A wearable pneumatic knee brace motion pattern recognition and assisted control method based on air pressure time sequence characteristics

Abstract

This invention discloses a control method for motion pattern recognition and assistance in wearable pneumatic knee braces based on air pressure temporal characteristics, relating to the fields of intelligent wearable assistive devices and motion state recognition. The invention addresses the problem that existing knee braces cannot recognize knee joint motion patterns in real time, achieve precise pressure adjustment, and provide safety protection. This invention uses an air pressure sensor to collect real-time signals of pressure changes inside the air bladder during knee joint movement, inputting the raw air pressure temporal data into the main control unit; processing the air pressure signals; segmenting the continuous air pressure data using a sliding window to construct fixed-length temporal input samples; inputting the preprocessed air pressure sequence into a motion recognition model composed of a convolutional neural network and a temporal feature extraction network, and classifying and recognizing the motion patterns; adjusting the inflation and deflation of the knee brace air bladder based on the recognition results; and also setting up an independent safety monitoring mechanism for pressure relief protection. This invention is used to optimize the motion patterns of wearable pneumatic knee braces.

Description

Technical Field

[0001] This invention relates to the field of smart wearable devices and sports assistance technology, and particularly to a method for identifying and assisting in the movement pattern of a wearable pneumatic knee brace based on air pressure time-series characteristics. This method achieves accurate identification of movement patterns by real-time acquisition, processing, and deep learning inference of the pressure change sequence inside the wearer's knee joint airbag. Furthermore, it dynamically adjusts the airbag pressure through an auxiliary control module, thereby providing intelligent assistance and protection for knee joint movement. Background Technology

[0002] Existing knee support devices mainly include open-type knee braces, sleeve-type knee braces, and patellar straps. The open design of open-type knee braces results in slightly lower stability, often leading to loosening or displacement during exercise, thus reducing their protective effect. Sleeve-type knee braces, while offering some flexible support, rely heavily on fixed pressure or manual adjustment, failing to achieve dynamic adaptation to the wearer's actual movement. Patellar straps offer limited protection for other knee joint structures, providing less overall knee joint stability support than other types of knee braces, and have lower versatility. Existing knee braces suffer from the following technical problems:

[0003] Insufficient Motion Pattern Recognition: Existing technologies lack the ability to recognize the wearer's motion state in real time, and cannot provide targeted assistance based on different motion patterns (such as walking, climbing stairs, running, etc.). Some existing technologies attempt to recognize motion states using inertial measurement units or accelerometers. For example, Chinese patent CN119167209A discloses a motion state recognition method for a smart knee brace, which collects knee motion data by setting a three-axis accelerometer on the knee brace and uses a random forest model for state classification. However, this method relies on the accelerometer as a single sensor and uses manual statistical features (such as maximum, minimum, average, and standard deviation) combined with traditional machine learning algorithms. Its generalization ability for complex motion scenarios is limited, and the recognition accuracy is insufficient to meet the needs of practical applications. Furthermore, this solution only addresses motion state recognition and does not integrate with the knee brace's auxiliary control functions.

[0004] Inaccurate air pressure control: Control methods are often switch-on or manual adjustment, making it difficult to precisely match the airbag pressure to the wearer's exercise needs, resulting in unstable assistive effects. Most existing pneumatic knee braces use fixed air pressure output or manual adjustment, lacking the function of dynamic closed-loop adjustment according to changes in exercise mode, making it difficult to achieve real-time matching of assistive force and exercise state. Even if some products have air pressure adjustment functions, they are mostly simple inflation and deflation controls, lacking the ability to finely adjust air pressure, which can easily lead to excessive or insufficient assistive force, affecting wearing comfort and assistive effect.

[0005] Insufficient safety protection mechanisms: The lack of real-time monitoring and protection mechanisms for airbag overpressure, leakage, and sensor malfunctions poses safety risks. Existing solutions lack systematic and independent monitoring and emergency response measures for abnormal situations such as excessive air pressure, airway leakage, and sensor failure, making it difficult to ensure wearer safety in a timely manner during equipment malfunctions. In the event of a control system malfunction or air pump failure, the airbag pressure may rise sharply, causing compression damage to the wearer's knee joint, posing a significant safety hazard.

[0006] Therefore, there is an urgent need for an intelligent pneumatic knee brace technology solution that can identify knee joint movement patterns in real time, achieve precise pressure adjustment, and provide safety protection. Summary of the Invention

[0007] The purpose of this invention is to address the problems of existing knee braces in being unable to recognize knee joint movement patterns in real time, achieve precise pressure adjustment, and provide safety protection. This invention collects the pressure change sequence of the knee joint airbag using a pneumatic pressure sensor, performs noise reduction and standardization using a data preprocessing module, and then uses a deep learning model for movement pattern recognition. An auxiliary control module generates target pressure control commands based on the recognition results, and an air pump and valve control module executes precise pressure adjustment operations, enabling the knee brace to intelligently assist different movement patterns. Simultaneously, a safety module monitors the system operation in real time to ensure safe use of the device. Therefore, this invention provides a control method for wearable pneumatic knee braces based on pneumatic pressure timing characteristics for movement pattern recognition and assistance.

[0008] The technical solution of this invention is:

[0009] This invention provides a control method for motion pattern recognition and assistance of wearable pneumatic knee braces based on air pressure timing characteristics, comprising the following steps:

[0010] Step 1: Collect the original air pressure sequence in the airbag during the wearer's exercise using an air pressure sensor;

[0011] Step 2: Standardize the data using the data preprocessing module to obtain the air pressure sequence;

[0012] Step 21: The median filtering algorithm is used to remove impulse noise and abnormal abrupt changes generated during the sampling process from the original air pressure sequence through the data preprocessing module;

[0013] Step 22: Input the median-filtered air pressure sequence into the low-pass filter module to suppress high-frequency disturbances and preserve the air pressure change trend during human movement;

[0014] Steps 2 and 3: Perform standardization processing on the low-pass filtered air pressure sequence to map the data in different sampling periods to a uniform numerical range, so as to serve as the air pressure time series input data for the subsequent motion pattern recognition model;

[0015] Step 3: The barometric pressure sequence is fed into a deep learning model for inference to obtain motion pattern recognition results;

[0016] The preprocessed barometric time series is input into a deep learning motion pattern recognition model. The deep learning motion pattern recognition model performs feature extraction and classification inference on the input sequence and outputs the motion pattern recognition result corresponding to the current wearer. The specific steps include:

[0017] Step 31: The barometric pressure time series input to the deep learning model is used to construct input samples using a fixed-length sliding window. Let the input samples be represented as:

[0018]

[0019] in, Indicates time The corresponding air pressure value, The input samples are arranged in chronological order and then input into the deep learning model, representing the sequence length.

[0020] The deep learning model first extracts local features from the input air pressure time series samples through multi-scale one-dimensional convolutional layers. The convolution operation expression is as follows:

[0021]

[0022] Wherein, the size of the convolution kernel , The weight parameter represents the weight at position i of the k-th scale convolution kernel. Represents the bias term parameter. Represents a modified linear unit. This represents the sampled value of the input pressure sequence within the corresponding receptive field.

