A method and system for monitoring the production quality of fiber fabrics

Abstract

The present invention relates to the field of artificial intelligence technology, and specifically to a method and system for monitoring the production quality of fiber fabrics. The present invention divides and numbers the production process, obtains the corresponding production machine data based on each process number and stores it, and labels the numbers of warm-keeping artificial fiber fabrics during the production process; obtains the quality inspection standard data, the artificial fiber fabrics that pass the quality inspection and those that do not pass the quality inspection according to the quality inspection content, and obtains the first image data and the second image data, obtains the first fiber feature and the second fiber feature through image feature analysis and conducts feature comparison to obtain the third fiber feature; analyzes the third fiber feature to obtain the problematic link, calls the production machine data of the process number through the number of the warm-keeping artificial fiber fabric, obtains the abnormal cause of the warm-keeping artificial fiber fabric that does not pass the quality inspection through time series anomaly analysis, and optimizes the process or hardware to improve the quality inspection pass rate.

Description

Technical Field

[0001] The present invention relates to the technical field of artificial intelligence, and specifically to a method and system for monitoring the production quality of fiber fabrics. Background Art

[0002] Thermal insulation fibers generally refer to fiber materials used to improve the warmth retention of clothing and home textile products. In addition to good warmth retention performance, these fiber materials usually also have characteristics such as light weight, softness, and breathability; thermal insulation fibers are generally divided into natural fiber materials and synthetic fiber materials; natural fiber materials such as cotton, goose down, or duck down, etc. However, these natural fibers have a certain acquisition cycle, and due to market demand problems, some natural thermal insulation fibers are in short supply and have high costs, resulting in a shortage of supply of natural thermal insulation fibers.

[0003] The emergence of artificial thermal insulation fibers has, to a certain extent, solved the supply problem. Artificial thermal insulation fibers are mainly based on natural polymer materials and are achieved through methods such as chemical treatment and mechanical processing, such as polyester fibers, ultra-fine synthetic fibers, or composite fibers, etc.; in addition to warmth retention performance, these artificial thermal insulation fibers, like polyester fibers, have the characteristics of high strength and wear resistance, ultra-fine synthetic fibers combine the softness and good texture of high-quality natural thermal insulation fibers, and composite fibers can combine the advantages of multiple fibers according to the various fiber types used.

[0004] Before these artificial thermal insulation fibers are made into clothing, they need to be made into corresponding fiber fabrics, and the production process of these fiber fabrics includes raw material preparation, spinning, weaving, dyeing, and fabric treatment; since artificial thermal insulation fibers do not have the inherent warmth retention advantage like natural fibers, after the production of thermal insulation fiber fabrics is completed, in order to make the clothing have corresponding warmth retention performance and other performances after production, product quality inspection is required for the produced fiber fabrics. In order to improve the passing rate of quality inspection, CN202210583793.9 proposes a quality monitoring system for the production of home textile fiber fabrics. By dividing the fabric into multiple regions, the quality, fabric density, and light transmittance of the fabric are detected to obtain the best fabric, and then based on linear fitting analysis, it is judged whether the fabric quality is qualified; CN202110283642.7 proposes a detection method and quality standard for thermal insulation bamboo fiber fabrics. By dividing into multiple grades, and using softness, deformability, and heat preservation as evaluation criteria, different grade quality inspection standards are adopted according to different grades for quality inspection; CN202210260738.6 proposes a wear resistance detection device for composite fiber fabrics. This device fixes the produced fiber fabric through a clamping groove, the telescopic rod pushes the clamping plate to clamp the fiber fabric, and then drives the slider to move on the slide rail to quickly detect the elastic coefficient of different fiber fabrics.

[0005] However, the current quality inspection method cannot analyze the reasons for traceability after quality inspection to optimize the production process of fiber fabrics and improve the passing rate of subsequent fiber fabric quality inspection. At present, there is little research on the attribution analysis of fiber fabric quality inspection based on artificial thermal insulation fibers.

