the trajectory error and / or evaluation results of the muscle physiological state to update the multi-channel stimulation pulse sequence; and a control module configured to control the multi-channel stimulator, the sensor module, the state evaluation module, and the adaptive adjustment module to repeatedly perform stimulation, monitoring, evaluation, and adjustment until the target motion trajectory is completed.

2. The system of claim 1, wherein, Channel The first pulse time is determined by the equation: wherein is a base stimulation frequency, is a channel fixed phase difference, is a time dither.

3. The system of claim 1, wherein, When time, using , ; When time, using , ; wherein, is a normalized ideal muscle activation curve, taking values in the range 0 to 1 ; is a narrow pulse width, used to preferentially activate slow muscle fibers; is a wide pulse width, used to recruit fast muscle fibers; is a low stimulation frequency; is a high stimulation frequency.

4. The system of claim 1, wherein, The brake advance time is 20 to 50 milliseconds.

5. The system of claim 1, wherein, The multi-channel stimulation pulse sequence comprises pulse parameters of the stimulation channels, wherein , is the number of muscles, the pulse parameters of each channel are represented by: wherein, is the pulse amplitude, is the pulse width, is the pulse frequency, is the phase difference.

Images