Garment damage positioning method and system based on nanofiber sensing grid

Abstract

The invention relates to a garment damage positioning method and system based on a nanofiber sensing grid. The method comprises the following steps: acquiring an original impact space-time distribution signal generated by an impact event and a sensor network real-time deformation signal through the nanofiber sensing grid; self-adaptive anti-interference preprocessing is carried out on the original shock wave space-time distribution signals, filtering parameters are dynamically adjusted, and pure shock wave signals are obtained; calculating plane coordinates of the impact points through a time difference positioning method and an optimization algorithm; inverting the penetration depth of an impact object through an energy attenuation model adaptive to the characteristics of a clothing protection medium, and forming a damage three-dimensional coordinate with the plane coordinate of the impact point; and dynamically matching a standardized damage grading threshold according to the energy grade of the pure shock wave signal, mapping the penetration depth in the three-dimensional coordinates of the damage to the grading threshold to obtain a quantitative damage grade, and transmitting the three-dimensional coordinates of the damage and the quantitative damage grade data to a terminal. According to the scheme, real-time accurate monitoring of clothing damage can be realized.

Description

Technical Field

[0001] This invention relates to the field of information monitoring technology, and in particular to a method and system for locating clothing damage based on a nanofiber sensing grid. Background Technology

[0002] In industrial operations, outdoor rescue, and military protection scenarios, clothing serves as a core protective barrier for the human body. Real-time and accurate monitoring of clothing damage directly impacts user safety and has become a core requirement in the field of protective equipment. Currently, clothing damage monitoring faces multiple challenges, including difficulties in integrating multi-source data, weak resistance to interference in complex environments, low positioning and assessment accuracy, and insufficient system coordination, hindering the intelligent upgrading of protective equipment.

[0003] At the data acquisition level, damage monitoring requires the simultaneous acquisition of multimodal signals such as impact and deformation. However, traditional acquisition methods often employ single-signal modes and centralized architectures, resulting in low acquisition efficiency, uneven load distribution, and inconsistent data formats. This leads to cumbersome preprocessing and difficulty in quickly generating standardized datasets. Regarding environmental interference, artifacts caused by human movement and parameter changes due to deformation of the protective medium make it difficult for traditional technologies to distinguish between effective signals and interference, resulting in insufficient signal-to-noise ratio. In the localization and assessment phase, existing technologies often calculate shock wave propagation velocity using fixed medium parameters, neglecting the impact of deformation, leading to large localization errors. Damage levels are mostly qualitative descriptions, lacking quantitative standards that dynamically match impact energy and medium characteristics. In terms of system integration, data silos exist in sensor acquisition, signal processing, and data transmission, making it difficult to form a closed loop for optimization execution and effect evaluation. Furthermore, the lack of a full-process traceability system results in low strategy iteration efficiency. These core deficiencies in existing technologies prevent damage monitoring from meeting the precise requirements for damage localization. Summary of the Invention

[0004] This invention achieves real-time and accurate monitoring of clothing damage through dual-signal collaborative acquisition, intelligent suppression of motion artifacts, dynamic calibration of dielectric coefficients, precise construction of three-dimensional coordinates, and dynamic matching of quantitative damage levels.

[0005] The technical solution proposed in this invention is: a method for locating clothing damage based on a nanofiber sensing mesh, the method comprising: The original spatiotemporal distribution signal of the impact event and the real-time deformation signal of the sensor network are collected by a nanofiber sensing grid. Based on a pre-defined motion artifact feature library, an adaptive anti-interference preprocessing method is used to perform spatiotemporal distribution signal of the original shock wave through a convolutional neural network model. The filtering parameters are dynamically adjusted according to the human motion state to filter out interference signals and artifacts, and obtain a pure shock wave signal. Based on the pure shock wave signal, the signal reception time difference of multiple sensor nodes is extracted, and dynamic medium coefficient calibration is performed by combining the real-time deformation signal of the sensor network. The plane coordinates of the impact point are calculated by the time difference positioning method and optimization algorithm. Energy correlation calculations are performed on the pure shock wave signal, and the plane coordinates of the impact point and the dynamic medium coefficient are integrated. The penetration depth of the impacting object is inverted by an energy attenuation model adapted to the characteristics of the protective medium of clothing, and combined with the plane coordinates of the impact point to form a three-dimensional damage coordinate system. Based on the energy level of the pure shock wave signal, a standardized damage grading threshold is dynamically matched, and the penetration depth in the three-dimensional coordinates of the damage is mapped to the grading threshold to obtain the quantitative damage level. The three-dimensional coordinates of the damage and the quantitative damage level are then transmitted to the terminal.

[0006] Preferably, the specific process for acquiring the original impact spatiotemporal distribution signal and the real-time deformation signal of the sensor network is as follows: By utilizing a pre-installed nanofiber sensing network and relying on the piezoelectric effect, the mechanical physical quantities generated by the impact event are converted in real time into the original impact spatiotemporal distribution signal in the form of an electrical signal containing amplitude and phase information. The nanofiber sensing network uniformly covers the key protective areas of the clothing. By leveraging the strain sensing characteristics of sensor networks, local deformations caused by human movement, such as stretching, bending, or impact, can be accurately captured, and real-time deformation signals including the degree and rate of deformation can be collected simultaneously. The high-precision clock synchronization module performs real-time calibration of the acquisition timing of the two types of signals based on the clock synchronization protocol, aligning the signal acquisition timestamps frame by frame to ensure that the acquisition times of the two are completely consistent.

[0007] Preferably, the construction process of the motion artifact feature library and the convolutional neural network model is as follows: Various interference signal samples generated in different scenarios were collected, and after wavelet transform denoising and maximum-minimum normalization preprocessing, key features were extracted and the categories were labeled. Feature samples are organized and stored in a unified data format system to form a structured motion artifact feature library containing different types and intensities of interference; Using the time and frequency domain features of the original signal as input and the filter parameter adjustment command as output, the convolutional neural network model is trained with feature library samples. The gradient descent method is used to optimize the loss function, and the number of iterations and learning rate are set until the model's prediction accuracy meets the preset requirements.

[0008] Preferably, the specific process of dynamically adjusting the filter parameters is as follows: By using the sliding window method for time-domain peak detection and the fast Fourier transform for frequency-domain analysis, the amplitude and frequency distribution of the real-time deformation signal of the sensor network can be accurately extracted. The extracted features are compared one by one with multiple preset thresholds established based on statistical analysis of a large amount of historical interference data. If the amplitude of the deformation signal exceeds the preset threshold of the corresponding range, the filter passband will be automatically narrowed proportionally. The greater the amplitude of exceeding the threshold, the narrower the passband. If interference components in a specific frequency band are detected, notch filtering technology is used to accurately attenuate the signal in that frequency band, thereby achieving dynamic adaptation and adjustment of the filtering parameters.

[0009] Preferably, the calibration process for the dynamic medium coefficient is as follows: Multiple control experiments were designed according to the deformation degree gradient and impact intensity gradient. Multiple parallel experiments were set up in each group to reduce errors. The deformation displacement, impact velocity and corresponding media propagation coefficient were recorded in detail under different deformation degrees and impact conditions. Based on experimental data, a nonlinear correlation model between the degree of deformation and the medium propagation coefficient was constructed using polynomial fitting. The peak displacement of the deformation signal is extracted in real time as the deformation parameter, and then substituted into the correlation model to calculate the calibrated medium coefficient. The calibrated medium coefficient is applied to the time difference positioning method to accurately correct the shock wave propagation velocity parameters.