[0006] In order to optimize the production process of fiber fabrics and improve the passing rate of product quality inspection, a fiber fabric production quality monitoring method and system are proposed. Summary of the Invention

[0007] The purpose of the present invention is to provide a fiber fabric production quality monitoring method and system. By dividing and numbering the production process, obtaining the corresponding production machine data based on each process number and storing it, and labeling the numbers of the thermal insulation artificial fiber fabrics during the production process; obtaining the quality inspection standard data, the artificial fiber fabrics that pass the quality inspection and those that do not pass the quality inspection according to the quality inspection content, and obtaining the first image data and the second image data, obtaining the first fiber characteristics and the second fiber characteristics through image feature analysis and comparing the characteristics to obtain the third fiber characteristics; analyzing the problem links based on the third fiber characteristics, calling the production machine data of the process number through the number of the thermal insulation artificial fiber fabric, obtaining the abnormal reasons for the artificial fiber fabrics that do not pass the quality inspection through time series anomaly analysis, and optimizing the process or hardware to improve the passing rate of quality inspection.

[0008] To achieve the above object, the present invention provides the following technical solutions:

[0009] A fiber fabric production quality monitoring method, comprising:

[0010] Obtain the production process of the thermal insulation artificial fiber fabric, divide it according to the production process, and generate a process number;

[0011] Obtain the production machine data, label the corresponding numbers for the production machine data according to the process number and store it; the thermal insulation artificial fiber fabric is produced and quality inspected according to the production process; during the production process, the artificial fiber fabric is numbered and labeled according to the number;

[0012] The production machine data is collected at a fixed time interval T, and the production machine data of different process numbers uses the same time interval T;

[0013] The production machine data is time series data including voltage data collected by each machine at time interval T, displacement data of each component of the production machine, dyeing temperature data, and dye color data;

[0014] Obtain the thermal insulation artificial fiber fabrics that pass the quality inspection, and obtain the quality inspection standard data of the thermal insulation artificial fiber fabrics according to the quality inspection content;

[0015] Obtain the first image data of the warm artificial fiber fabric, segment the first image data to generate multiple first warm fabric sub-images; obtain the first fiber characteristics based on the first warm fabric sub-images and the quality inspection standard data;

[0016] Further, the segmentation method for the first image data and the second image data is to obtain the number of length pixels and the number of width pixels of the first image data or the second image data, obtain the first segmentation common divisor based on the number of length pixels, and obtain the second segmentation common divisor based on the number of width pixels; screen and exclude the first segmentation common divisor and the second segmentation common divisor that are less than the segmentation threshold;

[0017] If there is the same data in the first segmentation common divisor and the second segmentation common divisor after exclusion, perform segmentation based on the same data; if there is no same data, then judge the data similarity of the first segmentation common divisor and the second segmentation common divisor, and select the first segmentation common divisor and the second segmentation common divisor with the greatest data similarity for segmentation;

[0018] Obtain the warm artificial fiber fabric that fails the quality inspection to obtain the second image data, segment the second image data to generate multiple second warm fabric sub-images; obtain the second fiber characteristics based on the second warm fabric sub-images and the quality inspection standard data;

[0019] Perform feature comparison calculation based on the first fiber characteristics and the second fiber characteristics to obtain the third fiber characteristics;

[0020] Further, the first fiber characteristics and the second fiber characteristics are identified by a warm fiber characteristics recognition model;

[0021] The warm fiber characteristics recognition model includes a warm fiber image input unit, a warm fiber characteristics classification unit, a warm fiber impurity comparison unit, a warm fiber pattern comparison unit, a warm fiber warmth retention comparison unit, and a warm fiber abnormality output unit;

[0022] The warm fiber image input unit preprocesses the first image data or the second image data to generate warm fiber input data; the warm fiber characteristics classification unit obtains the warm fiber input data, identifies and classifies the corresponding image characteristics based on the quality inspection standard data, and generates the first fiber characteristics or the second fiber characteristics;

