A multi-actuator cooperative target control method of an intelligent waist cushion

Abstract

The application discloses a kind of multi-actuator collaborative targeting control method and system of intelligent waist pad.The method is first based on anatomical static weight and muscle collaborative coupling matrix to calculate muscle function importance score, introduce historical habit and hardware state factor to generate integrated control priority;Subsequently, based on priority, construct trapezoidal motion envelope, according to the score dynamically set gradual time window, execute intermittent unloading cycle containing recovery phase;Finally, through staggered start strategy, air line bus pressure monitoring and mechanical resonance suppression algorithm, the action of multiple actuators is finely scheduled.The application realizes the precise hierarchical intervention to the core and auxiliary muscle groups of waist and back, while ensuring physiological comfort, solves the surge current and air line pressure loss problem caused by multiple load concurrency, improves the electrical safety and mechanical stability of system operation.

Description

Technical Field

[0001] This invention relates to the field of smart healthy home and ergonomic control technology, specifically a multi-actuator collaborative targeted control method for a smart lumbar support cushion. Background Technology

[0002] In modern, sedentary office or driving environments, the muscles in the lower back are prone to cumulative fatigue and stiffness, leading to chronic musculoskeletal injuries. Existing smart chairs or lumbar support systems are typically equipped with electromyography (EMG) sensors or pressure sensors to monitor the user's posture or muscle tension in real time. These systems generally employ a closed-loop feedback control mode, meaning that when the detected physiological signal values ​​exceed a preset threshold, the control unit activates built-in airbag components, vibration motors, or mechanical actuators to apply physical support or massage to the lower back. By altering the pressure distribution on the contact surface or providing physical stimulation, these systems alleviate muscle tension and help maintain the physiological curvature of the spine.

[0003] However, existing control strategies often overlook the complex collaborative mechanisms and compensatory effects among the muscle groups in the lower back, generally employing generalized triggering logic based on a single signal amplitude. Specifically, existing technologies struggle to distinguish the functional differences between deep core posture muscles and superficial accessory movement muscles in maintaining a seated posture, and fail to consider that the high activity of one muscle group might be compensating for the functional deficiencies of another fatigued muscle group when determining fatigue. This results in the system's inability to accurately identify the true source of fatigue. This crude control approach, lacking anatomical differentiation and synergistic relationship analysis, easily leads to the application of homogeneous intervention intensity to all monitored areas, preventing effective relief of truly fatigued areas and even exacerbating functional imbalances between muscles due to incorrect targeted intervention, thus failing to achieve efficient and precise fatigue repair. Summary of the Invention

[0004] The first aspect of this invention provides a multi-actuator collaborative targeted control method for an intelligent lumbar support, applied to a seating system comprising a microprocessor, multiple monitoring channels, and multiple actuators. This method addresses resource conflicts, mechanical resonance, and human adaptation issues during multi-actuator collaborative operation.

[0005] In this invention, a microprocessor performs a step of quantifying the importance of muscle function. The microprocessor assigns static base weights based on an anatomically defined muscle function lookup table; the erector spinae and multifidus muscles, located in the lumbar region and counteracting spinal flexion moments, are defined as core postural muscles and assigned static base weights within a first numerical range; the latissimus dorsi or lower trapezius muscles, located in the superficial back and responsible for upper limb movement, are defined as accessory postural muscles and assigned static base weights within a second numerical range. The microprocessor selects a fatigue index sequence within a specific time window and calculates the Pearson correlation coefficient by the ratio of the covariance of the fatigue index sequences of two different muscle groups to the product of their standard deviations; if the Pearson correlation coefficient is greater than a preset threshold, a strong synergistic compensatory relationship is determined. The microprocessor combines the static base weights, the current instantaneous fatigue level, and the muscle synergistic coupling matrix to calculate the muscle function importance score for each muscle group. This calculation process includes multiplying the static base weight of the target muscle group by the normalized fatigue index and adding the product of the synergistic gain coefficient and the fatigue transmission term of the surrounding muscle groups; the fatigue transmission term is a weighted sum of the fatigue indices of the surrounding muscle groups and the Pearson correlation coefficients.

[0006] In this invention, the microprocessor performs a comprehensive priority calculation step. The microprocessor calculates a historical trigger frequency factor, which is the ratio of the cumulative duration a specific channel is triggered and remains effective within a historical time window to the total system runtime. The microprocessor calculates a hardware state penalty factor, which is a normalized value of the current operating state parameters of the execution unit relative to a safety limit threshold. The microprocessor performs a weighted summation of the muscle function importance score, the historical trigger frequency factor, and the hardware state penalty factor to obtain the comprehensive control priority; the hardware state penalty factor has a negative weight, used to reduce the action priority of channels approaching their physical limits.

[0007] In this invention, the microprocessor executes timing control logic steps. The microprocessor generates hardware drive signals based on comprehensive control priorities and constructs a trapezoidal motion envelope including a rising edge, a hold segment, and a falling edge. When constructing the trapezoidal motion envelope, the microprocessor dynamically sets a gradual transition time window based on the muscle function importance score: when the muscle function importance score is greater than a preset emergency threshold, a shorter gradual transition time window is set to achieve a rapid response; when the muscle function importance score is less than or equal to the emergency threshold, a longer gradual transition time window is set to achieve a smooth transition. Furthermore, when the duration of the execution unit in the hold state reaches a preset safety threshold, the microprocessor forcibly initiates an unloading cycle. The unloading cycle sequentially executes a decay phase, a recovery phase, and a reloading phase; during the recovery phase, the microprocessor maintains a low-intensity state for a preset recovery time, utilizing the recovery time window to allow blood to flow back to the compressed soft tissue.

[0008] In this invention, the microprocessor performs resource conflict avoidance steps. The microprocessor schedules action requests from multiple execution units. The microprocessor executes an interleaved startup strategy, arranging all execution units to be started in descending order according to the comprehensive control priority and determining the startup sequence index; it calculates the startup delay using the product of the startup sequence index and the minimum phase offset time, and sequentially connects the power supply to each execution unit according to the startup delay.

[0009] When power resources are limited, the microprocessor executes priority-based power allocation logic. The microprocessor calculates the ideal total power demand for all channels in the request state; when the ideal total power demand exceeds the system's physical power limit, the microprocessor calculates the power scaling factor and normalizes the actual output power allocation based on the product weight of the ideal requested power of each channel and the overall control priority.

[0010] For pneumatic or vibratory actuators, resource conflict avoidance steps also include pneumatic bus pressure stabilization control logic or active mechanical resonance suppression logic. The pneumatic bus pressure stabilization control logic includes: allowing the opening of the solenoid valve corresponding to the next lower priority airbag only when the pneumatic bus pressure is greater than the minimum operating pressure threshold; if opening a new valve causes the pneumatic bus pressure to drop below the minimum operating pressure threshold, the newly opened valve is immediately closed and the task is placed in a waiting queue. The active mechanical resonance suppression logic includes: when the absolute value of the frequency difference between two actuators falls into the beat frequency interference range, the microprocessor forcibly adjusts the drive frequency of the actuator with the lower overall control priority, shifting it away from the interference source frequency by a minimum frequency shift.

[0011] A second aspect of this invention provides a multi-actuator collaborative targeting control system for an intelligent lumbar support. This system includes a microprocessor, a non-volatile memory, a multi-channel electromyography sensor, an actuator drive circuit, and a power management module. The non-volatile memory stores a computer program, and when the microprocessor executes the computer program, it implements the multi-actuator collaborative targeting control method for the intelligent lumbar support described in the first aspect.

[0012] This invention provides a multi-actuator collaborative targeted control method for an intelligent lumbar support. It has the following beneficial effects: 1. This invention, by combining anatomically defined static base weights with a muscle synergistic coupling matrix based on Pearson correlation coefficients, accurately quantifies the functional importance scores of different muscle groups during fatigue development. It can sensitively identify muscle areas that experience latent fatigue due to compensatory effects and guide the execution unit to perform differentiated targeted interventions based on the score. This achieves precise graded control of the core and accessory postural muscles of the human lower back, effectively solving the problem that traditional control strategies cannot cope with complex muscle compensation mechanisms, resulting in poor intervention effects.

[0013] 2. This invention establishes an adaptive balance between the urgency of physiological intervention and the perceived comfort of the human body by constructing a trapezoidal motion envelope and dynamically adjusting the gradual time window based on importance scores. At the same time, combined with intermittent unloading loop logic, the attenuation recovery process is forcibly executed after a long period of support reaches a safety threshold. The recovery window allows blood to return, thus providing effective fatigue relief while preventing the risk of muscle startle response caused by sudden changes in step force and local tissue ischemia caused by prolonged compression, ensuring the physiological safety of users during continuous use.

