Intelligent barbell strength training monitoring system and method based on pressure sensing

Abstract

The invention discloses an intelligent barbell strength training monitoring system and method based on pressure sensing, and belongs to the technical field of intelligent exercise monitoring. The method comprises the steps that a barbell sensing system matched with a pressure sensing glove and a body protection device is constructed; training basic parameters are obtained, a multi-sensing assembly is started to collect various pressure and attitude signals, and corrected pressure signals are obtained through filtering denoising and dynamic offset compensation; calculating the load weight, and fusing the multi-source signals to generate standardized training data; generating a visual interface containing a multi-dimensional curve and a structured report based on the standardized data; according to the method, wearable equipment physiological marker data, multi-sensor acquisition data and historical training records are fused, personalized training suggestions are generated through evaluation, meanwhile, protection early warning such as insufficient grip strength, pressure abnormity and posture imbalance is output, accurate monitoring, scientific evaluation and all-around safety protection of barbell strength training are achieved, and the training efficiency is improved. And professional training guidance and body protection are provided for trainees.

Description

Technical Field

[0001] This invention belongs to the field of intelligent motion monitoring technology, specifically an intelligent barbell strength training monitoring system and method based on pressure sensing. Background Technology

[0002] In the field of barbell strength training, accurate monitoring of training data, scientific evaluation of training effects, and prevention of sports injuries are core requirements. However, traditional monitoring methods rely heavily on manual recording of basic information such as load weight and repetitions. This is not only inefficient and prone to errors, but also fails to capture key data such as grip strength, body stress, and postural changes, making it difficult to support quantitative training analysis. While some existing barbell monitoring devices integrate sensors, they can only collect single barbell pressure signals, lacking the ability to monitor multi-dimensional factors such as pressure on the trainee's body contact points, posture, and hand grip strength. Furthermore, signal processing only performs simple filtering without accurately compensating for interference such as muscle tremors, resulting in insufficient data accuracy. Simultaneously, current technologies lack multi-source data fusion and evaluation capabilities, fail to combine physiological marker data with historical training records to generate personalized recommendations, and lack quantitative risk warning mechanisms. They cannot promptly identify problems such as insufficient grip strength and postural imbalances, easily leading to improper movements and sports injuries, and are no longer sufficient to meet the current needs for scientific and safe strength training. Summary of the Invention

[0003] To address the shortcomings of existing technologies, this invention proposes an intelligent barbell strength training monitoring system and method based on pressure sensing. The system constructs a barbell sensing system adapted to pressure-sensing gloves and body protection devices; acquires basic training parameters; activates multiple sensing components to collect various pressure and posture signals; obtains corrected pressure signals through filtering, noise reduction, and dynamic offset compensation; calculates the load weight; and fuses multi-source signals to generate standardized training data. Based on the standardized data, it generates a visual interface containing multi-dimensional curves and a structured report; integrates wearable device physiological marker data, multi-sensor data, and historical training records; and generates personalized training suggestions through evaluation. Simultaneously, it outputs protective warnings for insufficient grip strength, abnormal pressure, and postural imbalance, achieving accurate monitoring, scientific evaluation, and comprehensive safety protection for barbell strength training, providing trainees with professional training guidance and body protection.

[0004] To achieve the above objectives, the present invention provides the following technical solution:

[0005] A pressure-sensing-based intelligent barbell strength training monitoring method includes:

[0006] A barbell sensing system is constructed based on a calibrated pressure sensor and a wireless transmission module to generate pressure-weight conversion reference parameters and establish a stable communication link between the wireless transmission module and the terminal application. The barbell sensing system is adapted to the pressure-sensing gloves and body protection device worn by the trainee. The pressure-sensing gloves are used to collect hand grip pressure signals, and the body protection device is used to collect pressure signals and posture signals of the trainee's body contact points.

[0007] The training base parameters set by the trainee are obtained through the terminal application. The sensing components of the pressure sensor, pressure-sensitive gloves and body protection device are activated simultaneously to collect the original pressure electrical signal, hand grip pressure signal, body contact pressure signal and posture signal in real time. The signal is then filtered and denoised and dynamic offset compensation is performed to obtain the corrected pressure signal and posture signal.

[0008] The load weight is calculated based on the corrected original pressure electrical signal and pressure-weight conversion reference parameters. The corrected pressure signal and attitude signal are fused together, and the first dynamic feature and the second dynamic feature are extracted according to the load weight. The data is then transmitted to the terminal application in real time and converted into standardized training data.

[0009] Based on standardized training data and basic training parameters, a visual interface and structured training report are generated, including strength-time curves, grip strength change curves, and body stress / posture change curves.

[0010] The system calls upon physiological marker data acquired by wearable devices and integrates structured training reports, physiological marker data, data collected from pressure-sensitive gloves and body protection devices, as well as historical training records stored locally on the terminal application for evaluation. It then generates strength training suggestions and risk warnings and provides them back to the trainee. The risk warnings include protective warnings for insufficient hand grip strength, abnormal pressure at body contact points, and postural imbalance.

[0011] Specifically, the construction of the barbell sensing system includes:

[0012] The pressure sensor is encapsulated in a waterproof and shockproof housing and fixedly installed below the center grip area of ​​the barbell bar to collect the raw pressure electrical signal generated by the applied load in real time.

[0013] The wireless transmission module is integrated into the end weight clamp of the barbell bar and electrically connected to the pressure sensor through a built-in flexible circuit board. It is used to receive pressure electrical signals, hand grip pressure signals, body contact pressure signals and posture signals, and package the signals with corresponding timestamps and send them to the terminal application through its Bluetooth Low Energy protocol.

[0014] After constructing the barbell sensing system, a calibration process is performed: multiple weights of known standard weights are loaded onto the barbell bar in sequence, and the stable pressure electrical signal value output by the pressure sensor at each weight is recorded. The least squares method is used to linearly fit multiple sets of weight-signal value data to generate pressure-weight conversion reference parameters in the form of slope and intercept.

[0015] Zero-point calibration was performed on the grip force sensing components of the pressure-sensitive gloves and the pressure and posture sensing components of the body protection device, and the reference signal value under no external force was recorded.

[0016] The Bluetooth Low Energy protocol is used to establish pairing connections between terminal applications and wireless transmission modules, pressure-sensitive gloves, and body protection devices.

[0017] Specifically, the process of acquiring the corrected pressure signal includes:

[0018] The received original pressure electrical signal with timestamps, hand grip pressure signal, and body contact pressure signal are respectively input into a digital Butterworth band-stop filter with a preset cutoff frequency to obtain various types of denoised pressure signal sequences; the preset cutoff frequency is set according to the typical frequency range of human muscle tremors.

[0019] The static reference segment in each type of denoised pressure signal sequence is identified, and the mean value of the calculated static reference segment is used as the dynamic offset reference value of each type of signal; the static reference segment corresponds to the signal range when the barbell is placed statically on the support, the trainee holds it naturally, and the body protection device has no contact pressure.

[0020] The corrected pressure signal is obtained by subtracting the dynamic offset reference value from each data point in the denoised pressure signal sequence of each type in real time.

[0021] Specifically, the calculation of the load weight includes:

[0022] Based on the slope and intercept in the pressure-weight conversion reference parameters, the amplitude value of the corrected pressure signal is converted into instantaneous load weight.

[0023] The load weight of the current action is calculated by weighting multiple instantaneous load weights within the same action cycle, and the load weight is associated with its corresponding timestamp to generate a load weight time series.

[0024] Specifically, the extraction of the first dynamic feature includes:

[0025] Based on the load weight time series, the start and end points of each repetitive action are identified by finding local minimum points, and the action interval is verified by combining the abrupt change characteristics of the attitude signal, thus dividing the single repetitive action interval.