[0010] Preferably, the specific process for calculating the planar coordinates of the impact point is as follows: The sensor nodes are evenly and orthogonally arranged in the key protective area of ​​the garment lining according to the preset grid spacing. The node spacing is set according to the positioning accuracy requirements. The x and y axis spatial coordinate information of each node is accurately calibrated using a three-dimensional coordinate measuring device. Based on the spatial coordinates of each node and the signal reception time difference, a set of nonlinear equations corresponding to the time difference positioning method is constructed. Select the appropriate algorithm based on the requirements of solution efficiency and accuracy, and perform iterative calculations with the equation system residuals meeting the preset threshold as the termination condition. The plane coordinates of the impact point are obtained by minimizing the residuals of the system of equations.

[0011] Preferably, the specific process for forming the three-dimensional coordinates of the damage is as follows: A fixed integration time window is set according to the typical duration of the shock wave signal to ensure complete capture of shock energy information. Time-domain integration is performed on the pure shock wave signal to obtain shock energy-related parameters including signal integral amplitude, peak energy, and energy density. Common clothing protective media were selected, and a comparative experiment was conducted with multiple impact intensities. The measured values ​​of the penetration depth of the impactor were recorded under different energy parameters. The system was then calibrated to obtain an energy attenuation model that is adapted to the characteristics of clothing protective media. The impact energy-related parameters, the plane coordinates of the impact point, and the dynamically calibrated medium coefficient are all substituted into the energy attenuation model. The penetration depth of the impactor is then obtained by inverting the mapping relationship between the model input parameters and the measured depth. The impact penetration depth obtained by inversion is combined with the plane coordinates of the impact point to form a complete three-dimensional damage coordinate system that includes the plane coordinates of the x and y axes and the depth coordinates of the z axis.

[0012] Preferably, the specific process for obtaining the quantitative damage level and implementing the data transmission step is as follows: Based on the impact resistance parameters and damage risk levels of the protective medium for clothing, the energy levels of pure shock wave signals are divided into multiple preset ranges; A standardized damage grading threshold is configured for each energy range, and the threshold setting is matched with the damage limit of the protective medium. The energy range is determined based on the calculated impact energy parameters, and the corresponding classification threshold is automatically matched. The inverted penetration depth is compared with the matched grading threshold, and the quantitative damage level is obtained by mapping according to the preset rules. The three-dimensional coordinates and grade data of the damage are transmitted to the terminal device via a wireless transmission protocol, and the terminal analyzes and processes the received data.

[0013] The present invention also provides a clothing damage localization system based on a nanofiber sensing mesh, the system being used to execute the aforementioned clothing damage localization method based on a nanofiber sensing mesh.

[0014] The present invention also provides a computer-readable storage medium storing a computer program that is executed by a processor to implement the aforementioned method for locating clothing damage based on a nanofiber sensing mesh.

[0015] The beneficial effects of this invention are: 1. By synchronously acquiring impact spatiotemporal distribution signals and sensor network deformation signals, and combining motion artifact feature libraries and convolutional neural network models to dynamically adjust filtering parameters, the system effectively distinguishes between human motion interference and valid impact signals. This improves the signal-to-noise ratio from less than 20dB in traditional technologies to over 25dB, with interference attenuation ≥30dB, completely solving the pain point of high signal distortion in complex scenarios. This design ensures the purity of subsequent data processing, providing high-quality data support for positioning and inversion stages. It enables the system to operate stably in various scenarios, including daily activities and strenuous exercise, adapting to the protection and monitoring needs of multiple scenarios such as industrial operations and outdoor rescue.

[0016] 2. Based on real-time calibration of the medium propagation coefficient using deformation signals, and combined with time-of-flight positioning and energy attenuation model inversion of penetration depth, this method overcomes the accuracy bottleneck caused by fixed medium parameters in traditional technologies. The impact point planar positioning error is ≤2cm, and the penetration depth inversion error is ≤0.1mm, achieving precise three-dimensional damage characterization. This allows users to intuitively grasp core damage information. For example, in military scenarios, it can quickly determine the effectiveness of combat clothing protection; in industrial scenarios, it can accurately locate high-frequency damage areas, providing data support for the maintenance and optimization of protective equipment and significantly reducing the risk of protective failure.

[0017] 3. By dynamically matching grading thresholds according to the impact energy range and the characteristics of the protective medium, a quantitative damage standard of levels I-IV is formed. Combined with a full-process design including encrypted data transmission, terminal visualization, statistical analysis, and fault self-diagnosis, this not only achieves objective quantitative assessment of damage levels but also solves the problem that traditional qualitative descriptions are insufficient for evaluating protective effectiveness. Simultaneously, the closed-loop mechanism supports offline data caching, cross-platform synchronization, and historical traceability, providing a complete data chain for optimizing protective clothing performance and assessing risks in usage scenarios. This helps in developing targeted safety measures and promotes the transformation of protective monitoring from passive inspection to proactive early warning. Attached Figure Description

[0018] Figure 1 This is a flowchart of a method for locating clothing damage based on a nanofiber sensing mesh; Figure 2 This is a flowchart illustrating the location process of a clothing damage location method based on a nanofiber sensing mesh. Detailed Implementation

[0019] The following description is intended to disclose the present invention and enable those skilled in the art to implement it. The preferred embodiments described below are merely examples, and other obvious variations will occur to those skilled in the art. The basic principles of the invention defined in the following description can be applied to other embodiments, modifications, improvements, equivalents, and other technical solutions that do not depart from the spirit and scope of the invention.

[0020] It is understood that the term "a" should be understood as "at least one" or "one or more," that is, in one embodiment, the number of an element can be one, while in another embodiment, the number of the element can be multiple, and the term "a" should not be understood as a limitation on the number.

[0021] like Figure 1 and Figure 2As shown, dual signals are acquired through a nanofiber sensor network; a convolutional neural network model is constructed based on a motion artifact feature library to perform adaptive anti-interference preprocessing on the original signals; the dynamic medium coefficient is calibrated by combining deformation signals, and the planar coordinates of the impact point are calculated using a time-difference positioning method and optimization algorithm; the penetration depth is inverted through multi-parameter fusion to form the three-dimensional coordinates of the damage; and a quantitative damage level is obtained by dynamically matching a grading threshold and transmitting it to the terminal. Through dual signal acquisition, adaptive anti-interference, dynamic calibration positioning, three-dimensional coordinate construction, and grading transmission, a fully implementable clothing damage positioning technology system is constructed.

