Textile woven structure anomaly detection method

Abstract

The invention provides a textile woven structure anomaly detection method, which comprises the following steps: acquiring real-time tension monitoring data of warps and wefts in a jacquard textile weaving process, and establishing a tension distribution model in a normal weaving state according to the real-time tension monitoring data; if the classification result is in an abnormal state, spraying a pre-prepared chemical reagent on the surface of the abnormal position of the jacquard textile, and collecting spectral data of the surface of the textile after spraying; performing feature extraction and classification on the spectral data, and obtaining a comprehensive anomaly detection result by comparing a classification result with a tension position anomaly detection result; according to the comprehensive anomaly detection result, an abnormal area is accurately positioned and marked, and a sample in the abnormal area is cut and conveyed to a sample analysis device; and carrying out microstructure analysis on the cut sample, extracting fiber arrangement characteristics, judging the abnormal degree of fiber arrangement, and obtaining a microstructure abnormality evaluation result.

Description

Technical Field

[0001] The invention relates to the field of information technology, and in particular to a method for detecting abnormalities in a textile weaving structure. Background Art

[0002] Against the backdrop of the rapid development of the textile industry and the surge in demand for high-quality textiles, quality control in the production of jacquard textiles faces unprecedented challenges. In order to meet the market's expectations for high-quality textiles, ensuring that the tension and position deviation of the warp and weft threads during the weaving process are accurately monitored has become the key to improving product quality and production efficiency. However, achieving this goal is not easy, as it involves complex technical challenges such as the accuracy of real-time data acquisition, processing speed, and sensitivity of anomaly detection. This paper focuses on the core technical issues to be overcome, aiming to achieve real-time monitoring and anomaly detection of key quality parameters in the weaving process of jacquard textiles through technological innovation. First, real-time monitoring of the tension of the warp and weft threads requires the sensor to have extremely high sensitivity, and the data processing algorithm needs to be optimized to quickly identify deviations from the preset tension model to ensure that any tension anomalies can be captured in a timely manner. Secondly, the use of high-resolution image capture technology to accurately locate the actual position of the warp and weft threads during the weaving process requires high precision in image processing technology and perfect synchronization between the camera and the weaving machinery to accurately calculate the deviation from the theoretical position. To sum up, how to achieve real-time monitoring and abnormal detection of key quality parameters in the weaving process of jacquard textiles through high-tech means without affecting production efficiency, so as to ensure the quality of textiles and improve production efficiency at the same time, is a technical problem that needs to be solved urgently. Summary of the invention

[0003] The present invention provides a method for detecting abnormalities in a textile weaving structure, which mainly includes:

[0004] Acquire real-time tension monitoring data of the warp and weft during the weaving process of the jacquard textile, and establish a tension distribution model under normal weaving conditions according to the real-time tension monitoring data;

[0005] Collecting real-time images of the jacquard textile during weaving, accurately locating the positions of the warp and weft lines according to the real-time images, and obtaining the deviation between the actual positions of the warp and weft lines and the theoretical positions;

[0006] Classifying the abnormal state of the jacquard weaving structure according to the deviation between the tension distribution model and the real-time tension monitoring data, and the deviation between the actual position of the warp and weft lines and the theoretical position;

[0007] If the classification result is an abnormal state, spraying a pre-prepared chemical reagent on the surface of the abnormal position of the jacquard textile, and collecting spectral data of the textile surface after spraying;

[0008] Extracting features and classifying the spectral data, and obtaining a comprehensive anomaly detection result by comparing the classification result with the tension position anomaly detection result;

[0009] According to the comprehensive abnormality detection result, the abnormal area is accurately located and marked, and the sample of the abnormal area is cut and transported to the sample analysis device;

[0010] The cut samples are subjected to microstructure analysis to extract fiber arrangement characteristics, determine the degree of fiber arrangement abnormality, and obtain microstructure abnormality assessment results.

[0011] The technical solution provided by the embodiment of the present invention may have the following beneficial effects:

[0012] The present invention discloses a method for detecting abnormalities in textile weaving structures, which ensures that real-time monitoring and abnormality detection of key quality parameters are achieved without affecting production efficiency. Specifically, the sensor continuously monitors the tension of the warp and weft lines, establishes and maintains a reference model of the tension distribution under normal weaving conditions, so as to instantly identify any deviation from the ideal state. At the same time, the high-resolution camera is combined with the image processing algorithm to accurately locate the position of the warp and weft lines during the weaving process, quickly calculate the deviation value from the theoretical position, and ensure the geometric accuracy of the weaving structure. The tension data is further fused with the position deviation data to classify and mark the abnormal state of the weaving structure. Once the abnormal state is detected, the system immediately sprays a specific chemical reagent on the surface of the abnormal area and collects spectral data for analysis. Through cross-validation of spectral data analysis and tension position abnormality detection results, the abnormal area can be accurately located and its severity can be evaluated. Finally, the tension and position deviation in the weaving process can be effectively monitored and automatically adjusted to ensure the quality and production stability of the textile. By timely discovering and handling weaving abnormalities, not only can the production efficiency and product quality be significantly improved, but also the scrap rate can be greatly reduced and the production cost can be reduced. In addition, this method can not only monitor and adjust the tension and position deviation in the weaving process in real time to ensure the quality and stability of the textiles, but also detect and deal with weaving abnormalities in a timely manner, improve production efficiency and product quality, and reduce scrap rate. Its wide applicability enables it to be promoted and applied in the production of various types of textiles. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] Figure 1 The present invention is a flowchart of a method for detecting abnormalities in a textile weaving structure.

[0014] Figure 2 It is a schematic diagram of a method for detecting abnormalities in textile weaving structures according to the present invention.

