A control method and system for an electrohydraulic active ankle prosthesis

Abstract

This application discloses a control method and system for an electrohydraulic active ankle prosthesis, relating to the fields of lower limb gait recognition and prosthesis control. The method includes: extracting plantar pressure features based on plantar pressure signals; obtaining the division results of four gait stages—passive plantar flexion, passive dorsiflexion, active plantar flexion, and active dorsiflexion—based on the plantar pressure features and a trained SVM model; deriving the real-time gait stages based on the division results of the four gait stages, combined with the state division of a finite state machine based on ankle angle signals and plantar switches, through a weighted decision model; generating control signals for prosthesis movement based on the real-time gait stages and fluid pressure signals; and controlling the movement of the electrohydraulic active ankle prosthesis based on the control signals. This application can accurately identify lower limb gait, thereby achieving accurate control of the prosthesis based on the lower limb gait.

Description

Technical Field

[0001] This application relates to the field of lower limb gait recognition and prosthetic control, and in particular to a control method and system for an electro-hydraulic active ankle prosthesis. Background Technology

[0002] Limb loss causes significant inconvenience to the daily lives and work of people with disabilities. Prostheses are crucial tools for helping people with limb disabilities regain mobility, and the stability and flexibility of the ankle joint play a vital role in human movement and coordination. The ankle joint contributes 60% of the force propelling the body forward and upward. For patients with lower leg amputations, fitting an active ankle prosthesis, such as an electro-hydraulic active ankle prosthesis (i.e., an electrohydraulic ankle prosthesis), can restore basic motor function and help them regain confidence. Specifically, an electro-hydraulic active ankle prosthesis is an active ankle prosthesis based on the principle of electro-hydraulic actuation, meaning it is an active ankle prosthesis driven by an electro-hydraulic actuator.

[0003] However, achieving a perfect fit between a prosthesis and a person with disabilities requires more than just the mechanics and circuitry of the prosthesis itself. The key lies in ensuring that the prosthesis coordinates with the physiological movements of the person with disabilities during use, thereby facilitating stable and comfortable walking. In this regard, accurate gait recognition and appropriate control methods become crucial technical support. Accurate gait recognition technology helps us gain a deeper understanding of the walking patterns and habits of people with disabilities, providing crucial information for personalized adjustments to the prosthesis. By analyzing gait characteristics, we can determine important parameters such as gait phase, stride length, gait speed, and state stages for each individual, providing foundational data for subsequent prosthetic control. Simultaneously, appropriate control methods are key to achieving coordinated movement between the prosthesis and the person with disabilities. Suitable control algorithms and strategies ensure that the prosthesis moves with timing and force matching the user's physiological movements, providing a more stable and natural walking experience. Therefore, accurate lower limb gait recognition and its extended prosthetic control methods are not only a supplement to prosthetic design and manufacturing but also a key technology for ensuring a proper fit between the prosthesis and the person with disabilities, achieving compliant interactive movement. However, current control methods for electro-hydraulic active ankle prostheses cannot accurately identify lower limb gait, and therefore cannot achieve accurate control of the prosthesis based on lower limb gait. Summary of the Invention

[0004] The purpose of this application is to provide a control method and system for an electro-hydraulic active ankle prosthesis, which can accurately identify the lower limb gait and thus achieve accurate control of the prosthesis based on the lower limb gait.

[0005] To achieve the above objectives, this application provides the following solution:

[0006] In a first aspect, this application provides a control method for an electrostatic active ankle prosthesis, the control method for the electrostatic active ankle prosthesis comprising:

[0007] Acquire raw fluid pressure signals, raw ankle angle signals, and raw plantar pressure signals;

[0008] The original liquid pressure signal, the original ankle angle signal, and the original plantar pressure signal are preprocessed to obtain the liquid pressure signal, the ankle angle signal, and the plantar pressure signal; the plantar pressure signal includes the pressure signal at each point in the plantar pressure array.

[0009] The plantar pressure features are extracted based on the plantar pressure signals; the plantar pressure features include the overall foot area ground reaction force, the forefoot area ground reaction force, the midfoot area ground reaction force, the hindfoot area ground reaction force, the x-coordinate of the plantar pressure center trajectory, the y-coordinate of the plantar pressure center trajectory, and the maximum single-point impulse;

[0010] Based on the plantar pressure characteristics, the four gait phases of passive plantar flexion, passive dorsiflexion, active plantar flexion, and active dorsiflexion were divided into four stages based on the trained SVM model. The trained SVM model was obtained by training the SVM model with the overall foot area ground reaction force, forefoot area ground reaction force, midfoot area ground reaction force, hindfoot area ground reaction force, x-coordinate of the plantar pressure center trajectory, y-coordinate of the plantar pressure center trajectory, and maximum single-point impulse of different subjects in walking mode, and distinguishing the four gait phases of passive plantar flexion, passive dorsiflexion, active plantar flexion, and active dorsiflexion.

[0011] Based on the division results of the four gait stages of passive plantar flexion, passive dorsiflexion, active plantar flexion and active dorsiflexion, and combined with the state division of the finite state machine based on the ankle angle signal and the plantar switch, the real-time gait stage is obtained through a weighted decision model; the plantar switch is a pressure switch; the pressure signals of the two regions at the foremost and rearmost ends of the plantar pressure array are used as a pressure switch.

[0012] Based on the real-time gait phase and the fluid pressure signal, control signals for prosthetic movement are generated;

[0013] The electrostatic active ankle joint prosthesis movement is controlled according to the control signal of the prosthesis movement.

[0014] Optionally, the original fluid pressure signal, the original ankle angle signal, and the original plantar pressure signal are preprocessed to obtain the fluid pressure signal, ankle angle signal, and plantar pressure signal, specifically including:

[0015] Fourier transforms are performed on the original fluid pressure signal, the original ankle angle signal, and the original plantar pressure signal to obtain the frequency domain distributions of the original fluid pressure signal, the original ankle angle signal, and the original plantar pressure signal.

[0016] The frequency domain distributions of the original fluid pressure signal, the original ankle angle signal, and the original plantar pressure signal are subjected to low-pass filtering to obtain the fluid pressure signal, ankle angle signal, and plantar pressure signal.

[0017] Optionally, extracting plantar pressure features based on the plantar pressure signal specifically includes:

[0018] Based on the plantar pressure signal, the overall foot area ground reaction force, forefoot area ground reaction force, midfoot area ground reaction force, hindfoot area ground reaction force, x-coordinate of plantar pressure center trajectory, y-coordinate of plantar pressure center trajectory, and maximum single-point impulse are extracted using the plantar pressure center trajectory and pressure cloud map during walking.

[0019] Optionally, the weighted decision model is expressed as:

[0020]

[0021] Where, x i The plantar pressure signal, y i This indicates the ankle angle signal. This represents the weight of the plantar pressure signal in the i-th gait phase. This represents the weight of the ankle angle signal in the i-th gait phase, where i∈{1,2,3,4} represent the four gait phases respectively, i=1 represents passive plantar flexion, i=2 represents passive dorsiflexion, i=3 represents active plantar flexion, and i=4 represents active dorsiflexion. d This represents the weighted decision result, and `round` represents the rounding function, used to convert the weighted average result into the closest integer gait stage.

[0022] Optionally, control signals for prosthetic movement are generated based on the real-time gait phase and the fluid pressure signal, specifically including:

[0023] Based on the real-time gait phase and the fluid pressure signal, combined with the optimized typical event-triggered finite state machine and the electro-hydraulic active ankle joint prosthesis hydraulic model, kinematic and dynamic models, control signals for prosthesis movement are generated; the control signals for prosthesis movement include motor output power and high-speed switching valve opening.

[0024] Optionally, controlling the movement of the electro-hydraulic active ankle prosthesis according to the control signal for the prosthesis movement specifically includes:

[0025] The motor of the electro-hydraulic active ankle joint prosthesis is controlled according to the output power of the motor, and the high-speed switching valve of the electro-hydraulic active ankle joint prosthesis is controlled according to the opening degree of the high-speed switching valve. The electro-hydraulic active ankle joint prosthesis moves under the control of the motor and the high-speed switching valve.

[0026] Secondly, this application provides a control system for an electrohydraulic active ankle prosthesis, based on the aforementioned control method for an electrohydraulic active ankle prosthesis; the control system for the electrohydraulic active ankle prosthesis includes:

[0027] A liquid pressure sensor is installed on the hydraulic integration block of the electro-hydraulic active ankle prosthesis and connected to the internal hydraulic circuit of the electro-hydraulic active ankle prosthesis to acquire the raw liquid pressure signal.

[0028] An ankle angle sensor is installed at the ankle of an electro-hydraulic active ankle joint prosthesis to acquire raw ankle angle signals;

[0029] A multi-point plantar pressure array sensor is installed in the insole of an electro-hydraulic active ankle prosthesis to acquire raw plantar pressure signals.

