Textile fabric tearing strength detection method and system

Abstract

The invention discloses a textile fabric tearing strength detection method and system, and relates to the technical field of fabric detection.The method comprises the steps that fabric microstructure, macromechanics and component data are synchronously collected through a multi-mode sensing system, and detection modes are dynamically switched based on a real-time recognition result; inputting the space-time correlated defect data into the physical constraint enhanced prediction model to generate a crack propagation trend and a tear strength prediction value; wherein the prediction model is fused with a Griffith crack propagation theory and a hidden Markov model, and double constraints of a material mechanics law and dynamic defect evolution are realized; a non-contact detection module and a flexible contact detection module work cooperatively, and a process optimization scheme is output in combination with an edge and cloud distributed architecture.

Description

Technical Field

[0001] This application relates to the technical field of fabric detection, and more specifically, to a method and system for detecting the tearing strength of textile fabrics. Background Art

[0002] The tearing strength of textile fabrics is a core indicator to measure their durability and performance, directly affecting the quality safety and service life of end products. In fields such as clothing, outdoor equipment, and aerospace, the tearing failure of fabrics may lead to serious consequences, such as torn clothing affecting the wearing experience and damaged safety protection fabrics threatening life safety. Therefore, accurately detecting the tearing strength of fabrics and identifying potential failure hazards is a key link in ensuring product reliability and optimizing production processes.

[0003] Traditional detection methods mainly obtain the maximum tearing load of fabrics through physical destruction tests, such as the Elmendorf method and the trouser tearing method. Such methods require destroying samples and each single detection takes up to 5 - 10 minutes, consuming time and effort and having a high sample loss rate.

[0004] In view of the above problems, no effective solutions have been proposed yet. Summary of the Invention

[0005] Embodiments of this application provide a method and system for detecting the tearing strength of textile fabrics to solve the above technical problems.

[0006] This application provides a method for detecting the tearing strength of textile fabrics, including: Synchronously collecting fabric microstructure, macro-mechanics, and composition data through a multi-modal sensing system, and dynamically switching the detection mode based on real-time recognition results; Inputting the spatio-temporally correlated defect data into a physically constrained enhanced prediction model to generate crack propagation trends and predicted tearing strength values; wherein, the prediction model integrates Griffith crack propagation theory and hidden Markov model to achieve double constraints on material mechanics laws and dynamic defect evolution; Using a non-contact detection module and a flexible contact detection module to work together, and outputting a process optimization plan in combination with an edge and cloud distributed architecture.

[0007] Further, the realization of dynamically switching the detection mode includes: Obtaining fiber molecular bonding information through a near-infrared spectrometer, combining with the fabric surface texture captured by an industrial camera, and constructing an adaptive feature pyramid network for classification to output the fabric type and organizational structure; When identified as a knitted fabric, start the cyclic tensile load simulation module, synchronously extract the yarn slip trajectory data and molecular bonding strength attenuation data under tensile load, dynamically allocate weights through a dual-channel attention mechanism, and generate a joint mapping relationship between tensile load and slip displacement; When the coated fabric is recognized, the multi-scale residual convolutional network is activated to fuse the porosity distribution data of the penetrating infrared thermal imaging sequence and the laser profile scanning, and locate the critical area of fiber fracture under the coating.

[0008] Furthermore, the construction of the prediction model enhanced by physical constraints includes: Embed the Griffith energy release rate threshold in the state transition probability matrix of the hidden Markov model. When the predicted energy of the crack propagation path exceeds the material fracture toughness, adjust the state transition path to conform to physical laws; Design a joint loss function to dynamically weight and fuse the timing prediction error of the hidden Markov model and the energy constraint error calculated by the Griffith crack propagation theory, and synchronously optimize the matching degree between the crack evolution timing and physical laws through gradient backpropagation; According to the optimization result of the joint loss function, output the predicted value of the tearing strength that satisfies the double constraints of stress concentration and energy dissipation.

[0009] Furthermore, the implementation of the dual-channel attention mechanism includes: Use a temporal convolutional network to extract temporal features from the yarn slip trajectory data and generate a spatio-temporal encoding vector of the slip displacement; Use a graph convolutional network to model the intermolecular interaction relationship of the molecular bond strength attenuation data and generate a topological feature vector of the bond strength; Dynamically calculate the correlation weight between the spatio-temporal encoding vector and the topological feature vector through a cross-attention layer to generate an attention distribution map of the joint mapping relationship for indicating the parameter adjustment of the cyclic tensile load.

[0010] Furthermore, the adjustment of the state transition path includes: Calculate the stress intensity factor of the current crack propagation direction based on the Griffith crack propagation theory to generate an energy release rate constraint boundary; During the Viterbi decoding process of the crack propagation hidden Markov model, remove the state transition paths that exceed the constraint boundary; Reassign the probabilities of the state transition paths that conform to physical laws to ensure that the prediction of the crack propagation trend satisfies both dynamic defect evolution and material fracture mechanics laws.

[0011] Furthermore, the non-contact detection module includes: Emit ultrasonic waves with a set frequency through an air-coupled ultrasonic transducer, and use the wavelet packet decomposition algorithm to extract the yarn fracture characteristic frequency band in the echo signal; Based on a random forest classifier, distinguish the acoustic feature patterns of fiber debonding and monofilament fracture, and use the classification result as the criterion for defect type judgment. Synchronously use a laser profile scanner to obtain three-dimensional topography data of the fabric surface, and locate the stress concentration areas where the stiffness gradient exceeds the set threshold through a curvature mutation detection algorithm.

[0012] Furthermore, the implementation of the curvature mutation detection algorithm includes: Calculate the Gaussian curvature of the three-dimensional topography data to generate a curvature distribution heat map; Use a morphological gradient operator to extract the curvature mutation boundary and mark the potential tear starting points; Spatially match the curvature mutation region with the defect positions detected by ultrasound to verify the reliability of the stress concentration areas.

[0013] Furthermore, the implementation of the edge and cloud distributed architecture includes: Deploy a fault-tolerant detection model at the edge nodes in the weaving process, and use a bit-flip fault-tolerant mechanism to process the warp and weft yarn breakage data collected in real time; After receiving the abnormal data, the cloud platform simulates the tear life under the actual scenario through a digital twin model and inversely deduces the combination of process parameters that meet the target strength.