[0022] Furthermore, the original impact spatiotemporal distribution signal and the real-time deformation signal of the sensor network are collected. The specific details are as follows: Utilizing a pre-installed nanofiber sensor network, relying on the piezoelectric effect, the mechanical physical quantities generated by the impact event are converted in real-time into electrical signals containing amplitude and phase information, representing the original impact spatiotemporal distribution signal. The nanofiber sensor network uniformly covers key protective areas of the clothing (chest: a rectangular area extending horizontally to the left and right, and vertically upwards and downwards, 15cm from the center of the sternum, covering the heart and thoracic organs; back: a rectangular area extending horizontally to the left and right, and vertically to the top and bottom, 18cm from the scapula to the lumbar spine, with the spine as the midline, focusing on protecting the spine and back muscles; shoulders: A circular area with a diameter of 15cm centered on the acromion is designed to match the movement trajectory of the shoulder joint. Utilizing the strain sensing characteristics of a sensor network, it accurately captures local deformations caused by stretching, bending, or impact due to human movement. Simultaneously, it collects real-time deformation signals including the degree of deformation (0-50%, measurement resolution 0.1%) and the deformation rate (0-10cm / s, measurement accuracy ±0.05cm / s). A high-precision clock synchronization module, based on the PTPv2 protocol, performs real-time calibration of the acquisition timing of both types of signals, aligning the signal acquisition timestamps frame by frame (accurate to the millisecond level, timestamp error ≤10μs) to ensure complete consistency in acquisition time, with a timing deviation ≤500ms.

[0023] The nanofiber sensing network is an orthogonally woven structure, made of piezoelectric nanomaterials (either zinc oxide nanowires or barium titanate nanoparticles, with zinc oxide nanowires having an aspect ratio ≥50:1 and purity ≥99.5%; and barium titanate nanoparticles having a particle size / diameter of 50-200 nm and crystallinity ≥98%) and a flexible polymer substrate (polyvinylidene fluoride (PVDF) and polyurethane (PU) in a mass ratio of 7:3, with PVDF having a molecular weight of 500,000-800,000 and PU having a Shore hardness of A85-90) through melt blending (temperature 180-200℃, rotation speed 300-500 r / min) and wet spinning (spinneret diameter 0.1-0.2 mm, coagulation bath temperature 20-25℃, draw ratio 2-3 times). The fiber diameter is 5-20 μm (diameter variation coefficient ≤5%), and the sensing sensitivity error is ≤±3%. The node density in the critical protection area is 20-30 nodes / cm².2 5-10 per cm in non-critical areas 2 The node spacing is dynamically adjusted according to the positioning accuracy. When the positioning accuracy is ≤2cm, the node spacing is ≤3cm; when the positioning accuracy is ≤5cm, the node spacing is ≤5cm. The nodes use gold-plated electrodes (thickness 0.5-1μm, conductivity ≥4.5×10⁻⁶). 7 The conductivity and corrosion resistance are improved by bonding with conductive adhesive (bonding strength ≥ 0.5 N / cm). The clock synchronization module uses a DS3231 temperature-compensated crystal oscillator with frequency stability ≤ ±1 ppm and operating voltage of 3.3V. Synchronous sampling is achieved through hardware triggering (rising edge triggering, trigger level 3.3V). The acquisition device integrates a signal conditioning unit with a built-in low-noise preamplifier (model INA128, gain 100-1000 times, input noise ≤ 1 μVrms@1kHz) and a 0.1-10kHz passive filter circuit. The original impact spatiotemporal distribution signal sampling frequency is ≥ 1kHz, sampling accuracy is 16 bits, and voltage range is ±5V; the real-time deformation signal sampling frequency is ≥ 500Hz, deformation measurement range is 0-50%, and measurement accuracy is ≤ ±1%.

[0024] In detail, the specific implementation logic of this step is as follows: The nanofiber sensing network is embedded into the clothing lining (0.5-1mm breathable polyester fabric interlayer, air permeability ≥500mm / s) through a molding and bonding process (temperature 80-100℃, pressure 0.3-0.5MPa, heat and pressure holding for 10-15s), avoiding joint folds (such as elbow and knee joint folds with a radius ≤5mm), with a gap of ≤2mm between it and the human skin, and a movement restriction rate of ≤5% after wearing. After data acquisition is initiated, the sensor network is powered by a constant 3.3V voltage (ripple ≤10mV, power supply module is an LDO voltage regulator chip). When the impact signal is ≥0.5N, the piezoelectric nanomaterial converts physical quantities such as force and velocity into mV-level electrical signals (signal amplitude and impact force linear correlation coefficient ≥0.98, nonlinear error ≤2%). After being amplified to V level by a preamplifier and filtered by a filter circuit to remove high-frequency noise ≥10kHz (attenuation ≥40dB), the signal is transmitted to the STM32H743 data acquisition module via the SPI interface (clock frequency 1MHz, transmission bit width 8 bits). Simultaneously, the strain sensing channel, based on the resistance strain effect (sensitivity coefficient K=2.0-2.2, temperature coefficient ≤±50ppm / ℃), converts deformation displacement into resistance change (resistance change rate is proportional to deformation). This is converted into a voltage signal via a Wheatstone bridge (power supply voltage 5V, balance resistor accuracy ±0.1%), and transmitted in parallel with the impact signal, with a transmission delay ≤10μs. The clock synchronization module generates a synchronization pulse every 10ms to trigger synchronization sampling. Every 100ms, it reads the timing deviation value through the register. If the deviation exceeds 300ms, it starts the calibration process by resynchronizing the crystal oscillator clock. The time taken is ≤10ms, and the timing deviation after calibration is ≤100ms.

[0025] Specifically, the data acquisition process and equipment details are as follows: The data acquisition process is as follows: 1. Equipment check: Initiate a Bluetooth connection test via the terminal APP. The sensor network communication success rate is ≥99.5% (no more than 3 communication failure nodes), the battery voltage is ≥3.0V, and the sensor network insulation resistance is ≥10MΩ; 2. Initialization: Start the data acquisition and clock synchronization module, complete parameter configuration (sampling frequency 1kHz / 500Hz, gain 100 / 500 / 1000 times, impulse trigger threshold 0.5N) and self-calibration (calibrate the gain and offset through the internal reference signal, with a calibration error ≤0.5%); 3. Fitting confirmation 1. **Confirmation:** The terminal APP sends a capacitance detection command. A change in node capacitance ≤ 5pF is considered as no looseness or obstruction. The detection time is ≤ 10s. If the detection fails, the user is prompted to adjust the clothing fit. 2. **Acquisition Start:** After an impact event is triggered, the acquisition module simultaneously captures dual signals, and the clock synchronization module records the timestamp to complete timing alignment. 3. **Data Temporary Storage:** Data is stored in the Flash cache in the format "Timestamp (accurate to ms) - Impact signal amplitude (mV) - Deformation degree (%) - Deformation rate (cm / s) - Signal quality (SNR)" (erasable / write cycles ≥ 100,000, storage life ≥ 10 years). **Equipment Core Parameters:** Sensor network static power consumption ≤ 10mA, dynamic power consumption ≤ 30mA, response time ≤ 1ms; Clock synchronization module weight ≤ 10g, dimensions 30mm × 20mm × 5mm, built-in 1000mAh lithium battery, battery life ≥ 8 hours; Acquisition module supports BLE5.0 transmission (communication distance ≥ 10m, transmission delay ≤ 100ms, bit error rate ≤ 100ms). Built-in ≥16GB storage, capable of offline storage of ≥1000 sets of data, and adaptable to temperatures down to -20℃. 60℃, relative humidity 10% 90%, the network can withstand bending ≥10,000 times (bending radius ≥5mm) and washing ≥50 times (water temperature ≤40℃).