[0030] A signal and control circuit board, connected to the liquid pressure sensor, the ankle angle sensor, and the multi-point plantar pressure array sensor, is used to preprocess the original liquid pressure signal, the original ankle angle signal, and the original plantar pressure signal to obtain liquid pressure signal, ankle angle signal, and plantar pressure signal. Plantar pressure features are extracted from the plantar pressure signal. Based on these features, a trained SVM model is used to obtain the division results of four gait stages: passive plantar flexion, passive dorsiflexion, active plantar flexion, and active dorsiflexion. Based on these gait stage division results, combined with the finite state machine state partitioning based on the ankle angle signal and plantar switches, a weighted decision model is used to derive the real-time gait stage. Based on the real-time gait stage and the liquid pressure signal, a false gait stage is generated. The control signal for limb movement includes: plantar pressure signals comprising pressure signals at various points in the plantar pressure array; plantar pressure characteristics comprising: overall foot area ground reaction force, forefoot area ground reaction force, midfoot area ground reaction force, hindfoot area ground reaction force, x-coordinate of the plantar pressure center trajectory, y-coordinate of the plantar pressure center trajectory, and maximum single-point impulse; the trained SVM model is obtained by training the SVM model using the overall foot area ground reaction force, forefoot area ground reaction force, midfoot area ground reaction force, hindfoot area ground reaction force, x-coordinate of the plantar pressure center trajectory, y-coordinate of the plantar pressure center trajectory, and maximum single-point impulse of different subjects in walking modes, distinguishing four gait stages: passive plantar flexion, passive dorsiflexion, active plantar flexion, and active dorsiflexion; the plantar switch is a pressure switch; the pressure signals of the two regions at the foremost and rearmost ends of the plantar pressure array serve as a pressure switch;

[0031] The power supply and drive circuit board is connected to the signal and control circuit board, as well as the high-speed switching valve, motor, and battery of the electro-hydraulic active ankle joint prosthesis, and is used to control the movement of the electro-hydraulic active ankle joint prosthesis according to the control signal of the prosthesis movement.

[0032] Optionally, the ankle angle sensor is a potentiometer-type angle sensor, which is installed at the ankle of the electro-hydraulic active ankle joint prosthesis and fixed to the electro-hydraulic active ankle joint prosthesis through a stepped shaft and flange structure, and is used to obtain the real-time angle of the footplate of the electro-hydraulic active ankle joint prosthesis.

[0033] Optionally, the multi-point plantar pressure array sensor consists of 18 single-point resistive thin-film pressure sensors, which are installed in the insole of the electrostatic active ankle prosthesis and contact the silicone foot skin worn on the foot plate of the electrostatic active ankle prosthesis.

[0034] Optionally, the signal and control circuit board is equipped with a microcontroller. The microcontroller is used to preprocess the original fluid pressure signal, the original ankle angle signal, and the original plantar pressure signal to obtain the fluid pressure signal, ankle angle signal, and plantar pressure signal. Based on the plantar pressure signal, plantar pressure features are extracted. Based on the plantar pressure features, a trained SVM model is used to obtain the division results of four gait stages: passive plantar flexion, passive dorsiflexion, active plantar flexion, and active dorsiflexion. Based on the division results of the four gait stages, combined with the state division of the finite state machine based on the ankle angle signal and the plantar switch, a weighted decision model is used to obtain the real-time gait stage. Based on the real-time gait stage and the fluid pressure signal, a control signal for prosthetic movement is generated, and the control signal for prosthetic movement is transmitted to the power supply and drive circuit board.

[0035] According to the specific embodiments provided in this application, this application has the following technical effects:

[0036] This application provides a control method and system for an electro-hydraulic active ankle prosthesis. Based on extracted plantar pressure features including overall foot area ground reaction force, forefoot area ground reaction force, midfoot area ground reaction force, hindfoot area ground reaction force, x-coordinate of plantar pressure center trajectory, y-coordinate of plantar pressure center trajectory, and maximum single-point impulse, a trained SVM model is used to obtain the division results of four gait stages: passive plantar flexion, passive dorsiflexion, active plantar flexion, and active dorsiflexion. Then, based on the division results of these four gait stages, combined with ankle angle signals and plantar pressure data, a control method and system are developed. The finite state machine partitioning of the switch yields real-time gait stages through a weighted decision model. By incorporating tripedal pressure distribution characteristics into traditional plantar pressure gait features, the multivariate nonlinear characteristics in gait data can be better captured, improving the accuracy of gait segmentation. By employing a weighted decision strategy, combined with multi-source information and sensor reliability, precise control of key events and optimization of gait stages are achieved, ensuring that the prosthesis can adapt to the wearer's needs during walking. Ultimately, this achieves the effect of accurately identifying lower limb gait and thus enabling accurate control of the prosthesis based on lower limb gait. Attached Figure Description

[0037] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0038] Figure 1A flowchart illustrating a control method for an electrohydraulic active ankle prosthesis according to an embodiment of this application;

[0039] Figure 2 This is the overall flowchart of the ankle prosthesis gait detection method of this application;

[0040] Figure 3 This is a block diagram of the overall design of the controller in the ankle prosthesis of this application;

[0041] Figure 4 This is a simplified diagram of the dynamic structure of the ankle prosthesis in this application;

[0042] Figure 5 A diagram showing the correspondence between the four gait stages during human walking and the operating conditions of the ankle prosthesis hydraulic system;

[0043] Figure 6 This is the equivalent circuit diagram of the hydraulic drive system in passive state;

[0044] Figure 7 This is a block diagram showing the control of ankle joint impedance characteristics under passive conditions.

[0045] Figure 8 The equivalent circuit diagram of the hydraulic drive system under active plantar flexion operation;

[0046] Figure 9 This is a block diagram of the joint output torque control under active plantar flexion.

[0047] Figure 10 The equivalent circuit diagram of the hydraulic drive system under active dorsiflexion state;

[0048] Figure 11 A schematic diagram of ankle angle servoing for an ankle prosthesis in active dorsiflexion state;

[0049] Figure 12 This is a hardware layout diagram of the ankle prosthesis gait detection system of this application. Detailed Implementation

[0050] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0051] The purpose of this application is to provide a control method and system for an electro-hydraulic active ankle prosthesis, which can accurately identify the lower limb gait and thus achieve accurate control of the prosthesis based on the lower limb gait.

[0052] To make the above-mentioned objectives, features and advantages of this application more apparent and understandable, the application will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0053] like Figure 1 As shown, this application provides a control method for an electrostatic active ankle prosthesis, comprising:

[0054] Step 101: Acquire the raw fluid pressure signal, raw ankle angle signal, and raw plantar pressure signal.

[0055] Step 102: Preprocess the original liquid pressure signal, original ankle angle signal, and original plantar pressure signal to obtain the liquid pressure signal, ankle angle signal, and plantar pressure signal; the plantar pressure signal includes the pressure signal of each point in the plantar pressure array.

[0056] Step 102 specifically includes:

[0057] Fourier transforms were performed on the original fluid pressure signal, the original ankle angle signal, and the original plantar pressure signal to obtain the frequency domain distributions of the original fluid pressure signal, the original ankle angle signal, and the original plantar pressure signal.

[0058] Low-pass filtering is performed on the frequency domain distributions of the original fluid pressure signal, the original ankle angle signal, and the original plantar pressure signal to obtain the fluid pressure signal, ankle angle signal, and plantar pressure signal.

[0059] Step 103: Extract plantar pressure features based on plantar pressure signals; plantar pressure features include overall foot area ground reaction force, forefoot area ground reaction force, midfoot area ground reaction force, hindfoot area ground reaction force, x-coordinate of plantar pressure center trajectory, y-coordinate of plantar pressure center trajectory, and maximum single-point impulse.

[0060] Step 103 specifically includes:

[0061] Based on the plantar pressure signal, the ground reaction force of the entire foot area, the ground reaction force of the forefoot area, the ground reaction force of the midfoot area, the ground reaction force of the hindfoot area, the x-coordinate of the plantar pressure center trajectory, the y-coordinate of the plantar pressure center trajectory, and the maximum single-point impulse are extracted using the plantar pressure center trajectory and the pressure cloud map during walking.

[0062] Step 104: Based on the plantar pressure characteristics, obtain the classification results of the four gait stages of passive plantar flexion, passive dorsiflexion, active plantar flexion, and active dorsiflexion based on the trained SVM model; the trained SVM model is obtained by using the overall foot area ground reaction force, forefoot area ground reaction force, midfoot area ground reaction force, hindfoot area ground reaction force, x-coordinate of the plantar pressure center trajectory, y-coordinate of the plantar pressure center trajectory, and maximum single-point impulse of different subjects in the walking mode to train the SVM model and distinguish the four gait stages of passive plantar flexion, passive dorsiflexion, active plantar flexion, and active dorsiflexion.

[0063] In step 104, the weighted decision model is represented as follows:

[0064]

[0065] Where, x i Indicates plantar pressure signal, y i Indicates the ankle angle signal. This represents the weight of the plantar pressure signal in the i-th gait phase. S represents the weight of the ankle angle signal in the i-th gait phase, where i∈{1,2,3,4} represent the four gait phases respectively, i=1 represents passive plantar flexion, i=2 represents passive dorsiflexion, i=3 represents active plantar flexion, and i=4 represents active dorsiflexion. d This represents the weighted decision result, and `round` represents the rounding function, used to convert the weighted average result into the closest integer gait stage.