[0015] Figure 3 It is another schematic diagram of a method for detecting abnormalities in textile weaving structures according to the present invention. DETAILED DESCRIPTION

[0016] In order to make the purpose, technical solutions and advantages of the present invention more clear, the present invention is described in detail below with reference to the accompanying drawings and specific embodiments.

[0017] like Figure 1 -3. In this embodiment, a method for detecting abnormality in a textile weaving structure may specifically include:

[0018] Step S101, obtaining real-time tension monitoring data of the warp and weft during the weaving process of the jacquard textile, and establishing a tension distribution model under a normal weaving state according to the real-time tension monitoring data.

[0019] The data collected by the warp and weft tension sensors during the weaving process of jacquard textiles are obtained, and the data are preprocessed, including removing outliers and smoothing. The mean and standard deviation of the tension data are calculated based on the preprocessed data to determine the tension value range of the normal weaving state. The tension curve function is fitted using a polynomial regression method to obtain the relationship expression between tension and fabric parameters. The time series of the tension data is analyzed to predict the real-time tension values ​​of the warp and weft at different loom speeds. The actual collected tension value is compared with the predicted value to calculate the tension deviation.

[0020] Specifically, the data collected by the warp and weft tension sensors during the weaving process of jacquard textiles are obtained, and the data are preprocessed, including removing outliers and smoothing. The mean and standard deviation of the tension data are calculated to determine the tension value range of the normal weaving state. A tension distribution model is constructed for the combination of different yarn strengths and fabric densities. The tension curve function is fitted by the polynomial regression method to obtain the relationship expression between tension and fabric parameters. The warp and weft tension change rate is calculated according to the fabric width and yarn elongation parameters. A tension prediction model is established in combination with fabric structure information, including warp and weft density and tissue point diagram. The time series data is analyzed using the ARIMA algorithm to predict the real-time tension values ​​of the warp and weft at different loom speeds. The prediction results are dynamically updated by the sliding window method to improve the prediction accuracy. The actual collected tension data is compared with the predicted value to calculate the tension deviation. The tension deviation threshold is set, and when the deviation exceeds the threshold, the abnormal detection mechanism is triggered. The relationship between the tension deviation and the fabric surface flatness and wear resistance index is analyzed by the multivariate linear regression method. The main factors affecting the tension change are determined according to the regression coefficient. According to the tension deviation analysis results, the warp density and weft density parameters are adjusted. The PID control algorithm is used to adjust the loom speed in real time to achieve the dynamic balance of warp and weft tension. The proportional, integral and differential parameters of the PID controller are determined by the Ziegler-Nichols tuning method. The tension change trend is monitored in real time, and the tension balance control is completed when the tension fluctuation amplitude is reduced to the preset range. During the weaving process of jacquard textiles, the warp and weft tension sensors collect 100 data points per second. The collected data are preprocessed, the median filter is used to remove the outliers, and then the Gaussian filter is applied for smoothing. The mean and standard deviation of the processed data are calculated, and the tension value range of the normal weaving state is determined to be ±2 times the standard deviation of the mean. A tension distribution model is constructed for different yarn strengths, such as 20cN, 30cN, 40cN and fabric densities, such as 100 strands / inch, 120 strands / inch, and 140 strands / inch. The third-order polynomial regression is used to fit the tension curve function, and the relationship expression between tension and fabric parameters is obtained. According to the preset fabric width, such as 1.5 meters and yarn elongation, such as 2%, the change rate of warp and weft tension is calculated. Combined with fabric structure information, such as plain weave with a warp density of 200 strands / 10cm and a weft density of 180 strands / 10cm, a tension prediction model is established. The ARIMA (1, 1, 1) model is used to analyze time series data and predict the real-time tension values ​​of warp and weft at different loom speeds, such as 300rpm, 400rpm, and 500rpm. The prediction results are dynamically updated through a 60-second sliding window to improve the prediction accuracy. The actual collected tension data is compared with the predicted value to calculate the tension deviation. The tension deviation threshold is set to ±5%. When the deviation exceeds the threshold, the abnormal detection mechanism is triggered.The relationship between tension deviation and fabric surface flatness, such as roughness index Ra and wear resistance index, such as Martindale wear number, was analyzed by multivariate linear regression method. The main factors affecting tension change were determined according to the regression coefficient, such as the influence coefficient of surface flatness was 0.6 and the influence coefficient of wear resistance was 0.4. According to the results of tension deviation analysis, the warp density and weft density parameters were adjusted. The PID control algorithm was used to adjust the loom speed in real time to achieve the dynamic balance of warp and weft tension. The proportional, integral and differential parameters of the PID controller were determined by the Ziegler-Nichols setting method, and the initial values ​​were set to Kp=0.6, Ki=1.2 and Kd=0.075 respectively. The tension change trend was monitored in real time. When the tension fluctuation amplitude was reduced to within the preset range of ±2%, the tension balance control was completed. Data acquisition and processing were performed every 100 milliseconds during the whole process to ensure real-time and accuracy.

[0021] Step S102, collecting real-time images of the jacquard textile during the weaving process, accurately positioning the positions of the warp and weft lines according to the real-time images, and obtaining the deviations between the actual positions of the warp and weft lines and the theoretical positions.

[0022] A real-time image of a jacquard textile weaving process is acquired, and the real-time image is processed by a multi-scale image analysis method to obtain an enhanced image; edge detection is performed on the enhanced image to extract the contour of the warp and weft lines, and the contour is gradually made to fit the actual shape of the warp and weft lines by an iterative optimization method; a support vector machine algorithm is used to locate the intersection of the warp and weft lines according to the identified position information of the warp and weft lines, and a support vector machine model is trained to identify the intersection by extracting local image features as input; the actual direction of the warp and weft lines is fitted to ensure the continuity and smoothness of the curve; the theoretical position of the warp and weft lines is calculated, and the actual position of the fitted warp and weft lines is compared with the theoretical position, and the Euclidean distance is calculated to obtain the position deviation; if the position deviation exceeds a preset dynamic threshold, it is determined that the position of the warp and weft lines is abnormal.