[0026] Furthermore, a motion artifact feature library and a convolutional neural network model are constructed. The specific details are as follows: ≥1000 sets of various interference signal samples are collected from different scenarios (daily walking, running, jumping, bending over, mountain climbing, etc.), each set lasting ≥10 seconds (including the complete action cycle, with the action repeated ≥3 times). After wavelet transform denoising and minimax normalization preprocessing, key features in the time domain (peak value, kurtosis, pulse width, rise time) and frequency domain (center frequency, spectral bandwidth, spectral peak energy, harmonic distortion) are extracted and labeled (interference type: motion artifact feature library). (Motion / environmental interference; interference intensity: 1-5 levels); stored in the format of "sample identifier-interference type-intensity level-feature vector-acquisition scene" to form a structured motion artifact feature library; using 10-dimensional feature vectors as input and filter parameter adjustment instructions as output, a convolutional neural network model is trained using the feature library samples, and the mean squared error (MSE) loss function is optimized using the Adam optimizer, iterating for 50-100 rounds, with an initial learning rate of 0.001. If the validation set loss does not decrease for 5 consecutive rounds, it is halved until the prediction accuracy is ≥95% and the output error is ≤5%.

[0027] The interference samples cover 5 types and 5 intensity levels (Level 1 ≤ 0.1V, Level 5 ≥ 0.8V), with ≥ 20 samples of each type and intensity. Wavelet transform denoising uses the db4 wavelet basis, with 3-5 decomposition layers (5 layers when noise intensity ≥ 0.3V, 3 layers when 0.1-0.3V). Noise is removed using a heuristic thresholding method (the threshold equals the square root of the product of the signal-noise standard deviation and twice the natural logarithm and the signal length), resulting in an SNR improvement of ≥ 10dB after denoising. The maximum and minimum normalization calculation formula is: the normalized value equals the original value minus the minimum value, then divided by the difference between the maximum and minimum values, mapping the data to the [0,1] interval. The feature library is stored in SQLite, supporting fast retrieval within 50ms, and is updated with ≥ 100 new scene samples every 3 months. Convolutional Neural Network Model Structure: Input Layer (10 nodes) → Convolutional Layer 1 (3×3 kernels, 32 nodes, ReLU activation function) → Pooling Layer 1 (2×2 max pooling) → Convolutional Layer 2 (3×3 kernels, 64 nodes, ReLU activation function) → Pooling Layer 2 (2×2 max pooling) → Fully Connected Layer 1 (128 nodes, dropout 0.2) → Fully Connected Layer 2 (64 nodes, dropout 0.2) → Output Layer (3 nodes, Sigmoid activation function), outputting 3 types of filter parameter instructions (passband range, attenuation coefficient, filter type).

[0028] In detail, the specific implementation logic of this step is as follows: Features are extracted using a sliding window method with a 500ms window and a 100ms step size, with a window overlap rate of 50%, and each signal segment has 512 sampling points (at 1kHz sampling); the signal is first preprocessed by zero-mean normalization, and then time-domain features are extracted: the peak value is the maximum value of the signal within the window (e.g., 0.35V for running interference signal peak, 1.2V for impact signal peak), and the kurtosis is equal to the expected value of the fourth power of the difference between each data point in the signal and the signal mean, divided by the fourth power of the signal standard deviation (the kurtosis of impact signals is usually ≥3, and the kurtosis of motion interference signals is ≥3). Interference ≤ 2), pulse width is the duration for which the signal amplitude exceeds 50% of the peak value (impact signal < 100ms, interference signal > 500ms), rise time is the time for the signal to rise from 10% to 90% of the peak value (impact signal < 10ms); frequency domain features are obtained through 1024-point FFT transformation (Hanning window weighting, frequency resolution ≤ 1Hz), center frequency is the weighted average of spectral energy, spectral bandwidth is the frequency range between half-power points, peak energy is normalized to 0-1, and harmonic distortion is equal to the ratio of harmonic energy to fundamental energy (≤ 5% is considered valid). Sample annotation process: 3 professionals with more than 5 years of signal processing experience independently annotate the samples. The annotation results are verified by Kappa coefficient consistency (Kappa ≥ 0.8 is acceptable). The machine performs preliminary classification based on the K-means clustering algorithm (number of clusters = 10, iterations 100 times). Samples with annotation consistency ≥ 90% are directly included in the feature library. Inconsistent samples are determined by consultation among the 3 people. The model was trained by dividing the training set (700 sets), validation set (200 sets), and test set (100 sets) in a 7:2:1 ratio. The training batch size was 32. The MSE loss function gradually converged with iteration (initial loss ≥0.1, after training ≤0.01). The generalization performance was tested on different populations (height 150-190cm, weight 45-90kg, age 18-60 years), and the accuracy was ≥92%.

[0029] Specifically, the model training and optimization details are as follows: The model input contains 8 core features (4 time-domain + 4 frequency-domain) and 2 scene-aided features (action type encoding 0-9, environmental noise level 1-3). The input data is standardized to the [0,1] interval using Min-Max. The Adam optimizer is used during training, with parameters set to β1=0.9, β2=0.999, and ε=1e-7. Weights are iteratively adjusted using gradient descent, with weights initialized using a He normal distribution and bias initialized to 0. The dropout layer randomly deactivates 20% of neurons to suppress overfitting (accuracy difference between training and validation sets ≤3%). After training, the model is quantized to INT8 precision using the TensorFlowLite framework (quantization error ≤2%) and deployed on an NPU chip (computing power ≥1 TOPS, power consumption ≤5W). Single-sample inference time ≤10ms and batch inference (32 samples) time ≤200ms. Optimization mechanism: For every 1000 new interference samples accumulated, the dataset is re-divided and iterated for 50 rounds of training; every 3 months, an update package (≤5MB in size) is pushed through the terminal APP, supporting breakpoint resume and version rollback, and users can choose to upgrade.

[0030] Furthermore, the filtering parameters are dynamically adjusted, as detailed below: A sliding window method with a 100ms window and a 50ms step size is used to detect time-domain peak values; 1024-point FFT analysis is performed to analyze frequency domain characteristics and extract the amplitude and frequency distribution of the deformation signal; the signal is compared with three preset thresholds; if the amplitude exceeds the threshold, the passband is narrowed proportionally (narrowed by 20% for every 0.1V increase), with a maximum narrowing of 50%; if interference in a specific frequency band is detected, notch filtering is used to attenuate the signal by ≥40dB, achieving dynamic parameter adaptation.