[0023] Step 32: Input the output features of the convolutional layer into the temporal modeling layer to extract continuous temporal dependencies;

[0024] The temporal modeling layer employs a temporal convolutional network, which extracts temporal variation features within a 1-second range at a sampling frequency of 100Hz through dilated convolution. The dilated convolution expression is as follows:

[0025]

[0026] coefficient of expansion The convolution kernel length is n, where n=3. The weight parameter representing the i-th position of the convolution kernel. This represents the sampled value of the previous convolutional layer's output within the current receptive field;

[0027] Step 33: After the temporal modeling layer output is processed by the fully connected classification layer, the probability values ​​of each motion mode are output through the Softmax function.

[0028]

[0029] in: For the first Probability of motion pattern; The total number of sports mode categories, Represents the Euler number. , Represents the output of the fully connected layer;

[0030] Steps 3 and 4: Select the category with the highest probability value as the current motion pattern recognition result:

[0031]

[0032] in: For the final identification category;

[0033] Step 4: The auxiliary control module issues corresponding air pressure control commands based on the motion pattern recognition results;

[0034] Step 5: Achieve precise control of airbag pressure through the air pump and valve control module;

[0035] Step Six: The pneumatic knee brace actively responds to changes in movement patterns, achieving an assistive effect for the wearer during exercise;

[0036] Step 7: The safety module monitors the knee brace system in real time to prevent safety issues caused by equipment malfunctions.

[0037] Furthermore, the air pressure sensor mentioned in step one is an air pressure sensor installed inside the flexible airbag of the wearable pneumatic knee brace or at the air passage location connected to the flexible airbag. The air pressure sensor is used to detect in real time the changes in pressure inside the airbag caused by the wearer's knee joint during flexion, extension and load changes.

[0038] The pressure sensor converts the detected analog pressure signal into an electrical signal, and then converts it into digital pressure data through an analog-to-digital converter module.

[0039] The main control unit continuously reads digital pressure data at a sampling frequency of 10Hz-200Hz and forms a continuous pressure change sequence arranged in time order through a transfer function;

[0040] The continuous air pressure change sequence exhibits periodic dynamic changes as the wearer's movement changes, and serves as the data input for subsequent movement pattern recognition.

[0041] Furthermore, the median filtering algorithm described in step two uses a sliding window of length 2k+1 to process the original air pressure sequence. Let the original air pressure sequence be:

[0042]

[0043] The filtering result for the i-th sampling point is defined as:

[0044]

[0045] In the formula, is the output value after median filtering; k is the half-width of the window.

[0046] Preferably, the low-pass filtering in step two uses a digital low-pass filter to smooth the median-filtered air pressure sequence, and its recursive expression is:

[0047]

[0048] In the formula, Enter the current air pressure value; This is the current filter output value; This is the filtered output value from the previous moment; Let the filter coefficients satisfy: .

[0049] Preferably, the standardization process in step two employs a maximum-minimum-value normalization method to map the low-pass filtered air pressure sequence to a unified numerical range, the expression of which is:

[0050]

[0051] in, The output value is the standardized value; This is the air pressure value after low-pass filtering; The minimum value within the current sample window; This represents the maximum value within the current sample window.

[0052] Furthermore, step four also includes the auxiliary control module receiving the motion pattern recognition results output in step three, and calling the corresponding auxiliary control strategy according to different motion patterns to generate target air pressure control commands, and sending them to the air pump and valve control module.

[0053] Furthermore, the auxiliary control module in step four establishes a mapping relationship between motion modes and target auxiliary pressures, with different motion modes corresponding to different target pressure ranges:

[0054] When the motion pattern recognition result is walking on flat ground, the auxiliary control module outputs the first target pressure value;

[0055] When the motion pattern recognition result is climbing stairs, the auxiliary control module outputs a second target pressure value that is higher than the first target pressure value;

[0056] When the motion pattern recognition result is descending stairs, the auxiliary control module outputs a third target pressure value to mitigate the impact on the knee joint;

[0057] When the motion pattern recognition result is sitting, the auxiliary control module outputs a fourth target pressure value that is lower than the first target pressure value.

[0058] When the motion pattern is identified as running, the auxiliary control module outputs a momentarily increased fifth target pressure value.

[0059] Furthermore, in step five, the air pump and valve control module receives the target pressure control command output by the auxiliary control module, and controls the air pump and solenoid valve to operate based on the deviation between the target pressure value and the current real-time pressure value inside the airbag, so as to achieve precise adjustment of the pressure inside the airbag. The adjustment process is as follows:

[0060] The air pump and valve control module reads the current internal pressure value of the airbag in real time and calculates the pressure offset:

[0061]

[0062] in: The target pressure value output by the auxiliary control module; The current pressure value is collected in real time by the barometric pressure sensor; This represents the current pressure deviation.

[0063] When the pressure deviation meets:

[0064]

[0065] At this time, the air pump and valve control module starts the miniature air pump and closes the pressure relief solenoid valve to inflate the airbag; wherein: The preset pressure tolerance threshold is used;

[0066] When the pressure deviation meets the following conditions:

[0067]

[0068] At this time, the air pump and valve control module shuts down the miniature air pump and opens the pressure relief solenoid valve to deflate the airbag;

[0069] When the pressure deviation meets the following conditions:

[0070]

[0071] At this time, the air pump and valve control module keeps the air pump off and the solenoid valve in its current state, so that the airbag maintains its current pressure.

[0072] Furthermore, in step six, the pneumatic knee brace adjusts the internal pressure of the airbag under the action of the air pump and air valve control module, so that the flexible airbag applies dynamic support force to the wearer's knee joint, thereby providing an auxiliary effect during knee joint movement.

[0073] When the pressure inside the airbag increases, the flexible airbag expands and applies supporting pressure to the inside of the knee brace, which restrains the soft tissues around the knee joint, thereby improving knee joint stability.

[0074] The auxiliary force output by the flexible airbag is positively correlated with the internal pressure of the airbag.

[0075]

[0076] in: To provide auxiliary support; The internal pressure of the airbag; The structural transfer factor;

[0077] During walking or climbing stairs, the pneumatic knee brace increases support pressure during the knee flexion phase to reduce the output load on the quadriceps.

[0078] During the downstairs exercise, the pneumatic knee brace increases the cushioning and support pressure to reduce the impact load on the knee joint;

[0079] During running, the pneumatic knee brace delivers enhanced auxiliary pressure during the knee extension phase to improve running stability.

[0080] The auxiliary pressure is dynamically adjusted according to the movement pattern, so that the knee brace outputs support force in sync with the wearer's knee joint movement state;

[0081] During the switching of sports modes, the internal pressure of the airbag is adjusted gradually to avoid sudden changes in assistive force affecting wearing comfort.

[0082]

[0083] in: This is the gradual adjustment coefficient.

[0084] Furthermore, in step seven, the safety module collects real-time data on the airbag's internal pressure, control execution status, and sensor operating status, and monitors the system's operating status based on preset safety thresholds. When an abnormal state is detected, protective control is executed.

[0085] The safety module monitors the internal pressure of the airbag in real time and determines whether the current pressure exceeds the maximum allowable pressure threshold. When the real-time pressure exceeds the maximum safe pressure threshold, the air pump is immediately shut down and the pressure relief solenoid valve is opened.

[0086] Once the target pressure is established, if the real-time pressure remains below the target pressure threshold for a preset period of time, the system is determined to have a leakage anomaly.

[0087] If the pressure sensor output value remains unchanged for multiple consecutive sampling periods, the sensor is considered faulty. The determination formula is as follows:

[0088]

[0089] When the air pump runs continuously for more than the preset time threshold and fails to reach the target pressure, the air pump output will stop.

