A safety belt falling force detection method and system

Abstract

The application relates to the field of safety protection detection, and provides a safety belt falling force detection method and system. The method comprises the following steps: synchronously collecting the force state of a woven belt, and obtaining a user's personalized biomechanics parameter file based on a biomechanics identification model; performing time-frequency joint analysis and impact dynamics calculation on multi-dimensional aligned time sequence data flow, screening falling candidate events, classifying the falling candidate events, and obtaining a falling event characteristic vector; performing thermal compensation correction on the dynamic elastic modulus of the woven belt, and obtaining a temperature correction modulus field; based on an input edge end physical information neural network model and a cloud woven belt human body coupling mechanics model, obtaining a three-dimensional impact load approximate solution and a three-dimensional impact load accurate solution; calculating a personalized effective impact amount, comparing the personalized effective impact amount with a sub-position damage threshold database, and based on a fatigue damage accumulation model, outputting a falling force detection report. The application improves the physiological authenticity and accuracy of falling force detection.

Description

Technical Field

[0001] This invention relates to the field of safety protection testing technology, and in particular to a method and system for testing the force exerted by a seat belt during a fall. Background Technology

[0002] Falls from heights are a major cause of injuries and fatalities in industries such as construction, power, and chemicals. As a core piece of personal protective equipment (PPE) for preventing fall injuries, the stress state of safety belts during a fall directly determines their protective effect on the wearer and the structural integrity of the equipment itself. With increasingly stringent industrial safety regulations, the quantitative testing and assessment of the impact loads borne by safety belts in actual fall incidents has become a crucial technical requirement for accident analysis, equipment maintenance, and PPE management. Current industry standards clearly define the performance indicators of safety belts, including static strength, dynamic impact load, and temperature, and research on related testing methods and equipment continues to advance.

[0003] However, existing seat belt stress detection methods generally rely on discrete sampling of a few fixed measurement points using a single type of sensor in practical applications. This makes it impossible to achieve continuous distributed sensing of stress, strain, and temperature along the entire length of the webbing, which severely limits the accuracy of three-dimensional reconstruction of fall impact loads. At the same time, existing methods use fixed biomechanical parameters based on general population experience values ​​in the injury risk assessment stage, failing to dynamically correct for individual differences in age, body type, and body fat percentage. This results in significant deviations in the injury risk assessment results for special groups, making it difficult to support precise personal safety protection decisions. Summary of the Invention

[0004] This invention provides a method and system for detecting the force of a seatbelt falling, in order to overcome the shortcomings of the prior art.

[0005] This invention provides a method for detecting the force exerted by a seatbelt during a fall, comprising: S1. By deploying a heterogeneous sensor array at key stress nodes of the seat belt, the stress state of the webbing is collected synchronously, and the collected multi-dimensional aligned time-series data stream is input into the biomechanical identification model to obtain a personalized biomechanical parameter profile for the user. S2. Perform time-frequency joint analysis and impact dynamics calculation on the multidimensional aligned time-series data stream, filter the fall candidate events by the preset fall discrimination rule set, classify the fall candidate events by direction, and obtain the fall event feature vector. S3. Based on the multi-dimensional aligned temporal data stream, the dynamic elastic modulus of multiple positions of the webbing is thermally compensated and corrected to obtain the temperature correction modulus field; the feature vector of the fall event and the temperature correction modulus field are respectively input into the edge physical information neural network model and the cloud webbing human body coupling mechanical model to obtain the approximate solution and the accurate solution of the three-dimensional impact load. S4. Calculate the personalized effective impact amount based on the approximate solution of the three-dimensional impact load and the user's personalized biomechanical parameter file, compare it with the damage threshold database of different parts, and input the historical peak tension of each section in the accurate solution of the three-dimensional impact load into the fatigue damage accumulation model, and output a fall stress test report including the fall stress safety level and the ratio of the remaining bearing life of the webbing.

[0006] According to the method for detecting the force of a seatbelt falling according to the present invention, step S1 further includes: S11. Piezoresistive strain sensors, triaxial accelerometers and fiber Bragg grating sensors are deployed at the shoulder load-bearing section, waist buckle connection section and back hook point of the safety belt, respectively, and micro-thermal thin film sensor nodes are deployed at preset intervals along the entire length of the webbing to form a heterogeneous sensor array. S12. Align the clocks of each sensor node in the heterogeneous sensor array using a hardware time synchronization protocol to synchronize and quantize all channels at a uniform sampling rate to obtain structured data frames. S13. A multi-dimensional aligned timing data stream is formed by combining structured data frames from each channel; S14. Extract the data segment of the user in normal working activity state from the multidimensional aligned time-series data stream. Use the single-degree-of-freedom spring-damping system model to systematically identify the webbing tension signal and human body acceleration signal in the data segment of normal working activity state, and obtain the user stiffness, user damping and user soft tissue response time constant to form a personalized biomechanical parameter profile of the user.

[0007] According to the method for detecting the force of a seat belt fall provided by the present invention, in step S14, the systematic identification of the webbing tension signal and the human acceleration signal in the data segment of the normal working activity state is performed by a recursive Kalman filter algorithm. The recursive Kalman filter algorithm takes the webbing tension signal as the excitation input and the human acceleration signal as the response output to perform online back-calculation of the parameters of the single-degree-of-freedom spring-damped system model. The expression for the user soft tissue response time constant in step S14 is:

[0008] in, To calculate the obtained user soft tissue response time constant, For user stiffness, For user damping, The natural angular frequency of the human torso vibration.

[0009] According to the method for detecting the force of a seatbelt falling according to the present invention, step S2 further includes: S21. Perform time-frequency joint analysis on the acceleration signal in the multidimensional aligned time-series data stream to extract the time-frequency distribution feature vector of the impact energy. S22. Based on the slope of the rising edge and the duration of the peak value of the acceleration time-domain curve, candidate fall events are obtained by filtering from the impact energy time-frequency distribution feature vector using the fall discrimination rule set. S23. Perform time integration on the webbing tension signal in the candidate fall event, and calculate the equivalent fall height by combining the triaxial composite acceleration peak. S24. Based on the proportion of the three-axis acceleration components in the candidate fall events and the webbing strain asymmetry index output by the fiber Bragg grating sensor, the candidate fall events are classified by direction to obtain fall mode labels. S25, the feature vector of the fall event is composed of the impact energy time-frequency distribution feature vector, equivalent fall height, triaxial composite acceleration peak value, webbing strain asymmetry index and fall mode label combination.

