An exoskeleton imu sensor adaptive power consumption management optimization method

Abstract

The present application relates to energy consumption management technical field, especially to a kind of exoskeleton IMU sensor adaptive power consumption management optimization method, steps include: through the data of left knee and right knee, the instantaneous movement intensity index of each knee is calculated;The difference of left knee and right knee movement intensity is determined to determine the actual timing phase difference of both sides;In combination with the movement intensity of contralateral knee and its change gradient, the gait conversion urgency pre-judgment coefficient of this side knee is calculated;According to gait conversion urgency pre-judgment coefficient, the pre-judgment type promotion of this side knee sampling frequency is triggered, and the sampling frequency is stepped down when both sides are in low dynamic;Calculate the posture stability determination value and determine the quasi-static zero speed interval, turn off gyroscope in interval and reduce the resolution of accelerometer;According to information value density, determine the non-uniform time domain compression ratio of original data in local cache and the aggregation sending threshold value of wireless data packet.The present application can realize the significant reduction of power consumption on the premise of guaranteeing gait key data not lost.

Description

Technical Field

[0001] This invention relates to the field of energy management technology, and in particular to an adaptive power management optimization method for exoskeleton IMU sensors. Background Technology

[0002] Current power management strategies for exoskeleton IMU sensors generally employ fixed sampling and transmission frequencies, or simply perform threshold-based frequency tuning based on the motion amplitude collected by the sensor on the same side. However, during normal walking, the legs exhibit a strict alternating motion pattern. When one leg enters the low-dynamic support phase, the opposite leg is in the high-dynamic swing phase, resulting in a significant difference in the actual power consumption requirements of the two IMU sensors at any given moment. Existing methods do not consider this asymmetrical motion characteristic, leading to energy waste as the support-side sensor still collects a large amount of redundant data at a high frequency, while the swing-side sensor may miss crucial data during gait transitions due to its fixed low frequency. Furthermore, existing methods lack the ability to predict upcoming changes in the movement state of the own leg based on the movement trend of the opposite leg; frequency adjustment always lags behind actual movement changes, failing to prepare for frequency upsampling before gait transitions occur. Summary of the Invention

[0003] This invention provides an adaptive power management optimization method for exoskeleton IMU sensors, which can effectively solve the problems in the background art.

[0004] This invention provides an adaptive power management optimization method for exoskeleton IMU sensors, comprising the following steps: S1: Real-time acquisition of triaxial acceleration and triaxial angular velocity data of the left and right knees, gravity separation and sagittal projection of the acceleration data of each knee, and extraction of angular velocity scalar, tangential acceleration and normal acceleration of each knee; S2: Calculate the instantaneous motion intensity index of each knee within a sliding time window by combining angular velocity scalar and tangential acceleration; S3: Calculate the temporal cross-correlation function between the intensity of movement of the left and right knees to determine the actual temporal phase difference between the two sides; combine the intensity of movement of the opposite knee and its gradient of change to calculate the gait transition urgency prediction coefficient of the knee on this side. S4: Based on the gait transition urgency prediction coefficient, the sampling frequency of the knee on this side is increased in a predictive manner, and when both sides are in a low dynamic state, the sampling frequency is decreased in a stepwise manner. S5: Calculate the attitude stability judgment value based on the degree of deviation between normal acceleration and gravitational acceleration and the amplitude of angular velocity. Based on this, determine the quasi-static zero velocity interval, and shut down the gyroscope sampling path and reduce the accelerometer resolution within the interval. S6: Construct the data information value density based on the data distribution dispersion of angular velocity and the peak flatness of tangential acceleration, and determine the non-uniform temporal compression ratio of the original data in the local cache and the aggregation transmission threshold of wireless data packets accordingly.

[0005] Furthermore, in step S1, the gravity separation and sagittal projection of the acceleration data for each knee are specifically performed as follows: The attitude quaternion is updated using the original three-axis angular velocity data through quaternion differential equations and converted into a rotation matrix in the navigation coordinate system. The original three-axis acceleration data in the carrier coordinate system is rotated and projected to the navigation coordinate system. In the navigation coordinate system, the vertical axial acceleration and the front-to-back axial acceleration are respectively subjected to high-pass filtering with a cutoff frequency of 0.5Hz to remove the DC component of gravity and obtain the normal acceleration and tangential acceleration. After projecting the original three-axis angular velocity data onto the navigation coordinate system, only the angular velocity components around the inner and outer axes are extracted as angular velocity scalars, while the data of the other axes are discarded.

[0006] Furthermore, in step S2, the calculation of the instantaneous intensity index is specifically as follows: Calculate the angular acceleration scalar at the current time t: α(t)=[ω(t)-ω(tT s )] / T s ; In the formula, ω(t) is the angular velocity scalar at the current time t; T s The sampling period; Calculate the intensity index M of motion at the current time t. i (t): m i (t)=|α i (t)| / α0+|ω i (t)| / ω0+|a ti (t)| / g; M i (t)=(1 / N)×Σm i ; In the formula, the subscript i takes the form of L or R, representing the left knee or right knee respectively; α0 is the reference angular acceleration; ω0 is the reference angular velocity; a ti (t) represents tangential acceleration; g represents gravitational acceleration; T represents acceleration due to gravity. w Σm is the width of the sliding window; N is the number of sampling points within the window; i For time tT w All m at time t i The sum of (t).

[0007] Furthermore, in step S3, the calculation of the actual temporal phase difference and the gait transition urgency prediction coefficient of the lateral knee is specifically as follows: The left knee M built within the window L (t) and right knee M R The normalized cross-correlation function C of (t) LR (δ): ; In the formula, M Lavg and M Ravg These are the arithmetic mean values ​​of the intensity of movement of the left and right knees within the window, respectively. Calculate the actual timing phase difference δ argmax ; δ argmax =argmaxC LR (δ),(δ∈[-δ max ,0]); In the formula, δ max This represents the maximum search range; argmax indicates the value of the independent variable that makes the function within the parentheses reach its maximum value. Using the knee on the affected side as the target knee, the intensity index of the contralateral knee movement is denoted as M. op , calculate its gradient G. op (t): G op (t)=[M op (t)-M op (tT g )] / T g ; In the formula, the subscript 'op' represents the contralateral knee; when the knee on the same side is the left knee, 'op' represents the right knee, and when the knee on the same side is the right knee, 'op' represents the left knee; T g The time interval for gradient calculation; Calculate the predictive coefficient for gait transition urgency on this side of the knee: F pre (t)=P dec (t)×P phase (t)×P active (t); P dec (t)=min(1,-G op (t) / G max ); P phase (t)=max(0,1-|δ argmax -δ nRm | / δ tol ); P active (t)=min(1,M op (t) / M thr ); In the formula, G max The maximum reference gradient; δnRm δ represents the reference timing phase difference of bilateral knee joint movements during normal gait. tol For phase difference tolerance; M thr This is the effective threshold for contralateral exercise intensity.

[0008] Furthermore, in step S4, the specific operations of predictive upsampling and stepped downsampling are as follows: F pre (t) and threshold F thr Comparison shows that all three consecutive sampling points are ≥F thr The sampling frequency is switched to 200Hz and maintained until the peak value of ω(t) is reduced to less than half of the peak value. When F pre (t)≤threshold F thr And satisfy M i (t)<0.5 and M op When (t)>1.5, the sampling frequency is reduced by one level every 20 sampling periods, with each reduction being 10% of the current value, down to a minimum of 50Hz; during frequency reduction, if F pre (t) rises above the threshold F thr Then the frequency reduction will stop and the frequency will increase.

[0009] Further, step S5, calculating the attitude stability determination value and performing power reduction operations within the quasi-static zero-speed range, specifically involves: Calculate the deviation ε of gravity projection i (t): ε i (t)=|a ni (t)| / g; In the formula, a ni (t) represents the normal dynamic acceleration of the i-th knee; g represents the gravitational acceleration; Calculate the attitude stability judgment value S stb (t): S stb (t)=exp(-|ω i (t)| / ω Ref )×(1-ε i (t) / ε max ) In the formula, ω Ref ε is the reference angular velocity; max This represents the maximum permissible deviation. When S stb (t) All five consecutive sampling points are greater than the quasi-static decision threshold S thrWhen the system enters the quasi-static zero-speed range, it executes the following: the sampling clock gating of the gyroscope yaw and pitch axes is stopped; the accelerometer resolution is reduced from 16 bits to 10 bits, and the sampling axes are reduced to the normal and tangential axes; the accelerometer hardware threshold comparison is interrupted, and when the normal or tangential acceleration amplitude exceeds the wake-up threshold, an interrupt is generated to wake up the processor to exit the quasi-static zero-speed range and restore high-resolution sampling of all axes.