[0031] The time-domain peak detection employs an adaptive thresholding method (threshold equal to the baseline value plus 3 times the baseline standard deviation; the baseline value is the mean of the first 10 window signals, with an update period of 50ms), achieving a false detection rate ≤1% (≤10 false detections per 1000 test samples). The FFT frequency resolution is ≤1Hz, and spectral leakage is suppressed using a Hanning window (leakage rate ≤10%). Three threshold levels are determined based on K-means clustering of 500 historical data sets (100 iterations, silhouette coefficient ≥0.7): amplitude 0.1-0.3V (level 1, slight deformation), 0.3-0.7V (level 2, moderate deformation), 0.7-1V (level 3, severe deformation); frequency 1-10Hz (level 1, slow motion), 10-50Hz (level 2, fast motion), 50-100Hz (level 3, high-frequency vibration). The notch filter uses a second-order IIR structure, with a transfer function where the numerator is the square of s plus... The square of , the denominator is the square of s plus Divide by Q, multiply by s, and finally add The square of (where) It equals 2 multiplied by pi and then multiplied by the center frequency. (Q is the quality factor and is 10), center frequency locking accuracy ≤ 0.5Hz, bandwidth 0.5-5Hz, interference suppression ratio ≥ 30dB, and attenuation of effective components of impulse signal (1-10kHz) ≤ 5%.

[0032] In detail, the specific implementation logic of this step is as follows: After the dual signals are acquired by the ADC (sampling accuracy 16 bits), they are synchronously input into the filtering module. The hardware chip MAX291 is responsible for preliminary filtering (adjustable cutoff frequency, attenuation slope ≥24dB / oct), and the software algorithm runs on the ARM Cortex-M7 core (480MHz) for fine processing. Feature extraction is performed every 50ms: the deformed signal is traversed by the sliding window method, and the peak value in the time domain is extracted by the peak detection algorithm (threshold trigger + neighborhood comparison), retaining 2 decimal places; a 1024-point FFT transformation is performed on each signal segment, and zeros are padded to 1024 points to ensure frequency resolution. The three frequency components with the largest amplitude in the spectrum are extracted as key interference frequencies (arranged in descending order of amplitude). Interference level matching: Amplitude and frequency characteristics are compared with three threshold levels to comprehensively determine the interference intensity and type; Level 1 interference maintains a 1-10kHz passband, Level 2 narrows to 1-7kHz, and Level 3 narrows to 1-5kHz; when 50Hz power frequency interference (amplitude 0.1-0.3V) is detected, notch filtering is activated (center frequency 50Hz, bandwidth 2Hz); when 10-20Hz motion interference (amplitude 0.3-0.5V) is detected, the center frequency is set to the interference peak frequency, and the bandwidth is 5Hz. Parameter adjustment adopts a smooth switching mechanism, using a first-order low-pass filter to achieve parameter transition (transition time ≤ 5ms, transition function is y(t) equal to y0 plus (y1 minus y0) multiplied by (1 minus the negative t of the natural constant e divided by τ), where τ = 1ms), avoiding signal distortion (THD ≤ 3%). The total time from feature extraction to parameter adjustment is ≤ 50ms, meeting real-time requirements.

[0033] Specifically, the threshold establishment and filtering effect verification are as follows: The threshold is constructed using K-means clustering. The amplitude and frequency characteristics of 500 sets of historical data are clustered into 3 classes respectively. The mean μ and standard deviation σ of each class are calculated to determine the threshold range (e.g., amplitude level 1 = μ1 ± 0.05V). Each level corresponds to the optimal combination of filtering parameters, which is determined by grid search (passband range 1-15kHz step size 500Hz, attenuation coefficient 20-60dB step size 10dB). The parameters are stored in an XML configuration library and support dynamic updates. The filtering effect is evaluated using dual indicators: the processed impulse signal SNR ≥ 25dB, and the interference signal attenuation ≥ 30dB (50Hz power frequency ≥ 40dB). It can only be put into use after passing the standard signal source test (input 1kHz, 1V impulse signal and 50Hz, 0.5V interference signal). Optimization mechanism: For every 50 sets of impulse data collected, the pass rate of the filtered signal is statistically analyzed (SNR ≥ 25dB is considered passable). If the pass rate is < 95%, the threshold is re-clustered and updated.

[0034] Furthermore, the dynamic medium coefficient was calibrated, as detailed below: Sixty control experiments were designed based on deformation levels of 0%-50% (11 gradients in 5% increments) and impact velocities of 0-100 m / s (10 gradients in 10 m / s increments). Extreme combinations of 50% deformation and ≥80% impact velocity were excluded. Each group had three parallel experiments. Deformation displacement (accuracy ≤0.01 mm), impact velocity (accuracy ≤0.1 m / s), and measured values ​​of the medium propagation coefficient (accuracy ≤0.01 m / s) were recorded. Based on 180 data points, a correlation model (R²) was constructed using 3rd-5th order polynomial fitting. 2 ≥0.95); Real-time extraction of peak displacement of deformation signal (100ms / time, extraction algorithm is peak detection and threshold judgment, threshold ≥0.01mm), substitute into model to obtain calibration medium coefficient, correct shock wave propagation velocity, error ≤2%.

[0035] Deformation was controlled by an Instron 5944 tensile device (maximum tensile force 100N, displacement accuracy ±0.001mm, tensile speed adjustable from 0.1-10mm / min). Experimental samples were fixed in a dedicated fixture (flatness ≤0.5mm, clamping force 0-50N), and deformation was monitored by a Keyence GT2-P12K displacement sensor (range 0-10mm, accuracy ±0.002mm). Impact was simulated by a PCB086D05 pneumatic generator (impact velocity 0-100m / s, energy 0-50J), with the impact object being a 10mm GCr15 steel ball (HRC58-62, Ra≤0.8μm). The medium propagation coefficient was measured using an Optec DLS-C200 laser velocimeter (accuracy ≤±0.1m / s) and a Xilinx XC7K325T FPGA time measurement system (accuracy ≤±1ns). Polynomial fitting employed the least squares method, with model prediction bias ≤3%.

[0036] In detail, the specific implementation logic of this step is as follows: During the experimental phase, the nanofiber sensing network is fixed onto a 10cm × 10cm simulated protective medium sample (4 layers of Kevlar / single layer of UHMWPE, thickness 2-5mm, surface flatness ≤0.1mm), and the sample is fixed in a special fixture. The degree of deformation is controlled by a stretching device, with each set held for 5 minutes to stabilize (rebound rate ≤2%). A pneumatic impact generator is used to impact the central area of ​​the sample (deviation ≤0.5mm), with a 10-minute interval between each experimental group to avoid sample fatigue (fatigue is determined by a propagation coefficient deviation >5% after 3 consecutive tests). Data fitting is performed using MATLAB R2023a: a scatter plot of deformation degree versus propagation coefficient is plotted to initially determine the type of nonlinear relationship, followed by 3rd-5th order polynomial fitting, and the goodness of fit R is calculated. 2 Choose R 2 The largest order ≥ 0.95 (Kevlar 4th order, R) 2 ≥0.97; UHMWPE 3rd order, R 2≥0.96), to obtain the fitting equation (e.g., Kevlar: y equals 0.0002 x to the power of 4, minus 0.005 x to the power of 3, plus 0.03 x to the square of 4, minus 0.2 x, plus 2.8). Real-time calibration stage: every 100ms, the peak displacement is extracted from the deformation signal, and the deformation parameters are calculated using the formula "deformation degree equals peak position removed multiplied by the initial network length and then multiplied by 100%" (initial length is measured by laser, accuracy ≤0.1mm). These parameters are then substituted into the correlation model to obtain the calibration medium coefficient, replacing the fixed medium coefficient (Kevlar reference 2800m / s, UHMWPE reference 2500m / s). The parameters are corrected according to "corrected propagation speed equals calibration medium coefficient multiplied by reference propagation speed". Displacement, the deformation parameters are calculated using the formula "deformation degree equals peak position removed multiplied by the initial network length and then multiplied by 100%" (initial length is measured by laser, accuracy ≤0.1mm), substituted into the correlation model to obtain the calibration medium coefficient, replacing the fixed medium coefficient.