[0010] According to the method for detecting the force of a seatbelt falling according to the present invention, step S3 further includes: S31. Based on the distributed thermal field signal collected by the micro-thermal thin film sensor node in the multidimensional aligned time-series data stream, the dynamic elastic modulus at each position of the webbing is thermally compensated and corrected to obtain the temperature correction modulus field. S32. The feature vector of the fall event is inferred from the physical information neural network model of the temperature correction modulus field input edge, and an approximate solution of the three-dimensional impact load is output within a preset response time limit. S33. Upload the multidimensional aligned time-series data stream and the temperature-corrected modulus field to the cloud, input the cloud-based webbing human body coupling mechanics model for inverse analysis and solution, and obtain the accurate solution of the three-dimensional impact load.

[0011] According to the method for detecting the force of a seatbelt falling according to the present invention, in step S31, the expression for thermal compensation correction of the dynamic elastic modulus at each position of the webbing is as follows:

[0012] in, These are the spatial coordinates along the length of the webbing. For time, For the position coordinates of the webbing ,time The dynamic elastic modulus at that point. The nominal elastic modulus at the reference temperature. The temperature-modulus sensitivity coefficient of the webbing material. This is a distributed thermal field signal. This is a reference temperature.

[0013] According to the method for detecting the force of a seatbelt fall provided by the present invention, in step S32, the process of constructing the edge-end physical information neural network model further includes: S321. Using the Timoshenko beam theory governing equation as the physical constraint term in the cloud, and the data-driven loss term to form a composite loss function, the historical solution result set of cloud inverse finite element analysis is used as the training sample to train the pre-set basic neural network offline to obtain the physical information neural network. S322. Perform structural pruning and fixed-point quantization compression on the physical information neural network, and deploy the compressed model to the local low-power edge computing chip of the safety belt to obtain the edge physical information neural network model.

[0014] According to the method for detecting the force of a seatbelt falling according to the present invention, step S33 further includes: S331. Based on the fall mode label in the fall event feature vector, retrieve the corresponding posture template from the pre-built human posture library, and use the elastic basic contact model to parameterize the contact pressure distribution between the webbing and the human torso. S332. Using the distributed strain field of the fiber Bragg grating sensor in the multidimensional aligned time-series data stream as the boundary condition, the temperature-corrected modulus field is substituted into the Timoshenko beam discrete element control equation. With the goal of minimizing the norm of the difference between the simulated strain and the measured strain of each element, the Levenberg-Marquardt algorithm is used to iteratively solve the tension, shear force and bending moment of each section to obtain the accurate solution of the three-dimensional impact load.

[0015] According to the method for detecting the force of a seatbelt fall provided by the present invention, in step S4, the expression for the personalized effective impact amount is:

[0016] in, To achieve personalized and effective impact, The moment of the shock event. The time history of the resultant impact force is extracted from the approximate solution of the three-dimensional impact load. For the duration of the impact, The user's soft tissue response time constant in the user's personalized biomechanical parameter profile. For time; In step S4, the expression for the fatigue damage accumulation model is:

[0017] in, Spatial number index for the webbing cross section, For the first The ratio of remaining load-bearing life of the webbing at each cross section, For impact event indexing, This represents the total number of impact events experienced by the corresponding cross section. cross section spatial coordinates, cross section In the The peak tension corresponds to the impact event. For the first The number of load cycles experienced by the cross section during the impact event. For the cross section at the corresponding peak tension The theoretical fatigue life is as follows. The coefficient for accelerated aging due to heat.

[0018] The present invention also provides a seat belt fall force detection system for performing a seat belt fall force detection method as described in any of the above claims, comprising: The data acquisition module is used to synchronously acquire the stress state of the webbing through a heterogeneous sensor array deployed at key stress nodes of the seat belt, and input the acquired multidimensional aligned time-series data stream into the biomechanical identification model to obtain a personalized biomechanical parameter profile for the user. The identification module is used to perform time-frequency joint analysis and impact dynamics calculation on multidimensional aligned time-series data streams. It selects candidate fall events from a preset set of fall discrimination rules, classifies the candidate fall events by direction, and obtains the feature vector of the fall event. The calculation module is used to perform thermal compensation correction on the dynamic elastic modulus of multiple positions of the webbing based on the multidimensional aligned temporal data stream to obtain the temperature correction modulus field; the feature vector of the fall event and the temperature correction modulus field are respectively input into the edge physical information neural network model and the cloud webbing human body coupling mechanical model to obtain the approximate solution and the accurate solution of the three-dimensional impact load; Output module: Used to calculate personalized effective impact amount based on the approximate solution of three-dimensional impact load and the user's personalized biomechanical parameter file, compare it with the damage threshold database of different parts, and input the historical peak tension of each section in the accurate solution of three-dimensional impact load into the fatigue damage accumulation model, and output a fall stress test report including the fall stress safety level and the ratio of the remaining bearing life of the webbing.

[0019] This invention provides a method and system for detecting the force of a seatbelt falling from a height. Firstly, it deploys a heterogeneous sensor array at key stress points of the seatbelt, consisting of piezoresistive strain sensors, triaxial accelerometers, fiber Bragg grating sensors, and micro-thermal thin-film sensor nodes. A hardware time synchronization protocol is introduced to achieve precise time alignment of data from each channel, enabling the synchronous and complete capture of four physical quantities—stress, strain, temperature, and heat distribution—during the fall impact. This fundamentally solves the problem of insufficient sensing dimensions caused by discrete sampling points of a single sensor in existing technologies, providing a high-quality data foundation for subsequent load reconstruction. Furthermore, this invention converts the distributed thermal field signal into temperature... By correcting the modulus field and substituting it into the mechanical reconstruction model, the load calculation error of the webbing in a wide temperature range environment is effectively suppressed, and the reconstruction results can still maintain high reliability under extreme operating conditions such as high temperature and low temperature. The introduction of the edge physical information neural network model transforms the complex inverse finite element solution process into a fast inference task that can be completed on a low-power chip, enabling the system to output an approximate solution of the three-dimensional impact load within a preset response time limit even in a network-free environment, significantly compressing the response delay between the occurrence of the fall and the triggering of the warning. In addition, the accurate solution of the cloud-based webbing-human coupling mechanical model proposed in this invention is continuously fed back to update the edge model training sample set, forming a continuous self-optimization closed loop of edge-cloud collaboration. At the damage assessment level, the system uses a recursive Kalman filter algorithm to identify users' daily work activity data online, dynamically acquiring individual user-specific stiffness, damping, and soft tissue response time constants. These values ​​are then substituted into a personalized effective impact calculation formula, enabling the damage risk assessment results to truly reflect the individual differences among workers of different ages, body types, and physiological states. This effectively avoids the distortion of assessment results for specific groups by general experience parameters. Furthermore, this invention introduces the thermally accelerated aging coefficient into the calculation of the remaining load-bearing life ratio of the webbing in the modified Miner criterion, allowing equipment life management to simultaneously consider both mechanical fatigue damage and thermally induced material degradation failure mechanisms. The output drop stress test report and webbing life map provide quantifiable and traceable decision-making basis for enterprise safety management. Attached Figure Description

[0020] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0021] Figure 1 This invention provides a schematic flowchart of a method for detecting the force of a seatbelt falling. Figure 2This is a schematic diagram of a seatbelt fall force detection system provided by the present invention. Detailed Implementation

[0022] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, embodiments of this invention, and should not be construed as limiting the invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention. In the description of this invention, it should be understood that the terminology used is for descriptive purposes only and should not be construed as indicating or implying relative importance.