[0010] Further, in step S6, determining the non-uniform temporal compression ratio of the original data in the local cache and the aggregation transmission threshold of the wireless data packets specifically involves: Calculate the data distribution dispersion H w : H w =-Σp m ×log2(p m ), (m=1~K); In the formula, K is the number of intervals dividing the range of angular velocity amplitude; p m This represents the percentage of sampling points falling within the m-th interval; Calculate peak flatness Q pk : Q pk =max(|a t |) / [(1 / N)×Σ|a t |]; In the formula, max(|a t |) represents the maximum value of the absolute value sequence of tangential acceleration within the window; [(1 / N)×Σ|a t |] represents the arithmetic mean of the sequence of absolute values ​​of tangential acceleration within the window; Constructing the data information value density ρ ifo : ρ ifo =H w ×log2(1+Q pk )×tanh(M i (t) / M Ref ); In the formula, M Ref For reference purposes regarding exercise intensity; Determine the non-uniform temporal compression ratio: When ρ ifo <ρ Low When R = 0.8, cmp =floor(ρ Low / (ρ ifo +0.01); When ρ ifo ≥ρ Low At that time, R cmp =1; In the formula, floor() represents rounding down; Adjust the aggregated sending threshold: L thr =L bs ×[1+ρ ifo / (ρ ifo +ρ Ref )]; In the formula, L bs Basic aggregation threshold; ρ Ref For reference information density.

[0011] Furthermore, in step S4, after the predictive upsampling trigger, the following operations are also performed: The exercise intensity index M calculated for the left and right knee sides within the current window will be used respectively. L (t) and M R (t) Report to the main control unit; the main control unit calculates the relative ratio λ=M L / M R ; When λ>1.8, the transmission time slot length of the left knee is configured to be 1.8 times that of the right knee, and the allowable load of a single transmission of the left knee is configured to be 1.5 times that of the right knee. When λ < 0.55, the right knee transmission time slot length is configured to be 1.8 times that of the left knee, and the allowable load for a single transmission of the right knee is configured to be 1.5 times that of the left knee. When λ is between 0.55 and 1.8, the length of the transmission slot on both sides and the allowable load per transmission are equal; After allocation is completed, the master controller broadcasts the time slot allocation table and load configuration table for the next cycle. Each node starts sending within the specified time window. The time windows do not overlap, and the amount of data sent in a single session does not exceed the allowed load.

[0012] Furthermore, it also includes step S7: calculating the global health degradation factor and performing frequency derating correction, specifically: Calculate the energy degradation factor D Q : D Q =(Q Rtd -Q Rem ) / (Q Rtd +Q Rem ) In the formula, Q Rtd Q represents the battery's rated capacity. Rem This is the current remaining battery level; Calculate the temperature degradation factor D t : D t =1 / [1+exp(-(T j -T Ref ) / T0)] In the formula, T jFor internal junction temperature; T Ref T0 represents the ambient reference temperature; T0 represents the width of the temperature-sensitive feature. Calculate the global health degradation factor D hth : D hth =D Q +D t -D Q ×D t ; Execution frequency derating correction: f s,final =f s ×(1-D hth )+f s,min ×D hth ; f t,final =f t ×(1-D hth )+f t,min ×D hth ; In the formula, f s f is the sampling frequency; t f is the transmission frequency; s,min The minimum sustain sampling frequency; f t,min To maintain the minimum transmission frequency; Determine if emergency mode is triggered: D hth Emergency limp power mode is triggered when the power consumption is >0.85; 0.60 <D hth If the value is ≤0.85, it is marked as a sub-healthy state, and the wireless confirmation and retransmission mechanism is turned off and replaced with one-way broadcast transmission.

[0013] Furthermore, in step S7, after triggering the emergency limp power mode, the following steps are executed sequentially: f s,final The frequency is forcibly fixed at 25Hz, and the frequency increase request from step S4 is no longer responded to. Clock gating is applied to all sampling paths of the gyroscope and power supply is cut off to completely stop angular velocity data acquisition; The periodic wake-up function of the wireless communication module is disabled, and the radio frequency circuit is only activated during data transmission. The accelerometer sampling axis is reduced to retain only the sagittal plane normal axis, the resolution is reduced to 8 bits, and a hardware threshold comparison interrupt is configured. After the processor is woken up, it immediately reads the normal acceleration value and the current timestamp, encapsulates them into a minimum length broadcast frame, and bursts them with the maximum transmit power for a duration not exceeding 5ms. After transmission is complete, the processor clears the interrupt flag, puts the RF circuit and data processing module back into sleep mode, reduces the clock frequency to 1MHz, and only keeps the interrupt detection circuit and accelerometer powered.

[0014] The technical solution of this invention can achieve the following technical effects: This invention reduces data dimensionality by extracting effective motion components in the sagittal plane and discarding irrelevant axial data. It also constructs a continuous motion intensity index that integrates angular acceleration, angular velocity, and tangential acceleration to smoothly reflect gait dynamic changes. Based on this, it uses bilateral motion temporal correlation to calculate the gait transition urgency prediction coefficient, enabling the sensor sampling frequency to be adjusted ahead of the actual gait transition moment. This effectively overcomes the shortcomings of traditional methods in lag adjustment and significantly reduces the power consumption of invalid acquisition while ensuring gait phase determination capability. This significantly extends the continuous working endurance of the exoskeleton system under complex walking conditions and ensures that key transient data of gait transitions are not lost. Attached Figure Description

[0015] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0016] Figure 1 A flowchart illustrating the adaptive power management optimization method for exoskeleton IMU sensors. Detailed Implementation

[0017] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0018] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. The terminology used in this specification is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention. The term "and / or" as used herein includes any and all combinations of one or more of the associated listed items.

[0019] This invention relates to an adaptive power management optimization method for exoskeleton IMU sensors, such as... Figure 1As shown, the process includes multiple steps, S1 to S6. These steps construct a power consumption prediction and differentiated management system that fully utilizes the alternating movement patterns of the legs during human walking. The temporal correlation of bilateral knee joint movements is quantified into predictive coefficients with clear physical meaning. These coefficients precisely guide the real-time adjustment of the sampling frequency, wireless transmission strategy, and power supply status of each sensor, thereby achieving a significant reduction in power consumption while ensuring that key gait data is not lost. The specific details of each step are as follows: S1: Real-time acquisition of triaxial acceleration and triaxial angular velocity data of the left and right knees, gravity separation and sagittal projection of the acceleration data of each knee, and extraction of angular velocity scalar, tangential acceleration and normal acceleration of each knee; S2: Calculate the angular acceleration scalar based on the angular velocity scalar, and combine the angular acceleration scalar, angular velocity scalar, and tangential acceleration within a sliding time window to calculate the instantaneous motion intensity index of each knee. The purpose of this step is to compress the three-dimensional motion information of the knee joint in the sagittal plane (the sagittal plane is a section that divides the human body longitudinally into left and right parts) into a dimensionless continuous index. This index is much higher in the swing phase than in the stance phase, providing a quantitative basis for subsequent gait phase determination.

[0020] S3: Calculate the temporal cross-correlation function between the intensity indices of the left and right knees to determine the actual temporal phase difference between the two sides; combine the intensity of the contralateral knee's movement and its gradient to calculate the gait transition urgency prediction coefficient for the local knee. This step utilizes the physical characteristic of the strict alternation of the legs during human walking, that is, when the contralateral knee begins to decelerate from the peak of the swing phase, the local knee is about to transition from the support phase to the swing phase, thus achieving advanced prediction of gait transition.

[0021] S4: Based on the urgency prediction coefficient of gait transition, the sampling frequency of the knee on this side is pre-judged and increased. When both sides are in low dynamics, the sampling frequency is reduced stepwise. This step converts the prediction coefficient into the actual frequency adjustment action, so that the frequency increase operation is ahead of the moment of gait transition, which fundamentally solves the problem of delayed recognition in traditional methods.

[0022] S5: Calculate the attitude stability judgment value based on the deviation between normal acceleration and gravitational acceleration and the amplitude of angular velocity. Based on this, determine the quasi-static zero-velocity interval. Within the interval, shut down the gyroscope sampling path and reduce the accelerometer resolution, thereby significantly reducing the power consumption of invalid acquisition by identifying the quasi-static state of the knee joint.

[0023] S6: Construct the data information value density based on the data distribution dispersion of angular velocity and the peak flatness of tangential acceleration. Based on this, determine the non-uniform time-domain compression ratio of the original data in the local cache and the aggregation transmission threshold of wireless data packets. This enables the sensor to actively reduce the amount of data transmitted wirelessly and the packet transmission frequency within the low-value data window, further reducing wireless communication, the main source of power consumption.

[0024] In the core link consisting of S1 to S6, S1 to S3 are the perception and prediction layers, and S4 to S6 are the execution and optimization layers. The two layers form a closed loop through the prediction coefficient output by S3 and the information value density fed back by S6, so that the power management strategy can be adaptively adjusted according to the gait in real time.

[0025] Preferably, in step S1, the gravity separation and sagittal projection of the acceleration data of each knee are performed as follows: S1.1: Using the original three-axis angular velocity data, update the attitude quaternion using quaternion differential equations and convert it into a rotation matrix in the navigation coordinate system. The quaternion differential equations are as follows: ; In the formula, q is the attitude quaternion (q=[q0,q1,q2,q3)). T ω is a four-dimensional unit vector representing the rotational attitude of the vehicle coordinate system relative to the navigation coordinate system; ω is a three-axis angular velocity vector (measured by a gyroscope, in rad / s), with the form ω=[0,ω... x ,ω y ,ω z ] T (Extend the angular velocity vector into a pure quaternion form so that it can be multiplied with q in a quaternion manner); This represents quaternion multiplication; dq / dt is the rate of change of the attitude quaternion over time.