[0037] Specifically, model validation and parameter correction are as follows: Validation is performed using 10% of the reserved experimental data (18 data points), and deployment is only permitted if the relative error is ≤5%. For every 100 sets of impact data collected, 50 valid samples are extracted (impact velocity 20-80 m / s, deformation degree 5%-45%), and the polynomial coefficients are updated using the least squares method; a comprehensive calibration is performed every 3 months with 20 sets of control experiments. An anomaly monitoring mechanism is established: when the calibration medium coefficient exceeds 0.8 × -1.2 × the baseline value, the terminal APP prompts a check of the sensor network (anomaly is determined by ≥3 node interruptions or displacement ≥0.5 mm).

[0038] Furthermore, the planar coordinates of the impact point are calculated, as follows: Sensor nodes are orthogonally arranged at preset intervals, with ≥256 nodes in the critical area (16×16 grid, deviation ≤0.1cm) and ≥64 nodes in the non-critical area (8×8 grid); the x and y axis coordinates are calibrated using a Leica AT960-MR laser tracker (accuracy ≤±0.1mm) (clothing plane coordinate system: the midpoint of the neckline is the origin, the horizontal direction to the right is the x-axis, and the vertical direction downward is the y-axis); a nonlinear equation system is constructed based on the node coordinates and the signal reception time difference; the particle swarm optimization algorithm is selected, and the solution is iteratively obtained with a residual ≤0.01m as the termination condition to obtain the planar coordinates of the impact point (accuracy ≤2cm).

[0039] The nodes employ a modular design (4×4 grid / module, size 2cm×2cm), with modules connected by flexible wires (diameter ≤0.5mm, silver-plated copper wire, bending resistance ≥10000 times), and a 5mm overlap area reserved at the edges (signal overlap rate ≥90%). Calibration coordinates are stored in JSON format and encrypted with AES-256. The equations are based on the principle that "distance difference equals propagation speed multiplied by time difference," with a time difference measurement accuracy ≤1μs. Particle swarm optimization (PSO) algorithm parameters: population 50-100, iterations 50-100, inertia weight 0.4-0.9 (linearly decreasing), learning factor 1.5-2.0.

[0040] In detail, the specific implementation logic of this step is as follows: After receiving the impulse signal, the node triggers timestamp recording through a built-in comparator (trigger threshold 0.1V, response time ≤1μs). The timestamp is generated by a high-precision timer in the acquisition module (precision ≤0.1μs). The acquisition module aggregates the timestamps of all nodes through the SPI bus, calculates the time difference between any two nodes, and uses the 3σ criterion to remove outliers (values ​​exceeding the mean ±3σ are considered outliers, with an outlier rate ≤1%). An equation set is constructed: the first 8 nodes with the earliest signal reception times are selected, and 28 nonlinear equations (8×7 / 2) are constructed with a redundancy of 26 to reduce errors. Particle Swarm Optimization (PSO) Solution: 1. Initialize 50 candidate solutions (x∈[0,50]cm, y∈[0,80]cm); 2. Calculate the residual of the equation system for each candidate solution, and determine the global optimum gbest and the individual optimum pbest; 3. Update the position and velocity according to the position update rule (next position equals current position plus next velocity) and the velocity update rule (next velocity equals inertia weight multiplied by current velocity, plus the first learning factor multiplied by the first random number multiplied by the difference between the individual optimum and the current position, plus the second learning factor multiplied by the second random number multiplied by the difference between the global optimum and the current position) (r1, r2 are 0-1 random numbers, initial velocity ±2cm / iteration); 4. Adjust the inertia weight every 10 iterations (linearly reduced from 0.9 to 0.4); 5. Output the coordinates when the residual is <0.01m or after 100 iterations (e.g., the solution result (18.32cm, 42.15cm)).

[0041] Specifically, the accuracy verification and optimization are as follows: Testing at 10 known coordinate points (error ≤ 0.1mm), a deviation ≤ 2cm resulted in a pass rate ≥ 98%; the average deviation was 1.1cm when the node spacing was 3cm and 1.8cm when it was 5cm. When accuracy was not up to standard: 1. Reduce the node spacing by 20%; 2. Optimize algorithm parameters (e.g., non-linear adjustment of inertia weights); 3. Recalibrate the node coordinates. Node stability assurance: Coordinates are checked every 3 months and after garment washing; nodes with deviations exceeding 0.5mm are re-fixed; FPGA hardware acceleration accelerates the calculation, with a processing time ≤ 50ms.

[0042] Furthermore, a three-dimensional damage coordinate system was constructed, as detailed below: A fixed integration time window of 1-5 ms was set according to the impact energy (1 ms for <10J, 3 ms for 10-25J, and 5 ms for >25J). The trapezoidal integration method was used to calculate the impact energy parameters of the pure shock wave signal. Three protective media, Kevlar, ultra-high molecular weight polyethylene, and aramid fiber, were selected, and control experiments were conducted at 0-50J (in increments of 5J). Each group underwent five parallel experiments, and the penetration depth was measured (Olympus 38DLPLUS, accuracy ≤ ±0.01mm). An energy attenuation model (R0) was constructed based on the experimental data. 2 ≥0.96), substitute the parameters to invert the penetration depth (error ≤0.1mm), and combine with the plane coordinates to form a three-dimensional coordinate (x, y axis ≤2cm, z axis ≤0.1mm).

[0043] The integration time window was determined statistically from 100 sets of signals (median duration 800 μs for 0-10 J, 2.5 ms for 10-25 J, and 4.2 ms for 25-50 J), with an energy capture rate ≥95%. The integration amplitude is equal to half the sum of the signal amplitudes of all two adjacent sampling points, multiplied by the sampling interval, and finally summed. The impact energy is equal to the integration amplitude multiplied by 0.01 J / V・s (calibration error ≤1%). The impact source was an Instron CEAST9350 drop hammer tester. The penetration depth was equal to the original thickness minus the remaining thickness after impact (measured at 5 points and averaged), and outliers ±3σ were removed (removal rate ≤5%). The energy decay model: Kevlar used an exponential model (y equals 5.2 multiplied by the natural constant e raised to the power of -0.08 x, plus 0.1, R...). 2 ≥0.97), UHMWPE uses a power function model (y equals 0.05 multiplied by x raised to the power of 0.85, plus 0.05, R 2 ≥0.96).

