Search

CN-122005274-A - Rehabilitation exoskeleton self-adaptive control method and system based on physiological signals

CN122005274ACN 122005274 ACN122005274 ACN 122005274ACN-122005274-A

Abstract

The invention provides a physiological signal-based rehabilitation exoskeleton self-adaptive control method and system, which comprise the steps of obtaining multi-mode physiological signal data of a wearable actuating executing mechanism, preprocessing the multi-mode physiological signal data to obtain a synchronized multi-channel sensing signal sequence, extracting a characteristic data set, performing similarity calculation with a pre-built motion intention template library, determining motion intention type and intensity quantization value to select a control mode, simultaneously completing physiological state classification and obtaining state confidence coefficient, judging the safety of a pre-set motion track through a boundary detection algorithm, calculating a safety margin evaluation value, adjusting the track under a threshold value, and finally calculating target driving parameters to generate a real-time motion control instruction. The invention realizes the efficient browsing and analysis of the mass monitoring data, provides a comprehensive intelligent control solution for the wearable auxiliary equipment, and remarkably improves the safety, effectiveness and individuation level of rehabilitation training.

Inventors

  • FANG WENCHAO
  • NI JIAN
  • XU YUANHUA

Assignees

  • 浙江大学医学院附属第一医院(浙江省第一医院)

Dates

Publication Date
20260512
Application Date
20260414

Claims (11)