[0026] q0, q1, q2, and q3 are the four components of the attitude quaternion, which are dimensionless and satisfy q0 2 +q1 2 +q2 2 +q3 2 =1 unit quaternion constraint; ω x ω y ω z These are the angular velocity components around the X, Y, and Z axes in the carrier coordinate system, respectively, in rad / s.

[0027] The attitude quaternion q reflects the orientation of the IMU sensor relative to the navigation coordinate system at the current moment. The components of q change in real time with knee joint movement, where q0 represents the scalar component of rotation, and q1, q2, and q3 represent the rotational components about the three axes, respectively. By continuously recursively updating q, the sensor's attitude changes in three-dimensional space can be tracked in real time, providing a basis for subsequently rotating the acceleration from the carrier coordinate system to the navigation coordinate system.

[0028] S1.2: Using the rotation matrix R(q), the original triaxial acceleration data a in the carrier coordinate system is transformed. body =[a x ,a y ,a z ] T Rotate the projection onto the navigation coordinate system a nav =R(q)×a body The vertical acceleration a in the navigation coordinate system is obtained. up and horizontal forward and backward acceleration a north The navigation coordinate system corresponds to the coordinate directions of human movement.

[0029] S1.3: In the navigation coordinate system, the vertical axial acceleration and the front and rear axial acceleration are respectively subjected to high-pass filtering with a cutoff frequency of 0.5Hz to filter out the DC component of gravity and obtain the normal acceleration and tangential acceleration; after projecting the original three-axis angular velocity data to the navigation coordinate system, only the angular velocity components around the inner and outer axes are extracted as angular velocity scalars, and the other axial data are discarded.

[0030] The frequency domain separability of human gait frequency and gravity signal. Gravitational acceleration in the navigation coordinate system exhibits an approximately DC constant component (approximately 9.8 m / s² in the upward direction). 2 The dynamic acceleration generated by the knee joint during walking is mainly concentrated in the 1-3Hz frequency band (normal gait frequency is about 1-2Hz, and its harmonic components can reach 3-5Hz). Therefore, setting a cutoff frequency of 0.5Hz can effectively filter out the gravity DC bias while completely preserving the gait dynamic signal, so that the subsequently extracted "normal dynamic acceleration" and "tangential dynamic acceleration" truly reflect the vibration and impact generated by the knee joint movement, rather than the gravity projection component introduced by the sensor's static tilt angle. Without this filtering, the gravity component will continue to be superimposed on the exercise intensity index M. i In (t), the support phase M value is raised, the dynamic range of the swing phase and the support phase is compressed, and the sensitivity of gait phase discrimination is reduced.

[0031] Preferably, in step S2, the calculation of the instantaneous motion intensity index is specifically as follows: S2.1: Calculate the angular acceleration scalar α(t) at the current time t. Angular acceleration is the first derivative of angular velocity with respect to time, reflecting the degree of change in the angular velocity of the knee joint. It is calculated as follows: α(t)=[ω(t)-ω(tT s )] / T s ; In the formula, ω(t) is the angular velocity scalar at the current time t, in rad / s; T s The sampling period is α(t), in seconds; α(t) is the angular acceleration scalar, in rad / s. 2 .

[0032] ω(t) is the angular velocity scalar at the current sampling time; ω(tT) s Let ω be the angular velocity scalar at the previous sampling time. The difference between the two values ​​reflects the change in angular velocity within one sampling period. Divide by T. s The rate of change per unit time is obtained. The larger the |α(t)|, the more rapid the knee joint is undergoing acceleration or deceleration. The moment when the sign of α(t) changes from positive to negative or from negative to positive corresponds to the transition boundary of the gait phase, such as the transition from the acceleration phase of the swing phase to the deceleration phase, or the transition from the stationary phase of the support phase to the start of the swing phase.

[0033] S2.2: Calculate the instantaneous motion intensity component m at the current time t. i (t) and window-average exercise intensity index M i (t). The instantaneous motion intensity component is a dimensionless instantaneous value that characterizes the degree of motion activity of the knee joint at a certain moment; the window-averaged motion intensity index is its average value over a 0.3s window, used to smooth short-term fluctuations and provide stable input for cross-correlation calculations: m i (t)=|α i (t)| / α0+|ω i (t)| / ω0+|a ti (t)| / g; M i (t)=(1 / N)×Σm i ; In the formula, the subscript i takes the value L or R, representing the left knee or the right knee, respectively; α0 = 10 rad / s 2 The reference angular acceleration is determined based on the typical peak range of the knee joint swing phase angular acceleration in a normal person's gait (approximately 8~12 rad / s). 2 ), used to normalize the angular acceleration term to the dimensionless order of 0 to 3; ω0 = 2.0 rad / s is the reference angular velocity, which is based on the typical value of the knee joint swing phase angular velocity in normal human gait (about 1.5~2.5 rad / s), and is used to normalize the angular velocity term to the dimensionless order of 0~2. a ti (t) represents the tangential acceleration of the i-th knee extracted in step S1.3, in m / s². 2 g = 9.8 m / s 2 It is the acceleration due to gravity; T w =0.3s is the width of the sliding window, which is based on the fact that a complete gait cycle is about 1~1.2s, and 0.3s is about one-quarter of that, which can preserve sufficient waveform characteristics without being overly smoothed; N is the number of sampling points within the window, N=T w / T s ; m i (t) represents the instantaneous intensity component of motion, which is dimensionless; M i (t) is a dimensionless index of the average intensity of exercise within a window.

[0034] |α i (t)| / α0 is the normalized amplitude of angular acceleration, reflecting the intensity of dynamic change; |ω i (t)| / ω0 is the normalized amplitude of the angular velocity, reflecting the magnitude of the rotational kinetic energy; |a ti (t)| / g is the normalized amplitude of tangential acceleration, reflecting the linear impact generated by muscle exertion. After linear superposition of the three, all three values ​​are at a high level when the knee joint is undergoing intense movement, m i (t) takes a relatively large value; when the knee joint is at rest, all three terms are close to zero, m i (t) takes extremely small values. M i The physical meaning of (t) lies in compressing the complex multidimensional motion of the knee joint into a scalar value, the magnitude of which directly reflects the intensity of the knee joint motion at the current moment. i A larger value for (t) indicates more intense knee joint movement, requiring a higher sampling frequency from the sensor to capture motion details; M i A smaller (t) indicates that the motion tends to be smoother, allowing for a reduction in the sampling frequency to save power.

[0035] Preferably, in step S3, the calculation of the actual temporal phase difference and the gait transition urgency prediction coefficient of the knee on this side is specifically as follows: The left knee M built within the window L (t) and right knee M R The normalized cross-correlation function C of (t) LR(δ), the cross-correlation function is used to measure the similarity between two waveforms at different time offsets, and it is calculated as follows: ; In the formula, M Lavg and M Ravg These are the arithmetic mean values ​​of the intensity indices of the left and right knee movements within the window, used to shift the waveform center to zero and eliminate the influence of DC offset on the correlation. δ is the time offset independent variable, in seconds, which means that the right knee waveform is shifted by δ on the time axis and then compared with the left knee waveform. N is the number of sampling points within the window.

[0036] At the current time t, the system has already collected and stored data from tT. w All N sampling points up to t (at a sampling rate of 100Hz, T w =0.3s (30 sampling points). Cross-correlation calculates the similarity between two fully buffered waveform sequences, M in the formula. L (t-nT s ) and M R (t-nT s +δ) are all read from this acquired buffer. The cross-correlation function calculates the inner product similarity of two waveforms at different offsets by sliding. C LR The range of (δ) is [-1, 1]. The closer it is to 1, the more similar the shapes of the two waveforms are at that offset. The closer it is to -1, the opposite the shapes are. The closer it is to 0, the no correlation is indicated.

[0037] S3.2: In normal gait, the waveforms of the intensity of left and right knee movements are similar in shape but have a fixed time shift (approximately half a gait cycle). Therefore, at a specific δ point, C... LR (δ) will reach a maximum value close to 1, which is the actual temporal offset of the bilateral motion. The specific calculation is as follows: δ argmax =argmaxC LR (δ),(δ∈[-δ max ,0]); In the formula, δ max =0.6s is the maximum search range, which is based on the fact that a normal person's complete gait cycle is about 1.0~1.2s, and half of it is 0.5~0.6s. Therefore, the search range covers all possible offsets of half a gait cycle. argmax means making C LR (δ) The value of δ when it reaches its maximum value; δ argmaxThis represents the actual temporal phase difference between the two knee joint movements. A positive value indicates that the left knee movement leads the right knee movement, while a negative value indicates that the right knee movement leads the left knee movement. Since the knee joint movements in a normal gait follow a strict alternation pattern, with the contralateral knee as a reference, the movement of the knee on the current side will necessarily lag behind the contralateral knee (δ≤0). Therefore, the search range is limited to [-δ...]. max [0] has covered all effective gait phase differences. When the left knee is taken as the home side, a negative δ indicates that the right knee is leading; when the right knee is taken as the home side, a negative δ indicates that the left knee is leading. If it is necessary to predict both sides simultaneously, step S3 can be executed independently with the left knee and right knee as the home side respectively.