[0023] The warm fiber characteristics classification unit adopts three groups of characteristics recognition models. The first group of the characteristics recognition model is an impurity recognition model for obtaining impurity characteristics; the second group of the characteristics recognition model is a pattern characteristics recognition model for obtaining pattern characteristics; the third group of the characteristics recognition model is a warmth retention characteristics model for obtaining the void characteristics of the fiber fabric;

[0024] The first group of the feature recognition models adopts a convolutional layer with 2 layers of 3×3, a max pooling layer with 1 layer, and a convolutional layer with 1 layer of 1×1; the second group of the feature recognition models adopts a convolutional layer with 1 layer of 5×5, an average pooling layer with 1 layer, convolutional layers with 2 layers of 3×3, an average pooling layer with 1 layer, and a convolutional layer with 1 layer of 1×1; the third group of the feature recognition models adopts a convolutional layer with 1 layer of 3×3, an average pooling layer with 1 layer, a convolutional layer with 1 layer of 1×1, and an Attention mechanism;

[0025] The thermal insulation fiber impurity comparison unit extracts the impurity features of the first fiber feature and the second fiber feature for identification and comparison, and generates impurity anomaly data; the comparison calculation of the impurity features is as follows:

[0026]

[0027] Wherein, ICP is the impurity feature comparison result, IC2 is the impurity pixel size of the second fiber feature, IC1 is the impurity pixel size of the first fiber feature, ICN2 is the impurity pixel number of the second fiber feature, and ICN1 is the impurity pixel number of the first fiber feature;

[0028] The thermal insulation fiber pattern comparison unit extracts the pattern features of the first fiber feature and the second fiber feature for identification and comparison, and generates pattern anomaly data; the comparison calculation of the pattern features is as follows:

[0029]

[0030] Wherein, PS is the pattern similarity, PT2 is the pattern feature of the second fiber feature, PT1 is the pattern feature of the first fiber feature, CS is the pattern similarity, CT2 is the pattern color feature of the second fiber feature, and CT1 is the pattern color feature of the first fiber feature;

[0031] The thermal insulation fiber heat preservation comparison unit extracts the heat preservation features of the first fiber feature and the second fiber feature for identification and comparison, and generates heat preservation anomaly data; the comparison calculation of the heat preservation features is as follows:

[0032]

[0033] ω1 + ω2 + ω3 + ω4 = 1;

[0034] Wherein, WP is the comparison calculation result of the heat preservation feature, WN2 is the number of pores between fibers of the second fiber feature, WN1 is the number of pores between fibers of the first fiber feature, WKN2 is the pore pixel size between fibers of the second fiber feature, WKN1 is the pore pixel size between fibers of the first fiber feature, WKN2j The j-th pore pixel size between fibers of the fibers for the second fiber feature, WKN 1j The j-th pore pixel size between fibers of the fibers for the first fiber feature, ω1, ω2, ω3, ω4 are their respective calculation weights;

[0035] The thermal insulation fiber anomaly output unit outputs the impurity anomaly data, the pattern anomaly data, and the thermal insulation anomaly data;

[0036] The third fiber feature is the impurity anomaly data, the pattern anomaly data, and the thermal insulation anomaly data;

[0037] Conduct attribution analysis based on the third fiber feature to obtain the production process with problems, and call the production machine data corresponding to the process number according to the production process and the artificial fiber fabric number and analyze the anomaly data;

[0038] Further, calculate the probability of problems occurring in each link of the production process of the thermal insulation artificial fiber fabric that fails quality inspection according to the impurity anomaly data, the pattern anomaly data, and the thermal insulation anomaly data. The probability calculation is as follows:

[0039]

[0040] Among them, P is the probability, P i Is the problem probability of the i-th link, max() is the maximum value, Is the k-th anomaly data of the i-th link, Is the k-th anomaly reference value of the i-th link;

[0041] The anomaly data analysis is obtained based on the thermal insulation fiber fabric anomaly machine analysis model;

