An adaptive knee joint protection method and device based on predicted motion risk

Abstract

This invention relates to an adaptive knee joint protection method and device based on predicted motion risk. The method includes: using an embedded inertial measurement unit, pressure sensor, electromyography sensor, and depth camera to collect real-time user knee joint motion data, environmental data, and gait data; environmental data includes ground material, slope, and light intensity; gait data includes cadence, stride length, and gait cycle; inputting the knee joint motion data, environmental data, and gait data into a knee joint motion risk prediction model to predict the user's knee joint motion risk in the future; when the knee joint motion risk exceeds a preset threshold, controlling an exoskeleton through a knee joint assistive device to adjust the knee joint's flexion-extension damping and range of motion. This invention identifies motion intention through a risk prediction model and, based on a shape memory alloy variable stiffness actuator and gait patterns, achieves real-time monitoring and personalized protection of knee joint motion risk.

Description

Technical Field

[0001] This invention relates to the fields of biomedical engineering and rehabilitation robotics, specifically to an adaptive knee joint protection method and device based on predictive motion risk, and a computer device. Background Technology

[0002] Knee joints are prone to injury in daily life and sports, especially for the elderly and certain occupational groups. While traditional passive protective devices such as knee braces provide some support and protection, they cannot adaptively adjust to the user's real-time movement status and environment, resulting in limited protective effects and even hindering normal movement. Existing knee joint protection and rehabilitation training largely relies on the experience and judgment of doctors, making it difficult to achieve personalized and precise rehabilitation plans. Furthermore, it lacks effective prediction of future sports risks, hindering early intervention and prevention, and is ill-suited to complex and changing sports environments and the individualized needs of users.

[0003] To address the aforementioned issues, this invention proposes an adaptive knee joint protection method based on predicted motion risk. This method identifies motion intentions through a risk prediction model and utilizes a variable stiffness actuator based on shape memory alloy and gait patterns to achieve real-time monitoring and personalized protection of knee joint motion risks. Summary of the Invention

[0004] In view of the above problems, the present invention provides an adaptive knee joint protection method and device, and a computer device based on predicting motion risk.

[0005] According to one aspect of the present invention, an adaptive knee joint protection method based on predicted motion risk is provided, comprising:

[0006] The system utilizes an embedded inertial measurement unit, pressure sensor, electromyography sensor, and depth camera to collect real-time data on the user's knee joint motion, environment, and gait. The knee joint motion data includes angular velocity, angular acceleration, joint angle, ground reaction force data, and muscle activity data. The environment data includes ground material, slope, and light intensity. The gait data includes stride frequency, stride length, and gait cycle.

[0007] The knee joint motion data, environmental data, and gait data are input into the knee joint motion risk prediction model to predict the user's knee joint motion risk in the future period, wherein the future period is any value between 0.1 seconds and 5 seconds.

[0008] When the risk of knee joint movement exceeds a preset threshold, the exoskeleton is controlled by the knee joint assist device to adjust the flexion and extension damping and range of motion of the knee joint.

[0009] In one alternative approach, the knee joint motion risk prediction model employs a spatiotemporal attention-enhanced long short-term memory network model.

[0010] The loss function of the spatiotemporal attention-enhanced long short-term memory network model is:

[0011]

[0012] in, This is a time-varying class weight function based on gated recurrent units; ; It is the sigmoid activation function; Let be the hidden state of the GRU at time t-1; The regularization coefficient is used. It is the Frobenius norm; The total number of samples; Number of risk levels; Let be the true probability that the i-th sample belongs to the c-th category; Let be the predicted probability that the i-th sample belongs to the c-th category; The third derivative over time;

[0013] The output layer of the spatiotemporal attention-enhanced long short-term memory network model uses a dual-branch structure to predict acute injury risk values ​​and chronic strain indices, respectively. The fusion formula for the knee joint movement risk is as follows:

[0014]

[0015] in, This represents the risk value for acute injury. It is a chronic strain index; These are dynamic weighting coefficients; This is the time decay factor; The moment of commencement of the movement; This is the current time.

[0016] In an alternative approach, the method further includes:

[0017] A six-degree-of-freedom transformation matrix from the sensor coordinate system to the human anatomical coordinate system is established, and the spatial attitude is estimated in real time through extended Kalman filtering;

[0018] Wavelet packet decomposition was used to perform a 5-level decomposition of electromyographic signals, and energy entropy features were extracted as priors of motor intent.

[0019] Kinematic chain constraints are constructed using the 3D coordinates of the joints obtained by a depth camera, and optimization equations are established using the Lagrange multiplier method.

[0020] In one alternative approach, the state equation for the spatial attitude is:

[0021]

[0022] in, , These are quaternions, angular velocity, and gyroscope drift, respectively. Let k be the state vector at time k; This is the state transition function; This is the control input at time k; This is process noise; Let be the observation vector at time k; For observation functions; To observe noise;

[0023] The optimization equation is:

[0024]

[0025] in, The joint angle vector; This refers to the number of key points; The three-dimensional coordinate measurement value of the i-th joint point in the depth camera coordinate system; The three-dimensional coordinates of the i-th joint point are obtained by forward kinematics calculation. This is the Jacobian matrix used to describe the relationship between joint angular velocity and joint point velocity; The target joint velocity; To balance the weights between visual measurement error and motion intent tracking.