[0014] 3. This invention utilizes a phase-shift-based staggered start-up strategy and a closed-loop control logic for air circuit bus pressure to finely schedule the concurrent actions of multi-channel execution units on a microsecond-level time scale. This effectively disperses the superimposed transient surge current to different coordinate points on the time axis and prevents the sudden drop in air circuit pressure caused by multiple valves opening simultaneously. At the same time, it combines a mechanical resonance active suppression algorithm to eliminate beat frequency interference, achieving electrical stability and mechanical reliability of the system operation under limited power and hardware resources, and extending the service life of core components. Attached Figure Description

[0015] Figure 1 This is a flowchart of the process of the present invention. Detailed Implementation

[0016] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0017] Example: Please see the appendix Figure 1 This invention provides an adaptive adjustment smart lumbar support based on electromyography signal capture. The hardware structure of the smart lumbar support includes a support shell, an ABS plastic base, a filling layer, a sensing layer, a control circuit, and a support module.

[0018] The support shell and the ABS plastic base are fixedly connected by bolts, and the support shell and the ABS plastic base enclose a receiving cavity. The receiving cavity is filled with down plush toy cotton and soft elastic sponge. The soft elastic sponge is located below the down plush toy cotton to provide an elastic support base, and the down plush toy cotton is located above the soft elastic sponge to adapt to the user's waist curve.

[0019] The housing contains a circuit board, a battery, and a support module. The battery is a rechargeable lithium battery pack, electrically connected to the circuit board. The circuit board integrates a main control microprocessor, a Bluetooth communication module, a power management module, and actuator drive circuitry.

[0020] The support module comprises multiple execution units arranged in an array, the positions of which correspond to the projection positions of the erector spinae, quadratus lumborum, multifidus, pelvic floor muscles, diaphragm, and transverse abdominis muscles in the human lumbar anatomy. The specific structural forms of the execution units include a first execution unit and a second execution unit.

[0021] The first execution unit includes a silicone ball and a linear drive mechanism. The silicone ball is fixedly connected to the output end of the linear drive mechanism, which includes a servo motor and a lead screw and nut assembly. The output shaft of the servo motor is connected to the lead screw, and a nut that threads with the lead screw is fitted on the lead screw. The nut restricts rotation and allows movement along the lead screw axis, and the silicone ball is fixed to the nut. The servo motor receives pulse control signals sent by the circuit board, drives the lead screw to rotate, and thus causes the silicone ball to produce a linear displacement perpendicular to the surface of the support shell, achieving rigid physical support for a specific coordinate point.

[0022] The second actuator includes a telescopic bladder, a miniature air pump, and a solenoid valve assembly. The telescopic bladder is a sealed airbag made of thermoplastic polyurethane elastomer (TPU) material. The telescopic bladder is connected to the solenoid valve assembly via an air tube, and the input end of the solenoid valve assembly is connected to the miniature air pump. The miniature air pump and the solenoid valve assembly are electrically connected to the circuit board. By controlling the inflation and deflation process, the volume of the telescopic bladder is changed to achieve flexible planar support for a local area.

[0023] A sensing layer is positioned between the filling layer and the user contact surface. The sensing layer uses cotton-linen fabric as a flexible substrate, on which a high-density flexible piezoresistive array sensor is fixed. The high-density flexible piezoresistive array sensor employs a surface elastic matrix structure, integrating M rows × N columns (e.g., 8×8 or 16×16) of piezoresistive sensing units on a flexible substrate. Each piezoresistive sensing unit comprises an upper electrode layer, a lower electrode layer, and a pressure-sensitive material layer located between the two electrode layers. The resistance of the pressure-sensitive material layer decreases as the applied normal pressure increases.

[0024] The sensing layer also integrates non-contact capacitively coupled electrodes for acquiring electromyographic (EMG) signals. These electrodes are positioned at the gaps in the piezoresistive sensing units or stacked below them. Employing a high input impedance operational amplifier circuit, these electrodes sense changes in the surface charge of human skin, acquiring EMG signals from the lumbar muscles without direct skin contact and through clothing.

[0025] The outermost layer of the smart lumbar support is covered with high-elastic fabric, and the inner side of the high-elastic fabric is bonded with an inner shell of silicone and frosted fleece. A status indicator light strip is set on the outside of the lumbar support, which is connected to the circuit board to display the working status of the smart lumbar support.

[0026] The support module is fixedly installed on the inner side of the ABS plastic base. The module consists of multiple independently controlled actuators arranged in an array. Based on the anatomical distribution of the lumbar muscles, the support module is spatially divided into a core support area and a secondary support area. The core support area is located on both sides of the vertical central axis of the ABS plastic base, within a distance of 4 to 8 centimeters from the vertical central axis. This 4-8 centimeter area covers the L1 to L5 segments of the lumbar vertebrae, corresponding to the erector spinae and multifidus muscles. The secondary support area is located on the outer side and above and below the core support area, corresponding to the quadratus lumborum and transverse abdominis muscles.

[0027] A rigid support assembly is installed in the core support area, serving as the first execution unit. This rigid support assembly includes a silicone ball head, a guide sleeve, a push rod motor, and a position feedback sensor. The silicone ball head is made of solid silicone material with a Shore hardness of 60A to 75A and is hemispherical. The bottom of the silicone ball head is fixedly connected to the end of the push rod motor's output shaft via a threaded structure. The guide sleeve is fixed to the ABS plastic base with screws. The push rod motor's output shaft passes through the guide sleeve, and the inner diameter of the guide sleeve and the outer diameter of the push rod motor's output shaft are clearance-fitted to limit the radial displacement of the push rod motor's output shaft during extension and retraction, allowing the silicone ball head to perform linear reciprocating motion in a direction perpendicular to the surface of the ABS plastic base.

[0028] The linear stepper motor is a miniature linear stepper motor with an integrated lead screw and nut transmission mechanism, which converts the rotational motion of the motor into linear displacement. A position feedback sensor, either a linear potentiometer or an incremental encoder, is installed inside the linear stepper motor housing. This sensor continuously monitors the extension length of the linear stepper motor's output shaft and transmits this data via signal lines to a microprocessor on the circuit board. The microprocessor adjusts the pulse signal of the linear stepper motor based on the extension length data, forming a closed-loop position control. The maximum stroke of the linear stepper motor is set to 50 mm, with a single-step resolution better than 0.1 mm.

[0029] A flexible support assembly is provided in the auxiliary support area, which serves as the second execution unit. The flexible support assembly includes a multi-cavity telescopic bladder, an air distribution unit, and an air pressure monitoring unit. The multi-cavity telescopic bladder is made of two layers of high-strength thermoplastic polyurethane (TPU) film welded together using a high-frequency heat sealing process. The heat sealing line divides the interior of the multi-cavity telescopic bladder into independent air chambers, and each air chamber extends to the back of the ABS plastic base through an independent air guide tube.

[0030] The air distributor includes a miniature air pump and a multi-channel solenoid valve assembly. The outlet of the miniature air pump is connected to the main air inlet of the multi-channel solenoid valve assembly via a main air pipe. The multi-channel solenoid valve assembly consists of multiple normally closed two-position three-way solenoid valves connected in parallel. The outlet of each two-position three-way solenoid valve is connected to the corresponding air chamber guide tube via a quick-connect air pipe connector. The circuit board controls the opening and closing of each two-position three-way solenoid valve to independently inflate or depressurize a specific air chamber. A one-way pressure relief valve is installed on the main air pipe, which automatically opens to release pressure when the air pressure in the pipeline exceeds a safety threshold (e.g., 50 kPa).

[0031] The air pressure monitoring unit includes multiple MEMS air pressure sensors connected in series in each branch air tube to monitor the air pressure value inside each air chamber in real time and feed it back to the microprocessor. The microprocessor compares the real-time air pressure value with the preset target air pressure value and adjusts the speed of the micro air pump or the opening time of the solenoid valve through a PID algorithm to control the expansion degree of the multi-chamber telescopic bladder.

[0032] Rigid and flexible support components are spatially distributed in a mixed manner. Two sets of rigid support components are arranged symmetrically along the vertical central axis of the ABS plastic base, while four to six sets of flexible support components are arranged in a ring around the rigid support components. The rigid support components apply concentrated point pressure to deep muscles using the rigid displacement of the silicone ball head, while the flexible support components apply distributed planar pressure to superficial muscles using the deformation of the multi-chambered expansion bladder.