[0014] Furthermore, the method also includes an environmental adaptive calibration step: Establish a mapping relationship table between temperature and humidity and the attenuation of material strength, dynamically correct the detection results through Gaussian process regression, and apply a random correction coefficient to the predicted value of the tear strength of cotton fabrics when the environmental humidity exceeds the set threshold; Adopt an illumination invariance feature extraction algorithm, separate the brightness component in the HSV color space, and eliminate the influence of color temperature fluctuations on fiber texture recognition through histogram specification.

[0015] This application provides a textile fabric tear strength detection system, including: A detection mode switching module, which is used to synchronously collect fabric microstructure, macro-mechanics and composition data through a multi-modal sensing system, and dynamically switch the detection mode based on the real-time recognition results; A tear strength prediction module, which is used to input the spatio-temporally correlated defect data into a physically constrained enhanced prediction model to generate the crack propagation trend and the predicted value of the tear strength; among them, the prediction model integrates the Griffith crack propagation theory and the hidden Markov model to achieve double constraints on the mechanical laws of materials and the dynamic evolution of defects; A process optimization module, which is used to use a non-contact detection module and a flexible contact detection module to work together, and output a process optimization plan in combination with the edge and cloud distributed architecture.

[0016] Based on the embodiments provided in this application, a multi-modal sensing system is used to synchronously obtain fabric microstructure, macroscopic mechanics, and composition data, changing the defect of traditional detection relying on a single physical index (such as the maximum tearing load). By dynamically switching the detection mode according to the real-time recognition results, the system can adaptively adjust the detection strategy for different fabric characteristics, avoiding the detection blind spots of complex fabrics in the fixed process of traditional methods and improving the comprehensiveness and accuracy of the detection results. The prediction model with enhanced physical constraints combines Griffith crack propagation theory and hidden Markov model, for the first time combining the basic laws of material mechanics with the time series analysis of defect evolution. Compared with the "black box" prediction of traditional methods that only rely on statistical laws, this model not only follows the objective physical mechanism but also can capture the dynamic changes of defects related to time and space (such as the correlation of defect positions in continuous production), thus generating a tearing strength prediction result including the crack propagation trend, providing a scientific basis for early identification of potential failure hazards. The collaborative work of the non-contact detection module and the flexible contact detection module solves the problems of damage to vulnerable fabrics (such as silk and lace) by traditional rigid detection and the detection adaptability problems of complex-shaped fabrics (such as curved surfaces and folds). Combining the edge and cloud distributed architecture, the edge nodes process high-frequency detection data in real time and quickly respond to anomalies, and the cloud platform outputs a process optimization plan based on multi-dimensional data, forming an intelligent closed-loop of "detection - analysis - process adjustment" and promoting the upgrade of the quality control mode of textile production from "experience-driven" to "data-driven". BRIEF DESCRIPTION OF THE DRAWINGS

[0017] The drawings described herein are used to provide a further understanding of the embodiments of the present invention, form a part of this application, and the illustrative embodiments and descriptions thereof are used to explain this application without unduly limiting this application. In the drawings: Figure 1 It is a flowchart of an optional method for detecting the tearing strength of textile fabrics according to an embodiment of this application; Figure 2 It is a flowchart of another optional method for detecting the tearing strength of textile fabrics according to an embodiment of this application; Figure 3 It is a structural diagram of an optional system for detecting the tearing strength of textile fabrics according to an embodiment of this application.

[0018] The realization, functional characteristics, and advantages of the objectives of the present invention will be further described with reference to the embodiments and the drawings. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0019] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0020] Optionally, as Figure 1 shown, the present application provides a method for detecting the tearing strength of a textile fabric, including: S101, synchronously collecting fabric microstructure, macro-mechanics, and composition data through a multi-modal sensing system, and dynamically switching the detection mode based on the real-time recognition result; S102, inputting the spatiotemporally correlated defect data into a physically constrained enhanced prediction model to generate a crack propagation trend and a predicted value of the tearing strength; wherein, the prediction model integrates the Griffith crack propagation theory and the hidden Markov model to achieve double constraints of the material mechanics law and the dynamic defect evolution; S103, using a non-contact detection module and a flexible contact detection module to work together, and outputting a process optimization plan in combination with an edge and cloud distributed architecture.

[0021] Based on the embodiments provided by the present application, the multi-modal sensing system is used to synchronously obtain the fabric microstructure, macro-mechanics, and composition data, which changes the defect of traditional detection relying on a single physical index (such as the maximum tearing load). By dynamically switching the detection mode based on the real-time recognition result, the system can adaptively adjust the detection strategy according to different fabric characteristics, avoiding the detection blind area of complex fabrics by the fixed process of traditional methods, and improving the comprehensiveness and accuracy of the detection results. The physically constrained enhanced prediction model integrates the Griffith crack propagation theory and the hidden Markov model, and for the first time combines the basic laws of material mechanics with the time series analysis of defect evolution. Compared with the "black box" prediction of traditional methods that only rely on statistical laws, this model not only follows the objective physical mechanism but also can capture the spatiotemporal correlation of defect dynamic changes (such as the correlation of defect positions in continuous production), so as to generate a tearing strength prediction result including the crack propagation trend, providing a scientific basis for early identification of potential failure hazards. The collaborative work of the non-contact detection module and the flexible contact detection module solves the problems of damage to vulnerable fabrics (such as silk and lace) by traditional rigid detection and the detection adaptability problems of complex-shaped fabrics (such as curved surfaces and folds). Combining the edge and cloud distributed architecture, the edge node processes high-frequency detection data in real time and quickly responds to anomalies, and the cloud platform outputs a process optimization plan based on multi-dimensional data, forming an intelligent closed loop of "detection - analysis - process adjustment", promoting the upgrade of the quality control mode of textile production from "experience-driven" to "data-driven".

