Fabric production quality online detection and judgment method fused with multi-modal data

Abstract

The invention belongs to the technical field of textile industry detection, and relates to a fabric production quality online detection and judgment method fused with multi-modal data, which comprises the following steps: applying periodic stress to a fabric moving along a transmission path to form a local stress response area; thermal reaction signals of a local stress response area are collected, and a phase-locked thermodynamic map is generated; extracting abnormal hot spots in the phase-locked thermodynamic map, and generating a defect candidate point coordinate set; generating a photoelastic form graph at the position indicated by the defect candidate point coordinate set; the thermal signal characteristics of the abnormal hot spots are matched with the photoelastic form graph, and a defect diagnosis report is generated; generating a real-time quality treatment instruction according to the type and severity level of the flaw in the flaw diagnosis report; according to the method, the problem that the defect classification capability is weak due to lack of judgment basis due to single information acquisition dimension and difficulty in comprehensively evaluating defect characteristics from multiple angles in a traditional scheme is solved.

Description

Technical Field

[0001] This invention belongs to the technical field of textile industry testing, and relates to an online detection and judgment method for fabric production quality that integrates multimodal data. Background Technology

[0002] In textile production, internal structural defects such as yarn knots, broken yarns, poor interlayer bonding, and abnormal fiber bundles are often hidden within the fabric's internal structure and are difficult to identify with the naked eye. These defects affect the product's performance and lifespan, becoming a challenge in controlling product quality. At the same time, textile production lines operate at high speeds, with fabrics being continuously conveyed at speeds of several meters per second, which places high demands on the response time and detection accuracy of the detection system.

[0003] Existing detection solutions include surface defect detection based on visual imaging and simple infrared or ultrasonic single-sensor technology. Although optical visual inspection can achieve a certain degree of automation, its detection range is limited to the fabric surface and its ability to identify internal defects is poor. Single-sensor technology such as simple infrared scanning or ultrasonic thickness measurement has a certain indicativeness, but it lacks highly specific physical characterization, is prone to false alarms and missed detections, and is difficult to accurately determine the specific type and severity of defects.

[0004] Traditional solutions rely on a single dimension of information acquisition, making it difficult to comprehensively assess defect characteristics from multiple perspectives. This results in a weak ability to classify defects due to a lack of judgment criteria. Summary of the Invention

[0005] To address the aforementioned problems, this invention provides an online detection and judgment method for fabric production quality that integrates multimodal data.

[0006] A method for online detection and judgment of fabric production quality integrating multimodal data includes the following steps: S1. Periodic stress is applied to the fabric moving along the conveying path to form a local stress response zone; S2. Collect thermal reaction signals in the local stress response zone and generate a phase-locked thermogram characterizing the location of the defect and the heat. S3. Extract abnormal hot spots in the phase-locked thermal image where the signal intensity is greater than the preset reference temperature change threshold, and record their location information in the fabric to generate a set of defect candidate point coordinates. S4. At the location indicated by the coordinate set of defect candidate points, use the cross-polarized light component to obtain the light field signal that changes polarization state when passing through the fabric, and generate a photoelastic morphology map that characterizes the stress distribution of defect points. S5. Match the thermal signal characteristics of abnormal hot spots with the photoelastic morphology diagram, and make a judgment based on the preset association rule library to generate a defect diagnosis report including the defect location, type and severity level. S6. Based on the type and severity level of the defects in the defect diagnosis report, compare them with the preset handling level and generate a real-time quality handling instruction.

[0007] A further aspect of the present invention involves forming a localized stress response zone, comprising the following steps: An ultrasonic phased array generator is used to generate a standing wave field with alternating high-pressure and low-pressure zones inside the fabric. When the fabric passes through this standing wave field, it generates a differentiated stress response at the defect location, forming a local stress response zone.

[0008] A further aspect of the present invention generates a phase-locked thermogram, comprising the following steps: Temperature change signals in local stress response zones were acquired using an infrared focal plane array. The acoustic heating effect signal was separated and extracted from the temperature change signal using a lock-in amplification algorithm; The acoustic heating effect signal is recombined according to its spatial location distribution to generate a phase-locked thermogram.

[0009] A further aspect of the present invention generates a set of candidate defect point coordinates, comprising the following steps: In the phase-locked thermal map, pixel clusters whose heat intensity exceeds the reference temperature change threshold are identified as abnormal hot spots; The depth of the defect in the fabric thickness direction is calculated based on the heat intensity of the abnormal hot spot and the preset heat diffusion model. By combining the location information of abnormal hot spots in the fabric conveying direction and the vertical conveying direction with the defect depth, a set of defect candidate point coordinates including three-dimensional location information is generated.

[0010] A further aspect of the present invention generates a photoelastic morphology diagram, comprising the following steps: The light field signal that changes polarization state due to stress birefringence after passing through the fabric is captured by an image acquisition device, and the light field signal is manifested as photoelastic stripes. The fringe features of the light field signal are digitally reconstructed to generate a photoelastic morphology map.

[0011] In a further embodiment of the present invention, the stripe features include stripe shape, stripe density, and stripe direction.

[0012] A further aspect of the present invention generates a defect diagnosis report, comprising the following steps: The thermal signal features are generated by extracting the hot spot area and heat intensity of the candidate defect points from the phase-locked thermal map. The thermal signal features and stripe features are matched against a pre-defined association rule base to determine the defect type and severity level, and generate a defect diagnosis report.

[0013] In a further aspect of this invention, the preset association rule base includes the following rules: When the hot spot area is greater than the first area threshold and the heat intensity is less than the first classification intensity threshold, the defect type is determined to be the first defect type. When the hot spot area is less than the second area threshold and the heat intensity is greater than the second classification intensity threshold, the defect type is determined to be the second defect type.

[0014] In a further embodiment of the present invention, the preset association rule base also includes: When the heat intensity or stripe density is less than the first classification threshold, the severity level is the first severity level. When the heat intensity or stripe density is greater than the first grading threshold and less than the second grading threshold, the severity level is the second severity level. When the heat intensity or stripe density exceeds the second-level threshold, the severity level is the third-level severity level.

[0015] A further aspect of the present invention generates real-time quality handling instructions, comprising the following steps: Based on the severity level in the defect diagnosis report, the defect is classified and processed. When the severity level is the first severity level, the first-level processing is performed, and when the severity level is the second severity level, the second-level processing is performed. When the severity level is the third severity level, level three processing shall be performed.

[0016] In summary, the present invention has the following beneficial technical effects: 1. By synchronously acquiring and correlating multiple dimensions of physical signals, such as thermal response and optical anomalies generated by acoustic excitation, defects hidden deep within the fabric can be located and characterized through clear physical manifestations, thereby improving the accuracy and specificity of detection and reducing false alarm and missed detection rates.

[0017] 2. By using ultrasonic standing wave fields as a stress excitation method, various hidden structural weaknesses can be stimulated deep into the internal space of the fabric, so that internal defects that are originally difficult to be directly perceived by external observation equipment can be transformed into detectable physical signals, reducing the probability of overlooking potential quality hazards in the entire production chain.

[0018] 3. The real-time automatic classification and handling mechanism transforms quality control from passive post-event inspection to proactive process intervention, making handling decisions based on the type and severity of defects, thus avoiding unnecessary production interruptions and efficiency losses. Attached Figure Description

[0019] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the accompanying drawings used in the description of the embodiments or the prior art will be briefly introduced below. The drawings are used to provide a further understanding of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0020] Figure 1 A flowchart illustrating an embodiment of this application is disclosed.