[0038] S3.3: Taking the knee on this side as the target knee, the intensity index of the movement of the contralateral knee is denoted as M. op , calculate its gradient G. op (t): G op (t)=[M op (t)-M op (tT g )] / T g ; In the formula, the subscript op represents the opposite knee. When the knee on the same side is the left knee, op represents the right knee, and when the knee on the same side is the right knee, op represents the left knee. T g =0.1s is the gradient calculation time interval, and its value is determined to be sufficient to filter out high-frequency sampling noise above 10Hz, while avoiding excessive smoothing of feature abrupt changes within about 0.1~0.2s during gait transition.

[0039] According to the physical laws of bilateral alternating movement in the human body, when the contralateral knee is at the end of the swing phase (i.e., the contralateral knee joint transitions from rapid flexion and extension to extension and braking), its intensity of movement rapidly decreases from its peak. op (t) shows a large negative value. At this time, the knee on this side is about to transition from the support phase to the swing phase. This is the most reliable precursor signal for predicting the gait transition of the knee on this side, which is about 0.1 to 0.2 seconds earlier than directly detecting the movement changes of the knee on this side itself. This allows the present invention to achieve advanced prediction.

[0040] S3.4: Calculate the predictive coefficient F for the gait transition urgency of the knee on this side. pre (t). This coefficient is obtained by multiplying three independent sub-factors, each of which corresponds to a physical judgment condition: F pre (t)=P dec (t)×P phase (t)×P active (t); P dec (t)=min(1,-G op (t) / Gmax ); P phase (t)=max(0,1-|δ argmax -δ nRm | / δ tol ); P active (t)=min(1,M op (t) / M thr ); In the formula, G max =8.0s -1 The maximum reference gradient is determined based on the typical upper limit of the rate of decrease in intensity at the end of the swing phase in normal individuals (approximately 5-10 seconds). -1 ), used to normalize the deceleration intensity; δ nRm =0.5s is the reference timing phase difference of bilateral knee joint movement under normal gait, which corresponds to half a gait cycle of bilateral alternating movement when a normal person walks; δ tol =0.20s is the phase difference tolerance, which is based on the typical range of gait cycle variation in normal people (approximately ±0.15~0.20s). Exceeding this range indicates that there is an abnormality in the gait. M thr =0.3 is the effective threshold for contralateral movement intensity. Its value is based on the boundary between the intensity of the support phase and the swing phase (the M value of the support phase is usually between 0.2 and 0.5, and the swing phase is usually between 2 and 8). A value below 0.3 indicates that the knee joint is basically still and is not enough to trigger the prediction.

[0041] The three sub-factors correspond to three independent conditions: P dec (t) reflects whether the contralateral knee is decelerating (G) op (t) < 0 and the larger the absolute value, the closer the factor is to 1); P phase (t) reflects whether the bilateral phase difference is within the normal range (δ) argmax Compared with the benchmark value δ nRm The smaller the deviation, the closer the factor is to 1); P active (t) reflects whether the current exercise intensity of the contralateral knee is sufficient (M) op The larger (t) is, the closer the factor is to 1).

[0042] F pre The range of F(t) is [0,1]. pre (t) The closer it is to 1, the more simultaneously all three conditions are met, meaning that the knee on this side will transition from the support phase to the swing phase in a very short time (approximately 0.1~0.2s), requiring an immediate increase in the sampling frequency to capture the gait transition transient. F pre(t) The closer it is to 0, the less likely a gait transition is to occur, allowing for a lower sampling frequency to save power. Compared to traditional methods based on the amplitude threshold of the local motion, this invention achieves true advance prediction by considering the motion trend of the opposite side.

[0043] Preferably, in step S4, the specific operations of predictive upsampling, stepped downsampling, and bandwidth allocation are as follows: F pre (t) and threshold F thr =0.75 comparison, all three consecutive sampling points are ≥F thr When the opposite knee is in the rapid braking phase at the end of the swing phase, and the current knee is about to enter the pre-swing phase, the sampling frequency is switched to 200Hz and maintained until the peak value of ω(t) drops to less than half of the peak value. The reason for maintaining the frequency until the peak value of ω(t) drops to less than half of the peak value before exiting the high-frequency state is that the most critical data in the gait transition transient is not the peak time itself, but the complete process from the start of the swing phase to the arrival of the peak value.

[0044] When F pre (t)≤threshold F thr And satisfy M i (t)<0.5 (this knee is in the support phase) and M op When (t)>1.5 (the contralateral knee is in the swing phase), it indicates that the knee of this side is in the middle of the support phase (low dynamics), and the contralateral knee is in the middle of the swing phase (high dynamics). Both sides are in a stable alternating motion phase, and there is still a long time before the next gait transition of this knee. Therefore, the sampling frequency can be reduced by one level every 20 sampling cycles, with each reduction being 10% of the current value. Gradual reduction allows the attitude calculation filter sufficient time to adapt to the new sampling rate, with a minimum of 50Hz. If F... pre (t) rises above the threshold F thr The frequency reduction process will then be terminated immediately and the frequency increase process will begin.

[0045] Preferably, in step S5, calculating the attitude stability determination value and performing power reduction operations within the quasi-static zero-speed range specifically involves: S5.1: Calculate the gravity projection deviation ε i (t): ε i (t)=|a ni (t)| / g; In the formula, a ni (t) represents the normal dynamic acceleration of the i-th knee extracted in step S1.3, in m / s². 2 a ni (t) is the normal dynamic acceleration after high-pass filtering, which has filtered out the DC component of gravity. When the knee joint is completely stationary, a ni (t) should be equal to 0, εi (t)=0; when there is vertical impact or vibration, a ni (t) is not 0, ε i (t)>0; g=9.8m / s 2 This is the acceleration due to gravity.

[0046] Gravity projection deviation ε i (t) is the ratio of normal dynamic acceleration to gravitational acceleration, used to measure the degree of vertical vibration at the current moment, ε i The smaller the value of ε, the more stable the knee joint is in the vertical direction, and the more likely it is to be in the quasi-static phase of the support phase; i The larger the (t) value, the more vertical impact or vibration is present, and the knee joint is in a dynamic phase, so power consumption reduction should not be performed.

[0047] S5.2: Calculate the attitude stability judgment value S stb (t): S stb (t)=exp(-|ω i (t)| / ω Ref )×(1-ε i (t) / ε max ) In the formula, ω Ref =1.5rad / s is the reference angular velocity, which is based on the typical upper limit of the knee joint micro-motion angular velocity in the mid-stance phase of the stance phase. The knee joint angular velocity in the mid-stance phase of a normal person is usually below 0.5rad / s, and 1.5rad / s provides about three times the margin. ε max =0.3 is the maximum allowable deviation, and its value is based on the fact that the vertical dynamic acceleration in the middle of the support phase generally does not exceed 30% of the gravitational acceleration.

[0048] S stb (t) The closer to 1, the more likely the knee joint is to simultaneously meet the conditions of "no rotation" and "no vibration," and is in a highly stable state (i.e., the quasi-static zero-velocity range in the middle of the support phase). S stb The closer (t) is to 0, the more likely it is that at least one condition is not met, at which point the knee joint is in motion or under impact, and power reduction should not be performed.

[0049] S5.3: When S stb (t) All five consecutive sampling points are greater than the quasi-static decision threshold S thrWhen the velocity reaches 0.85, the system is determined to have entered the quasi-static zero-speed range. Upon entry, the following power reduction operations are performed: the sampling clock gating of the gyroscope's yaw and pitch axes is stopped; the accelerometer resolution is reduced from 16 bits to 10 bits, and the sampling axes are reduced from three axes to two axes: normal and tangential; the accelerometer hardware threshold comparison is interrupted, and the system will wake up when the normal or tangential acceleration amplitude exceeds the wake-up threshold of 1.5 m / s². 2 An interrupt is generated to wake up the processor, exit the quasi-static zero-speed range, and resume full-axis high-resolution sampling. The power saving effect of this set of operations is significant: turning off the gyroscope reduces the total power consumption of the sensor, reducing the resolution of the accelerometer further saves the total power consumption of the sensor, and reducing the sampling axis further saves the total power consumption of the sensor. The combination of these three factors can significantly reduce the power consumption of the sensor in this range.

[0050] Preferably, in step S6, determining the non-uniform temporal compression ratio of the original data in the local cache and the aggregation transmission threshold of the wireless data packets specifically involves: S6.1: Calculate the data distribution dispersion H w : H w =-Σp m ×log2(p m ), (m=1~K); In the formula, K=8 is the number of intervals to divide the angular velocity amplitude range, which is [-10, 10] rad / s. The range is evenly divided into 8 intervals, each with a width of 2.5 rad / s. p m This represents the proportion of the number of sampling points within the current window that fall within the m-th interval out of the total number of sampling points N.