[0090] If a serious anomaly is detected, rapid depressurization is performed to restore the airbag to a safe pressure range;

[0091] After the anomaly is resolved, the control module re-enters standby mode and waits for the next motion pattern recognition result.

[0092] Compared with the prior art, the present invention has the following advantages:

[0093] 1. The present invention has accurate motion pattern recognition: By collecting the time-series characteristics of the air pressure inside the knee joint airbag and combining it with a deep learning model for inference, the present invention can accurately identify the wearer's motion pattern (such as walking on flat ground, going up and down stairs, running, sitting still, etc.) in real time, overcoming the shortcomings of existing knee braces that lack the ability to adapt to motion states.

[0094] 2. This invention achieves intelligent auxiliary control: Based on the identified motion pattern, this invention can dynamically generate corresponding target pressure control commands, realize closed-loop pressure regulation and gradual control, so that the airbag provides timely and controllable support to the knee joint, and improves the motion assistance effect.

[0095] 3. Precise air pressure regulation of the present invention: Combining air pump, air valve and PWM control technology, the present invention can finely regulate the airbag pressure, realize rapid inflation and deflation and steady pressure maintenance, and ensure that the assist force of the knee brace matches the actual needs of the knee joint in different stages of exercise.

[0096] 4. This invention achieves a balance between comfort and flexibility: the knee brace uses a flexible airbag to apply support, and the pressure adjustment is gradual to avoid sudden pressure causing discomfort to the wearer, while ensuring the freedom of movement of the knee joint and significantly improving wearing comfort.

[0097] 5. The safety protection of this invention is more complete: the safety module monitors the airbag pressure, sensor status and control execution status in real time, and realizes overpressure protection, pressure relief protection, sensor abnormality alarm and motion mode abnormality freeze, effectively reducing the safety risks during the use of the knee brace.

[0098] 6. The system intelligence and automation of the present invention: The present invention realizes full-process automation from motion sensing to air pressure control, without the need for manual intervention, and can adapt to a variety of sports scenarios, which is significantly better than traditional fixed pressure or manually adjustable knee braces.

[0099] 7. This invention has a wide range of applications: it is applicable to daily life and sports scenarios, and can take into account the needs of knee joint protection, sports assistance and rehabilitation training, and has high application and promotion value. Attached Figure Description

[0100] Figure 1 is a flowchart illustrating a method for motion pattern recognition and auxiliary control of wearable pneumatic knee braces based on air pressure timing characteristics provided by the present invention.

[0101] Figure 2 is a schematic diagram of the pneumatic knee pad airbag configuration scheme in the embodiment of the present invention;

[0102] Figure 3 is a schematic diagram of the pneumatic knee brace mechanical structure in the embodiment of the present invention;

[0103] Figure 4 is a schematic diagram of the circuit connection and air circuit connection of the pneumatic knee pad control module in the embodiment of the present invention. Detailed Implementation

[0104] Specific implementation method one: Combining Figures 1 to 4 This embodiment describes the following steps:

[0105] Step 1: Collect the original air pressure sequence in the airbag during the wearer's exercise using an air pressure sensor;

[0106] Step 2: Standardize the data using the data preprocessing module to obtain the air pressure sequence;

[0107] Step 21: The median filtering algorithm is used to remove impulse noise and abnormal abrupt changes generated during the sampling process from the original air pressure sequence through the data preprocessing module;

[0108] Step 22: Input the median-filtered air pressure sequence into the low-pass filter module to suppress high-frequency disturbances and preserve the air pressure change trend during human movement;

[0109] Steps 2 and 3: Perform standardization processing on the low-pass filtered air pressure sequence to map the data in different sampling periods to a uniform numerical range, so as to serve as the air pressure time series input data for the subsequent motion pattern recognition model;

[0110] Step 3: The barometric pressure sequence is fed into a deep learning model for inference to obtain motion pattern recognition results;

[0111] The preprocessed barometric time series is input into a deep learning motion pattern recognition model. The deep learning motion pattern recognition model performs feature extraction and classification inference on the input sequence and outputs the motion pattern recognition result corresponding to the current wearer. The specific steps include:

[0112] Step 31: The barometric pressure time series input to the deep learning model is used to construct input samples using a fixed-length sliding window. Let the input samples be represented as:

[0113]

[0114] in, Indicates time The corresponding air pressure value, The input samples are arranged in chronological order and then input into the deep learning model, representing the sequence length.

[0115] The deep learning model first extracts local features from the input air pressure time series samples through multi-scale one-dimensional convolutional layers. The convolution operation expression is as follows:

[0116]

[0117] Wherein, the size of the convolution kernel , The weight parameter represents the weight at position i of the k-th scale convolution kernel. Represents the bias term parameter. Represents a modified linear unit. This represents the sampled value of the input pressure sequence within the corresponding receptive field.

[0118] Step 32: Input the output features of the convolutional layer into the temporal modeling layer to extract continuous temporal dependencies;

[0119] The temporal modeling layer employs a temporal convolutional network, which extracts temporal variation features within a 1-second range at a sampling frequency of 100Hz through dilated convolution. The dilated convolution expression is as follows:

[0120]

[0121] coefficient of expansion The convolution kernel length is n, where n=3. The weight parameter representing the i-th position of the convolution kernel. This represents the sampled value of the previous convolutional layer's output within the current receptive field;

[0122] Step 33: After the temporal modeling layer output is processed by the fully connected classification layer, the probability values ​​of each motion mode are output through the Softmax function.

[0123]

[0124] in: For the first Probability of motion pattern; The total number of sports mode categories, Represents the Euler number. , Represents the output of the fully connected layer;

[0125] Steps 3 and 4: Select the category with the highest probability value as the current motion pattern recognition result:

[0126]

[0127] in: For the final identification category;

[0128] Step 4: The auxiliary control module issues corresponding air pressure control commands based on the motion pattern recognition results;

[0129] Step 5: Achieve precise control of airbag pressure through the air pump and valve control module;

[0130] Step Six: The pneumatic knee brace actively responds to changes in movement patterns, achieving an assistive effect for the wearer during exercise;

[0131] Step 7: The safety module monitors the knee brace system in real time to prevent safety issues caused by equipment malfunctions.

[0132] This invention utilizes a pressure sensor embedded in the internal airbag of a wearable pneumatic knee brace to collect real-time pressure changes within the airbag during knee joint movement. The collected raw pressure time-series data is then input to the main control unit. First, the pressure signal undergoes median filtering, smoothing, and normalization to eliminate noise interference and improve data stability. Then, a sliding window is used to segment the continuous pressure data, constructing a fixed-length time-series input sample. The pre-processed pressure sequence is input into a motion recognition model composed of a convolutional neural network and a temporal feature extraction network to classify and identify movement patterns such as walking, climbing stairs, descending stairs, sitting, and running. Based on the recognition results, corresponding auxiliary control strategies are invoked to control a micro-pump and solenoid valve to regulate the inflation and deflation of the knee brace airbag, achieving dynamic output of auxiliary support force under different movement states. Simultaneously, the system incorporates an independent safety monitoring mechanism to detect abnormal pressure states in real time, automatically implementing pressure relief protection when the pressure exceeds a set threshold.

[0133] This invention utilizes only a single air pressure sensor to achieve motion pattern recognition and assisted control linkage. It has the advantages of simple structure, high real-time performance, comfortable wear and high level of intelligence, and is applicable to the fields of rehabilitation assistance, knee joint protection and lower limb assistive wearable devices.