[0022] Furthermore, asFigure 2 As shown, the implementation of the dynamic switching detection mode includes: S201, obtaining fiber molecular bonding information through a near-infrared spectrometer, combining the surface texture of the fabric captured by an industrial camera, and constructing an adaptive feature pyramid network for classification to output the fabric type and organizational structure; S202, when it is identified as a knitted fabric, start the cyclic tensile load simulation module, synchronously extract the yarn slip trajectory data and the molecular bonding strength attenuation data under the tensile load, dynamically allocate weights through a dual-channel attention mechanism, and generate a joint mapping relationship between the tensile load and the slip displacement; S203, when it is identified as a coated fabric, activate the multi-scale residual convolution network, fuse the porosity distribution data of the penetrating infrared thermal imaging sequence and the laser profile scanning, and locate the critical area of fiber fracture under the coating.

[0023] In some embodiments of the present application, the fabric type is classified through multi-modal data, and a differential detection strategy is started for knitted / coated fabrics. Specifically, Knitted fabric detection scenario (such as stretch knitted fabric): When the near-infrared spectrometer collects fiber molecular bonding information, the preset scanning wavelength range is 900 - 1700 nm (covering the characteristic absorption peaks of cotton / spandex), and the industrial camera captures the coil structure texture at a resolution of 12 million pixels. When training the adaptive feature pyramid network, the preset classification threshold of the "coil density" of the knitted fabric is ≥ 25 turns / cm (to distinguish from the warp and weft floating point structure of the woven fabric).

[0024] When it is identified as a knitted fabric, the cyclic tensile load simulation module is started: The preset number of tensile cycles is 50 times (simulating the daily stretching frequency of human joint activities), and the single load range is ±15% of the nominal strength of the fabric (for example, for a fabric with a nominal tear strength of 60 N, the load reciprocates between 51 N and 69 N). Synchronously, a high-speed camera (frame rate 1000 fps) is used to track the yarn slip trajectory with an accuracy of 0.02 mm, combined with the molecular bonding strength attenuation data (judged by the change rate of the amide I band absorption peak of the near-infrared spectrum), and a "load-slip" joint mapping relationship is generated through a dual-channel attention mechanism to accurately locate the weak connection points of the coil structure.

[0025] Coated fabric detection scenario (such as outdoor waterproof and breathable fabric): When processing transmissive infrared thermal imaging data with a multi-scale residual convolutional network, the preset detection wavelength is 8 - 14 μm (the transparent spectral range for penetrating the coating material), and the laser profilometer acquires the surface porosity distribution with an accuracy of 5 μm. When positioning the critical area of fiber fracture under the coating, the preset porosity anomaly threshold is ≥3% (that is, the local porosity increase by more than 3% compared to the normal area is marked as the defect area), and the authenticity of stress concentration is verified through the temperature gradient change (preset ΔT≥2℃) in the thermal imaging sequence to avoid misjudgment caused by interference on the coating surface. Based on the embodiments provided in this application, through the collaborative data acquisition of a near-infrared spectrometer and an industrial camera, an adaptive feature pyramid network is constructed to achieve precise classification of fabric types and organizational structures, changing the extensive mode of traditional detection relying on manual experience or single-index classification.

[0026] For the cyclic tensile load simulation module initiated for knitted fabrics, synchronously fuse the yarn slip trajectory and the molecular bond strength attenuation data, and dynamically allocate weights through a dual-channel attention mechanism, enabling the detection system to capture the slip failure mode of the unique coil structure of knitted fabrics under cyclic loads. This process correlates the macroscopic mechanical behavior (tensile load-displacement curve) with the microscopic molecular bond characteristics (such as the hydrogen bond breaking process). For the multi-scale residual convolutional network of coated fabrics, it fuses transmissive infrared thermal imaging and laser profilometry data, can penetrate the coating to locate the critical area of fiber fracture, breaking through the bottleneck of traditional visual inspection limited by the surface state of the fabric, and providing a new technical path for the internal defect detection of functional coated fabrics.

[0027] Furthermore, the construction of a prediction model enhanced by physical constraints includes: Embed the Griffith energy release rate threshold in the state transition probability matrix of the hidden Markov model. When the predicted energy of the crack propagation path exceeds the fracture toughness of the material, adjust the state transition path to conform to physical laws; In this embodiment, the preset Griffith energy release rate thresholds include but are not limited to 0.5 J / m², 0.7 J / m², etc.

[0028] Design a joint loss function to dynamically weight and fuse the temporal prediction error of the hidden Markov model and the energy constraint error calculated by the Griffith crack propagation theory, and synchronously optimize the matching degree between the crack evolution time series and physical laws through gradient backpropagation; According to the optimization result of the joint loss function, output the tear strength prediction value that satisfies the dual constraints of stress concentration and energy dissipation.

[0029] In a specific implementation manner, the joint loss function is: ; ; wherein, is the weighting factor of the timing error and the physical constraint, with an initial value of 0.6 (prioritizing the timing continuity of crack propagation). When the physical constraint error exceeds the relative timing prediction error in 10 consecutive detections, it is automatically adjusted to 0.5 (balancing the two types of errors); is the relative timing prediction error of the hidden Markov model; is the total number of time steps; is the predicted crack propagation displacement value at the time step, output by the hidden Markov model, in mm, such as 0.3 mm, reflecting the model's prediction of the crack propagation amount in the time dimension; is the measured crack propagation displacement value at the time step, collected by a laser displacement sensor, such as 0.31 mm, serving as the benchmark data for verifying the prediction results; is the physical constraint error of Griffith's theory; is the total number of crack propagation paths; is the measured energy release rate of the th crack path, collected by an air-coupled ultrasonic transducer to obtain the stress wave energy and converted in combination with the fabric thickness, in J / m², such as the measured value of 0.6 J / m when cotton fibers break; is the energy release rate threshold of Griffith's crack propagation theory, set differently according to fiber types. For cotton fiber fabrics, it is 0.5 J / m² (critical energy for hydrogen bond breakage), and for polyester fiber fabrics, it is 2.0 J / m² (critical energy for covalent bond breakage); is the physical constraint tolerance coefficient, allowing a reasonable fluctuation of 10% in the energy release rate near the threshold (for example, a threshold of 0.5 J / m² can accept a range of 0.45 - 0.55 J / m) to avoid misjudgment caused by detection noise.