[0026] In one alternative embodiment, the knee joint assist device includes a variable stiffness actuator based on a shape memory alloy, wherein the variable stiffness actuator establishes a two-way coupling relationship between the temperature field and the stress field through an SMA phase transformation dynamics model.

[0027] The SMA phase transition dynamics model predicts the drive current optimizer in the finite time domain of each control cycle.

[0028] In one alternative approach, the bidirectional coupling relationship is as follows:

[0029]

[0030] in, The density of the material; Specific heat capacity; For temperature; Thermal conductivity; For time; For stress; In response to the situation; Latent heat of phase transition; For stress divergence; External force density; For acceleration;

[0031] The finite-time domain optimization expression for each control cycle is as follows:

[0032]

[0033] in, Current state The extent of injury to the lower knee joint; Let be the expected risk value at time k; It is a positive definite weight matrix; Let be the driving current at time k.

[0034] In an alternative approach, the method further includes:

[0035] A user gait pattern library is established using a hidden Markov model. The user gait pattern library includes normal gait, abnormal forward lean gait, excessive internal rotation gait, excessive external rotation gait, hip abduction gait, foot drop gait, scissor gait, short leg gait, staggering gait, forward lunge gait, and high leg lift gait.

[0036] In an alternative approach, the step of establishing a user gait pattern library using a hidden Markov model further includes:

[0037] A left-right type HMM structure is initialized for each gait pattern contained in the user gait pattern library; wherein the number of hidden states is automatically determined by the Bayesian information criterion.

[0038] The parameters of the left-right type HMM structure are re-estimated according to the Baum-Welch algorithm;

[0039] The forward probability of the observed sequence belonging to a left-right HMM structure is calculated in real time, and the optimal state sequence of the forward probability is decoded by the Viterbi algorithm.

[0040] According to another aspect of the present invention, an adaptive knee protection device based on predicted motion risk is provided, comprising:

[0041] The data acquisition module is used to collect real-time user knee joint motion data, environmental data, and gait data using an embedded inertial measurement unit, pressure sensor, electromyography sensor, and depth camera. The knee joint motion data includes angular velocity, angular acceleration, joint angle, ground reaction force data, and muscle activity data. The environmental data includes ground material, slope, and light intensity. The gait data includes stride frequency, stride length, and gait cycle.

[0042] The risk prediction module is used to input the knee joint motion data, environmental data and gait data into the knee joint motion risk prediction model to predict the user's knee joint motion risk in the future period, wherein the future period is any value between 0.1 seconds and 5 seconds.

[0043] The knee joint protection module is used to control the exoskeleton through the knee joint assist device to adjust the flexion and extension damping and range of motion of the knee joint when the risk of knee joint movement exceeds a preset threshold.

[0044] According to another aspect of the present invention, a computer device is provided, comprising: a processor, a memory, a communication interface, and a communication bus, wherein the processor, the memory, and the communication interface communicate with each other through the communication bus;

[0045] The memory is used to store at least one executable instruction, which causes the processor to perform the operation corresponding to the above-described adaptive knee joint protection method based on predicted motion risk.

[0046] According to the solution provided by this invention, an embedded inertial measurement unit, pressure sensor, electromyography sensor, and depth camera are used to collect real-time user knee joint motion data, environmental data, and gait data. The knee joint motion data includes angular velocity, angular acceleration, joint angle, ground reaction force data, and muscle activity data. The environmental data includes ground material, slope, and light intensity. The gait data includes stride frequency, stride length, and gait cycle. The knee joint motion data, environmental data, and gait data are input into a knee joint motion risk prediction model to predict the user's knee joint motion risk in the future, where the future cycle is any value between 0.1 seconds and 5 seconds. When the knee joint motion risk exceeds a preset threshold, the exoskeleton is controlled via a knee joint assistive device to adjust the knee joint's flexion-extension damping and range of motion. This invention identifies movement intention through a risk prediction model and, based on a shape memory alloy variable stiffness actuator and gait patterns, achieves real-time monitoring and personalized protection of knee joint motion risk.

[0047] The above description is merely an overview of the technical solution of the present invention. In order to better understand the technical means of the present invention and to implement it in accordance with the contents of the specification, and in order to make the above and other objects, features and advantages of the present invention more apparent and understandable, specific embodiments of the present invention are described below. Attached Figure Description

[0048] Various other advantages and benefits will become apparent to those skilled in the art upon reading the following detailed description of preferred embodiments. The accompanying drawings are for illustrative purposes only and are not intended to limit the invention. Furthermore, the same reference numerals denote the same parts throughout the drawings. In the drawings:

[0049] Figure 1 A flowchart illustrating an adaptive knee joint protection method based on predicted motion risk according to an embodiment of the present invention is shown.