[0026] Within each repetitive motion interval, the maximum value of the load weight is extracted as the peak force, and the minimum value of the load weight is extracted as the valley force. The difference between the peak force and the valley force is calculated as the net output force of that repetitive motion.

[0027] The peak force, trough force, and net output force of each repetitive motion, along with supplementary indicators such as hand grip strength and pressure at the body contact points, are used as the first dynamic feature set.

[0028] Specifically, extracting the second dynamic feature includes:

[0029] Based on the load weight time series, the first derivative of the load weight change over time is calculated to obtain the force change rate series;

[0030] Within the single repetitive action interval, the positive maximum rate of change is extracted from the force change rate sequence as the explosive force index of the centripetal phase, and the negative maximum rate of change is extracted as the explosive force index of the centrifugal phase.

[0031] Based on the load weight time series, the duration during which the load weight exceeds a preset threshold weight within a single repetitive movement interval is calculated as the muscle tension time.

[0032] The peak value of the rate of change of hand grip strength, the offset angle and the offset rate of body posture are calculated simultaneously. When the posture offset angle exceeds the preset safety threshold, it is marked as a posture imbalance feature.

[0033] The explosive power index during the concentric phase, the explosive power index during the eccentric phase, the duration of muscle tension, and the characteristics of postural imbalance were used as the second set of dynamic features.

[0034] Specifically, the evaluation process involves calling upon physiological biomarker data acquired by wearable devices and integrating structured training reports, physiological biomarker data, and historical training records stored locally on the terminal application. This evaluation is implemented using a deep neural network model. The deep neural network model includes a multilayer perceptron branch for receiving the first and second dynamic features from the structured training report, a one-dimensional convolutional neural network branch for processing the physiological biomarker data, a convolutional recurrent neural network branch for processing multi-dimensional sensor data collected by pressure-sensitive gloves and body protection devices, and a long short-term memory network branch for processing long-term trends in historical training records. The physiological biomarker data includes heart rate and heart rate variability.

[0035] The multilayer perceptron branch contains three fully connected layers. The first layer maps the input dynamic features to a 64-dimensional space and uses the ReLU activation function. The second layer reduces the dimensions to 32, and the third layer reduces the dimensions to 16.

[0036] The one-dimensional convolutional neural network branch contains two convolutional layers. The first layer uses 16 convolutional kernels of length 3, and the second layer uses 32 convolutional kernels of length 3. Each convolutional layer is followed by a max pooling layer, and finally flattened into a one-dimensional vector.

[0037] The convolutional recurrent neural network branch performs convolutional feature extraction and recurrent temporal analysis on the time series data of grip force, body pressure, and posture. The number of hidden layer units is 64, and the output is a multidimensional feature vector.

[0038] The Long Short-Term Memory (LSTM) network branch processes the key indicator sequences of the most recent 10 training sessions, and its hidden layer unit number is 32.

[0039] The output vectors of the multilayer perceptron branch, the one-dimensional convolutional neural network branch, the convolutional recurrent neural network branch, and the long short-term memory network branch are concatenated in the concatenation layer. Through a fusion network containing two fully connected layers, the final output layer uses the sigmoid activation function to generate a scalar value representing the probability of risk warning and a multidimensional vector containing strength training suggestions.

[0040] Specifically, the method also includes a compensation mechanism based on environmental parameters:

[0041] In the barbell sensing system, a temperature sensor is integrated in the same position as the pressure sensor. Miniature temperature sensors are also integrated next to the sensing components of the pressure-sensing gloves and body protection devices to collect real-time temperature data of the environment in contact with the barbell bar, hands, and body parts.

[0042] Temperature data is sent to the terminal application via the stable communication link.

[0043] In the terminal application, a pre-stored temperature-pressure sensitivity coefficient lookup table is invoked. For the received real-time temperature data, the corresponding compensation coefficient is obtained from the temperature-pressure sensitivity coefficient lookup table to perform temperature drift compensation on the corrected pressure signal, thereby obtaining the temperature-compensated pressure electrical signal. The temperature-pressure sensitivity coefficient lookup table records the zero-point drift and sensitivity change of various pressure sensors at different temperatures.

[0044] The load weight is recalculated based on the temperature-compensated pressure signal.

[0045] Specifically, the method also includes training action norm recognition:

[0046] Predefined standard movement template features; the standard movement template features include standardized load weight time series, force change rate series, grip strength change series, and body stress / posture change series for standard squat, bench press, and deadlift movements.

[0047] During real-time training, the load weight time series, force change rate series, grip strength change series, and body pressure / posture change series within the current single repetitive movement interval are extracted as features to be compared. The dynamic time warping algorithm is used to calculate the minimum alignment path distance between the features to be compared and the standard movement template features to obtain the movement similarity score.

[0048] If the similarity score of the action is lower than the preset standard threshold, or if the changes in grip strength or body pressure / posture deviate from the standard template beyond the preset safety range, then a non-standard action marker and a protective warning will be added to the generated structured training report. When generating strength training suggestions, priority will be given to correcting the form of the action, and specific protective operation suggestions for adjusting hand grip strength and correcting body posture will be given.

[0049] A pressure-sensing-based intelligent barbell strength training monitoring system includes: a barbell sensing module, a pressure-sensing glove, a body protection device, a terminal application module, and an assessment module.

[0050] The barbell sensing module is used to collect raw pressure signals and ambient temperature signals in real time during training, complete the initial signal encapsulation and wireless transmission, and generate a pressure-weight conversion benchmark through a calibration process.

[0051] The pressure-sensing glove is a protective sensing device worn by the trainee's hand. It has a built-in grip force sensing component and a miniature temperature sensor to collect hand grip force pressure signals and hand ambient temperature signals in real time. After preliminary signal processing, the signals are wirelessly transmitted to the terminal application.

[0052] The body protection device is a protective sensing device worn by the trainee on the body contact area. It has built-in pressure sensing components, posture sensing components and miniature temperature sensors to collect pressure signals, training posture signals and ambient temperature signals of the body contact area in real time. After completing the initial signal processing, it is wirelessly transmitted to the terminal application, and at the same time, it realizes the basic protection of the body contact area by buffering and impact resistance.

[0053] The terminal application module serves as the interactive entry point for trainees and the data storage and display center. It is used to receive parameters set by trainees, receive data transmitted by hardware, store key information, generate visual interfaces and reports, and provide training suggestions and risk warnings to trainees.

[0054] The evaluation module is used to purify, convert, and extract features from the collected raw data, and integrate multi-source data from the barbell sensing module, pressure-sensing glove module, body protection device module, and wearable device for intelligent evaluation. At the same time, it identifies the standardization of training movements and generates protective warnings.

[0055] Compared with the prior art, the beneficial effects of the present invention are:

[0056] 1. This invention proposes an intelligent barbell strength training monitoring system based on pressure sensing, and optimizes and improves its architecture, operation steps and processes. The system has the advantages of simple process, low investment and operating costs and low production costs.

[0057] 2. This invention proposes an intelligent barbell strength training monitoring method based on pressure sensing. By constructing a barbell sensing system adapted to pressure-sensing gloves and body protection devices, and combining signal processing methods such as filtering and noise reduction and dynamic offset compensation, it achieves multi-dimensional and accurate acquisition of load weight, hand grip strength, body pressure, and posture during barbell training. At the same time, it extracts multiple types of dynamic features to generate standardized data, and through a visualization interface of multi-dimensional curves and structured reports, trainees can intuitively and comprehensively grasp their own training status. This breaks through the limitations of traditional monitoring methods that are singular and have vague data, and provides accurate and detailed quantitative data for strength training.

