Intelligent control exoskeleton system for knee joint of hemiplegic patient

Abstract

The application discloses a hemiplegic knee joint intelligent control exoskeleton system and relates to the technical field of medical devices.The system synchronously acquires bilateral electromyography, angles, pressure and gait information through an acquisition unit; a prediction unit takes healthy side data as input, and outputs a target gait phase and flexion and extension angles of the affected side through an individualized model; a decision unit determines a helping torque according to a comparison of the bilateral electromyography proportion and threshold value; a control unit performs position-force hybrid control according to the target angles and the helping torque; an adjustment unit adjusts parameters according to a proportional trend; and a safety unit monitors multi-source signals and switches a compliant mode when a threshold value is triggered.The application predicts gait requirements of the affected side knee joint through the characteristics of the healthy side movement, determines the helping torque of the affected side knee joint according to the proportion of the bilateral electromyography and the threshold value of active participation, and further combines position-force hybrid control, parameter adjustment and compliant support switching in an abnormal state, so that the assisting movement of the affected side knee joint can be matched with the gait rhythm of the patient and the active force state.

Description

Technical Field

[0001] This invention relates to the field of medical device technology, specifically to an intelligent exoskeleton system for controlling the knee joint of hemiplegic patients. Background Technology

[0002] Hemiplegic patients often experience unilateral limb motor dysfunction due to central nervous system damage, commonly manifesting as insufficient muscle strength in the affected lower limb, decreased knee flexion and extension control, poor stability during the stance phase, and insufficient knee flexion during the swing phase. As a crucial joint for weight-bearing support, cushioning, stability, and swing flexion and extension during gait, the knee joint's motor control directly impacts gait symmetry, walking safety, and rehabilitation outcomes in hemiplegic patients. Therefore, utilizing knee exoskeletons to assist in the control of the affected knee joint is an important technical direction in gait rehabilitation training for hemiplegic patients.

[0003] Currently, control schemes for knee exoskeletons and related gait training devices for hemiplegic patients mainly revolve around preset trajectory-driven and passive rehabilitation training. The systems typically use pre-set knee flexion-extension angle curves to control the exoskeleton, guiding the patient's affected knee to complete movements such as extension during the support phase and flexion during the swing phase. Angle sensors and plantar pressure sensors detect the patient's gait phase, triggering corresponding assistive movements. Regarding bilateral limb management, most systems still employ a relatively uniform control logic, insufficiently considering the differences between the movement state of the healthy side and the actual force exertion capacity of the affected knee. Feedback information during training is mostly presented as data indicators such as angle, pressure, and steps, making it difficult to achieve dynamic adjustments that match the patient's level of active participation.

[0004] The limitations of existing technologies include at least the following problems: Knee exoskeleton devices for hemiplegic patients generally use a pre-set, fixed knee flexion and extension trajectory drive method. The affected knee joint mostly completes movements under the traction of the device, making it difficult to flexibly adjust according to the patient's real-time movement intentions and active force levels; the recognition of movement intentions on the affected side relies heavily on kinematic or biomechanical signals such as knee joint angle and plantar pressure. These signals usually only change significantly after the movement has occurred, making it difficult to reflect the patient's upcoming knee joint movement needs in a timely manner, resulting in a delayed triggering of assistance; there are significant functional differences between the healthy and affected sides in hemiplegic patients. The healthy side usually has a strong compensatory capacity, while the affected knee joint requires more precise support and guidance. Existing unified control strategies cannot simultaneously meet the needs of low intervention on the healthy side and moderate assistance on the affected side, easily leading to overcompensation on the healthy side and insufficient active stimulation on the affected side; furthermore, information feedback during training is mostly limited to numerical or curve displays, lacking closed-loop regulation based on the degree of active participation, making it difficult to maintain the patient's level of active participation during long-term training. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides an intelligent exoskeleton system for the knee joint of hemiplegic patients. This system solves the problems of existing exoskeleton devices for the knee joint of hemiplegic patients, such as strong dependence on fixed trajectory, delayed assistance timing, poor bilateral adaptability, and difficulty in dynamically adjusting assistance parameters according to training status, in terms of knee joint drive mode, movement intention recognition, differentiated control of the healthy and affected sides, and training feedback adjustment.

[0006] To achieve the above objectives, this invention provides the following technical solution: an intelligent exoskeleton system for controlling the knee joint of a hemiplegic patient, comprising: a data acquisition unit for simultaneously acquiring electromyographic signals, joint angles, plantar pressure distribution, and gait phase information of the patient's unaffected and affected lower limbs; a prediction unit for using the electromyographic signals, joint angles, plantar pressure distribution, and gait phase information of the unaffected lower limb as input, and outputting the target gait phase and target knee flexion-extension angle of the affected knee joint in the current gait cycle through an individually trained prediction model; and a decision unit for calculating the activation levels of the corresponding muscles on the unaffected and affected sides, comparing the ratio of the activation level of the affected side to the activation level of the unaffected side with an active participation judgment threshold. When the ratio is lower than the active participation judgment threshold, the assist torque of the affected knee joint is determined based on the ratio difference; the control unit is used to perform position-force hybrid control on the affected knee joint drive module with the target knee joint flexion-extension angle as the position control target and the assist torque of the affected knee joint as the force control target, and outputs the torque to assist the flexion-extension of the affected knee joint; the adjustment unit is used to adjust the calculation parameters of the assist torque of the affected knee joint and the active participation judgment threshold according to the trend of the ratio change of the activation level of the corresponding muscle on the affected side to the activation level of the corresponding muscle on the healthy side in a continuous gait cycle; the safety unit is used to monitor the amplitude of the electromyographic signal of the affected lower limb, the angle of the affected knee joint, the plantar pressure distribution of the affected side, and the output torque. When the monitoring results reach the corresponding preset threshold, the output torque is reduced and the system switches to compliant support mode.

[0007] Furthermore, the training steps of the individualized prediction model are as follows: Collect electromyographic signals, joint angles, plantar pressure distribution, and gait phase information of the unaffected lower limb during multiple consecutive steps in a calibrated walking state; simultaneously collect the actual gait phase and actual knee flexion-extension angle of the affected knee joint in a calibrated assisted state; use the electromyographic signals, joint angles, plantar pressure distribution, and gait phase information of the unaffected lower limb as input samples, and the actual gait phase and actual knee flexion-extension angle of the affected knee joint as output labels; train to obtain the mapping relationship between the unaffected gait characteristics and the target state of the affected knee joint, and form an individualized prediction model.

[0008] Further, the steps for constructing the input sample are as follows: the electromyographic signal of the unaffected lower limb is filtered, rectified, and feature extracted to obtain the temporal features of muscle activation on the unaffected side; the flexion and extension angular velocity of the unaffected knee joint is calculated based on the joint angle on the unaffected side; the heel strike time and toe lift-off time are extracted based on the plantar pressure distribution on the unaffected side; and the temporal features of muscle activation on the unaffected side, the flexion and extension angular velocity of the unaffected knee joint, the heel strike time, the toe lift-off time, and the gait phase information on the unaffected side are combined to form the input sample.