[0030] Based on the above formula, Griffith's crack propagation theory is embedded in the machine learning loss function. Taking the detection of aramid fireproof fabric as an example, by dynamically weighting and fusing the timing error and the energy constraint error (such as automatically increasing the physical constraint weight to 0.5 when continuous energy anomalies are detected), the prediction error of aramid fabrics is reduced, and all prediction results meet Griffith's theory threshold (for example, the fracture toughness of aramid fibers is 5.0 J / m², and the measured energy release rate is ≥4.5 J / m²). In the detection of a certain fireproof clothing fabric, the model corrects the wrong path of "sudden fracture without stress concentration" (the proportion of such paths drops from 35% to 5%), identifies microcracks caused by high-speed impact under the coating, and warns of fabric failure 2 hours earlier than traditional methods.

[0031] Based on the embodiments provided in this application, by embedding the Griffith energy release rate threshold in the state transition probability matrix of the hidden Markov model, the deep integration of the material fracture mechanics theory and the sequential analysis of defect evolution is realized for the first time. Traditional machine learning models often ignore the physical law constraints when predicting crack propagation and may output unreasonable results that violate the stress-strain relationship. However, in this solution, by introducing the energy release rate constraint boundary in the state transition path, the model prediction process simultaneously satisfies the objective physical mechanism of "stress concentration leading to energy dissipation". The design of the joint loss function dynamically weights the time series prediction error and the physical law constraint error, and synchronously optimizes the time series characteristics of crack evolution and the material mechanics matching degree through gradient backpropagation, so that the generated tear strength prediction value not only reflects the dynamic change trend of defects (such as the spatio-temporal correlation of defect positions in continuous production), but also conforms to the energy conservation principle of the Griffith crack propagation theory.

[0032] Further, the implementation of the dual-channel attention mechanism includes: Using a temporal convolutional network to extract the temporal features from the yarn slip trajectory data and generate the spatio-temporal encoding vector of the slip displacement; Using a graph convolutional network to model the intermolecular interaction relationship from the molecular bond strength decay data and generate the topological feature vector of the bond strength; Dynamically calculating the correlation weight between the spatio-temporal encoding vector and the topological feature vector through a cross-attention layer to generate the attention distribution map of the joint mapping relationship for indicating the parameter adjustment of the cyclic tensile load.

[0033] In a specific implementation manner, the correlation weight of the dual-channel attention mechanism is calculated based on the following formula: ; where, is the yarn slip trajectory feature of the th tensile cycle, including displacement (mm), slip velocity (mm / s), and slip acceleration (mm / s²); P is the total number of tensile cycles, and for the knitted fabric, the stretching frequency of daily wear is simulated, and the scenario value is 50 times (such as the knee bending cycle of Lycra stretch fabric); Q is the total number of molecular bond nodes, corresponding to the number of bonding points of the fiber molecular chain. The cellulose chain of cotton fiber contains about 200 hydrogen bond nodes, so the scenario value is 200; is the strength feature of the qth molecular bond node, including bond energy (such as 0.5 eV for the hydrogen bond of cotton fiber), bond length (nm), and dihedral angle (°), which is detected by a near-infrared spectrometer; is the attention weight matrix of the slip trajectory feature; is the attention weight matrix of the molecular bond feature; is the bias vector, which optimizes the GELU activation function input. It is initially 0 and is adaptively adjusted through training. It is the associated weight of the qth bonding node in the pth cycle. For example, when the weight of a spandex node is 0.85, it is determined to be a high-risk point for failure.

[0034] Based on the above formula, this formula dynamically calculates the correlation weight between the yarn slip trajectory and the molecular bonding strength, realizing the cross-modal deep correlation analysis of knitted fabrics from macroscopic slip failure to microscopic bonding attenuation. Taking Lycra stretch fabric detection as an example, the traditional method can only detect obvious yarn breaks with a high missed detection rate. This formula focuses on the coupling area of "sudden increase in slip displacement + sudden drop in bonding strength" through attention weight (such as triggering high-precision detection when the weight of a coil node exceeds 0.8), thereby improving the detection accuracy of coil connection failure. Specifically, when the weight of the 150th bonding node of the spandex molecular chain suddenly increases by 40% in the 30th stretching cycle, the system can identify the risk of hydrogen bond breakage caused by slippage of the node in advance, and issue a warning 5 cycles earlier than the traditional single signal detection, avoiding the tearing of stretch fabrics due to missed detection.

[0035] Based on the embodiments provided in this application, a dual-channel attention mechanism for knitted fabrics is used to extract the temporal characteristics of the yarn slip trajectory through a temporal convolutional network, and a cross-scale correlation analysis model of "macroscopic slip behavior-microscopic bonding state" is constructed by combining the topological characteristics of the intermolecular interaction modeling with a graph convolutional network. Traditional detection methods for tear strength analysis of knitted fabrics usually only focus on the macroscopic displacement data under tensile loads, while this solution dynamically calculates the correlation weights of the two types of features through a cross-attention layer, which can accurately capture the attenuation law of the molecular bonding strength during yarn slip. For example, when the yarn slip rate in a certain area is strongly correlated with the probability of bond breakage of adjacent molecules, the system can automatically identify the area as a potential failure starting point. The establishment of this joint mapping relationship provides a more refined feature dimension for the failure mechanism analysis of knitted fabrics, enabling the detection system to perform customized defect diagnosis based on the coil structure characteristics of knitted fabrics, avoiding the problem of insufficient adaptability of the traditional "unified parameter" detection mode to special fabric structures.

[0036] Furthermore, the adjustment of the state transfer path includes: Based on Griffith crack growth theory, the stress intensity factor of the current crack growth direction is calculated to generate the energy release rate constraint boundary; In the Viterbi decoding process of the crack growth hidden Markov model, the state transition paths that exceed the constraint boundary are removed; The probability of state transition paths that conform to physical laws is redistributed to ensure that the crack propagation trend prediction satisfies both the dynamic defect evolution and the laws of material fracture mechanics.

[0037] In some embodiments of the present application, the Griffith theory constraint is embedded in the hidden Markov model, and the prediction is ensured to conform to the physical laws by presetting the energy threshold and the state transition rule.