[0058] 3. This invention proposes a smart barbell strength training monitoring method based on pressure sensing. It integrates physiological marker data from wearable devices, multi-sensor data, and historical training records for comprehensive evaluation. This method can generate personalized strength training suggestions tailored to the trainee's actual situation, helping to optimize training programs and improve training results. It can also provide targeted warnings for insufficient grip strength, abnormal body pressure, and postural imbalance, promptly reminding trainees to avoid the risk of sports injuries caused by improper movements and inappropriate physiological loads. This achieves the dual goals of scientific guidance for strength training and comprehensive safety protection. Attached Figure Description

[0059] Figure 1 This is a schematic diagram of an intelligent barbell strength training monitoring method based on pressure sensing according to the present invention.

[0060] Figure 2 This is an architecture diagram of an intelligent barbell strength training monitoring system based on pressure sensing, according to the present invention. Detailed Implementation

[0061] Example 1

[0062] This embodiment uses a typical barbell squat training exercise as an example to illustrate the complete implementation process of this method in a real-world scenario. The trainee is a male fitness enthusiast weighing 80 kg, whose training goal is muscle hypertrophy. The training location is a home gym with an ambient temperature of approximately 23 degrees Celsius and moderate humidity. The trainee uses a standard Olympic barbell with symmetrical 20 kg weight plates at both ends, for a total barbell load of 60 kg. The trainee wears a wristband with heart rate monitoring, pressure-sensitive gloves, and body protection devices for the waist and knees. The trainee's mobile phone has the corresponding terminal application installed.

[0063] Please see Figure 1 The present invention provides an embodiment of an intelligent barbell strength training monitoring method based on pressure sensing, the method comprising steps S1 to S5:

[0064] S1: Construct a barbell sensing system based on a calibrated pressure sensor and a wireless transmission module, generate pressure-weight conversion reference parameters, and establish a stable communication link between the wireless transmission module and the terminal application; the barbell sensing system is adapted to the pressure-sensing gloves and body protection device worn by the trainee, the pressure-sensing gloves are used to collect hand grip pressure signals, and the body protection device is used to collect pressure signals and posture signals of the trainee's body contact points.

[0065] In this embodiment, a high-precision resistance strain gauge pressure sensor is selected, with a range of 0-500 kg, a full-scale output of 2 mV / V, and a nonlinearity error of less than 0.05%. To adapt to the sweat, vibration, and other environmental factors that may exist in a gym, the sensor is encapsulated in a cylindrical aluminum alloy shell. The shell is filled with silicone for waterproofing and shockproofing, and the surface of the shell is anodized to increase wear resistance. This encapsulation is securely installed on the lower middle part of the barbell bar using high-strength double-sided tape and matching clamps, that is, directly below the area where the trainee's hands normally grip, ensuring that the sensor can directly sense the pressure changes caused by the loaded weight and the force applied by the trainee. The wireless transmission module uses a low-power Bluetooth chip, version 5.0, which supports the Bluetooth Low Energy protocol. The wireless transmission module is integrated into a flat plastic box, the size of which is designed to fit perfectly into the internal cavity of the metal weight clamp at the end of the barbell. The wireless transmission module is connected to the pressure sensor via a flexible printed circuit board approximately 30 cm long. The flexible circuit board is laid along the surface of the barbell and secured with nylon cable ties to prevent it from loosening due to shaking during training. The wireless transmission module has a built-in rechargeable lithium polymer battery with a capacity of 500 mAh, which can be charged via a magnetic interface. A single charge can support approximately 40 hours of continuous operation. After being powered on, the wireless transmission module enters broadcast mode, waiting to pair with the terminal application.The calibration process is as follows: First, the barbell pressure sensor is calibrated. Before training begins, the barbell is placed horizontally on the support frame of the squat rack. The calibration mode is activated via the terminal application. The trainee loads standard calibration weights of known weights onto both ends of the barbell in sequence. In this example, 20 kg, 40 kg, 60 kg, 80 kg, and 100 kg are loaded onto the empty bar in sequence. After each weight is loaded, the pressure sensor output signal is allowed to stabilize for approximately 5 seconds. The terminal application records the raw value of the stable pressure electrical signal corresponding to that weight, in the form of an analog-to-digital converter reading. After loading is complete, the terminal application obtains five sets of weight-signal value data pairs. The least squares method is used to perform linear fitting on these data, generating a fitted straight line with the signal value on the x-axis and the weight on the y-axis. The slope of this straight line represents the pressure change per unit weight. The weight change corresponding to each unit, and the intercept corresponding to the theoretical zero-point output of the sensor under no-load conditions, together constitute the pressure-weight conversion reference parameters and are stored in the local configuration file of the terminal application; the second is pressure-sensing glove calibration, where the trainee wears gloves naturally, keeping the hands in a state of no external force grip, and initiates zero-point calibration of the glove grip force sensing component through the terminal application, recording the reference signal value under no external force as the reference for grip force pressure signal correction; the third is body protection device calibration, where the trainee wears the body protection device to the designated parts of the waist and knee joints, keeping the body in a natural, relaxed, and non-compression state, and initiates zero-point calibration of the device's pressure sensing component and posture sensing component through the terminal application, recording the reference signal values ​​under no contact pressure and no posture deviation, respectively, and storing them in the terminal application as correction references. Finally, a stable communication link is established. The trainee opens the terminal application on their mobile phone. The application automatically scans for nearby low-power Bluetooth devices and identifies a device named "SmartBarbell_01", a pressure-sensing glove named "Glove_Sensor_01", and a body protection device named "Body_Protect_01" in the device list. After clicking to connect in sequence, the pairing and binding of all devices is completed. After successful pairing, the application interface displays that the devices are connected and the signal strength is good. At this point, a smart barbell sensing system including a calibrated pressure sensor, a wireless transmission module, a pressure-sensing glove, and a body protection device is completed, and a stable data communication link with low power consumption and low latency is established with the user terminal.

[0066] S2: The training base parameters set by the trainee are obtained through the terminal application. The sensing components of the pressure sensor, pressure-sensing gloves and body protection device are activated simultaneously to collect the original pressure electrical signal, hand grip pressure signal, body contact part pressure signal and posture signal in real time. The signal is then filtered and denoised and dynamic offset compensation is performed to obtain the corrected pressure signal and posture signal.

[0067] Furthermore, the basic training parameters include training movements, training goals, planned number of sets, planned number of repetitions, and preset load weights. Among these, the training goals include at least one of the following: muscle strength increase, muscle hypertrophy, and explosive power improvement.

[0068] Furthermore, based on the selected training goal, the system automatically sets a set of matching monitoring parameters: when muscle strength growth is selected, the system uses the maintenance and progress of peak strength as the core monitoring indicator; when muscle hypertrophy is selected, the system uses muscle tension time and total training volume as the core monitoring indicators; when explosive power improvement is selected, the system uses the concentric phase explosive power indicator as the core monitoring indicator. When generating the strength training suggestions, the system will prioritize comparing the completion status of the core monitoring indicators corresponding to the selected training goal with historical data to provide personalized suggestions for that goal, rather than general suggestions.