[0009] Further, the steps for determining the assist torque of the affected knee joint are as follows: The electromyographic signal of the healthy lower limb is filtered, rectified, and integrated to obtain the corresponding muscle activation level on the healthy side; the electromyographic signal of the affected lower limb is filtered, rectified, and integrated to obtain the corresponding muscle activation level on the affected side; the ratio of the corresponding muscle activation level on the affected side to the corresponding muscle activation level on the healthy side is calculated; the ratio is compared with the active participation judgment threshold; when the ratio is lower than the active participation judgment threshold, the assist torque of the affected knee joint is determined based on the difference between the ratio and the active participation judgment threshold, the corresponding muscle activation level on the healthy side, and the assist gain coefficient, and the assist torque of the affected knee joint does not exceed the preset assist upper limit.

[0010] Furthermore, the adjustment steps for the active participation judgment threshold and the assist gain coefficient are as follows: Read the ratio data of the activation level of the corresponding muscle on the affected side to the activation level of the corresponding muscle on the healthy side within multiple consecutive gait cycles; calculate the ratio change trend based on the ratio data; when the ratio change trend continues to rise, increase the active participation judgment threshold and decrease the assist gain coefficient; when the ratio change trend continues to fall, maintain the active participation judgment threshold and increase the assist gain coefficient; when the ratio change trend does not change significantly, maintain the current values ​​of the active participation judgment threshold and the assist gain coefficient.

[0011] Furthermore, the steps of the position-force hybrid control are as follows: read the flexion-extension angle of the target knee joint and calculate the position control torque component based on the flexion-extension angle; read the assist torque of the affected knee joint and calculate the force control torque component based on the assist torque of the affected knee joint; perform weighted fusion of the position control torque component and the force control torque component to obtain the output torque; input the output torque into the affected knee joint drive module to drive the affected knee joint to complete the flexion-extension assisted movement.

[0012] Furthermore, the real-time correction steps for the target knee flexion and extension angle are as follows: real-time acquisition of the affected knee angle and the plantar pressure distribution on the affected side; comparison of the affected knee angle and the plantar pressure distribution on the affected side with the reference knee angle and the reference plantar pressure distribution corresponding to the target gait stage; when there is a deviation in the comparison results, the angle deviation and pressure deviation are calculated; the target knee flexion and extension angle is corrected based on the angle deviation and pressure deviation.

[0013] Furthermore, the switching steps for the compliant support mode are as follows: Real-time monitoring of the amplitude of electromyography (EMG) signals of the affected lower limb, the angle of the affected knee joint, the plantar pressure distribution of the affected foot, and the output torque; comparing the amplitude of the EMG signals of the affected lower limb with the spasticity threshold, comparing the angle of the affected knee joint with the safe angle range, comparing the plantar pressure distribution of the affected foot with the pressure threshold range, and comparing the output torque with the maximum torque limit; when any monitoring result reaches the corresponding threshold, reducing the output torque to the support torque; and switching the affected knee joint drive module to the passive compliant support state.

[0014] Furthermore, the output torque feedback steps are as follows: the servo motor receives the control command corresponding to the output torque; the harmonic reducer transmits the torque output by the servo motor to the affected knee joint; the torque sensor detects the actual output torque of the affected knee joint drive module in real time; and the actual output torque is fed back to the control unit to correct the output torque of the next control cycle.

[0015] Furthermore, the initial setting steps for the active participation judgment threshold are as follows: In the initialization phase, electromyographic signals of the unaffected lower limb of the patient under calibrated walking conditions are collected, and the baseline of the corresponding muscle activation level on the unaffected side is calculated; in the initialization phase, electromyographic signals of the affected lower limb of the patient under resting conditions are collected, and the baseline of the corresponding muscle activation level on the affected side is calculated; based on the baseline of the corresponding muscle activation level on the unaffected side and the baseline of the corresponding muscle activation level on the affected side, the initial value of the active participation judgment threshold for the corresponding muscle activation level on the affected side relative to the corresponding muscle activation level on the unaffected side is set.

[0016] The present invention has the following beneficial effects: (1) The intelligent exoskeleton system for knee joint control of hemiplegic patients collects electromyographic signals, joint angles, plantar pressure distribution and gait phase information of the healthy and affected lower limbs, and uses the data of the healthy side as the input of the prediction model to output the target gait stage and target knee flexion and extension angle of the affected knee joint. This makes the auxiliary control of the affected knee joint no longer dependent on the preset fixed trajectory. The movement rhythm of the healthy lower limb can be transformed into the stage target and angle target of the affected knee joint in the current gait cycle. The exoskeleton auxiliary movements match the patient's own walking rhythm, reducing the passive drag of the affected side caused by the universal trajectory drive, as well as the training discomfort caused by the inconsistency between the device traction and the patient's movement intention.

[0017] (2) The intelligent exoskeleton system for knee joint control in hemiplegic patients calculates the corresponding muscle activation levels on the healthy and affected sides and compares the ratio of muscle activation on the affected side to the healthy side with the active participation judgment threshold. When the force exerted on the affected side is insufficient, the knee joint assist torque is determined based on the ratio difference, so that the exoskeleton outputs compensatory assistance around the actual force exertion gap on the affected side. The knee joint assist torque on the affected side is constrained by the preset assist upper limit, reducing the equipment dependence caused by excessive assistance. The ratio change trend in the continuous gait cycle is used to adjust the assist torque calculation parameters and the active participation judgment threshold, so that the assist output is updated with the change of active force exertion on the affected side, improving the problem that fixed parameter assistance is difficult to adapt to changes in training state.

[0018] (3) The intelligent exoskeleton system for knee joint control of hemiplegic patients uses the target knee joint flexion and extension angle as the position control target and the assist torque of the affected knee joint as the force control target to perform position and force hybrid control on the drive module of the affected knee joint. The exoskeleton outputs an auxiliary torque that matches the force gap of the affected side while following the target knee joint movement. By monitoring the amplitude of electromyography signal, the angle of the affected knee joint, the pressure distribution of the affected foot, and the output torque, the system reduces the output torque and switches to a compliant support mode when the monitoring results reach the corresponding preset threshold, thereby reducing the adverse effects of muscle spasm, abnormal joint angle, sudden change in foot force, and excessive torque on the training process of the affected knee joint.

[0019] Of course, any product implementing this invention does not necessarily need to achieve all of the advantages described above at the same time. Attached Figure Description

[0020] Figure 1 This is a block diagram of the intelligent exoskeleton system for controlling the knee joint of hemiplegic patients according to the present invention.

[0021] Figure 2 This is a flowchart illustrating the training steps of the prediction model in the intelligent knee joint control exoskeleton system for hemiplegic patients of the present invention, which has undergone individualized training.