[0033] A gap is provided between adjacent actuators, and the gap is filled with high-density sound-absorbing cotton. The high-density sound-absorbing cotton is used to isolate the mechanical vibration of adjacent actuators and absorb the noise generated by motor operation and airflow. The overall thickness of the support module in the non-working state is less than the depth of the internal cavity of the support shell, so that the silicone ball head and the multi-cavity telescopic bladder do not protrude from the surface of the lumbar pad when contracted, thus maintaining the flatness of the lumbar pad surface.

[0034] The sensing layer is positioned between the filling layer and the outermost layer of high-elastic fabric. The sensing layer employs a multi-layer flexible composite structure, using cotton linen or polyethylene terephthalate (PET) flexible film as the substrate material. A high-density flexible piezoresistive array sensor and a non-contact capacitively coupled electromyographic electrode array are integrated onto the substrate material.

[0035] The high-density flexible piezoresistive array sensor adopts an M-row × N-column matrix topology, where M and N are both positive integers greater than or equal to 8. The sensor consists of an upper electrode layer, a lower electrode layer, and a piezoresistive material layer sandwiched between the upper and lower electrode layers. The upper electrode layer contains M parallel row lines, and the lower electrode layer contains N parallel column lines perpendicular to the row lines. The overlapping areas of the row and column lines define independent piezoresistive sensing units. The piezoresistive material layer is made of carbon nanotube-doped polymer film or force-sensitive conductive rubber, and its volume resistivity decreases with increasing normal pressure.

[0036] When a user's lower back rests against the surface of the sensing layer, the piezoresistive sensing unit experiences pressure, causing the spacing between the conductive particles in the pressure-sensitive material layer to decrease, forming a conductive path. The circuit board applies voltage excitation to the row lines via a multiplexer and reads current signals from the column lines. By calculating the resistance change of each piezoresistive sensing unit, it reconstructs the pressure distribution matrix of the lumbar contact surface. The physical resolution of the pressure distribution matrix corresponds to the geometric center spacing of the piezoresistive sensing units; for example, a geometric center spacing set to 10 mm to 20 mm is used to determine the specific coordinates of the L1 to L5 segments of the lumbar spine.

[0037] A non-contact capacitively coupled electromyographic (EMG) electrode array is distributed in the gap area of ​​the piezoresistive sensing unit. The number of non-contact EMG electrodes is at least six, corresponding to the anatomical locations of the bilateral erector spinae, bilateral quadratus lumborum, and bilateral multifidus muscles. These electrodes do not directly contact the skin; instead, they acquire EMG signals through capacitive coupling. Each electrode includes a metal inductive plate and a pre-conditioning circuit. The metal inductive plate is made of flexible copper foil or conductive fabric.

[0038] In operation, the metal induction plates and the skin surface of the user's waist form a parallel-plate capacitor model, with the user's clothing acting as the dielectric layer of the capacitor. The alternating current biopotential signal generated by the contraction of human muscles is coupled to the metal induction plates through the clothing medium.

[0039] The pre-amplifier signal conditioning circuit is positioned adjacent to the metal sensing electrode plate. It employs an ultra-high input impedance operational amplifier to construct an impedance transformation buffer. The operational amplifier's input impedance is set to be greater than 10¹²Ω to match the high source impedance characteristics of the skin-clothing-electrode interface and prevent signal attenuation. To suppress 50Hz power frequency interference and environmental electromagnetic noise, an active shielding layer is placed on the side of the metal sensing electrode plate away from the human body. This active shielding layer is connected to the output of the operational amplifier. Through bootstrap circuit technology, the potential of the active shielding layer follows the potential of the metal sensing electrode plate in real time, eliminating the parasitic capacitance effect between the metal sensing electrode plate and the active shielding layer. Furthermore, the sensing layer is connected to a reference ground with the non-measurement area on the back of the human body via a large area of ​​conductive fabric, constructing a feedback loop for the right leg drive (RLD) circuit, further suppressing common-mode interference voltage.

[0040] The signal outputs of the high-density flexible piezoresistive array sensor and the non-contact capacitively coupled electromyographic electrode array converge at a flexible flat ribbon cable interface, which connects to the analog front end of the circuit board. Through this layout, the sensing layer can simultaneously acquire pressure distribution data reflecting posture and electromyographic signals reflecting muscle physiological state without requiring the user to remove clothing. Furthermore, the two sensors are physically aligned in spatial coordinates via the substrate material.

[0041] The adaptive adjustment smart lumbar support system based on electromyography (EMG) signal capture constructs a closed-loop control circuit integrating perception, decision-making, and execution. The smart lumbar support system collects the user's physiological and physical data in real time through multimodal sensors at the hardware layer. This data is then processed and analyzed by the algorithm model built into the microprocessor, ultimately transforming it into control commands to drive the support module, thereby achieving active adjustment of the lumbar muscle state.

[0042] The system's workflow specifically includes four main logical stages: data acquisition, signal processing and feature extraction, multimodal feature fusion and decision-making, and hierarchical collaborative control.

[0043] During the data acquisition phase, the system is in standby monitoring mode. When a user sits on the smart lumbar support, a high-density flexible piezoresistive array sensor detects a change in pressure. If the pressure value exceeds a preset seating threshold (e.g., ...), the system will detect the change. The microprocessor switches to operating mode. At this time, the microprocessor's analog-to-digital converter (ADC) operates at a set sampling frequency (e.g., ...). The system simultaneously acquires the voltage signal output from the piezoresistive sensing unit and the electromyographic analog signal output from the non-contact capacitively coupled electromyographic electrodes. For the acquisition of electromyographic signals, the system uses a differential amplifier circuit to filter out common-mode interference signals.

[0044] In the signal processing and feature extraction stage, the microprocessor performs digital filtering and feature operations on the acquired raw data. For pressure data, the microprocessor constructs a pressure distribution matrix and extracts the pressure peak coordinates, pressure gradient, and contact area features to identify the user's sitting posture and body shape. For electromyographic signals, the microprocessor first passes the signal through a bandpass filter (e.g., ...). Motion artifacts are removed, and then the root mean square (RMS) value is calculated in the time domain to quantify muscle activation intensity. The median frequency (MF) is calculated in the frequency domain using a fast Fourier transform (FFT). When muscles are fatigued, the conduction velocity of muscle fibers decreases, causing the power spectrum of the electromyographic signal to shift towards lower frequencies, which is reflected in a decrease in the median frequency value.

[0045] In the multimodal feature fusion and decision-making stage, the microprocessor executes the fatigue determination algorithm. The system introduces an adaptive weighting mechanism to fuse electromyographic (EMG) features with stress features. Specifically, the microprocessor calculates the signal-to-noise ratio (SNR) of the current EMG signal in real time and dynamically adjusts the EMG feature weight coefficients based on the SNR. Weighting coefficients for pressure characteristics When the signal-to-noise ratio is higher than a preset quality threshold (e.g., ... When the microprocessor increases The value is preferentially determined based on electromyography (EMG) signals to assess fatigue; when the signal-to-noise ratio is below the quality threshold, the microprocessor reduces... Value and increase The fatigue level is primarily determined based on pressure distribution characteristics. The microprocessor compares the fused feature values ​​with a pre-set fatigue threshold database. The fatigue threshold database is stratified based on the user's Body Mass Index (BMI), and the corresponding decision threshold is applied to users with different BMI categories to determine the real-time fatigue level of each lumbar muscle group (erector spinae, multifidus, quadratus lumborum).

[0046] In the hierarchical collaborative control phase, the microprocessor generates control strategies based on the fatigue level and physiological importance of each muscle group. The microprocessor calculates a comprehensive priority score for each execution unit, which characterizes the urgency of the muscles at the corresponding location requiring support. Based on the priority ranking, the microprocessor generates a queue of drive commands containing action type, action amplitude, and action timing. Action types include the ejection of a silicone ball and the inflation of a telescopic bladder; action amplitude corresponds to the number of motor steps or the activation time of the air pump. The drive circuit receives the drive command queue and outputs corresponding pulse width modulation (PWM) signals or high / low level signals to control the servo motor of the first execution unit and the solenoid valve group of the second execution unit, respectively.

[0047] A servo motor drives a silicone ball to provide physical support to the deep muscle area, while a solenoid valve controls a telescopic bladder to provide flexible support to the superficial muscle area. Simultaneously, sensors continuously monitor changes in muscle condition. Once the electromyographic signal characteristics return to the preset normal range or the pressure distribution uniformity reaches the set standard, the microprocessor sends a stop or reset command, completing one adaptive adjustment cycle.

[0048] The overall control method for intelligent lumbar support cushions based on multimodal biosignal feedback is implemented by a microprocessor executing a main control program. The overall control method for intelligent lumbar support cushions includes the following steps executed in sequence: Step S100: System initialization and seating status monitoring.