  1. 1. A physiological signal-based rehabilitation exoskeleton adaptive control method, which is characterized by comprising the following steps: acquiring multi-mode physiological signal data acquired by a wearable actuating executing mechanism, and performing time stamp alignment, resampling and standardized preprocessing on the multi-mode physiological signal data to obtain a synchronized multi-channel sensing signal sequence; Based on the synchronized multichannel sensing signal sequence, extracting a characteristic data set, performing similarity calculation on the characteristic data set and a pre-built motion intention template library, determining a current motion intention type and an intention intensity quantization value, determining a control mode type according to the current motion intention type and the intention intensity quantization value, and performing physiological state classification to obtain a current physiological state type and state confidence; Judging the safety of the preset motion trail through a boundary detection algorithm according to the current physiological state category, the state confidence coefficient and the preset motion trail, calculating a safety margin evaluation value, and adjusting the preset motion trail when the safety margin evaluation value is lower than a safety threshold value to obtain an adjusted motion trail; and calculating target driving parameters of an actuating mechanism according to the safety margin evaluation value, the adjusted motion trail, the control mode type and the intention intensity quantization value, and generating a real-time motion control instruction.
  2. 2. The method of claim 1, wherein the acquiring the multi-modal physiological signal data acquired by the wearable actuation actuator, performing time-stamp alignment, resampling and normalization preprocessing on the multi-modal physiological signal data to obtain the synchronized multi-channel sensing signal sequence, comprises: Acquiring multi-modal physiological signal data acquired by a wearable actuation execution mechanism, wherein the multi-modal physiological signal data comprises an articulation kinematics signal, an articulation dynamics signal, a muscle electrophysiology signal and a peripheral circulation physiological signal, the articulation kinematics signal comprises an articulation angle and an angular velocity, the articulation dynamics signal comprises an articulation moment and a movement resistance, the muscle electrophysiology signal comprises a multi-channel surface electromyographic signal, and the peripheral circulation physiological signal comprises an optical volume pulse signal; the starting point of a motor motion cycle is taken as a time reference anchor point, time stamps of signals of all channels in the multi-mode physiological signal data are aligned, signals with different sampling rates are interpolated into a unified time grid, and signal data after time synchronization are obtained; And respectively carrying out linear trending treatment and z-score standardization treatment on each channel signal in the signal data after time synchronization, eliminating direct current components and amplitude differences, and obtaining the multichannel sensing signal sequence after synchronization.
  3. 3. The method of claim 2, wherein the extracting a feature data set based on the synchronized multichannel sense signal sequence comprises: Applying a sliding window segmentation technology to the synchronized multichannel sensing signal sequence, setting the window length to be 0.5-3 seconds and the overlapping rate to be 50-75%, and obtaining a time window signal segment; calculating the mean value, variance, peak value, signal envelope value and inter-channel pearson correlation coefficient of each channel based on the time window signal segments to obtain a time domain feature vector; Performing spectrum analysis on each channel signal based on the time window signal segments, and calculating the energy duty ratio, the peak frequency position and the spectrum centroid in the predefined functional frequency band to obtain a frequency domain feature vector; Combining the degrees of freedom of joints in the joint kinematics signals, moment values in the joint dynamics signals and myoelectricity envelope values in the muscle electrophysiology signals to construct a six-dimensional to ten-dimensional state space vector; and splicing the time domain feature vector, the frequency domain feature vector and the state space vector according to columns to obtain the feature data set.
  4. 4. A method according to claim 3, wherein said computing similarity of the feature dataset to a library of pre-built motion intent templates, determining a current motion intent category and intent intensity quantization value, comprises: extracting a standard movement intention template from the pre-established movement intention template library, wherein the standard movement intention template is formed by preprocessing and extracting features of a standard active movement signal acquired clinically; comparing the current time window feature vector sequence in the feature data set with each standard motion intention template in the pre-built motion intention template library one by one, and calculating Euclidean distance, cosine similarity or dynamic time warping distance between the current time window feature vector sequence and each standard motion intention template to obtain similarity scores for the templates; selecting a template corresponding to the maximum value from all the similarity scores to obtain a best matching template identifier and a highest similarity score; based on the motion intention type corresponding to the best matching template identification, obtaining the current motion intention type; And carrying out normalization processing on the highest similarity score, and mapping the highest similarity score to a 0-1 interval to obtain the intent intensity quantized value.
  5. 5. The method of claim 4, wherein the determining a control pattern type from the current athletic intent category and the intent intensity quantization value comprises: comparing the highest similarity score with a preset effective intention threshold, judging that the effective active movement intention is detected when the highest similarity score is higher than the preset effective intention threshold, otherwise judging that the effective active movement intention is not detected; When it is determined that no effective active movement intention is detected, setting the control mode type to a passive assist mode; When it is determined that a valid active locomotion intent is detected and the intent intensity quantization value is below a first intensity threshold, setting the control mode type to an active cooperative mode; When it is determined that an effective active motion intent is detected and the intent intensity quantization value is above a second intensity threshold, the control mode type is set to an adaptive hybrid mode.
  6. 6. The method according to claim 5, wherein the similarity calculation adopts an elastic alignment matching method based on a time sequence similarity measure, and specifically comprises: Performing time axis nonlinear alignment on the time sequence feature vector sequences in the feature data set and each standard template sequence in the pre-built motion intention template library, constructing an accumulated distance matrix through a dynamic programming algorithm, and calculating an optimal alignment path to obtain a minimum distance measurement value allowing time warping; Determining whether effective pattern matching exists or not through a self-adaptive threshold judging mechanism based on the minimum distance measurement value, wherein the self-adaptive threshold is dynamically adjusted according to the statistical characteristics of the distance distribution in the pre-built motion intention template library, and matching confidence and pattern category identification are obtained; And according to the matching confidence coefficient and the pattern category identification, when the matching confidence coefficient exceeds a confirmation threshold value, taking the signal pattern corresponding to the matching event as a new sample into the pre-built motion intention template library, updating a representative template through cluster analysis, and realizing incremental learning and evolution optimization of the pre-built motion intention template library.