[0022] Figure 3 This is a flowchart of the initial setting steps for the active participation in the judgment threshold in the intelligent control exoskeleton system for the knee joint of hemiplegic patients of the present invention. Detailed Implementation

[0023] Please see Figure 1This invention provides a technical solution: an intelligent exoskeleton system for controlling the knee joint of a hemiplegic patient, comprising: a data acquisition unit for simultaneously acquiring electromyographic signals, joint angles, plantar pressure distribution, and gait phase information of the patient's unaffected and affected lower limbs; a prediction unit for taking the electromyographic signals, joint angles, plantar pressure distribution, and gait phase information of the unaffected lower limb as input, and outputting the target gait phase and target knee flexion-extension angle of the affected knee joint in the current gait cycle through an individualized prediction model; and a decision unit for calculating the activation levels of the corresponding muscles on the unaffected and affected sides, comparing the ratio of the activation level of the affected side to the activation level of the unaffected side with an active participation judgment threshold, and determining the threshold if the ratio is lower than a certain threshold. When actively participating in the judgment threshold, the assist torque of the affected knee joint is determined based on the proportional difference; the control unit is used to perform position-force hybrid control on the affected knee joint drive module with the target knee joint flexion-extension angle as the position control target and the assist torque of the affected knee joint as the force control target, and outputs the torque to assist the flexion-extension of the affected knee joint; the adjustment unit is used to adjust the calculation parameters of the assist torque of the affected knee joint and the active participation judgment threshold according to the proportional change trend of the activation level of the corresponding muscle on the affected side to the activation level of the corresponding muscle on the healthy side in a continuous gait cycle; the safety unit is used to monitor the amplitude of the electromyographic signal of the affected lower limb, the angle of the affected knee joint, the plantar pressure distribution of the affected side, and the output torque. When the monitoring results reach the corresponding preset threshold, the output torque is reduced and the system switches to a compliant support mode.

[0024] The acquisition unit includes a surface electromyography (EMG) sensor, an inertial measurement unit, and a plantar pressure sensor. The EMG sensor is attached to the surface of the quadriceps femoris, hamstring, tibialis anterior, and gastrocnemius muscles of the healthy and affected lower limbs, respectively, to acquire the EMG signals of the corresponding muscles. The inertial measurement units are fixed to the thigh and calf of the healthy lower limb and the thigh and calf of the affected lower limb near the knee joint, respectively, to collect joint angles and motion postures; Foot pressure sensors are embedded in the forefoot, midfoot, and heel areas of the insole on both the healthy and affected sides to collect pressure distribution data in each area of ​​the foot. The sampling frequency of the surface electromyography sensor is set to no less than 1000Hz, and the sampling frequency of the inertial measurement unit and the plantar pressure sensor is set to 100Hz-200Hz. All kinds of signals are transmitted to the prediction unit, decision unit and safety unit after being timestamped. The affected knee joint drive module includes a servo motor, a harmonic reducer, and a torque sensor. The servo motor provides power output, the harmonic reducer is used to reduce the speed and increase the torque, and the torque sensor is used to detect the output torque in real time and feed it back to the control unit. The prediction model for individualized training uses an LSTM neural network model. The input data includes electromyographic features of the healthy lower limb, joint motion features of the healthy side, plantar pressure features of the healthy side, and gait phase features of the healthy side. The output data includes the target gait phase and target knee flexion-extension angle of the affected knee joint. The initial value of the active participation judgment threshold is set based on the baseline of the activation level of the muscles on the healthy and affected sides of the patient. The initial value of the assist gain coefficient is set at 0.5-0.8, which can be dynamically adjusted according to the patient's recovery status. The preset upper limit of assistance is determined based on the patient's weight and the muscle strength level on the affected side, and is converted by combining the lower limb equivalent force arm. Gait phase information includes the support phase and the swing phase. The support phase is further divided into three sub-phases: heel strike, mid-support phase, and heel lift-off phase. The swing phase is further divided into two sub-phases: acceleration swing and deceleration swing.

[0025] Specifically, such as Figure 2 As shown, the training steps for the individualized prediction model are as follows: The electromyographic signals, joint angles, plantar pressure distribution, and gait phase information of the unaffected lower limb were collected during multiple consecutive steps in a calibrated walking state. Specifically: The calibrated walking state is defined as the patient walking at a comfortable and uniform pace on a flat surface, with a minimum of 50 steps. During the data collection process, the patient does not wear an exoskeleton or wears an exoskeleton but does not activate the assist function. The collected electromyographic signals of the healthy lower limb covered four key muscles: quadriceps femoris, hamstrings, tibialis anterior, and gastrocnemius. The collection time was consistent with the walking time. The joint angles of the healthy hip, knee, and ankle joints, as well as the pressure distribution data of the forefoot, midfoot, and heel of the healthy foot, were collected simultaneously. At the same time, the gait phase information of the healthy side was determined by the pressure changes of the plantar pressure sensor. In one implementation, when the pressure on the healthy heel reaches a preset pressure threshold (e.g., 50N), it is determined that the heel has struck the ground and the foot enters the support phase. When the pressure on the healthy heel is lower than the preset pressure threshold and the pressure on the forefoot is still higher than the preset pressure threshold, it is determined that the heel leaves the ground and enters the later stage of support. When the pressure on the forefoot of the healthy side is lower than the preset pressure threshold and continues for a preset time, it is determined that the toes leave the ground and the swing phase begins. Simultaneously collect the actual gait phase and actual knee flexion-extension angle of the affected knee joint under calibrated assisted state, specifically as follows: The calibration assist state is when the patient wears this system and the system is in calibration acquisition mode. It does not output active assistance, but only provides necessary mechanical support and collects the status data of the affected knee joint. It simultaneously collects the actual flexion and extension angle of the affected knee joint (collected through the affected side inertial measurement unit) and the actual gait phase. The actual gait phase is determined by the combined distribution of plantar pressure on the affected side and the angle of the affected knee joint. The determination criteria are consistent with the determination criteria of the healthy side gait phase. The data collection process was carried out simultaneously with the healthy side calibration walking data collection process to ensure that the signal from the healthy side at each step corresponded one-to-one with the actual state of the affected side. In one implementation, for each step the patient completes on the healthy side, the flexion and extension angle change curve of the affected knee joint and the corresponding gait stage are recorded simultaneously to form a set of matching data. The input samples are electromyography (EMG) signals of the unaffected lower limb, joint angles of the unaffected side, plantar pressure distribution of the unaffected side, and gait phase information of the unaffected side. The output labels are the actual gait phase and actual knee flexion-extension angle of the affected knee joint. Specifically: The collected electromyographic signals of the unaffected lower limb were filtered, rectified, and feature extracted to obtain the temporal features of muscle activation. The joint angles on the healthy side were smoothed to obtain time-series data of joint angles; the plantar pressure distribution on the healthy side was normalized to obtain pressure percentage data for each region. The healthy side gait phase information is converted into numerical labels (e.g., the support phase is recorded as 0 and the swing phase as 1); the healthy side data processed above are combined into input samples, and each input sample corresponds to the healthy side state at one gait moment; The actual gait phase (numerical label) and actual knee flexion-extension angle of the affected knee joint are used as output labels to form a sample set that matches input and output, with a sample set size of no less than 1000 sets; The training process yields a mapping relationship between the healthy side's gait characteristics and the target state of the affected knee joint, forming an individualized prediction model, specifically: The sample set is divided into training set, validation set and test set in a ratio of 7:2:1. The training set is used to train the LSTM neural network model, the validation set is used to adjust the model hyperparameters (including the number of hidden layers, the number of neurons and the learning rate), and the test set is used to verify the model accuracy. During training, mean squared error is used as the loss function, and gradient descent is used to optimize the model parameters. Training is stopped when the loss function value on the validation set no longer decreases for 100 consecutive epochs. After training is completed, when the prediction error on the test set meets the preset error threshold, the individualized prediction model is obtained. In one implementation, the predicted deviation of the flexion and extension angle of the affected knee joint is not greater than a preset angle threshold, and the prediction result of the target gait phase meets the preset accuracy requirement. The predictive model can be retrained or updated based on the patient's recovery progress to maintain the model's match with the patient's current gait status.