[0023] Specifically, real-time images of jacquard textiles in the weaving process are obtained from a high-speed camera, and the real-time images are processed by a multi-scale image analysis method. According to different lighting conditions and fabric structures, the image is decomposed at multiple scales by wavelet transform, and denoising, contrast enhancement and distortion correction are performed at different scales. Combined with the image pyramid technology, the processing results of each scale are fused to obtain an enhanced image. Edge detection is performed on the processed image, and the contours of the warp and weft are extracted by combining the Canny operator and the Sobel operator. The active contour model is introduced, and the energy function is constructed according to the color and width characteristics of the warp and weft. The contour is gradually fitted to the actual shape of the warp and weft through an iterative optimization algorithm to obtain the precise position information of the warp and weft. According to the identified position information of the warp and weft, the support vector machine algorithm is used to locate the intersection of the warp and weft combined with the pre-set fabric structure parameters and the warp and weft density data. By extracting local image features as input, the support vector machine model is trained to identify the intersection. The actual direction of the warp and weft is fitted with the B-spline curve to ensure the continuity and smoothness of the curve. The theoretical position of the warp and weft is calculated according to the fabric design parameters and the kinematics of the loom. The actual position of the fitted longitude and latitude lines is compared with the theoretical position, and the Euclidean distance is calculated to obtain the position deviation. A dynamic threshold is set to automatically adjust the deviation threshold according to the fabric type and production speed. The sliding window averaging method is used to smooth the deviation data between consecutive frames and output the real-time position deviation results of the longitude and latitude lines. During the weaving process of jacquard textiles, a high-speed camera collects real-time images with a resolution of 1920x1080 at a rate of 120 frames per second. The image is decomposed into four scales by wavelet transform, corresponding to resolutions of 1920x1080, 960x540, 480x270 and 240x135, respectively. At each scale, adaptive median filtering is used to remove noise, contrast-limited adaptive histogram equalization is used to enhance image contrast, and a distortion correction algorithm based on phase consistency is used to correct lens distortion. The image pyramid technology fuses the processing results of each scale with weights of 0.4, 0.3, 0.2, and 0.1 to obtain the enhanced image. Edge detection was performed on the fused image. The high and low thresholds of the Canny operator were set to 50 and 150, and the convolution kernel size of the Sobel operator was 3x3. The energy function of the active contour model contains image gradient terms, curve smoothing terms, and color constraint terms, with weights of 0.5, 0.3, and 0.2, respectively. The gradient descent method was used for iterative optimization, with a maximum number of iterations of 100 and a convergence threshold of 0.001. The support vector machine algorithm used a radial basis kernel function with a kernel parameter γ = 0.1 and a penalty parameter C = 10. The training samples included 10,000 positive samples and 30,000 negative samples. The B-spline curve used cubic splines with a control point interval of 10 pixels. The kinematic model of the loom takes into account the loom speed, weft beating frequency, and warp and weft tension to calculate the theoretical position.The dynamic threshold of position deviation sets the reference value according to the fabric type, such as 2mm for plain fabric and 3mm for satin fabric, and adjusts linearly with the production speed. The threshold increases by 0.5mm for every 100 rpm increase. The window size of the sliding window averaging method is 5 frames, and each frame slides 1 frame. Data collection and processing are performed every 8.33 milliseconds (120Hz) to ensure real-time and accuracy.

[0024] Step S103, classifying the abnormal state of the jacquard weaving structure according to the deviation between the tension distribution model and the real-time tension monitoring data, and the deviation between the actual position of the warp and weft lines and the theoretical position.

[0025] The tension deviation value between the tension distribution model and the real-time tension monitoring data is obtained, and the position deviation value between the actual position of the warp and weft lines and the theoretical position is obtained. A multidimensional feature vector is established according to the tension deviation value, the position deviation value, the yarn type, the loom speed, the weft beating frequency, the warp and weft tension and the fabric width. The multidimensional feature vector is normalized by the Z-score standardization method to obtain a standardized feature vector. The standardized feature vector is subjected to principal component analysis, and the principal component whose cumulative contribution rate exceeds the preset threshold is selected to construct a reduced dimension feature space. The Mahalanobis distance is used to calculate the distance value between the current state and the preset normal state sample set, and the abnormality degree is divided according to the distance value. The abnormality degree and the position deviation distribution of the intersection of the warp and weft lines are preliminarily classified. Based on the preliminary classification results, the pre-established abnormal state knowledge base is combined to perform refined classification to obtain the final classification result of the abnormal state of the jacquard weaving structure.