[0066] Step 105: Based on the division results of the four gait stages of passive plantar flexion, passive dorsiflexion, active plantar flexion and active dorsiflexion, and combined with the state division of the finite state machine based on the ankle angle signal and the plantar switch, the real-time gait stage is obtained through a weighted decision model; the plantar switch is a pressure switch; the pressure signals of the two regions at the foremost and rearmost ends of the plantar pressure array are used as a pressure switch.

[0067] Step 106: Generate control signals for prosthetic movement based on real-time gait phase and fluid pressure signals.

[0068] Step 106 specifically includes:

[0069] Based on real-time gait phase and fluid pressure signals, combined with an optimized finite state machine triggered by typical events, as well as the hydraulic, kinematic, and dynamic models of the electro-hydraulic active ankle prosthesis, control signals for prosthesis movement are generated. The control signals for prosthesis movement include motor output power and high-speed switching valve opening.

[0070] Step 107: Control the movement of the electro-hydraulic active ankle prosthesis according to the control signal of the prosthesis movement.

[0071] Step 107 specifically includes:

[0072] The motor of the electro-hydraulic active ankle prosthesis is controlled by the motor output power, and the high-speed switching valve of the electro-hydraulic active ankle prosthesis is controlled by the opening degree of the high-speed switching valve. The electro-hydraulic active ankle prosthesis moves under the control of the motor and the high-speed switching valve.

[0073] The following specific embodiment illustrates the technical solution of the control method for an electrostatic active ankle prosthesis of this application:

[0074] This application relates to a control method for an electro-hydraulic active ankle prosthesis, specifically a gait detection and control method for wearable ankle prostheses. The gait detection and control method includes the acquisition and preprocessing of raw signals from sensors such as plantar pressure and ankle angle; feature extraction of plantar pressure, including pressure distribution in the tripedal region and pressure center traces; misclassification optimization of the training set of the gait phase SVM classification model; top-level state classification combining plantar pressure array and ankle angle information; prosthesis dynamics model; and top-middle-bottom multi-layer control. Ultimately, the prosthesis can effectively identify and judge human gait information to determine control strategies, adjust walking damping, and control output timing.

[0075] The working principle of hydraulic drive for ankle prostheses is usually as follows: the hydraulic pump is driven by a drive motor to draw hydraulic oil from the tank and supply it to the hydraulic cylinder through pipelines according to the state of the reversing valve to provide power to the foot.

[0076] like Figure 2 The diagram shows the overall flowchart of the ankle prosthesis gait detection method of this application. It consists of steps such as sensor raw signal acquisition, raw signal preprocessing, plantar pressure feature extraction, gait phase classification model (i.e., classification SVM model) deployment, ankle angle finite state machine determination (i.e., ankle angle threshold determination), top-level state classification based on multi-sensor information fusion (i.e., multi-source information fusion state classification), and prosthesis dynamics model and control. It reflects the entire process from human-machine walking information acquisition to human-machine collaborative control.

[0077] The acquisition of raw sensor signals is divided into the acquisition of plantar pressure, ankle angle, and fluid pressure signals. Plantar pressure refers to the pressure generated between the sole of the foot and the ground when a person walks, runs, jumps, or stands. With each step, the sole of the foot bears the pressure of the body weight. The plantar pressure signal is acquired through a multi-point plantar pressure array sensor installed in the insole. Each point represents the plantar pressure in that area (i.e., the pressure value at each point in the array), used to determine the gait stage (see Plantar Pressure Feature Extraction for details; gait stage determination can be achieved through feature extraction and data analysis combined with machine learning). The plantar pressure is a 3×6 sensor array, and each area can sense the corresponding pressure value based on strain. The ankle angle signal (i.e., the specific angle change data of the ankle angle) is acquired through a potentiometer-type angle sensor installed at the prosthetic ankle, used to set the switching threshold (based on experimental data obtained from multiple tests, the angle change during gait switching is very clear, allowing for the setting of the switching threshold). The fluid pressure signal is acquired through a fluid pressure sensor installed on the hydraulic integrated block and connected to the oil circuit, used to detect the internal state of the prosthetic power system (the prosthesis is hydraulically driven; the internal state of the prosthesis can be determined by measuring the oil pressure at the upper and lower ends of the hydraulic cylinder and the motor outlet).

[0078] In the preprocessing of the original signals, Fourier transforms are performed on signals such as plantar pressure (i.e., Fourier transforms are performed on plantar pressure signals, ankle angle signals, and fluid pressure signals to reduce high-frequency noise). This yields the frequency domain distribution of the signals. The Fourier transform results show that signals generated by plantar pressure, etc., are concentrated in the low-frequency band, with almost no high-frequency information. Therefore, signals above 5Hz are filtered. Subsequent analysis and feature extraction of plantar pressure signals focus on time domain information. Then, the data is standardized and preprocessed, and the sliding window method is used to extract data features, eliminating scale differences between features and ensuring that the optimization algorithm converges faster, thereby improving the training speed and performance of the classifier. In short, the original signal preprocessing involves sequentially low-pass filtering the frequency domain distributions of plantar pressure signals, ankle angle signals, and fluid pressure signals to eliminate the influence of high-frequency noise and obtain more realistic sensor data.

[0079] In plantar pressure feature extraction, the first step is to analyze the characteristics of the plantar pressure signal (similar to an exploratory analysis and pattern summarization of preprocessed signals). In gait analysis, the analysis of the Center of Pressure (COP) is typically used to assess foot stability and the health of walking posture. The COP describes the trajectory of the point of application of the plantar reaction force on the ground during standing or walking, changing over time. COP is usually measured using devices such as pressure sensor arrays or force platforms, representing the pressure distribution applied at different locations on the sole of the foot. However, relying solely on COP as an analytical indicator may not provide detailed information on the pressure distribution between different areas within the foot. By observing the pressure cloud map during walking (plotting the sensor array as an XY plane image, with the pressure at each point represented on the image to obtain the pressure cloud map), the foot is divided into three regions—the forefoot, midfoot, and hindfoot—based on its physiological structure and functional characteristics. Each region bears a pressure load that is staggered over time. For example, the hindfoot experiences the greatest impact force upon landing; during the propulsion phase of walking, the pressure on the forefoot reaches its maximum, while the midfoot provides stability and cushioning throughout the gait cycle. Furthermore, in the stance phase (a medical term for a gait phase), the pressure contour map shows significant maxima at hindfoot landing and forefoot push-off. Therefore, seven features—local reaction force in the three foot regions, the plantar pressure center trace, and the maximum single-point impulse—are extracted as the basis for classification.

[0080] Specifically, the ground reaction force in the foot area is extracted based on data from each point in the plantar pressure array: Foot pressure center trace: Maximum single-point impulse (i.e., the pressure value at the point of maximum pressure in the plantar pressure array at a certain moment): x and y are the pressure coordinates of each point in the plantar pressure array. In the formula, x(i) represents the value of the i-th data point, MAV represents the mean absolute value, N represents the total number of data points, and P(x,y) represents the pressure value of the point with coordinates (x,y) in the plantar pressure matrix. i ,y i ) represents the pressure value at the i-th point, x i and y i COP represents the X-axis and Y-axis coordinates of the i-th point in the plantar pressure matrix, respectively. X and COP Y P represents the X-axis and Y-axis coordinates of the center of plantar pressure, respectively. i A represents the pressure value at the i-th pressure point. i f represents the area of ​​the i-th pressure point. s This indicates the sampling frequency, used to calculate impulse.

[0081] The challenge in deploying gait phase classification models lies in the fact that gait data often contains complex nonlinear relationships. Human gait can be assumed to be a nonlinear function of human-computer interaction information, which is difficult to obtain through mathematical modeling. Machine learning methods, however, have strong nonlinear modeling capabilities and can better capture the nonlinear features in gait data. Among them, SVM seeks the optimal trade-off between model complexity and learning ability based on limited sample information, and has excellent generalization ability.

[0082] The SVM model was trained using seven gait features from different subjects in different walking modes (including passive plantar flexion, passive dorsiflexion, active plantar flexion, and active dorsiflexion, with cyclic switching between the four gait modes during walking). These features included global foot ground reaction force, forefoot ground reaction force, midfoot ground reaction force, hindfoot ground reaction force, COP trajectory x-coordinate, COP trajectory y-coordinate, and maximum single-point pressure (i.e., maximum single-point impulse). This allowed the model to distinguish four gait stages: passive plantar flexion, passive dorsiflexion, active plantar flexion, and active dorsiflexion. The model's recognition accuracy was validated using 5-fold cross-validation, and the misclassification cost function was adjusted to improve the recognition accuracy of the active gait stage. This resulted in an SVM classification model optimized for individuals (because the model can be trained on datasets from different subjects, incorporating the subject's unique walking habits and information, thus yielding more accurate and personalized gait stage recognition). The forefoot region includes the toes (phalanges) and metatarsals, and is the part of the foot closest to the toes; the midfoot region is located between the forefoot and hindfoot regions and is mainly composed of tarsal bones; the hindfoot region includes the calcaneus and talus, and is the part of the foot closest to the ankle; the sum of the pressure values ​​of the corresponding plantar pressure array sensors in each foot region is the ground reaction force value of that region.