  7. 7. The method according to claim 6, wherein the extracting of the frequency domain feature vector comprises a rapid spectrum analysis method based on sparsity assumption, specifically comprising: Performing segmentation windowing processing on the synchronized multichannel sensing signal sequence, setting the window length to be 256-2048 sampling points and the overlapping rate to be 50%, and obtaining a multichannel time window signal matrix; Based on the multi-channel time window signal matrix, carrying out random hash bucket division on each channel signal through a sparse frequency spectrum recovery algorithm, calculating energy amplitude of each frequency component, comparing the energy amplitude with an energy threshold, judging that the corresponding frequency component is a significant energy frequency component when the energy amplitude is larger than the energy threshold, and positioning all the significant energy frequency components by utilizing hash collision to obtain a sparse frequency spectrum coefficient set, wherein the energy threshold is set to be 10-50% of the average value of the energy amplitudes of all the frequency components; Dividing the sparse frequency spectrum coefficient set according to a predefined functional frequency band, wherein the predefined functional frequency band comprises a low frequency band of 0-5Hz, a middle frequency band of 5-50Hz and a high frequency band of 50-250Hz, and calculating the energy duty ratio, the peak frequency position and the cross-channel coherence index of each frequency band to obtain the frequency domain feature vector.
  8. 8. The method of claim 7, wherein said classifying the physiological state to obtain a current physiological state class and a state confidence comprises: Extracting the frequency domain feature vector from the feature data set, constructing a Gaussian mixture probability model, and obtaining prestored Gaussian mixture probability model parameters, wherein the Gaussian mixture probability model parameters comprise mean value vectors, covariance matrixes and mixing weights of basic states, and the basic states comprise a fatigue state, a normal state and an excited state; Substituting the frequency domain feature vector into the mixed Gaussian probability model, calculating the Gaussian probability density value of the frequency domain feature vector in each basic state, and calculating the posterior probability of each basic state through a Bayesian formula to obtain state posterior probability distribution; Based on the state posterior probability distribution, selecting a state with highest posterior probability as the current physiological state category through a maximum posterior criterion, and taking a posterior probability value corresponding to the state with highest posterior probability as the state confidence; According to the current physiological state category, a preset state-parameter mapping table is queried, a motion speed adjustment coefficient, a moment limiting parameter and a motion amplitude recommended value aiming at the current physiological state category are obtained, and a personalized control parameter recommended value is generated.
  9. 9. The method of claim 8, wherein the determining the safety of the preset motion trajectory by a boundary detection algorithm according to the current physiological state category, the state confidence level and the preset motion trajectory, calculating a safety margin evaluation value, and adjusting the preset motion trajectory when the safety margin evaluation value is lower than a safety threshold value to obtain an adjusted motion trajectory, comprises: Constructing a multi-dimensional joint space, wherein the dimensions of the multi-dimensional joint space comprise angles, angular velocities and moments of various degrees of freedom of joints, defining a dynamic safety area and a tabu area boundary in the multi-dimensional joint space according to the current physiological state category and the state confidence, performing octree space subdivision on the multi-dimensional joint space, and establishing a hierarchical index structure to obtain a spatial index database; Acquiring current joint state point coordinates and a direction vector of the preset motion track, and inquiring a space region to which the current joint state point coordinates belong in the space index database through a point positioning algorithm to obtain a region type identifier, wherein the region type identifier comprises a safety region identifier, a boundary region identifier and a tabu region identifier; based on the region type identifier and the direction vector of the preset motion track, transmitting a detection ray along the direction vector of the preset motion track, and obtaining the shortest distance from the current position to the tabu boundary through the intersection calculation of the detection ray and the tabu region boundary, wherein the shortest distance is used as the safety margin evaluation value; When the safety margin evaluation value is lower than a safety threshold value, correcting the preset motion track, and adjusting track parameters through a gradient descent method to enable the corrected track to keep a safety distance from a tabu boundary, so as to obtain the adjusted motion track.
  10. 10. The method according to claim 1, further comprising real-time anomaly monitoring based on the feature data set and a historical database, generating an early warning signal and recommending intervention when anomalies are detected, comprising in particular: Carrying out layering segmentation on the historical data in the historical database according to a plurality of time scales of day, time and time to obtain a multi-stage time interval dividing structure; Calculating signal statistics, abnormal event counts and representative feature samples in corresponding time intervals for each time interval in the multi-level time interval dividing structure to obtain interval feature abstracts, and constructing k-d tree space indexes based on the interval feature abstracts; extracting a current time window feature vector from the feature data set, calculating a deviation degree from a historical normal mode, judging that an abnormality is detected when the deviation degree exceeds an abnormality detection threshold, and taking the current time window feature vector as an abnormality feature vector; Based on the abnormal feature vector, initiating k neighbor query in the k-d tree space index, setting a k value to be 5-20, and returning to a history time interval set with the nearest feature distance; And recursively entering a next-level index to locate each coarse-granularity time interval in the historical time interval set until a specific historical abnormal record is located, returning the specific historical abnormal record and corresponding follow-up development information, and generating the early warning signal and the recommended intervention measure.
  11. 11. A physiological signal-based rehabilitation exoskeleton adaptive control system, comprising: the data acquisition and preprocessing module is used for acquiring multi-mode physiological signal data acquired by the wearable actuating executing mechanism, and performing time stamp alignment, resampling and standardized preprocessing on the multi-mode physiological signal data to obtain a synchronized multi-channel sensing signal sequence; The motion intention recognition and state classification module is used for extracting a characteristic data set based on the synchronized multichannel sensing signal sequence, carrying out similarity calculation on the characteristic data set and a pre-built motion intention template library, determining a current motion intention type and an intention intensity quantization value, determining a control mode type according to the current motion intention type and the intention intensity quantization value, and carrying out physiological state classification to obtain a current physiological state type and state confidence; The safety track planning module is used for judging the safety of the preset motion track through a boundary detection algorithm according to the current physiological state category, the state confidence coefficient and the preset motion track, calculating a safety margin evaluation value, and adjusting the preset motion track when the safety margin evaluation value is lower than a safety threshold value to obtain an adjusted motion track; and the control instruction generation module is used for calculating a target driving parameter of an actuating mechanism according to the safety margin evaluation value, the adjusted motion trail, the control mode type and the intention intensity quantization value to generate a real-time motion control instruction.