[0051] Data distribution dispersion H w H is used to measure the uniformity of the distribution of angular velocity amplitude within the current window. w A larger value indicates a more dispersed distribution of angular velocity amplitudes within the window, resulting in richer information about amplitude variations and higher data acquisition value. In gait, the angular velocity amplitude distribution during the oscillation phase is relatively dispersed (from slow to fast and back to slow), H... w Larger; the angular velocity of the supporting phase is concentrated near zero, H w Smaller.

[0052] S6.2: Calculate peak flatness Q pk : Q pk =max(|a t |) / [(1 / N)×Σ|a t |]; In the formula, max(|a t |) represents the maximum value of the absolute tangential acceleration sequence within the window, in m / s². 2 ; [(1 / N)×Σ|a t |] represents the arithmetic mean of the absolute values ​​of tangential accelerations within the window, in m / s². 2 .

[0053] The ratio of peak value to mean value reflects the prominence of extreme values ​​in the signal relative to the average level. During the gait swing phase, the forward and backward swing of the thighs generates sharp acceleration pulses with amplitudes much higher than the average level within the window; during the support phase, the acceleration signal is smoother, with the maximum value not significantly different from the average value. Q pk The larger the value, the more significant the impact pulse event within the window. These pulses typically correspond to specific key events in gait (such as heel strike and toe lift). The more important the key time markers in the data, the less likely they are to be compressed or discarded.

[0054] S6.3: Constructing the data information value density ρ ifo : ρ ifo =H w ×log2(1+Q pk )×tanh(M i (t) / M Ref ); In the formula, M Ref =1.0 is the reference intensity of the movement. Its value is based on the typical dividing value of the intensity index between the swing phase and the support phase -- M value > 1 usually corresponds to the swing phase, and < 1 usually corresponds to the support phase.

[0055] ρ ifo The data acquisition value of the current window is quantified by combining three dimensions: angular velocity distribution richness, acceleration pulse intensity, and current motion intensity. Only when all three dimensions are high (dispersed angular velocity distribution, presence of impact pulses, and overall intense motion) is the data truly valuable; if any dimension is insufficient, ρ... ifo It will be suppressed. ρ ifo The larger the value, the richer the gait information contained in the data within the current window, and the higher the collection value. More raw data should be retained and sent at a higher frequency.

[0056] S6.4: Determine the non-uniform temporal compression ratio: When ρ ifo <ρ Low When R = 0.8, cmp =floor(ρ Low / (ρ ifo +0.01); When ρ ifo ≥ρ Low At that time, R cmp =1; In the formula, ρLow =0.8 is the compression baseline threshold; 0.01 is the minimum protection constant, used to prevent ρ ifo =0 when the denominator is zero; floor() means round down, making R equal to zero. cmp It is a positive integer.

[0057] Compression ratio R cmp This determines the actual amount of data sent. When R cmp When =1, there is no compression; R cmp When =2, retain 1 / 2 of the data; R cmp When the value is 3, retain 1 / 3 of the data. This is typically done within the low-value window of the support phase (ρ). ifo ≈0.3), R cmp =2, data volume halved; when standing still (ρ) ifo ≈0.05), R cmp It can reach 16, significantly reducing the amount of data.

[0058] S6.5: Adjust the aggregation sending threshold: L thr =L bs ×[1+ρ ifo / (ρ ifo +ρ Ref )]; In the formula, L bs =64 bytes is the basic aggregation threshold, and the value is determined based on the minimum effective payload of a typical wireless data packet; ρ Ref =1.0 is the reference information density.

[0059] L thr The unit is bytes, and the value range is 64 to 128 bytes. L thr A larger value indicates that more data is accumulated in the buffer before being sent, resulting in a lower packet transmission frequency, lower RF power consumption, but increased latency.

[0060] The adjustment logic for the aggregation transmission threshold is as follows: when the data value is high, the threshold is appropriately increased to package more valid data into a single transmission, reducing the number of packet transmissions and thus lowering RF startup power consumption; when the data value is low, the threshold is decreased to promptly transmit a small amount of valid data, preventing low-value data from occupying too much buffer space. In the formula [1+ρ ifo / (ρ ifo +ρ Ref )] in ρ ifo When it approaches 0, it approaches 1 (L) thr =L bs ), in ρ ifo As it approaches infinity, it approaches 2(L) thr =2×L bsThis ensures that the adjustment range is between 1 and 2 times the base value, avoiding buffer overflow or excessively frequent sending.

[0061] Preferably, in step S4, after the predictive up-frequency trigger, the following operation is also performed: After the predictive frequency increase is triggered, the exercise intensity index M calculated for the left and right knee sides within the current window will be used. L (t) and M R (t) Report to the main control unit; the main control unit calculates the relative ratio λ=M L / M R λ reflects the relative intensity of current movement of both knee joints.

[0062] When λ > 1.8, it indicates that the intensity of the left knee movement is more than 1.8 times that of the right knee, indicating that the left knee is in the high-dynamic phase of the swing phase while the right knee is in the low-dynamic phase of the support phase. At this time, the angular velocity and tangential acceleration data collected by the left knee IMU contain rich gait phase transition information, while the right knee IMU data is mainly composed of static gravity components and has high information redundancy. Therefore, the transmission time slot length and single transmission load of the left knee are configured to be 1.8 times and 1.5 times that of the right knee, respectively. This prioritizes bandwidth resources for the high-dynamic side, ensuring that its critical data is uploaded to the main control with a lower packet loss rate and shorter transmission delay, while preventing the low-dynamic side from frequently occupying the wireless channel during the support phase.

[0063] When λ < 0.55, it indicates that the intensity of the right knee movement is more than 1.8 times that of the left knee. The physical states are reversed compared to the previous situation, with the right knee in the high-dynamic swing phase and the left knee in the low-dynamic support phase. The bandwidth allocation strategy is executed in reverse, configuring the right knee's transmission time slot length and single transmission load to be 1.8 times and 1.5 times that of the left knee, respectively.

[0064] When λ is between 0.55 and 1.8, it is determined that both knee joints are in similar motion states, that is, both sides are simultaneously in the swing phase (running or fast walking) or both sides are simultaneously in the support phase (standing or very slow walking). At this time, the data value of either side does not have a significant advantage, so an equal allocation strategy is adopted, with the same transmission time slot length and load on both sides.

[0065] After allocation, the master controller broadcasts the time slot allocation table and load configuration table for the next cycle. Each node starts transmitting within the specified time window. The time windows do not overlap to avoid wireless collisions, and the amount of data transmitted in a single transmission does not exceed the allowed load to avoid prolonged channel occupation. The thresholds of 0.55 and 1.8 can be adjusted according to the specific use case.

[0066] Preferably, the present invention further includes step S7, calculating the global health degradation factor and performing frequency derating correction. The purpose of this step is to globally drate the power consumption strategy of the entire system when the sensor faces harsh conditions such as low battery or high temperature, in order to ensure minimum functional maintenance. Specifically, this step involves: S7.1: Calculate the energy degradation factor D Q : D Q =(Q Rtd -Q Rem ) / (Q Rtd +Q Rem ) In the formula, Q Rtd =500mAh is the rated capacity of the battery; Q Rem The current remaining battery power, in mAh, is read by the battery power monitoring chip. D Q Used to quantify the degree of degradation of the current remaining power relative to the rated capacity; the closer to 0, the more sufficient the battery power, and the less need for derating due to power supply issues; D Q The closer it is to 1, the more likely the battery is to run out, requiring a significant reduction in power consumption to extend battery life.

[0067] S7.2: Calculate the temperature degradation factor D t : D t =1 / [1+exp(-(T j -T Ref ) / T0)] In the formula, T j This is the internal junction temperature, in °C, read by the temperature sensor integrated inside the IMU. T Ref =25°C is the ambient reference temperature; T0 = ​​15°C is the width of the temperature-sensitive feature.

[0068] D t D is used to quantify the negative impact of the current junction temperature rise on sensor operation. t The closer the value is to 0.5, the closer the junction temperature is to room temperature, and the thermal noise and leakage current effects are at the baseline level; D t The closer a value is to 1, the higher the junction temperature (over 50°C), requiring a reduction in the operating frequency to decrease self-heating and ensure conversion accuracy.

[0069] S7.3: Calculate the global health degradation factor D by considering the degradation effects of both electrical charge and temperature. hth : D hth =D Q +D t -D Q×D t ; D hth The closer to 0, the better the overall health of the sensor, and no performance limitations are required; D hth The closer the value is to 1, the more severe the degradation state the sensor is in, such as when its power is about to run out, the junction temperature is too high, or both, requiring significant derating.