[0026] The steps for constructing the input sample are as follows: The electromyographic signals of the unaffected lower limb were filtered, rectified, and feature extracted to obtain the temporal features of muscle activation on the unaffected side, specifically as follows: First, the electromyography signal is subjected to bandpass filtering and power frequency notch filtering. The bandpass filtering range is 5Hz-500Hz, and the power frequency notch filtering frequency is 50Hz to remove high-frequency noise, low-frequency interference, and power frequency interference (such as environmental electromagnetic interference and power frequency noise). The filtered electromyographic signal is subjected to full-wave rectification to convert the negative signal into a positive signal; Finally, the rectified signal is processed by sliding window integration. The window size is set to 100ms and the step size is set to 50ms to obtain the electromyographic signal integral value in each window, i.e. the temporal characteristics of the activation level of the healthy side muscle. In one embodiment, the electromyographic signal of the healthy quadriceps femoris muscle was processed to obtain muscle activation timing characteristics of 0.8-3.5 mV·s. The activation levels of different time phases were significantly different, with the activation level during the support phase being higher than that during the swing phase. The flexion-extension angular velocity of the healthy knee joint is calculated based on the angle of the healthy joint, specifically as follows: The real-time flexion and extension angles of the healthy knee joint are collected using an inertial measurement unit at a frequency of 100Hz, meaning that one angle value is collected every 10ms. The knee flexion-extension angular velocity is calculated using the difference method. The formula is as follows: ; in, The angular velocity of knee flexion and extension (° / s). The current knee flexion and extension angle (°) is given. The angle of knee flexion and extension at the previous moment (°). The sampling time interval is 0.01s. The calculated angular velocity values ​​are smoothed to remove instantaneous fluctuations; In one embodiment, the flexion-extension angular velocity of the healthy knee joint during the support phase is -50° / s to 20° / s (the negative sign indicates extension, and the positive sign indicates flexion), and the flexion-extension angular velocity during the swing phase is 20° / s to 80° / s. The heel strike time and toe lift-off time were extracted based on the plantar pressure distribution on the healthy side. Specifically: The plantar pressure sensor collects pressure data of the forefoot, midfoot, and heel of the healthy side in real time, and sets the heel pressure threshold to 50N and the forefoot pressure threshold to 30N. When the heel pressure suddenly increases from below 50N to above 50N and lasts for more than 100ms, it is determined as the moment of heel strike. When the heel pressure drops from above 50N to below 10N and the forefoot pressure drops from above 30N to below 30N for more than 50ms, it is considered the moment when the toes leave the ground. In one implementation, the time when the patient's healthy heel strikes the ground corresponds to 0% of the gait cycle, and the time when the toes leave the ground corresponds to 60%-70% of the gait cycle. There may be slight differences among different patients depending on their own walking habits. The input sample consists of the temporal features of muscle activation on the healthy side, the flexion and extension angular velocity of the knee joint on the healthy side, the heel strike time, the toe lift-off time, and the gait phase information on the healthy side. Specifically: The activation time-series features of the four key muscles on the healthy side (one set of time-series data for each muscle), the time-series data of the knee flexion and extension angular velocity on the healthy side, the heel strike time (numerical, in ms), the toe lift-off time (numerical, in ms), and the gait phase information on the healthy side (numerical label 0 or 1) are fused together, and each input sample corresponds to a comprehensive feature of a gait moment. During the fusion process, all data are normalized to the 0-1 range.

[0027] This implementation scheme integrates the temporal characteristics of muscle activation on the healthy side, the flexion and extension angular velocity of the healthy knee joint, the heel strike time, the toe lift-off time, and the gait phase information on the healthy side into input samples. This enables the prediction model to simultaneously acquire muscle exertion, joint movement, and changes in plantar contact on the healthy lower limb. This avoids gait judgment lag caused by relying on only a single angle signal, and makes the predicted target gait stage and target flexion and extension angle of the affected knee joint closer to the patient's current gait state.

[0028] Specifically, the steps for determining the assist torque of the affected knee joint are as follows: The electromyographic signals of the unaffected lower limb were filtered, rectified, and integrated to obtain the activation level of the corresponding muscles on the unaffected side, specifically as follows: The processing flow is consistent with the electromyography signal processing flow in the construction of the healthy side input sample. First, the electromyography signal of the healthy lower limb is processed by bandpass filtering (5Hz-500Hz) and power frequency notch filtering (50Hz) to remove interference signals. Then, full-wave rectification is performed to convert the negative electromyography signal into a positive signal. Using a 100ms sliding window integration, the muscle activation levels of the quadriceps, hamstrings, tibialis anterior and gastrocnemius muscles on the healthy side are obtained. The average value of the activation levels of the four muscles is taken as the final value of the corresponding muscle activation level on the healthy side. In one implementation, the average activation level of the healthy side muscles was 1.2-2.8 mV·s, and the average value of the support phase during walking was higher than that of the swing phase. The electromyographic signals of the affected lower limb were filtered, rectified, and integrated to obtain the activation level of the corresponding muscles on the affected side, specifically as follows: The activation levels of the quadriceps, hamstrings, tibialis anterior and gastrocnemius muscles on the affected side were obtained by using bandpass filtering (5Hz-500Hz), power frequency notch filtering (50Hz), full wave rectification and 100ms sliding window integration. The average value of the activation levels of the four muscles was taken as the final value of the activation level of the corresponding muscle on the affected side. Due to insufficient muscle strength on the affected side, the average level of muscle activation is usually lower than that on the healthy side; In one implementation, the average level of muscle activation on the affected side is 0.5-1.8 mV·s; The ratio of the activation level of the corresponding muscle on the affected side to the activation level of the corresponding muscle on the healthy side is calculated as follows: The proportion is calculated using division. The formula is: ; in, This represents the ratio of muscle activation levels on the affected side to the healthy side. This represents the average activation level of the corresponding muscle on the affected side. This represents the average activation level of the corresponding muscle on the healthy side. Proportion It is usually between 0 and 1; when A value close to 1 indicates that the activation level of the muscles on the affected side is close to that on the healthy side; when A value close to 0 indicates a lower level of muscle activation on the affected side; when When it is greater than 1, it can be The amplitude is limited to 1 or used as an abnormal electromyographic state for judgment; In one implementation, during the initial stage of patient recovery The value is 0.3-0.5, indicating the later stage of recovery. The value can be increased to 0.7-0.9; The proportion is compared with the threshold for judging active participation, specifically: The threshold for actively participating in the judgment is denoted as , The initial value is 0.6 (which can be dynamically adjusted via the adjustment unit). The comparison process involves calculating the ratio. and Compare sizes; when When the affected side actively participates sufficiently, no assistance is needed or only minimal assistance is required. when If the affected side is not actively involved enough, it is necessary to provide corresponding assistance based on the ratio difference. In one implementation, if , If so, it is determined that the affected side is not actively involved enough, and the assist torque needs to be calculated. When the proportion is lower than the active participation judgment threshold, the assist torque of the affected knee joint is determined based on the difference between the proportion and the active participation judgment threshold, the normalized activation level of the corresponding muscle on the healthy side, and the assist gain coefficient. The assist torque of the affected knee joint does not exceed the preset assist upper limit, specifically as follows: First, calculate the ratio difference: ; in, The proportional difference ranges from 0 to 0.6; the assist gain coefficient is denoted as... The initial value is 0.6 (which can be dynamically adjusted via the adjustment unit); Assist torque of the affected knee joint The calculation formula is: ; in, This represents the normalized activation level of the corresponding muscle on the healthy side. The electromyography-assisted conversion coefficient is expressed in N·m. Preset boost limit Based on the patient's weight, leg length, and affected side muscle strength level, the calculation formula is as follows: ; in, The patient's weight (in kg). The acceleration due to gravity (value 9.8 m / s²) 2 ), The equivalent force arm of the lower leg (unit: m). This is the upper limit coefficient, and its value is set according to the patient's weight, the muscle strength level on the affected side, and the training stage. Calculated assist torque If it exceeds Then take As the final assist torque.