[0023] The embodiments of the present invention are described below with reference to the figures.

[0024] like Figure 1 As shown, the present invention provides a method for detecting the force of a seatbelt falling, comprising: S1. By deploying a heterogeneous sensor array at key stress nodes of the seat belt, the stress state of the webbing is collected synchronously, and the collected multi-dimensional aligned time-series data stream is input into the biomechanical identification model to obtain a personalized biomechanical parameter profile for the user.

[0025] Existing seatbelt detection solutions typically deploy a single type of sensor at only a few fixed measuring points, acquiring a limited dimension of physical quantities that cannot simultaneously describe the mechanical and thermal states of the webbing during impact. This invention posits that the webbing undergoes four physical processes simultaneously during an actual drop impact: tension, acceleration, strain distribution, and temperature distribution. The lack of information in any one of these processes can lead to systematic deviations in subsequent load reconstruction. Therefore, in step S1, this invention integrates four heterogeneous sensor types—piezoresistive strain gauges, triaxial accelerometers, fiber Bragg grating sensors, and micro-thermal thin-film sensors—into a single acquisition framework. A hardware time synchronization protocol ensures strict alignment of the four types of signals on the time axis, fundamentally resolving the timing inconsistency problem in multi-source data fusion.

[0026] Step S1 further includes: S11. Piezoresistive strain sensors, triaxial accelerometers, and fiber Bragg grating sensors are deployed at the shoulder load-bearing section, waist buckle connection section, and back attachment point of the safety belt, respectively. Micro-thermal thin film sensor nodes are deployed along the entire length of the webbing at a preset interval to form a heterogeneous sensor array.

[0027] In step S11, the present invention deploys different types of sensors at three key stress points of the safety belt: a piezoresistive strain sensor is deployed at the shoulder load-bearing section. This sensor is based on the principle that the resistance value of a material changes after being stressed, converting the tension on the webbing cross section into an electrical signal output, with a range covering 0 to 50 kN and an accuracy of ±0.5% FS; a triaxial accelerometer is deployed at the waist buckle connection section. This sensor simultaneously senses the acceleration components in the three orthogonal directions of X, Y, and Z, with a sampling rate of not less than 10 kHz, used to capture the transient motion state of the human torso during the fall impact; a fiber Bragg grating sensor is deployed at the back D-ring hook point. This sensor utilizes the response characteristics of the periodic change in refractive index in the fiber core to external strain, sensing the strain distribution along the length of the webbing with sub-millimeter spatial resolution, while simultaneously outputting temperature information to eliminate the interference of thermal expansion on the strain reading. In addition, the present invention also deploys a micro-thermal thin-film sensor node every 10 cm along the entire length of the webbing. This node converts the local temperature of the webbing surface into a resistance change signal output, forming a distributed thermal field acquisition channel covering the entire length of the webbing. The above four types of sensors together constitute a heterogeneous sensor array, enabling full-domain sensing of four physical quantities: webbing stress, strain, temperature, and heat distribution.

[0028] S12. Align the clocks of each sensor node in the heterogeneous sensor array using a hardware time synchronization protocol, and quantize all channels synchronously at a uniform sampling rate to obtain structured data frames.

[0029] Since the physical acquisition circuits of the four types of sensors are independent of each other, and the clocks of each node have drift differences, this invention introduces a hardware time synchronization module based on the IEEE 1588 precision time protocol in step S12. By periodically exchanging timestamp messages between each sensor node and the main control unit, the clock error of each node is converged to within 100ns. After clock alignment is completed, this invention synchronously triggers sampling of all channels in the heterogeneous sensor array at a uniform sampling rate of not less than 10kHz. Each sampling result is encapsulated into a structured data frame in the format of "timestamp + sensor identifier + physical quantity value + check code" and written to the local anti-vibration flash memory cache.

[0030] S13. A multidimensional aligned timing data stream is formed by combining structured data frames from each channel.

[0031] In step S13, the present invention aligns and arranges the structured data frames continuously generated by each channel according to the timestamp order, and merges them into a multi-dimensional aligned time-series data stream containing five types of signals: piezoresistive tension, triaxial acceleration, fiber Bragg grating strain, fiber Bragg grating temperature, and distributed thermal field, denoted as... In this data stream, each channel corresponds to the same physical time node at any given moment.

[0032] S14. Extract the data segment of the user in normal working activity state from the multidimensional aligned time-series data stream. Use the single-degree-of-freedom spring-damping system model to systematically identify the webbing tension signal and human body acceleration signal in the data segment of normal working activity state, and obtain the user stiffness, user damping and user soft tissue response time constant to form a personalized biomechanical parameter profile of the user.

[0033] In step S14, the systematic identification of the webbing tension signal and human acceleration signal in the data segment of normal operation activity state is carried out by recursive Kalman filtering algorithm. The recursive Kalman filtering algorithm uses the webbing tension signal as excitation input and the human acceleration signal as response output to perform online back-calculation of the parameters of the single-degree-of-freedom spring-damped system model.

[0034] The expression for the user soft tissue response time constant in step S14 is as follows:

[0035] in, To calculate the obtained user soft tissue response time constant, For user stiffness, For user damping, The natural angular frequency of the human torso vibration.

[0036] Traditional injury assessment methods describe the response characteristics of human soft tissue using fixed, general biomechanical parameters, neglecting individual differences among workers of different ages, body types, and body fat percentages. This invention argues that during daily work activities, the webbing of the safety belt worn by the user is always under stress, and the dynamic relationship between the webbing tension signal and the human body acceleration signal naturally contains complete information about the user's soft tissue biomechanical characteristics. Based on this, step S14 of this invention provides an online identification mechanism centered on recursive Kalman filtering, transforming the daily work process itself into a data source for continuously calibrating the user's biomechanical parameters without requiring additional specialized testing procedures, thus achieving dynamic acquisition and continuous updating of individual user parameters.

[0037] In step S14, the present invention identifies and extracts data segments from the multidimensional aligned temporal data stream indicating that the user is in a normal working state. Specifically, the judgment criteria are: the peak value of the triaxial composite acceleration falls within the range of 0.2g to 2g (g is the gravitational acceleration), and does not trigger any fall candidate judgment conditions in the fall discrimination rule set. Continuous data segments that meet these conditions are marked as normal working state data segments. Subsequently, the present invention separates the webbing tension signal and the human body acceleration signal from this data segment, simplifying the human torso into a single-degree-of-freedom spring-damped system model. This model uses stiffness and damping as two parameters to describe the mechanical response characteristics of the human soft tissue to external forces.