[0049] After the smart lumbar support cushion is powered on, the microprocessor executes a self-test program to check the electrical connections of the servo motor, miniature air pump, and sensor array. Upon successful self-test, the system enters a low-power standby mode. In this mode, the microprocessor controls a high-density flexible piezoresistive array sensor to scan at a low frequency (e.g., 10Hz). The microprocessor reads the pressure values ​​of each piezoresistive sensing unit in real time and compares these values ​​with a preset seating trigger threshold (e.g., 5kPa). When the pressure value within the effective pressure distribution area exceeds the seating trigger threshold, and this excess state persists for more than a preset anti-accidental touch time window (e.g., 1 second), the microprocessor determines that the user has sat down and immediately switches the system to normal operating mode.

[0050] Step S200: Real-time acquisition and preprocessing of multimodal data.

[0051] After entering normal operating mode, the microprocessor increases the sampling frequency to the operating frequency (e.g., The microprocessor acquires two signals in parallel: the first is the pressure distribution signal from a high-density flexible piezoresistive array sensor, and the second is the raw electromyographic signal from a non-contact capacitively coupled electromyographic electrode array. The microprocessor preprocesses the acquired raw electromyographic signals... Notch filters remove power frequency interference and, through to A bandpass filter is used to remove baseline drift and high-frequency noise, resulting in a preprocessed electromyographic signal. Simultaneously, the microprocessor performs two-dimensional interpolation smoothing on the pressure distribution signal, generating a resolution of [resolution value missing]. The real-time pressure distribution matrix.

[0052] Step S300: Heterogeneous feature extraction and fatigue state fusion determination.

[0053] The microprocessor performs feature engineering on the data acquired in step S200. For the preprocessed electromyographic signals, the microprocessor calculates a preset time window (e.g., ...). The root mean square (RMS) and median frequency (MF) within the pressure distribution matrix are calculated. For the real-time pressure distribution matrix, the microprocessor extracts the pressure peak, pressure gradient, and contact area features. The microprocessor then uses an adaptive weighted algorithm based on signal-to-noise ratio to fuse electromyographic features with pressure features to calculate the muscle fatigue index at the current moment. .

[0054] Simultaneously, the microprocessor estimates the user's body shape characteristics based on the real-time pressure distribution matrix. Specifically, the total pressure value is calculated by integrating the real-time pressure distribution matrix; the effective contact area is calculated based on the number of non-zero pressure units; the average contact pressure is obtained by dividing the total pressure value by the effective contact area; and the microprocessor determines the user's body shape category based on the positive correlation between the average contact pressure and body mass index (BMI). The microprocessor then matches a corresponding personalized fatigue threshold from a pre-set database based on the body shape category. The microprocessor will display the muscle fatigue index. With personalized fatigue threshold Compare. If And the duration exceeds the judgment period (e.g.) If the corresponding muscle group is fatigued, then it is determined that the muscle group is in a state of fatigue.

[0055] Step S400: Dynamic priority calculation and drive instruction queue generation.

[0056] When fatigued muscle groups are detected, the microprocessor initiates decision-making logic. Based on a pre-defined muscle function importance weighting table and real-time calculated fatigue levels, the microprocessor calculates a weighted product method to determine the overall priority score for each unit to be executed. The calculation formula is as follows:

[0057] In the formula, The functional weight coefficients are the corresponding muscle parts of the unit to be executed. These functional weight coefficients are pre-set based on the contribution of muscles to maintaining spinal stability in anatomy (e.g., the erector spinae muscle is set to...). Multifidus muscle is set as ); The current fatigue level coefficient for the corresponding muscle group (range of values) to Microprocessors are scored according to their overall priority. All execution units requiring actions are sorted in descending order to generate a driver instruction queue. The driver instruction queue contains the ID of each execution unit, the target action type, the target parameter value, and the execution time tag.

[0058] Step S500: Multi-actuator timing coordinated targeted control.

[0059] The microprocessor parses the drive instruction queue and sends control signals according to the timeline. The microprocessor defines the execution time of the highest priority instruction in the drive instruction queue as time T0. At time T0, the microprocessor immediately sends a pulse signal to the first execution unit (servo motor) to drive the silicone ball to provide rigid support to the core muscle group. At a delay after time T0 (e.g., T0+),... The microprocessor sends control signals to the second execution unit (solenoid valve assembly and air pump) to drive the telescopic bladder to provide flexible pneumatic support for the auxiliary muscle groups. Through this time-sharing startup strategy, the microprocessor controls the total instantaneous current peak of the power supply within a safe range and prevents mechanical interference caused by simultaneous operation of adjacent execution units.

[0060] Step S600: Closed-loop feedback and state maintenance.

[0061] After the execution unit completes its action, the microprocessor continues to execute steps S200 to S300, monitoring the muscle fatigue index in real time. The changing trend. If detected If the pressure drops below the recovery threshold, the microprocessor generates a hold command to maintain the current support state. If the user leaves (the pressure value is below the seating trigger threshold), the microprocessor controls all servo motors to reset to zero and controls all solenoid valves to open for exhaust. The system then returns to the low-power standby mode in step S100.

[0062] The microprocessor executes a feature extraction program to construct a 12-dimensional feature vector containing physiological electrical signal features and physical pressure features. eigenvectors As input variables for the fatigue assessment model, the feature vector The mathematical expression is The process of constructing feature vectors includes electromyographic signal feature extraction, pressure distribution feature extraction, and vector normalization.

[0063] Step S310: Extraction of time-frequency domain features of electromyographic signals.

[0064] The microprocessor performs sliding time window analysis on the denoised electromyographic signal, extracting the feature components of the first six dimensions. The length of the sliding time window is set to... Sampling points (e.g.) or ).

[0065] The microprocessor calculates the root mean square (RMS) value of the signal within the sliding time window as a characteristic component. Calculate the mean absolute value (MAV) as the characteristic component. .

[0066] and The root mean square (RMS) value represents the effective intensity of muscle contraction and the number of motor units recruited. The calculation formula is as follows:

[0067] Average absolute value The calculation formula is as follows:

[0068] In the formula, For discrete electromyography signals in the first... The amplitude of each sampling point This represents the total number of data points within the sliding time window.

[0069] The microprocessor calculates the waveform length (WL) as a characteristic component. Waveform length Waveform length reflects the complexity of the signal and the cumulative amount of frequency changes. The calculation formula is as follows:

[0070] The microprocessor performs a Fast Fourier Transform (FFT) on the electromyographic signal to obtain the power spectral density function. Based on power spectral density function The microprocessor calculates the median frequency (MF) as a characteristic component. Calculate the average power frequency (MPF) as a characteristic component. Median frequency The following integral relationship is satisfied:

[0071] Average power frequency The calculation formula is:

[0072] In the formula, Represents frequency variables. Indicates frequency as The power spectral density value at that time. The microprocessor calculates the total energy of the power spectrum as a characteristic component. .

[0073] Step S320: Extraction of spatial features of pressure distribution.

[0074] The microprocessor is based on the pressure distribution matrix output by a high-density flexible piezoresistive array sensor. Extract the physical feature components from the 7th to the 11th dimensions.

[0075] The horizontal coordinate of the center of stress (COP) calculated by the microprocessor and vertical coordinates As feature components , As feature components The center of pressure represents the projected location of the user's torso center of gravity. The calculation formula is as follows:

[0076]

[0077] In the formula, Represents the first element in the pressure distribution matrix. Line number The pressure value measured by the piezoresistive sensing unit of the column, and These represent the total number of rows and columns of the sensors, respectively.

[0078] Microprocessor statistical stress value The number of sensing units exceeding a preset zero-point threshold is multiplied by the physical area of ​​a single sensing unit to obtain the effective contact area as a feature component. The microprocessor extracts the maximum pressure value from the pressure distribution matrix. As feature components The microprocessor calculates the coefficient of non-uniformity (variance) of the pressure distribution as a characteristic component. The coefficient of non-uniformity is used to assess the balance of sitting posture:

[0079] In the formula, The set of effective contact areas, The total number of effective contact units, This represents the average pressure value within the effective contact area.

[0080] Step S330: Time dimension features and vector normalization.

[0081] The microprocessor records the duration of the current pose maintenance. Duration As the feature component of the 12th dimension Duration is used to introduce the time integral effect and quantify static fatigue accumulation.

[0082] After constructing the original feature vector Then, the microprocessor uses the min-max normalization method to process the features in each dimension to obtain the standard input feature vector. The normalization formula is:

[0083] In the formula, For the normalized first 3D eigenvalues These are the original calculated values.