[0083] The finite state machine based on ankle angle determines four gait states by using two single points on the plantar pressure sensor (the pressure values ​​of the foremost and rearmost regions of the plantar pressure array as a simple pressure switch) to determine the ankle angle sensor switching threshold through key gait events such as heel strike (HS), maximum dorsiflexion (MD), and toe lift-off (TO). This part is actually the traditional ankle angle threshold switching approach of finite state machines, but here it is part of sensor fusion, which results in higher recognition accuracy. The specific implementation steps of this part include:

[0084] Standing Phase Phase Control: The three states (CP, CD, TS) are the three phases of the standing phase, and each state is described as follows: (1) The CP phase begins when the heel touches the ground and ends when the entire foot touches the ground. During the CP phase, the prosthesis output characteristic is impedance servo to prevent the toes from touching the ground prematurely and to provide a certain degree of shock absorption and cushioning for the human body. (2) The CD phase begins when the entire foot touches the ground and ends before the TS phase or before the toes leave the ground, depending on whether the ankle angle exceeds the maximum dorsiflexion angle. During the CD phase, the prosthesis output characteristic is impedance servo to maintain smooth body movement and energy recovery from the accumulator. (3) The TS phase is initiated only when the entire foot touches the ground and the ankle angle exceeds the maximum dorsiflexion angle; otherwise, the prosthesis will remain in the CD phase until the foot leaves the ground. During the TS phase, the prosthesis output torque propels the human body forward.

[0085] Swing phase control: Three states (SW1, SW2, SW3) are used for the control of the swing phase. The specific description of each state is as follows: (4) SW1 starts from the toes leaving the ground and extends for a given delay t H (5) SW2 begins after SW1 and ends after the ankle joint angle reaches the predetermined value. During this period, the ankle joint angle reaches the state before the CP phase, preparing for the heel strike in the next cycle. (6) SW3 begins after SW2 and ends when the heel strikes in the next cycle.

[0086] The top-level state classification based on multi-sensor information fusion uses plantar pressure center point and tripedal zone information measured by plantar pressure array sensors to understand and simulate the dynamics of human walking. It combines this with traditional finite state machine states based on ankle angle and plantar switch to achieve the division of the four gait stages of the ankle joint during human walking. Figure 3 The top-level control section.

[0087] This application integrates information from two sensors (i.e., plantar pressure signal and ankle angle signal) to complete the state switching of the prosthetic finite state machine (simply put, a weighted control strategy). Its state transition trigger information (the fused output is four real-time gait phase information, and the transition of gait phase is the state transition trigger information, i.e., the transition of gait phase is synchronized with the transition of controller state phase) not only considers the gait phase division results based on plantar pressure characteristics, but also combines the ankle joint angle measurement and traditional plantar switch signal in the traditional finite state machine. Specifically, it is a weighted allocation method for the importance of the two sensor data obtained through experimental analysis. Assume that S = {CP, CD, TS, PF} is the set of gait phases, corresponding to (1) passive plantar flexion, (2) passive dorsiflexion, (3) active plantar flexion and (4) active dorsiflexion. Considering that the sensitivity and importance of the plantar pressure and ankle joint angle sensors may be different in different gait phases, this application introduces a weighted decision model to divide the gait phases.

[0088] Let x i For data acquired from plantar pressure array sensors, y i The data is obtained from an ankle angle sensor, where i∈{1,2,3,4} represents the four gait stages. The weighted decision model can be represented as:

[0089]

[0090] in, and These are the weights of the plantar pressure array sensor data and the ankle angle data in the i-th gait phase, respectively, S. d It is the weighted decision result. round is a rounding function used to convert the weighted average result into the closest integer gait stage, that is, to obtain the real-time gait stage, and then switch to the underlying servo model of the corresponding gait stage.

[0091] Plantar pressure experiments show that plantar pressure changes are more pronounced during passive plantar flexion, passive dorsiflexion, and active plantar flexion. At these stages, the ankle angle triggering condition of the traditional finite state machine exhibits poor reproducibility in terms of its changing trend and threshold switching between the two passive states. Therefore, the weighting of the plantar pressure array sensor... and The accuracy should be relatively high. Conversely, during active dorsiflexion, the plantar pressure array has lower recognition accuracy, and changes in ankle angle are more critical; therefore, the sensor's weighting is less important. It will be higher. The weights in the weighted decision model. and Optimization can be achieved through experiments using actual gait data from different populations, minimizing the error between the model's predicted gait phases and the actual gait phases to ensure the model's accuracy and robustness.

[0092] Based on the gait characteristics and actual working conditions of the prosthesis, and combining hydraulic, kinematic, and dynamic models, servo impedance, torque, and displacement controllers were designed for the prosthesis's dynamics and control. These controllers can finely adjust the torque output and real-time damping of the prosthetic joint using a PWM-PFM strategy (a general control strategy). (Torque output is controlled by a motor, and damping is controlled by a high-speed switching valve; the combination of the two yields a more realistic and natural gait.) This simulates the physical behavior and mechanical response of the human ankle joint at different gait stages, achieving mechanical characteristics and dynamic responses similar to those of the human ankle joint. The specific dynamic model can be found [link to specific model]. Figure 4 Among them, gait characteristics refer to the medical gait characteristics of the ankle joint during human walking, with the working condition being walking on flat ground. The combination of these two factors determines the basis for gait stage division and the specific ankle joint movements within each gait stage, providing factual evidence for the gait stages obtained through sensor fusion.

[0093] The above section introduced sensor fusion-based state recognition and gait segmentation. The following section introduces the hydraulic model (prosthetic hydraulic formula), which describes the correspondence between the four gait stages of human walking and the working conditions of the ankle prosthesis hydraulic system. This aims to serve as a theoretical basis for subsequent kinematic and dynamic models. After optimizing the finite state machine state triggered by events to achieve the segmentation of the four gait stages of the ankle joint during human walking, the calculations determine how to accurately control the underlying actuators (motor and high-speed switching valve).

[0094] like Figure 5 The diagram illustrates the correspondence between the four gait stages during human walking and the operating conditions of the ankle prosthesis hydraulic system. Both the physiological gait divisions and the operating stages of an active ankle prosthesis consist of passive plantar flexion, passive dorsiflexion, active plantar flexion, and active dorsiflexion, which can be categorized into four parts:

[0095] 1. During the passive plantar flexion phase, the upper and lower chambers of the hydraulic cylinder are connected via valves 3 and 4. When the heel is impacted by the ground, hydraulic oil flows from the lower chamber to the upper chamber. The dynamic model of the hydraulic cylinder at this time is obtained from the following formula:

[0096] m A a A =F A -(P3-P2)A a -F s ;

[0097] Where, m A For load and piston mass, aA 为 The acceleration of the hydraulic cylinder piston rod, P3 and P2 represent the pressures in the upper and lower chambers of the hydraulic cylinder, respectively, and F A For external force, F s For internal friction, A a It is the area of ​​the hydraulic cylinder where the oil acts.

[0098] To ensure the consistency between the hydraulic cylinder's movement and the human body's movement, this application adjusts the overall hydraulic damping by controlling the valve opening of a high-speed switching valve to achieve the PWM duty cycle of the bionic regulating valve 3 in the prosthetic hydraulic system. The damping of the hydraulic system will be adjusted accordingly. The average flow rate through valve 3... It can be represented as:

[0099]

[0100] Where ρ represents the density of the oil, and c d For flow coefficient, This is the equivalent flow cross-sectional area of ​​the on / off valve, and its calculation formula is:

[0101]

[0102] Where θ represents the switching angle of the high-speed switching valve, and D is the diameter of the high-speed switching valve. The average displacement of the high-speed switching valve is controlled by a PWM wave to open and close it. Controlling the duty cycle of the PWM wave controls the opening time of the high-speed switching valve over a period of time, which is equivalent to controlling the size of the throttle orifice. Therefore, the output torque of the prosthesis can be adjusted by regulating the PWM duty cycle. Simultaneously, the DC motor continuously operates, charging the high-voltage accumulator A. The volume change of the accumulator is as follows:

[0103]

[0104] Where V1 and V'1 represent the volume of the accumulator at different times, P1 and P'1 represent the pressure of the accumulator at different times, V0 is the accumulator capacity, and P... a Let ΔV be the charging pressure of accumulator A, ΔV be the volume change of the accumulator, and n be the exponent of the gas variation process inside the accumulator's gas bladder.