[0050] Figure 2 A schematic diagram of the framework of an adaptive knee joint protection device based on predicted motion risk according to an embodiment of the present invention is shown;

[0051] Figure 3 A schematic diagram of the structure of a computer device according to an embodiment of the present invention is shown. Detailed Implementation

[0052] Exemplary embodiments of the invention will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the invention are shown in the drawings, it should be understood that the invention may be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that this invention will be thorough and complete, and will fully convey the scope of the invention to those skilled in the art.

[0053] Figure 1 A flowchart illustrating an adaptive knee joint protection method based on predicted motion risk according to an embodiment of the present invention is shown. Specifically, as... Figure 1 As shown, it includes the following steps:

[0054] Step S101: The user's knee joint motion data, environmental data, and gait data are collected in real time using an embedded inertial measurement unit, pressure sensor, electromyography sensor, and depth camera. The knee joint motion data includes angular velocity, angular acceleration, joint angle, ground reaction force data, and muscle activity data. The environmental data includes ground material, slope, and light intensity. The gait data includes stride frequency, stride length, and gait cycle.

[0055] In this embodiment, an embedded inertial measurement unit and integrated sensors are used to minimize interference with the user's movement and improve wearability. The pressure sensor, electromyography (EMG) sensor, and depth camera are all non-invasive, avoiding harm to the human body. The depth camera provides environmental information (such as ground texture and slope) to better predict movement risks. Gait data is used to identify the user's gait patterns, such as the presence of abnormal gait, thereby enabling risk assessment and personalized protection.

[0056] For example, when a user climbs stairs, an inertial measurement unit (IMU) measures the angular velocity and angular acceleration of the knee joint (reflecting the range and speed of knee movement). A pressure sensor measures the ground reaction force on the stair steps (reflecting the load on the knee joint). Electromyography (EMG) measures the activity potentials of the quadriceps femoris muscle (reflecting muscle exertion). A depth camera acquires information about the stair slope and the three-dimensional position of the user's knee joint. The IMU data is used to calculate the knee joint's posture angle, and the pressure sensor data is filtered to obtain a smooth ground reaction force curve. The EMG data is filtered and amplified to extract muscle activity features. The depth camera data and IMU data are fused to obtain the knee joint's motion trajectory. By analyzing the IMU, pressure sensor, and EMG data, it is determined whether the user exhibits abnormal movement patterns such as excessive knee flexion or excessive inversion during stair climbing. The stair slope information is used to predict the stress on the knee joint over a future period, assessing the risk of injury. Based on the risk assessment results, an exoskeleton is used to adjust the knee joint's flexion-extension damping and range of motion to protect the knee joint.

[0057] In an alternative approach, the method further includes:

[0058] A six-degree-of-freedom transformation matrix from the sensor coordinate system to the human anatomical coordinate system is established, and the spatial attitude is estimated in real time through extended Kalman filtering;

[0059] Wavelet packet decomposition was used to perform a 5-level decomposition of electromyographic signals, and energy entropy features were extracted as priors of motor intent.

[0060] Kinematic chain constraints are constructed using the 3D coordinates of the joints obtained by a depth camera, and optimization equations are established using the Lagrange multiplier method.

[0061] In this embodiment, a six-degree-of-freedom transformation matrix from the sensor coordinate system to the human anatomical coordinate system is established, improving the accuracy and stability of posture estimation. Wavelet packet decomposition is used to perform a five-level decomposition of the electromyographic signal, extracting energy entropy features as a priori motion intent to more accurately capture the user's motion intent. When predicting knee joint motion risk, the user's actual motion intent is considered, thereby improving prediction accuracy. Kinematic chain constraints are constructed using the three-dimensional coordinates of joint points obtained from a depth camera, and optimization equations are established using the Lagrange multiplier method to ensure the accuracy and consistency of the kinematic model. This allows for tracking and adjusting the knee joint's motion state during the control process of the knee joint assistive device.

[0062] Specifically, firstly, the relationship between the sensor coordinate system and the human anatomical coordinate system is determined through a calibration process, establishing a six-degree-of-freedom transformation matrix. This transformation matrix converts the sensor-measured data into the human anatomical coordinate system, thus achieving a unified coordinate system. Using the Extended Kalman Filter (EKF) algorithm, sensor data, and a kinematic model, the spatial posture of the knee joint is estimated in real time. Wavelet packet decomposition is performed on the electromyographic signals to extract energy entropy features. Energy entropy features reflect the complexity and uncertainty of muscle activity and, as prior information about movement intention, help to more accurately predict the user's movement intention. Three-dimensional coordinates of joint points are obtained through a depth camera to construct kinematic chain constraints, which ensure that the movement of joint points conforms to human kinematics. The Lagrange multiplier method is used to introduce the kinematic chain constraints into the optimization equation to establish an optimization objective function. This objective function balances the weights between visual measurement errors and movement intention tracking, ensuring that the control strategy of the knee joint assistive device both conforms to the user's movement intention and effectively protects the knee joint.