[0084] These are the minimum and maximum limits of the dimensional features obtained when the user first uses the system for calibration, or the corresponding statistical limit values ​​in the system's pre-set database.

[0085] The microprocessor executes a dynamic weight adaptive fusion algorithm to convert the 12-dimensional standard input feature vector. Mapped to muscle fatigue index The dynamic weighted adaptive fusion algorithm dynamically adjusts the weight ratio of physiological electrical signal features and physical pressure features in the decision-making process based on environmental noise and signal quality.

[0086] Step S340: Signal-to-noise ratio estimation and signal quality quantization.

[0087] Microprocessors based on power spectral density function Calculate the signal-to-noise ratio of electromyographic signals The microprocessor defines the effective signal bandwidth as... to (For example to The noise frequency band is defined as follows: to (For example to Signal-to-noise ratio The calculation formula is as follows:

[0088] In the formula, This is the bandwidth correction factor. The value is equal to the ratio of the effective signal bandwidth to the noise bandwidth. (Introduction) Based on the assumption that noise is uniformly distributed in the frequency domain (white noise), the narrowband noise power estimation is extended to an equivalent broadband noise power, thereby estimating the full-band signal-to-noise ratio when noise in the same frequency band cannot be completely separated.

[0089] Step S350: Adaptive fusion weight coefficient calculation.

[0090] The microprocessor uses a variant of the Sigmoid function to adjust the signal-to-noise ratio. Global weight coefficients mapped to electromyographic feature groups Global weight coefficients The value ranges from 0 to 1. When the signal-to-noise ratio When the global weight coefficient is higher than the preset threshold, Approaching 1; when the signal-to-noise ratio When the global weight coefficient is below the preset threshold, It approaches 0. The calculation formula is as follows:

[0091] In the formula, This is the sensitivity coefficient, which controls the gradient of the weights as the signal-to-noise ratio changes. A typical value is [value missing]. to .

[0092] The signal-to-noise ratio threshold inflection point, for example, set to That is, when the signal-to-noise ratio is At that time, the weighting of electromyographic features and pressure features is equal.

[0093] The microprocessor simultaneously calculates the global weight coefficients of the pressure feature group. , Satisfy the normalization constraint:

[0094] Step S360: Calculate the weighted fatigue index.

[0095] The microprocessor combines a pre-defined feature contribution vector Calculate the muscle fatigue index Feature contribution vector Each element in Represents the corresponding eigencomponents Correlation coefficient with muscle fatigue level. Feature contribution vector. The solution is obtained in advance through logistic regression analysis or linear discriminant analysis (LDA) on a large amount of sample data, and the weight normalization condition is satisfied to ensure the output. It is within the range of 0 to 1.

[0096] Muscle fatigue index The fusion calculation formula is as follows:

[0097] In the formula, Indexes (1 to 6) of the characteristic components of the electromyographic signal. Indexes for the characteristic components of the pressure distribution (7 to 12). and These are the normalized eigenvalues.

[0098] microprocessors will Compared to personalized fatigue thresholds determined based on the user's Body Mass Index (BMI) The comparison is performed, and a binary fatigue assessment result is output. When the user is wearing heavy clothing, causing a low signal-to-noise ratio in the electromyography (EMG) signal, the fusion calculation formula automatically reduces the summation weight of EMG feature terms, smoothly switching to a focus primarily on pressure distribution features (such as posture maintenance time). and pressure non-uniformity coefficient The fatigue assessment model.

[0099] The microprocessor obtains the muscle fatigue index for a single frame. Then, the fatigue duration determination logic is executed. The fatigue duration determination logic constructs a time sliding window integral model, performs time-domain filtering and energy accumulation calculation on the muscle fatigue index, so as to filter out transient interference signals caused by body posture adjustment and ensure that the triggering of control commands is based on continuous and stable physiological state changes.

[0100] Step S370: Sliding window data buffering and smoothing.

[0101] The microprocessor allocates a first-in-first-out (FIFO) circular buffer in its internal memory to store the most recently accessed data. The muscle fatigue index is calculated from the frame.

[0102] The value is determined by the sampling frequency. and preset smooth time window Decision, that is .

[0103] The microprocessor performs a moving average filter on the data within the circular buffer and calculates the smoothing fatigue index at the current moment. Moving average filtering is used to eliminate high-frequency random noise interference. The calculation formula is as follows:

[0104] In the formula, For the current moment, The sampling interval is... This represents the raw fatigue index at a historical moment.

[0105] Step S380: Calculation of cumulative fatigue load energy.

[0106] The microprocessor employs an inverse time-limit algorithm to assess the relationship between fatigue severity and duration. This algorithm establishes a non-linear inverse relationship between fatigue severity and the duration required to trigger a action, known as the smoothed fatigue index. The greater the magnitude of the exceedance of the threshold, the shorter the duration required to trigger adjustment.

[0107] Microprocessor defines fatigue energy accumulation value Fatigue energy accumulation value This characterizes the integral of fatigue load exceeding the physiological threshold over time. The microprocessor updates this value in each control cycle. The iterative calculation formula is as follows:

[0108] Among them, increment The calculation logic is as follows:

[0109] In the formula, The basic threshold is used to determine the fatigue threshold. .

[0110] To recover the attenuation coefficient, recover the attenuation coefficient Used to set the rate of fatigue load decay. To ensure the system can respond to cumulative fatigue, the decay coefficient is restored. Set to less than the basic judgment threshold The value, for example ,in for to The constant between. When calculated When it is less than 0, the microprocessor will Reset to 0.

[0111] Step S390: Status determination and flag output.

[0112] The microprocessor calculates the current fatigue energy accumulation value. With the preset trigger energy threshold Compare. Trigger energy threshold. The setting rule is: preset a minimum response time. and a full-scale fatigue amplitude ,but This setting ensures that even slight over-fatigue (e.g., amplitude exceeding the limit) can be managed. ), as long as the duration reaches It can also trigger actions.

[0113] like The microprocessor determines that the current muscle group has entered a specific fatigue state and sets the corresponding fatigue state flag. Set to 1 and lock the fatigue status flag. Until Drop to reset threshold The following is the reset threshold. Set to trigger energy threshold A certain proportion (e.g.) This hysteresis comparison mechanism can prevent the system from experiencing frequent start-stop oscillations near the critical state. When the fatigue state flag is... When the value is 1, the microprocessor is allowed to proceed to the subsequent control instruction generation stage.

[0114] The microprocessor executes a body shape recognition and classification subroutine, which establishes a mapping relationship between the user's physical characteristics and system control parameters. The microprocessor analyzes the static pressure distribution matrix acquired by a high-density flexible piezoresistive array sensor, extracts anthropometry-related geometric and physical features, and determines the user's body shape category. The body shape category determines the subsequent fatigue assessment threshold and the spatial mapping relationship of the execution unit's actions.

[0115] Step S410: Anthropometry feature extraction.

[0116] The microprocessor obtains the average pressure distribution matrix after the user is seated and stabilized. The microprocessor is based on the average pressure distribution matrix. Calculate two features: effective seated width and average load intensity .

[0117] Effective sitting width This characterizes the user's pelvic width and hip dimensions. The microprocessor calculates the effective sitting posture width by searching the column index boundaries of non-zero elements in the pressure distribution matrix. The calculation formula is as follows:

[0118] In the formula, For pressure values ​​greater than the noise threshold (e.g.) The rightmost column index of the sensing unit. The leftmost column index for sensor units whose pressure values ​​are greater than the noise threshold. The physical resolution of the sensor array in the column direction (e.g.) ).

[0119] Average load intensity Used to estimate a user's weight and body fat percentage. Average load intensity. Defined as the average pressure within the effective contact area, the calculation formula is as follows:

[0120] In the formula, The pressure value of the matrix unit. The set of effective contact areas, The effective contact area. The effective contact area is determined statistically. The number of elements in the set is multiplied by the physical area of ​​a single sensing unit.

[0121] Step S420: Body size clustering and classification decision.

[0122] The microprocessor has a pre-built body shape classification lookup table generated based on the K-Means clustering algorithm. The body shape classification lookup table divides the two-dimensional feature space... It is divided into three discrete regions, corresponding to three standard body type categories: small, medium, and large.

[0123] The microprocessor calculates the result of step S410 and The input lookup table is used for matching. The decision logic is as follows: like Less than the first width threshold (For example )and Less than the first pressure threshold (For example The microprocessor determines that the user is of small build; like Greater than the second width threshold (For example or Greater than the second pressure threshold (For example The microprocessor determines that the user is of a large physique; If the above two conditions are not met, the microprocessor determines that the user is of medium build.