[0038] This invention employs a recursive Kalman filter algorithm to identify the two parameters mentioned above online. Specifically, the noodle shop uses the webbing tension signal as the system excitation input and the human body acceleration signal as the system response output. After receiving a new frame of tension and acceleration data at each sampling moment, the recursive Kalman filter algorithm first predicts the current state based on the parameter estimates from the previous moment and the system state equation. Then, it calculates the residual between the predicted value and the measured acceleration, and applies a weighted correction to the estimated values ​​of stiffness and damping based on the Kalman gain. This process is repeated frame by frame until the parameter estimates converge, ultimately outputting the user stiffness. With user damping .

[0039] After obtaining the user stiffness and user damping, this invention substitutes these two values ​​into the formula for calculating the user soft tissue response time constant, and combines this with the natural angular frequency of human torso vibration to calculate the user soft tissue response time constant. This constant reflects the low-pass filtering characteristic duration of a specific user's soft tissue to impact loads. Ultimately, user stiffness, user damping, and the user's soft tissue response time constant together constitute the user's personalized biomechanical parameter profile. .

[0040] In addition, the present invention employs a sliding window mean strategy to continuously optimize and update the user's personalized biomechanical parameter profile. That is, whenever a new batch of identification results of normal working activity data segments is accumulated, the stored value in the profile is replaced with the mean of the most recent identification values, so that the parameter profile continuously approaches the user's true individual biomechanical characteristics as the wearing time increases.

[0041] S2. Perform time-frequency joint analysis and impact dynamics calculation on the multidimensional aligned time-series data stream. Select candidate fall events by filtering the preset fall discrimination rule set, and classify the candidate fall events by direction to obtain the fall event feature vector.

[0042] Step S2 further includes: S21. Perform time-frequency joint analysis on the acceleration signal in the multidimensional aligned time-series data stream to extract the time-frequency distribution feature vector of the impact energy.

[0043] In step S21, this invention extracts the acceleration signal output from the triaxial accelerometer from the multidimensional aligned time-series data stream and simultaneously applies two time-frequency analysis methods: short-time Fourier transform and wavelet packet decomposition. The short-time Fourier transform performs a segmented Fourier transform on the acceleration signal according to a fixed time window, outputting the energy distribution of the signal in the 5-500Hz frequency band within each time window. Wavelet packet decomposition, on the other hand, performs multi-level decomposition of the acceleration signal in both the time and frequency domains, extracting nodal energy coefficients from the high-frequency impact component and the low-frequency trend component. This invention then concatenates the results of these two analyses into a time-frequency distribution feature vector of impact energy. This vector describes the energy distribution pattern of the acceleration signal in both the time and frequency axes.

[0044] S22. Based on the slope of the rising edge and the duration of the peak value of the acceleration time-domain curve, candidate fall events are obtained by filtering from the impact energy time-frequency distribution feature vector using the fall discrimination rule set.

[0045] In step S22, the present invention constructs a set of fall discrimination rules based on prior knowledge of impact dynamics. This set of rules includes two core criteria: first, the slope of the rising edge of the acceleration time-domain curve must be greater than 5000 m / s². 3 First, the rate at which acceleration jumps from the baseline to the peak must meet this lower limit, reflecting the rapid loading characteristics of a real fall impact; second, the duration of the peak acceleration must be less than 30ms, reflecting the pulse-like and short characteristics of a fall impact. This invention uses the time-frequency distribution characteristic vector of impact energy... The rising slope and peak duration of each time window are compared with the two criteria mentioned above. The data segment corresponding to the time window that satisfies both criteria is marked as a candidate event for a fall, while the data segment that does not satisfy either criterion is classified as daily vibration interference and excluded.

[0046] S23. Perform time integration on the webbing tension signal in the candidate fall event, and combine it with the triaxial composite acceleration peak value to calculate the equivalent fall height.

[0047] In step S23, the present invention extracts the webbing tension signal output by the piezoresistive strain sensor from the multidimensional aligned time-series data stream corresponding to the fall candidate event, using the start time of the fall candidate event as the starting point. Until the end time For the integration interval, the webbing tension signal is integrated over time to obtain the impact impulse; subsequently, the triaxial composite acceleration peak is synthesized from the triaxial acceleration components output by the triaxial accelerometer. According to the impact dynamics equations, the impact velocity is obtained by dividing the impact impulse by the equivalent mass. Then Substituting the kinematic equations of free fall, the equivalent fall height can be calculated. This value represents the equivalent free fall distance corresponding to this fall event.

[0048] The expression for the equivalent fall height calculated in step S23 is:

[0049] in, This is the equivalent fall height; This refers to the impact velocity upon impact. It is the acceleration due to gravity; The starting point of the candidate crash event. The moment the impact ends; For a moment The net value of the triaxial composite acceleration.

[0050] S24. Based on the proportion of the three-axis acceleration components in the candidate fall events and the webbing strain asymmetry index output by the fiber Bragg grating sensor, the candidate fall events are classified by direction to obtain fall mode labels.

[0051] In step S24, this invention calculates the energy proportion of the X, Y, and Z axis acceleration components output by the triaxial accelerometer in the candidate fall event during the impact duration, and determines the principal axis component of the fall direction based on the triaxial energy proportion. Simultaneously, this invention extracts the strain values ​​output by the fiber Bragg grating sensor at the measuring points on the left and right sides of the webbing from the multidimensional aligned time-series data stream, and calculates the webbing strain asymmetry index. The strain asymmetry index is calculated by dividing the difference between the strain on the left and the strain on the right by the sum of the two. The index ranges from -1 to 1; a larger absolute value indicates greater asymmetry in the stress distribution on both sides of the webbing. The specific expression for the webbing strain asymmetry index is:

[0052] in, The strain asymmetry index of the webbing; For a moment The strain value output by the fiber Bragg grating sensor at the measuring point on the left side of the webbing; For a moment The strain value output by the fiber Bragg grating sensor at the measuring point on the right side of the webbing.

[0053] After calculation, this invention combines the triaxial acceleration component ratio with the webbing strain asymmetry index. The combined input fall direction classification rule, when the vertical acceleration component is dominant and When the acceleration is close to zero, it is classified as a vertical fall; when the forward acceleration component is dominant, it is classified as a forward-leaning fall; when the lateral acceleration component is dominant, it is classified as a vertical fall. When the absolute value is large, it is classified as lateral swing, and the final output is a fall mode label. .

[0054] S25, the feature vector of the fall event is composed of the impact energy time-frequency distribution feature vector, equivalent fall height, triaxial composite acceleration peak value, webbing strain asymmetry index and fall mode label combination.