[0070] S7.4: Execution frequency derating correction: f s,final =f s ×(1-D hth )+f s,min ×D hth ; f t,final =f t ×(1-D hth )+f t,min ×D hth ; In the formula, f s Sampling frequency, in Hz; f t The wireless transmission frequency determined in step S4 based on the bandwidth contention weight, in Hz; f s,min =25Hz is the minimum sustain sampling frequency, which is determined based on the Nyquist theorem requiring at least 10 times the sampling frequency of gait (approximately 2Hz). f t,min =5Hz is the minimum sustain transmission frequency, and the value is determined based on the minimum sustain requirement of low-frequency heartbeat packets; f s,final and f t,final The final execution frequency, in Hz.

[0071] f s,final and f t,final All calculations use linear interpolation, and the calculation method is based on the condition of good health (D). hth =0) Fully executes the original frequency, completely degenerates (D) hth =1) Reduce to the lowest sustaining frequency for a smooth transition between intermediate states. Linear interpolation minimizes computation while ensuring a smooth transition, making it suitable for real-time execution on the sensor's local processor.

[0072] S7.5: Determine if emergency mode is triggered: D hth When the value is >0.85, it corresponds to a scenario where the battery level is about 10% and the junction temperature is about 60°C. At this time, the sensor is in a severely degraded state and it is necessary to trigger the emergency limp power mode. 0.60 <D hthWhen the value is ≤0.85, it indicates that the battery is in a sub-healthy state. At this time, the wireless confirmation and retransmission mechanism should be turned off and replaced with one-way broadcast transmission to save receiving power consumption.

[0073] In step S7, after triggering the emergency limp power mode, in order to maintain basic functions with the lowest power consumption under severe degradation conditions, the following steps are executed sequentially: First, f s,final The frequency is forcibly fixed at 25Hz, and the upsampling request from step S4 is no longer responded to. Fixing the frequency at 25Hz means that even during gait transition transients, data is collected at the lowest sampling rate, which significantly reduces the amount of information but can maintain the most basic gait cycle count.

[0074] Second, clock gating is applied to all sampling paths of the gyroscope and power is cut off to completely stop angular velocity data acquisition. Angular velocity data is discarded because single-axis accelerometer data is sufficient to detect the start and stop of walking and gait cycle counting. Although angular velocity data can provide richer motion information, its power consumption is too high.

[0075] Third, disable the periodic wake-up function of the wireless communication module; the RF circuit should only be activated momentarily during data transmission. Periodic wake-up is used to monitor the master controller's commands; disabling this function in emergency mode can significantly reduce RF standby power consumption.

[0076] Fourth, the accelerometer sampling axis is reduced to retain only the normal axis, the resolution is reduced to 8 bits, a hardware threshold comparison interrupt is configured, and the wake-up threshold is 1.5 m / s². 2 An 8-bit resolution corresponds to approximately 0.125g of resolution within a ±16g range, sufficient to distinguish between walking impacts and ambient noise; 1.5m / s 2 The wake-up threshold can effectively filter out minute movements of the human body.

[0077] Fifth, after the processor is woken up, it immediately reads the normal acceleration value and the current timestamp, encapsulates them into a minimum length broadcast frame (without any handshake retransmission field, and the length is about 1 / 3 of the standard data packet), and bursts it with the maximum transmit power for a duration not exceeding 5ms to reduce radio frequency occupancy time.

[0078] Sixth, after transmission is complete, the processor clears the interrupt flag, puts the RF circuit and data processing module back into sleep mode, and reduces the clock frequency to 1MHz (approximately 48~72MHz during normal operation), retaining power only for the interrupt detection circuit and accelerometer. After these operations, the sensor power consumption can be reduced from approximately 45mW during normal operation to below approximately 5mW, providing an additional approximately 2 hours of continuous operation when the battery is about to run out.

[0079] Preferably, the present invention also includes a coordinate system calibration operation after the sensor module is reinstalled. This step is performed once after each reinstallation of the sensor module, earlier than step S1, and specifically includes: The wearer is required to remain stationary for 5 seconds, and acceleration data from 500 sampling points is collected and averaged to calculate the direction of the gravity vector in the IMU sensor's local coordinate system. In a stationary standing posture, all acceleration measured by the IMU comes from gravity; therefore, the direction of gravity is the direction of the acceleration vector in the local coordinate system. An initial rotation matrix is ​​established between the local and navigation coordinate systems, aligning the Z-axis of the local coordinate system with the celestial direction of the navigation coordinate system (opposite to gravity). In the subsequent data processing step S1.2, this rotation matrix is ​​used to rotate all IMU output data from the local coordinate system to the navigation coordinate system, eliminating the influence of installation posture differences (such as sensor misalignment, asymmetrical leg installation angles, etc.) on the algorithm.

[0080] This step addresses the challenge of ensuring perfectly horizontal and symmetrical IMU sensors on exoskeletons during actual installation. In cases of asymmetry, the left knee sensor might be deflected by 5° relative to the knee flexion-extension axis, while the right knee sensor might be deflected by -3°. Without calibration, these installation deviations can be misinterpreted as gait abnormalities, leading to inaccurate prediction coefficient calculations. A simple 5-second static standing calibration eliminates the impact of these installation deviations.

[0081] The following section demonstrates the complete execution of this method on a certain model of exoskeleton dual-knee IMU sensor system, providing a more detailed explanation of each step: Hardware configuration and initial parameters: The exoskeleton has a nine-axis IMU sensor (including a three-axis accelerometer and a three-axis gyroscope) installed at each knee joint, transmitting data to the main controller in the waist via ANT+ wireless connection. Battery rated capacity Q Rtd =500mAh, initially fully charged. Sensor initial sampling frequency 100Hz, initial wireless transmission frequency 10Hz. Gait data acquisition window width T w =0.3s, sampling period T s =0.01s (corresponding to 100Hz), number of sampling points N=30 within the window. Reference angular acceleration α0=10rad / s² 2 The reference angular velocity ω0 = 2.0 rad / s.

[0082] The subject, wearing an exoskeleton, walked on a level surface at a normal speed (approximately 4.5 km / h). At a certain moment t=5.0 s, the IMU in the left knee recorded the raw triaxial acceleration as [-0.2, 9.1, 0.8] m / s². 2 (Carrier coordinate system), after being rotated and projected onto the navigation coordinate system using attitude quaternions, the vertical acceleration is obtained as 9.5 m / s². 2 Horizontal acceleration 0.6 m / s² 2 After a 0.5Hz high-pass filter, the normal dynamic acceleration anL =0.3m / s 2 tangential acceleration a tL =0.5m / s 2 The angular velocity is projected to obtain the angular velocity scalar ω. L =0.2 rad / s. The corresponding data from the right knee IMU within the same window is: normal dynamic acceleration a. nR =0.4m / s 2 tangential acceleration a tR =2.8m / s 2 angular velocity scalar ω R =3.8 rad / s. In the above data, the right knee angular velocity of 3.8 rad / s is in the typical range of the middle of the swing phase, and the left knee angular velocity of 0.2 rad / s is in the typical rest range of the support phase.

[0083] For the left knee, the angular acceleration α at the current moment L (5.0)=[ω L (5.0)-ω L [(4.99)] / 0.01. Based on continuous historical data, ω L At 4.99s, it is 0.19 rad / s, ω L At 5.00 s, it is 0.20 rad / s, therefore α L (5.0)=(0.20-0.19) / 0.01=1.0rad / s 2 .

[0084] Instantaneous motion intensity component: m L (5.0) = 1.0 / 10 + 0.2 / 2.0 + 0.5 / 9.8 = 0.1 + 0.1 + 0.051 = 0.251. The arithmetic mean of the 30 sampling points within the window is used to obtain the left knee M. L (5.0)≈0.25.

[0085] For the right knee, the current angular velocity ω R =3.8 rad / s (in the middle of the oscillation phase), angular acceleration α R (5.0) = (3.8 - 3.5) / 0.01 = 30 rad / s 2 tangential acceleration a tR =2.8m / s 2 Instantaneous motion intensity component: m R (5.0) = 30 / 10 + 3.8 / 2.0 + 2.8 / 9.8 = 3.0 + 1.9 + 0.286 = 5.186. The arithmetic mean of the 30 sampling points within the window is used to obtain the right knee M. R (5.0)≈5.2.

[0086] At this time, the left knee is in the support phase (M)L =0.25), the right knee is in the swing phase (M R =5.2), the intensity of movement of the left and right knees differed by about 20 times, which is consistent with the physical law of bilateral differences in normal gait.

[0087] With the left knee as the opposing knee and the right knee as the contralateral knee, calculate M. L (t) and M R (t) is the normalized cross-correlation function in the range [-0.6s, 0.6s]. Within the t=5.0s window, the peak of the right knee motion intensity waveform occurs approximately 0.45s before the left knee, i.e., δ argmax = -0.45s (a negative value indicates that the right knee moves ahead of the left knee).

[0088] Calculate the gradient of motion intensity change in the contralateral knee (right knee). M is calculated 0.1 s before t=5.0 s. R (4.9) = 5.5, current M R (5.0) = 5.2, therefore: G op (5.0) = (5.2 - 5.5) / 0.1 = -3.0s -1 .