[0038] For example, in the scenario of crack prediction for cotton fiber fabrics: When constructing the hidden Markov model, according to the measured value of the fracture toughness of cotton fibers, the Griffith energy release rate threshold is preset to 0.5 J / m² (determined by the standard single-filament fracture test, reflecting the critical energy for the rupture of hydrogen bonds between cotton fiber molecules). When the predicted energy of the crack propagation path exceeds this threshold by more than 10% (i.e., ≥ 0.55 J / m²), a state transition path adjustment is triggered: Calculate the constraint boundary based on the stress intensity factor formula (the formula is not explicitly shown, only the logic is described), remove the unreasonable paths of "sudden fracture" that exceed the boundary by 20% (such as the prediction of direct fracture without stress concentration), and perform probability reallocation on the paths that conform to energy conservation (for example, retain the paths of gradual energy release, and the probability weight is increased by 30%).

[0039] When dynamically weighting the joint loss function, the initial weight is preset as "time series prediction error: energy constraint error = 0.6:0.4". When the proportion of the energy constraint error exceeds 60% in 5 consecutive detections (such as the scenario of fabric aging resulting in a decrease in toughness), the weight is automatically adjusted to 0.5:0.5 to ensure that the model synchronously optimizes the matching degree between the crack evolution time series and the physical laws.

[0040] Based on the embodiments provided in the present application, in the Viterbi decoding process of the hidden Markov model for crack propagation, the stress intensity factor is calculated based on the Griffith theory and the energy release rate constraint boundary is generated, effectively filtering out the state transition paths that do not conform to the material fracture mechanics laws. When the traditional time series prediction model processes crack propagation data, it may generate prediction results that violate physical common sense due to data noise or model deviation (such as crack mutations without stress concentration). However, this solution ensures that the crack propagation trend prediction simultaneously meets the requirements of dynamic defect evolution laws and material fracture toughness by removing the paths that exceed the constraint boundary and performing probability reallocation on the effective paths. For example, when it is detected that the stress intensity factor in a certain area does not reach the material fracture threshold, the system will automatically suppress the unreasonable prediction of "sudden crack propagation", and instead preferentially select the path that conforms to the gradual change of the energy release rate, making the prediction result closer to the real physical process and improving the reliability and scientific basis of defect evolution prediction.

[0041] Furthermore, the non-contact detection module includes: Emitting ultrasonic waves with a set frequency through an air-coupled ultrasonic transducer, and extracting the characteristic frequency band of yarn fracture in the echo signal by using the wavelet packet decomposition algorithm; In the fiber fabric detection scenario (such as pure cotton canvas), the set frequency can be 200 kHz. In the polyester fiber fabric detection scenario (such as polyester industrial fabric), the set frequency can be 350 kHz.

[0042] Based on the acoustic feature patterns of the random forest classifier to distinguish fiber debonding and monofilament fracture, the classification results are used as the defect type criterion; Synchronously use a laser profilometer to obtain the three-dimensional topography data of the fabric surface, and locate the stress concentration area where the stiffness gradient exceeds the set gradient threshold through the curvature mutation detection algorithm.

[0043] In this embodiment, for low-risk fabrics (such as silk fabrics): set the gradient threshold: 8 N / mm². Because silk fibers are soft and fine and have low stiffness, slight abnormal fiber arrangement can cause stiffness mutation. When detecting the splitting defect of silk fabrics, when the stiffness gradient of a certain area exceeds 8 N / mm² and the curvature heat map shows a red highlight (Gaussian curvature > 0.008 mm⁻²), it is determined as a high-risk splitting point (the splitting of silk fabrics is usually caused by the accumulation of fiber slippage resulting in a sudden increase in stiffness).

[0044] For high-rigidity fabrics (such as aramid fireproof fabrics): set the gradient threshold: 20 N / mm² (aramid fibers themselves have high stiffness, and a larger gradient change is required to reflect defects). When detecting fiber fracture of aramid fabrics, when the stiffness gradient reaches 22 N / mm² and the ultrasonic detection captures a 350 kHz high-frequency fracture signal at the same position, the system confirms that this area is the stress concentration area (when aramid fibers fracture, it is accompanied by significant stiffness mutation and high-frequency stress waves).

[0045] Based on the embodiments provided in this application, in the non-contact detection module, the collaborative work of the air-coupled ultrasonic transducer and the laser profilometer constructs a dual non-destructive detection system of "acoustic feature analysis - topography defect location". The feature frequency band in the ultrasonic echo signal is extracted through the wavelet packet decomposition algorithm, and the acoustic modes of fiber debonding and monofilament fracture are distinguished by combining the random forest classifier, enabling the system to identify deep structural defects without contacting the fabric. This is crucial for the detection of vulnerable materials such as medical non-woven fabrics and cultural relic protection fabrics, avoiding physical damage that may be caused by traditional contact detection. The synchronously collected laser three-dimensional topography data locates the stress concentration area through curvature mutation detection, forming a cross-verification in the spatial dimension with the ultrasonic detection results, solving the problem of false defect judgment that may exist in a single non-contact detection technology (such as only relying on ultrasound), and providing technical support for multi-modal fusion for the internal defect detection of complex structure fabrics (such as multi-layer composite fabrics).

[0046] Further, the implementation of the curvature mutation detection algorithm includes: Perform Gaussian curvature calculation on the three-dimensional topography data to generate a curvature distribution heat map; The morphological gradient operator is used to extract the curvature mutation boundary and mark the potential starting point of tearing. The curvature mutation region is spatially matched with the defect position detected by ultrasonic inspection to verify the reliability of the stress concentration region.

[0047] In some embodiments of the present application, through the collaborative detection of ultrasonic and laser, the preset characteristic frequency band and the stiffness gradient threshold are used to locate the stress concentration region.