[0055] In step S25, the present invention uses the impact energy time-frequency distribution feature vector obtained in steps S21 to S24 to calculate the impact energy time-frequency distribution feature vector. Equivalent fall height Triaxial composite peak acceleration Webbing strain asymmetry index and fall mode labels Concatenate the data according to a fixed field order to form a feature vector of the fall event. This vector, in a structured form, fully describes the time-frequency characteristics, kinematic parameters, and directional attributes of this candidate fall event, serving as the input for the edge-end physical information neural network model inference and cloud-based inverse analysis solution in step S3.

[0056] S3. Based on the multidimensional aligned temporal data stream, the dynamic elastic modulus of multiple positions of the webbing is thermally compensated and corrected to obtain the temperature correction modulus field; the feature vector of the fall event and the temperature correction modulus field are respectively input into the edge physical information neural network model and the cloud webbing human body coupling mechanical model to obtain the approximate solution and the accurate solution of the three-dimensional impact load.

[0057] Step S3 further includes: S31. Based on the distributed thermal field signal collected by the micro-thermal thin film sensor node in the multidimensional aligned time-series data stream, the dynamic elastic modulus at each position of the webbing is thermally compensated and corrected to obtain the temperature-corrected modulus field.

[0058] In step S31, the expression for thermal compensation correction of the dynamic elastic modulus at each position of the webbing is as follows:

[0059] in, These are the spatial coordinates along the length of the webbing. For time, For the position coordinates of the webbing ,time The dynamic elastic modulus at that point. The nominal elastic modulus at the reference temperature. The temperature-modulus sensitivity coefficient of the webbing material. This is a distributed thermal field signal. This is a reference temperature.

[0060] Furthermore, the elastic modulus of the webbing material decreases with increasing temperature. While this property has limited impact under normal temperature conditions, in scenarios such as high-temperature operations, near-fire operations, or continuous frictional heat generation, temperature variations can reach tens of degrees Celsius, corresponding to modulus changes of several percentage points. If a constant modulus is still used in the inverse finite element method, a non-negligible systematic error will be introduced. Based on this, step S31 of this invention introduces a micro-thermal thin-film sensor array distributed along the entire length of the webbing to convert the webbing thermal field information into point-by-point dynamic elastic modulus correction values. This ensures that the material parameters of the mechanical reconstruction model remain consistent with the actual physical state at every moment and at every location. Simultaneously, local abnormal hotspot information is incorporated into equipment lifespan management, forming a closed-loop evaluation system that links thermal, mechanical, and lifespan assessments.

[0061] In step S31, the present invention extracts the distributed thermal field signal output by each micro-thermal thin-film sensor node along the entire length of the webbing from the multi-dimensional aligned time-series data stream. This signal records the real-time temperature value at every 10cm interval of the webbing, using spatial coordinates and time as dual indices along the webbing length. Subsequently, the present invention uses 23°C as the reference temperature and the nominal elastic modulus of the webbing material at the reference temperature as the benchmark value. The difference between the real-time temperature value at each node and the reference temperature is multiplied by the temperature-modulus sensitivity coefficient of the webbing material (approximately -0.8% / °C for polyester fiber and approximately -1.1% / °C for nylon) to obtain the modulus correction ratio at each location. Then, this correction ratio is multiplied by the nominal elastic modulus, and the dynamic elastic modulus of the webbing at each spatial coordinate and each time point is calculated point by point. All calculation results are arranged according to the dual indices of spatial coordinates and time to form a temperature-corrected modulus field.

[0062] Furthermore, when the temperature of any node in the distributed thermal field signal rises beyond a preset threshold within a preset time window, this invention records the spatial coordinates of the node and the corresponding timestamp as thermal anomaly warning information and writes it into the alarm queue.

[0063] S32. The feature vector of the fall event is used to infer the physical information neural network model of the temperature correction modulus field input edge, and the approximate solution of the three-dimensional impact load is output within the preset response time limit.

[0064] Furthermore, high-precision solutions for inverse finite element analysis require numerous matrix operations, with a typical latency of seconds for a single solution on a cloud server, which cannot meet the real-time requirement of millisecond-level early warning after a fall. Relying solely on lightweight models at the edge carries the risk of insufficient accuracy under extreme conditions or rare fall scenarios. Therefore, this invention functionally separates the two computational approaches: the edge-side physical information neural network model undertakes the real-time response task, sacrificing a small amount of accuracy for millisecond-level output; the cloud-based inverse finite element arithmetic undertakes the post-fall high-precision reconstruction task, with its results continuously supplementing the training sample set of the edge model. This allows the edge model to continuously approach the accuracy of the cloud model as historical data accumulates, forming a self-optimizing edge-cloud collaborative architecture over time.

[0065] In step S32, the present invention concatenates the fall event feature vector output in step S2 with the temperature correction modulus field output in step S31 as a joint input, and sends it to the edge physical information neural network model that has been deployed on the local low-power edge computing chip of the seat belt. After receiving the above joint input, the model directly performs forward inference operation and outputs the three-dimensional impact load approximation solution within a preset response time limit (set to 50ms in a specific embodiment). The final obtained three-dimensional impact load approximation solution uses the normal force, tangential force and torque of each section of the webbing as fields, covering the entire time course of the fall impact.

[0066] In step S32, the construction process of the edge-end physical information neural network model further includes: S321. Using the Timoshenko beam theory governing equation as the physical constraint term in the cloud, and the data-driven loss term to form a composite loss function, the pre-set basic neural network is trained offline using the historical solution set of cloud inverse finite element analysis as the training sample to obtain the physical information neural network.

[0067] In step S321, this invention selects a basic neural network in the cloud and embeds the Timoshenko beam theory governing equations as physical constraints into the training loss function. The Timoshenko beam theory treats the discrete webbing element as a beam element simultaneously subjected to bending and shear deformation, and its governing equations describe the differential relationships between section rotation, lateral displacement, bending moment, and shear force. For the physical constraint term, the calculation method is as follows: substituting the output of the neural network at each training sample point into the aforementioned differential relationship, the sum of squared residuals is calculated as the physical penalty loss. The data-driven loss term is the mean square error of the difference between the load prediction value output by the neural network and the corresponding label value in the historical solution set of the inverse finite element analysis in the cloud. Based on the above two losses, this invention weights and sums the two losses to form a composite loss function, and uses this composite loss function to perform backpropagation iterative training on the basic neural network until convergence, thus obtaining the physical information neural network.

[0068] Furthermore, the expression for the physical information neural network composite loss function in step S321 is as follows:

[0069]

[0070]

[0071] in, This represents the total value of the composite loss function. This is a data-driven loss term that measures the deviation between the neural network output and the finite element history solution. The physical constraint loss term measures the degree to which the neural network output violates the Timoshenko beam governing equations; These are the physical constraint weighting coefficients, used to balance the relative contributions of the two losses; The total number of training samples; For the first The neural network output load prediction value for each sample; For the first Historical solution results of cloud-based inverse finite element analysis for each sample; This is the residual operator for the Timoshenko beam control equation, used to calculate the difference between the two ends of the equation after substituting the neural network output into the control equation. It is an L2 norm.