[0021] Figure 2 Structural schematic diagrams of embodiments of this application are disclosed. Detailed Implementation

[0022] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. All other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0023] The following is in conjunction with the appendix Figure 1 - Figure 2 A preferred description of the present invention is provided below.

[0024] See attached document Figure 1 This invention proposes an online detection and judgment method for fabric production quality that integrates multimodal data, including the following steps: S1. Periodic stress is applied to the fabric moving along the conveying path to form a local stress response zone; S2. Collect thermal reaction signals in the local stress response zone and generate a phase-locked thermogram characterizing the location of the defect and the heat. S3. Extract abnormal hot spots in the phase-locked thermal image where the signal intensity is greater than the preset reference temperature change threshold, and record their location information in the fabric to generate a set of defect candidate point coordinates. S4. At the location indicated by the coordinate set of defect candidate points, use the cross-polarized light component to obtain the light field signal that changes polarization state when passing through the fabric, and generate a photoelastic morphology map that characterizes the stress distribution of defect points. S5. Match the thermal signal characteristics of abnormal hot spots with the photoelastic morphology diagram, and make a judgment based on the preset association rule library to generate a defect diagnosis report including the defect location, type and severity level. S6. Based on the type and severity level of the defects in the defect diagnosis report, compare them with the preset handling level and generate a real-time quality handling instruction.

[0025] In one embodiment of the present invention, step S1 includes the following steps: An ultrasonic phased array generator is used to generate a standing wave field with alternating high-pressure and low-pressure zones inside the fabric. When the fabric passes through this standing wave field, it generates a differentiated stress response at the defect location, forming a local stress response zone.

[0026] Specifically, on the fabric production line's conveyor path, a section with a smooth fabric surface and stable operation is selected. An ultrasonic phased array generator is installed above or below this section using a non-contact installation method, ensuring a fixed, minute gap (e.g., 5 mm) between the generator and the fabric to avoid physical contact affecting the normal fabric conveyance. Subsequently, the ultrasonic phased array generator is activated, controlling the excitation phase and amplitude of its multiple internal piezoelectric transducer units to emit focused ultrasonic energy beams. These energy beams interfere at a predetermined depth within the fabric, forming a spatially stable standing wave field. This standing wave field consists of a series of alternating high-pressure and low-pressure zones distributed along the fabric width, serving as the stress excitation source applied to the fabric.

[0027] When fabric passes through this static standing wave field at a production line speed, for example, 2 meters per second, each area of ​​the fabric will experience periodic high-frequency compression and stretching. In structurally intact areas, this stress is absorbed and transmitted uniformly. However, in locations with internal structural weaknesses, such as fiber breaks, sparse yarn areas, or areas with weak interlayer bonding, the material's mechanical properties change, leading to a different response to high-frequency stress compared to the surrounding normal areas. This difference manifests as abnormally intensified vibrations and increased friction between fibers. This inconsistent mechanical behavior ultimately forms, on a macroscopic scale, specific regions with differentiated stress response characteristics—the localized stress response zone. The formation of this zone transforms the invisible internal structural problems into dynamic physical phenomena to be explored.

[0028] The ultrasonic phased array generator is a device that electronically controls and manipulates the direction and focus of the ultrasonic beam. It can synthesize a specific sound field by adjusting the excitation delay time of multiple independent transmitting units, and in this step, it is used to generate a standing wave field. A standing wave field is a wave field with a fixed spatial energy distribution, possessing continuous vibration peaks and troughs at specific locations, i.e., high-pressure and low-pressure regions. The stress excitation source refers to the periodically varying compressive and tensile forces, on the micrometer scale, applied by the standing wave field to the passing fabric fibers. Internal structural weaknesses refer to the general term for various defects that may occur during fabric production. Structurally, these can be classified as point defects such as knots, linear defects such as broken yarns, and surface defects such as thin weft threads.

[0029] Differential stress response refers to the phenomenon where, when a region with internal structural weaknesses is subjected to a stress excitation source, its internal vibration amplitude and frictional heat generation differ from those of the surrounding normal tissue. Its data characteristics manifest as localized abrupt changes in measurable thermal or optical signals in subsequent steps. A localized stress response zone is a specific physical area on the fabric that includes internal structural weaknesses and exhibits a differential stress response, serving as a direct physical marker of the presence of a defect.

[0030] For example, to generate an effective localized stress response zone, the fabric conveying speed on a standard textile production line is 2 meters per second. The selected ultrasonic phased array generator is set to a working frequency of 40 kHz. This frequency setting is based on empirical values ​​derived from acoustic characteristic experiments on over 100 common cotton-linen blended fabrics, ensuring effective coupling of ultrasonic energy into the fabric interior; this is used only as an example. The generator is installed 5 millimeters above the fabric surface. Upon startup, the control system adjusts the phase of the 128 array elements within the generator, forming a stable standing wave field at the center of the fabric thickness. The standing wave field covers a range of 1.5 meters along the fabric width, with an alternating interval of 2 millimeters between high-pressure and low-pressure zones. For instance, when a section of fabric with internal defects, such as tiny knots, passes through the standing wave field, the fibers in that defective area will experience abnormal vibration and friction due to structural discontinuity. This response contrasts sharply with the uniform vibration of the surrounding smooth area, thus creating a localized stress response zone with a diameter of approximately 1 millimeter at the knot location, where the physical properties have changed.

[0031] In one embodiment of the present invention, step S2 includes the following steps: The temperature change signal of the local stress response zone is acquired using an infrared focal plane array; the acoustic heating effect signal is separated and extracted from the temperature change signal using a lock-in amplification algorithm; the acoustic heating effect signal is recombined according to its spatial distribution to generate a lock-in thermogram.

[0032] Specifically, an infrared focal plane array is installed directly below or above the localized stress response zone in the fabric conveying path. This infrared focal plane array works by containing tiny thermal pixels, each capable of converting received infrared radiation into an electrical signal. During installation, the operating frequency and phase of the infrared focal plane array are synchronized and locked with the transmission frequency and phase of the ultrasonic phased array generator. This synchronization is achieved by establishing a common reference clock signal between the two devices, ensuring that each data acquisition moment of the infrared focal plane array corresponds to a specific stage of the periodic change in the standing wave field.

[0033] Subsequently, the high-speed acquisition mode of the infrared focal plane array was activated, enabling it to continuously capture images of the local stress response area at a high frame rate, such as 10,000 frames per second. During the acquisition process, each frame recorded the temperature distribution of the area at that instant. Due to the presence of environmental background thermal noise, such as radiant heat generated by machine operation and ambient temperature fluctuations, these acquired raw temperature images contained a large amount of irrelevant thermal signals, submerging the weak acoustic heating effect signals generated by the actual ultrasonic standing wave field excitation.

[0034] Therefore, to separate the real signal from the noise, lock-in amplification (LIA) technology is employed. The process involves: first, establishing a reference signal with the same phase and frequency as the transmitted signal from the ultrasonic phased array generator; then, digitizing each frame of temperature image acquired by the infrared focal plane array. Specifically, for each pixel in the image, the temperature curve of that pixel over time in the acquisition sequence is extracted. This temperature curve is then synchronously demodulated with the reference signal, i.e., the product of the two is calculated, and the product is integrated over time. Only those components whose temperature changes are in phase with the reference signal will accumulate after integration; random noise unrelated to the reference signal will cancel each other out during integration. Through this mathematical process, environmental noise is significantly reduced, and only the nanosecond to microsecond-level instantaneous temperature change signal generated solely by the ultrasonic standing wave field excitation is extracted. This extracted signal is called the acoustic heating effect signal.