[0026] Specifically, the deviation value between the tension distribution model and the real-time tension monitoring data, as well as the deviation value between the actual position of the warp and weft and the theoretical position, are obtained. For different fabric structures and warp and weft densities, a multidimensional feature vector is established, including parameters such as tension deviation, position deviation, yarn type, loom speed, weft beating frequency, warp and weft tension, and fabric width. The Z-score standardization method is used to normalize each parameter to eliminate the differences between different dimensions and obtain the standardized feature vector. The principal component analysis is performed on the standardized feature vector, and the principal components with a cumulative contribution rate of more than 95% are selected to construct the feature space after dimensionality reduction. The Mahalanobis distance is used to calculate the distance between the current state and the preset normal state sample set, and the normal state sample set is obtained through historical data statistics. According to the calculated distance value, a threshold is set to divide the abnormality degree and obtain the abnormality degree index. Combined with the abnormality degree index and the position deviation distribution of the intersection of the warp and weft lines, the X-means clustering algorithm is used to preliminarily classify the abnormal state of the jacquard weaving structure. The X-means algorithm automatically determines the optimal number of clusters through the Bayesian information criterion, overcoming the limitation of the fixed number of clusters. The clustering results are iteratively optimized to obtain the preliminary classification results of abnormal states and the feature descriptions of each category. Based on the preliminary classification results, the random forest algorithm is used for refined classification in combination with the pre-established abnormal state knowledge base. The abnormal state knowledge base contains feature descriptions and processing suggestions for known abnormal types. The random forest consists of multiple decision trees, with the maximum depth of each tree set to 10 and the minimum number of sample splits set to 5. The classification results of multiple trees are combined through a voting mechanism to obtain the final classification results of the abnormal state of the jacquard weaving structure, including the abnormal type, severity and corresponding processing suggestions. During the weaving process of jacquard textiles, tension and position data are collected in real time to construct a 10-dimensional feature vector, including warp tension deviation -5N to +5N, weft tension deviation -3N to +3N, warp position deviation -2mm to +2mm, weft position deviation -1.5mm to +1.5mm, yarn type (cotton, polyester, blended), loom speed 400-800rpm, weft beating frequency 100-200 times / minute, warp tension 20-50N, weft tension 15-40N and fabric width 1.53m. The Z-score method is used for standardization to convert the data of each dimension into a distribution with a mean of 0 and a standard deviation of 1. The principal component analysis of the standardized feature vector is performed, and the first five principal components are selected, with a cumulative contribution rate of 96.8%. The Mahalanobis distance is used to calculate the distance between the current state and the normal state sample set, which contains 1000 historical normal samples. The distance threshold is set to 10.5, and it is judged as abnormal if it is greater than this value. The X-means clustering algorithm initially sets 210 cluster centers, which are finally determined as 6 clusters through iterative optimization, corresponding to 6 abnormal states. The random forest algorithm constructs 100 decision trees, with a maximum depth of 10 and a minimum sample split number of 5 for each tree.Input features include abnormality index, clustering results and warp and weft intersection position deviation distribution (20x20 grid). The final classification results include abnormality type, such as warp breakage, weft loss, structural deviation, etc., severity, including minor, moderate, severe and treatment suggestions, such as adjusting tension, replacing damaged parts, and re-threading. This is done once per second to achieve real-time monitoring and classification of abnormal jacquard weaving structure status.

[0027] Step S104: if the classification result is an abnormal state, spray a pre-prepared chemical reagent on the surface of the abnormal position of the jacquard textile, and collect spectral data of the textile surface after spraying.

[0028] The abnormal state classification result of the jacquard textile is obtained, and a corresponding reagent formula is selected from a preset chemical reagent database; a chemical reagent is prepared according to the reagent formula, and the chemical reagent is used for spraying treatment; the coordinate information of the abnormal position is obtained, the area and shape of the spraying area are determined, and a spraying path planning is generated; according to the spraying path planning, the surface of the abnormal position of the jacquard textile is sprayed; a spectrum acquisition device is used to perform a spectral scan on the textile surface after the spraying treatment to obtain spectral data; the spectral data is preprocessed, and the spectral curve is subjected to dimension reduction processing by a principal component analysis method; according to the spectral curve after the dimension reduction processing, characteristic bands are extracted to generate spectral feature vectors; a support vector machine algorithm is used to establish a corresponding relationship between the spectral feature vector and the abnormal state, and the accuracy of the abnormal state classification result is judged.

[0029] Specifically, according to the classification results of abnormal states of jacquard textiles, the corresponding reagent formula is selected from the preset chemical reagent database. Fluorescent brighteners are used for warp breakage, acid dyes are used for weft loss, and vat dyes are used for tissue structure deviation. Chemical reagents are prepared according to the formula ratio through precision mixing equipment, and the area and shape of the spraying area are calculated according to the coordinate information of the abnormal position to generate the spraying path planning. High-precision spraying equipment is used to adjust the nozzle position according to the spraying path planning. The initial spraying pressure is set to 0.5MPa, the spraying speed is set to 500mm / s, and the parameters are adjusted in real time through closed-loop control. Precise spraying is performed on the surface of the abnormal position of the jacquard textile, and the spraying area is controlled within 1.2 times the range of the abnormal area. After spraying is completed, the waiting time is set according to the characteristics of the chemical reagent, such as 30 seconds for fluorescent brighteners, 60 seconds for acid dyes, and 90 seconds for vat dyes. Start the spectrum acquisition equipment, set the spectrum data acquisition range to 300-800nm, the spectrum resolution to 0.5nm, and the acquisition time to 500ms. The surface of the textile after spraying was spectrally scanned, and the ambient temperature and humidity data were recorded at the same time. The spectrometer was calibrated with a standard white plate, and calibration was performed every 10 measurements to ensure the accuracy and repeatability of the data. The collected spectral data was preprocessed, including denoising, baseline correction and standardization. The principal component analysis method was used to reduce the dimension of the spectral curve, and the principal components with a cumulative contribution rate of more than 95% were selected. Combined with the textile material information and environmental parameters, characteristic bands were extracted, such as the characteristic peak of fluorescent brighteners near 450nm, acid dyes near 520nm, and vat dyes near 630nm. Spectral feature vectors were generated, and the support vector machine algorithm was used to establish the corresponding relationship between spectral features and abnormal states to verify the accuracy of the abnormal state classification results. In the production process of jacquard textiles, when the warp break abnormality was detected during real-time monitoring, the fluorescent brightener formula (concentration 5%, pH 8.5) was immediately selected from the chemical reagent database. The precision mixing equipment mixes 50ml of reagent according to the formula ratio, and calculates the elliptical spraying area of ​​10cm2 based on the abnormal coordinates (X: 150mm, Y: 300mm). The high-precision spraying equipment adjusts the nozzle position, and the initial spraying pressure is set to 0.5MPa and the speed is 500mm / s. The spraying effect is monitored in real time, and the pressure is fine-tuned to 0.48MPa and the speed is reduced to 480mm / s to ensure uniform coverage. After the spraying is completed, the timer is set for 30 seconds to wait for the chemical reaction. Subsequently, the spectrum acquisition equipment is started, and the scanning range of 300-800nm, 0.5nm resolution, and 500ms acquisition time are set. The ambient temperature is recorded as 25℃ and the relative humidity is 60%. After every 10 measurements, the spectrometer is calibrated with a standard white plate (reflectivity 99.9%). The collected spectral data is denoised by Savitzky-Golay filtering, and the baseline is corrected by polynomial fitting. The principal component analysis method reduces the spectral data to 5 principal components, with a cumulative contribution rate of 97.3%.Special attention was paid to the characteristic peak near 450nm, whose relative intensity was 0.85. The support vector machine algorithm was used to match the spectral feature with the warp break state, where the kernel function was RBF, C = 1.0, gamma = 0.1, and the confidence level reached 95.7%, verifying the accuracy of abnormal state classification. It only took 90 seconds from abnormal detection to result verification, achieving rapid response and accurate analysis of abnormal states of jacquard textiles.