[0048] In a specific implementation manner, the multi-modal defect matching degree score is calculated based on the following formula: ; where is the total number of defect points to be matched. Each time of detection, high-confidence candidate points are extracted from the laser scanning and ultrasonic detection results. For example, the value is 10 to ensure coverage of the potential defect regions on the surface and inside of the fabric; is the summation index, corresponding to the nth defect point (n = 1, 2,..., N). For example, the 3rd point simultaneously satisfies the surface curvature mutation (stiffness gradient 16 N / mm²) and the abnormal ultrasonic signal (energy mutation in the 250 kHz frequency band), and its reliability is preferably verified; is the coordinate of the curvature mutation point located by the nth laser profile scan, collected by the laser profile scanner (accuracy 5 μm), with the unit of mm, reflecting the region where the surface stiffness gradient of the fabric exceeds 15 N / mm², such as the micro-deformation position (12.5 mm, 8.3 mm) at the intersection of the warp and weft of the woven fabric; is the coordinate of the defect point located by the nth air-coupled ultrasonic detection, located by the ultrasonic transducer (accuracy 0.1 mm), with the unit of mm, capturing the acoustic characteristic position of fiber fracture or debonding, such as the fiber fracture point (12.8 mm, 8.1 mm) under the coated fabric; is the Gaussian kernel width (spatial matching scale factor), which is dynamically valued according to the fabric type, 0.3 mm for knitted fabric (to adapt to the natural deformation error of the coil structure), and 0.8 mm for coated fabric (to compensate for the propagation delay error of the ultrasonic signal by the coating); is the multi-modal defect matching degree score, ranging from [0, 1]. When ≥ 0.8, it is determined as a high-reliability stress concentration area. For example, when the coordinate difference of a certain region is 0.3 mm, κ = 0.91 is calculated, and it is confirmed that there is a 0.2 mm micro-crack under the coating, solving the problem of missed detection in traditional detection.

[0049] Based on the above formula, the matching degree of the defect positions between laser scanning and ultrasonic detection is quantitatively calculated through the Gaussian kernel function, and a cross-modal reliability verification system is constructed, solving the problem of detecting hidden defects in coated fabrics and thick fabrics. Taking outdoor waterproof and breathable fabrics as an example, the traditional single ultrasonic detection has a high missed detection rate due to coating shielding. However, by setting the Gaussian kernel width σ = 0.8 mm exclusive to coated fabrics (compensating for the 0.5 mm positioning error caused by the coating), the detection accuracy of microcracks 0.2 mm under the coating is improved. In specific implementation, when the laser scanning detects a stiffness gradient of 16 N / mm² (exceeding the threshold of 15 N / mm²) at (15.2 mm, 8.3 mm), and the ultrasonic wave captures a fiber fracture signal at (15.5 mm, 8.0 mm), the calculated matching degree κ = 0.88. The system confirms that this area is the critical fracture area and guides the process to adjust the warp tension by ±10 cN, avoiding the waterproof failure accident caused by fiber fracture under the coating. For canvas with a thickness of 3 mm, through the adaptive adjustment of σ = 1.0 mm, the interlayer debonding detection rate is increased from 60% to 98%, significantly improving the safety performance detection ability of thick fabrics.

[0050] For example, in the detection scenarios of yarn breakage and fiber debonding: The air-coupled ultrasonic transducer emits ultrasonic waves at 250 kHz (the preset frequency band covers the fracture characteristic responses of cotton / polyester fibers). The wavelet packet decomposition algorithm extracts the frequency band of 200 - 300 kHz as the yarn breakage characteristic (when the energy ratio of this frequency band ≥ 40%, it is determined as a fracture signal), and the frequency band of 100 - 150 kHz as the fiber debonding characteristic (when the phase difference exceeds 180°, it is marked as debonding). The random forest classifier presets the classification confidence threshold as 70% (that is, when the classification consistency rate of a single decision tree ≥ 70%, the final defect type is output).

[0051] The laser profilometer acquires three-dimensional topography data with a resolution of 10 μm, and the preset threshold for Gaussian curvature calculation is (corresponding to the abnormal fiber arrangement at the 0.2 mm scale). The morphological gradient operator uses a 3×3 pixel window to extract the curvature mutation boundary. When the spatial deviation between the curvature mutation area and the defect position detected by ultrasonic is ≤ 0.5 mm, it is determined as a high-reliability stress concentration area (for example, when there are abnormal ultrasonic echoes and surface micro-convex deformations in a certain area at the same time, it is marked as the starting point of tearing).

[0052] Based on the embodiments provided in this application, the curvature mutation detection algorithm calculates the Gaussian curvature and performs morphological gradient processing on the three-dimensional topography data obtained by laser profile scanning, and can accurately locate the regions with abnormal surface stiffness gradients of the fabric. These regions are often the starting points of tearing failure. Traditional visual inspection, due to relying on gray-scale or texture features, is difficult to capture the stiffness differences caused by microscopic curvature changes. By spatially matching the curvature mutation boundaries with the defect positions detected by ultrasonic inspection, an "abnormal surface topography - internal structure defect" correlation verification mechanism is formed. For example, when the curvature mutation boundary in a certain region highly coincides with the fiber fracture characteristic frequency band in the ultrasonic echo signal, the system can determine that this region is a high-risk tearing point. This spatial registration and cross-verification of multi-modal data significantly improve the accuracy of stress concentration region localization, avoiding missed or misjudgments caused by insufficient data dimensions in a single detection technology, and providing a more reliable defect localization basis for accurate prediction of fabric tearing strength.

[0053] Furthermore, the implementation of the edge and cloud distributed architecture includes: Deploy a fault-tolerant detection model at the edge node in the weaving process, and use the bit-flip fault-tolerant mechanism to process the warp and weft yarn breakage data collected in real time; After receiving the abnormal data, the cloud platform simulates the tearing life under the actual scenario through the digital twin model, and inversely deduces the combination of process parameters that meet the target strength.

[0054] In some embodiments of this application, the edge node performs real-time fault tolerance processing, and the cloud inversely deduces process parameters, which involves presetting abnormal determination thresholds and parameter adjustment ranges.

[0055] For example, in the real-time monitoring scenario of the weaving process: the fault-tolerant detection model deployed at the edge node presets the abnormal determination rule for warp and weft yarn breakage as follows: the tension load fluctuation of 5 consecutive sampling points (sampling frequency 100Hz) exceeds ±20% of the nominal value (for example, when the nominal tension is 100cN, an alarm is triggered when >120cN or <80cN is continuously detected). The bit-flip fault-tolerant mechanism automatically filters out short-term noise (such as instantaneous fluctuations caused by equipment vibration, and abnormal signals with a duration <50ms are ignored).