[0089] Three sub-factors: P dec (5.0)=min(1,-(-3.0) / 8.0)=min(1,0.375)=0.375; P phase (5.0)=max(0,1-0.05 / 0.20)=0.75; P active (5.0) = min(1, 5.2 / 0.3) = 1. Final left knee gait transition urgency prediction coefficient: F pre (5.0) = 0.375 × 0.75 × 1 = 0.281.

[0090] F pre (5.0) = 0.281 <F thr =0.75, frequency boost not yet triggered. The contralateral knee only begins to decelerate slowly at t=5.0s (-3.0s). -1 The deceleration intensity was insufficient to trigger the prediction, which is in line with physical expectations.

[0091] Observe subsequent sampling points: at t=5.01s, the right knee continues to decelerate, M R (5.01) = 4.8, G op (5.01) = (4.8 - 5.2) / 0.01 = -40s -1 P dec =min(1,40 / 8)=1,P phase =0.75, P active =min(1,4.8 / 0.3)=1,Fpre (5.01) = 1 × 0.75 × 1 = 0.75; At t=5.02s, M R (5.02) = 4.2, G op (5.02) = (4.2 - 4.8) / 0.01 = -60s -1 F pre (5.02) = 0.75; At t=5.03s, M R (5.03) = 3.5, G op (5.03) = (3.5 - 4.2) / 0.01 = -70s -1 F pre (5.03) = 0.75.

[0092] Three consecutive sampling points (5.01s, 5.02s, 5.03s) F pre The values ​​are all equal to 0.75, reaching the trigger threshold. At t=5.03s, a predictive boost of the left knee IMU sampling frequency is triggered, switching from the current 100Hz to 200Hz. This moment is t. tRig =5.03s. The process of the right knee M value rapidly decreasing from 5.2 to 3.5 corresponds to the physical process of knee joint extension braking at the end of the swing phase.

[0093] Continue monitoring the angular velocity ω of the left knee. L At t=5.18s, ω L The gait transition peak reached 3.9 rad / s, and t was recorded. peak =5.18s. Actual predicted lead time Δ act =5.18-5.03=0.15s, which is exactly equal to the preset target lead time Δ. tgt =0.15s. This lead time means that the sampling frequency has been increased from 100Hz to 200Hz 0.15s before the left knee actually begins to flex and extend rapidly, preparing for capturing complete kinematic data of the left knee swing phase initiation.

[0094] At t=5.50s (the left knee has entered the middle of the support phase—approximately 0.3~0.4s after the left knee touches the ground, at which point the knee joint angular velocity has decreased to near zero), the normal dynamic acceleration a of the left knee is... nL =0.15m / s 2 Gravity projection deviation ε L =0.15 / 9.8≈0.0153. Angular velocity scalar ω L =0.05rad / s.

[0095] Attitude stability judgment value: S stb=exp(-0.05 / 1.5)×(1-0.0153 / 0.3)=0.918. S stb =0.918>S thr =0.85, and this state is maintained for more than 5 sampling points (50ms), indicating that it has entered the quasi-static zero-velocity interval.

[0096] Power consumption reduction was implemented by disabling gyroscope sampling on the yaw and pitch axes, reducing the accelerometer bit size from 16 to 10, retaining only the normal and tangential axes, and enabling hardware wake-up interrupts. During this phase, power consumption decreased from approximately 45mW to approximately 18mW, a saving of approximately 60%.

[0097] Within the t=5.50s window (support phase), the left knee angular velocity is concentrated in the low amplitude range of [-0.1, 0.3] rad / s. Calculate the data distribution dispersion H. w =1.2 bits (3 bits when uniformly distributed across 8 intervals; 1.2 bits indicates that the angular velocity value is mainly concentrated in 1-2 amplitude intervals). Tangential acceleration peak flatness Q pk =0.9 / 0.3=3.0. Current exercise intensity M L =0.15.

[0098] Data information value density: ρ ifo =1.2×log2(1+3.0)×tanh(0.15 / 1.0)=1.2×2.0×0.149=0.358. ρ ifo =0.358<ρ Low =0.8, compression is needed. Compression ratio R cmp =floor(0.8 / (0.358+0.01))=floor(2.17)=2, meaning one sample is retained every two sampling points. Aggregate transmission threshold L thr =64×[1+0.358 / (0.358+1.0)]=64×1.264≈81 bytes, which is a moderate increase compared to the basic threshold of 64 bytes.

[0099] Assuming that after 1 hour of continuous walking, the remaining battery power Q of the left knee IMU is... Rem =180mAh (corresponding to approximately 36% battery capacity), current internal junction temperature T j =52°C (This temperature can be reached inside the IMU during summer outdoor activities or high-intensity walking).

[0100] Electricity degradation factor: D Q =(500-180) / (500+180)≈0.471; Temperature degradation factor: D t=1 / [1+exp(-(52-25) / 15)]=1 / [1+exp(-1.8)]=0.858; Global health degradation factor: D hth =0.471+0.858-0.471×0.858=0.925. D hth =0.925>0.85, triggering emergency limp power mode.

[0101] If the current gait requires a sampling frequency f s =200Hz (oscillating phase), wireless transmission frequency f t =15Hz, then: Final sampling frequency: f s,final =200×(1-0.925)+25×0.925=15+23.125=38.125Hz; Final wireless transmission frequency: f t,final =15×(1-0.925)+5×0.925=1.125+4.625=5.75Hz.

[0102] This means that the sampling frequency needs to be dated from 200Hz to about 38Hz, and the transmission frequency needs to be dated from 15Hz to about 6Hz to match the minimum functional maintenance requirements under harsh health conditions.

[0103] Summary of power management performance: In this embodiment, during a 30-minute test of normal walking, compared to the traditional method with a fixed sampling frequency of 100Hz and a transmission frequency of 10Hz, this method achieves significant power savings: During the quasi-static zero-velocity interval in the mid-phase of the support phase (approximately 40% of the gait cycle time), the gyroscope shutdown and accelerometer resolution reduction reduce the power consumption of the sensor on this side by approximately 60%, resulting in a total power saving of approximately 27mW on both sides. Within the support phase window where data information value density is low, a non-uniform compression ratio of 2:1 reduces the amount of wireless data transmitted by approximately 50%, and correspondingly reduces transmission power consumption by approximately 40%. When the motion intensity is asymmetrical on both sides, differentiated bandwidth allocation reduces the transmission time slot of the less active side by about 44% and the total wireless power consumption on both sides by about 35%.

Claims

1. An adaptive power management optimization method for exoskeleton IMU sensors, characterized by the following steps: include: S1: Real-time acquisition of triaxial acceleration and triaxial angular velocity data of the left and right knees, gravity separation and sagittal projection of the acceleration data of each knee, and extraction of angular velocity scalar, tangential acceleration and normal acceleration of each knee; S2: Calculate the instantaneous motion intensity index of each knee within a sliding time window by combining angular velocity scalar and tangential acceleration; S3: Calculate the temporal cross-correlation function between the intensity of movement of the left and right knees to determine the actual temporal phase difference between the two sides; combine the intensity of movement of the opposite knee and its gradient of change to calculate the gait transition urgency prediction coefficient of the knee on this side. S4: Based on the gait transition urgency prediction coefficient, the sampling frequency of the knee on this side is increased in a predictive manner, and when both sides are in a low dynamic state, the sampling frequency is decreased in a stepwise manner. S5: Calculate the attitude stability judgment value based on the degree of deviation between normal acceleration and gravitational acceleration and the amplitude of angular velocity. Based on this, determine the quasi-static zero velocity interval, and shut down the gyroscope sampling path and reduce the accelerometer resolution within the interval. S6: Construct the data information value density based on the data distribution dispersion of angular velocity and the peak flatness of tangential acceleration, and determine the non-uniform temporal compression ratio of the original data in the local cache and the aggregation transmission threshold of wireless data packets accordingly.

2. The adaptive power consumption management optimization method for exoskeleton IMU sensors according to claim 1, characterized in that, In step S1, the gravity separation and sagittal projection of the acceleration data for each knee are performed as follows: The attitude quaternion is updated using the original three-axis angular velocity data through quaternion differential equations and converted into a rotation matrix in the navigation coordinate system. The original three-axis acceleration data in the carrier coordinate system is rotated and projected to the navigation coordinate system. In the navigation coordinate system, the vertical axial acceleration and the front-to-back axial acceleration are respectively subjected to high-pass filtering with a cutoff frequency of 0.5Hz to remove the DC component of gravity and obtain the normal acceleration and tangential acceleration. After projecting the original three-axis angular velocity data onto the navigation coordinate system, only the angular velocity components around the inner and outer axes are extracted as angular velocity scalars, while the data of the other axes are discarded.