[0069] In this embodiment, when the trainee is ready to start training, a new training session is first created on the main interface of the terminal application. The system pops up a parameter setting page, where the trainee needs to set the basic parameters for this training session and select squat as the training exercise. Next, in the training goal options, among the three options of "muscle strength increase," "muscle hypertrophy," and "explosive power improvement," "muscle hypertrophy" is selected. When the training goal is muscle hypertrophy, the core monitoring indicators inside the system will be automatically set to muscle tension time and total training volume. This means that in data processing and evaluation, the system will pay special attention to the time the muscles are under load and the total amount of work done in this training session, and provide personalized suggestions based on the completion of these indicators. Then, the trainee enters the planned number of sets for this training session as five sets, with ten repetitions per set. The trainee can also optionally enter a preset load weight of 60 kg for this training session, but this weight is for reference only, and the actual load of the system will be based on the transmission. Based on sensor measurements, after all parameters are set, the trainee clicks the "Start Training" button. At this time, the terminal application immediately sends a start command to the barbell sensing system via a stable Bluetooth link. After receiving the start command, the wireless transmission module wakes up the pressure sensor, enabling it to start working at a sampling frequency of 100 Hz. All sensing components start working synchronously. The pressure sensor collects raw pressure electrical signals at a sampling frequency of 100 Hz, the pressure-sensing glove collects hand grip pressure signals in real time, and the body protection device simultaneously collects pressure signals from the waist and knee joint contact points, as well as body posture angle signals. At the same time, the high-precision real-time clock inside the wireless transmission module timestamps each collected pressure signal sample with millisecond accuracy. These timestamped pressure signal data packets are instantly sent to the terminal application via Bluetooth Low Energy protocol. The background service of the terminal application continues to run, receiving and caching these continuous streams of raw data, waiting for further processing.

[0070] S3: Calculate the load weight based on the corrected original pressure electrical signal and pressure-weight conversion reference parameters, extract the first dynamic feature and the second dynamic feature according to the load weight, and transmit them to the terminal application in real time to generate standardized training data.

[0071] S4: Based on standardized training data and basic training parameters, generate a visual interface and structured training report that includes strength-time curves, grip strength change curves, and body stress / posture change curves.

[0072] Furthermore, the specific implementation method for generating the visualization interface containing the force-time curve is as follows:

[0073] (1) On the graphical interface of the terminal application, the load weight time series is plotted in real time with time as the horizontal axis and load weight as the vertical axis to form the main force-time curve;

[0074] (2) In the same chart, the force change rate sequence, grip force change curve and body contact pressure change curve are displayed in superimposed with different colors / line types to form a multi-dimensional auxiliary analysis curve. At the same time, the change of body posture offset angle is displayed in the form of a combination of numerical values ​​and curves.

[0075] (3) Provide interactive functions for each identified repetitive action interval. When the user clicks or hovers over an interval, a floating window pops up on the interface to dynamically display the detailed values ​​of the first and second dynamic features of the repetition.

[0076] (4) Provide zoom and smoothing controls to allow users to zoom in on the force-time curve locally and apply smoothing algorithms of different intensities to observe the detailed trends of force changes more clearly.

[0077] In this embodiment, the terminal application uses standardized training data to generate and update the user interface on the mobile phone screen in real time, specifically including:

[0078] On the main training interface, a dynamically updated chart occupies the primary area. The horizontal axis represents time, and the vertical axis represents the load weight. As training progresses, a curve is plotted in real time—this is the main strength-time curve. It visually displays the waveform of the load weight fluctuating over time throughout the training process, with each peak and trough corresponding to one repetition. Within the same chart, the system uses a dashed line, overlaid with a different color, to display the strength change rate sequence, forming an auxiliary analysis curve. This curve helps trainees intuitively understand their exertion rhythm, grip strength changes, knee / lower back pressure, and postural deviations. After a training session, each repetition interval is clearly marked on the chart. Trainees can click on an interval, such as the one where they felt the exertion was not smooth. A floating window will pop up, dynamically displaying detailed data for that repetition: peak strength 85 kg, trough strength 20 kg, net output strength 65 kg, concentric explosive force index 320 kg / s, eccentric explosive force index -280 kg / s, muscle tension time 1.2 seconds, etc. This provides data support for trainees to conduct immediate technical review. In addition, the interface provides zoom and smoothing controls. If trainees find the curve fluctuations too fragmented, they can apply a mild smoothing algorithm to make the overall trend clearer. If they want to observe the subtle differences between any two movements, they can zoom in on a specific area of ​​the chart using gestures. After the trainee completes all five sets of squats and clicks "End Training," the system automatically generates a structured training report for this session. This report includes not only all the raw data and extracted features but also a summary analysis at the session level.

[0079] Specifically, the structured training report includes:

[0080] Training session metadata: basic information such as training date, start / end time, total training duration, training actions, training objectives, load used, number of sets / attacks completed, etc.

[0081] Detailed data for each repetition: Records the full dynamic characteristics of each repetition in chronological order of timestamps, including load weight, peak / trough force, net output force, concentric / eccentric explosive force, muscle tension time, grip strength index, body stress / posture index, etc.

[0082] Session summary analysis: Calculate the average peak strength, average muscle tension time, and total training volume for all repetitions; count the number of postural imbalances, insufficient grip strength, and abnormal body stress; mark the intervals of non-standard movements; and initially record potential risk points.

[0083] S5: Call upon the physiological marker data obtained by the wearable device, and integrate the data collected from the structured training report, physiological marker data, pressure-sensing gloves and body protection device, as well as the historical training records stored locally on the terminal application for evaluation, generate strength training suggestions and risk warnings and provide feedback to the trainee. The risk warnings include protective warnings for insufficient hand grip strength, abnormal pressure at body contact points, and postural imbalance.

[0084] Furthermore, the storage format and synchronization mechanism for updating and storing the structured training report in the historical training record are as follows:

[0085] (1) The structured training report is stored in JSON format. The top-level fields include training session metadata, data arrays for each repetition, and evaluation summary of the session. The training session metadata includes date, duration, and training type. The data array includes timestamp, load weight, first and second dynamic features. The evaluation summary includes average peak force, total number of repetitions, and risk warning records.

[0086] (2) After each training session, the terminal application first encrypts the complete structured training report and stores it in the local SQLite database;

[0087] (3) When the terminal application detects an available Wi-Fi network, it automatically synchronizes the newly added training report locally to the cloud server. The cloud server allocates independent data storage space for each user and maintains data version management to ensure that historical training records can be fully restored when the terminal device is changed.

[0088] In this embodiment, during the training process, the smart bracelet worn by the trainee is synchronized with the mobile terminal application via Bluetooth, providing real-time physiological marker data such as heart rate and heart rate variability. After the training is completed, the terminal application automatically integrates the following multi-source data: the structured training report for this session, the time-series data of physiological markers throughout the training, the full collection data of the pressure-sensing gloves / body protection device, and the historical training records of the trainee's 10 most recent squat training sessions stored locally in the terminal application.

[0089] In this embodiment, the evaluation process is implemented using a deep neural network model. This deep neural network model includes a multilayer perceptron branch for receiving the first and second dynamic features from the structured training report, a one-dimensional convolutional neural network branch for processing physiological biomarker data, a convolutional recurrent neural network branch for processing multi-dimensional sensor data collected by pressure-sensitive gloves and body protection devices, and a long short-term memory network branch for processing long-term trends in historical training records. The four branches are in parallel architecture to process different types of data. The physiological biomarker data includes heart rate and heart rate variability.

[0090] The first is the multilayer perceptron branch, which is used to process the structured features of this training. It receives all the first and second dynamic features in this report as input. The multilayer perceptron branch contains three fully connected layers: the first layer maps the input dynamic features to a 64-dimensional space and introduces non-linearity using the ReLU activation function; the second layer compresses the dimension to 32 dimensions; the third layer further compresses it to 16 dimensions, and finally outputs a 16-dimensional vector, which can be regarded as a highly abstract representation of the action pattern of this training.