[0044] In detail, the specific implementation logic of this step is as follows: Determine the energy range based on the peak amplitude of the impact signal (<0.5V corresponds to <10J, 0.5-1.0V corresponds to 10-25J, >1.0V corresponds to >25J), and match the integration window; perform trapezoidal integration on the clean signal (SNR≥25dB) (e.g., integration amplitude 2.8V). s, impact energy = 2.8 × 0.01 = 0.028 J), calculate peak energy and energy density, and remove abnormal parameters exceeding 0-50 J (proportion ≤ 1%). Model correction: call the calibration medium coefficient, and correct the model parameters according to "m correction equals m reference multiplied by (calibration medium coefficient divided by reference medium coefficient)"; substitute the impact energy and plane coordinates into the model, and iteratively invert the penetration depth: set the initial penetration depth to 0, calculate the corresponding energy value and compare it with the actual energy, adjust by 0.005 mm when the error is > 0.01 mm, repeat the iteration until the error is ≤ 0.01 mm (e.g., Kevlar 18J impact inversion depth is 2.1 mm). The three-dimensional coordinate combination is in the format "(x,y,z)", with the x and y axes retaining 2 decimal places (cm) and the z axis retaining 3 decimal places (mm), such as (15.23,30.45,1.234).

[0045] Specifically, model calibration and coordinate verification are as follows: the model is calibrated every 50 sets of data, and the experiment is repeated when the medium is changed; the three-dimensional coordinate deviation is tested at 5 known penetration depth gradients (0.5mm, 1.0mm, 2.0mm, 3.0mm, 5.0mm), and the pass rate is ≥98%; the integral window parameters are updated every 100 sets of data, and calibration is triggered when the model drift is ≥0.05mm.

[0046] Furthermore, the quantitative damage level is obtained and the data is transmitted, as follows: The energy level is divided into 3 intervals (0-10J, 10-25J, 25-50J), with an overlap of ≤5% between adjacent intervals; a standardized damage grading threshold is configured for each interval (referencing the AIS standard); the interval is determined based on the impact energy, the threshold is matched, and the damage level (Level I-IV) is mapped according to the penetration depth; the data is transmitted to the terminal via BLE5.0 or LoRa protocol, with a parsing delay of ≤50ms.

[0047] The energy ranges correspond to the protective medium states: Range 1 (fracture strength ≥ 80%), Range 2 (30%-80%), and Range 3 (< 30%). Grading thresholds are: Range 1 (Level I < 0.5mm, Level IV ≥ 2.0mm), Range 2 (Level I < 1.0mm, Level IV ≥ 3.0mm), and Range 3 (Level I < 2.0mm, Level IV ≥ 5.0mm), supporting ±10% fine-tuning. The transmission protocol supports AES-128 encryption (key updated 24 hours a day). The terminal supports 2D / 3D visualization, local / cloud storage (≥ 10,000 records), and Level IV damage audible and visual alarm (≥ 80dB, 2Hz flashing).

[0048] Specifically, the specific implementation logic of this step is as follows: First, determine the interval according to the impact energy, and use the peak energy as the standard in case of anomalies; Damage level mapping: Level I (no damage, protection efficiency ≥ 95%), Level II (local deformation, 85% - 95%), Level III (partial penetration, 50% - 85%), Level IV (complete penetration, < 50%). Data encapsulation: Organize fields in JSON format (timestamp, device SN, medium type, three-dimensional coordinates, damage level, energy, SNR, battery level). Example: {"timestamp":"20240520143025689","device_sn":"SN2024050001","medium_type":"Kevlar","3d_coordinate":{"x":"15.23","y":"30.45","z":"1.234"},"damage_level":"Level II","energy":"18.5J","snr":"32dB","battery_level":"75%"}, and transmit after CRC32 check (polynomial is 0xEDB88320). Terminal parsing: Parse JSON data through the rapidjson library (efficiency ≥ 1000 records / second), mark the impact position with 2D icons (different colors for different levels), restore the damage depth with 3D models, trigger an audible and visual alarm for Level IV damage, and store the data in the local SQLite database and synchronize it to the cloud.

[0049] Specifically, the threshold configuration and data processing are as follows: The threshold supports default and manual fine-tuning (export in CSV format), the data encapsulation contains 12 core fields, and retransmission in case of check failure is ≤ 3 times (abnormal ratio ≤ 0.5%). Transmission and storage are encrypted, and user access requires authentication (password / biometric, accuracy ≥ 99.2%), and it will be locked for 15 minutes if there are 3 consecutive failures. Update the threshold library quarterly, incorporating the GB / T23463 - 2021 standard and ≥ 100 groups of new experimental data.

[0050] In addition, the terminal device also supports data export, offline caching, and statistical analysis functions: Data export supports Excel and PDF formats (encrypted with AES-128), offline caching ≥ 500 groups (circular overwrite), and automatic synchronization after reconnecting; Statistical analysis generates reports on a daily / weekly / monthly basis, supports multi-dimensional filtering and export, providing support for protection optimization.

[0051] The system has a device self-check and fault prompt function. The daily automatic self-check is ≤ 1 minute, and the self-check covers key links such as the connectivity of the sensing network and the status of the acquisition module. When a fault is detected, a warning pop-up window + a sound prompt of ≥ 70dB will be displayed, providing troubleshooting suggestions, and supporting the generation of encrypted fault logs with a single click.

[0052] In terms of power consumption optimization, the system adopts an intelligent sleep-wake mechanism, entering a low-power mode (power consumption ≤5mA) after 5 minutes of inactivity, and waking up within ≤100ms upon detecting an impact or operation. The battery is a 2000mAh 18650 lithium battery, supporting Type-C fast charging (10W), providing 4 hours of battery life with 30 minutes of charging, and ≥72 hours of standby time on a full charge.

[0053] The processes described above with reference to the flowcharts in the embodiments disclosed in this invention can be implemented as computer software programs. The embodiments disclosed in this invention include a computer program product comprising a computer program carried on a computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via a communication component, and / or installed from a removable medium. When the computer program is executed by a central processing unit (CPU), it performs the functions defined in the methods of this application. It should be noted that the computer-readable medium described above in this application can be a computer-readable signal medium or a computer-readable storage medium, or any combination of the two. A computer-readable storage medium can be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of computer-readable storage media may include, but are not limited to: electrical connections having one or more wire segments, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof. In this application, a computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in connection with an instruction execution system, apparatus, or device. In this application, a computer-readable signal medium may include a data signal propagated in baseband or as part of a carrier wave, carrying computer-readable program code. Such propagated data signals can take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. A computer-readable signal medium can also be any computer-readable medium other than a computer-readable storage medium, which can send, propagate, or transmit a program for use by or in connection with an instruction execution system, apparatus, or device. The program code contained on a computer-readable medium may be transmitted using any suitable medium, including but not limited to: wireless segments, wire segments, optical fibers, RF, etc., or any suitable combination thereof.

[0054] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of the present invention. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0055] Those skilled in the art should understand that the embodiments of the present invention described above and shown in the accompanying drawings are merely examples and do not limit the present invention. The purpose of the present invention has been fully and effectively achieved. The functions and structural principles of the present invention have been shown and explained in the embodiments. Without departing from the stated principles, the implementation of the present invention may have any variations or modifications.