[0042] The fiber fabric anomaly machine analysis model includes a machine data input unit, a machine anomaly feature extraction unit, a machine anomaly feature classification unit, and a machine anomaly feature output unit;

[0043] The machine data input unit preprocesses the production machine data to generate machine anomaly input data;

[0044] The machine anomaly feature extraction unit extracts the temporal anomaly features of the machine anomaly input data to generate machine anomaly feature data; the machine anomaly feature classification unit classifies the machine anomaly feature data to generate machine anomalies features; the machine anomaly feature output unit outputs the anomaly labels of the machine anomaly features;

[0045] The abnormal machine analysis model of the thermal insulation fiber fabric uses 4 one-dimensional convolutional layers and 4 LSTM units to obtain abnormal data. The 4 one-dimensional neural networks include 2 convolutional layers of 1*5 and 2 convolutional layers of 1*3. The 4 one-dimensional convolutional layers are directly connected to the 4 LSTM units, and classification output is performed in the Softmax layer;

[0046] Optimize the production process according to the abnormal data;

[0047] The present invention also proposes a production quality monitoring system for fiber fabrics, including a process number division module, a production machine data acquisition module, a thermal insulation artificial fiber fabric number marking module, a quality inspection marked data acquisition module, a first fiber feature and a second fiber feature acquisition module, a third fiber feature acquisition module, an abnormal data analysis module and a process optimization module, specifically:

[0048] The process number division module obtains the production process of the thermal insulation artificial fiber fabric and divides it to generate a process number;

[0049] The production machine data acquisition module marks the numbers of the production machine data according to the process number and stores them;

[0050] The thermal insulation artificial fiber fabric number marking module produces and performs quality inspection on the thermal insulation artificial fiber fabric according to the production process; during the production process, the artificial fiber fabric is numbered and marked according to the number;

[0051] The quality inspection marked data acquisition module obtains the thermal insulation artificial fiber fabric that has passed the quality inspection and obtains the quality inspection standard data of the thermal insulation artificial fiber fabric according to the quality inspection content;

[0052] The first fiber feature and the second fiber feature acquisition module obtains the first image data of the thermal insulation artificial fiber fabric, segments the first image data to generate a plurality of first thermal insulation fabric sub-images; obtains the first fiber feature according to the first thermal insulation fabric sub-images and the quality inspection standard data; obtains the thermal insulation artificial fiber fabric that has not passed the quality inspection and obtains the second image data, segments the second image data according to the second image data to generate a plurality of second thermal insulation fabric sub-images; obtains the second fiber feature according to the second thermal insulation fabric sub-images and the quality inspection standard data;

[0053] The third fiber feature acquisition module performs feature comparison calculation according to the first fiber feature and the second fiber feature to obtain the third fiber feature;

[0054] Further, the attribution analysis of the third fiber feature is that the third fiber feature is impurity abnormal data, pattern abnormal data and thermal insulation abnormal data obtained by the first fiber feature and the second fiber feature acquisition module;

[0055] Calculate the probabilities of problems occurring in each link of the production process of the unqualified warm artificial fiber fabric based on the impurity abnormal data, pattern abnormal data, and warmth retention abnormal data. The probability calculation is as follows:

[0056]

[0057] where P is the probability, P i is the problem probability of the i-th link, max() is the maximum value, is the k-th abnormal data of the i-th link, is the k-th abnormal reference value of the i-th link;

[0058] The abnormal data analysis module performs attribution analysis based on the third fiber characteristic to obtain the production process with problems, and calls the production machine data corresponding to the process number according to the production process and the artificial fiber fabric number and analyzes the abnormal data;

[0059] Furthermore, the abnormal data analysis is obtained based on the abnormal machine analysis model of the warm fiber fabric;

[0060] The abnormal machine analysis model of the fiber fabric includes a machine data input unit, a machine abnormal feature extraction unit, a machine abnormal feature classification unit, and a machine abnormal feature output unit;

[0061] The machine data input unit preprocesses the production machine data to generate machine abnormal input data;

[0062] The machine abnormal feature extraction unit extracts the time-series abnormal features of the machine abnormal input data to generate machine abnormal feature data; the machine abnormal feature classification unit classifies the machine abnormal feature data to generate machine abnormal features; the machine abnormal feature output unit outputs the abnormal labels of the machine abnormal features;

[0063] The process optimization module optimizes the production process according to the abnormal data.