Description

Rehabilitation exoskeleton self-adaptive control method and system based on physiological signals Technical Field The invention belongs to the technical field of medical rehabilitation instruments, and particularly relates to a rehabilitation exoskeleton self-adaptive control method and system based on physiological signals. Background The rehabilitation exoskeleton is an important auxiliary device in the current nerve rehabilitation field, and can assist patients with limb dysfunction to perform rehabilitation training and promote nerve function recovery. However, the traditional rehabilitation exoskeleton device mostly adopts a fixed mode for control, lacks the sensing and adapting ability to the real-time physiological state of the patient, and is difficult to be adjusted in a personalized way according to the actual demands of the patient. The rehabilitation exoskeleton in clinical application at present mostly depends on a preset motion track to carry out mechanical assistance, so that the active motion intention of a patient cannot be identified, and safety evaluation and adjustment cannot be carried out according to the physiological state of the patient. The mechanical rehabilitation training mode has limited effect, and can cause the safety problems of muscle spasm, joint injury and the like due to lack of a safety guarantee mechanism. In recent years, with the development of biological signal acquisition and processing technology, attempts have been made to introduce physiological data such as myoelectric signals into the control of rehabilitation exoskeleton. However, most of these schemes use only a single signal source, and it is difficult to cope with complex and variable clinical environments and individual differences of patients by using a simple threshold judgment method. Especially for the nervous system injury patient, the physiological signal is weak, unstable and has a large amount of noise, so that the prior art is difficult to realize accurate and reliable control. In addition, the existing rehabilitation exoskeleton system generally lacks an effective safety guarantee mechanism. Especially when the patient has dystonia or limited joint movement, the fixed movement pattern may cause discomfort or even injury, seriously affecting the continuity and safety of rehabilitation training. Therefore, there is a need for a rehabilitation exoskeleton control method that can integrate multiple physiological signals, achieve accurate intention recognition, provide safety guarantee, and be adaptively adjusted, so as to improve the effect and safety of rehabilitation training. Disclosure of Invention The invention aims to overcome the defects of the prior art, and provides a rehabilitation exoskeleton self-adaptive control method and system based on physiological signals, which can effectively integrate multi-mode physiological signals, realize accurate identification of the movement intention of a patient, real-time evaluation of physiological states and dynamic planning of safety tracks, thereby providing personalized, safe and effective rehabilitation training. In order to achieve the above purpose, the present invention provides the following technical solutions: The invention provides a physiological signal-based rehabilitation exoskeleton self-adaptive control method, which comprises the following steps: acquiring multi-mode physiological signal data acquired by a wearable actuating executing mechanism, and performing time stamp alignment, resampling and standardized preprocessing on the multi-mode physiological signal data to obtain a synchronized multi-channel sensing signal sequence; Based on the synchronized multichannel sensing signal sequence, extracting a characteristic data set, performing similarity calculation on the characteristic data set and a pre-built motion intention template library, determining a current motion intention type and an intention intensity quantization value, determining a control mode type according to the current motion intention type and the intention intensity quantization value, and performing physiological state classification to obtain a current physiological state type and state confidence; Judging the safety of the preset motion trail through a boundary detection algorithm according to the current physiological state category, the state confidence coefficient and the preset motion trail, calculating a safety margin evaluation value, and adjusting the preset motion trail when the safety margin evaluation value is lower than a safety threshold value to obtain an adjusted motion trail; and calculating target driving parameters of an actuating mechanism according to the safety margin evaluation value, the adjusted motion trail, the control mode type and the intention intensity quantization value, and generating a real-time motion control instruction. The invention has the beneficial effects that: 1. the invention integrates various physiological informatio