[0091] The second branch is a one-dimensional convolutional neural network (CNN) used to process physiological time-series data. It takes the sequence of heart rate and heart rate variability changes over time during the training process as input. The one-dimensional CNN branch contains two convolutional layers: the first layer uses 16 convolutional kernels of length 3 to slide and extract local features over time; the second layer uses 32 convolutional kernels of length 3 to extract deeper features. Each convolutional layer is followed by a max pooling layer to reduce the data dimensionality and retain significant features. Finally, the output is flattened into a one-dimensional vector.

[0092] The third is the convolutional recurrent neural network branch, which performs convolutional feature extraction and recurrent temporal analysis on the time series data of grip force, body pressure, and posture. The number of hidden layer units is 64, and the output is a multidimensional feature vector.

[0093] The third branch is the Long Short-Term Memory (LSTM) network, which is used to analyze historical trends and read the key indicator sequence of the user's most recent 10 squat training sessions, such as the average peak strength and average muscle tension time of each training session. The LTM network has memory units and is good at processing this type of data with time-series dependencies. Its hidden layer unit number is set to 32, which can capture long-term change patterns in the user's strength level, endurance, etc.

[0094] The output vectors from the multilayer perceptron branch, the one-dimensional convolutional neural network branch, the convolutional recurrent neural network branch, and the long short-term memory network branch are fed into a concatenation layer and fused into a comprehensive feature vector. This comprehensive vector is then further integrated and analyzed through a fusion network containing two fully connected layers. Finally, the output layer uses the sigmoid activation function to generate two key outputs: a scalar value representing the probability of over-fatigue, action risk, etc., in this training; and a multi-dimensional vector whose dimensions correspond to different training suggestion directions, and the value of the vector represents the strength or relevance of each suggestion.

[0095] Furthermore, based on the output of the neural network and combined with the training objectives, the system generates final strength training suggestions and risk warnings, which are then intuitively fed back to the trainee through the terminal application interface.

[0096] In this embodiment, the model analysis showed that: the average muscle tension time in this training session was 1.3 seconds, which is better than the historical average of 1.1 seconds; the total training volume met the target; the heart rate zone was within the effective stimulation range for muscle hypertrophy; the grip strength was generally stable, but there was a slight decrease in the last set; the lower back pressure was normal, but the knee joint experienced abnormal pressure in the last two sets; and there were 3 instances of body posture deviation during squats, such as an angle of 18°, which exceeded the 15° safety threshold. At the same time, the heart rate variability data showed a significant increase in fatigue in the later stages of training.

[0097] The risk warning generated based on this is:

[0098] Posture imbalance warning: The last two sets of squats were detected to have 3 instances of body posture deviation exceeding the safety threshold, accompanied by abnormal knee joint pressure, indicating a potential risk of joint damage.

[0099] Muscle fatigue warning: Heart rate variability data shows a significant increase in fatigue in the later stages of training, with the average rate of decline during the eccentric phase exceeding the safe threshold, indicating insufficient control during the eccentric phase. It is recommended to stop high-load training immediately.

[0100] Grip strength tip: A slight decrease in hand grip strength was observed in the last set. It is recommended to use grip strength assistive devices in subsequent training to avoid incorrect movements due to insufficient grip strength.

[0101] Based on this, the personalized training recommendations are as follows: This squat workout significantly stimulated muscle hypertrophy, provided sufficient time for muscle tension, and achieved the target training volume. It is recommended to allow 48-72 hours of full recovery, focusing on knee relaxation and stretching, and increasing protein intake to promote muscle synthesis. For the next workout, the load can be slightly increased to 62.5 kg. During the workout, focus on posture control, and reduce the number of repetitions per set to 8, prioritizing proper form. If necessary, use knee braces and grip strength enhancement devices.

[0102] Furthermore, the structured training report is encrypted and stored in the local SQLite database of the terminal application in JSON format. The top-level fields include training session metadata, single and repeated data arrays, evaluation summary, and risk warning records. When the terminal application detects an available Wi-Fi network, it automatically synchronizes the newly added training report to the cloud server. The cloud allocates independent data storage space for each user and maintains data version management to ensure that trainees can fully restore historical training records when changing terminal devices, thus ensuring the continuity and long-term value of the data.

[0103] The method also includes a compensation mechanism based on environmental parameters:

[0104] A1: In the barbell sensing system, a temperature sensor is integrated in the same position as the pressure sensor. Miniature temperature sensors are integrated next to the sensing components of the pressure-sensing gloves and body protection devices to collect real-time temperature data of the environment in contact with the barbell bar, hands, and body parts.

[0105] A2: The temperature data is sent to the terminal application via the stable communication link;

[0106] A3: In the terminal application, the pre-stored temperature-pressure sensitivity coefficient lookup table is called. For the received real-time temperature data, the corresponding compensation coefficient is obtained from the temperature-pressure sensitivity coefficient lookup table to perform temperature drift compensation on the corrected pressure signal, and the temperature-compensated pressure electrical signal is obtained. The temperature-pressure sensitivity coefficient lookup table records the zero-point drift and sensitivity change of various pressure sensors at different temperatures.

[0107] A4: Recalculate the load weight based on the pressure signal after temperature compensation.

[0108] In this embodiment, during the half hour that the trainee is squatting, the local temperature of the barbell may slowly rise from 23 degrees Celsius to 24 degrees Celsius due to the grip of the hands and the gentle breeze from the indoor air conditioning. The temperature of the hands and knees also fluctuates slightly. Each temperature sensor collects and transmits temperature data in real time, and the terminal application automatically completes temperature drift compensation for all sensor signals.

[0109] The method also includes training action norm recognition:

[0110] B1: Predefined standard movement template features; the standard movement template features include standardized load weight time series, force change rate series, grip strength change series, and body stress / posture change series of standard squat, bench press, and deadlift movements. This template is based on a large amount of standard movement data from professional athletes.

[0111] B2: During real-time training, the load weight time series, force change rate series, grip strength change series, and body pressure / posture change series within the current single repetitive action interval are extracted as features to be compared. The dynamic time warping algorithm is used to calculate the minimum alignment path distance between the features to be compared and the standard action template features to obtain the action similarity score. The dynamic time warping algorithm is existing technology in this field and is not an inventive solution of this application, so it will not be described in detail here.

[0112] B3: If the similarity score of the action is lower than the preset standard threshold (85 points in this embodiment), or if the change in grip strength or body pressure / posture deviates from the standard template by more than the preset safety range, then an action non-standard mark and protective warning will be added to the generated structured training report. When generating strength training suggestions, priority will be given to correcting the form of the action, and specific protective operation suggestions for adjusting hand grip strength and correcting body posture will be given.

[0113] In this embodiment, before training begins, the trainee selects "squat" as the training exercise in the application. The system then loads a predefined standard squat template feature, which is an idealized load weight time series and force change rate sequence based on a large amount of data from professional athletes. Each time the trainee repeats the exercise, the system extracts the comparison features of that exercise and uses a dynamic time warping algorithm to elastically compare them with the standard template, calculating an exercise similarity score. In this embodiment, the trainee's first few sets of exercises were standard, with scores all above 90 points. However, in the last set at exhaustion, a slight forward lean and knee valgus occurred, causing abnormal fluctuations in the force curve and knee joint pressure curve. The dynamic time warping algorithm calculated that the similarity score for this exercise was only 75 points, lower than the system's preset standard threshold of 85 points. Therefore, in the generated structured training report, this repetition was marked as an exercise non-standard. Meanwhile, in the generated strength training suggestions, the system prioritized adding corrective suggestions such as "Please pay attention to the stability of the movement when the last set is exhausted. Leaning forward can easily lead to excessive stress on the lower back, and knee valgus can increase the risk of joint injury. It is recommended to prioritize the form of the movement during training. You can appropriately reduce the number of repetitions per set or use knee braces. When exerting force, pay attention to tightening your core to keep your body upright, and exert force along the direction of your toes to avoid knee valgus."