[0134] Specific Implementation Method Two: Combining Figure 1 This embodiment describes a pressure sensor that is installed inside the flexible airbag of the wearable pneumatic knee brace or in an air passage connected to the flexible airbag. The pressure sensor is used to detect in real time the pressure changes inside the airbag caused by the wearer's knee joint during flexion, extension, and load changes.

[0135] The pressure sensor converts the detected analog pressure signal into an electrical signal, and then converts it into digital pressure data through an analog-to-digital converter module.

[0136] The main control unit continuously reads digital pressure data at a sampling frequency of 10Hz-200Hz and forms a continuous pressure change sequence arranged in time order through a transfer function;

[0137] The continuous air pressure change sequence exhibits periodic dynamic changes as the wearer's movement changes, and serves as the data input for subsequent movement pattern recognition.

[0138] This implementation uses only a single air pressure sensor to simultaneously achieve motion sensing and pressure feedback, eliminating the need for additional inertial measurement units or angle sensors, thus reducing hardware costs and system complexity. The sensor is directly placed inside the airbag or connected to the air passage, collecting the direct mechanical action signal generated by knee joint movement on the airbag, eliminating data synchronization issues common in multi-sensor fusion. The sampling frequency is adjustable from 10Hz to 200Hz, covering various motion scenarios and ensuring no loss of detail in air pressure changes. A clear mechanical coupling relationship exists between knee flexion, extension, and load changes and airbag pressure, providing a reliable physical basis for subsequent motion pattern recognition based on air pressure timing.

[0139] Specific implementation method three: Combining Figure 1 This embodiment describes a median filtering algorithm in step two that uses a sliding window of length 2k+1 to process the original air pressure sequence. Let the original air pressure sequence be:

[0140]

[0141] The filtering result for the i-th sampling point is defined as:

[0142]

[0143] In the formula, is the output value after median filtering; k is the half-width of the window.

[0144] This implementation removes impulse noise and anomalous abrupt changes from the original barometric pressure sequence. A sliding window of length 2k+1 is used. All sampling points within the window are sorted by value, and the median is taken as the filtered output for the current point. Compared to mean filtering, this method better preserves the edge features of the barometric pressure signal and avoids signal distortion caused by individual outliers. The half-width k of the window can be set according to the actual sampling frequency and noise level. A larger k value results in stronger filtering but also greater loss of signal details; a balance must be struck between noise reduction and fidelity preservation. After median filtering, spikes caused by instantaneous impacts to the airbag or sensor sampling jitter during knee joint movement are effectively suppressed, providing a cleaner signal foundation for subsequent low-pass filtering and deep learning recognition.

[0145] Specific implementation method four: Combination Figure 1 This embodiment describes a method where, in step two, a digital low-pass filter is used to smooth the median-filtered air pressure sequence. The recursive expression for this method is:

[0146]

[0147] In the formula, Enter the current air pressure value; This is the current filter output value; This is the filtered output value from the previous moment; Let the filter coefficients satisfy: .

[0148] This implementation method can suppress high-frequency disturbances in the air pressure signal while preserving the main air pressure change trends during human movement. A first-order digital low-pass filter is used, where the filter coefficient α in the recursive expression is between 0 and 1. A smaller α results in a stronger filtering effect but a slower response to signal changes, while a larger α results in a faster response but a weaker ability to suppress high-frequency noise. After removing impulse noise through median filtering, the signal may still contain high-frequency fluctuations caused by sensor noise or external vibrations. These high-frequency components contribute little to motion pattern recognition and may even interfere with the model's judgment of the overall trend. Through low-pass filtering, the air pressure curve becomes smoother, and the trends of air pressure rise and fall during knee flexion and extension become clearer, providing stable time-series data for subsequent standardization processing and deep learning model input.

[0149] Specific Implementation Method Five: Combining Figure 1 This embodiment describes a standardization process in step two that uses a maximum / minimum value normalization method to map the low-pass filtered air pressure sequence to a uniform numerical range. The expression for this method is:

[0150]

[0151] in, The output value is the standardized value; This is the air pressure value after low-pass filtering; The minimum value within the current sample window; This represents the maximum value within the current sample window.

[0152] This implementation eliminates dimensional differences in air pressure data across different sampling periods or among different wearers, mapping the low-pass filtered air pressure sequence to a uniform numerical range (typically 0 to 1). Because wearers vary in body size, fit, and exercise intensity, the range of raw air pressure amplitudes collected may differ. Directly inputting this data into a deep learning model can lead to poor convergence or decreased generalization ability. By employing a maximum and minimum value normalization method, using the maximum and minimum values ​​within the current sample window as benchmarks, each sampling point is mapped to a fixed interval, ensuring all samples have the same numerical distribution range. After standardization, the air pressure sequence maintains consistency in amplitude scale, eliminating individual differences and dimensional influences. During model training, this allows for greater focus on the waveform shape of air pressure changes rather than absolute amplitude, thereby improving the accuracy of motion pattern recognition and adaptability to different users.

[0153] Specific Implementation Method Six: Combination Figure 1 In this embodiment, step four further includes receiving the motion pattern recognition result output in step three, and calling the corresponding auxiliary control strategy according to different motion patterns to generate a target air pressure control command, which is then sent to the air pump and valve control module.

[0154] This implementation establishes a linkage between motion pattern recognition results and pneumatic actuators, converting the recognized motion state into specific control commands. The auxiliary control module receives the motion pattern recognition results (such as walking, going upstairs, going downstairs, sitting, running, etc.) output in step three, and calls a preset auxiliary control strategy based on the knee joint stress characteristics under different motion modes to generate corresponding target air pressure values. For example, when going upstairs, the knee joint needs to overcome gravity and requires greater support force, therefore a higher target pressure is output; when going downstairs, the knee joint bears a greater impact load and needs buffer support, so the target pressure is set to a moderate value; when sitting, the knee joint is unloaded, so a lower pressure is output to avoid compression. The generated target air pressure control commands are sent to the air pump and valve control module via a communication interface, enabling the recognition results to drive the actuator's action, achieving a closed-loop linkage from perception to control, and preventing the recognition results from being used only for display or recording without contributing to the actual auxiliary effect of the knee brace.

[0155] Specific implementation method seven: Combination Figure 1This embodiment describes a method where, in step four, the auxiliary control module establishes a mapping relationship between motion modes and target auxiliary pressures, with different motion modes corresponding to different target pressure ranges.

[0156] When the motion pattern recognition result is walking on flat ground, the auxiliary control module outputs the first target pressure value;

[0157] When the motion pattern recognition result is climbing stairs, the auxiliary control module outputs a second target pressure value that is higher than the first target pressure value;

[0158] When the motion pattern recognition result is descending stairs, the auxiliary control module outputs a third target pressure value to mitigate the impact on the knee joint;

[0159] When the motion pattern recognition result is sitting, the auxiliary control module outputs a fourth target pressure value that is lower than the first target pressure value.

[0160] When the motion pattern is identified as running, the auxiliary control module outputs a momentarily increased fifth target pressure value.