[0035] After extraction, the acoustic heating effect signal is reorganized according to its spatial location on the fabric. For each two-dimensional coordinate position perpendicular to the fabric conveying direction, the mean or peak value of the acoustic heating effect signal at that position is calculated throughout the entire acquisition period to characterize the thermal response intensity at that location. Simultaneously, since the intensity of the acoustic heating effect is correlated with the depth of internal structural defects, it is also necessary to invert or estimate the equivalent depth information of the heat source based on the attenuation law of the thermal signal. All these thermal response intensity and depth information at various spatial locations are organized into a two-dimensional or three-dimensional image, where different pixel brightness or color represents the thermal response intensity at that location; this image is the phase-locked thermal map.

[0036] The output signal after phase-locked amplification is: ; Where P represents the output of the lock-in amplifier, in millikelvin; u(t) represents the real-time temperature deviation signal acquired by the infrared focal plane array, in millikelvin; T represents the integration time window, in seconds; ω represents the angular frequency of the reference signal, which is directly related to the operating frequency of the ultrasonic phased array generator, in radians per second. The phase of the reference signal is represented by radians; t represents the time variable, in seconds.

[0037] Among them, the infrared focal plane array is an optical sensor that can simultaneously acquire the infrared radiation distribution of a two-dimensional area, convert it into electronic signals, and form temperature image data. Key parameters of this device include spatial resolution and temporal resolution, i.e., the physical size in millimeters corresponding to each pixel and the acquisition frame rate. Its spatial resolution is set based on actual measurements of the defect scale range of n common textile fabrics, enabling it to distinguish heat sources from defects larger than 0.5 millimeters. Lock-in amplification (LLA) is a signal processing method that extracts weak components with the same frequency and phase as the reference signal from the noise background through demodulation and integration synchronized with the reference signal. This technology is based on the autocorrelation property of orthogonal signals after integration and their orthogonality to irrelevant noise. The acoustic heating effect signal refers to the heat change generated by abnormal vibration and inter-fiber friction in the internal region of the fabric when it is subjected to a stress excitation source. Its data structure is a temperature deviation field with respect to time and spatial location. The background thermal signal refers to thermal radiation from the environment and machines that is not generated by ultrasonic standing wave field excitation. It is uncorrelated with the reference signal, spatially irregularly distributed, and its amplitude is usually much larger than that of the acoustic heating effect signal. Phase-locked thermal imaging is image data generated by organizing the extracted acoustic-induced thermal effect signals according to spatial location. Each pixel or pixel cluster corresponds to a specific location on the fabric, and the size of the pixel value indicates the intensity of the thermal response at that location. The entire image forms a spatial mapping that characterizes the location of defects and the distribution of heat intensity.

[0038] For example, on an actual operating textile production line, an ultrasonic phased array generator is installed with a working frequency of 40 kHz and an initial emission phase of 0 degrees. An infrared focal plane array deployed below it uses the same 40 kHz reference frequency and 0-degree phase for synchronous acquisition. The spatial resolution of the infrared focal plane array is 0.1 mm per pixel, and the acquisition frame rate is 10,000 frames per second. An image is acquired at a specific location on the fabric, with an acquisition period of 100 milliseconds. In the original acquired infrared image, the temperature fluctuations around this location range from ±5 Kelvin, which includes approximately ±4.8 Kelvin of ambient background thermal signal and only approximately ±0.2 Kelvin of acoustic heating effect real signal. Using a 40 kHz reference signal with a 0-degree phase, phase-locked amplification is performed, and the integration time window is set to 100 milliseconds, i.e., the complete acquisition period, the P-value of a pixel location in the image is calculated over the entire period. The temperature curve of this pixel during the acquisition period exhibits high-frequency oscillations, with the amplitude and phase of the oscillations consistent with the reference signal. In the mathematical operations of lock-in amplification, this consistent component is enhanced, and the integral result will be positive, for example, several milliklvin. For adjacent pixels considered normal, their temperature curves contain no components related to the reference signal, only random noise. After the same lock-in amplification process, the integral result is approximately zero milliklvin. Arranging the lock-in amplified results of all pixels in the entire image according to pixel position yields a thermogram. Areas with abnormal defects appear as bright spots of 5 to 15 milliklvin, while normal areas remain at low values ​​close to zero milliklvin. This image is the lock-in thermogram, which reveals the thermal manifestation of invisible internal structural defects.

[0039] In one embodiment of the present invention, step S3 includes the following steps: In the phase-locked thermal image, pixel clusters with heat intensity exceeding the reference temperature change threshold are identified as abnormal hot spots. Based on the heat intensity of the abnormal hot spots and the preset heat diffusion model, the defect depth in the fabric thickness direction is calculated. The location information of the abnormal hot spots in the fabric conveying direction and the vertical conveying direction is combined with the defect depth to generate a set of defect candidate point coordinates including three-dimensional location information.

[0040] Specifically, the phase-locked thermal spectrum is analyzed to identify the locations of physical signal anomalies, and each anomalous hot spot is recorded using spatial coordinates, providing a clear detection target for subsequent optical verification steps.

[0041] First, a baseline judgment standard is established. Before the production line is fully operational, fabric samples that have been verified as qualified and free from any visible or known defects need to be completely collected and analyzed. These qualified samples are then tested using an ultrasonic phased array generator and an infrared focal plane array in the same manner as in step S2, generating corresponding phase-locked thermal maps. In these maps, the thermal response intensity at all pixel locations is uniformly distributed, with no obvious thermal signal anomalies. The average thermal response intensity of all pixels in the phase-locked thermal maps of these qualified samples is calculated; this average value is called the baseline temperature change threshold. The significance of the baseline temperature change threshold lies in the fact that it represents the normal thermal response level of the fabric structure under the same ultrasonic excitation when the fabric structure is intact. Thermal signal intensities exceeding this threshold indicate abnormal characteristics in the fabric structure at that location.

[0042] A real-time scanning process is then initiated. As the fabric to be inspected on the production line passes through the inspection device, the infrared focal plane array continuously acquires phase-locked thermal images. The scanning program traverses each newly acquired thermal image pixel by pixel. For each pixel location in the image, the thermal response intensity value at that pixel is compared with a reference temperature change threshold. Pixel locations with thermal response intensities greater than the reference temperature change threshold are marked as abnormal hotspot candidates. After completing the threshold comparison for a single pixel, pixel cluster identification is further performed. The principle of pixel cluster identification is that if multiple adjacent pixels are all marked as abnormal hotspot candidates, they should be merged and considered as the same physical defect source.

[0043] Specifically, a connectivity analysis algorithm is used to scan all marked anomalous pixels in the heatmap, identifying adjacent and continuous pixel groups, and defining each continuous pixel group as an anomalous hotspot. This avoids incorrectly splitting a single defect into multiple independent anomalous points due to changes in the spatial gradient of its thermal signal.

[0044] After identifying abnormal hotspots, their locations are converted into coordinates of the fabric in three-dimensional space.

[0045] The first coordinate dimension is the longitudinal position along the fabric conveying direction. Since the fabric moves at a constant speed, the detection system knows the time offset of the current moment relative to the detection device, and can calculate the distance of the abnormal hot spot from the reference point in the conveying direction by multiplying the time by the fabric speed.