[0063] In one alternative approach, the state equation for the spatial attitude is:

[0064]

[0065] in, , These are quaternions, angular velocity, and gyroscope drift, respectively. Let k be the state vector at time k; This is the state transition function; This is the control input at time k; This is process noise; Let be the observation vector at time k; For observation functions; To observe noise;

[0066] The optimization equation is:

[0067]

[0068] in, The joint angle vector; This refers to the number of key points; The three-dimensional coordinate measurement value of the i-th joint point in the depth camera coordinate system; The three-dimensional coordinates of the i-th joint point are obtained by forward kinematics calculation. This is the Jacobian matrix used to describe the relationship between joint angular velocity and joint point velocity; The target joint velocity; To balance the weights between visual measurement error and motion intent tracking.

[0069] In this embodiment, quaternions, angular velocity, and gyroscope drift are used as state vectors to more accurately describe the spatial attitude of the knee joint. Quaternions avoid the gimbaling problem that may occur with Euler angles, improving the stability of the attitude representation. Gyroscope drift compensates for sensor errors, further improving the accuracy of attitude estimation. A state equation is used to describe the system's changes over time, allowing the attitude estimation to be dynamically adjusted according to the motion state. By minimizing the difference between the Jacobian matrix and the target joint velocity in the optimization equation, the attitude estimation accurately reflects the actual motion state and responds to the user's motion intentions.

[0070] Specifically, a state transition function is constructed based on the kinematic model to describe the change of the state vector over time. For example, quaternions are updated using an integral method, and angular velocity is updated based on angular acceleration (assuming the gyroscope drift changes slowly). The control input is determined according to the specific application scenario (e.g., angular acceleration is selected as the control input), and a covariance matrix of the process noise is set to describe the uncertainty of the state transition function. The observation vectors of the joints, including the 3D coordinates obtained from the depth camera measurement, are determined, and an observation function is constructed to map the state vector to the observation vector. For example, quaternions are used to transform the joint coordinates in the human anatomical coordinate system to the depth camera coordinate system. A covariance matrix of the observation noise is set to describe the uncertainty of the sensor measurement. The state vector and state covariance matrix are initialized. The state vector and state covariance matrix at the current moment are predicted based on the state transition function. The state vector and state covariance matrix are updated based on the observation equation and the observation vector. The prediction and update steps are repeated to achieve real-time attitude estimation. Using joint angle vectors and human kinematics models, the three-dimensional coordinates of each joint point in the depth camera coordinate system are calculated, the Jacobian matrix between joint angular velocity and joint point velocity is calculated, and the optimal joint angle vector is obtained by solving the optimization equation using numerical optimization methods (such as gradient descent).

[0071] Step S102: Input the knee joint motion data, environmental data, and gait data into the knee joint motion risk prediction model to predict the user's knee joint motion risk in the future period, wherein the future period is any value between 0.1 seconds and 5 seconds.

[0072] In this embodiment, traditional protection methods typically react based on the current state. However, this application predicts knee joint movement risks over future periods, taking protective measures to avoid or mitigate injury before potential risks occur, thus providing more effective and proactive knee joint protection. The knee joint movement risk prediction model considers the user's individual movement data, environmental data, and gait data. The prediction results are used to control the exoskeleton, achieving personalized adaptive protection. By collecting and processing data in real time and predicting future trends in periods of 0.1 to 5 seconds, it can quickly respond to changes in the user's movement state, ensuring timely protection.

[0073] For example, while a user is running, the embedded system collects the following data in real time: angular velocity increases rapidly, joint angle approaches maximum flexion angle, and muscle activity intensity increases significantly. The ground is asphalt with a slightly uphill slope and good lighting. Step frequency increases, and stride length increases. After preprocessing, the data is input into a knee joint motion risk prediction model. Assuming a future period of 0.5 seconds, the model predicts a high risk of acute injury and a slightly elevated chronic strain index within the next 0.5 seconds, ultimately calculating a knee joint motion risk exceeding a preset threshold. Determining a high risk of knee injury within the next 0.5 seconds, protective measures are immediately triggered. The exoskeleton increases knee joint flexion and extension damping, limiting excessive flexion, absorbing impact, and reducing pressure on the knee joint. The exoskeleton appropriately restricts the range of motion of the knee joint to prevent excessive stretching or twisting. These protective measures reduce the risk of knee injury during running.

[0074] In one alternative approach, the knee joint motion risk prediction model employs a spatiotemporal attention-enhanced long short-term memory network model.

[0075] The loss function of the spatiotemporal attention-enhanced long short-term memory network model is:

[0076]

[0077] in, This is a time-varying class weight function based on gated recurrent units; ; It is the sigmoid activation function; Let be the hidden state of the GRU at time t-1; The regularization coefficient is used. It is the Frobenius norm; The total number of samples; Number of risk levels; Let be the true probability that the i-th sample belongs to the c-th category; Let be the predicted probability that the i-th sample belongs to the c-th category; The third derivative over time;

[0078] The output layer of the spatiotemporal attention-enhanced long short-term memory network model uses a dual-branch structure to predict acute injury risk values ​​and chronic strain indices, respectively. The fusion formula for the knee joint movement risk is as follows:

[0079]

[0080] in,; This represents the risk value for acute injury. It is a chronic strain index; These are dynamic weighting coefficients; This is the time decay factor; The moment of commencement of the movement; This is the current time.