[0161] This implementation establishes differentiated air pressure control targets based on the biomechanical needs of the knee joint in different movement modes to achieve precise matching of assist force. When walking on flat ground, the knee joint flexion and extension range is moderate, and the required assist force is relatively balanced, so the first target pressure value is set at the baseline level. When climbing stairs, the knee joint needs to overcome its own weight to do upward work, and the quadriceps muscle bears a large load, requiring stronger support assistance, so the second target pressure value is higher than the first target pressure value. When descending stairs, the knee joint bears a large impact load, requiring cushioning assistance to reduce joint impact, so the third target pressure value takes into account both support and cushioning. When sitting, the knee joint is unloaded, and excessively high air pressure will cause unnecessary compression, so the fourth target pressure value is lower than the baseline level to maintain basic fit. When running, the knee joint flexion and extension frequency is fast and the impact force is large, requiring a fifth target pressure value that increases instantaneously to provide enhanced support in key phases.

[0162] By establishing a mapping relationship between this movement pattern and the target pressure, the auxiliary force output by the knee brace changes synchronously with the actual force requirements of the knee joint, avoiding the problems of insufficient support or excessive assistance in different movement patterns caused by using a uniform pressure value.

[0163] Specific implementation method eight: Combination Figure 1 In this embodiment, the air pump and valve control module in step five receives the target pressure control command output by the auxiliary control module, and controls the air pump and solenoid valve to operate based on the deviation between the target pressure value and the current real-time pressure value inside the airbag, so as to achieve precise adjustment of the pressure inside the airbag. The adjustment process is as follows:

[0164] The air pump and valve control module reads the current internal pressure value of the airbag in real time and calculates the pressure offset:

[0165]

[0166] in: The target pressure value output by the auxiliary control module; The current pressure value is collected in real time by the barometric pressure sensor; This represents the current pressure deviation.

[0167] When the pressure deviation meets the following conditions:

[0168]

[0169] At this time, the air pump and valve control module starts the miniature air pump and closes the pressure relief solenoid valve to inflate the airbag; wherein: The preset pressure tolerance threshold is used;

[0170] When the pressure deviation meets the following conditions:

[0171]

[0172] At this time, the air pump and valve control module shuts down the miniature air pump and opens the pressure relief solenoid valve to deflate the airbag;

[0173] When the pressure deviation meets the following conditions:

[0174]

[0175] At this time, the air pump and valve control module keeps the air pump off and the solenoid valve in its current state, so that the airbag maintains its current pressure.

[0176] This implementation achieves precise airbag pressure regulation through closed-loop control, enabling the actual pressure value to quickly and stably follow the target pressure command. The air pump and valve control module reads the current internal pressure value of the airbag in real time and calculates the deviation from the target pressure value. When the deviation is greater than the preset allowable error threshold ΔP, it indicates that the current pressure is lower than the target value. The control module starts the micro air pump to inflate and closes the pressure relief solenoid valve, causing the pressure to rise. When the deviation is less than -ΔP, it indicates that the current pressure is higher than the target value. The control module shuts off the air pump and opens the pressure relief solenoid valve to release air, causing the pressure to drop. When the deviation is within the range of -ΔP to ΔP, it indicates that the pressure has reached near the target value. The control module keeps the air pump off and the solenoid valve in its current state to maintain pressure. By setting the error threshold ΔP, the increased energy consumption and wear of the actuator caused by frequent opening and closing of the air pump and solenoid valve are avoided, while ensuring the accuracy of pressure control. This three-stage adjustment method of inflation, deflation, and pressure maintenance allows the airbag pressure to be dynamically adjusted according to the movement mode, providing the knee joint with auxiliary support force that matches the movement requirements.

[0177] Specific Implementation Method Nine: Combining Figure 1 This embodiment describes the pneumatic knee brace in step six, where the air pump and valve control module adjusts the internal pressure of the airbag, allowing the flexible airbag to apply dynamic support to the wearer's knee joint, thereby providing assistance during knee joint movement.

[0178] When the pressure inside the airbag increases, the flexible airbag expands and applies supporting pressure to the inside of the knee brace, which restrains the soft tissues around the knee joint, thereby improving knee joint stability.

[0179] The auxiliary force output by the flexible airbag is positively correlated with the internal pressure of the airbag.

[0180]

[0181] in: To provide auxiliary support; The internal pressure of the airbag; The structural transfer factor;

[0182] During walking or climbing stairs, the pneumatic knee brace increases support pressure during the knee flexion phase to reduce the output load on the quadriceps.

[0183] During the downstairs exercise, the pneumatic knee brace increases the cushioning and support pressure to reduce the impact load on the knee joint;

[0184] During running, the pneumatic knee brace delivers enhanced auxiliary pressure during the knee extension phase to improve running stability.

[0185] The auxiliary pressure is dynamically adjusted according to the movement pattern, so that the knee brace outputs support force in sync with the wearer's knee joint movement state;

[0186] During the switching of sports modes, the internal pressure of the airbag is adjusted gradually to avoid sudden changes in assistive force affecting wearing comfort.

[0187]

[0188] in: This is the gradual adjustment coefficient.

[0189] This implementation transforms airbag pressure regulation into actual knee joint assistance, achieving a physical transfer from pressure control to movement assistance. When the airbag pressure increases, the flexible airbag expands and applies supporting pressure to the inner side of the knee brace, constraining the soft tissues around the knee joint and thus improving joint stability. The assisting support force F and the airbag pressure P are positively correlated: F=K·P, where K is the structural transmission coefficient, depending on the mechanical transmission efficiency of the airbag and knee brace structure. The timing and mode of application of the assisting force vary depending on the movement mode: when walking or climbing stairs, the supporting pressure is increased during the knee flexion phase to assist the quadriceps muscles and reduce muscle burden; when descending stairs, the cushioning support pressure is increased to absorb the impact load on the knee joint; during running, the enhanced assisting pressure is output during the extension phase to improve movement stability. When switching movement modes, the pressure is adjusted according to… Adjust the gradient method. The coefficient is gradually adjusted to avoid discomfort caused by sudden changes in assistive force. Through this dynamic adjustment, the support force output by the knee brace is synchronized with the knee joint's movement, achieving the effect of sports assistance.

[0190] Specific Implementation Method Ten: Combining Figure 1 This embodiment describes a method where, in step seven, the safety module collects real-time data on the airbag's internal pressure, control execution status, and sensor operating status. It also monitors the system's operating status based on preset safety thresholds and executes protective controls when an abnormal state is detected.

[0191] The safety module monitors the internal pressure of the airbag in real time and determines whether the current pressure exceeds the maximum allowable pressure threshold. When the real-time pressure exceeds the maximum safe pressure threshold, the air pump is immediately shut down and the pressure relief solenoid valve is opened.

[0192] Once the target pressure is established, if the real-time pressure remains below the target pressure threshold for a preset period of time, the system is determined to have a leakage anomaly.

[0193] If the pressure sensor output value remains unchanged for multiple consecutive sampling periods, the sensor is considered faulty. The determination formula is as follows:

[0194]

[0195] When the air pump runs continuously for more than the preset time threshold and fails to reach the target pressure, the air pump output will stop.

[0196] If a serious anomaly is detected, rapid depressurization is performed to restore the airbag to a safe pressure range;

[0197] After the anomaly is resolved, the control module re-enters standby mode and waits for the next motion pattern recognition result.