[0046] The second coordinate dimension is the lateral position perpendicular to the fabric conveying direction. This information comes directly from the pixel position of the infrared focal plane array, where the pixel's position in the image row direction corresponds to the fabric width direction, and the pixel's position in the image column direction corresponds to the fabric length direction. Multiplying the pixel's column coordinate by the physical size of a unit pixel gives the lateral position of the hotspot.

[0047] The third coordinate dimension characterizes the depth of the defect, i.e., the distance of the defect from the surface along the fabric thickness direction. This depth cannot be directly read from the planar image; instead, it is inverted based on the principles of thermal diffusion physics. The principle is as follows: an ultrasonic standing wave field excites internal structural defects to generate heat, which diffuses outward along the fabric thickness direction. In the temperature distribution measured by an infrared focal plane array, the closer the heat source is to the detection surface, the higher the measured thermal signal intensity; the farther the heat source is from the detection surface, the lower the measured thermal signal intensity. By establishing a mathematical mapping relationship between the heat source depth and the measured thermal signal intensity, the equivalent depth can be inverted based on the thermal reaction intensity at the anomalous hot spot. Let the thermal reaction intensity at the anomalous hot spot be I. According to the solution of the thermal diffusion equation, there is an inversion relationship between the equivalent depth d of the heat source and the thermal reaction intensity I.

[0048] The 3D coordinates of all identified anomalous hotspots are collected to form a dataset. Each element in this dataset includes complete spatial information of the anomalous hotspot, including its longitudinal position along the fabric conveying direction, its lateral position perpendicular to the conveying direction, and its equivalent depth derived from the thermal reaction intensity. This dataset serves as the defect candidate point coordinate set, marking specific locations requiring focused inspection in subsequent optical inspection steps, enabling photoelastic detection to pinpoint areas with thermal anomalies.

[0049] The formula relating the depth of the heat source to the intensity of the thermal reaction is as follows: ; Where d represents the equivalent depth of the heat source, in millimeters; The thermal diffusivity of the fabric material is expressed in square millimeters per second. This represents the equivalent time from the generation of heat by the heat source to the diffusion of heat to the detection surface, measured in seconds. This equivalent time corresponds to the measured thermal reaction intensity; the lower the thermal reaction intensity, the... The larger the value, the deeper the heat source.

[0050] In some embodiments, the direct relationship between the heat source depth d and the measured thermal signal intensity I can also be used: ; in, This indicates the measured intensity of the thermal response, expressed in millikelvin. This indicates the reference thermal response intensity of the defect on the tested surface, expressed in millikelvin. represents the attenuation coefficient, which is related to the thermal properties of the fabric and is expressed as the reciprocal of the value in millimeters; d represents the equivalent depth of the defect, also expressed in millimeters.

[0051] The formula for calculating the longitudinal position coordinate is: ; Where X represents the longitudinal coordinate of the abnormal hot spot along the conveying direction, in millimeters; v represents the fabric conveying speed, in millimeters per second; and t represents the time interval from the reference time to the detection of the abnormal hot spot, in seconds.

[0052] The horizontal position coordinates are the pixel positions along the fabric width direction, and their calculation formula is as follows: ; Where Y represents the horizontal coordinate of the anomalous hotspot in millimeters; j represents the column pixel number of the anomalous hotspot in the infrared focal plane array image, dimensionless; and p represents the physical size of a unit pixel in millimeters.

[0053] The reference temperature change threshold refers to the numerical value of temperature change representing the normal thermal response level, obtained through statistical analysis of test data from verified qualified fabrics. This threshold is a scalar value, not a matrix or vector. The reference temperature change threshold can be set based on the statistical average of lock-in thermograms generated by at least 50 batches of qualified fabric samples under the same test conditions. This sample size is chosen based on a 95% confidence level requirement in industrial statistics. Abnormal hot spots refer to pixel locations or continuous regions formed by multiple adjacent pixels in the lock-in thermogram where the thermal response intensity exceeds the reference temperature change threshold. These hot spots have a clear spatial location, quantifiable thermal response intensity, and a physical correlation with fabric structural defects. A pixel is the basic unit in an infrared focal plane array image, possessing row and column coordinates and a corresponding thermal response intensity value. A pixel cluster is a collection of one or more adjacent pixels, defined by the principle that these pixels are connected through four or eight adjacent nodes (top, bottom, left, and right) and all satisfy the condition of exceeding the reference temperature change threshold.

[0054] The three-dimensional coordinate system is a spatial reference frame established based on the direction of fabric movement and geometry on the production line. Its three dimensions are the longitudinal position along the conveying direction, the transverse position perpendicular to the conveying direction, and the depth position obtained by thermal signal attenuation inversion.

[0055] The thermal intensity associated with defect depth refers to the physical correspondence between the thermal reaction intensity and the distance between the defect and the detected surface; its data structure is a functional relationship. The defect candidate point coordinate set is a dataset where each element includes the complete three-dimensional coordinate information of the abnormal hot spot and its corresponding thermal reaction intensity value. The data structure of the set is in list or matrix form, with the number of rows equal to the total number of detected abnormal hot spots, and the number of columns representing the three-dimensional coordinate heating reaction intensity.

[0056] For example, taking the fabric sample in step S2, assuming a baseline temperature change threshold of 3.5 millikrvin is determined based on preliminary statistics of 50 batches of qualified fabric samples. When a section of fabric to be inspected passes through the detection device, the phase-locked thermal image acquired by the infrared focal plane array shows a bright spot with an area of ​​approximately 0.3 square centimeters at a certain point in the image, with a pixel value of 8.7 millikrvin. Assuming the highest pixel position of this bright spot is at row 150 and column 350, the values ​​of the four adjacent pixel positions are 7.2, 7.9, 8.1, and 8.3 millikrvin, respectively, all exceeding the threshold, forming a connected pixel cluster. The unit pixel size of the infrared focal plane array is 0.1 millimeters. Through connectivity analysis, it is confirmed that these five pixels should be merged into an anomalous hot spot.

[0057] In the time dimension, if the time offset between the acquisition time of this frame image and the reference starting point of the production line is 2.5 seconds, and the fabric conveying speed is 2000 mm per second, the vertical coordinate X = 2000 × 2.5 = 5000 mm is calculated, indicating that the abnormal hot spot is 5 meters away from the reference starting point of the production line along the conveying direction.

[0058] The column number of the center pixel of the abnormal hot spot is 350. The horizontal coordinate Y = 350 × 0.1 = 35 mm is calculated, which means that the abnormal hot spot is located 35 mm away from the fabric width.

[0059] For depth inversion, the highest thermal response intensity of the anomalous hotspot, 8.7 milliklvin, is substituted into the depth inversion formula. Assuming the reference thermal response intensity of the fabric is 25, and the attenuation coefficient... If it is 0.1, then: 8.7 = 25 × exp(-0.1 × d), thus calculating d = 10.54 mm. This indicates that the equivalent depth of the defect is approximately 10.5 mm, meaning it exists approximately 10.5 mm inside the fabric in the thickness direction from the surface.

[0060] The complete information of this anomalous hotspot is recorded as a coordinate tuple: (5000 mm, 35 mm, 10.5 mm, 8.7 mKelvin), representing the location of the defect candidate point in three-dimensional space and its thermal reaction intensity. During the detection process, subsequent frames are scanned. If other anomalous hotspots are found, their coordinates are calculated in the same way, and these are aggregated into a defect candidate point coordinate set. Assuming that three anomalous hotspots are identified in this detection cycle, the defect candidate point coordinate set includes three records. This coordinate set will serve as a guide for the inspection location in photoelasticity testing.