[0081] In this embodiment, a spatiotemporal attention-enhanced long short-term memory network model is used to better capture key features in time series data, adaptively adjust the weights of different time steps and spatial dimensions, and use a dual-branch structure to predict acute injury risk values ​​and chronic strain indices, while considering both short-term and long-term exercise risks.

[0082] Step S103: When the risk of knee joint movement exceeds a preset threshold, the exoskeleton is controlled by the knee joint assist device to adjust the flexion and extension damping and range of motion of the knee joint.

[0083] In this embodiment, when the predicted movement risk exceeds a preset threshold, the exoskeleton control algorithm is activated. The exoskeleton control algorithm adjusts the flexion-extension damping and range of motion of the knee joint to reduce the stress on the knee joint. For example, in high-risk situations, damping is increased to limit rapid and violent movements, while in low-risk situations, damping is reduced to increase the degree of freedom of movement.

[0084] For example, if the predicted motion risk value is 0.6, below the preset threshold, the exoskeleton maintains low flexion-extension damping, allowing for more natural movement and unrestricted range of motion. However, when the user climbs a steep slope or experiences an unstable gait due to slippery surfaces, causing the predicted motion risk value to exceed the preset threshold (0.9), the exoskeleton automatically increases flexion-extension damping to slow the knee's flexion speed and reduce impact on the knee joint. Alternatively, the exoskeleton slightly restricts excessive flexion or extension of the knee joint to prevent excessive stress on the knee joint at unsafe angles. Despite these protective interventions, the exoskeleton strives to maintain the user's freedom of movement, avoiding excessive constraints and thus improving the user experience. The user perceives the assistive device providing support and stability to the knee joint when the risk increases, preventing potential sports injuries.

[0085] In one alternative embodiment, the knee joint assist device includes a variable stiffness actuator based on a shape memory alloy, wherein the variable stiffness actuator establishes a two-way coupling relationship between the temperature field and the stress field through an SMA phase transformation dynamics model.

[0086] The SMA phase transition dynamics model predicts the drive current optimizer in the finite time domain of each control cycle.

[0087] In this embodiment, the variable stiffness characteristics of shape memory alloy (SMA) are utilized to dynamically adjust the support force of the exoskeleton based on the predicted knee joint movement risk. This approach better meets the human body's movement needs than traditional exoskeletons with fixed stiffness or simple damping adjustment, avoiding excessive constraint or insufficient support. The behavior of the SMA actuator is more accurately predicted using an SMA phase transition dynamics model, enabling precise control of the knee joint assist torque. Specifically, the SMA actuator, temperature sensor, current control circuit, and control algorithm are integrated into the knee joint assist device, allowing real-time control and data monitoring of the SMA actuator via an embedded system or host computer.

[0088] In one alternative approach, the bidirectional coupling relationship is as follows:

[0089]

[0090] in, The density of the material; Specific heat capacity; For temperature; Thermal conductivity; For time; For stress; In response to the situation; Latent heat of phase transition; For stress divergence; External force density; For acceleration;

[0091] The finite-time domain optimization expression for each control cycle is as follows:

[0092]

[0093] in, Current state The extent of injury to the lower knee joint; Let be the expected risk value at time k; It is a positive definite weight matrix; Let be the driving current at time k.

[0094] In this embodiment, the bidirectional coupling relationship describes the interaction between the temperature field and stress field of the shape memory alloy (SMA), more realistically reflecting its physical characteristics. Optimization within the finite time domain of each control cycle allows for timely adjustment of the SMA actuator's drive current to adapt to changing motion risks, thus providing more effective protection. The optimization objective function directly considers the degree of injury to the knee joint, enabling the control strategy to minimize the risk of motion injury.

[0095] For example, when a user encounters a slope while walking, the knee joint motion risk prediction model predicts that in the next control cycle, due to the larger slope angle, the shear force on the knee joint will increase, resulting in a higher risk of motion injury. A two-way coupling model is used to predict the temperature and stress distribution of the SMA under different drive currents. The optimal drive current is calculated based on the finite-time domain optimization expression to minimize the damage to the knee joint. The current control circuit applies the calculated drive current to the SMA actuator. The SMA actuator adjusts its stiffness according to the drive current to provide additional support to the knee joint, reducing the shear force and minimizing the risk of motion injury. If the slope angle decreases in the next control cycle, reducing the motion risk, the optimization algorithm recalculates the drive current to restore the stiffness of the SMA actuator to a lower level.

[0096] In an alternative approach, the method further includes:

[0097] A user gait pattern library is established using a hidden Markov model. The user gait pattern library includes normal gait, abnormal forward lean gait, excessive internal rotation gait, excessive external rotation gait, hip abduction gait, foot drop gait, scissor gait, short leg gait, staggering gait, forward lunge gait, and high leg lift gait.

[0098] In this embodiment, the assist mode of the exoskeleton is controlled by combining information from the gait database, providing the right assist force at the right time, thereby improving user experience and safety.