[0064] Compared with the prior art, the beneficial effects of the present invention are:

[0065] 1. In order to clarify the reasons for the production of unqualified warm artificial fiber fabrics, the characteristic data of qualified and unqualified warm artificial fiber fabrics are obtained through an image detection method. Based on the characteristic comparison, the characteristic data of the unqualified warm artificial fiber fabrics are obtained. Based on the characteristic data, the main reasons for the unqualified can be quickly found, and subsequent attribution analysis can be carried out based on the main reasons, providing a solid data basis for subsequent attribution analysis and the improvement of the quality inspection pass rate.

[0066] 2. Based on the first fiber features that pass quality inspection and the second fiber features that fail quality inspection through image recognition, through feature comparison, the feature data of the anomalies caused by the failed quality inspection can be accurately obtained. By calculating the attribution analysis probability based on this feature data, the link that causes the anomaly can be quickly and accurately obtained, and the stored production machine data can be quickly called through the numbered data for subsequent production line anomaly analysis, so as to find the problematic part of the production line for optimization and improve the quality inspection pass rate.

[0067] 3. In order to analyze the reasons for the failed quality inspection in the production process, the production process data with problems is quickly called through the number of the warm artificial fiber fabric. Through the abnormal time series analysis of the warm fiber fabric abnormal machine analysis model, the real reason for the abnormal data is obtained, and the corresponding results are output so that the relevant personnel can quickly adopt the corresponding solutions for optimization to avoid the occurrence of the same problems in the future and improve the pass rate of quality inspection. Brief Description of the Drawings

[0068] Figure 1 It is a schematic diagram of the method flow of the present invention;

[0069] Figure 2 It is a diagram of the fiber production workshop of the present invention;

[0070] Figure 3 It is a schematic diagram of the process of the warm fiber feature recognition model of the present invention;

[0071] Figure 4 It is a schematic diagram of the process of the warm fiber fabric abnormal machine analysis model of the present invention;

[0072] Figure 5 It is a schematic diagram of the system process of the present invention. Detailed Embodiment

[0073] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0074] In thermal insulation fibers, due to the fact that natural fibers have a certain acquisition cycle and some fibers face the problems of supply falling short of demand and high costs caused by market demand; therefore, artificial thermal insulation fibers have, to a certain extent, solved the supply and demand problems of natural fibers. Before artificial thermal insulation fibers are made into products, they need to be processed into corresponding fabrics. Since they do not have the thermal insulation advantages of natural fibers, it is necessary to conduct quality inspections on the fabrics. However, the current quality inspection methods cannot be traced back to optimize the production process and improve the quality inspection passing rate. For this reason, the present invention provides a method for monitoring the production quality of fiber fabrics, and the technical solution is as follows:

[0075] Refer to Figure 1 As shown, obtain the production process of the thermal insulation artificial fiber fabric, divide it according to the production process, and generate a process number;

[0076] Obtain the production machine data, label the corresponding numbers for the production machine data according to the process number, and store them; the thermal insulation artificial fiber fabric is produced and quality inspected according to the production process; the number of the thermal insulation artificial fiber fabric is labeled during the production process;

[0077] Obtain the thermal insulation artificial fiber fabric that has passed the quality inspection, and obtain the quality inspection standard data of the thermal insulation artificial fiber fabric according to the quality inspection content;

[0078] Obtain the first image data of the thermal insulation artificial fiber fabric, segment the first image data to generate multiple first thermal insulation fabric sub-images; obtain the first fiber characteristics according to the first thermal insulation fabric sub-images and the quality inspection standard data;