[0124] Step S430: Adaptive calibration of control parameters.

[0125] The microprocessor calibrates the system's control parameters based on the determined body type.

[0126] First, the microprocessor calibrates the fatigue judgment threshold. Because users of different body types have varying muscle endurance, the microprocessor sets a lower fatigue threshold for smaller body types and a higher fatigue threshold for larger body types.

[0127] Secondly, the microprocessor calibrates the spatial mapping coordinates of the execution unit. Because the physical distances of spinal anatomical landmarks (such as lumbar vertebrae L1 to L5) relative to the seat center differ for users of different body types, and because the lateral distances of the erector spinae muscles relative to the midline of the spine differ, the microprocessor must adjust the trigger position of the execution unit. The microprocessor introduces a coordinate scaling factor. Perform a linear transformation on the target driving coordinates.

[0128] Assume the theoretical target action coordinates calculated by the algorithm are Corrected actual driving coordinates The calculation formula is as follows:

[0129]

[0130] In the formula, This is the horizontal scaling factor. This is the vertical scaling factor. For large bodies, and Set to greater than The value (e.g.) ), to expand the support range; for small bodies, and Set to less than The value (e.g.) ), to reduce the support range.

[0131] and This is a compensation constant based on the offset of the body center. The microprocessor uses the corrected actual drive coordinates. Control the servo motor or select the airbag array to ensure that the physical support points are accurately applied to the user's muscle belly, avoiding accidental contact with bones.

[0132] The microprocessor executes a threshold mapping program, which maps the body type category determined in step S420 and the user's historical interaction data to specific fatigue judgment thresholds. The threshold mapping program uses a combination of lookup table addressing and dynamic compensation algorithms to determine the personalized action threshold for each monitoring channel.

[0133] Step S440: Addressing and matching of the baseline threshold matrix.

[0134] The microprocessor's internal non-volatile memory stores a pre-calibrated reference threshold matrix. Benchmark threshold matrix The dimension is The number 3 corresponds to three body size categories: small, medium, and large. The corresponding number of muscle groups monitored.

[0135] The microprocessor outputs the body type index based on step S420. Muscle location index corresponding to the current monitoring channel From the benchmark threshold matrix Read the benchmark fatigue threshold :

[0136] Benchmark threshold matrix The values ​​in the calculation are calibrated based on the electrical conductivity of human tissue. For larger individuals, a thicker layer of subcutaneous fat acts as a low-pass filter, attenuating the high-frequency components of the electromyography signal and reducing the signal-to-noise ratio, thus affecting the calculated fatigue index. The dynamic range is compressed. Therefore, the microprocessor sets a lower baseline fatigue threshold for users with large physiques (e.g., ), setting a higher baseline fatigue threshold for users with smaller builds (e.g. This is to compensate for the physical attenuation during signal transmission.

[0137] Step S450: Dynamic compensation calculation based on interactive feedback.

[0138] Microprocessors introduce user preference correction items For the benchmark fatigue threshold Dynamic adjustments will be made. User preference modification items. This is calculated based on users' historical intervention behaviors in the system's automatic operation.

[0139] The microprocessor monitors whether the user has performed an intervention within a preset time window (e.g., 30 seconds) after the system automatically triggers a support action, and updates the cumulative correction step size according to the intervention type. .

[0140] The logical correction is as follows: If the microprocessor detects that the user has performed a "manual undo" operation, it indicates that the current fatigue threshold is too low, causing the system to trigger falsely when the user is not fatigued. In this case, the microprocessor will accumulate the correction step size. Increase by one positive unit step (e.g.) ( ), to increase the action threshold.

[0141] If the microprocessor detects that the user has performed a "manual boost" operation, it indicates that the current fatigue threshold is too high, causing the system to become unresponsive when the user is fatigued. In this case, the microprocessor will accumulate the correction step size. Increase by a negative unit step (e.g.) This is to lower the action threshold.

[0142] Final dynamic fatigue threshold The calculation formula is as follows:

[0143]

[0144] In the formula, For the first The correction step size during secondary user intervention is set to a value of [value]. or .

[0145] This represents the total number of interventions.

[0146] Forgetting factor (e.g., value) Forgetting factor By exponentially decreasing the weight of long-standing historical operations on current decisions, a dynamic fatigue threshold is ensured. It mainly follows the user's recent changes in physical fitness and preferences.

[0147] Step S460: Generation of the hysteresis comparison interval.

[0148] Microprocessor based on dynamic fatigue threshold An asymmetric hysteresis comparison interval is constructed to prevent execution units from frequently starting and stopping in critical states.

[0149] Microprocessor calculates action trigger threshold and action reset threshold :

[0150]

[0151] In the formula, The hysteresis coefficient is the coefficient of performance. The typical range of values ​​is to .

[0152] The logic control rules are as follows: When the muscle fatigue index is calculated Rise and exceed the action trigger threshold At that time, the microprocessor generates control instructions for "startup support"; Only when the muscle fatigue index The value drops and falls below the action reset threshold. At that time, the microprocessor generates a "stop support" control instruction; When muscle fatigue index At the action reset threshold With action trigger threshold During this period, the microprocessor maintains the current operating state of the execution unit. The hysteresis comparison mechanism can filter out signal fluctuation interference when the muscle is in a sub-fatigue state.

[0153] When a microprocessor detects signs of fatigue in multiple muscle groups simultaneously, it does not simply apply the same intensity of intervention to all areas. Instead, it executes a muscle function importance quantification program. Based on human biomechanical models and the principle of muscle synergy, this program calculates the priority weight for each monitored muscle group, thereby generating a control strategy with a sequence and intensity gradient.

[0154] Step S510: Allocation of biomechanical static weights.

[0155] The microprocessor internally stores a muscle function lookup table based on anatomical definitions. This lookup table assigns a static base weight to the muscle group corresponding to each monitoring channel. Static base weights The numerical value reflects the contribution of the corresponding muscle in maintaining sitting balance.

[0156] For example, the microprocessor defines the erector spinae and multifidus muscles located in the lumbar region as core postural muscles and assigns them a high static baseline weight (e.g., This is because the erector spinae and multifidus muscles are directly responsible for counteracting the flexion torque of the spine. The microprocessor defines the latissimus dorsi or trapezius bundles located in the superficial layers of the back as accessory postural muscles, assigning them lower static base weights (e.g., ...). By introducing static base weights. When resources are limited (such as limited airbag inflation power), the system prioritizes the functional recovery of core posture muscles.

[0157] Step S520: Analysis of muscle synergy and compensatory effects.

[0158] Microprocessors calculate the correlation of fatigue development among different muscle groups to quantify the synergistic or compensatory behaviors between muscles. When a dominant muscle becomes fatigued, accessory muscles often compensate by increasing their contraction intensity, resulting in a highly positive correlation between the changes in fatigue indicators of the two.

[0159] The microprocessor selects the current moment Previous specific time window (e.g., 60 seconds), obtain the first Fatigue index sequence of individual muscle groups and the Fatigue index sequence of individual muscle groups The microprocessor calculates the Pearson correlation coefficient between the two. :

[0160] In the formula, Let be the covariance of the two fatigue index sequences. and These are the standard deviations of the two sequences. The microprocessor uses the correlation coefficient. Constructing a muscle co-coupling matrix .like If the correlation exceeds a preset threshold (e.g., 0.7), the microprocessor determines the first... Muscle groups and the first Muscle groups exhibit strong synergistic compensatory relationships. This correlation analysis supports the technical feature of "topological correlation analysis based on multidimensional physiological signals" in the claims.

[0161] Step S530: Dynamic Importance Score Calculation. The microprocessor combines the static base weights, the current instantaneous fatigue level, and the synergistic coupling effect to calculate the first... Overall importance score of individual muscle groups The calculation formula is as follows:

[0162] In the formula, For the first Static baseline weights of muscle groups For the first The current normalized fatigue index of the muscle group. This is a synergistic gain coefficient (e.g., a value of 0.3), used to adjust the weighting of the influence of peripheral muscle fatigue on the importance of the current muscle.

[0163] Summation term in the formula It characterizes the "fatigue transmission effect": even the first The muscle group itself has a low fatigue index, but if it is combined with the first muscle group... Peripheral muscle groups with strong synergistic relationships (the first) The muscle groups are in a state of high fatigue. Comprehensive importance score of muscle groups It will also be significantly increased. This mechanism ensures that the system can identify "risk nodes" that, although not yet severely fatigued, are under tremendous compensatory pressure, thereby enabling preventative intervention.