[0029] The steps for actively participating in the judgment threshold and adjusting the boost gain coefficient are as follows: The ratio of the activation level of the corresponding muscles on the affected side to the activation level of the corresponding muscles on the healthy side was read over multiple consecutive gait cycles. Specifically: The proportion of continuous collection of 10-20 complete gait cycles Value, one for each gait cycle. Value (taken within the gait cycle) (average) During the reading process, outliers are removed (such as those caused by sudden actions of the patient). (Value abruptly changes, exceeding the normal range of 0-1). In one implementation, 15 gait cycles are continuously collected to obtain 15 [data / data / etc.]. The values ​​are 0.42, 0.45, 0.43, 0.47, 0.46, 0.49, 0.48, 0.51, 0.50, 0.53, 0.52, 0.55, 0.54, 0.57, and 0.56, respectively. The trend of proportional change is calculated based on the proportional data, specifically as follows: Linear fitting was used to analyze multiple consecutive gait cycles. The values ​​are fitted to obtain the fitted line: ; in, The gait cycle number is (1, 2, ..., n, where n is the number of consecutively collected gait cycles). The slope of the fitted line, The intercept; By slope Determine the trend of the proportion change: when At that time, the trend of the proportion change was continuously rising; when At that time, the trend of the judgment ratio was a continuous decline; when At that time, the trend of the proportion change was not significantly different; In one implementation, the above 15 Values ​​obtained after fitting If so, the trend of the proportion change is determined to be a continuous increase; When the trend of the ratio change continues to rise, the threshold for active participation judgment is increased, and the assist gain coefficient is decreased, specifically as follows: The continuously rising trend in the proportion indicates that the patient's ability to actively exert force on the affected side is improving, necessitating a higher threshold for judging active participation. Reduce the boost gain coefficient This reduces system assistance, allowing the affected side to retain more space for active exertion; The adjustment range is: Each increase is 0.05-0.1. Each reduction is 0.05-0.1; After adjustment Not exceeding 0.9, Not less than 0.3; In one implementation, the initial , The proportion trend continues to rise, after adjustment , ; When the trend of the ratio change continues to decline, maintain the active participation judgment threshold and increase the assist gain coefficient, specifically as follows: A continuous downward trend in the proportion indicates a decline in the patient's ability to actively exert force on the affected side or fatigue, necessitating the maintenance of the active participation threshold. Unchanged, increase boost gain coefficient This increases system support and prevents patients from becoming overly fatigued; The adjustment range is: Each increase is 0.05-0.1, after adjustment Not exceeding 0.9; In one implementation, the initial , The proportion trend continued to decline, after adjustment , ; When the trend of proportional change does not change significantly, maintain the current values ​​of the active participation judgment threshold and the assist gain coefficient, specifically as follows: The proportional change trend shows no significant change, indicating that the patient's active force exertion ability on the affected side is stable and no adjustment is needed. and To maintain the current value and keep the system's assistance stable, avoiding frequent adjustments that could interfere with the patient's recovery; In one implementation, the initial , If the proportional trend does not change significantly, then it will continue to be maintained. , .

[0030] In this implementation plan, the active force exertion status of the affected side is determined by the ratio of the activation level of the corresponding muscle on the affected side to the activation level of the corresponding muscle on the healthy side. The active participation judgment threshold and the assist gain coefficient are adjusted according to the trend of the ratio change over multiple gait cycles, so that the assist parameters can be updated with the change of active force exertion on the affected side, reducing the assist deviation caused by fixed parameters.

[0031] Specifically, the steps of position-force hybrid control are as follows: The target knee joint flexion and extension angles are read, and the position control torque component is calculated based on these angles. Specifically: The control unit reads the target flexion-extension angle of the affected knee joint from the prediction unit in real time. At the same time, the actual flexion and extension angle of the affected knee joint is read through the inertial measurement unit. ; The position control torque component is calculated using a proportional-integral-derivative (PID) control algorithm. The PID control algorithm formula is: ; in, This is a proportionality coefficient, with a value ranging from 5 to 10; This is the integral coefficient, with a value ranging from 0.1 to 0.5; These are the differential coefficients, with values ​​ranging from 0.5 to 2.0. The function of the position control torque component is to make the actual angle of the affected knee joint approach the target angle; In one implementation, the control unit takes the angle error between the target angle and the actual angle as input and calculates the position control torque component in each control cycle according to the set PID parameters. The assist torque of the affected knee joint is read, and the force control torque component is calculated based on the assist torque of the affected knee joint. Specifically: The control unit reads the assist torque of the affected knee joint from the decision unit in real time. and will As a reference value for the force control torque component; In one implementation, the force controls the torque component. =q× Where q is the force control correction coefficient; The function of force-controlled torque component is to provide flexible assistance to the affected knee joint, matching the patient's active force level; The position control torque component and the force control torque component are weighted and fused to obtain the output torque, which is as follows: Set position control weight coefficient Sum of force control weighting coefficients ,satisfy ; Adjust weighting coefficients based on gait phase: During the support period, the position control weighting coefficient Force control weighting coefficient Prioritize ensuring knee joint stability; During the swing period, the position control weighting coefficient Force control weighting coefficient Prioritize ensuring flexibility in support; Output torque The calculation formula is: ; In one implementation, the support period , , , ,but ; The output torque is input into the affected knee joint drive module to drive the affected knee joint to complete flexion and extension assisted movements, specifically as follows: Output torque calculated by the control unit The target torque at the affected knee joint end is calculated based on the target torque at the knee joint end, the reduction ratio of the harmonic reducer, and the transmission efficiency. The calculation formula is as follows: ; in, The reduction ratio, Transmission efficiency (not less than 90%); The control unit converts the target torque at the servo motor end into a control command (voltage signal) for the servo motor. After receiving the control command, the servo motor outputs the corresponding speed and torque. The torque is amplified by the harmonic reducer and transmitted to the mechanical joint of the affected knee joint, driving the knee joint to complete the flexion and extension movements; During the driving process, the torque sensor detects the actual output torque in real time and feeds it back to the control unit for torque correction in the next control cycle; In one implementation, the target output torque at the knee joint end... Reduction ratio Transmission efficiency The target torque at the servo motor end After receiving the control command, the servo motor outputs the corresponding speed and torque, and the torque transmitted to the knee joint through the reducer is about 11.29 N·m, which matches the patient's walking rhythm.