[0061] In one embodiment of the present invention, step S4 includes the following steps: The light field signal, which changes polarization due to stress birefringence after passing through the fabric, is captured by an image acquisition device. This light field signal manifests as photoelastic stripes. The stripe features of the light field signal are digitally reconstructed to generate a photoelastic morphology map. The stripe features include stripe shape, stripe density, and stripe direction.

[0062] Specifically, firstly, based on the three-dimensional coordinate information of each defect candidate point recorded in the defect candidate point coordinate set, the physical position of that point in the production line coordinate system is calculated. For the longitudinal coordinate, the time delay of that point relative to the optical detection device is deduced based on the current movement speed of the fabric to determine the scanning time. For the transverse coordinate and depth information, these directly correspond to the positions in the fabric width and thickness directions. Using this coordinate information, the optical positioning system is driven, for example through a two-dimensional scanning mirror or a linear motion mechanism, to align the irradiation point of the cross-polarized light with the position of the defect candidate point.

[0063] A cross-polarized light system is then established, consisting of three optical components arranged in sequence: a polarizer, the fabric sample to be tested, and an analyzer. First, the light source is activated, allowing its emitted natural light to pass through the polarizer. The polarizer is a linear polarizer, characterized by allowing only light components oscillating in a specific direction, such as the horizontal, to pass through, while blocking light components perpendicular to that direction. After passing through the polarizer, the light becomes linearly polarized, its electric field vector oscillating only in a fixed direction. This linearly polarized light enters the fabric and interacts with the fabric's internal fiber structure and stress distribution. In normal areas with no stress or uniform stress distribution, the light passes directly through the fabric, and the polarization direction remains unchanged. However, in defective areas with non-uniform stress distribution, the fiber orientation and material density within the fabric change, causing this area to become an anisotropic optical medium. Light travels at different speeds in different directions in such an anisotropic medium; this phenomenon is known as stress birefringence.

[0064] Stress birefringence can be understood as follows: when the fibers and components of a fabric are subjected to uneven stress, its structure undergoes strain. This strain alters the refractive index of light within the material. At stress concentration points, there is a difference in refractive index along the stress direction and in the two principal directions perpendicular to the stress direction. Incident linearly polarized light can be decomposed into two orthogonally polarized components along the two principal directions. These two components propagate at different speeds due to their different refractive indices as they pass through the stress region, thus accumulating different phases. When these two components exit the fabric, a phase difference has already been established between them. When passing through an analyzer, the analyzer's polarization direction is set orthogonal to the polarizer's direction; that is, if the polarizer allows horizontally polarized light to pass, the analyzer only allows vertically polarized light to pass. In a cross-polarization configuration, when the fabric is stress-free or has uniform stress, the light emitted from the fabric still maintains the polarizer's polarization direction, perpendicular to the analyzer's polarization direction, and therefore cannot pass through the analyzer, resulting in a dark field. However, when there is uneven stress in the fabric, due to the phase difference introduced by stress birefringence, the light emitted from the fabric no longer maintains a single polarization direction, but becomes elliptically polarized light. This elliptically polarized light can be decomposed into components along the analyzer direction, so some light energy can pass through the analyzer. The intensity of the light passing through the analyzer varies with spatial position, and is related to the stress intensity and stress direction at that position, forming an alternating pattern of bright and dark stripes, i.e., photoelastic stripes.

[0065] An image acquisition device, such as a high-speed camera, is activated, its optical axis aligned with the defective area, to acquire the light field signal transmitted through a cross-polarized light system. The high-speed camera continuously captures photoelastic stripe images at a sufficiently high frame rate, such as 1000 frames per second. Each frame records the photoelastic stripe distribution in that area at that moment. Because the fabric is in continuous motion during transport, by acquiring multiple frames, photoelastic stripe data of the defective area from different angles or over multiple time periods can be obtained. The geometric shape, spatial density, and orientation of the stripes in these images contain rich information about the stress distribution at the defective point.

[0066] A series of acquired photoelastic fringe images were digitally processed. The process included: first, image enhancement and noise filtering to enhance the contrast and clarity of the fringes; then, fringe feature extraction was performed on each frame, including parameters such as fringe position, shape, direction, and spacing between adjacent fringes. Fringe shapes can be categorized into several typical types, such as diffuse fringes (dispersed areas formed by multiple adjacent fringes), star-shaped fringes (a group of fringes radiating outwards from the center point), and concentrated fringes (high-density clusters of fringes). The fringe spacing is positively correlated with the stress intensity at that location; a smaller spacing indicates a larger stress gradient and a higher degree of stress concentration. The fringe direction is perpendicular to the principal stress direction; analyzing the fringe direction allows determination of the principal stress direction.

[0067] All extracted stripe feature information is spatially integrated and organized. Repeated stripe features across multiple frames are fused and averaged to form a stable, comprehensive image representing the stress distribution pattern of the defect. This comprehensive image includes the spatial distribution of the stripe pattern, clearly showing the stress distribution characteristics of the defect area. This digitally processed and reconstructed image is the photoelastic morphology map. The function of the photoelastic morphology map is to provide detailed information about the stress distribution pattern of the defect for subsequent steps, becoming an important basis for defect type determination.

[0068] The formula relating phase difference to stress is: ; in, This represents the phase difference caused by stress birefringence, expressed in radians. This indicates the wavelength of incident light in a vacuum, measured in micrometers. The refractive index difference n1-n2 is dimensionless; This indicates the effective thickness through which light can propagate in the fabric, measured in micrometers.

[0069] The phase difference δ is related to the intensity of transmitted light through the following relationship: ; in, This indicates the intensity of light transmitted through the cross-polarization analyzer, expressed in relative units of luminous flux. This represents the initial intensity of the incident light. This is the phase difference mentioned above. When equal hour, An equal value of 0 corresponds to dark stripes in a stripe image; when equal hour, equal This corresponds to the bright stripes in the fringe image. The phase difference between adjacent stripes is π.

[0070] The relationship between fringe spacing and stress concentration factor is as follows: ; Where N represents the total number of fringes passing through the observation area, which is dimensionless; σ represents the maximum stress in the area, in Pascals; d represents the thickness dimension of the optical observation, in millimeters; C represents the stress optical constant of the material, in the reciprocal of Pascals; and λ represents the wavelength of light, in nanometers.

[0071] The cross-polarized light assembly is an optical system unit composed of a polarizer and an analyzer, used to generate light with a specific polarization state and analyze the polarization characteristics of transmitted light. The polarizer is an optical element, typically a polarizing film or polarizing crystal, whose function is to convert unpolarized or natural light into linearly polarized light; it has a fixed transmission axis direction. The analyzer has the same structure and principle as the polarizer; it is also a polarizing element, and its function is to analyze the polarization state of light, allowing only polarization components parallel to its transmission axis to pass through. Stress birefringence is an optical phenomenon caused by the spatial variation of the refractive index of a fabric due to the stress distribution within the fabric. The characteristic properties of this effect are that its intensity is positively correlated with the magnitude of the stress, and its direction is related to the direction of the principal stress. Photoelastic fringes are alternating bright and dark patterns caused by stress birefringence under cross-polarized light observation, including geometric parameters such as the shape, density, and direction of the fringes. The shape of the fringes can be classified into diffuse, star-shaped, and concentrated types, each corresponding to different stress distribution characteristics. The fringe density is the spacing between adjacent fringes, reflecting the magnitude of the stress gradient.