[0072] S322. Perform structural pruning and fixed-point quantization compression on the physical information neural network, and deploy the compressed model to the local low-power edge computing chip of the safety belt to obtain the edge physical information neural network model.

[0073] In step S322, the present invention performs structural pruning on the physical information neural network, that is, removes neuron connections with small absolute weight values ​​in the network according to a preset pruning rate, compressing the number of network parameters by about 60%; then performs INT8 fixed-point quantization on the remaining parameters, converting the weights originally stored as 32-bit floating-point numbers into 8-bit integer representations, further compressing the model size and inference computation volume, and finally writes the compressed model into a local low-power edge computing chip (ARM Cortex-M series or FPGA) to obtain the edge physical information neural network model.

[0074] S33. Upload the multidimensional aligned time-series data stream and the temperature-corrected modulus field to the cloud, input the cloud-based webbing human body coupling mechanics model for inverse analysis and solution, and obtain the accurate solution of the three-dimensional impact load.

[0075] Step S33 further includes: S331. Based on the fall mode label in the fall event feature vector, retrieve the corresponding posture template from the pre-built human posture library, and use the elastic basic contact model to parameterize the contact pressure distribution between the webbing and the human torso.

[0076] Furthermore, existing methods for reconstructing webbing mechanics typically assume a fixed pressure distribution between the webbing and the human torso, failing to consider the impact of changes in human posture on the contact interface under different fall directions. However, in reality, the contact area, contact center location, and pressure distribution between the webbing and the human torso differ significantly under vertical fall, forward-leaning fall, and lateral swing modes. Using a uniform contact model for all fall directions would distort the boundary conditions of the internal forces at each cross-section, thus affecting the accuracy of the three-dimensional impact load solution. Therefore, this invention uses the fall mode labels classified in step S2 as prerequisites for contact modeling. It retrieves the corresponding posture template from a pre-built human posture library and then uses the Winkler elastic basic contact model to parameterize the contact pressure distribution under that posture. This ensures that the boundary conditions relied upon for the inverse finite element solution remain physically consistent with the actual fall posture, avoiding a disconnect between the contact model and the actual fall state.

[0077] In step S331, the present invention reads the fall pattern label from the fall event feature vector and retrieves the corresponding posture template from a pre-built human posture library based on the label. This template describes the contact area distribution between the human torso and the webbing under a specific fall direction. Subsequently, the present invention uses the Winkler elastic basic contact model to parameterize the pressure distribution within the contact area. This model discretizes the contact interface into several elastic support points, and the contact pressure at each support point is proportional to the normal deformation of the webbing at that point, thereby transforming the continuous contact pressure distribution into a finite number of undetermined parameters.

[0078] S332. Using the distributed strain field of the fiber Bragg grating sensor in the multidimensional aligned time-series data stream as the boundary condition, the temperature-corrected modulus field is substituted into the Timoshenko beam discrete element control equation. With the goal of minimizing the norm of the difference between the simulated strain and the measured strain of each element, the Levenberg-Marquardt algorithm is used to iteratively solve the tension, shear force and bending moment of each section to obtain the accurate solution of the three-dimensional impact load.

[0079] In step S332, the present invention extracts the distributed strain field output by the fiber Bragg grating sensor at each measuring point of the webbing from the multidimensional aligned time-series data stream and uses it as the boundary condition for mechanical solution; the dynamic elastic modulus at the corresponding position of each cross section in the temperature-corrected modulus field is substituted into the Timoshenko beam discrete element control equation, replacing the original constant modulus, so that the control equation uses temperature-corrected material parameters at each moment and at each cross section.

[0080] Specifically, this invention discretizes the webbing along its length into 200 equal-length elements. The L2 norm of the difference between the simulated strain calculated by the governing equation and the measured strain of the fiber Bragg grating for each element is used as the optimization objective. The Levenberg-Marquardt algorithm is used to iteratively solve for the tension, shear force and bending moment of each section.

[0081] In each iteration, the Levenberg-Marquardt algorithm calculates the Jacobian matrix of the objective function with respect to the internal forces at each unknown cross-section. It then adaptively adjusts the search direction and step size based on the damping factor, dynamically switching between gradient descent and the Gauss-Newton method until the L2 norm converges to within 0.5%. Finally, it outputs the complete time histories of tension, shear force, and bending moment at each cross-section, forming the exact solution for the three-dimensional impact load. This exact solution is then fed back to the edge processing unit to supplement the training sample set of the physical information neural network, used to update the model weights during the next round of offline training.

[0082] S4. Calculate the personalized effective impact amount based on the approximate solution of the three-dimensional impact load and the user's personalized biomechanical parameter file, compare it with the damage threshold database of different parts, and input the historical peak tension of each section in the accurate solution of the three-dimensional impact load into the fatigue damage accumulation model, and output a fall stress test report including the fall stress safety level and the ratio of the remaining bearing life of the webbing.

[0083] Current seatbelt life management methods typically rely solely on the cumulative number of impacts or visual inspection results to determine whether the equipment needs replacement, failing to incorporate the accelerated aging effect of the webbing material under thermal conditions into quantitative assessments. The standard Miner criterion describes the remaining life of the material by linearly accumulating the proportion of fatigue damage under each impact load, based on the assumption that material properties remain constant throughout the service life. However, under conditions such as near-fire operations and continuous frictional heat generation, the molecular chain breakage rate of the webbing material increases rapidly, resulting in an actual fatigue damage rate under the same load that is much higher than under normal temperature conditions.

[0084] Based on the aforementioned technical problems, this invention transforms the distributed thermal field signal and thermal anomaly early warning information recorded in real time in step S31 into a thermally accelerated aging coefficient specific to each cross-section. This coefficient is multiplied into the damage accumulation term of the standard Miner criterion, allowing the fatigue damage rate of each cross-section to be dynamically adjusted according to its actual thermal exposure history. Thus, the remaining load-bearing life of each cross-section of the webbing is simultaneously driven by both mechanical fatigue damage and thermally induced material degradation. The thermal field distribution information originates from step S31, is transmitted through the thermal anomaly early warning queue to the life model in step S4, forming a complete thermo-mechanical coupled life assessment data chain.

[0085] In step S4, the expression for the personalized effective impact amount is:

[0086] in, To achieve personalized and effective impact, The moment of the shock event. The time history of the resultant impact force is extracted from the approximate solution of the three-dimensional impact load. For the duration of the impact, The user's soft tissue response time constant in the user's personalized biomechanical parameter profile. For time.