[0030] Step S105 , extracting features and classifying the spectral data, and obtaining a comprehensive anomaly detection result by comparing the classification result with the tension position anomaly detection result.

[0031] Acquire spectral data and preprocess the spectral data, wherein the preprocessing includes denoising, baseline correction and standardization. Use a one-dimensional convolutional neural network to classify the preprocessed spectral data, wherein the one-dimensional convolutional neural network includes a convolution layer, a pooling layer and a fully connected layer. Construct a tension position abnormality feature vector according to the tension position abnormality detection result, and the tension value in the tension position abnormality feature vector is standardized to the interval [0, 1]. Use a gradient boosting decision tree algorithm to classify the tension position abnormality feature vector to obtain a tension position abnormality classification result. Combine the spectral classification result and the tension position abnormality classification result to construct a fusion feature vector.

[0032] Specifically, the collected spectral data were preprocessed, including denoising, baseline correction and standardization. The wavelet transform method was used to enhance the spectral features, and characteristic wavelength points were extracted from the range of 300-800nm. The correlation coefficient method was used to select the key characteristic wavelength, calculate the characteristic peak intensity and peak area, and generate the spectral feature vector. The spectral feature vector was subjected to principal component analysis, and the principal components with a cumulative contribution rate of more than 95% were selected to construct the feature space after dimensionality reduction. The spectral data were classified using a one-dimensional convolutional neural network, and the network structure contained 3 convolutional layers, 2 pooling layers and 2 fully connected layers. The convolution kernel size was set to 5, the step size was 1, and the pooling window size was 2. The ReLU activation function and the Softmax output layer were used, and the cross entropy loss function was used to optimize the network parameters through the back propagation algorithm to obtain the spectral classification results. The tension position anomaly detection results were obtained, including the tension anomaly threshold, position deviation, anomaly type and anomaly severity. The tension position anomaly feature vector was constructed, the tension value was standardized to the [0, 1] interval, the position deviation was normalized, and the anomaly type and severity were uniquely encoded. The gradient boosting decision tree algorithm was used to classify the tension position anomaly, and the number of trees was set to 100, the maximum depth was set to 5, and the learning rate was set to 0.1. The fusion feature vector was constructed by combining the spectral classification results and the tension position anomaly classification results. The output probability values ​​and confidence levels of the two classifiers were used as new features. The random forest algorithm was used for comprehensive classification, and the number of decision trees was set to 200, the maximum depth was set to 15, and the minimum number of leaf node samples was set to 5. The final comprehensive anomaly detection results, including anomaly type, anomaly severity, and comprehensive anomaly probability, were obtained by the average probability method. According to the detection results, the loom parameters, such as tension control, warp let-off speed, and fabric density, were automatically adjusted to form a closed-loop control. In the production process of jacquard textiles, the spectral acquisition equipment obtained the reflectance spectral data in the range of 300-800nm, with a sampling interval of 0.5nm. The Savitzky-Golay filter algorithm was applied to the raw data for denoising, with the window size set to 11 and the polynomial order set to 3. The progressive polynomial fitting method was used for baseline correction, with a fitting order of 5. The spectral features were enhanced by wavelet transform, and db4 wavelet was selected with 3 decomposition layers. The correlation coefficient method was used to select the key characteristic wavelengths, the threshold was set to 0.95, and 20 characteristic wavelength points were finally selected. The input of the one-dimensional convolutional neural network was the processed spectral data, which contained 3 convolutional layers with kernel sizes of 5, 3, and 3, 16, 32, and 64 channels, 2 maximum pooling layers with a window size of 2, and 2 fully connected layers with 128 and 64 nodes. The network was trained on 1000 samples for 50 rounds, with a learning rate of 0.001 and a batch size of 32, and finally achieved a classification accuracy of 97% on the test set.The data obtained by the tension position abnormality detection include warp tension (20-50N), weft tension (15-40N), warp position deviation (-2 to +2mm) and weft position deviation (-1.5 to.

[0033] +1.5mm). The gradient boosting decision tree algorithm uses these features for classification, trained on 5000 samples, and finally obtained a classification accuracy of 95%. The random forest algorithm integrates spectral and tension position features, trains 200 decision trees on 1000 samples, with a maximum depth of 15, and obtains the final anomaly detection results. The loom parameters are automatically adjusted according to the detection results. For example, when the warp tension is detected to be too high (>45N), the warp feeding speed is increased by 5%. It only takes 3 seconds from data collection to parameter adjustment, realizing real-time monitoring and optimization of the jacquard textile production process.