3. The adaptive power consumption management optimization method for exoskeleton IMU sensors according to claim 1, characterized in that, In step S2, the calculation of the instantaneous intensity index is specifically as follows: Calculate the angular acceleration scalar at the current time t: α(t)=[ω(t)-ω(tT s )] / T s ; In the formula, ω(t) is the angular velocity scalar at the current time t; T s The sampling period; Calculate the intensity index M of motion at the current time t. i (t): m i (t)=|a i (t)| / α0+|ω i (t)| / ω0+|a ti (t)| / g; M i (t)=(1 / N)×Σm i ; In the formula, the subscript i takes the form of L or R, representing the left knee or right knee respectively; α0 is the reference angular acceleration; ω0 is the reference angular velocity; a ti (t) represents tangential acceleration; g represents gravitational acceleration; T represents acceleration due to gravity. w Σm is the width of the sliding window; N is the number of sampling points within the window; i For time tT w All m at time t i The sum of (t).

4. The adaptive power consumption management optimization method for exoskeleton IMU sensors according to claim 3, characterized in that, In step S3, the calculation of the actual temporal phase difference and the gait transition urgency prediction coefficient of the lateral knee is specifically as follows: The left knee M built within the window L (t) and right knee M R The normalized cross-correlation function C of (t) LR (δ): ； In the formula, M Lavg and M Ravg These are the arithmetic mean values ​​of the intensity of movement of the left and right knees within the window, respectively. Calculate the actual timing phase difference δ argmax ; δ argmax =argmaxC LR (δ),(δ∈[-δ max ,0]); In the formula, δ max This represents the maximum search range; argmax indicates the value of the independent variable that makes the function within the parentheses reach its maximum value. Using the knee on the affected side as the target knee, the intensity index of the contralateral knee movement is denoted as M. op , calculate its gradient G. op (t): G op (t)=[M op (t)-M op (t-T g )] / T g ; In the formula, the subscript 'op' represents the contralateral knee; when the knee on the same side is the left knee, 'op' represents the right knee, and when the knee on the same side is the right knee, 'op' represents the left knee; T g The time interval for gradient calculation; Calculate the predictive coefficient for gait transition urgency on this side of the knee: F pre (t)=P dec (t)×P phase (t)×P active (t); P dec (t)=min(1,-G op (t) / G max ); P phase (t)=max(0.1-|δ argmax -d nRm | / d tol ); P active (t)=min(1,M op (t) / M thr ); In the formula, G max The maximum reference gradient; δ nRm δ represents the reference timing phase difference of bilateral knee joint movements during normal gait. tol For phase difference tolerance; M thr This is the effective threshold for contralateral exercise intensity.

5. The adaptive power consumption management optimization method for exoskeleton IMU sensors according to claim 4, characterized in that, In step S4, the specific operations of predictive upsampling and stepped downsampling are as follows: F pre (t) and threshold F thr Comparison shows that all three consecutive sampling points are ≥F thr The sampling frequency is switched to 200Hz and maintained until the peak value of ω(t) is reduced to less than half of the peak value. When F pre (t)≤threshold F thr And satisfy M i (t)<0.5 and M op When (t)>1.5, the sampling frequency is reduced by one level every 20 sampling periods, with each reduction being 10% of the current value, down to a minimum of 50Hz; during frequency reduction, if F pre (t) rises above the threshold F thr Then the frequency reduction will stop and the frequency will increase.

6. The adaptive power consumption management optimization method for exoskeleton IMU sensors according to claim 1, characterized in that, Step S5, calculating the attitude stability judgment value and performing power reduction operation in the quasi-static zero-speed range, specifically involves: Calculate the deviation ε of gravity projection i (t): ε i (t)=|a ni (t)| / g; In the formula, a ni (t) represents the normal dynamic acceleration of the i-th knee; g represents the gravitational acceleration; Calculate the attitude stability judgment value S stb (t): S stb (t)=exp(-|ω i (t)| / ω Ref )×(1-e i (t) / e max ) In the formula, ω Ref Used as reference angular velocity; ε max This represents the maximum permissible deviation. When S stb (t) All five consecutive sampling points are greater than the quasi-static decision threshold S thr When the system enters the quasi-static zero-speed range, it executes the following: the sampling clock gating of the gyroscope yaw and pitch axes is stopped; the accelerometer resolution is reduced from 16 bits to 10 bits, and the sampling axes are reduced to the normal and tangential axes; the accelerometer hardware threshold comparison is interrupted, and when the normal or tangential acceleration amplitude exceeds the wake-up threshold, an interrupt is generated to wake up the processor to exit the quasi-static zero-speed range and restore high-resolution sampling of all axes.

7. The adaptive power consumption management optimization method for exoskeleton IMU sensors according to claim 1, characterized in that, In step S6, the non-uniform temporal compression ratio of the original data in the local cache and the aggregation transmission threshold of the wireless data packets are determined as follows: Calculate the data distribution dispersion H w : H w =-Σp m ×log2(p m ),(m=1~K); In the formula, K is the number of intervals dividing the range of angular velocity amplitude; p m This represents the percentage of sampling points falling within the m-th interval; Calculate peak flatness Q pk : Q pk =max(|a t |) / [(1 / N)×Σ|a t |]; In the formula, max(|a t |) represents the maximum value of the absolute value sequence of tangential acceleration within the window; [(1 / N)×Σ|a t |] represents the arithmetic mean of the sequence of absolute values ​​of tangential acceleration within the window; Constructing the data information value density ρ ifo : ρ ifo =H w ×log2(1+Q pk )×tanh(M i (t) / M Ref ); In the formula, M Ref For reference purposes regarding exercise intensity; Determine the non-uniform temporal compression ratio: When ρ ifo <ρ Low = 0.8, R cmp = floor(ρ Low / (ρ ifo + 0.01)); When ρ ifo ≥ρ Low At that time, R cmp =1; In the formula, floor() represents rounding down; Adjust the aggregated sending threshold: L thr =L bs ×[1+ρ ifo / (r ifo +r Ref )]; In the formula, L bs Basic aggregation threshold; ρ Ref For reference information density.

8. The adaptive power consumption management optimization method for exoskeleton IMU sensors according to claim 1, characterized in that, In step S4, after the predictive upsampling trigger, the following operations are also performed: The exercise intensity index M calculated for the left and right knee sides within the current window will be used respectively. L (t) and M R (t) Report to the main control unit; the main control unit calculates the relative ratio λ=M L / M R ; When λ>1.8, the transmission time slot length of the left knee is configured to be 1.8 times that of the right knee, and the allowable load of a single transmission of the left knee is configured to be 1.5 times that of the right knee. When λ < 0.55, the right knee transmission time slot length is configured to be 1.8 times that of the left knee, and the allowable load for a single transmission of the right knee is configured to be 1.5 times that of the left knee. When λ is between 0.55 and 1.8, the length of the transmission slot on both sides and the allowable load per transmission are equal; After allocation is completed, the master controller broadcasts the time slot allocation table and load configuration table for the next cycle. Each node starts sending within the specified time window. The time windows do not overlap, and the amount of data sent in a single session does not exceed the allowed load.

9. The adaptive power consumption management optimization method for exoskeleton IMU sensors according to claim 1, characterized in that, It also includes step S7: calculating the global health degradation factor and performing frequency derating correction, specifically: Calculate the energy degradation factor D Q : D Q =(Q Rtd -Q Rem ) / (Q Rtd +Q Rem ) In the formula, Q Rtd Q represents the battery's rated capacity. Rem This is the current remaining battery level; Calculate the temperature degradation factor D t : D t =1 / [1+exp(-(T j -T Ref ) / T0)] In the formula, T j For internal junction temperature; T Ref T0 represents the ambient reference temperature; T0 represents the width of the temperature-sensitive feature. Calculate the global health degradation factor D hth : D hth =D Q +D t -D Q ×D t ; Execution frequency derating correction: f s,final =f s ×(1-D hth )+f s,min ×D hth ; f t,final =f t ×(1-D hth )+f t,min ×D hth ; In the formula, f s f is the sampling frequency; t f is the transmission frequency; s,min f is the minimum sustain sampling frequency; t,min To maintain the minimum transmission frequency; Determine if emergency mode is triggered: D hth Emergency limp power mode is triggered when the power consumption is >0.85; 0.60 <D hth If the value is ≤0.85, it is marked as a sub-healthy state, and the wireless confirmation and retransmission mechanism is turned off and replaced with one-way broadcast transmission.

10. The adaptive power consumption management optimization method for exoskeleton IMU sensors according to claim 9, characterized in that, In step S7, after triggering the emergency limp power mode, the following steps are executed sequentially: f s,final The frequency is forcibly fixed at 25Hz, and the frequency increase request from step S4 is no longer responded to. Clock gating is applied to all sampling paths of the gyroscope and power supply is cut off to completely stop angular velocity data acquisition; The periodic wake-up function of the wireless communication module is disabled, and the radio frequency circuit is only activated during data transmission. The accelerometer sampling axis is reduced to retain only the sagittal plane normal axis, the resolution is reduced to 8 bits, and a hardware threshold comparison interrupt is configured. After the processor is woken up, it immediately reads the normal acceleration value and the current timestamp, encapsulates them into a minimum length broadcast frame, and bursts them with the maximum transmit power for a duration not exceeding 5ms. After transmission is complete, the processor clears the interrupt flag, puts the RF circuit and data processing module back into sleep mode, reduces the clock frequency to 1MHz, and only keeps the interrupt detection circuit and accelerometer powered.

Images