[0114] The construction of the barbell sensing system includes:

[0115] S1.1: The pressure sensor is encapsulated in a waterproof and shockproof housing and fixedly installed below the center grip area of ​​the barbell bar for real-time acquisition of the original pressure electrical signal generated by the applied load;

[0116] S1.2: The wireless transmission module is integrated into the end counterweight clamp of the barbell bar and electrically connected to the pressure sensor through the built-in flexible circuit board. It is used to receive pressure electrical signals, hand grip pressure signals, body contact pressure signals and posture signals, and package the signals with the corresponding timestamps through its Bluetooth Low Energy protocol and send them to the terminal application.

[0117] S1.3: After constructing the barbell sensing system, perform the calibration process: load multiple weights of known standard weights onto the barbell bar in sequence, record the stable pressure electrical signal value output by the pressure sensor under each weight, and use the least squares method to linearly fit multiple sets of weight-signal value data to generate pressure-weight conversion benchmark parameters in the form of slope and intercept.

[0118] S1.4: Perform zero-point calibration on the grip force sensing component of the pressure-sensitive glove and the pressure and posture sensing component of the body protection device, and record the reference signal value under no external force.

[0119] S1.5: Establishes pairing connections between terminal applications and wireless transmission modules, pressure-sensing gloves, and body protection devices based on the Bluetooth Low Energy protocol.

[0120] The process of acquiring the corrected pressure signal includes:

[0121] S3.1: The received original pressure electrical signal with timestamps, hand grip pressure signal, and body contact pressure signal are respectively input to a digital Butterworth band-stop filter with a preset cutoff frequency to obtain various types of denoised pressure signal sequences; the preset cutoff frequency is set according to the typical frequency range of human muscle tremors. The digital Butterworth band-stop filter is prior art in this field and is not an inventive solution of this application, so it will not be described in detail here.

[0122] In this embodiment, the design of the digital Butterworth band-stop filter is based on the fact that during strength training, in addition to active exertion, there may be involuntary micro-twitches in the muscles, with a typical frequency range between 8 Hz and 12 Hz. Therefore, the filter uses this frequency range as the stopband to effectively filter out noise interference in this frequency band and obtain a relatively smooth denoised pressure signal sequence.

[0123] S3.2: Identify the static reference segment in each type of denoised pressure signal sequence, and use the mean value of the calculated static reference segment as the dynamic offset reference value of each type of signal; the static reference segment corresponds to the signal range when the barbell is placed statically on the support, the trainee holds it naturally, and the body protection device has no contact pressure.

[0124] In this embodiment, before and after each set of exercises, the barbell is usually returned to the support frame of the squat rack. At this time, the barbell is in a static state. The system detects intervals with extremely low signal change rates and determines them as rest intervals by combining the timestamps. It automatically marks the signal in this interval as a static reference segment, calculates the average value of all signal samples in the static reference segment, and uses this average value as the dynamic offset reference value for this training session. This dynamic offset reference value represents the sensor's output reference when the barbell is unloaded and stationary in the current environment and current installation state.

[0125] S3.3: Subtract the dynamic offset reference value from each data point in the denoised pressure signal sequence of each type in real time, in order to eliminate static deviations caused by sensor zero drift, installation preload, etc., to obtain the corrected pressure electrical signal, and store the corrected pressure electrical signal in association with the corresponding timestamp.

[0126] The calculated load weight includes:

[0127] C1: Based on the slope and intercept in the pressure-weight conversion reference parameters, the amplitude value of the corrected original pressure signal is converted into instantaneous load weight;

[0128] In this embodiment, based on the slope and intercept in the pressure-weight conversion reference parameters, the amplitude values ​​of the corrected pressure electrical signal are substituted one by one into the linear conversion formula to calculate the instantaneous load weight at each moment. For example, if the corrected signal value at any moment is A, the slope is K, and the intercept is B, then the instantaneous load weight is equal to K multiplied by A plus B. The linear conversion formula is prior art and not an inventive solution of this application, and will not be elaborated upon here.

[0129] C2: Calculate the load weight of the current action by weighted averaging of multiple instantaneous load weights within the same action cycle, and associate the load weight with its corresponding timestamp to generate a load weight time series.

[0130] In this embodiment, due to the high sampling frequency of the sensor, the instantaneous load weight value fluctuates slightly. In order to obtain a more stable value that represents the load of a single action, the system will perform a weighted average of multiple instantaneous load weight values ​​within the identified single action cycle. The weight is higher for values ​​closer to the middle of the action and slightly lower for values ​​closer to the start and end points. The calculated weighted average is the load weight of that action. The system associates the instantaneous load weight corresponding to all time points or the load weight of each action cycle with its timestamp and arranges them in chronological order to form the load weight time series for this training.

[0131] The extraction of the first dynamic feature includes:

[0132] D1: Based on the load weight time series, the start and end points of each repetitive action are identified by finding local minimum points, and the action interval is verified by combining the abrupt change characteristics of the attitude signal, thus dividing the single repetitive action interval.

[0133] In this embodiment, the system first performs motion segmentation based on the load weight time series. It identifies the start and end points of each repetitive motion by finding local minima in the sequence. In a squat, the lowest point usually corresponds to the moment of maximum load weight. However, due to factors such as the barbell's inertia during movement, the pressure sensed by the sensor will dynamically change. The system determines the lowest point of the motion by detecting the inflection point where the weight change trend changes from decreasing to increasing, and defines the interval between two adjacent lowest points as a complete repetitive motion interval. This process divides the continuous training data stream into independent repetitive motions.

[0134] D2: Within each repetitive motion interval, extract the maximum value of the load weight as the peak force, extract the minimum value of the load weight as the valley force, and calculate the difference between the peak force and the valley force as the net output force of that repetitive motion.

[0135] In this embodiment, within each identified repetitive movement interval, the system extracts the following first dynamic features: firstly, peak force, which is the maximum load weight within the interval, representing the maximum external resistance overcome by the trainee in that movement; secondly, trough force, which is the minimum load weight within the interval, usually occurring at the moment of movement transition; and finally, net output force, which is the difference between peak force and trough force, reflecting the net work actually done by the trainee against gravity in that repetition.

[0136] D3: The peak force, trough force, and net output force of each repetitive motion, along with supplementary indicators such as hand grip strength and pressure at the body contact points, are used as the first dynamic feature set.

[0137] Extracting the second dynamic features includes:

[0138] E1: Based on the load weight time series, calculate the first derivative of the load weight change over time to obtain the force change rate series. Within the single repetitive action interval, extract the positive maximum change rate from the force change rate series as the centripetal phase explosive force index and extract the negative maximum change rate as the centrifugal phase explosive force index. The calculation process of the first derivative is prior art in this field and is not an inventive solution of this application, so it will not be described in detail here.

[0139] In this embodiment, the system performs differential processing on the load weight time series, calculates the first derivative of the load weight change over time, and obtains a force change rate sequence. This force change rate sequence reflects the speed at which the trainee exerts force. Within a single repetitive movement interval, two key indicators are extracted from the force change rate sequence: one is the concentric phase explosive force indicator, i.e., the positive maximum rate of change. In squats, this corresponds to the instantaneous rate of force increase when squatting from the lowest point, and is an important parameter for evaluating explosive force; the other is the eccentric phase explosive force indicator, i.e., the negative maximum rate of change. This corresponds to the instantaneous rate of force decrease during the squat, reflecting the ability to control the lowering weight.