[0105] 2. During the passive backbend stage, the upper and lower chambers of the hydraulic cylinder are connected via valves 3 and 4. Hydraulic oil flows from the upper chamber to the lower chamber. Similar to the above, the dynamic equations and pressure equations of the hydraulic cylinder are:

[0106] m A a A =F A -(P2-P3)A a -F s

[0107]

[0108] Where, m A a represents the mass of the piston. A F represents the acceleration of the piston. AA represents the force acting on the piston. a F represents the area of ​​the piston that experiences force. s β represents the static friction force of the system. e The bulk modulus of hydraulic oil, v A This represents the piston speed, Q2 and Q3 represent the flow rates in different pipes, Q2 corresponds to the pipe where P2 is located, and Q3 corresponds to the pipe where P3 is located. a Indicates the position of the piston. This represents the rate of change of P2. V3 represents the rate of change of P3, V2 represents the volume of accumulator B, and V3 represents the volume of accumulator C. At this time, by adjusting the PWM duty cycle of valve 3, the damping of the hydraulic system circuit can be adjusted in real time, achieving real-time adjustment of the ankle joint torque and thus real-time adjustment of the ankle joint impedance characteristics. Furthermore, the motor continuously operates, charging the high-voltage accumulator A, causing the pressure in the high-voltage accumulator to continuously increase.

[0109]

[0110] Among them, P acc P represents the power of the energy storage device. a q represents the charging pressure of accumulator A. acc The flow rate of the accumulator is represented by P1 and P'1, which represent the pressure of accumulator A at different times. V1 and V'1 represent the volume of accumulator A at different times. ΔV represents the volume difference at different times. At the same time, the human body's gravity does work on the hydraulic cylinder, and accumulator C recovers the energy brought by the gravitational potential energy of the human body.

[0111]

[0112] Where V0 is the energy storage capacity, P a The charging pressure of accumulator A, P c V represents the charging pressure of accumulator C, V3 and V'3 represent the volume of accumulator C at different times, and P3 and P'3 represent the pressure of accumulator C at different times.

[0113] 3. During the active plantar flexion phase, the upper chamber of the hydraulic cylinder is connected to the low-pressure accumulator B via valve 1. At this time, the high-pressure accumulator releases the high-pressure oil accumulated during the two passive phases. The released energy drives the hydraulic cylinder to output a momentary high thrust, causing the ankle to propel the body forward. The dynamic equations and pressure equations of the prosthesis are:

[0114] m A a A =F A +(P2-P3)A a -F s

[0115]

[0116] During this process, the pressure in the lower chamber of the hydraulic cylinder can be adjusted by regulating the PWM duty cycle of valve 3, making the output of the ankle prosthesis approximate the output of a biological ankle joint. Specifically, in this state, the hydraulic pump and accumulator simultaneously act on the actuator. The power of the hydraulic pump is:

[0117] P pum p = P1V pumpn η0 / 60;

[0118] Among them, V pump P is the displacement of the hydraulic pump. pump Let η0 represent the power of the hydraulic pump, η0 represent the overall efficiency of the hydraulic pump, and n represent the rotational speed of the hydraulic pump. At this point, the total input power of the hydraulic system is the sum of the input power of the hydraulic pump and the power of the accumulator.

[0119] 4. During the active dorsiflexion phase, the hydraulic system is similar to a traditional hydraulic drive system. An electric motor drives the hydraulic pump, which in turn moves the hydraulic cylinder, preparing for the next gait cycle. The dynamic equations and pressure equations of the hydraulic cylinder are as follows:

[0120] m a a A =F A +(P3-P2)A a -F s

[0121]

[0122] This allows us to derive the correspondence between the four gait stages during human walking and the operating conditions of the ankle prosthesis hydraulic system, revealing the influence of motor input and valve opening on the displacement, velocity, acceleration, and output force of the hydraulic cylinder. Combined with... Figure 4 The kinematics of prostheses can be used to construct a complete mathematical model of prostheses.

[0123] like Figure 3 The diagram shown is an overall design block diagram of the controller in the ankle prosthesis of this application. The controller aims to achieve accurate fitting of the gait of a real biological ankle joint. The system consists of three core logic layers: a decision fusion layer (top layer), a state transition control layer (middle layer), and a servo impedance control layer (bottom layer).

[0124] The top layer integrates information from the plantar pressure array sensor and the ankle angle sensor, performing gait phase identification, i.e., multi-sensor gait phase identification and segmentation (top-level state classification based on multi-sensor information fusion). The plantar pressure sensor analyzes characteristic changes such as pressure distribution and pressure center traces in the tripedal region to provide accurate determination of each phase of the gait cycle, generating gait phase information. The ankle angle sensor and plantar switch provide gait phase information through traditional finite state machine logic and partitioning methods. This information is fused using a decision-level advantage weight algorithm to obtain the final gait phase trigger signal, providing an accurate trigger point for system state transitions.

[0125] The middle layer uses a finite state machine with optimized key event triggering to process the gait phase information of the top layer. At this layer, the system, combined with preset threshold constraints, identifies the current gait phase and controls the state transition, determined by a finite state machine based on ankle angle and plantar pressure. The state transition control layer responds to the fused triggering information provided by the decision layer and ensures smooth transfer and switching of the prosthesis between key events in the gait cycle, achieving continuity and naturalness in the prosthetic gait.

[0126] Based on the gait characteristics and actual working conditions of the prosthesis, the underlying control logic incorporates servo impedance, torque, and displacement controllers, combining hydraulic and dynamic models. These controllers can finely adjust the torque output and real-time damping of the prosthetic joints using a PWM-PFM strategy to simulate the physical behavior and mechanical response of the human ankle joint at different gait stages, achieving similar mechanical characteristics and dynamic response to the human ankle joint. The overall control strategy ensures the coordinated operation of each controller layer to simulate the complex movements of the biological ankle joint under different ground surfaces and walking conditions.

[0127] The following section introduces the kinematic and dynamic models, namely the kinematic and dynamic formulas for prostheses:

[0128] like Figure 4 The diagram shown is a simplified kinetic structure of the ankle prosthesis of this application. It analyzes the relationship between the motion of the hydraulic cylinder and the motion of the ankle joint, a crucial link in the control and actuation of the prosthesis. Since the connection between the hydraulic cylinder and the carbon fiber footplate is rigid, the ankle joint angle and the hydraulic cylinder displacement have a fixed mapping relationship. By controlling the force, displacement, and velocity of the hydraulic cylinder, the output of the ankle prosthesis can be controlled. The mechanical structure of the prosthesis can be equivalently represented as a crank-connecting rod mechanism. The hydraulic cylinder performs linear translational motion, and the connecting rod BC rotates about a fixed axis around point C, thus it can be equivalent to the fixed-axis rotation of AC. Rods AB and BC have fixed lengths. Furthermore, the rotation angle about point C is defined as the ankle joint rotation angle.

[0129] The geometric relationship between the hydraulic cylinder displacement and the ankle joint rotation angle is as follows:

[0130]

[0131] Among them, L BC and L AB d represents the lengths of links BC and AB, respectively. x and d y Let α be the fixed length on the ankle prosthesis, β be the angle between link AB and the x-axis, β be the angle between link BC and the horizontal direction, and y be the displacement of the hydraulic cylinder. These values ​​can be calculated using the following formula:

[0132]

[0133] The relationship between the hydraulic cylinder's movement speed and the ankle joint's speed is as follows:

[0134]

[0135] v BA v represents the magnitude of the relative velocity of point B with respect to point A. A This represents the magnitude of the linear velocity at point A, v. B This represents the linear velocity of rod BC, from which we can know:

[0136] v B =ω BC L BC ;

[0137] ω can be calculated AB and v A :

[0138]

[0139] Where, ω BC ω represents the angular velocity of rod BC. AB This represents the angular velocity of rod AB.

[0140] Hydraulic cylinder acceleration a A With ankle joint angular acceleration α BA The relationship is:

[0141]

[0142] in, These represent the axial acceleration and tangential acceleration of rods BC and AB, respectively.

[0143]

[0144] Where, α BC Let α represent the acceleration of point B relative to point C. AB This represents the acceleration of point A relative to point B.

[0145] Hydraulic cylinder output force F AThe relationship between the ankle joint torque T and the ankle joint torque T is as follows:

[0146]

[0147] have:

[0148]

[0149] in, F represents the axial force on rod BC. BC This represents the force acting on rod BC. Therefore, the hydraulic cylinder outputs force F. A for:

[0150]

[0151] This shows that the angle, angular velocity, angular acceleration, and torque of the ankle joint can all be derived from the displacement, velocity, acceleration, and output force of the hydraulic cylinder. Figure 5 The hydraulic system model in the model constitutes the overall system model, which can realize precise drive control of the entire prosthesis.

[0152] Hydraulic and dynamic models are used in combination: by constructing hydraulic and dynamic models, after identifying the real-time motion state of the prosthesis (i.e., the state output of the optimized finite state machine based on sensor fusion), the required angle, angular velocity, angular acceleration, and torque of the ankle joint can be inferred from the motion state and phase. Then, the displacement, velocity, acceleration, and output force of the hydraulic cylinder can be obtained. Finally, data such as the motor output power and valve opening in the underlying servo controller can be obtained, and the overall real-time and accurate control can be achieved through feedback.

[0153] Output force F in the dynamic formula A The hydraulic pump displacement V in the hydraulic formula pump The pressure difference (P1-P2, P2-P3, etc.) before and after the high-speed switching valve is the final control target obtained from the prosthetic dynamics model and control steps.