[0032] In this implementation scheme, by weighted and fused the position control torque component corresponding to the target knee joint flexion and extension angle and the force control torque component corresponding to the assist torque of the affected knee joint, the affected knee joint drive module can guide the knee joint flexion and extension according to the target angle, and can also provide compensation torque according to the degree of insufficient active force exertion on the affected side. This makes the output torque constrained by both the target angle and the force exertion gap on the affected side, reducing the control deviation caused by single position control or single force control.

[0033] Specifically, the real-time correction steps for the target knee flexion-extension angle are as follows: Real-time acquisition of the knee joint angle and plantar pressure distribution on the affected side, specifically: The actual flexion and extension angles of the affected knee joint are collected in real time using an inertial measurement unit located near the knee joint on the lower leg of the affected side, with a collection frequency of 100Hz. The pressure distribution data of the forefoot, midfoot, and heel are collected in real time using a plantar pressure sensor on the affected side, with the collection frequency consistent with the angle collection frequency. The collected data underwent simple filtering to remove transient noise; The affected knee angle and plantar pressure distribution were compared with the reference knee angle and reference plantar pressure distribution corresponding to the target gait phase. Specifically: First, determine the target gait phase of the affected knee joint (output by the prediction unit), and then call the reference knee joint angle range and reference plantar pressure distribution range corresponding to the target gait phase; The reference data were determined based on data collected during patients' calibrated walking states; The reference knee angle range corresponding to the target gait phase is determined by the patient's calibrated walking data, and the reference angle range during the support phase is smaller than that during the swing phase. The comparison process is as follows: Determine whether the actual angle of the affected knee joint is within the reference angle range, and whether the pressure distribution in different areas of the affected foot is within the reference pressure distribution range; If both are within the range, then the comparison is considered to be without bias; If any one of them is outside the range, then the comparison is considered to have a bias. In one implementation, the target gait phase is the support phase (heel strike), with a reference angle range of 0-20° and an actual angle of 17°. The reference heel pressure is 40-60%, and the actual heel pressure is 35%. If the comparison is found to be biased, then the comparison is considered to have a deviation. When there are deviations in the comparison results, the angular deviation and pressure deviation are calculated, specifically as follows: Angle deviation The calculation formula is: ; in, The median value of the reference knee angle for the target gait phase; Pressure deviation The calculation formula is: ; in, This represents the actual pressure percentage in the corresponding area of ​​the affected foot. This is the median value of the reference pressure percentage for this region; In one implementation, the target gait phase is the support phase (heel strike), with a reference angle of 10° and an actual angle of 17°. =17-10=7°; If the median heel pressure is 50% and the actual heel pressure is 35%, then... ; The target knee flexion-extension angle is corrected based on angle deviation and pressure deviation, specifically as follows: Set angle correction coefficient and pressure correction factor , The value range is 0.3-0.7. The value range is 0.1-0.3; Corrected target knee flexion-extension angle The calculation formula is: ; The corrected target angle must be within the reference angle range of this gait phase. If it exceeds the range, the boundary value of the reference range shall be used.

[0034] In this implementation scheme, the flexion and extension angle of the target knee joint is corrected by the actual angle of the affected knee joint and the pressure distribution of the affected foot. This allows the target angle output by the prediction unit to be corrected according to the actual foot landing state and joint movement state of the affected side, reducing the control deviation caused by continuing to drive according to the original target angle when the affected side has gait deviation, abnormal foot landing, or insufficient knee joint movement.

[0035] Specifically, the steps to switch to the compliant support mode are as follows: Real-time monitoring of electromyographic signal amplitude of the affected lower limb, knee joint angle, plantar pressure distribution, and output torque on the affected side, specifically: The safety unit receives data transmitted from the acquisition unit and the control unit in real time; Among them, the amplitude of electromyography (EMG) signal of the affected lower limb is acquired by the surface EMG sensor of the acquisition unit to monitor the EMG signal amplitude of the quadriceps femoris and hamstring muscles in real time (to avoid misdiagnosis caused by abnormal signals of a single muscle). The angle of the affected knee joint was collected by an inertial measurement unit; The plantar pressure distribution on the affected side was collected by a plantar pressure sensor. The output torque is acquired by the torque sensor of the drive module; The safety unit receives data on the angle of the affected knee joint, the pressure distribution of the affected foot, and the output torque at 100Hz. It also receives the amplitude characteristics of the electromyographic signal of the affected lower limb after processing within a preset time window, which is used to obtain the data required for abnormal condition judgment. The amplitude of the electromyographic signal of the affected lower limb was compared with the spasticity judgment threshold; the angle of the affected knee joint was compared with the safe angle range; the plantar pressure distribution of the affected foot was compared with the pressure threshold range; and the output torque was compared with the maximum torque limit. Specifically: The comparison process involves judging four parameters simultaneously. If any one parameter meets the abnormal condition, the monitoring result is determined to have reached the corresponding preset threshold. When any monitoring result reaches the corresponding preset threshold, the output torque will be reduced to the support torque, specifically as follows: The supporting torque is set based on the patient's weight, the supporting capacity of the affected side, and the current gait stage of the knee joint. The calculation formula is as follows: ; in, The patient's weight (in kg). The acceleration due to gravity (value 9.8 m / s²) 2 ), The equivalent force arm of the lower leg (unit: m). The support coefficient (value 0.005-0.01) is used to maintain the basic support posture of the knee joint and reduce the risk of instability caused by insufficient support. The descent process adopts a linear descent method with a descent time set to 200ms, which means that the current output torque is linearly reduced to the support torque to avoid sudden torque changes that could cause impact on the knee joint. In one implementation, the patient weighs 60 kg, and the lower leg equivalent force arm is 0.3 m. Then the supporting torque The current output torque is approximately 11.29 N·m, which linearly decreases from 11.29 N·m to 1.411 N·m within 200 ms, with a decrease rate of (11.29-1.411) / 0.2≈49.38 N·m / s; Switch the affected knee joint drive module to passive compliant support mode, specifically as follows: During the switching process, the active assist output of the servo motor is reduced, the drive module is switched to low impedance follow-up control state, and the damping parameters are set by the control unit to keep the affected knee joint in a flexible support within a limited angle range. This allows the knee joint to achieve smooth flexion and extension under the patient's own force or external force, while avoiding rapid shaking or jamming of the knee joint. Meanwhile, the safety unit continuously monitors various parameters. If the abnormal state is resolved (e.g., the electromyographic signal amplitude recovers to below 5mV and the duration is >100ms), the system automatically exits the compliant support mode and resumes normal auxiliary control. If the abnormal state persists, the system will issue an alarm and maintain a compliant support state until manual intervention is required.