[0140] E2: Based on the load weight time series, calculate the duration during which the load weight exceeds a preset threshold weight within a single repetitive movement interval, and use this as the muscle tension time;

[0141] In this embodiment, the system presets a threshold weight, such as 70% of the maximum load. Within a single repetitive movement interval, it calculates the total duration for which the load weight exceeds this threshold weight. This time approximately represents the time the target muscle group is under high mechanical tension, which is one of the key factors for stimulating muscle hypertrophy.

[0142] E3: Simultaneously calculate the peak value of the rate of change of hand grip strength, the offset angle and offset rate of body posture. When the posture offset angle exceeds the preset safety threshold, such as 15°, it is marked as a posture imbalance feature.

[0143] E4: The explosive power index during the concentric phase, the explosive power index during the eccentric phase, the duration of muscle tension, and the characteristics of postural imbalance are used as the second set of dynamic features.

[0144] Furthermore, after extracting the second dynamic feature, an enhanced analysis is performed on the centrifugation stage:

[0145] (1) Calculate the eccentric phase time from the peak force point to the trough force point in each repetitive action interval;

[0146] (2) Calculate the average rate of decrease of the load weight during the centrifugation stage, i.e. (peak force - valley force) / centrifugation stage time;

[0147] (3) Set a safe centrifugation rate threshold based on the user's historical best data or group standards, for example, the rate of decrease per second should not exceed 1.5% of its maximum weight in a single cycle;

[0148] (4) If the average descent rate of the centrifugation stage calculated in real time exceeds the safe centrifugation rate threshold, then when generating the risk warning, an additional specific warning message about insufficient control of the centrifugation stage and potential soft tissue damage risk will be generated.

[0149] In this embodiment, the extracted first and second dynamic features are immediately packaged and transmitted to the terminal application via Bluetooth after each repeated action. After receiving these feature data, the terminal application performs standardization processing on them in conjunction with the training baseline parameters set at the beginning of training. For example, all strength data are divided by the trainee's weight to obtain relative strength values; time data is converted into standard units; and different types of feature data are labeled with uniform tags. Finally, a standardized training data stream with a uniform format and clear meaning is generated to provide input for visualization interface and in-depth analysis.

[0150] Example 2

[0151] Please see Figure 2 Another embodiment of the present invention provides: an intelligent barbell strength training monitoring system based on pressure sensing, comprising:

[0152] Barbell sensing module, pressure-sensing gloves, body protection device, terminal application module, evaluation module;

[0153] The barbell sensing module 10 is used to collect raw pressure signals and ambient temperature signals in real time during training, complete the initial signal encapsulation and wireless transmission, and generate a pressure-weight conversion benchmark through a calibration process.

[0154] The pressure-sensing glove 20 is a protective sensing device worn by trainees' hands. It has a built-in grip force sensing component and a miniature temperature sensor to collect hand grip force pressure signals and hand ambient temperature signals in real time. After preliminary signal processing, the signals are wirelessly transmitted to the terminal application.

[0155] The body protection device 30 is a protective sensing device worn by the trainee on the body contact parts. It has built-in pressure sensing components, posture sensing components and miniature temperature sensors to collect pressure signals, training posture signals and ambient temperature signals of the body contact parts in real time. After completing the initial signal processing, it is wirelessly transmitted to the terminal application, while realizing the basic protection of the body contact parts by buffering and impact resistance.

[0156] The terminal application module 40 serves as the interactive entry point for trainees and the central hub for data storage and display. It is used to receive parameters set by trainees, receive data transmitted from hardware, store key information, generate visual interfaces and reports, and provide training suggestions and risk warnings to trainees, thereby enabling bidirectional data flow.

[0157] The evaluation module 50 is used to purify, convert, and extract features from the collected raw data. It integrates multi-source data from the barbell sensing module, pressure-sensing glove module, body protection device module, and wearable device for intelligent evaluation, while also identifying the standardization of training movements and generating protective warnings.

[0158] The embodiments of the present invention have been described above with reference to the accompanying drawings. However, the present invention is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments under the guidance of the present invention without departing from the spirit and scope of the present invention. All of these variations are within the protection scope of the present invention.

Claims

1. A method of monitoring a smart barbell strength training based on pressure sensing, characterized in that, include: A barbell sensing system is constructed based on a calibrated pressure sensor and a wireless transmission module, generating pressure-weight conversion reference parameters and establishing a stable communication link between the wireless transmission module and the terminal application. The barbell sensing system is compatible with the pressure-sensing gloves and body protection device worn by the trainee. The pressure-sensing gloves are used to collect hand grip pressure signals, and the body protection device is used to collect pressure signals and posture signals of the trainee's body contact points. The training base parameters set by the trainee are obtained through the terminal application. The sensing components of the pressure sensor, pressure-sensitive gloves and body protection device are activated simultaneously to collect the original pressure electrical signal, hand grip pressure signal, body contact pressure signal and posture signal in real time. The signal is then filtered and denoised and dynamic offset compensation is performed to obtain the corrected pressure signal and posture signal. The load weight is calculated based on the corrected original pressure electrical signal and pressure-weight conversion reference parameters. The corrected pressure signal and attitude signal are fused together, and the first dynamic feature and the second dynamic feature are extracted according to the load weight. The data is then transmitted to the terminal application in real time and converted into standardized training data. Based on standardized training data and basic training parameters, a visual interface and structured training report are generated, including strength-time curves, grip strength change curves, and body stress / posture change curves. The system calls upon physiological marker data acquired by wearable devices and integrates it with structured training reports, physiological marker data, data collected from pressure-sensitive gloves and body protection devices, as well as historical training records stored locally on the terminal application for evaluation. It then generates strength training suggestions and risk warnings and provides them back to the trainee. The risk warnings include protective warnings for insufficient hand grip strength, abnormal pressure at body contact points, and postural imbalance.

2. The intelligent barbell strength training monitoring method based on pressure sensing as described in claim 1, characterized in that, The construction of the barbell sensing system includes: The pressure sensor is encapsulated in a waterproof and shockproof housing and fixedly installed below the center grip area of ​​the barbell bar to collect the raw pressure electrical signal generated by the applied load in real time. The wireless transmission module is integrated into the end weight clamp of the barbell bar and electrically connected to the pressure sensor through a built-in flexible circuit board. It is used to receive pressure electrical signals, hand grip pressure signals, body contact pressure signals and posture signals, and package the signals with corresponding timestamps and send them to the terminal application through its Bluetooth Low Energy protocol. After constructing the barbell sensing system, a calibration process is performed: multiple weights of known standard weights are loaded onto the barbell bar in sequence, and the stable pressure electrical signal value output by the pressure sensor at each weight is recorded. The least squares method is used to linearly fit multiple sets of weight-signal value data to generate pressure-weight conversion reference parameters in the form of slope and intercept. Zero-point calibration was performed on the grip force sensing components of the pressure-sensitive gloves and the pressure and posture sensing components of the body protection device, and the reference signal value under no external force was recorded. The Bluetooth Low Energy protocol is used to establish pairing connections between terminal applications and wireless transmission modules, pressure-sensitive gloves, and body protection devices.

3. The intelligent barbell strength training monitoring method based on pressure sensing as described in claim 2, characterized in that, The process of acquiring the corrected pressure signal includes: The received original pressure electrical signal with timestamps, hand grip pressure signal, and body contact pressure signal are respectively input into a digital Butterworth band-stop filter with a preset cutoff frequency to obtain various types of denoised pressure signal sequences; the preset cutoff frequency is set according to the typical frequency range of human muscle tremors. The static reference segment in each type of denoised pressure signal sequence is identified, and the mean value of the calculated static reference segment is used as the dynamic offset reference value of each type of signal; the static reference segment corresponds to the signal range when the barbell is placed statically on the support, the trainee holds it naturally, and the body protection device has no contact pressure. The corrected pressure signal is obtained by subtracting the dynamic offset reference value from each data point in the denoised pressure signal sequence of each type in real time.