[0198] This implementation allows for independent monitoring of the knee brace system's operational safety. When an abnormality is detected, timely protective controls are implemented to prevent injury to the wearer from equipment malfunction. The safety module collects real-time data on the airbag's internal pressure, control execution status, and sensor operating status, and makes judgments based on preset safety thresholds: when the real-time pressure exceeds the maximum safe pressure threshold, the air pump is immediately shut off and the pressure relief solenoid valve is opened to prevent overpressure damage to the knee joint; if the real-time pressure remains below the target pressure threshold for a preset time after the target pressure is established, a leakage abnormality is identified, preventing auxiliary system failure due to insufficient air pressure; if the pressure sensor output value remains unchanged for multiple consecutive sampling cycles, a sensor abnormality is identified, preventing misjudgment by the control system due to sensor failure; if the air pump runs continuously for more than a preset time threshold without reaching the target pressure, the air pump output is stopped to prevent damage or overheating from prolonged idling; in the event of a serious abnormality, rapid pressure relief is performed to restore the airbag to the safe pressure range. After the abnormality is resolved, the control module re-enters standby mode, awaiting the next motion mode recognition result. This independent safety monitoring mechanism ensures that the knee brace responds promptly in various abnormal situations, protecting the wearer's safety.

[0199] Combination Figures 1 to 4 Description of embodiments of the present invention:

[0200] This invention collects the air pressure sequence in the wearer's airbag and uses a neural network model to infer the movement pattern, thereby achieving precise control of the airbag pressure for the target movement pattern and thus achieving excellent sports assistance effect.

[0201] like Figure 1 As shown, the method specifically includes the following steps:

[0202] Step 1: By using a highly sensitive air pressure sensor placed inside the knee pad airbag, the sequence of pressure changes inside the airbag during the wearer's knee joint movement is collected in real time;

[0203] Step 2: Perform median filtering and low-pass filtering to denoise the collected air pressure sequence, and then perform standardization to improve the accuracy of subsequent identification;

[0204] Step 3: Input the processed air pressure sequence into the deep learning model, extract temporal features, and classify and output the current motion pattern;

[0205] Step 4: The auxiliary control module generates target air pressure control commands based on the recognition results, maps different motion modes to airbag pressure ranges, and uses closed-loop control to adjust the pressure;

[0206] Step 5: The air pump and valve control module performs inflation, deflation, or pressure holding operations according to the target pressure to achieve precise adjustment of the airbag pressure;

[0207] Step Six: Knee braces provide support, cushioning, and stability to the knee joint through dynamic pressure output, improving the safety and comfort of knee joint movement;

[0208] Step 7: The safety module monitors the airbag pressure, sensor status, and control execution status in real time, providing functions such as overpressure protection, pressure relief protection, sensor malfunction alarm, and motion mode malfunction freeze.

[0209] The location of the air pressure sensor mentioned in step one of this invention mainly refers to placing the air pressure sensor inside the flexible airbag of the wearable pneumatic knee brace or at the air passage location connected to the flexible airbag. The air pressure sensor is used to detect in real time the pressure changes inside the airbag caused by the wearer's knee joint during flexion, extension, and load changes. The air pressure sensor converts the detected analog air pressure signal into an electrical signal, and then converts it into digital pressure data through an analog-to-digital converter. The main control unit continuously reads the digital pressure data at a sampling frequency of 10Hz-200Hz, and forms a continuous air pressure change sequence arranged in chronological order through a simple transfer function. The continuous air pressure change sequence exhibits periodic dynamic changes with the wearer's movement state and serves as the data input for subsequent movement pattern recognition. The collected air pressure sequence is as follows:

[0210] p = [19.7802, 19.7314, 19.7314, 19.8291, 19.8291, 19.7314, 19.5849, 19.4872, 19.4872, 19.4872, 19.4872, 19.5849, 19.6337, 19.6337, 19.6825, 19.7314, 19.7314, 19.7314, 19.7802, 19.8291, 19.8291, 20, 20.3907, 20.3907, 20.3907 , 20.7326, 20.6838, 20.6349, 20.6349, 20.6349, 20.5372, 20.3907, 20.3419, 20.3419, 20.1954, 20.0977, 20.0488, 20, 19.8779, 19.9512, 20.0977, 20.293, 20.5372, 20.6349, 20.8303, 21.0745, 21.2454, 21.4896, 21.7338, 21.8803, 2 2.0757, 22.1734, 22.442, 22.5885, 22.6374, 22.5885, 22.6374, 22.6374, 22.735, 22.735, 22.6374, 22.6374, 22.442, 22.1245, 21.7827, 21.2943, 20.7326, 20.3907, 20.2442, 20.1465, 19.8779, 19.6825, 19.4872, 19.1941, 18.9499, 18.9011, 18.9011, 19.6337, 19.7314, 19.7314, 20, 20.4396, 20.7326, 20.3419, 20.4884, 20.5372, 20.6838, 21.0745, 21.2454, 21.2454, 21.1233, 20.9768, 20.7326, 20.5372, 20.3907, 20.2442, 20.0488, 19.9512, 19.8291, 19.8291).

[0211] Step two of this invention includes data preprocessing of the original air pressure change sequence obtained in step one, specifically including: using a median filtering algorithm to remove impulse noise and abnormal abrupt changes generated during the sampling process from the original air pressure sequence; inputting the median-filtered air pressure sequence into a low-pass filtering module to suppress high-frequency disturbance components and retain the main air pressure change trends during human movement; and performing standardization processing on the low-pass-filtered air pressure sequence to map data from different sampling periods to a unified numerical range, so as to serve as input data for the subsequent motion pattern recognition model.

[0212] Step 21: The median filtering algorithm uses a sliding window of length 2k+1 to process the original air pressure sequence. Let the original air pressure sequence be:

[0213]

[0214] The filtering result for the i-th sampling point is defined as:

[0215]

[0216] In the formula, is the output value after median filtering; k is the half-width of the window.

[0217] Step 22: Low-pass filtering. A digital low-pass filter is used to smooth the median-filtered air pressure sequence. Its recursive expression is:

[0218]

[0219] In the formula, Enter the current air pressure value; This is the current filter output value; This is the filtered output value from the previous moment; Let the filter coefficients satisfy: .

[0220] Steps 2 and 3: Standardization processing uses a maximum and minimum value normalization method to map the low-pass filtered air pressure sequence to a uniform numerical range. The expression is as follows:

[0221]

[0222] in, The output value is the standardized value; This is the air pressure value after low-pass filtering; The minimum value within the current sample window; This represents the maximum value within the current sample window.

[0223] Step three of this invention includes inputting the pre-processed air pressure time series into a deep learning motion pattern recognition model, which then performs feature extraction and classification reasoning on the input sequence and outputs the motion pattern recognition result corresponding to the current wearer.

[0224] The barometric pressure time series input to the deep learning model is constructed using a fixed-length sliding window. Let the input sample be represented as:

[0225]

[0226] in Indicates time The corresponding air pressure value, This indicates the sequence length. The input samples are arranged in chronological order and then input into the deep learning model.

[0227] The deep learning model first extracts local features from the input air pressure time series samples through multi-scale one-dimensional convolutional layers. The convolution operation expression is as follows:

[0228]

[0229] Wherein, the size of the convolution kernel , The weight parameter represents the weight at position i of the k-th scale convolution kernel. Represents the bias term parameter. Represents a modified linear unit. This represents the sampled value of the input pressure sequence within the corresponding receptive field.

[0230] Step 32: Input the output features of the convolutional layer into the temporal modeling layer to extract continuous temporal dependencies;

[0231] The temporal modeling layer employs a temporal convolutional network, which extracts temporal variation features within a 1-second range at a sampling frequency of 100Hz through dilated convolution. The dilated convolution expression is as follows:

[0232]

[0233] coefficient of expansion The convolution kernel length is n, where n=3. The weight parameter representing the i-th position of the convolution kernel. This represents the sampled value of the previous convolutional layer's output within the current receptive field;

[0234] Step 33: After the temporal modeling layer output is processed by the fully connected classification layer, the probability values ​​of each motion mode are output through the Softmax function.