[0034] Step S106, accurately locate and mark the abnormal area according to the comprehensive abnormality detection result, cut the abnormal area sample and transmit it to the sample analysis device.

[0035] According to the abnormality detection result, the abnormal area is located by using the U-Net segmentation algorithm, which includes a downsampling layer and an upsampling layer, to obtain the center coordinates, area and shape parameters of the abnormal area, and generate a digital contour description of the abnormal area. If the abnormal area exists, the contour of the abnormal area is extracted by using the Canny edge detection algorithm; according to the digital contour description of the abnormal area, the cutting device is controlled to cut the sample of the abnormal area, and the cut sample is transported to the corresponding analysis device by a mechanical arm, and the analysis device includes a fiber fineness analyzer, a strength tester or a microscope.

[0036] Specifically, according to the comprehensive anomaly detection results, the anomaly type, severity and probability information are extracted, and the U-Net segmentation algorithm is used in combination with the fabric coordinates to accurately locate the abnormal area. The U-Net network structure contains 4 downsampling layers and 4 upsampling layers. The cross entropy loss function is used to train 50 rounds on 1000 annotated images, and the learning rate is set to 0.001. The center coordinates, area and shape parameters of the abnormal area are calculated to generate a digital contour description of the abnormal area. The abnormal area is imaged in real time using a high-resolution camera with an image resolution of 4096*3072 pixels. The Canny edge detection algorithm is used to extract the contour, and the threshold is set to a low threshold of 50 and a high threshold of 150. The laser marking method is used to mark the abnormal area, with a laser power range of 0.5-5W and a wavelength of 532nm. The width of the marking line is linearly related to the severity of the abnormality, ranging from 0.1 to 1mm. According to the digital contour description of the abnormal area, the high-precision cutting equipment is controlled to cut the sample of the abnormal area. The cutting equipment is driven by a servo motor, with a position accuracy of ±0.01mm. The cutting accuracy is dynamically adjusted according to the type of fabric, with cotton fabric set to ±0.5mm and silk fabric increased to ±0.2mm. The cutting speed is inversely proportional to the fabric thickness, ranging from 100 to 500mm / s. A six-axis robotic arm is used to grab the cut samples, and the grabbing force is controlled at 5-20N. The sample is fixed with a vacuum adsorption device, and the vacuum degree is -80kPa. The sample is placed on an anti-static conveyor belt with a conveyor belt speed range of 0.1-1m / s. The sample is identified by radio frequency identification technology, and the tag frequency is 915MHz. The sample is transported to the corresponding analysis device, such as a fiber fineness analyzer, a strength tester or a microscope system. At the entrance of the analysis device, a sample pretreatment unit is set, including dust removal, temperature and humidity adjustment, and flattening operations to ensure that the sample state meets the analysis requirements. On the jacquard fabric production line, an abnormality detection identified a warp break abnormality with an abnormality probability of 98.7% and a severity of 7.5 (out of 10). The U-Net segmentation algorithm locates the abnormal area within 0.3 seconds and outputs the coordinates of the abnormal center (X: 1250mm, Y: 750mm), with an area of ​​25 square centimeters and an irregular ellipse. The 4K high-resolution camera captures the image of the abnormal area, and the Canny algorithm extracts the edge, with a low threshold of 45 and a high threshold of 140. The 3W power and 532nm wavelength laser completes the marking of the abnormal area within 0.5 seconds, and the marking line width is 0.6mm. The high-precision cutting equipment sets the cutting accuracy to ±0.3mm. Considering that the sample is a cotton and nylon blended fabric, the cutting speed is adjusted to 300mm / s. The six-axis robotic arm grabs the cut sample with a force of 15N and fixes it with adsorption at a vacuum degree of -75kPa. The anti-static conveyor belt speed is set to 0.5m / s, and the sample is transported 50cm to the 915MHzRFID reader, giving the sample a unique ID: TF20240711-001.The sample then enters the fiber strength tester. Before the test, the dust removal device cleans the sample surface with a pressure of 0.3MPa, the constant temperature and humidity unit adjusts the environment to 20°C and 65% relative humidity, and the sample flattening device flattens the sample to a standard state with a force of 5N.

[0037] Step S107, performing microstructure analysis on the cut sample, extracting fiber arrangement characteristics, determining the degree of fiber arrangement abnormality, and obtaining a microstructure abnormality assessment result.

[0038] The image of the cropped sample is subjected to median filtering to obtain a first image after noise removal; an adaptive histogram equalization method is used for the first image to obtain a second image with enhanced contrast; a skeleton extraction algorithm is applied to the second image to extract fiber features and calculate fiber diameter, length and density parameters; a multidimensional feature vector is constructed based on the fiber features, and the multidimensional feature vector includes fiber diameter distribution, length distribution and density distribution; a dimensionality reduction process is performed on the multidimensional feature vector to obtain a reduced dimensionality feature vector; the reduced dimensionality feature vector is input into a pre-trained convolutional neural network, and a fiber arrangement abnormality degree score and an abnormality type probability distribution are obtained from the convolutional neural network.