[0079] Obtain the thermal insulation artificial fiber fabric that has not passed the quality inspection and obtain the second image data, segment the second image data according to the second image data to generate multiple second thermal insulation fabric sub-images; obtain the second fiber characteristics according to the second thermal insulation fabric sub-images and the quality inspection standard data;

[0080] Perform feature comparison calculation according to the first fiber characteristics and the second fiber characteristics to obtain the third fiber characteristics;

[0081] Conduct attribution analysis according to the third fiber characteristics to obtain the production process with problems, call the production machine data corresponding to the process number according to the production process and the number of the thermal insulation artificial fiber fabric, and analyze the abnormal data;

[0082] Optimize the production process according to the abnormal data.

[0083] In order to improve the product quality inspection rate, conduct attribution analysis on the thermal insulation fiber fabric that has not passed the quality inspection to improve the production process; the present invention conducts attribution analysis from three aspects in total;

[0084] First, during the production process, all data of the production machines are collected. To facilitate subsequent attribution analysis results, all data are numbered according to the process, and stored according to the corresponding numbers for subsequent calling and analysis. At the same time, the artificial fiber fabrics are numbered during the production process to facilitate quickly finding the problematic data stored.

[0085] Secondly, after passing the quality inspection, images of the warm-keeping fiber fabrics that have passed the quality inspection and those that have not are obtained. Corresponding data features are obtained based on the image data, and the features of the warm-keeping fiber fabrics that have passed the quality inspection are compared with those that have not passed the quality inspection to obtain different features of the warm-keeping fiber fabrics that have not passed the quality inspection. Attribution analysis is performed on these different features to obtain the corresponding processes where problems occur.

[0086] Finally, traceability is carried out based on the results of the attribution analysis and the numbers of the warm-keeping fiber fabrics to obtain the corresponding stored data, and this section of data is analyzed to obtain the abnormal data features and categories. Based on this abnormal feature, relevant personnel are required to make corresponding optimizations and adjustments to the production process.

[0087] Through the above method, without affecting the process, the specific reasons for the warm-keeping fiber fabrics that have not passed the quality inspection can be obtained. Automatic traceability and analysis are carried out based on this reason to obtain the reasons for abnormal problems, so as to achieve the effect of improving the quality inspection pass rate and the quality of the product line.

[0088] Example 1

[0089] For specific illustration, the present invention is elaborated from the following examples:

[0090] Obtain the production process of the warm-keeping artificial fiber fabrics, divide according to the production process, and generate process numbers.

[0091] Obtain the production machine data, label the corresponding numbers to the production machine data according to the process numbers and store them; the warm-keeping artificial fiber fabrics are produced and quality-inspected according to the production process; during the production process, the artificial fiber fabrics are numbered and labeled according to the numbers; the number labeling can be carried out on the fabrics or by electronic labeling; the production workshop is as shown in Figure 2 shown.

[0092] The production machine data is collected at fixed time intervals T (for example, but not limited to 1 second, 2 seconds, 3 seconds). The production machine data with different process numbers uses the same time interval T; in order to improve the accuracy of subsequent model analysis of abnormal data and facilitate improving the quality inspection efficiency of the warm-keeping fiber fabrics, the data is collected at the same interval during data acquisition to ensure that the collection data frequencies of each process are exactly the same, reduce the parameter operation complexity of the model, improve the recognition accuracy of the model, and ensure the accuracy of abnormal problem discovery and the quality inspection rate.

[0093] The production machine data is time-series data, including voltage data collected by each machine at time interval T, displacement data of each component of the production machine, dyeing temperature data, and dye color data;

[0094] Obtain the warm artificial fiber fabric that has passed quality inspection, and obtain the quality inspection standard data of the warm artificial fiber fabric according to the quality inspection content;

[0095] Obtain the first image data of the warm artificial fiber fabric, segment the first image data to generate multiple first warm fabric sub-images; obtain the first fiber features according to the first warm fabric sub-images and the quality inspection standard data; the first image data is an image of the warm artificial fiber fabric that has passed quality inspection;