[0056] After receiving the abnormal data, the cloud digital twin model presets the simulation parameters as follows: simulate the "tension-bending" composite stress scenario of the fabric in actual use at a time step of 0.1 seconds (such as the sudden impact condition of automotive airbag fabric). When inversely deducing the process parameters, the adjustment range of the warp yarn tension is limited to ±10cN of the current value (to avoid exceeding the stable adjustment range of the loom tension system), ensuring the engineering feasibility of the optimization plan.

[0057] Based on the embodiments provided in this application, the design of the edge and cloud distributed architecture realizes the hierarchical collaboration of "real-time detection - anomaly handling - process optimization" in the textile production process. Fault-tolerant detection models are deployed at the edge nodes in the weaving process, and a bit-flip fault-tolerant mechanism is used to process the weft and warp break data collected frequently. It can complete data noise reduction and anomaly identification within milliseconds, ensuring the real-time detection during the high-speed operation of the production line - this solves the problem of lag in quality control caused by network latency in the traditional cloud centralized processing mode. After receiving the anomaly data uploaded by the edge nodes, the cloud platform simulates the tear life of the fabric in the actual usage scenario (such as a complex outdoor stress environment) through a digital twin model, and inversely deduces the combination of process parameters (such as warp tension, weft density, etc.) that meet the target strength, transforming the detection data from "quality judgment basis" into "production process optimization input". For example, when the predicted value of the tear strength of a certain batch of fabric fluctuates, the cloud system can automatically trace back to the abnormal loom tension parameters and generate a specific adjustment plan.

[0058] Further, the method also includes an environment adaptive calibration step: Establish a mapping relationship table between temperature and humidity and material strength attenuation, dynamically correct the detection results through Gaussian process regression. When the environmental humidity exceeds the preset humidity sensitivity threshold, apply a random correction coefficient to the predicted value of the tear strength of cotton fabrics; Among them, the preset humidity sensitivity threshold can include but is not limited to 80%RH, 70%RH, etc.

[0059] Adopt an illumination-invariant feature extraction algorithm, separate the brightness component in the HSV color space, and eliminate the influence of color temperature fluctuation on fiber texture recognition through histogram specification.

[0060] In some embodiments of this application, the influence of environmental interference on the detection results is dynamically corrected through preset temperature and humidity thresholds and illumination processing algorithms.

[0061] For example, in the detection scenario of a high-humidity workshop (such as the finishing section of a cotton spinning mill): When establishing the temperature and humidity mapping relationship table, the preset humidity sensitivity threshold for cotton fabrics is 80%RH (when exceeding this threshold, the strength of cotton fibers decreases significantly due to moisture absorption). The Gaussian process regression model dynamically updates the correction coefficient with the historical data of the most recent 24 hours (including parameters such as temperature and humidity, fabric strength, production batch, etc.) (for example, when the measured humidity is 85%, the strength prediction value is automatically multiplied by a correction coefficient of 0.9 to reflect the strength decrease after moisture absorption).

[0062] In the light invariance processing, when separating the brightness component in the HSV color space, the preset allowable range of color temperature fluctuation is 5000K - 7000K (covering the common color temperature ranges of workshop LED lighting and natural light). The brightness values are normalized to the gray scale range of 120 - 180 through histogram specification (to eliminate the influence of brightness differences under different light sources on fiber texture recognition, for example, to avoid misjudging shadow areas as fiber breaks).

[0063] Based on the embodiments provided in this application, the environmental adaptive calibration step establishes a dynamic mapping relationship between temperature, humidity and material strength attenuation through Gaussian process regression. For humidity-sensitive materials such as cotton fabrics, a correction coefficient is automatically applied when the environmental humidity exceeds the threshold, solving the problem of deviation in detection results caused by fluctuations in environmental parameters in traditional detection methods. For example, in a high-humidity workshop environment, the hygroscopic expansion of cotton fibers will change their mechanical properties, and the dynamic correction mechanism of this solution can compensate for this change in real time to ensure the stability of detection results. The light invariance feature extraction algorithm separates the brightness component in the HSV color space and performs histogram specification, eliminating the influence of color temperature fluctuations of different light sources (such as sunlight, LED, sodium lamp) on fiber texture recognition, and avoiding misjudgment of defects (such as misidentifying shadow areas as fiber breaks) caused by changes in lighting conditions in traditional visual detection. The combination of these two technologies enables the detection system to maintain stable performance in complex industrial environments (such as temperature and humidity fluctuations of ±20%, light color temperature change > 1000K), without relying on the constant temperature and humidity laboratory environment required for traditional detection, significantly improving the scene adaptability of the device and the reliability of detection results.

[0064] Optionally, as shown in FIG. 3, this application provides a textile fabric tear strength detection system, including: A detection mode switching module 301, configured to synchronously collect fabric microstructure, macro-mechanics and composition data through a multi-modal sensing system, and dynamically switch the detection mode based on real-time recognition results; A tear strength prediction module 302, configured to input spatio-temporally correlated defect data into a physically constrained enhanced prediction model to generate crack propagation trends and tear strength prediction values; wherein, the prediction model integrates Griffith crack propagation theory and hidden Markov model to achieve double constraints of material mechanics laws and dynamic defect evolution; A process optimization module 303, configured to use a non-contact detection module and a flexible contact detection module to work together, and output a process optimization scheme in combination with an edge and cloud distributed architecture.

[0065] It should be noted that in this application, the embodiments implemented on the side of the textile fabric tear strength detection system can be mutually referred to the embodiments implemented on the side of the textile fabric tear strength detection method, and this application will not elaborate one by one.

[0066] The above are only the preferred embodiments of the present invention, and do not limit the patent scope of the present invention accordingly. Any equivalent structure or equivalent process transformation made by using the content of the specification and drawings of the present invention, or directly or indirectly applied in other related technical fields, shall be similarly included in the patent protection scope of the present invention.