[0164] Step S540: Execute sequence generation. The microprocessor performs a comprehensive importance score for all monitoring channels. Sort the data in descending order to generate a priority control list. Select the microprocessor based on its overall importance score. highest Muscle groups (e.g.) (as a target for intervention. For those who have not yet entered the pre-intervention phase) Even if the fatigue index of a muscle group exceeds the baseline threshold, the microprocessor will temporarily suspend the corresponding execution action or only output low-intensity maintenance support to concentrate the system's energy and mechanical computing power to solve the most urgent risk of posture instability.

[0165] Microprocessors in obtaining muscle function importance scores Based on this, a comprehensive priority calculation program is executed. This program incorporates the user's historical habit factors and hardware constraint penalty factors to correct the purely physiological dimension scores, outputting a comprehensive control priority for driving the execution unit. .

[0166] Step S550: Weighted fusion of multidimensional features.

[0167] The microprocessor calculates the value for each monitoring channel based on a preset decision fusion formula. Comprehensive control priority The decision fusion formula is as follows:

[0168] In the formula, The muscle function importance score calculated in step S530.

[0169] , , These are the weighting coefficients, and the weighting coefficients satisfy the normalization condition. Typical parameter configuration is as follows: The above configuration ensures that real-time physiological signals play a dominant role in control decisions, while taking into account personalized habits and device safety.

[0170] This refers to the historical trigger frequency factor. The microprocessor calculates the historical trigger frequency factor. This quantifies the degree to which users rely on support actions at specific locations. The microprocessor calculates statistics within a preset historical time window (e.g., the past 168 hours). The channel is triggered and maintains a valid cumulative duration. Total system runtime The ratio, i.e. Higher The numerical value indicates that the current area is a common fatigue point for the user, and the microprocessor uses this value to improve the response sensitivity of the corresponding channel.

[0171] This is a hardware state penalty factor. The microprocessor acquires the data in real time. The operating status parameters of the channel execution unit specifically include the air pressure value inside the airbag. Or motor coil temperature value The microprocessor normalizes the operating state parameters relative to the safety limit threshold to obtain... When the operating status parameters approach the safety limit threshold, Approaching This significantly reduces the priority of comprehensive control. This is to prevent the equipment from overheating or over-inflating.

[0172] Step S560: Resource conflict resolution and power allocation.

[0173] The microprocessor prioritizes the overall control. Resolve resource conflicts when multiple channels send concurrent requests.

[0174] The microprocessor calculates the ideal total power demand for all channels in a requesting state. That is, the ideal requested power for all channels. The sum of .

[0175] When the ideal total power demand Exceeding the system's physical power limit When the pump reaches its maximum flow rate or the power supply reaches its maximum current (e.g., at maximum flow rate or maximum current), the microprocessor executes a weighted attenuation algorithm. This algorithm allocates the actual output power based on the "demand-priority" product weight of each channel. The calculation formula is as follows:

[0176] In the formula, This represents the total number of channels currently requesting the action. The weighted attenuation algorithm ensures high overall control priority. The channel can obtain a larger proportion of power resources, thereby prioritizing the intervention effect in the core fatigue area, while reducing the action of secondary areas.

[0177] Step S570: Perform signal mapping and output.

[0178] The microprocessor will allocate the actual output power. It is mapped to specific hardware driver signals.

[0179] For pneumatic actuators, the microprocessor will output the actual power. It can be converted into a pulse width modulation (PWM) signal duty cycle to control the on / off state of the solenoid valve, or into a voltage command to adjust the speed of the air pump.

[0180] For electromechanical actuators, the microprocessor will output the actual power. Convert to the target feed speed of the stepper motor or the torque limit value of the servo motor.

[0181] The microprocessor sends the generated hardware drive signals to the drive circuit, which drives the actuator to apply physical support.

[0182] The microprocessor executes a timing control logic program to shape and schedule the hardware drive signals generated in step S570 in the time dimension. The timing control logic program generates a trapezoidal motion envelope containing a rising edge, a hold segment, and a falling edge to avoid muscle startle reactions caused by step signals.

[0183] Step S610: Generation of the trapezoidal motion envelope.

[0184] The microprocessor builds connections to the current state value With the target state value The transition function. The microprocessor uses an interpolation algorithm to determine the current state value. With the target state value A series of transition setpoints are generated between them. .

[0185] The microprocessor uses a linear ramp function to calculate the first... Reference setpoint for each control cycle The calculation formula is as follows:

[0186]

[0187] In the formula, This is the setpoint for the previous control cycle.

[0188] This is a sign function used to determine the direction of numerical change.

[0189] This is a single-step increment.

[0190] To control the cycle frequency.

[0191] The time window is gradually adjusted. The microprocessor calculates the overall importance score based on step S530. Set the fade time window The microprocessor employs an inverse proportional mapping strategy: when the overall importance score is... Greater than the preset emergency threshold (e.g.) When ), the microprocessor will Set as a fast response value (e.g.) (seconds); when the overall importance score When the threshold is less than or equal to the emergency threshold, the microprocessor will... Set to a slow transition value (e.g.) (seconds). The inverse proportional mapping strategy establishes a dynamic balance between the urgency of physiological intervention and the comfort perceived by the human body, ensuring that the system can intervene quickly in high-risk situations and act gently in low-risk situations.

[0192] Step S620: Closed-loop feedback correction.

[0193] The microprocessor incorporates a PID closed-loop feedback algorithm to eliminate the nonlinear response error of the actuator. The microprocessor acquires feedback values ​​from the sensors. Calculate the feedback value With reference set point Deviation between .

[0194] The microprocessor calculates the corrected control quantity. The calculation formula is as follows:

[0195] In the formula, This is the proportionality coefficient. The integral coefficient is... These are the differential coefficients. To address the inflation delay characteristic of pneumatic actuators, the microprocessor has pre-defined integral separation logic: when the deviation... When the absolute value of the integral term exceeds the preset integral separation threshold, the microprocessor forces the integral term coefficient to be separated. Set to zero, and adjust using only the proportional and derivative terms. The integral separation logic prevents integral oversaturation and system overshoot oscillation caused by airway transmission delays, ensuring that the airbag pressure smoothly approaches the target value.

[0196] Step S630: Intermittently unload the loop logic.

[0197] The duration of the execution unit in the hold state Reaching the preset safety threshold At a time (e.g., 10 minutes), the microprocessor forces the start of the unload cycle.

[0198] The unloading cycle includes the following timing phases: 1. Attenuation phase: The output intensity of the microprocessor-controlled execution unit decreases by a ramp to 50% of the target intensity or returns to zero; II. Recovery Phase: The microprocessor maintains a low-intensity state for a period of time. (For example, 60 seconds), the microprocessor uses the recovery phase time window to allow blood to flow back to the compressed soft tissue, restoring the oxygen supply to the local tissue; III. Reloading Phase: The microprocessor re-enlarges the output intensity to the target state value. The intermittent unloading loop logic prevents the execution unit from continuously compressing the same muscle area for a long time, thus avoiding obstruction of local blood circulation, and thereby meeting the metabolic needs of soft tissue while providing continuous fatigue support.

[0199] The microprocessor executes a resource conflict avoidance program to schedule action requests from multiple execution units. This program addresses issues such as transient current surges, sudden drops in gas pressure, and mechanical resonance caused by the simultaneous startup of multiple loads.

[0200] Step S640: Interleaved start control based on phase offset.

[0201] When the microprocessor receives start commands from multiple execution units (such as multiple air pumps or vibration motors) within the same control cycle, the microprocessor executes an interleaved start strategy.

[0202] The microprocessor calculates the integrated control priority based on step S550. Sort all execution units to be started in descending order and determine the startup sequence index for each execution unit. ( Indicates the highest priority.

[0203] The microprocessor allocates a startup delay to each execution unit to be started. The calculation formula is as follows:

[0204] In the formula, This is the minimum phase offset time.

[0205] The value is set according to the transient response recovery time of the power system, and is usually set to 20 milliseconds to 50 milliseconds.

[0206] Microprocessors utilize startup delay The next load is powered on only after the startup current of the previous load has decayed to the steady-state current. The staggered startup strategy distributes the originally superimposed transient peak current to different coordinate points on the time axis, preventing power supply voltage drops from triggering microprocessor resets.

[0207] Step S650: Pressure stabilization control of the air circuit bus.

[0208] The microprocessor monitors the air bus pressure in real time via a pressure sensor. The microprocessor internally stores the minimum operating pressure threshold required to ensure the normal operation of pneumatic components. .