[0087] In step S4, the expression for the fatigue damage accumulation model is:

[0088] in, Spatial number index for the webbing cross section, For the first The ratio of remaining load-bearing life of the webbing at each cross section, For impact event indexing, This represents the total number of impact events experienced by the corresponding cross section. cross section spatial coordinates, cross section In the The peak tension corresponds to the impact event. For the first The number of load cycles experienced by the cross section during the impact event. For the cross section at the corresponding peak tension The theoretical fatigue life is as follows. The coefficient for accelerated aging due to heat.

[0089] In specific step S4, the present invention extracts the impact resultant force time history from the three-dimensional impact load approximation solution, takes the start time of the impact event as the lower limit of integration and the start time plus the impact duration as the upper limit of integration, multiplies the impact resultant force time history point by point with the exponential weighted function with the user's soft tissue response time constant as the characteristic parameter, and then integrates to obtain the personalized effective impact amount, as shown in the above formula.

[0090] The physical meaning of the exponential weighting function is as follows: the response of human soft tissue to impact force has a time delay, and the response amplitude increases exponentially with time. The larger the user's soft tissue response time constant, the slower the soft tissue response to the same impact force. The smaller the integral result, the stronger the buffering characteristics of the user's soft tissue to impact.

[0091] After calculating the personalized effective impact amount, this invention compares the personalized effective impact amount with the injury threshold database established by body part and population category, and outputs the first level of the four levels (green, yellow, orange, red) as the fall force safety level based on the comparison results.

[0092] Subsequently, this invention extracts the peak tension of each impact event from the accurate solution of the three-dimensional impact load section by section and inputs it into the fatigue damage accumulation model. This model is based on the modified Miner criterion. The core idea of ​​the Miner criterion is that the material accumulates a certain proportion of fatigue damage under each cycle of load, and the material fails when the sum of the proportions of each damage reaches 1.

[0093] Based on the standard Miner criterion, this invention introduces the thermal accelerated aging coefficient of the corresponding node in the thermal anomaly warning information recorded in step S31 into the denominator. The thermal accelerated aging coefficient is calculated based on the cumulative thermal exposure history of the cross section. When its value is greater than 1, it indicates that the cross section is accelerating the consumption of fatigue life due to thermal material degradation.

[0094] Based on the principle of fatigue life calculation, this invention calculates the remaining load-bearing life ratio for each of the 200 discrete sections of the webbing. When the remaining load-bearing life ratio of any section is lower than the first preset threshold, the section is marked as a forced replacement warning; when it is lower than the second preset threshold, an immediate shutdown lockout command is triggered.

[0095] Ultimately, this invention summarizes and encapsulates the fall stress safety level, the remaining bearing life ratio of each section, the forced replacement warning, and the shutdown lock command into a fall stress test report, which is then uploaded to the cloud safety management platform via a wireless communication module.

[0096] like Figure 2 As shown, the present invention also provides a seatbelt fall force detection system, comprising: Acquisition module 100: Used to synchronously acquire the stress state of the webbing through a heterogeneous sensor array deployed at key stress nodes of the seat belt, and input the acquired multidimensional aligned time-series data stream into the biomechanical identification model to obtain a personalized biomechanical parameter profile for the user. Identification module 200: used to perform time-frequency joint analysis and impact dynamics calculation on multidimensional aligned time-series data streams, filter candidate fall events from a preset fall discrimination rule set, classify the candidate fall events by direction, and obtain the fall event feature vector; Calculation module 300: used to perform thermal compensation correction on the dynamic elastic modulus of multiple positions of the webbing based on multidimensional aligned temporal data stream to obtain temperature correction modulus field; input the fall event feature vector and the temperature correction modulus field into the edge physical information neural network model and the cloud webbing human body coupling mechanical model respectively to obtain the approximate solution and the accurate solution of the three-dimensional impact load; Output module 400: It is used to calculate the personalized effective impact amount based on the approximate solution of the three-dimensional impact load and the user's personalized biomechanical parameter file, compare it with the damage threshold database of different parts, and input the historical peak tension of each section in the accurate solution of the three-dimensional impact load into the fatigue damage accumulation model, and output a fall stress test report including the fall stress safety level and the ratio of the remaining bearing life of the webbing.

[0097] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0098] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the parts that contribute to the prior art, can be embodied in the form of software products. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., including several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the seat belt fall force detection method described in various embodiments or some parts of embodiments.

[0099] This invention achieves personalized and accurate assessment of damage risk by constructing a user-specific biomechanical parameter profile, significantly improving the physiological realism of fall stress detection. Secondly, it introduces a dual-modal solution mechanism combining edge-side physical information neural networks and cloud-based inverse analysis, balancing millisecond-level real-time emergency warning with high-fidelity load reconstruction. Furthermore, through thermal compensation correction and a fatigue damage accumulation model, it effectively eliminates the interference of ambient temperature on material properties, achieving refined health management of the webbing throughout its entire life cycle, and significantly improving the intelligence and reliability of seat belt protection.

[0100] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for detecting the force exerted by a seatbelt during a fall, characterized in that, include: S1. By deploying a heterogeneous sensor array at key stress nodes of the seat belt, the stress state of the webbing is collected synchronously, and the collected multi-dimensional aligned time-series data stream is input into the biomechanical identification model to obtain a personalized biomechanical parameter profile for the user. S2. Perform time-frequency joint analysis and impact dynamics calculation on the multidimensional aligned time-series data stream, filter the fall candidate events by the preset fall discrimination rule set, classify the fall candidate events by direction, and obtain the fall event feature vector. S3. Based on the multi-dimensional aligned temporal data stream, the dynamic elastic modulus of multiple positions of the webbing is thermally compensated and corrected to obtain the temperature correction modulus field; the feature vector of the fall event and the temperature correction modulus field are respectively input into the edge physical information neural network model and the cloud webbing human body coupling mechanical model to obtain the approximate solution and the accurate solution of the three-dimensional impact load. S4. Calculate the personalized effective impact amount based on the approximate solution of the three-dimensional impact load and the user's personalized biomechanical parameter file, compare it with the damage threshold database of different parts, and input the historical peak tension of each section in the accurate solution of the three-dimensional impact load into the fatigue damage accumulation model, and output a fall stress test report including the fall stress safety level and the ratio of the remaining bearing life of the webbing.

2. The method for detecting the force of a seatbelt falling according to claim 1, characterized in that, Step S1 further includes: S11. Piezoresistive strain sensors, triaxial accelerometers and fiber Bragg grating sensors are deployed at the shoulder load-bearing section, waist buckle connection section and back hook point of the safety belt, respectively, and micro-thermal thin film sensor nodes are deployed at preset intervals along the entire length of the webbing to form a heterogeneous sensor array. S12. Align the clocks of each sensor node in the heterogeneous sensor array using a hardware time synchronization protocol to synchronize and quantize all channels at a uniform sampling rate to obtain structured data frames. S13. A multi-dimensional aligned timing data stream is formed by combining structured data frames from each channel; S14. Extract the data segment of the user in normal working activity state from the multidimensional aligned time-series data stream. Use the single-degree-of-freedom spring-damping system model to systematically identify the webbing tension signal and human body acceleration signal in the data segment of normal working activity state, and obtain the user stiffness, user damping and user soft tissue response time constant to form a personalized biomechanical parameter profile of the user.