[0099] In an alternative approach, the step of establishing a user gait pattern library using a hidden Markov model further includes:

[0100] A left-right type HMM structure is initialized for each gait pattern contained in the user gait pattern library; wherein the number of hidden states is automatically determined by the Bayesian information criterion.

[0101] The parameters of the left-right type HMM structure are re-estimated according to the Baum-Welch algorithm;

[0102] The forward probability of the observed sequence belonging to a left-right HMM structure is calculated in real time, and the optimal state sequence of the forward probability is decoded by the Viterbi algorithm.

[0103] In this embodiment, the Bayesian Information Criterion (BIC) is used to automatically determine the number of hidden states in the Hidden MM, avoiding the subjectivity of manual setting and adapting to the complexity of different gait patterns. The Baum-Welch algorithm, as an HMM parameter estimation method, can effectively learn the parameters of the HMM from the training data, enabling the model to accurately describe gait patterns. Forward probabilities are calculated in real time, and the Viterbi algorithm is used to decode the optimal state sequence for real-time gait identification. It can not only identify the current gait pattern but also decode the stage (state) of the gait at each time step.

[0104] For example, a Hidden Model (HMM) is created to describe the "forward gait." This HMM is determined to be a left-right structure, and the number of hidden states is automatically determined to be 4 using the BIC criterion, corresponding to the four phases of the forward gait: stance phase initiation, forward shift of body center of gravity, stance phase end, and swing phase. The state transition probability matrix, observation probability matrix, and initial state probability vector are randomly initialized. The HMM is trained using a large amount of "forward gait" data, and the state transition probability matrix, observation probability matrix, and initial state probability vector are iteratively updated using the Baum-Welch algorithm. After multiple iterations, the parameters of the HMM converge, accurately describing the "forward gait." When the user walks, gait data is collected in real time, and feature vectors are extracted. The forward probability of this feature vector belonging to the "forward gait" HMM is calculated, and the Viterbi algorithm is used to decode the optimal state sequence. For example, the Viterbi algorithm decodes the following state sequence: State 1 -> State 2 -> State 3 -> State 4. If the forward probability exceeds a preset threshold, it is determined that the user is in a forward-leaning gait, and the exoskeleton is controlled to adjust the knee joint flexion and extension damping to prevent the user from falling.

[0105] According to the solution provided by this invention, an embedded inertial measurement unit, pressure sensor, electromyography sensor, and depth camera are used to collect real-time user knee joint motion data, environmental data, and gait data. The knee joint motion data includes angular velocity, angular acceleration, joint angle, ground reaction force data, and muscle activity data. The environmental data includes ground material, slope, and light intensity. The gait data includes stride frequency, stride length, and gait cycle. The knee joint motion data, environmental data, and gait data are input into a knee joint motion risk prediction model to predict the user's knee joint motion risk in the future, where the future cycle is any value between 0.1 seconds and 5 seconds. When the knee joint motion risk exceeds a preset threshold, the exoskeleton is controlled via a knee joint assistive device to adjust the knee joint's flexion-extension damping and range of motion. This invention identifies movement intention through a risk prediction model and, based on a shape memory alloy variable stiffness actuator and gait patterns, achieves real-time monitoring and personalized protection of knee joint motion risk.

[0106] Figure 2 A schematic diagram of the framework of an adaptive knee joint protection device based on predicted motion risk according to an embodiment of the present invention is shown. The adaptive knee joint protection device based on predicted motion risk includes:

[0107] The data acquisition module 210 is used to collect real-time user knee joint motion data, environmental data, and gait data using an embedded inertial measurement unit, pressure sensor, electromyography sensor, and depth camera; wherein, the knee joint motion data includes angular velocity, angular acceleration, joint angle, ground reaction force data, and muscle activity data; the environmental data includes ground material, slope, and light intensity; and the gait data includes stride frequency, stride length, and gait cycle.

[0108] The risk prediction module 220 is used to input the knee joint motion data, environmental data and gait data into the knee joint motion risk prediction model to predict the user's knee joint motion risk in the future period, wherein the future period is any value between 0.1 seconds and 5 seconds.

[0109] The knee joint protection module 230 is used to control the exoskeleton through the knee joint assist device to adjust the flexion and extension damping and range of motion of the knee joint when the risk of knee joint movement exceeds a preset threshold.

[0110] Figure 3 The diagram shows a structural schematic of an embodiment of the computer device of the present invention. The specific embodiments of the present invention do not limit the specific implementation of the computer device.

[0111] like Figure 3 As shown, the computer device may include: a processor 302, a communications interface 304, a memory 306, and a communications bus 308.

[0112] The processor 302, communication interface 304, and memory 306 communicate with each other via communication bus 308. Communication interface 304 is used to communicate with other network elements such as clients or other servers. The processor 302 executes program 310, specifically performing the relevant steps in the above-described embodiment of the adaptive knee joint protection method based on predicted motion risk.

[0113] Specifically, program 310 may include program code that includes computer operation instructions.