[0154] The underlying control layer is a well-known technology. This application maps different underlying controls to their appropriate gait stages to achieve effective control. For details, please refer to the following description of the underlying controller:

[0155] The following section introduces the design of the underlying controller (underlying servo controller), namely the servo impedance, torque, and displacement controller:

[0156] 1. Joint Impedance Controller Design

[0157] Because the ankle prosthesis operates at a high frequency, and the output torque of the ankle joint in passive state is related to both the ankle joint's operating speed and the drive signal of the high-speed switching valve, speed disturbances have a significant impact on the ankle joint's torque output, and both the input signal and the disturbance signal change rapidly. Therefore, the damping control system must ensure the system's speed and robustness. Thus, a PID combined with PWM-PFM control strategy is used to control the prosthesis's output torque. This control strategy improves the system's linearity, minimizes the dead zone of the high-speed switching valve's flow characteristics, and linearizes it, which is beneficial for accurately controlling the system's flow rate, thereby achieving real-time and accurate adjustment of the prosthesis's damping.

[0158] In the passive state, the prosthesis hydraulic drive system can be equivalent to, for example, Figure 6 A simplified system is proposed, powered by the weight of the human body; the downward pressure of the body causes the hydraulic fluid to flow, and the resistance of the hydraulic system is controlled by a high-speed switching valve. Based on this system, the following design is implemented: Figure 7 Impedance control system.

[0159] The impedance characteristic curve represents the relationship between the ankle joint output torque and the ankle joint angle under the target motion mode and gait speed. Specifically, the resistance torque corresponding to the characteristic angle is used as the input signal, which passes through a PID controller, a PWM-PFM driver, and a high-speed switching valve, ultimately acting on the hydraulic cylinder to control the pressure difference between the two chambers. Simultaneously, the pressure in both chambers of the hydraulic cylinder serves as the feedback signal, achieving a closed-loop system. The hydraulic cylinder acts as an actuator to move the prosthesis. During this process, the external environment exerts a load force on the hydraulic cylinder rod through the prosthesis joint hinge, ultimately affecting the ankle joint's output angular acceleration, angular velocity, and angle. The ankle joint's angular velocity also affects the hydraulic fluid flow rate, thus disturbing the pressure in the two chambers of the hydraulic cylinder.

[0160] 2. Design of Joint Output Torque Controller

[0161] In the active plantar flexion state, such as Figure 8 The high-speed switching valve in the prosthetic hydraulic drive system shown controls the oil flow rate by controlling the damping between the high-pressure accumulator and the hydraulic cylinder, thereby controlling the pressure in the hydraulic cylinder's actuating chamber. The power source of this system is the charged high-pressure accumulator and the hydraulic pump. Based on this system, the following design is implemented... Figure 9 The torque control system shown is shown.

[0162] The output characteristic curve is derived from the relationship between the ankle joint output torque and ankle joint angle under the target motion mode and walking speed. Specifically, the output torque corresponding to the characteristic angle is used as the input signal, which passes through a PID controller, a PWM-PFM driver, and a high-speed switching valve, ultimately acting on the hydraulic cylinder to control the pressure difference between the two chambers. Simultaneously, the pressure in both chambers of the hydraulic cylinder serves as the feedback signal to achieve system closed-loop control. The hydraulic cylinder acts as an actuator to move the prosthesis. During this process, the external environment exerts a load force on the hydraulic cylinder rod through the prosthesis joint hinge, ultimately affecting the ankle joint's output angular acceleration, angular velocity, and angle.

[0163] 3. Joint Angle Servo Controller Design

[0164] During the swing phase of active dorsiflexion, the ankle prosthesis primarily undergoes active dorsiflexion, and the hydraulic system can be simplified as follows: Figure 10 The simplified system is shown. In this process, since the load on the prosthesis is relatively small and the main task is to restore the ankle joint angle, a classic PID-based joint angle servo system is used to control the rotation angle of the ankle prosthesis. The system is designed as follows: Figure 11 The ankle angle servo system shown.

[0165] The system feedback is mainly based on the ankle joint rotation angle measured by the angle sensor, while the system input is the target ankle joint angle θ during the active dorsiflexion phase of the swing phase. d The controller's output is the input of the PWM driver to control the speed of the DC motor, thereby controlling the EHA actuator and achieving control of the ankle angle.

[0166] Based on the aforementioned control method for an electrostatic active ankle prosthesis, this application also provides a control system for an electrostatic active ankle prosthesis, comprising:

[0167] A liquid pressure sensor is installed on the hydraulic integration block of the electro-hydraulic active ankle prosthesis and connected to the internal hydraulic circuit of the electro-hydraulic active ankle prosthesis to acquire raw liquid pressure signals.

[0168] An ankle angle sensor is installed at the ankle of an electro-hydraulic active ankle prosthesis to acquire raw ankle angle signals.

[0169] A multi-point plantar pressure array sensor is installed in the insole of an electro-hydraulic active ankle prosthesis to acquire raw plantar pressure signals.

[0170] The signal and control circuit board is connected to the liquid pressure sensor, ankle angle sensor, and multi-point plantar pressure array sensor, respectively. It is used to preprocess the raw liquid pressure signal, raw ankle angle signal, and raw plantar pressure signal to obtain the liquid pressure signal, ankle angle signal, and plantar pressure signal. Plantar pressure features are extracted from the plantar pressure signal. Based on these features, a trained SVM model is used to obtain the division results of four gait stages: passive plantar flexion, passive dorsiflexion, active plantar flexion, and active dorsiflexion. Based on these gait stage division results, combined with the finite state machine state partitioning based on the ankle angle signal and plantar switch, a weighted decision model is used to derive the real-time gait stage. Based on the real-time gait stage and the liquid pressure signal, control signals for prosthetic limb movement are generated. The foot pressure signal includes the pressure signal at each point in the foot pressure array; the foot pressure characteristics include the overall foot area ground reaction force, forefoot area ground reaction force, midfoot area ground reaction force, rearfoot area ground reaction force, x-coordinate of the foot pressure center trajectory, y-coordinate of the foot pressure center trajectory, and maximum single-point impulse; the trained SVM model is obtained by training the SVM model using the overall foot area ground reaction force, forefoot area ground reaction force, midfoot area ground reaction force, rearfoot area ground reaction force, x-coordinate of the foot pressure center trajectory, y-coordinate of the foot pressure center trajectory, and maximum single-point impulse of different subjects in walking modes, and distinguishing the four gait stages of passive plantar flexion, passive dorsiflexion, active plantar flexion, and active dorsiflexion; the foot switch is a pressure switch; the pressure signals of the two regions at the foremost and rearmost ends of the foot pressure array are used as a pressure switch.

[0171] The power supply and drive circuit board is connected to the signal and control circuit board, as well as the high-speed switching valve, motor, and battery of the electro-hydraulic active ankle prosthesis, and is used to control the movement of the electro-hydraulic active ankle prosthesis according to the control signal of the prosthesis movement.

[0172] Among them, the ankle angle sensor is a potentiometer-type angle sensor. The potentiometer-type angle sensor is installed at the ankle of the electro-hydraulic active ankle joint prosthesis and is fixed to the electro-hydraulic active ankle joint prosthesis through a stepped shaft and flange structure. It is used to obtain the real-time angle of the footplate of the electro-hydraulic active ankle joint prosthesis.

[0173] The multi-point plantar pressure array sensor consists of 18 single-point resistive thin-film pressure sensors, which are installed in the insole of the electrostatic active ankle prosthesis and contact the silicone foot skin worn on the foot plate of the electrostatic active ankle prosthesis.

[0174] A microcontroller is installed on the signal and control circuit board. The microcontroller is used to preprocess the original fluid pressure signal, original ankle angle signal, and original plantar pressure signal to obtain the fluid pressure signal, ankle angle signal, and plantar pressure signal. Plantar pressure features are extracted based on the plantar pressure signal. Based on the plantar pressure features, the four gait stages of passive plantar flexion, passive dorsiflexion, active plantar flexion, and active dorsiflexion are divided into four stages based on a trained SVM model. Based on the division of the four gait stages, combined with the state division of the finite state machine based on the ankle angle signal and the plantar switch, the real-time gait stage is obtained through a weighted decision model. Based on the real-time gait stage and fluid pressure signal, the control signal for the prosthesis movement is generated and transmitted to the power supply and drive circuit board.

[0175] The following specific embodiment illustrates the technical solution of the control system for an electro-hydraulic active ankle prosthesis of this application:

[0176] This application relates to a control system for an electro-hydraulic active ankle prosthesis, specifically a gait detection and control system for wearable ankle prostheses. The gait detection and control system is installed on the active ankle prosthesis and comprises components such as a liquid pressure sensor, an ankle angle sensor, a plantar pressure array sensor, a signal and control circuit board, and a power supply and drive circuit board. It detects information about the prosthesis and the wearer, and based on gait segmentation results and real-time gait phase control, adjusts the damping and output of the prosthesis's drive components in real time during walking. Ultimately, the two systems combine to achieve compliant control throughout the entire walking cycle and satisfy the dynamic coupling relationship between the prosthesis and the wearer.