4. The intelligent barbell strength training monitoring method based on pressure sensing as described in claim 3, characterized in that, The calculated load weight includes: Based on the slope and intercept in the pressure-weight conversion reference parameters, the amplitude value of the corrected original pressure electrical signal is converted into instantaneous load weight. The load weight of the current action is calculated by weighting multiple instantaneous load weights within the same action cycle, and the load weight is associated with its corresponding timestamp to generate a load weight time series.

5. The intelligent barbell strength training monitoring method based on pressure sensing as described in claim 4, characterized in that, The extraction of the first dynamic feature includes: Based on the load weight time series, the start and end points of each repetitive action are identified by finding local minimum points, and the action interval is verified by combining the abrupt change characteristics of the attitude signal, thus dividing the single repetitive action interval. Within each repetitive motion interval, the maximum value of the load weight is extracted as the peak force, and the minimum value of the load weight is extracted as the valley force. The difference between the peak force and the valley force is calculated as the net output force of that repetitive motion. The peak force, trough force, and net output force of each repetitive motion, along with supplementary indicators such as hand grip strength and pressure at the body contact points, are used as the first dynamic feature set.

6. The intelligent barbell strength training monitoring method based on pressure sensing as described in claim 5, characterized in that, Extracting the second dynamic features includes: Based on the load weight time series, the first derivative of the load weight change over time is calculated to obtain the force change rate series; Within the single repetitive action interval, the positive maximum rate of change is extracted from the force change rate sequence as the explosive force index of the centripetal phase, and the negative maximum rate of change is extracted as the explosive force index of the centrifugal phase. Based on the load weight time series, the duration during which the load weight exceeds a preset threshold weight within a single repetitive movement interval is calculated as the muscle tension time. The peak value of the rate of change of hand grip strength, the offset angle and the offset rate of body posture are calculated simultaneously. When the posture offset angle exceeds the preset safety threshold, it is marked as a posture imbalance feature. The explosive power index during the concentric phase, the explosive power index during the eccentric phase, the duration of muscle tension, and the characteristics of postural imbalance were used as the second set of dynamic features.

7. The intelligent barbell strength training monitoring method based on pressure sensing as described in claim 6, characterized in that, The evaluation process involves calling upon physiological marker data acquired by wearable devices and integrating structured training reports, physiological marker data, and historical training records stored locally on the terminal application. The evaluation process is implemented through a deep neural network model. The deep neural network model includes a multilayer perceptron branch for receiving the first and second dynamic features from the structured training report, a one-dimensional convolutional neural network branch for processing physiological marker data, a convolutional recurrent neural network branch for processing multi-dimensional sensor data collected by pressure-sensitive gloves and body protection devices, and a long short-term memory network branch for processing long-term trends in historical training records. The physiological marker data include heart rate and heart rate variability; The multilayer perceptron branch contains three fully connected layers. The first layer maps the input dynamic features to a 64-dimensional space and uses the ReLU activation function. The second layer reduces the dimensions to 32, and the third layer reduces the dimensions to 16. The one-dimensional convolutional neural network branch contains two convolutional layers. The first layer uses 16 convolutional kernels of length 3, and the second layer uses 32 convolutional kernels of length 3. Each convolutional layer is followed by a max pooling layer, and finally flattened into a one-dimensional vector. The convolutional recurrent neural network branch performs convolutional feature extraction and recurrent temporal analysis on the time series data of grip force, body pressure, and posture. The number of hidden layer units is 64, and the output is a multidimensional feature vector. The Long Short-Term Memory (LSTM) network branch processes the key indicator sequences of the most recent 10 training sessions, and its hidden layer unit number is 32. The output vectors of the multilayer perceptron branch, the one-dimensional convolutional neural network branch, the convolutional recurrent neural network branch, and the long short-term memory network branch are concatenated in the concatenation layer. Through a fusion network containing two fully connected layers, the final output layer uses the sigmoid activation function to generate a scalar value representing the probability of risk warning and a multidimensional vector containing strength training suggestions.

8. The intelligent barbell strength training monitoring method based on pressure sensing as described in claim 7, characterized in that, The method also includes a compensation mechanism based on environmental parameters: In the barbell sensing system, a temperature sensor is integrated in the same position as the pressure sensor. Miniature temperature sensors are also integrated next to the sensing components of the pressure-sensing gloves and body protection devices to collect real-time temperature data of the environment in contact with the barbell bar, hands, and body parts. Temperature data is sent to the terminal application via the stable communication link. In the terminal application, the pre-stored temperature-pressure sensitivity coefficient lookup table is called. For the received real-time temperature data, the corresponding compensation coefficient is obtained from the temperature-pressure sensitivity coefficient lookup table to perform temperature drift compensation on the corrected pressure signal, and the temperature-compensated pressure signal is obtained. The temperature-pressure sensitivity coefficient lookup table records the zero-point drift and sensitivity changes of various pressure sensors at different temperatures. The load weight is recalculated based on the temperature-compensated pressure signal.

9. The intelligent barbell strength training monitoring method based on pressure sensing as described in claim 8, characterized in that, The method also includes training action norm recognition: Predefined standard action template features; The standard movement template features include standardized load weight time series, force change rate series, grip strength change series, and body stress / posture change series for standard squat, bench press, and deadlift movements. During real-time training, the load weight time series, force change rate series, grip strength change series, and body pressure / posture change series within the current single repetitive movement interval are extracted as features to be compared. The dynamic time warping algorithm is used to calculate the minimum alignment path distance between the features to be compared and the standard movement template features to obtain the movement similarity score. If the similarity score of the action is lower than the preset standard threshold, or if the changes in grip strength or body pressure / posture deviate from the standard template beyond the preset safety range, then non-standard action and protective warning signs will be added to the generated structured training report. When generating strength training suggestions, priority will be given to correcting the form of the action, and specific protective operation suggestions for adjusting hand grip strength and correcting body posture will be given.

10. A pressure-sensing-based intelligent barbell strength training monitoring system, used to implement the pressure-sensing-based intelligent barbell strength training monitoring method according to any one of claims 1-9, characterized in that, include: Barbell sensing module, pressure-sensing gloves, body protection device, terminal application module, evaluation module; The barbell sensing module is used to collect raw pressure signals and ambient temperature signals in real time during training, complete the initial signal encapsulation and wireless transmission, and generate a pressure-weight conversion benchmark through a calibration process. The pressure-sensing glove is a protective sensing device worn by the trainee's hand. It has a built-in grip force sensing component and a miniature temperature sensor to collect hand grip force pressure signals and hand ambient temperature signals in real time. After preliminary signal processing, the signals are wirelessly transmitted to the terminal application. The body protection device is a protective sensing device worn by the trainee on the body contact area. It has built-in pressure sensing components, posture sensing components and miniature temperature sensors to collect pressure signals, training posture signals and ambient temperature signals of the body contact area in real time. After completing the initial signal processing, it is wirelessly transmitted to the terminal application, and at the same time, it realizes the basic protection of the body contact area by buffering and impact resistance. The terminal application module serves as the interactive entry point for trainees and the data storage and display center. It is used to receive parameters set by trainees, receive data transmitted by hardware, store key information, generate visual interfaces and reports, and provide training suggestions and risk warnings to trainees. The evaluation module is used to purify, convert, and extract features from the collected raw data, and integrate multi-source data from the barbell sensing module, pressure-sensing glove module, body protection device module, and wearable device for intelligent evaluation. At the same time, it identifies the standardization of training movements and generates protective warnings.

Images