[0114] Processor 302 may be a central processing unit (CPU), an application-specific integrated circuit (ASIC), or one or more integrated circuits configured to implement embodiments of the present invention. The computer device includes one or more processors, which may be processors of the same type, such as one or more CPUs; or processors of different types, such as one or more CPUs and one or more ASICs.

[0115] Memory 306 is used to store program 310. Memory 306 may include high-speed RAM memory, and may also include non-volatile memory, such as at least one disk storage device.

[0116] According to the solution provided by this invention, an embedded inertial measurement unit, pressure sensor, electromyography sensor, and depth camera are used to collect real-time user knee joint motion data, environmental data, and gait data. The knee joint motion data includes angular velocity, angular acceleration, joint angle, ground reaction force data, and muscle activity data. The environmental data includes ground material, slope, and light intensity. The gait data includes stride frequency, stride length, and gait cycle. The knee joint motion data, environmental data, and gait data are input into a knee joint motion risk prediction model to predict the user's knee joint motion risk in the future, where the future cycle is any value between 0.1 seconds and 5 seconds. When the knee joint motion risk exceeds a preset threshold, the exoskeleton is controlled via a knee joint assistive device to adjust the knee joint's flexion-extension damping and range of motion. This invention identifies movement intention through a risk prediction model and, based on a shape memory alloy variable stiffness actuator and gait patterns, achieves real-time monitoring and personalized protection of knee joint motion risk.

[0117] Those skilled in the art will understand that modules in the device of the embodiments can be adaptively changed and placed in one or more devices different from that embodiment. Modules, units, or components in the embodiments can be combined into a single module, unit, or component, and further, they can be divided into multiple sub-modules, sub-units, or sub-components. Except where at least some of such features and / or processes or units are mutually exclusive, any combination of all features disclosed in this specification (including the accompanying claims, abstract, and drawings) and all processes or units of any method or device so disclosed can be employed. Unless expressly stated otherwise, each feature disclosed in this specification (including the accompanying claims, abstract, and drawings) may be replaced by an alternative feature that serves the same, equivalent, or similar purpose. Furthermore, those skilled in the art will understand that although some embodiments herein include certain features included in other embodiments but not others, combinations of features from different embodiments are intended to be within the scope of the invention and form different embodiments. For example, in the following claims, any of the claimed embodiments can be used in any combination. The invention can be implemented by means of hardware comprising several different elements and by means of a suitably programmed computer. In the unit claims listing several devices, several of these devices may be embodied by the same hardware item. Unless otherwise specified, the steps in the above embodiments should not be construed as limiting the order of execution.

Claims

1. An adaptive knee joint protection method based on predicted motion risk, characterized in that, include: The system utilizes an embedded inertial measurement unit, pressure sensor, electromyography sensor, and depth camera to collect real-time data on the user's knee joint motion, environment, and gait. The knee joint motion data includes angular velocity, angular acceleration, joint angle, ground reaction force data, and muscle activity data. The environment data includes ground material, slope, and light intensity. The gait data includes stride frequency, stride length, and gait cycle. The knee joint motion data, environmental data, and gait data are input into the knee joint motion risk prediction model to predict the user's knee joint motion risk in the future period, wherein the future period is any value between 0.1 seconds and 5 seconds. When the risk of knee joint movement exceeds a preset threshold, the exoskeleton is controlled by the knee joint assist device to adjust the flexion and extension damping and range of motion of the knee joint. The knee joint motion risk prediction model adopts a spatiotemporal attention-enhanced long short-term memory network model. The loss function of the spatiotemporal attention-enhanced long short-term memory network model is: ； in, This is a time-varying class weight function based on gated recurrent units; ; It is the sigmoid activation function; Let be the hidden state of the GRU at time t-1; The regularization coefficient is used. It is the Frobenius norm; The total number of samples; Number of risk levels; Let be the true probability that the i-th sample belongs to the c-th category; Let be the predicted probability that the i-th sample belongs to the c-th category; The third derivative over time; The output layer of the spatiotemporal attention-enhanced long short-term memory network model uses a dual-branch structure to predict acute injury risk values ​​and chronic strain indices, respectively. The fusion formula for the knee joint movement risk is as follows: ； in, This represents the risk value for acute injury. It is a chronic strain index; These are dynamic weighting coefficients; This is the time decay factor; The moment of commencement of the movement; The current time; The knee joint assist device includes a variable stiffness actuator based on shape memory alloy, which establishes a two-way coupling relationship between the temperature field and the stress field through the SMA phase transformation dynamics model. The SMA phase transition dynamics model predicts the drive current optimizer in the finite time domain of each control cycle. The bidirectional coupling relationship is as follows: ； in, The density of the material; Specific heat capacity; For temperature; Thermal conductivity; For time; For stress; In response to the situation; Latent heat of phase transition; For stress divergence; External force density; For acceleration; The finite-time domain optimization expression for each control cycle is as follows: ； in, Current state The extent of injury to the lower knee joint; Let be the expected risk value at time k; It is a positive definite weight matrix; Let be the driving current at time k.