[0235]

[0236] in: For the first Probability of motion pattern; The total number of sports mode categories, Represents the Euler number. , Represents the output of the fully connected layer;

[0237] The category with the highest probability value is taken as the current motion pattern recognition result:

[0238]

[0239] in: This is for the final category identification.

[0240] Step four of the present invention includes the auxiliary control module receiving the motion mode recognition result output in step three, and calling the corresponding auxiliary control strategy according to different motion modes to generate a target air pressure control command and send it to the air pump and valve control module.

[0241] Step 41: The auxiliary control module establishes a mapping relationship between movement modes and target auxiliary pressures, with different movement modes corresponding to different target pressure ranges. When the movement mode recognition result is walking on flat ground, the auxiliary control module outputs a first target pressure value; when the movement mode recognition result is climbing stairs, it outputs a second target pressure value higher than the first target pressure value; when the movement mode recognition result is descending stairs, it outputs a third target pressure value to reduce knee joint impact; when the movement mode recognition result is sitting still, it outputs a fourth target pressure value lower than the first target pressure value; and when the movement mode recognition result is running, it outputs a fifth target pressure value that increases instantaneously.

[0242] Step five of the present invention includes the air pump and valve control module receiving the target pressure control command output by the auxiliary control module, and controlling the air pump and solenoid valve to work according to the deviation between the target pressure value and the current real-time pressure value inside the airbag, so as to achieve precise adjustment of the pressure inside the airbag.

[0243] The air pump and valve control module reads the current internal pressure value of the airbag in real time and calculates the pressure offset:

[0244]

[0245] in: The target pressure value output by the auxiliary control module; The current pressure value is collected in real time by the barometric pressure sensor; This represents the current pressure deviation.

[0246] When the pressure deviation meets the following conditions:

[0247]

[0248] At this time, the control module starts the miniature air pump and closes the pressure relief solenoid valve to inflate the airbag. Specifically: This is the preset pressure tolerance threshold.

[0249] When the pressure deviation meets the following conditions:

[0250]

[0251] At this time, the control module shuts down the miniature air pump and opens the pressure relief solenoid valve to deflate the airbag.

[0252] When the pressure deviation meets the following conditions:

[0253]

[0254] At this time, the control module keeps the air pump off and the solenoid valve in its current state, so that the airbag maintains its current pressure.

[0255] Step six of this invention involves the pneumatic knee brace adjusting the internal pressure of the airbag under the action of the air pump and valve control module. This allows the flexible airbag to apply dynamic support to the wearer's knee joint, thereby providing assistance during knee joint movement. When the internal pressure of the airbag increases, the flexible airbag expands and applies supporting pressure to the inner side of the knee brace, restraining the soft tissues around the knee joint and thus improving knee joint stability. The auxiliary force output by the flexible airbag is positively correlated with the internal pressure of the airbag.

[0256]

[0257] in: To provide auxiliary support; The internal pressure of the airbag; is the structural transfer factor.

[0258] During walking or climbing stairs, the pneumatic knee brace increases support pressure during knee flexion to reduce the load on the quadriceps. During descending stairs, it increases cushioning support pressure to reduce impact load on the knee joint. During running, it delivers enhanced assist pressure during knee extension to improve running stability. The assist pressure dynamically adjusts with the movement mode, ensuring the knee brace's support force changes synchronously with the wearer's knee joint movement. During mode switching, the internal pressure of the airbag adjusts gradually to avoid sudden changes in assist force affecting wearing comfort.

[0259]

[0260] in: This is the gradual adjustment coefficient.

[0261] Step seven of this invention includes a safety module that collects real-time data on the airbag's internal pressure, control execution status, and sensor operating status, and monitors the system's operating status according to a preset safety threshold. When an abnormal state is detected, protective control is executed. The safety module detects the airbag's internal pressure in real-time and determines whether the current pressure exceeds the maximum allowable pressure threshold. When the real-time pressure exceeds the maximum safe pressure threshold, the air pump is immediately shut down and the pressure relief solenoid valve is opened. Once the target pressure is established, if the real-time pressure remains below the target pressure threshold for a preset time, a system leak is determined. If the pressure sensor output value remains unchanged for multiple consecutive sampling periods, a sensor malfunction is determined, using the following formula:

[0262]

[0263] If the air pump runs continuously for more than a preset time threshold without reaching the target pressure, the air pump output will stop. If a serious anomaly is detected, rapid depressurization will be performed to restore the airbag to a safe pressure range. After the anomaly is resolved, the control module will re-enter standby mode and wait for the next motion mode recognition result.

[0264] While the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the invention. Those skilled in the art can make other changes within the spirit of the invention and apply it to fields not mentioned in the invention. Of course, all such changes made in accordance with the spirit of the invention should be included within the scope of protection claimed by the invention.

Claims

1. A control method for a wearable pneumatic knee-pad motion pattern recognition and assistance based on air pressure timing characteristics, characterized in that: Includes the following steps: Step 1: Collect the original air pressure sequence in the airbag during the wearer's exercise using an air pressure sensor; Step 2: Standardize the data using the data preprocessing module to obtain the air pressure sequence; Step 21: The median filtering algorithm is used to remove impulse noise and abnormal abrupt changes generated during the sampling process from the original air pressure sequence through the data preprocessing module; Step 22: Input the median-filtered air pressure sequence into the low-pass filter module to suppress high-frequency disturbances and preserve the air pressure change trend during human movement; Steps 2 and 3: Perform standardization processing on the low-pass filtered air pressure sequence to map the data in different sampling periods to a uniform numerical range, so as to serve as the air pressure time series input data for the subsequent motion pattern recognition model; Step 3: The barometric pressure sequence is fed into a deep learning model for inference to obtain motion pattern recognition results; The preprocessed barometric time series is input into a deep learning motion pattern recognition model. The deep learning motion pattern recognition model performs feature extraction and classification inference on the input sequence and outputs the motion pattern recognition result corresponding to the current wearer. The specific steps include: Step 31: The barometric pressure time series input to the deep learning model is used to construct input samples using a fixed-length sliding window. Let the input samples be represented as: wherein, indicates the time corresponding to the air pressure value, indicates the sequence length, and the input samples are input into the deep learning model in time sequence. The deep learning model first extracts local features from the input air pressure time series samples through multi-scale one-dimensional convolutional layers. The convolution operation expression is as follows: wherein the size of the convolution kernel , represents the weight parameter of the i-th position of the k-th scale convolution kernel, represents the bias parameter, represents the rectified linear unit, represents the sampling value of the input air pressure sequence within the corresponding receptive field range; Step 32: Input the output features of the convolutional layer into the temporal modeling layer to extract continuous temporal dependencies; The temporal modeling layer employs a temporal convolutional network, which extracts temporal variation features within a 1-second range at a sampling frequency of 100Hz through dilated convolution. The dilated convolution expression is as follows: coefficient of expansion The convolution kernel length is n, where n=3. The weight parameter representing the i-th position of the convolution kernel. This represents the sampled value of the previous convolutional layer's output within the current receptive field; Step 33: After the temporal modeling layer output is processed by the fully connected classification layer, the probability values ​​of each motion mode are output through the Softmax function. in: For the first Probability of motion pattern; The total number of sports mode categories, Represents the Euler number. , Represents the output of the fully connected layer; Steps 3 and 4: Select the category with the highest probability value as the current motion pattern recognition result: in: For the final identification category; Step 4: The auxiliary control module issues corresponding air pressure control commands based on the motion pattern recognition results; Step 5: Achieve precise control of airbag pressure through the air pump and valve control module; Step Six: The pneumatic knee brace actively responds to changes in movement patterns, achieving an assistive effect for the wearer during exercise; Step 7: The safety module monitors the knee brace system in real time to prevent safety issues caused by equipment malfunctions.