[0209] When performing a multi-channel inflation task, the microprocessor executes the following conditional gating logic: The microprocessor is enabled first, corresponding to the highest overall control priority. The airbag solenoid valve; The microprocessor continuously compares the pressure on the gas bus. With minimum working pressure threshold ; Only when the air circuit bus pressure Greater than the minimum working pressure threshold Only when the microprocessor allows the activation of the solenoid valve corresponding to the next lower priority airbag; If opening a new valve causes pressure on the main gas line... Falling to the minimum working pressure threshold The microprocessor immediately closes the newly opened valve and places the corresponding inflation task in the waiting queue. In subsequent clock cycles, the microprocessor polls the air bus pressure. The system will continue until the pressure is restored before attempting to restart the tasks in the waiting queue. The conditional gating logic prevents a failure state where all airbags fail to inflate due to a diversion effect.

[0210] Step S660: Active suppression of mechanical resonance.

[0211] When the microprocessor drives multiple vibration motors or pulsating airbags simultaneously, the microprocessor executes a frequency domain interference suppression algorithm.

[0212] The microprocessor obtains the drive frequencies of any two execution units operating simultaneously. and The microprocessor calculates the absolute value of the frequency difference. .

[0213] The microprocessor determines the absolute value of the frequency difference. Does it fall within the preset beat frequency interference range? (e.g., 0Hz to 3Hz).

[0214] If the absolute value of the frequency difference When the signal falls within the beat frequency interference range, the microprocessor forcibly adjusts the drive frequency of the execution unit with the lower priority in the overall control. and offset by a safe frequency band. Adjusted frequency The calculation is as follows:

[0215] In the formula, For the minimum frequency shift (e.g., 5Hz). These are symbolic functions. The microprocessor uses symbolic functions... Determine the frequency shift direction to ensure the adjusted frequency. Stay away from sources of interference The active frequency shift strategy eliminates mechanical beat frequency oscillations caused by close frequency proximity, reducing the risk of fatigue damage to mechanical structures.

[0216] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A multi-actuator collaborative targeted control method for an intelligent lumbar support, the method being applied to a seating system comprising a microprocessor, multiple monitoring channels, and multiple actuators, characterized in that, The method includes the following steps: The steps for quantifying the importance of muscle function are as follows: The microprocessor assigns static basic weights based on an anatomically defined muscle function lookup table, calculates the fatigue development correlation between monitoring channels to construct a muscle synergistic coupling matrix, and combines the static basic weights, the current instantaneous fatigue level, and the muscle synergistic coupling matrix to calculate the muscle function importance score for each muscle group. Comprehensive priority calculation steps: The microprocessor incorporates the user's historical habit factor and hardware constraint penalty factor to correct the muscle function importance score and calculate the comprehensive control priority for driving the execution unit; Timing control logic execution steps: The microprocessor generates hardware drive signals based on the comprehensive control priority, and constructs a trapezoidal action envelope containing rising edge, hold segment and falling edge, and performs timing shaping on the hardware drive signals; Resource conflict avoidance steps: The microprocessor schedules the action requests of multiple execution units, executes an interleaved start strategy to disperse transient current, and adjusts the actual output power or operation frequency of the execution units according to physical resource constraints.

2. The multi-actuator collaborative targeted control method for an intelligent lumbar support according to claim 1, characterized in that, In the step of quantifying the importance of muscle function, the allocation of static base weights specifically includes: The microprocessor defines the erector spinae and multifidus muscles, which are located in the lumbar region and counteract the flexion moment of the spine, as core postural muscles and assigns them static base weights ranging from 0.8 to 1.

0. The microprocessor defines the latissimus dorsi or lower trapezius muscles, which are located in the superficial back and are responsible for upper limb movement, as auxiliary postural muscles and assigns them a static base weight of 0.4 to 0.

6.

3. The multi-actuator collaborative targeted control method for an intelligent lumbar support according to claim 1, characterized in that, In the step of quantifying the importance of muscle function, calculating the muscle function importance score for each muscle group specifically includes: The microprocessor selects a fatigue index sequence within a specific time window, calculates the ratio of the covariance of two different muscle group fatigue index sequences to the product of their standard deviations, and obtains the Pearson correlation coefficient. If the Pearson correlation coefficient is greater than a preset threshold, it is determined that there is a strong cooperative compensatory relationship; The microprocessor multiplies the static baseline weights of the target muscle group with the normalized fatigue index, and adds the product of the synergistic gain coefficient and the fatigue transmission term of the surrounding muscle groups to obtain the muscle function importance score; wherein, the fatigue transmission term is the weighted sum of the fatigue index of the surrounding muscle groups and the Pearson correlation coefficient.

4. The multi-actuator collaborative targeted control method for an intelligent lumbar support according to claim 1, characterized in that, In the comprehensive priority calculation step, the calculation of the comprehensive control priority specifically adopts the following decision fusion strategy: The microprocessor calculates the historical trigger frequency factor, which is the ratio of the cumulative duration during which a specific channel is triggered and remains effective within a historical time window to the total system runtime. The microprocessor calculates a hardware state penalty factor, which is a normalized value of the current running state parameter of the execution unit relative to a safety limit threshold. The microprocessor performs a weighted summation of the muscle function importance score, the historical trigger frequency factor, and the hardware state penalty factor to obtain the comprehensive control priority; wherein the hardware state penalty factor corresponds to a negative weight.

5. The multi-actuator collaborative targeted control method for an intelligent lumbar support according to claim 1, characterized in that, The resource conflict avoidance steps also include priority-based power allocation logic: The microprocessor calculates the ideal total power demand for all channels in a requesting state; When the ideal total power demand exceeds the upper limit of the system's physical power, the microprocessor calculates the power scaling factor; The microprocessor uses the power scaling factor to attenuate the ideal requested power for each channel and introduces a logarithmic correction based on the comprehensive control priority to preserve the power allocation ratio of high-priority channels.

6. The multi-actuator collaborative targeted control method for an intelligent lumbar support according to claim 1, characterized in that, In the timing control logic execution step, the construction of the trapezoidal action envelope specifically includes: The microprocessor uses an interpolation algorithm to generate a transition setpoint between the current state value and the target state value; The microprocessor dynamically sets a transition time window based on the muscle function importance score: when the muscle function importance score is greater than a preset emergency threshold, a shorter transition time window is set to achieve a rapid response; when the muscle function importance score is less than or equal to the emergency threshold, a longer transition time window is set to achieve a smooth transition.

7. The multi-actuator collaborative targeted control method for an intelligent lumbar support according to claim 1, characterized in that, The timing control logic execution steps also include intermittent unloading loop logic: When the duration of the execution unit in the hold state reaches a preset safety threshold, the microprocessor forcibly starts the unload cycle; The unloading cycle sequentially executes a decay phase, a recovery phase, and a reloading phase. During the recovery phase, the microprocessor maintains a low-intensity state for a preset recovery time, utilizing the time window of the recovery time to allow blood to flow back to the compressed soft tissue.

8. The multi-actuator collaborative targeted control method for an intelligent lumbar support according to claim 1, characterized in that, In the resource conflict avoidance step, the execution of the staggered startup strategy specifically includes: The microprocessor sorts all execution units to be started in descending order according to the comprehensive control priority to determine the startup sequence index; The microprocessor calculates the startup delay using the product of the startup sequence index and the minimum phase offset time; The microprocessor sequentially turns on the power supply of each execution unit according to the startup delay, ensuring that the startup current of the previous load decays to the steady-state current before the next load starts to be powered on.

9. The multi-actuator collaborative targeted control method for an intelligent lumbar support according to claim 1, characterized in that, The resource conflict avoidance steps also include air circuit bus pressure stabilization control logic or mechanical resonance active suppression logic: The specific control logic for stabilizing the air circuit bus pressure is as follows: the solenoid valve corresponding to the next lower priority airbag is allowed to be opened only when the air circuit bus pressure is greater than the minimum working pressure threshold; if opening a new valve causes the air circuit bus pressure to drop below the minimum working pressure threshold, the newly opened valve is immediately closed and the task is placed in the waiting queue. The active suppression logic for mechanical resonance is as follows: when the absolute value of the frequency difference between two execution units falls into the beat frequency interference range, the microprocessor forcibly adjusts the driving frequency of the execution unit with the lower priority of the integrated control, so that it shifts away from the frequency of the interference source by a minimum frequency shift.

10. A multi-actuator collaborative targeting control system for an intelligent lumbar support, characterized in that, The system includes a microprocessor, a non-volatile memory, a multi-channel electromyography sensor, an execution unit drive circuit, and a power management module; the non-volatile memory stores a computer program, and when the microprocessor executes the computer program, it implements a multi-actuator collaborative targeting control method for an intelligent lumbar support as described in any one of claims 1 to 9.

Images