2. The adaptive knee joint protection method based on predicted motion risk according to claim 1, characterized in that, The method further includes: A six-degree-of-freedom transformation matrix from the sensor coordinate system to the human anatomical coordinate system is established, and the spatial attitude is estimated in real time through extended Kalman filtering; Wavelet packet decomposition was used to perform a 5-level decomposition of electromyographic signals, and energy entropy features were extracted as priors of motor intent. Kinematic chain constraints are constructed using the 3D coordinates of the joints obtained by a depth camera, and optimization equations are established using the Lagrange multiplier method.

3. The adaptive knee joint protection method based on predicted motion risk according to claim 2, characterized in that, The state equation for the spatial attitude is: ； in, , These are quaternions, angular velocity, and gyroscope drift, respectively. Let k be the state vector at time k; This is the state transition function; This is the control input at time k; This is process noise; Let be the observation vector at time k; For observation functions; To observe noise; The optimization equation is: ； in, The joint angle vector; This refers to the number of key points; The three-dimensional coordinate measurement value of the i-th joint point in the depth camera coordinate system; The three-dimensional coordinates of the i-th joint point are obtained by forward kinematics calculation. This is the Jacobian matrix used to describe the relationship between joint angular velocity and joint point velocity; The target joint velocity; To balance the weights between visual measurement error and motion intent tracking.

4. The adaptive knee joint protection method based on predicted motion risk according to claim 1, characterized in that, The method further includes: A user gait pattern library is established using a hidden Markov model. The user gait pattern library includes normal gait, abnormal forward lean gait, excessive internal rotation gait, excessive external rotation gait, hip abduction gait, foot drop gait, scissor gait, short leg gait, staggering gait, forward lunge gait, and high leg lift gait.

5. The adaptive knee joint protection method based on predicted motion risk according to claim 4, characterized in that, The step of establishing a user gait pattern library using a hidden Markov model further includes: A left-right type HMM structure is initialized for each gait pattern contained in the user gait pattern library; wherein the number of hidden states is automatically determined by the Bayesian information criterion. The parameters of the left-right type HMM structure are re-estimated according to the Baum-Welch algorithm; The forward probability of the observed sequence belonging to a left-right HMM structure is calculated in real time, and the optimal state sequence of the forward probability is decoded by the Viterbi algorithm.

6. An adaptive knee joint protection device based on predicted motion risk, characterized in that, include: The data acquisition module is used to collect real-time user knee joint motion data, environmental data, and gait data using an embedded inertial measurement unit, pressure sensor, electromyography sensor, and depth camera. The knee joint motion data includes angular velocity, angular acceleration, joint angle, ground reaction force data, and muscle activity data. The environmental data includes ground material, slope, and light intensity. The gait data includes stride frequency, stride length, and gait cycle. The risk prediction module is used to input the knee joint motion data, environmental data and gait data into the knee joint motion risk prediction model to predict the user's knee joint motion risk in the future period, wherein the future period is any value between 0.1 seconds and 5 seconds. The knee joint protection module is used to control the exoskeleton through the knee joint assist device to adjust the flexion and extension damping and range of motion of the knee joint when the risk of knee joint movement exceeds a preset threshold. The knee joint motion risk prediction model adopts a spatiotemporal attention-enhanced long short-term memory network model. The loss function of the spatiotemporal attention-enhanced long short-term memory network model is: ； in, This is a time-varying class weight function based on gated recurrent units; ; It is the sigmoid activation function; Let be the hidden state of the GRU at time t-1; The regularization coefficient is used. It is the Frobenius norm; The total number of samples; Number of risk levels; Let be the true probability that the i-th sample belongs to the c-th category; Let be the predicted probability that the i-th sample belongs to the c-th category; The third derivative over time; The output layer of the spatiotemporal attention-enhanced long short-term memory network model uses a dual-branch structure to predict acute injury risk values ​​and chronic strain indices, respectively. The fusion formula for the knee joint movement risk is as follows: ； in, This represents the risk value for acute injury. It is a chronic strain index; These are dynamic weighting coefficients; This is the time decay factor; The moment of commencement of the movement; The current time; The knee joint assist device includes a variable stiffness actuator based on shape memory alloy, which establishes a two-way coupling relationship between the temperature field and the stress field through the SMA phase transformation dynamics model. The SMA phase transition dynamics model predicts the drive current optimizer in the finite time domain of each control cycle. The bidirectional coupling relationship is as follows: ； in, The density of the material; Specific heat capacity; For temperature; Thermal conductivity; For time; For stress; In response to the situation; Latent heat of phase transition; For stress divergence; External force density; For acceleration; The finite-time domain optimization expression for each control cycle is as follows: ； in, Current state The extent of injury to the lower knee joint; Let be the expected risk value at time k; It is a positive definite weight matrix; Let be the driving current at time k.

7. A computer device, comprising: The processor, memory, communication interface, and communication bus are provided, wherein the processor, memory, and communication interface communicate with each other via the communication bus. The memory is used to store at least one executable instruction, which causes the processor to perform the operation corresponding to the adaptive knee joint protection method based on predictive motion risk as described in any one of claims 1-5.

Images