[0036] In this implementation scheme, the amplitude of electromyography signal on the affected side, the angle of the knee joint on the affected side, the pressure distribution of the plantar pressure on the affected side, and the output torque are monitored and thresholds are compared simultaneously. When any monitoring result reaches the corresponding threshold, the output torque is reduced and the system switches to a low-impedance follow-up control state. This enables the system to weaken the active drive in time when there is muscle spasm, abnormal joint angle, abnormal plantar pressure, or excessive torque, thereby reducing the adverse effects of continuous active drive on the knee joint under abnormal conditions.

[0037] Specifically, the steps for output torque feedback are as follows: The servo motor receives control commands corresponding to the output torque, specifically: The control unit will calculate the output torque The pulse width modulation (PWM) control command is converted into a servo motor. The duty cycle of the PWM command is linearly related to the output torque. The larger the duty cycle, the greater the output torque of the servo motor. Control commands are transmitted to the servo motor driver via a communication bus (such as a CAN bus). After receiving the command, the driver drives the servo motor to run. In one implementation, the output torque The corresponding PWM instruction has a duty cycle of 60%. After receiving the instruction, the driver controls the servo motor to run at a speed of 50 r / min. The harmonic reducer transmits the torque output from the servo motor to the affected knee joint, specifically as follows: The harmonic reducer is rigidly connected to the output shaft of the servo motor, and the reduction ratio is set to 1:50. It is used to convert the high speed and low torque of the servo motor into low speed and high torque. The torque output by the servo motor is amplified by a harmonic reducer and then transmitted to the mechanical joint of the affected knee via a drive shaft, driving the knee joint to flex and extend; the transmission efficiency of the harmonic reducer... At least 90%, the actual torque at the knee joint is the output torque of the servo motor × reduction ratio × transmission efficiency, reducing torque transmission loss; In one embodiment, the servo motor outputs a torque of 0.2508 N·m. After being amplified by a harmonic reducer (1:50) and considering a transmission efficiency of 90%, the torque transmitted to the knee joint is approximately 0.2508 × 50 × 0.9 ≈ 11.286 N·m, which is close to the target output torque at the knee joint. The torque sensor detects the actual output torque of the affected knee joint drive module in real time, specifically: A torque sensor is installed on the drive shaft between the harmonic reducer and the mechanical joint of the knee joint to detect the actual torque transmitted to the knee joint in real time. The detection frequency is 100Hz, which is consistent with the system sampling frequency; The detected actual output torque data is converted into a digital signal by an A / D converter and then transmitted to the control unit. The torque sensor has a measurement accuracy of no less than ±0.1 N·m; In one embodiment, the target output torque is 11.29 N·m, and the actual output torque detected by the torque sensor is 11.25 N·m, with a detection error of 0.04 N·m, which is within the allowable range. The actual output torque is fed back to the control unit to correct the output torque for the next control cycle, specifically as follows: The control unit receives the actual output torque from the torque sensor. Calculate the actual output torque and the target output torque. Deviation: ; The output torque for the next control cycle is corrected using a proportional control algorithm. The correction formula is as follows: ; in, The feedback correction factor has a value range of 0.1-0.3. The corrected output torque ensures that the deviation between the actual output torque and the target output torque does not exceed 0.5 N·m; In one implementation, , , , The output torque in the next control cycle The corrected output torque is used to reduce the torque deviation in the next control cycle.

[0038] In this implementation scheme, the actual output torque of the affected knee joint drive module is detected in real time by a torque sensor, and the actual output torque is fed back to the control unit to correct the output torque of the next control cycle, so that the actual torque at the knee joint end can be close to the target torque calculated by the control unit, and to compensate for the torque deviation caused by transmission loss, changes in patient resistance and mechanical clearance.

[0039] Specifically, such as Figure 3 As shown, the initial setting steps for actively participating in the judgment threshold are as follows: During the initialization phase, electromyographic signals of the unaffected lower limb were acquired during a calibrated walking state, and the baseline activation level of the corresponding muscles on the unaffected side was calculated. Specifically: The initialization phase involves the patient wearing the system for the first time, performing a calibration walk on a flat surface, taking 30 steps, and collecting electromyographic signals from the quadriceps femoris, hamstrings, tibialis anterior, and gastrocnemius muscles of the unaffected lower limb. The collected electromyographic signals were filtered, rectified, and integrated to obtain the activation level of the corresponding muscle on the healthy side at each step. Calculate the average muscle activation level over 30 steps, and use it as the baseline for the corresponding muscle activation level on the healthy side. ; In one implementation, the average activation level of the healthy side muscles over 30 steps is 2.0 mV·s. ; During the initialization phase, electromyographic signals of the affected lower limb were collected from the patient's affected lower limb at rest, and the baseline activation level of the corresponding muscle on the affected side was calculated. Specifically: The resting state is when the patient is sitting in a chair with the affected lower limb naturally relaxed, without exertion or movement, and the data collection time is 10 seconds; Electromyographic signals of the quadriceps femoris, hamstrings, tibialis anterior, and gastrocnemius muscles of the affected lower limb were collected and processed by filtering, rectification, and integration. The average muscle activation level over 10 seconds was calculated and used as the baseline for the corresponding muscle activation level on the affected side. ; Muscle activation levels on the affected side are typically lower at rest. In one implementation, the average level of muscle activation on the affected side within 10 seconds is 0.3 mV·s. ; Based on the baseline activation levels of the corresponding muscles on the healthy side and the corresponding muscles on the affected side, an initial threshold value for the active participation judgment of the corresponding muscle activation level on the affected side relative to the corresponding muscle activation level on the healthy side is set, specifically as follows: Actively participate in determining the initial value of the threshold The calculation formula is: ; in, This is the initial coefficient, with a value range of 0.5-0.6; This formula combines the resting baseline of the affected side and the walking baseline of the healthy side, so that the initial threshold can take into account both assist triggering and over-assist control. If the calculated initial value of the active participation judgment threshold exceeds the preset range, the initial value of the active participation judgment threshold will be limited to the range of 0.4-0.9. In one implementation, , , ,but The initial threshold for active participation in the judgment was 0.65, which is within the range of 0.4-0.9. If the patient's muscle strength on the affected side is extremely weak, the threshold can be appropriately lowered. Adjust the initial threshold value.

[0040] In this implementation plan, the initial value of the active participation judgment threshold is determined by calibrating the baseline of muscle activation level during walking on the healthy side and the baseline of muscle activation level during resting on the affected side. This allows the threshold setting to be combined with the individual muscle exertion foundation of the patient, avoiding the problem of the affected side being unable to trigger assistance or triggering assistance too frequently when using a uniform threshold. This serves as the initial parameter for subsequent assistance torque calculation and parameter adjustment.

[0041] Although preferred embodiments of the invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including both the preferred embodiments and all changes and modifications falling within the scope of the invention.