2. The control method for wearable pneumatic knee brace motion pattern recognition and assistance based on air pressure timing characteristics according to claim 1, characterized in that: The air pressure sensor mentioned in step one is an air pressure sensor installed inside the flexible airbag of the wearable pneumatic knee brace or at the air passage location connected to the flexible airbag. The air pressure sensor is used to detect the changes in internal pressure of the airbag caused by the wearer's knee joint during flexion, extension and load changes in real time. The pressure sensor converts the detected analog pressure signal into an electrical signal, and then converts it into digital pressure data through an analog-to-digital converter module. The main control unit continuously reads digital pressure data at a sampling frequency of 10Hz-200Hz and forms a continuous pressure change sequence arranged in time order through a transfer function; The continuous air pressure change sequence exhibits periodic dynamic changes as the wearer's movement changes, and serves as the data input for subsequent movement pattern recognition.

3. The control method for wearable pneumatic knee brace motion pattern recognition and assistance based on air pressure timing characteristics according to claim 2, characterized in that: The median filtering algorithm described in step two uses a sliding window of length 2k+1 to process the original air pressure sequence. Let the original air pressure sequence be: The filtering result for the i-th sampling point is defined as: In the formula, is the output value after median filtering; k is the half-width of the window.

4. The control method for wearable pneumatic knee brace motion pattern recognition and assistance based on air pressure timing characteristics according to claim 3, characterized in that: The low-pass filtering described in step two uses a digital low-pass filter to smooth the median-filtered air pressure sequence. Its recursive expression is: In the formula, Enter the current air pressure value; This is the current filter output value; This is the filtered output value from the previous moment; Let the filter coefficients satisfy: .

5. The control method for wearable pneumatic knee brace motion pattern recognition and assistance based on air pressure timing characteristics according to claim 4, characterized in that: The standardization process described in step two employs a maximum-minimum-value normalization method to map the low-pass filtered air pressure sequence to a uniform numerical range. Its expression is: in, The output value is the standardized value; This is the air pressure value after low-pass filtering; The minimum value within the current sample window; This represents the maximum value within the current sample window.

6. The control method for motion pattern recognition and assistance of wearable pneumatic knee brace based on air pressure timing characteristics according to claim 5, characterized in that: Step four also includes the auxiliary control module receiving the motion pattern recognition results output in step three, and calling the corresponding auxiliary control strategy according to different motion patterns to generate target air pressure control commands and send them to the air pump and valve control module.

7. The control method for wearable pneumatic knee brace motion pattern recognition and assistance based on air pressure timing characteristics according to claim 6, characterized in that: Step four's auxiliary control module establishes a mapping relationship between motion modes and target auxiliary pressures, with different motion modes corresponding to different target pressure ranges: When the motion pattern recognition result is walking on flat ground, the auxiliary control module outputs the first target pressure value; When the motion pattern recognition result is climbing stairs, the auxiliary control module outputs a second target pressure value that is higher than the first target pressure value; When the motion pattern recognition result is descending stairs, the auxiliary control module outputs a third target pressure value to mitigate the impact on the knee joint; When the motion pattern recognition result is sitting, the auxiliary control module outputs a fourth target pressure value that is lower than the first target pressure value. When the motion pattern is identified as running, the auxiliary control module outputs a momentarily increased fifth target pressure value.

8. The control method for wearable pneumatic knee brace motion pattern recognition and assistance based on air pressure timing characteristics according to claim 7, characterized in that: In step five, the air pump and valve control module receives the target pressure control command output by the auxiliary control module, and controls the air pump and solenoid valve to operate based on the deviation between the target pressure value and the current real-time pressure value inside the airbag, so as to achieve precise adjustment of the pressure inside the airbag. The adjustment process is as follows: The air pump and valve control module reads the current internal pressure value of the airbag in real time and calculates the pressure offset: in: The target pressure value output by the auxiliary control module; The current pressure value is collected in real time by the barometric pressure sensor; This represents the current pressure deviation. When the pressure deviation meets: At this time, the air pump and valve control module starts the miniature air pump and closes the pressure relief solenoid valve to inflate the airbag; wherein: The preset pressure tolerance threshold is used; When the pressure deviation meets: At this time, the air pump and valve control module shuts down the miniature air pump and opens the pressure relief solenoid valve to deflate the airbag; When the pressure deviation meets: At this time, the air pump and valve control module keeps the air pump off and the solenoid valve in its current state, so that the airbag maintains its current pressure.

9. The control method for wearable pneumatic knee brace motion pattern recognition and assistance based on air pressure timing characteristics according to claim 8, characterized in that: In step six, the pneumatic knee brace adjusts the internal pressure of the airbag under the action of the air pump and air valve control module, so that the flexible airbag applies dynamic support force to the wearer's knee joint, thereby providing assistance during knee joint movement. When the pressure inside the airbag increases, the flexible airbag expands and applies supporting pressure to the inside of the knee brace, which restrains the soft tissues around the knee joint, thereby improving knee joint stability. The auxiliary force output by the flexible airbag is positively correlated with the internal pressure of the airbag. in: To provide auxiliary support; The internal pressure of the airbag; The structural transfer factor; During walking or climbing stairs, the pneumatic knee brace increases support pressure during the knee flexion phase to reduce the output load on the quadriceps. During the downstairs exercise, the pneumatic knee brace increases the cushioning and support pressure to reduce the impact load on the knee joint; During running, the pneumatic knee brace delivers enhanced auxiliary pressure during the knee extension phase to improve running stability. The auxiliary pressure is dynamically adjusted according to the movement pattern, so that the knee brace outputs support force in sync with the wearer's knee joint movement state; During the switching of sports modes, the internal pressure of the airbag is adjusted gradually to avoid sudden changes in assistive force affecting wearing comfort. in: This is the gradual adjustment coefficient.

10. The control method for motion pattern recognition and assistance of wearable pneumatic knee brace based on air pressure timing characteristics according to claim 9, characterized in that: In step seven, the safety module collects real-time data on the airbag's internal pressure, control execution status, and sensor operating status, and monitors the system's operating status based on preset safety thresholds. When an abnormal state is detected, protective controls are executed. The safety module monitors the internal pressure of the airbag in real time and determines whether the current pressure exceeds the maximum allowable pressure threshold. When the real-time pressure exceeds the maximum safe pressure threshold, the air pump is immediately shut down and the pressure relief solenoid valve is opened. Once the target pressure is established, if the real-time pressure remains below the target pressure threshold for a preset period of time, the system is determined to have a leakage anomaly. If the pressure sensor output value remains unchanged for multiple consecutive sampling periods, the sensor is considered faulty. The determination formula is as follows: When the air pump runs continuously for more than the preset time threshold and fails to reach the target pressure, the air pump output will stop. If a serious anomaly is detected, rapid depressurization is performed to restore the airbag to a safe pressure range; After the anomaly is resolved, the control module re-enters standby mode and waits for the next motion pattern recognition result.

Images