[0072] High-speed cameras are used to continuously capture rapidly changing optical signals at high frame rates. The parameters of a high-speed camera are set based on the fabric's movement speed and the spatial resolution requirements of photoelastic stripe details, typically using frame rates of 1000 frames per second or higher. Polarization state refers to the oscillation pattern of the electric field vector of a light wave in a plane perpendicular to the propagation direction; it can be linear, circular, or elliptical polarization. The light field signal refers to a two-dimensional image sequence acquired from a high-speed camera, including photoelastic stripe information. The photoelastic morphology map is a comprehensive image that characterizes the stress distribution pattern of defect points after digital processing and feature extraction. Its data structure is a two-dimensional matrix, where each element represents the stripe feature intensity or stress-related parameters at that location.

[0073] For example, continuing with the identified defect candidate point, the coordinates of which are (5000 mm, 35 mm, 10.5 mm, 8.7 mKelvin). The fabric conveying speed is 2000 mm per second. When the fabric moves to a point where the longitudinal distance between the defect point and the optical detection device is zero, cross-polarized light detection is triggered.

[0074] A cross-polarized light system was established, using a red laser with a wavelength of 635 nm as the light source. A polarizer was used to generate horizontally linearly polarized light. This linearly polarized light irradiated the fabric and interacted with its internal structure. In the area where the defect was located, stress concentration caused by yarn knots led to abnormal fiber arrangement, forming an anisotropic optical structure. The refractive indices of light in the two principal directions in this region were 1.505 and 1.490, with a difference Δn = 0.015. Assuming the effective propagation thickness of light in the fabric was 1 mm, the phase difference δ = (2π / 635 × 10⁻⁶) / ( ... -6)×0.015×1=9932×0.015≈149 radians≈47.4π.

[0075] Since δ≈47π, according to the principle of fringe formation, there should be approximately 47 bright and dark fringes in the observation area. The fringe spacing is λ / Δn=635 / 0.015≈42.3 micrometers.

[0076] A high-speed camera is activated to acquire an image sequence at a preset frame rate, for example, 2000 frames per second. Within a set acquisition time, such as 50 milliseconds, multiple image sequences can be obtained for analysis. By digitally processing and feature fusion of the acquired image sequence, a photoelastic morphology map of the defect can be generated. In this example, the generated photoelastic morphology map is expected to clearly show the stress distribution around the defect. For example, if a star-shaped stripe distribution is observed, it indicates that there may be a point stress concentration source at that location, and the surrounding stress exhibits a characteristic of gradually decreasing from the center outwards.

[0077] In one embodiment of the present invention, step S5 includes the following steps: The thermal spot area and heat intensity of the candidate defect points are extracted from the phase-locked thermal image to generate thermal signal features. The thermal signal features and stripe features are matched with the preset association rule library to determine the defect type and severity level, and generate a defect diagnosis report.

[0078] Specifically, by performing in-depth physical property diagnosis on the identified defect candidate points, the specific type and severity of the defect are ultimately determined. This process includes: using the defect candidate point coordinate set generated in step S3 as the core index to initiate the data association process. For each coordinate entry in the defect candidate point coordinate set, based on the three-dimensional coordinate information of that entry, the corresponding region of the defect on the phase-locked thermal map and the corresponding image on the photoelastic morphology map are located. Each potential defect point has paired data: a thermal signal data package describing its thermal response characteristics and an optical morphology data package describing its stress distribution pattern.

[0079] Next, feature extraction is performed on these two sets of paired data. For thermal signal features, the hot spot area at the candidate defect location is extracted from the phase-locked thermal image, which is the physical area covered by the pixel cluster exceeding the reference temperature change threshold; at the same time, the peak intensity of the hot spot is extracted, which is the maximum thermal reaction value in the region. For photoelastic morphology features, the system automatically identifies and classifies the macroscopic shape of the photoelastic stripes through image analysis algorithms, such as determining whether it is diffuse, star-shaped, concentrated, or linear; at the same time, the stripe density is calculated, which is the number of stripes per unit length in the region of highest stress concentration. The image analysis algorithm includes the following steps: First, the original photoelastic stripe image is subjected to noise reduction filtering and contrast enhancement, such as Gaussian filtering and histogram equalization, to improve image quality. Then, through image binarization and skeletonization thinning algorithms, the stripes are converted into skeleton lines with a single pixel width to represent their geometric structure. Based on the extracted skeleton lines, their shape features such as orientation consistency and centroid distribution are calculated, and the macroscopic shape of the stripes is determined to be a preset category such as diffuse, starburst, concentrated, or linear using predefined classification rules or classifiers. The stripe density calculation includes: delineating an analysis window in the region of highest stress concentration, counting the total pixel length of the skeleton lines within the window, and dividing it by the physical area of ​​the window to obtain the number of stripes per unit length, which serves as a quantitative indicator of stripe density.

[0080] After feature extraction, the system performs the core collaborative diagnostic judgment. This judgment process is based on a pre-set association rule base. This rule base was established before system deployment by conducting complete experiments (S1 to S4 steps) on a large number of known types of fabric defect samples, summarizing the intrinsic physical relationship between their thermal and optical reactions. The thermal signal features and photoelastic morphology features extracted from the current defect candidate point are used as input for matching queries in the rule base.

[0081] The rule base includes the following rules: A first area threshold and a second area threshold are set to determine the size of the hot spot area; a first classification intensity threshold and a second classification intensity threshold are set to distinguish heat intensity. When the hot spot area is greater than the first area threshold and the heat intensity is less than the first classification intensity threshold, it is determined to be a first defect type; when the hot spot area is less than the second area threshold and the heat intensity is greater than the second classification intensity threshold, it is determined to be a second defect type.

[0082] Meanwhile, the severity level is determined by a multi-level threshold based on signal strength or stripe density: when it is less than the first level threshold, it is the first severity level, such as mild; when it is between the first and second level thresholds, it is the second severity level, such as moderate; when it is greater than the second level threshold, it is the third severity level, such as severe.

[0083] It should be noted that the first and second classification intensity thresholds are intended to assist in determining the defect type, while the first and second grading thresholds are intended to assist in determining the severity level. The first and second classification intensity thresholds are classification boundaries set based on the thermal-optical physical mechanisms of different defect types, used to map abnormal responses to specific defect types. The first and second grading thresholds are grading boundaries set based on the quality tolerance standards within the same defect type, used to quantify the severity of the defect. The classification thresholds and grading thresholds are independent of each other, and their values ​​are calibrated separately based on experimental data from a large number of known samples. For example, a defect might be classified as a fiber bundle stress concentration type because its thermal intensity is higher than the second classification intensity threshold, and simultaneously classified as a moderate severity level because its thermal intensity is lower than the second grading threshold.