[0177] like Figure 12 The diagram shown is a hardware layout diagram of the ankle prosthesis gait detection system of this application. It is installed on the active ankle prosthesis 601 and consists of components such as a liquid pressure sensor 602, an ankle angle sensor 603, a plantar pressure array sensor 604, a signal and control circuit board 605, and a power supply and drive circuit board 606.

[0178] The multi-point plantar pressure array sensor 604 is installed inside the insole and contacts the silicone foot skin worn on the prosthetic footplate to ensure the force-bearing area. It consists of 18 single-point resistive thin-film pressure sensors, each point representing the plantar pressure in its area, used to determine the human body pressure on that area; the potentiometer-type angle sensor 603 is installed at the ankle of the ankle prosthesis 601 and is fixed to the prosthesis through a stepped shaft and flange structure, used to record the real-time angle of the prosthetic footplate; the liquid pressure sensor 602 is connected to the internal hydraulic circuit of the ankle prosthesis 601, used to detect the real-time liquid pressure in the upper and lower chambers of the prosthetic hydraulic cylinder and the front and rear ends of the high-speed switching valve.

[0179] The signal and control circuit board 605 and the power supply and drive circuit board 606 are fixed to one side of the active ankle prosthesis 601. The signal and control circuit board 605, while connected to the aforementioned sensors, also serves as the computation and processing center for the prosthesis signals. It houses a microcontroller whose main function is to acquire data from various sensors and transmit it to the microcontroller, performing feature extraction and state classification, and transmitting control signals for specific components to the drive circuit board. The power supply and drive circuit board 606 is connected to the prosthesis's high-speed switching valve, motor, and battery, primarily responsible for connecting and driving the high-power components of the prosthesis. Based on the states and component control signals calculated by the signal and control circuit board, it can perform precise control (PWM control) of the corresponding structures of the prosthesis (motor and high-speed switching valve), thereby achieving actuation. Separating the drive circuit and control circuit improves the circuit's resistance to electromagnetic interference and facilitates placement on the prosthesis. In the control circuit, interfaces for various external sensors and the DSP microcontroller are provided, while interfaces are reserved for future sensor expansion, thus completing the overall data acquisition and calculation functions of the prosthesis. In the drive circuit, an overall power supply and high-speed switching valve and motor drive are designed, and pluggable terminal interfaces are selected to facilitate the connection of the battery, motor, and high-speed switching valve terminals.

[0180] This application provides a gait detection and control method and system for wearable ankle prostheses, which is specifically divided into two levels: a gait detection and control method and a supporting gait detection and control system. These two systems work together to determine human gait information, adjust walking damping and output timing, and achieve compliant control of the prosthesis throughout the entire gait phase, satisfying the dynamic coupling relationship between the prosthesis and the wearer. The gait detection and control method for wearable ankle prostheses includes: acquiring and preprocessing raw signals; extracting plantar pressure features and classifying gait using a machine learning classification model; fusing plantar pressure and ankle angle information to achieve top-level controller decision-making; and constructing an active ankle prosthesis hydraulic and dynamic model. The control method consists of a decision fusion layer, a state control layer, and a servo impedance control layer, with its core being a finite state machine for optimizing key event transitions and a bottom-level servo controller. This involves completing gait segmentation using plantar pressure feature extraction and a machine learning classification model. Building upon traditional plantar pressure gait feature center (COP) traces and other features, it incorporates tripedal pressure distribution characteristics based on plantar physiological structures to further capture multivariate nonlinear features in gait data, achieving more accurate gait segmentation. Finally, it integrates plantar pressure and ankle angle information to implement top-level controller decisions. This involves fusing multi-source information and sensor reliability at different stages to trigger key information transfers in a finite state machine, using formulas... A weighted decision-making strategy is used to achieve overall gait stage information fusion. Here, i∈{1,2,3,4} represents the four gait stages. and These are the weights of the plantar pressure array sensor data and the ankle angle data in the i-th gait phase, respectively, S. d This is a weighted decision result, where `round` is a rounding function, ultimately yielding the decision for the state stage. This application relates to a gait detection and control system for wearable ankle prostheses, installed on an active ankle prosthesis. It comprises a liquid pressure sensor that monitors the pressure in the hydraulic system to adjust the output force of the hydraulic cylinder; an ankle angle sensor that provides real-time joint angle information and the prosthesis's position and posture during the gait cycle; a plantar pressure array sensor that detects the pressure distribution and characteristics of different parts of the wearer's foot; and circuit boards for signals and drives. This sensor combination can comprehensively monitor and control various key parameters of the hydraulically driven prosthesis, ensuring the prosthesis's stability, flexibility, and functionality.

[0181] This application specifically discloses a method and system for lower limb motion recognition and human-computer interaction control in active electrohydraulic ankle joint prostheses. Designed for wearable active electrohydraulic ankle joint prostheses, this application includes a gait detection and control method, as well as a supporting gait detection and control system. The system includes components such as a liquid pressure sensor, an ankle angle sensor, a plantar pressure array sensor, a signal and control circuit board, and a power supply and drive circuit board. The combination of these sensors and control circuits enables comprehensive monitoring and precise control of key parameters of the prosthesis, thereby ensuring the stability, flexibility, and functionality of the prosthesis. The gait detection and control method of this application includes the acquisition and preprocessing of raw signals, extraction of plantar pressure features, gait segmentation based on machine learning, and fusion of plantar pressure and ankle... The top-level control decision based on angular information, combined with the pressure distribution of the three-foot region, can better capture the multivariate nonlinear characteristics in gait data, significantly improving the accuracy of gait segmentation. Through the optimization of weighted decision strategy and finite state machine, the real-time adjustment and compliant control of the prosthesis are completed, ensuring the dynamic adaptation of the prosthesis during walking and improving the overall motion experience. The construction of hydraulic and dynamic models provides detailed dynamic information of prosthesis movement, enabling the prosthesis to better achieve bottom-level walking damping and output timing control to adapt to the wearer's gait changes, further improving the overall performance and comfort of the prosthesis.

[0182] Compared to existing classic prosthetic finite state machines that use sensor thresholds for state switching, where plantar pressure only acts as a single-point switch between the sole and toes, failing to reveal the wealth of human walking characteristics, and where all triggering information for typical events is directly provided by sensors integrated into the prosthetic system, this application's prosthetic detection system optimizes key event triggering. It uses plantar pressure array sensors to measure the plantar pressure center point and tripedal zone information to understand and simulate human walking dynamics, achieving the division of the ankle joint into four gait stages during walking. Combined with the traditional finite state machine state division based on ankle angle and plantar switch, it achieves multi-sensor information fusion at the decision layer to complete identification. Furthermore, this application constructs a specific prosthetic hydraulic and dynamic model, enabling more precise control of the prosthetic actuator through the underlying model. Compared to traditional methods, this application can more accurately control the prosthetic power output and damping transition timing, resulting in a better and more natural assistive effect. The advantages of this application are described in detail below:

[0183] (1) This application realizes comprehensive gait detection and control. By combining gait detection and control methods and systems, it can comprehensively monitor various key parameters of the prosthesis, independently and accurately judge human gait information, formulate corresponding control strategies, regulate walking damping and output timing, and realize compliant control of the prosthesis in the entire gait phase.

[0184] (2) This application optimizes the feature capture of nonlinear plantar data. Based on traditional plantar pressure gait features, it adds the pressure distribution features of the three-foot region, which can better capture the multivariate nonlinear features in gait data and improve the accuracy of gait segmentation.

[0185] (3) This application optimizes the finite state machine controller for prosthetics: by adopting a weighted decision-making strategy and combining multi-source information and sensor reliability, it achieves precise control of key events and optimization of gait phases, ensuring that the prosthesis can adapt to the wearer's needs during walking. Existing finite state machine control for prosthetics completes multi-state switching through sensor thresholds and empirical formulas, while the optimized state machine in this application utilizes SVM machine learning, introduces plantar pressure tripedal zone features, and fuses multi-sensor states to obtain more accurate key state recognition for different individuals.

[0186] (4) This application constructs a hydraulic and dynamic model of the prosthesis, providing detailed dynamic information on the movement of the prosthesis, and dynamically adjusting the walking damping and output timing of the prosthesis to adapt to the wearer's gait changes, thereby improving the overall performance and comfort of the prosthesis. Figure 5The figure shows the working state of the prosthesis in the four-gait stage of this application. The key point is that by constructing a hydraulic and dynamic model, after identifying the real-time motion state of the prosthesis, the required angle, angular velocity, angular acceleration, and torque of the ankle joint can be inferred from the motion state and phase. Then, the displacement, velocity, acceleration, and output force of the hydraulic cylinder can be obtained. Finally, the data such as the motor output power and valve opening in the underlying servo controller can be obtained, and the overall real-time accurate control can be achieved through feedback.

[0187] This document uses specific examples to illustrate the principles and implementation methods of this application. The descriptions of the above embodiments are only for the purpose of helping to understand the methods and core ideas of this application. Furthermore, those skilled in the art will recognize that, based on the ideas of this application, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of this application.