Claims

1. A method for detecting the tearing strength of a textile fabric, characterized in that, Including: Synchronously collect fabric microstructure, macro-mechanics and composition data through a multi-modal sensing system, and dynamically switch the detection mode based on real-time recognition results; Input the defect data with spatio-temporal correlation into a prediction model with enhanced physical constraints to generate prediction values of crack propagation trend and tearing strength; wherein, the prediction model integrates Griffith crack propagation theory and hidden Markov model to achieve double constraints of material mechanics laws and dynamic defect evolution; Adopt a non-contact detection module and a flexible contact detection module to work together, and output a process optimization scheme in combination with an edge and cloud distributed architecture.

2. The method for detecting the tearing strength of a textile fabric according to claim 1, wherein, The realization of dynamically switching the detection mode includes: Obtain fiber molecular bonding information through a near-infrared spectrometer, combine the fabric surface texture captured by an industrial camera, and construct an adaptive feature pyramid network for classification to output fabric type and organizational structure; When it is identified as a knitted fabric, start a cyclic tensile load simulation module, synchronously extract the yarn slip trajectory data and molecular bonding strength attenuation data under the tensile load, dynamically allocate weights through a dual-channel attention mechanism, and generate a joint mapping relationship between the tensile load and the slip displacement; When it is identified as a coated fabric, activate a multi-scale residual convolution network, fuse the porosity distribution data of the penetrative infrared thermal imaging sequence and laser profile scanning, and locate the critical area of fiber fracture under the coating.

3. The method for detecting the tearing strength of a textile fabric according to claim 1, characterized in that, The construction of a prediction model with enhanced physical constraints includes: Embed the Griffith energy release rate threshold in the state transition probability matrix of the hidden Markov model. When the predicted energy of the crack propagation path exceeds the material fracture toughness, adjust the state transition path to conform to physical laws; Design a joint loss function, dynamically weight and fuse the time series prediction error of the hidden Markov model and the energy constraint error calculated by the Griffith crack propagation theory, and synchronously optimize the matching degree of crack evolution time series and physical laws through gradient backpropagation; According to the optimization result of the joint loss function, output the tearing strength prediction value that satisfies the double constraints of stress concentration and energy dissipation.

4. The textile fabric tearing strength detection method according to claim 2, characterized in that The realization of the dual-channel attention mechanism includes: Use a temporal convolutional network to extract temporal features from the yarn slip trajectory data to generate a spatio-temporal encoding vector of the slip displacement; Use a graph convolutional network to model the intermolecular interaction relationship of the molecular bonding strength attenuation data to generate a topological feature vector of the bonding strength; Dynamically calculate the correlation weight between the spatio-temporal encoding vector and the topological feature vector through a cross-attention layer to generate an attention distribution map of the joint mapping relationship for indicating the parameter adjustment of the cyclic tensile load.

5. The method for detecting the tearing strength of a textile fabric according to claim 3, wherein The adjustment of the state transition path includes: Calculate the stress intensity factor of the current crack propagation direction based on the Griffith crack propagation theory to generate an energy release rate constraint boundary; During the Viterbi decoding process of the crack propagation hidden Markov model, remove the state transition paths that exceed the constraint boundary; Perform probability reallocation on the state transition paths that conform to physical laws to ensure that the crack propagation trend prediction satisfies both dynamic defect evolution and material fracture mechanics laws.

6. The method for detecting the tearing strength of a textile fabric according to claim 1, wherein, The non-contact detection module includes: Ultrasonic waves of a set frequency are emitted through an air-coupled ultrasonic transducer, and the wavelet packet decomposition algorithm is used to extract the characteristic frequency band of yarn breakage in the echo signal; Based on a random forest classifier, the acoustic feature patterns of fiber debonding and monofilament breakage are distinguished, and the classification results are used as the criterion for defect type judgment; At the same time, a laser profile scanner is used to obtain the three-dimensional topography data of the fabric surface, and the stress concentration area where the stiffness gradient exceeds the set gradient threshold is located through the curvature mutation detection algorithm.

7. The method for detecting the tearing strength of a textile fabric according to claim 6, characterized in that, The implementation of the curvature mutation detection algorithm includes: Calculating the Gaussian curvature of the three-dimensional topography data to generate a curvature distribution heat map; Using a morphological gradient operator to extract the curvature mutation boundary and mark the potential tear starting point; Performing spatial matching between the curvature mutation area and the defect position detected by ultrasound to verify the reliability of the stress concentration area.

8. The method for detecting the tearing strength of a textile fabric according to claim 1, characterized in that, The implementation of the edge and cloud distributed architecture includes: Deploying a fault-tolerant detection model at the edge nodes in the weaving process, and using a bit-flip fault-tolerant mechanism to process the warp and weft yarn breakage data collected in real time; After receiving the abnormal data, the cloud platform simulates the tear life under the actual scenario through a digital twin model and inversely deduces the combination of process parameters that meet the target strength.

9. The method for detecting the tearing strength of a textile fabric according to claim 1, wherein, The method further includes an environmental adaptive calibration step: Establishing a mapping relationship table between temperature and humidity and material strength attenuation, dynamically correcting the detection results through Gaussian process regression, and applying a random correction coefficient to the predicted value of the tear strength of cotton fabrics when the environmental humidity exceeds the preset humidity sensitivity threshold; Using an illumination-invariant feature extraction algorithm to separate the luminance component in the HSV color space, and eliminating the influence of color temperature fluctuations on fiber texture recognition through histogram specification.

10. A textile fabric tearing strength detection system, characterized in that, Including: A detection mode switching module for synchronously collecting fabric microstructure, macro-mechanical and composition data through a multi-modal sensing system, and dynamically switching the detection mode based on the real-time recognition results; A tear strength prediction module for inputting the spatio-temporally correlated defect data into a prediction model with enhanced physical constraints to generate the crack propagation trend and the predicted value of the tear strength; among them, the prediction model integrates the Griffith crack propagation theory and the hidden Markov model to achieve double constraints on material mechanics laws and dynamic defect evolution; A process optimization module for using a non-contact detection module and a flexible contact detection module to work together, and outputting a process optimization scheme in combination with the edge and cloud distributed architecture.

Images