[0084] For example, if a candidate defect exhibits a large-area, sheet-like thermal signal, and its corresponding photoelastic morphology shows widely distributed, directional, diffuse photoelastic stripes, then the system classifies the defect as interlayer pseudo-adhesion, i.e., the first type of defect. This is because when interlayer pseudo-adhesion is subjected to ultrasonic excitation, large-area weak friction generates sheet-like heat sources. Simultaneously, this planar structural defect leads to uneven stress distribution but does not form concentrated stress points, thus presenting diffuse stripes. Another rule is: if the thermal signal exhibits a small-area but high-intensity point-like strong thermal signal, and its photoelastic morphology shows star-shaped or highly concentrated photoelastic stripes radiating outward from the center point, then the system classifies the defect as fiber bundle stress concentration, i.e., the second type of defect, such as yarn knots or foreign objects. This is because point-like defects generate concentrated high-frequency vibrations and friction under ultrasonic excitation, forming strong hot spots. Simultaneously, this point acts as a stress concentration source, and its stress field diffuses outward, forming star-shaped or concentrated stripe patterns.

[0085] While determining the type of defect, the system also assesses the severity level of the defect based on the extracted feature quantification values. The severity level determination is also based on a rule base. For example, for defects of the fiber bundle stress concentration type, the higher the thermal signal intensity or the greater the density of the photoelastic stripes, the more severe the stress concentration, and the higher the severity level, which can be divided into three levels: slight, moderate, and severe.

[0086] Finally, the system integrates all diagnostic results. For each defect point in the defect candidate coordinate set, structured information is generated, including its 3D location, the defect type determined through collaborative diagnosis, and the severity level assessed based on feature quantification values. This electronic document, which integrates all diagnostic information, is the defect diagnosis report.

[0087] It should be noted that thermal signal features are a set of key parameters extracted from phase-locked thermal maps to describe the thermal response characteristics of defects. Their data structure typically includes two values: hot spot area and intensity, in square millimeters and millikelvin, respectively. The hot spot area is set based on industrial classification standards for defect size, such as planar defects larger than 1 square centimeter. Stripe features are a set of key parameters extracted from photoelastic morphology diagrams to describe the stress distribution pattern of defects. The pre-defined association rule base is a decision knowledge base stored in the computer system, whose functional attribute is to provide a mapping relationship from multimodal features to defect types. Its data structure can be a lookup table or logical rules, and its content is based on statistical analysis and physical modeling of experimental data from more than n sets of known defect samples of different types. Sheet-like thermal signals refer to thermal regions on phase-locked thermal maps that are large in area, irregular in shape, and relatively uniform in intensity. Diffuse photoelastic stripes refer to stripe patterns on photoelastic morphology diagrams that lack a clear center, have a wide distribution range, and random orientation. Point-like strong thermal signals refer to hot spots on a phase-locked loop thermogram that are very small, almost pixel-like, but with an intensity value much higher than the background. Starburst or concentrated photoelastic stripes refer to patterns on a photoelastic morphology map where all stripes converge at a central point or radiate outwards from the central point. A defect diagnosis report is a structured electronic data file that provides a complete identity profile for each diagnosed defect. Its data structure typically consists of records with multiple fields, including defect ID, 3D coordinates, defect type, and severity level.

[0088] For example, consider the defect candidate point detected in step S4. The location of this defect candidate point is (5000 mm, 35 mm, 10.5 mm). First, associate the multimodal data of this point. From the phase-locked thermal map in step S2, the thermal signal features of this location are extracted as follows: hot spot area 0.2 square millimeters, intensity 8.7 milliklvin. Suppose that from the photoelastic morphology map in step S4, the stripe features of this location are extracted as follows: the stripe shape is star-shaped, and the stripe density reaches 25 stripes per millimeter in the central region. Combine these two features, namely the point-like strong thermal signal and the star-shaped shape, and match them with the preset association rule base. The rule base contains a threshold-based rule: if the hot spot area is less than a second area threshold and the intensity is greater than a second classification intensity threshold, it is determined to be the second defect type. The current defect features satisfy this rule, therefore the system determines that the defect type is the second defect type, namely fiber bundle stress concentration.

[0089] Next, a severity level assessment is performed. The severity level classification for the second defect type in the rule base is as follows: if the stripe density is greater than the second level threshold, for example, 20 stripes per millimeter, it is classified as the third severity level. Since the current defect has a stripe density of 25 stripes per millimeter, the system determines its severity level as severe.

[0090] Finally, the system generates a diagnostic record for the defect. The record contains the following information: Defect ID 001, Location (5000, 35, 10.5) mm, Defect Type: Fiber Bundle Stress Concentration, Severity Level: Severe. This record is added to the summary list to form the final defect diagnostic report, which is then passed to subsequent steps to trigger appropriate remedial measures.

[0091] In one embodiment of the present invention, step S6 includes the following steps: Based on the severity level in the defect diagnosis report, the defect is classified and processed. When the severity level is the first severity level, the first-level processing is performed, and when the severity level is the second severity level, the second-level processing is performed. When the severity level is the third severity level, level three processing shall be performed.

[0092] Specifically, a graded system for defect handling is established. Before the system is put into use, defect conditions corresponding to different handling levels are defined based on production process requirements and product quality acceptance standards. The core of this system is the handling level setting. This setting maps all possible diagnosed defect types and severity levels to three distinct handling levels: monitoring level, warning level, and shutdown level. The monitoring level corresponds to defects that have a minor impact on the final product quality but need to be recorded for subsequent quality traceability and process optimization. The warning level corresponds to defects of moderate severity; although they do not cause immediate production interruption, they are beyond acceptable limits and need to be clearly marked on the product for subsequent sorting or rework. The shutdown level corresponds to severe defects that may cause equipment damage or generate large-scale continuous scrap, requiring immediate interruption of the production process for investigation and repair.

[0093] The system performs graded processing based on the severity level: when the severity level is the first severity level, it performs level one processing, i.e., monitoring level, and only records defect information; when the severity level is the second severity level, it performs level two processing, i.e., warning level, and activates alarm equipment and marks on the fabric; when the severity level is the third severity level, it performs level three processing, i.e. shutdown level, and sends a shutdown signal to the main control system of the production line.

[0094] Once the system receives a defect diagnosis report, it initiates a tiered handling procedure. This procedure analyzes each defect record in the report in real time, extracting the defect type and severity level. Subsequently, this information is compared with preset handling levels to determine the appropriate handling level for the defect.

[0095] If a defect is determined to be at the monitoring level, such as a defect of type fiber bundle stress concentration but severity level of minor, the system will perform a data logging operation. For example, the complete diagnostic information of the defect, including its three-dimensional coordinates, type, severity level, and diagnostic timestamp, will be stored in a long-term quality database, without triggering any other physical actions.

[0096] If a defect is classified as a warning level defect, such as a pseudo-adhesion between layers with a medium severity level, the system will activate alarms and marking devices on the production line. For example, a control signal will illuminate a flashlight installed next to the operator's station and trigger a buzzer to emit a short warning sound to alert the operator. Simultaneously, the system calculates the time required for the defect to reach the marking system position based on its longitudinal coordinates on the fabric. When the defect moves below the marking system, the system activates an ink jetting system to spray small ink dots at a fixed distance from the fabric edge, which can be identified by subsequent processes, as a physical marker of the defect location.

[0097] If a defect is classified as a shutdown-level defect, such as a fiber bundle stress concentration defect with a severity level of "severe," or if more than three warning-level defects are detected consecutively, the system will immediately take intervention measures. For example, it will generate an emergency stop signal and send it directly to the production line's main control system, such as a programmable logic controller (PLC), via the industrial control bus. Upon receiving this stop signal, the main control system will immediately stop the fabric conveyor motors and related production equipment according to a preset safety stop procedure, thereby interrupting the entire production process.