3. The method for detecting the force of a seatbelt falling according to claim 2, characterized in that, In step S14, the system identification of the webbing tension signal and human acceleration signal in the data segment of normal operation activity state is carried out by recursive Kalman filtering algorithm. The recursive Kalman filtering algorithm takes the webbing tension signal as the excitation input and the human acceleration signal as the response output to perform online back-reasoning of the parameters of the single degree of freedom spring-damped system model. The expression for the user soft tissue response time constant in step S14 is: in, To calculate the obtained user soft tissue response time constant, For user stiffness, For user damping, The natural angular frequency of the human torso vibration.

4. The method for detecting the force of a seatbelt falling according to claim 1, characterized in that, Step S2 further includes: S21. Perform time-frequency joint analysis on the acceleration signal in the multidimensional aligned time-series data stream to extract the time-frequency distribution feature vector of the impact energy. S22. Based on the slope of the rising edge and the duration of the peak value of the acceleration time-domain curve, candidate fall events are obtained by filtering from the impact energy time-frequency distribution feature vector using the fall discrimination rule set. S23. Perform time integration on the webbing tension signal in the candidate fall event, and calculate the equivalent fall height by combining the triaxial composite acceleration peak. S24. Based on the proportion of the three-axis acceleration components in the candidate fall events and the webbing strain asymmetry index output by the fiber Bragg grating sensor, the candidate fall events are classified by direction to obtain fall mode labels. S25, the feature vector of the fall event is composed of the impact energy time-frequency distribution feature vector, equivalent fall height, triaxial composite acceleration peak value, webbing strain asymmetry index and fall mode label combination.

5. The method for detecting the force of a seatbelt falling according to claim 1, characterized in that, Step S3 further includes: S31. Based on the distributed thermal field signal collected by the micro-thermal thin film sensor node in the multidimensional aligned time-series data stream, the dynamic elastic modulus at each position of the webbing is thermally compensated and corrected to obtain the temperature correction modulus field. S32. The feature vector of the fall event is inferred from the physical information neural network model of the temperature correction modulus field input edge, and an approximate solution of the three-dimensional impact load is output within a preset response time limit. S33. Upload the multidimensional aligned time-series data stream and the temperature-corrected modulus field to the cloud, input the cloud-based webbing human body coupling mechanics model for inverse analysis and solution, and obtain the accurate solution of the three-dimensional impact load.

6. The method for detecting the force of a seatbelt falling according to claim 5, characterized in that, In step S31, the expression for thermal compensation correction of the dynamic elastic modulus at various positions of the webbing is: in, These are the spatial coordinates along the length of the webbing. For time, For the position coordinates of the webbing ,time The dynamic elastic modulus at that point. The nominal elastic modulus at the reference temperature. The temperature-modulus sensitivity coefficient of the webbing material. This is a distributed thermal field signal. This is a reference temperature.

7. The method for detecting the force of a seatbelt falling according to claim 5, characterized in that, In step S32, the construction process of the edge-end physical information neural network model further includes: S321. Using the Timoshenko beam theory governing equation as the physical constraint term in the cloud, and the data-driven loss term to form a composite loss function, the historical solution result set of cloud inverse finite element analysis is used as the training sample to train the pre-set basic neural network offline to obtain the physical information neural network. S322. Perform structural pruning and fixed-point quantization compression on the physical information neural network, and deploy the compressed model to the local low-power edge computing chip of the safety belt to obtain the edge physical information neural network model.

8. The method for detecting the force of a seatbelt falling according to claim 5, characterized in that, Step S33 further includes: S331. Based on the fall mode label in the fall event feature vector, retrieve the corresponding posture template from the pre-built human posture library, and use the elastic basic contact model to parameterize the contact pressure distribution between the webbing and the human torso. S332. Using the distributed strain field of the fiber Bragg grating sensor in the multidimensional aligned time-series data stream as the boundary condition, the temperature-corrected modulus field is substituted into the Timoshenko beam discrete element control equation. With the goal of minimizing the norm of the difference between the simulated strain and the measured strain of each element, the Levenberg-Marquardt algorithm is used to iteratively solve the tension, shear force and bending moment of each section to obtain the accurate solution of the three-dimensional impact load.

9. The method for detecting the force of a seatbelt falling according to claim 1, characterized in that, In step S4, the expression for the personalized effective impact amount is: in, To achieve personalized and effective impact, The moment of the shock event. The time history of the resultant impact force is extracted from the approximate solution of the three-dimensional impact load. For the duration of the impact, The user's soft tissue response time constant in the user's personalized biomechanical parameter profile. For time; In step S4, the expression for the fatigue damage accumulation model is: in, Spatial number index for the webbing cross section, For the first The ratio of remaining load-bearing life of the webbing at each cross section, For impact event indexing, This represents the total number of impact events experienced by the corresponding cross section. cross section spatial coordinates, cross section In the The peak tension corresponds to the impact event. For the first The number of load cycles experienced by the cross section during the impact event. For the cross section at the corresponding peak tension The theoretical fatigue life is as follows. The coefficient for accelerated aging due to heat.

10. A seatbelt fall force detection system, used to perform a seatbelt fall force detection method as described in any one of claims 1 to 9, characterized in that, include: The data acquisition module is used to synchronously acquire the stress state of the webbing through a heterogeneous sensor array deployed at key stress nodes of the seat belt, and input the acquired multidimensional aligned time-series data stream into the biomechanical identification model to obtain a personalized biomechanical parameter profile for the user. The identification module is used to perform time-frequency joint analysis and impact dynamics calculation on multidimensional aligned time-series data streams. It selects candidate fall events from a preset set of fall discrimination rules, classifies the candidate fall events by direction, and obtains the feature vector of the fall event. The calculation module is used to perform thermal compensation correction on the dynamic elastic modulus of multiple positions of the webbing based on the multidimensional aligned temporal data stream to obtain the temperature correction modulus field; the feature vector of the fall event and the temperature correction modulus field are respectively input into the edge physical information neural network model and the cloud webbing human body coupling mechanical model to obtain the approximate solution and the accurate solution of the three-dimensional impact load; Output module: Used to calculate personalized effective impact amount based on the approximate solution of three-dimensional impact load and the user's personalized biomechanical parameter file, compare it with the damage threshold database of different parts, and input the historical peak tension of each section in the accurate solution of three-dimensional impact load into the fatigue damage accumulation model, and output a fall stress test report including the fall stress safety level and the ratio of the remaining bearing life of the webbing.

Images