Claims

1. A method for locating clothing damage based on a nanofiber sensing mesh, characterized in that, The method includes: The original spatiotemporal distribution signal of the impact event and the real-time deformation signal of the sensor network are collected by a nanofiber sensing grid. Based on a pre-defined motion artifact feature library, an adaptive anti-interference preprocessing method is used to perform spatiotemporal distribution signal of the original shock wave through a convolutional neural network model. The filtering parameters are dynamically adjusted according to the human motion state to filter out interference signals and artifacts, and obtain a pure shock wave signal. Based on the pure shock wave signal, the signal reception time difference of multiple sensor nodes is extracted, and dynamic medium coefficient calibration is performed by combining the real-time deformation signal of the sensor network. The plane coordinates of the impact point are calculated by the time difference positioning method and optimization algorithm. Energy correlation calculations are performed on the pure shock wave signal, and the plane coordinates of the impact point and the dynamic medium coefficient are integrated. The penetration depth of the impacting object is inverted by an energy attenuation model adapted to the characteristics of the protective medium of clothing, and combined with the plane coordinates of the impact point to form a three-dimensional damage coordinate system. Based on the energy level of the pure shock wave signal, a standardized damage grading threshold is dynamically matched, and the penetration depth in the three-dimensional coordinates of the damage is mapped to the grading threshold to obtain the quantitative damage level. The three-dimensional coordinates of the damage and the quantitative damage level are then transmitted to the terminal.

2. The method for locating clothing damage based on a nanofiber sensing mesh according to claim 1, characterized in that, The specific process for acquiring the original impact spatiotemporal distribution signal and the real-time deformation signal of the sensor network is as follows: By utilizing a pre-installed nanofiber sensing network and relying on the piezoelectric effect, the mechanical physical quantities generated by the impact event are converted in real time into the original impact spatiotemporal distribution signal in the form of an electrical signal containing amplitude and phase information. The nanofiber sensing network uniformly covers the key protective areas of the clothing. By leveraging the strain sensing characteristics of sensor networks, local deformations caused by human movement, such as stretching, bending, or impact, can be accurately captured, and real-time deformation signals including the degree and rate of deformation can be collected simultaneously. The high-precision clock synchronization module performs real-time calibration of the acquisition timing of the two types of signals based on the clock synchronization protocol, aligning the signal acquisition timestamps frame by frame to ensure that the acquisition times of the two are completely consistent.

3. The method for locating clothing damage based on a nanofiber sensing mesh according to claim 2, characterized in that, The construction process of the motion artifact feature library and convolutional neural network model is as follows: Various interference signal samples generated in different scenarios were collected, and after wavelet transform denoising and maximum-minimum normalization preprocessing, key features were extracted and the categories were labeled. Feature samples are organized and stored in a unified data format system to form a structured motion artifact feature library containing different types and intensities of interference; Using the time and frequency domain features of the original signal as input and the filter parameter adjustment command as output, the convolutional neural network model is trained with feature library samples. The gradient descent method is used to optimize the loss function, and the number of iterations and learning rate are set until the model's prediction accuracy meets the preset requirements.

4. The method for locating clothing damage based on a nanofiber sensing mesh according to claim 3, characterized in that, The specific process of dynamically adjusting the filtering parameters is as follows: By using the sliding window method for time-domain peak detection and the fast Fourier transform for frequency-domain analysis, the amplitude and frequency distribution of the real-time deformation signal of the sensor network can be accurately extracted. The extracted features are compared one by one with multiple preset thresholds established based on statistical analysis of a large amount of historical interference data. If the amplitude of the deformation signal exceeds the preset threshold of the corresponding range, the filter passband will be automatically narrowed proportionally. The greater the amplitude of exceeding the threshold, the narrower the passband. If interference components in a specific frequency band are detected, notch filtering technology is used to accurately attenuate the signal in that frequency band, thereby achieving dynamic adaptation and adjustment of the filtering parameters.

5. The method for locating clothing damage based on a nanofiber sensing mesh according to claim 4, characterized in that, The specific process for calibrating the dynamic medium coefficient is as follows: Multiple control experiments were designed according to the deformation degree gradient and impact intensity gradient. Multiple parallel experiments were set up in each group to reduce errors. The deformation displacement, impact velocity and corresponding media propagation coefficient were recorded in detail under different deformation degrees and impact conditions. Based on experimental data, a nonlinear correlation model between the degree of deformation and the medium propagation coefficient was constructed using polynomial fitting. The peak displacement of the deformation signal is extracted in real time as the deformation parameter, and then substituted into the correlation model to calculate the calibrated medium coefficient. The calibrated medium coefficient is applied to the time difference positioning method to accurately correct the shock wave propagation velocity parameters.

6. The method for locating clothing damage based on a nanofiber sensing mesh according to claim 5, characterized in that, The specific steps for calculating the plane coordinates of the impact point are as follows: The sensor nodes are evenly and orthogonally arranged in the key protective area of ​​the garment lining according to the preset grid spacing. The node spacing is set according to the positioning accuracy requirements. The x and y axis spatial coordinate information of each node is accurately calibrated using a three-dimensional coordinate measuring device. Based on the spatial coordinates of each node and the signal reception time difference, a set of nonlinear equations corresponding to the time difference positioning method is constructed. Select the appropriate algorithm based on the requirements of solution efficiency and accuracy, and perform iterative calculations with the equation system residuals meeting the preset threshold as the termination condition. The plane coordinates of the impact point are obtained by minimizing the residuals of the system of equations.

7. The method for locating clothing damage based on a nanofiber sensing mesh according to claim 6, characterized in that, The specific steps for forming the three-dimensional coordinates of the damage are as follows: A fixed integration time window is set according to the typical duration of the shock wave signal to ensure complete capture of shock energy information. Time-domain integration is performed on the pure shock wave signal to obtain shock energy-related parameters including signal integral amplitude, peak energy, and energy density. Common clothing protective media were selected, and a comparative experiment was conducted with multiple impact intensities. The measured values ​​of the penetration depth of the impactor were recorded under different energy parameters. The system was then calibrated to obtain an energy attenuation model that is adapted to the characteristics of clothing protective media. The impact energy-related parameters, the plane coordinates of the impact point, and the dynamically calibrated medium coefficient are all substituted into the energy attenuation model. The penetration depth of the impactor is then obtained by inverting the mapping relationship between the model input parameters and the measured depth. The impact penetration depth obtained by inversion is combined with the plane coordinates of the impact point to form a complete three-dimensional damage coordinate system that includes the plane coordinates of the x and y axes and the depth coordinates of the z axis.

8. The method for locating clothing damage based on a nanofiber sensing mesh according to claim 7, characterized in that, The specific process for obtaining the quantitative damage level and implementing the data transmission step is as follows: Based on the impact resistance parameters and damage risk levels of the protective medium for clothing, the energy levels of pure shock wave signals are divided into multiple preset ranges; A standardized damage grading threshold is configured for each energy range, and the threshold setting is matched with the damage limit of the protective medium. The energy range is determined based on the calculated impact energy parameters, and the corresponding classification threshold is automatically matched. The inverted penetration depth is compared with the matched grading threshold, and the quantitative damage level is obtained by mapping according to the preset rules. The three-dimensional coordinates and grade data of the damage are transmitted to the terminal device via a wireless transmission protocol, and the terminal analyzes and processes the received data.

9. A clothing damage localization system based on a nanofiber sensing mesh, characterized in that, The system is used to perform a method for locating clothing damage based on a nanofiber sensing mesh, as described in any one of claims 1-8.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that is executed by a processor to implement a method for locating clothing damage based on a nanofiber sensing mesh as described in any one of claims 1-8.

Images