[0098] After making any of the above-mentioned decision-making processes, the system needs to translate this decision into a standardized instruction that can be directly recognized and executed by downstream equipment. This instruction clearly includes the type of action to be performed, such as recording, marking, or stopping, as well as relevant parameters, such as the marking location and the stoppage reason code. This standardized electronic signal is the real-time quality handling instruction. This instruction is sent to the corresponding execution unit, such as a data server, marking controller, or production line main control system, thereby completing the closed-loop processing of the quality issue.

[0099] The handling level is a multi-dimensional set of decision boundaries, whose functional attribute defines the mapping relationship from defect characteristics to handling levels. Its data structure can be a table containing multiple rules, each specifying the defect type, severity range, and corresponding handling level. Monitoring level, alert level, and shutdown level are three predefined handling response levels, characterized by progressively increasing intervention intensity.

[0100] For example, the defect diagnosis report generated in step S5 includes a record: defect ID001, location (5000, 35, 10.5) mm, defect type fiber bundle stress concentration, severity level severe.

[0101] After receiving the report, the system analyzes the record, extracting the defect type as fiber bundle stress concentration and the severity level as severe. Then, the system queries the treatment level table. The table contains the following rule: any defect whose severity level is determined to be severe has a treatment level of shutdown. Based on this rule, the defect is determined to be at the shutdown level, and the system executes the shutdown treatment procedure.

[0102] See appendix Figure 2 This invention also proposes an online detection and judgment system for fabric production quality that integrates multimodal data, including the following modules: The stress excitation module is used to apply periodic stress to the fabric moving along the conveying path, forming a local stress response zone; The thermal response acquisition module is used to acquire the thermal response signal of the local stress response zone and generate a phase-locked thermogram characterizing the location of the defect and the heat. The anomaly detection module is used to extract abnormal hot spots in the phase-locked thermal image where the signal intensity is greater than the preset reference temperature change threshold, and record their location information in the fabric to generate a set of defect candidate point coordinates. The optical morphology detection module is used to acquire the light field signal that changes polarization state when passing through the fabric at the position indicated by the coordinate set of defect candidate points using a cross-polarized light component, and generate a photoelastic morphology map characterizing the stress distribution of defect points. The collaborative diagnosis and judgment module is used to match the thermal signal characteristics of abnormal hot spots with the photoelastic morphology diagram, and make judgments based on the preset association rule library to generate a defect diagnosis report including the defect location, type and severity level. The graded handling execution module compares the type and severity level of defects in the defect diagnosis report with the preset handling levels to generate real-time quality handling instructions.

[0103] Each of the modules can be implemented in whole or in part through software, hardware, or a combination thereof. It supports hardware embedded in or independent of the processor in the computer device, and also supports software stored in the memory of the computer device, so that the processor can call and execute the operations corresponding to each of the above modules.

[0104] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit them. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be included within the protection scope of the present invention.

Claims

1. A method for online detection and judgment of fabric production quality integrating multimodal data, characterized in that, Includes the following steps: S1. Periodic stress is applied to the fabric moving along the conveying path to form a local stress response zone; S2. Collect thermal reaction signals in the local stress response zone and generate a phase-locked thermogram characterizing the location of the defect and the heat. S3. Extract abnormal hot spots in the phase-locked thermal image where the signal intensity is greater than the preset reference temperature change threshold, and record their location information in the fabric to generate a set of defect candidate point coordinates. S4. At the location indicated by the coordinate set of defect candidate points, use the cross-polarized light component to obtain the light field signal that changes polarization state when passing through the fabric, and generate a photoelastic morphology map that characterizes the stress distribution of defect points. S5. Match the thermal signal characteristics of abnormal hot spots with the photoelastic morphology diagram, and make a judgment based on the preset association rule library to generate a defect diagnosis report including the defect location, type and severity level. S6. Based on the type and severity level of the defects in the defect diagnosis report, compare them with the preset handling level and generate a real-time quality handling instruction.

2. The method for online detection and judgment of fabric production quality by integrating multimodal data according to claim 1, characterized in that, The formation of a localized stress response zone includes the following steps: An ultrasonic phased array generator is used to generate a standing wave field with alternating high-pressure and low-pressure zones inside the fabric. When the fabric passes through this standing wave field, it generates a differentiated stress response at the defect location, forming a local stress response zone.

3. The method for online detection and judgment of fabric production quality by integrating multimodal data according to claim 2, characterized in that, Generating a lock-in thermogram includes the following steps: Temperature change signals in local stress response zones were acquired using an infrared focal plane array. The acoustic heating effect signal was separated and extracted from the temperature change signal using a lock-in amplification algorithm; The acoustic heating effect signal is recombined according to its spatial location distribution to generate a phase-locked thermogram.

4. The method for online detection and judgment of fabric production quality by integrating multimodal data according to claim 1, characterized in that, Generating a set of candidate defect point coordinates includes the following steps: In the phase-locked thermal map, pixel clusters whose heat intensity exceeds the reference temperature change threshold are identified as abnormal hot spots; The depth of the defect in the fabric thickness direction is calculated based on the heat intensity of the abnormal hot spot and the preset heat diffusion model. By combining the location information of abnormal hot spots in the fabric conveying direction and the vertical conveying direction with the defect depth, a set of defect candidate point coordinates including three-dimensional location information is generated.

5. The method for online detection and judgment of fabric production quality by integrating multimodal data according to claim 1, characterized in that, Generating a photoelastic morphology diagram includes the following steps: The light field signal that changes polarization state due to stress birefringence after passing through the fabric is captured by an image acquisition device, and the light field signal is manifested as photoelastic stripes. The fringe features of the light field signal are digitally reconstructed to generate a photoelastic morphology map.

6. The online detection and judgment method for fabric production quality integrating multimodal data according to claim 5, characterized in that, Stripe characteristics include stripe shape, stripe density, and stripe direction.

7. The method for online detection and judgment of fabric production quality by integrating multimodal data according to claim 1, characterized in that, Generating a defect diagnosis report includes the following steps: The thermal signal features are generated by extracting the hot spot area and heat intensity of the candidate defect points from the phase-locked thermal map. The thermal signal features and stripe features are matched against a pre-defined association rule base to determine the defect type and severity level, and generate a defect diagnosis report.

8. The method for online detection and judgment of fabric production quality by integrating multimodal data according to claim 7, characterized in that, The default association rule base includes the following rules: When the hot spot area is greater than the first area threshold and the heat intensity is less than the first classification intensity threshold, the defect type is determined to be the first defect type. When the hot spot area is less than the second area threshold and the heat intensity is greater than the second classification intensity threshold, the defect type is determined to be the second defect type.

9. The method for online detection and judgment of fabric production quality by integrating multimodal data according to claim 8, characterized in that, The pre-defined association rule base also includes: When the heat intensity or stripe density is less than the first classification threshold, the severity level is the first severity level. When the heat intensity or stripe density is greater than the first grading threshold and less than the second grading threshold, the severity level is the second severity level. When the heat intensity or stripe density exceeds the second-level threshold, the severity level is the third-level severity level.

10. The method for online detection and judgment of fabric production quality by integrating multimodal data according to claim 9, characterized in that, Generating real-time quality handling instructions includes the following steps: Based on the severity level in the defect diagnosis report, the defect is classified and processed. When the severity level is the first severity level, the first-level processing is performed, and when the severity level is the second severity level, the second-level processing is performed. When the severity level is the third severity level, level three processing shall be performed.

Images