[0039] Specifically, a high-resolution electron microscope was used to image the cut samples, and the magnification was set to 1000-3000 times. The sample surface was scanned at multiple points using a spiral scanning method. Each sample was scanned at 9 points with a point spacing of 100 microns. After obtaining the fiber microstructure image, the noise was removed by median filtering, and the image contrast was enhanced by adaptive histogram equalization to improve the image clarity. The skeleton extraction algorithm was used to identify and extract the fibers from the processed microstructure image, and the parameters such as fiber diameter, length and density were calculated. The fast marching method was used to analyze the fiber arrangement angle and intersection distribution, and the fiber surface morphology characteristics were extracted by combining morphological opening and closing operations. The fiber breakage was identified by edge detection and breakpoint connection algorithms, and the breakage rate and breakage morphological characteristics were calculated. Based on the extracted fiber features, a multidimensional feature vector was constructed, including indicators such as fiber diameter distribution, length distribution, density distribution, arrangement angle distribution, intersection density, surface roughness and breakage rate. The Z-score standardization method was used to process the feature data to eliminate dimensional differences. The principal component analysis was performed on the standardized feature vector, and the principal component with a cumulative contribution rate of more than 95% was selected. The reduced feature vector is input into a pre-trained convolutional neural network, which contains 4 convolutional layers, where the convolution kernel sizes are 3×3, 3×3, 5×5, and 5×5, 3 maximum pooling layers, where the pooling window size is 2×2, and 2 fully connected layers, where the number of nodes in the fully connected layers is 128 and 64, respectively. 1000 normal fiber arrangement samples and 5000 abnormal samples of different degrees are used for training, using the cross entropy loss function, a learning rate of 0.001, a batch size of 64, and 50 rounds of training. The network outputs the fiber arrangement abnormality score (0-10 points) and the probability distribution of the abnormal type. According to the preset abnormality threshold, where mild abnormality corresponds to 3-5 points, moderate abnormality corresponds to 5-7 points, and severe abnormality corresponds to 7-10 points, combined with the probability of abnormal type, the microstructure abnormality is comprehensively evaluated. On the jacquard fabric production line, a sample suspected of having abnormal fiber arrangement is sent to the microstructure analysis system. The sample was spirally scanned by a high-resolution electron microscope at a magnification of 2000 times, and a total of 9 points of images were collected, with a resolution of 4096×3072 pixels for each point. The image was denoised by median filtering with a window size of 7×7, and then enhanced using contrast-limited adaptive histogram equalization (CLAHE) with a limit threshold set to 0.01. The skeleton extraction algorithm was applied to the processed image, and 2763 fibers were extracted. The improved fast marching method was used to analyze the fiber arrangement, and the main arrangement angle was 43.2° with a standard deviation of 5.7°. The Canny algorithm was used for edge detection with a low threshold of 30 and a high threshold of 90, and 37 fiber breakage points were identified. The constructed feature vector contained 12 dimensions. After Z-score standardization, the principal component analysis retained the first 5 principal components, with a cumulative contribution rate of 96.8%.The convolutional neural network completed training within 2 hours, and the accuracy of the test set reached 97.3%. For this sample, the network output abnormality score was 6.8 points, which is a moderate abnormality. The highest probability abnormality type is "fiber breakage", with a probability of 0.72. The comprehensive evaluation results show that the sample has moderate fiber arrangement abnormality, mainly manifested as local fiber breakage and arrangement angle deviation. It is recommended to adjust the weaving tension and combing process.

[0040] It will be apparent to those skilled in the art that the present application is not limited to the details of the exemplary embodiments described above, and that the present application can be implemented in other specific forms without departing from the spirit or essential features of the present application. Therefore, the embodiments should be considered exemplary and non-limiting in all respects, and the scope of the present application is defined by the appended claims rather than the above description, and it is intended that all changes falling within the meaning and scope of the equivalent elements of the claims be included in the present application. Any reference numeral in a claim should not be considered as limiting the claim to which it relates.

Claims

1. A method for detecting abnormalities in textile weaving structures, characterized in that: The method comprises: Acquire real-time tension monitoring data of the warp and weft during the weaving process of the jacquard textile, and establish a tension distribution model under normal weaving conditions according to the real-time tension monitoring data; Collecting real-time images of the jacquard textile during weaving, accurately locating the positions of the warp and weft lines according to the real-time images, and obtaining the deviation between the actual positions of the warp and weft lines and the theoretical positions; Classifying the abnormal state of the jacquard weaving structure according to the deviation between the tension distribution model and the real-time tension monitoring data, and the deviation between the actual position of the warp and weft lines and the theoretical position; If the classification result is an abnormal state, spraying a pre-prepared chemical reagent on the surface of the abnormal position of the jacquard textile, and collecting spectral data of the textile surface after spraying; Extracting features and classifying the spectral data, and obtaining a comprehensive anomaly detection result by comparing the classification result with the tension position anomaly detection result; According to the comprehensive abnormality detection result, the abnormal area is accurately located and marked, and the sample of the abnormal area is cut and transported to the sample analysis device; The cut samples are subjected to microstructure analysis to extract fiber arrangement characteristics, determine the degree of fiber arrangement abnormality, and obtain microstructure abnormality assessment results.

2. The method according to claim 1, wherein: The method of acquiring real-time tension monitoring data of the warp and weft during the weaving process of the jacquard textile and establishing a tension distribution model under a normal weaving state according to the real-time tension monitoring data comprises: Acquire data collected by warp and weft tension sensors during the weaving process of jacquard textiles, and pre-process the data, including removing outliers and smoothing; Calculate the mean and standard deviation of the tension data according to the pre-processed data, and determine the tension value range of the normal weaving state; The polynomial regression method is used to fit the tension curve function and the relationship expression between tension and fabric parameters is obtained; Analyze the time series of the tension data to predict the real-time tension values ​​of the warp and weft threads at different loom speeds; Compare the actual collected tension value with the predicted value and calculate the tension deviation.