Claims

1. A control method for an electro-hydraulic active ankle prosthesis, characterized in that, The control method for an electrostatic active ankle prosthesis includes: Acquire raw fluid pressure signals, raw ankle angle signals, and raw plantar pressure signals; The original liquid pressure signal, the original ankle angle signal, and the original plantar pressure signal are preprocessed to obtain the liquid pressure signal, the ankle angle signal, and the plantar pressure signal; the plantar pressure signal includes the pressure signal at each point in the plantar pressure array. Plantar pressure features are extracted based on the plantar pressure signal; the plantar pressure features include overall foot area ground reaction force, forefoot area ground reaction force, midfoot area ground reaction force, rearfoot area ground reaction force, x-coordinate of plantar pressure center trajectory, y-coordinate of plantar pressure center trajectory, and maximum single-point impulse; the extraction of plantar pressure features based on the plantar pressure signal specifically includes: extracting overall foot area ground reaction force, forefoot area ground reaction force, midfoot area ground reaction force, rearfoot area ground reaction force, x-coordinate of plantar pressure center trajectory, y-coordinate of plantar pressure center trajectory, and maximum single-point impulse based on the plantar pressure signal using the plantar pressure center trajectory and pressure cloud map during walking; Based on the plantar pressure characteristics, the four gait phases of passive plantar flexion, passive dorsiflexion, active plantar flexion, and active dorsiflexion were divided into four stages based on the trained SVM model. The trained SVM model was obtained by training the SVM model with the overall foot area ground reaction force, forefoot area ground reaction force, midfoot area ground reaction force, hindfoot area ground reaction force, x-coordinate of the plantar pressure center trajectory, y-coordinate of the plantar pressure center trajectory, and maximum single-point impulse of different subjects in walking mode, and distinguishing the four gait phases of passive plantar flexion, passive dorsiflexion, active plantar flexion, and active dorsiflexion. Based on the division of gait into four stages—passive plantar flexion, passive dorsiflexion, active plantar flexion, and active dorsiflexion—and combined with the finite state machine state division based on the ankle angle signal and the plantar switch, the real-time gait stages are derived through a weighted decision model. The plantar switch is a pressure switch; the pressure signals from the foremost and rearmost regions of the plantar pressure array serve as a single pressure switch. The weighted decision model is expressed as follows: Where, x i The plantar pressure signal, y i This indicates the ankle angle signal. This represents the weight of the plantar pressure signal in the i-th gait phase. This represents the weight of the ankle angle signal in the i-th gait phase, where i∈{1,2,3,4} represent the four gait phases respectively, i=1 represents passive plantar flexion, i=2 represents passive dorsiflexion, i=3 represents active plantar flexion, and i=4 represents active dorsiflexion. d This represents the weighted decision result, and `round` represents the rounding function, used to convert the weighted average result into the closest integer gait stage. Based on the real-time gait phase and the fluid pressure signal, control signals for prosthetic movement are generated; The electrostatic active ankle joint prosthesis movement is controlled according to the control signal of the prosthesis movement.

2. The control method for an electro-hydraulic active ankle prosthesis according to claim 1, characterized in that, Preprocessing the original fluid pressure signal, the original ankle angle signal, and the original plantar pressure signal to obtain the fluid pressure signal, ankle angle signal, and plantar pressure signal specifically includes: Fourier transforms are performed on the original fluid pressure signal, the original ankle angle signal, and the original plantar pressure signal to obtain the frequency domain distributions of the original fluid pressure signal, the original ankle angle signal, and the original plantar pressure signal. The frequency domain distributions of the original fluid pressure signal, the original ankle angle signal, and the original plantar pressure signal are subjected to low-pass filtering to obtain the fluid pressure signal, ankle angle signal, and plantar pressure signal.

3. The control method for an electro-hydraulic active ankle prosthesis according to claim 1, characterized in that, Based on the real-time gait phase and the fluid pressure signal, control signals for prosthetic movement are generated, specifically including: Based on the real-time gait phase and the fluid pressure signal, combined with the optimized typical event-triggered finite state machine and the electro-hydraulic active ankle joint prosthesis hydraulic model, kinematic and dynamic models, control signals for prosthesis movement are generated; the control signals for prosthesis movement include motor output power and high-speed switching valve opening.

4. The control method for an electro-hydraulic active ankle prosthesis according to claim 3, characterized in that, Controlling the movement of the electro-hydraulic active ankle prosthesis according to the control signal of the prosthesis movement specifically includes: The motor of the electro-hydraulic active ankle joint prosthesis is controlled according to the output power of the motor, and the high-speed switching valve of the electro-hydraulic active ankle joint prosthesis is controlled according to the opening degree of the high-speed switching valve. The electro-hydraulic active ankle joint prosthesis moves under the control of the motor and the high-speed switching valve.

5. A control system for an electro-hydraulic active ankle prosthesis, characterized in that, A control method for an electrostatic active ankle prosthesis based on any one of claims 1-4; The control system for the electrostatic active ankle prosthesis includes: A liquid pressure sensor is installed on the hydraulic integration block of the electro-hydraulic active ankle prosthesis and connected to the internal hydraulic circuit of the electro-hydraulic active ankle prosthesis to acquire the raw liquid pressure signal. An ankle angle sensor is installed at the ankle of an electro-hydraulic active ankle joint prosthesis to acquire raw ankle angle signals; A multi-point plantar pressure array sensor is installed in the insole of an electro-hydraulic active ankle prosthesis to acquire raw plantar pressure signals. A signal and control circuit board, connected to the liquid pressure sensor, the ankle angle sensor, and the multi-point plantar pressure array sensor, is used to preprocess the original liquid pressure signal, the original ankle angle signal, and the original plantar pressure signal to obtain liquid pressure signal, ankle angle signal, and plantar pressure signal. Plantar pressure features are extracted from the plantar pressure signal. Based on these features, a trained SVM model is used to obtain the division results of four gait stages: passive plantar flexion, passive dorsiflexion, active plantar flexion, and active dorsiflexion. Based on these gait stage division results, combined with the finite state machine state partitioning based on the ankle angle signal and plantar switches, a weighted decision model is used to derive the real-time gait stage. Based on the real-time gait stage and the liquid pressure signal, a false gait stage is generated. The control signal for limb movement includes: plantar pressure signals comprising pressure signals at various points in the plantar pressure array; plantar pressure characteristics comprising: overall foot area ground reaction force, forefoot area ground reaction force, midfoot area ground reaction force, hindfoot area ground reaction force, x-coordinate of the plantar pressure center trajectory, y-coordinate of the plantar pressure center trajectory, and maximum single-point impulse; the trained SVM model is obtained by training the SVM model using the overall foot area ground reaction force, forefoot area ground reaction force, midfoot area ground reaction force, hindfoot area ground reaction force, x-coordinate of the plantar pressure center trajectory, y-coordinate of the plantar pressure center trajectory, and maximum single-point impulse of different subjects in walking modes, distinguishing four gait stages: passive plantar flexion, passive dorsiflexion, active plantar flexion, and active dorsiflexion; the plantar switch is a pressure switch; the pressure signals of the two regions at the foremost and rearmost ends of the plantar pressure array serve as a pressure switch; The power supply and drive circuit board is connected to the signal and control circuit board, as well as the high-speed switching valve, motor, and battery of the electro-hydraulic active ankle joint prosthesis, and is used to control the movement of the electro-hydraulic active ankle joint prosthesis according to the control signal of the prosthesis movement.

6. The control system for an electro-hydraulic active ankle prosthesis according to claim 5, characterized in that, The ankle angle sensor is a potentiometer-type angle sensor, which is installed at the ankle of the electro-hydraulic active ankle joint prosthesis and fixed to the electro-hydraulic active ankle joint prosthesis through a stepped shaft and flange structure. It is used to obtain the real-time angle of the footplate of the electro-hydraulic active ankle joint prosthesis.

7. The control system for an electro-hydraulic active ankle prosthesis according to claim 5, characterized in that, The multi-point plantar pressure array sensor consists of 18 single-point resistive thin-film pressure sensors, which are installed in the insole of the electrostatic active ankle joint prosthesis and contact the silicone foot skin worn on the foot plate of the electrostatic active ankle joint prosthesis.

8. The control system for an electro-hydraulic active ankle prosthesis according to claim 5, characterized in that, The signal and control circuit board is equipped with a microcontroller. The microcontroller is used to preprocess the original fluid pressure signal, the original ankle angle signal, and the original plantar pressure signal to obtain the fluid pressure signal, ankle angle signal, and plantar pressure signal. Based on the plantar pressure signal, plantar pressure features are extracted. Based on the plantar pressure features, a trained SVM model is used to obtain the division results of four gait stages: passive plantar flexion, passive dorsiflexion, active plantar flexion, and active dorsiflexion. Based on the division results of the four gait stages, combined with the state division of the finite state machine based on the ankle angle signal and the plantar switch, a weighted decision model is used to obtain the real-time gait stage. Based on the real-time gait stage and the fluid pressure signal, a control signal for prosthetic limb movement is generated and transmitted to the power supply and drive circuit board.

Images