[0042] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. An intelligent exoskeleton system for controlling the knee joint of hemiplegic patients, characterized in that, include: The acquisition unit is used to simultaneously acquire electromyographic signals, joint angles, plantar pressure distribution, and gait phase information of the patient's healthy and affected lower limbs. The prediction unit takes the electromyographic signal of the healthy lower limb, the joint angle of the healthy side, the plantar pressure distribution of the healthy side, and the gait phase information of the healthy side as input, and outputs the target gait phase and the target knee flexion and extension angle of the affected knee joint in the current gait cycle through an individualized prediction model. The decision unit is used to calculate the activation level of the corresponding muscles on the healthy side and the corresponding muscles on the affected side. It compares the ratio of the activation level of the corresponding muscles on the affected side to the activation level of the corresponding muscles on the healthy side with the active participation judgment threshold. When the ratio is lower than the active participation judgment threshold, the assist torque of the affected knee joint is determined based on the ratio difference. The control unit is used to perform position-force hybrid control on the affected knee joint drive module, with the target knee joint flexion-extension angle as the position control target and the affected knee joint assist torque as the force control target, and outputs torque to assist the flexion-extension of the affected knee joint. The adjustment unit is used to adjust the calculation parameters of the knee joint assist torque and the active participation judgment threshold of the affected side based on the trend of the ratio of the activation level of the corresponding muscle on the affected side to the activation level of the corresponding muscle on the healthy side in a continuous gait cycle. The safety unit is used to monitor the amplitude of electromyographic signals of the affected lower limb, the angle of the affected knee joint, the pressure distribution of the affected foot, and the output torque. When the monitoring results reach the corresponding preset threshold, the output torque is reduced and the system switches to compliant support mode.

2. The intelligent exoskeleton system for controlling the knee joint of hemiplegic patients according to claim 1, characterized in that, The training steps for the individualized prediction model are as follows: The electromyographic signals, joint angles, plantar pressure distribution, and gait phase information of the unaffected lower limb were collected during multiple consecutive steps in a calibrated walking state. Simultaneously collect the actual gait phase and actual knee flexion-extension angle of the affected knee joint under calibrated assisted state; The input samples are the electromyographic signal of the healthy lower limb, the joint angle of the healthy side, the plantar pressure distribution of the healthy side, and the gait phase information of the healthy side. The output labels are the actual gait phase of the affected knee joint and the actual knee flexion and extension angle. The training process yields a mapping relationship between the gait characteristics of the healthy side and the target state of the knee joint on the affected side, and forms an individualized prediction model.

3. The intelligent exoskeleton system for controlling the knee joint of hemiplegic patients according to claim 2, characterized in that, The steps for constructing the input sample are as follows: The electromyographic signals of the unaffected lower limb were filtered, rectified, and feature extracted to obtain the temporal features of muscle activation on the unaffected side. Calculate the flexion-extension angular velocity of the healthy knee joint based on the angle of the healthy joint. Extract the heel strike time and toe lift time based on the plantar pressure distribution on the healthy side; The input sample consists of the temporal features of muscle activation on the healthy side, the flexion and extension angular velocity of the knee joint on the healthy side, the heel strike time, the toe lift-off time, and the gait phase information on the healthy side.

4. The intelligent exoskeleton system for controlling the knee joint of hemiplegic patients according to claim 1, characterized in that, The steps for determining the assist torque of the affected knee joint are as follows: The electromyographic signals of the unaffected lower limb were filtered, rectified, and integrated to obtain the activation level of the corresponding muscles on the unaffected side. The electromyographic signals of the affected lower limb were filtered, rectified, and integrated to obtain the activation level of the corresponding muscles on the affected side. Calculate the ratio of the activation level of the corresponding muscle on the affected side to the activation level of the corresponding muscle on the healthy side; Compare the proportion with the threshold for judging active participation; When the proportion is lower than the active participation judgment threshold, the assist torque of the affected knee joint is determined based on the difference between the proportion and the active participation judgment threshold, the activation level of the corresponding muscle on the healthy side, and the assist gain coefficient, and the assist torque of the affected knee joint does not exceed the preset assist upper limit.

5. The intelligent control exoskeleton system for knee joint of hemiplegic patient according to claim 4, characterized in that, The steps for actively participating in the judgment threshold and adjusting the boost gain coefficient are as follows: Read the ratio of the activation level of the corresponding muscle on the affected side to the activation level of the corresponding muscle on the healthy side within multiple consecutive gait cycles; Calculate the trend of proportion changes based on the proportion data; When the trend of the ratio change continues to rise, increase the threshold for active participation judgment and decrease the assist gain coefficient; When the trend of the ratio change continues to decline, maintain the active participation judgment threshold and increase the assist gain coefficient; When the trend of proportional change does not change significantly, maintain the current values ​​of the active participation judgment threshold and the assist gain coefficient.

6. The intelligent control exoskeleton system for knee joint of hemiplegic patient according to claim 1, characterized in that, The steps of position-force hybrid control are as follows: Read the target knee joint flexion and extension angle, and calculate the position control torque component based on the target knee joint flexion and extension angle; Read the assist torque of the affected knee joint and calculate the force control torque component based on the assist torque of the affected knee joint; The position control torque component and the force control torque component are weighted and fused to obtain the output torque; The output torque is input into the affected knee joint drive module to drive the affected knee joint to complete flexion and extension assisted movements.

7. The intelligent control exoskeleton system for knee joint of hemiplegic patient according to claim 1, characterized in that, The steps for real-time correction of the target knee flexion-extension angle are as follows: Real-time acquisition of knee joint angle and plantar pressure distribution on the affected side; Compare the affected knee angle and plantar pressure distribution with the reference knee angle and reference plantar pressure distribution corresponding to the target gait stage; When there are discrepancies in the comparison results, calculate the angle deviation and pressure deviation; The target knee flexion and extension angle is corrected based on angular and pressure deviations.

8. The intelligent control exoskeleton system for knee joint of hemiplegic patient according to claim 1, characterized in that, The steps to switch to the compliant support mode are as follows: Real-time monitoring of electromyographic signal amplitude of the affected lower limb, knee joint angle of the affected side, plantar pressure distribution and output torque of the affected side; The amplitude of electromyography signal of the affected lower limb is compared with the spasticity judgment threshold, the angle of the affected knee joint is compared with the safe angle range, the pressure distribution of the affected foot is compared with the pressure threshold range, and the output torque is compared with the maximum torque limit. When any monitoring result reaches the corresponding threshold, the output torque will be reduced to the support torque; Switch the affected knee joint drive module to passive compliant support mode.

9. The intelligent control exoskeleton system for knee joint of hemiplegic patient according to claim 1, characterized in that, The steps for output torque feedback are as follows: The servo motor receives control commands corresponding to the output torque. The harmonic reducer transmits the torque output by the servo motor to the affected knee joint; The torque sensor detects the actual output torque of the drive module on the affected knee joint in real time. The actual output torque is fed back to the control unit to correct the output torque for the next control cycle.

10. The intelligent exoskeleton system for controlling the knee joint of hemiplegic patients according to claim 1, characterized in that, The initial setting steps for actively participating in the judgment threshold are as follows: During the initialization phase, electromyographic signals of the unaffected lower limb of the patient were collected during a calibrated walking state, and the baseline activation level of the corresponding muscle on the unaffected side was calculated. During the initialization phase, electromyographic signals of the affected lower limb were collected in the patient's resting state, and the baseline activation level of the corresponding muscle on the affected side was calculated. Based on the baseline activation levels of the corresponding muscles on the healthy side and the baseline activation levels of the corresponding muscles on the affected side, an initial value for the active participation judgment threshold of the activation level of the corresponding muscles on the affected side relative to the activation level of the corresponding muscles on the healthy side is set.

Images