3. The method according to claim 1, wherein: The method of collecting real-time images during the weaving process of the jacquard textile, accurately positioning the positions of the warp and weft lines according to the real-time images, and obtaining the deviation between the actual positions of the warp and weft lines and the theoretical positions includes: Acquire a real-time image of the jacquard textile during the weaving process, and process the real-time image using a multi-scale image analysis method to obtain an enhanced image; Performing edge detection on the enhanced image, extracting the contours of the longitude and latitude lines, and gradually fitting the contours to the actual shapes of the longitude and latitude lines through an iterative optimization method; Based on the identified longitude and latitude position information, the support vector machine algorithm is used to locate the intersection of the longitude and latitude lines. The support vector machine model is trained to identify the intersection by extracting local image features as input; Fit the actual direction of longitude and latitude lines to ensure the continuity and smoothness of the curve; Calculate the theoretical position of the longitude and latitude lines, compare the actual position of the longitude and latitude lines after fitting with the theoretical position, and calculate the Euclidean distance to obtain the position deviation; If the position deviation exceeds a preset dynamic threshold, it is determined that there is an abnormality in the position of the longitude and latitude lines.

4. The method according to claim 1, wherein: The abnormal state of the jacquard weaving structure is classified according to the deviation between the tension distribution model and the real-time tension monitoring data, and the deviation between the actual position of the warp and weft lines and the theoretical position, including: Obtain the tension deviation value between the tension distribution model and the real-time tension monitoring data, and obtain the position deviation value between the actual position of the longitude and latitude lines and the theoretical position; Establishing a multidimensional feature vector according to the tension deviation value, the position deviation value, the yarn type, the loom speed, the weft beating frequency, the warp and weft tensions, and the fabric width; The multidimensional feature vector is normalized by using a Z-score normalization method to obtain a normalized feature vector; Performing principal component analysis on the standardized feature vector, selecting principal components whose cumulative contribution rate exceeds a preset threshold, and constructing a reduced-dimensional feature space; The Mahalanobis distance is used to calculate the distance value between the current state and the preset normal state sample set, and the abnormality degree is divided according to the distance value; Preliminary classification of the degree of anomaly and the distribution of position deviations of intersection points of longitude and latitude lines; Based on the preliminary classification result, a refined classification is performed in combination with a pre-established abnormal state knowledge base to obtain a final classification result of the abnormal state of the jacquard weaving structure.

5. The method according to claim 1, wherein: If the classification result is an abnormal state, a pre-prepared chemical reagent is sprayed on the surface of the abnormal position of the jacquard textile, and spectral data of the textile surface after spraying is collected, including: Obtain the classification result of the abnormal state of the jacquard textile and select the corresponding reagent formula from the preset chemical reagent database; Prepare chemical reagents according to the reagent formula, wherein the chemical reagents are used for spraying treatment; Obtain the coordinate information of the abnormal position, determine the area and shape of the spraying area, and generate the spraying path planning; According to the spraying path planning, spraying treatment is performed on the surface of the abnormal position of the jacquard textile; Use spectrum acquisition equipment to perform spectrum scanning on the surface of the textile after spraying to obtain spectrum data; Preprocessing the spectral data, and performing dimension reduction processing on the spectral curve using a principal component analysis method; According to the spectrum curve after dimension reduction processing, characteristic bands are extracted to generate spectrum feature vectors; A support vector machine algorithm is used to establish a corresponding relationship between the spectral feature vector and the abnormal state, and to determine the accuracy of the abnormal state classification result.

6. The method according to claim 1, wherein: The feature extraction and classification of the spectral data, and obtaining a comprehensive anomaly detection result by comparing the classification result with the tension position anomaly detection result, include: Acquiring spectral data, and preprocessing the spectral data, wherein the preprocessing includes denoising, baseline correction and standardization; A one-dimensional convolutional neural network is used to classify the preprocessed spectral data, wherein the one-dimensional convolutional neural network comprises a convolutional layer, a pooling layer and a fully connected layer; Constructing a tension position abnormality feature vector according to the tension position abnormality detection result, wherein the tension value in the tension position abnormality feature vector is standardized to the interval [0, 1]; Using a gradient boosting decision tree algorithm to classify the tension position abnormality feature vector, to obtain a tension position abnormality classification result; The spectral classification result and the tension position abnormality classification result are combined to construct a fusion feature vector.

7. The method according to claim 1, wherein: The method of accurately locating and marking the abnormal area according to the comprehensive abnormality detection result, cutting the abnormal area sample and transmitting it to the sample analysis device comprises: According to the abnormality detection result, the abnormal area is located using a U-Net segmentation algorithm, wherein the U-Net segmentation algorithm includes a downsampling layer and an upsampling layer, obtains the center coordinates, area and shape parameters of the abnormal area, and generates a digital contour description of the abnormal area; If the abnormal area exists, the Canny edge detection algorithm is used to extract the contour of the abnormal area; According to the digital contour description of the abnormal area, the cutting equipment is controlled to cut samples from the abnormal area, and the cut samples are transported to the corresponding analysis device through a robotic arm. The analysis device includes a fiber fineness analyzer and a strength tester or a microscope.

8. The method according to claim 1, wherein: The microstructure analysis of the cut sample is performed to extract the fiber arrangement characteristics, determine the degree of fiber arrangement abnormality, and obtain the microstructure abnormality evaluation result, including: Performing median filtering on the image of the cropped sample to obtain a first image after noise removal; Adopting an adaptive histogram equalization method on the first image to obtain a second image with enhanced contrast; Applying a skeleton extraction algorithm to the second image to extract fiber features and calculate fiber diameter, length and density parameters; Constructing a multidimensional feature vector based on the fiber characteristics, wherein the multidimensional feature vector includes fiber diameter distribution, length distribution and density distribution; Performing dimensionality reduction processing on the multidimensional feature vector to obtain a reduced-dimensional feature vector; The dimension-reduced feature vector is input into a pre-trained convolutional neural network, and the fiber arrangement abnormality degree score and abnormality type probability distribution are obtained from the convolutional neural network.

Images