[0096] Further, the segmentation method for the first image data and the second image data is to obtain the number of length pixels and the number of width pixels of the first image data or the second image data, obtain the first segmentation common divisor according to the number of length pixels, and obtain the second segmentation common divisor according to the number of width pixels; screen and exclude the first segmentation common divisor and the second segmentation common divisor that are less than the segmentation threshold; the segmentation threshold is the pixel size, which is defaulted to 332*332, and can be modified according to the actual situation in the specific implementation process;

[0097] If there are identical data in the first segmentation common divisor and the second segmentation common divisor after exclusion, then segment according to the identical data; if there is no identical data, then judge the data similarity of the first segmentation common divisor and the second segmentation common divisor, and select the first segmentation common divisor and the second segmentation common divisor with the greatest data similarity for segmentation; for example, the first segmentation common divisors are 200, 300, 350, 400, and the second segmentation common divisors are 250, 340, 380, 450. Judge according to the difference of the data. The difference between 350 of the first segmentation common divisor and 340 of the second segmentation common divisor is 10, which is the minimum of all differences. Therefore, it is judged that the data similarity is the highest. Therefore, select 350*340 as the standard for segmenting the data;

[0098] Since the image data of the warm fiber fabric may have the situation of inconsistent image sizes during the input process, in the specific implementation process, the shooting positions of the cameras are taken at the same height to obtain the first image data and the second image data. At the same time, to ensure the accuracy of the output feature data, a unified segmentation standard is adopted for the image data, so that the first image data and the second image data can be analyzed using unified image data, ensuring the unity of data dimensions, improving the recognition accuracy of the model, and ensuring the accuracy of quality inspection problem analysis;

[0099] Obtain the warm artificial fiber fabric that fails the quality inspection and obtain the second image data. Segment it based on the second image data to generate multiple second warm fabric sub-images. Obtain the second fiber characteristics based on the second warm fabric sub-images and the quality inspection standard data.

[0100] Perform feature comparison calculation based on the first fiber characteristics and the second fiber characteristics to obtain the third fiber characteristics.

[0101] The first fiber characteristics are the fiber characteristics of the fabric that passes the quality inspection, the second fiber characteristics are the fiber characteristics of the fabric that fails the quality inspection, and the third fiber characteristics are the fiber characteristics after comparison. Fiber characteristics include impurity characteristics, pattern characteristics, and void characteristics.

[0102] Furthermore, the first fiber characteristics and the second fiber characteristics are identified by a warm fiber characteristics recognition model.

[0103] Refer to Figure 3 As shown, the warm fiber characteristics recognition model includes a warm fiber image input unit, a warm fiber characteristics classification unit, a warm fiber impurity comparison unit, a warm fiber pattern comparison unit, a warm fiber warmth retention comparison unit, and a warm fiber anomaly output unit.

[0104] The warm fiber image input unit preprocesses the first image data or the second image data to generate warm fiber input data. The warm fiber characteristics classification unit obtains the warm fiber input data, identifies and classifies the corresponding image characteristics based on the quality inspection standard data, and generates the first fiber characteristics or the second fiber characteristics.

[0105] The warm fiber characteristics classification unit uses three groups of characteristics recognition models. The first group of characteristics recognition models is an impurity recognition model for obtaining impurity characteristics. The second group of characteristics recognition models is a pattern characteristics recognition model for obtaining pattern characteristics. The third group of characteristics recognition models is a warmth retention characteristics model for obtaining the void characteristics of the fiber fabric. The first group of characteristics recognition models uses 2 layers of 3*3 convolutional layers, 1 layer of max pooling layer, and 1 layer of 1*1 convolutional layer. The second group of characteristics recognition models uses 1 layer of 5*5 convolutional layer, 1 layer of average pooling layer, 2 layers of 3*3 convolutional layers, 1 layer of average pooling layer, and 1 layer of 1*1 convolutional layer. The third group of characteristics recognition models uses 1 layer of 3*3 convolutional layer, 1 layer of average pooling layer, 1 layer of 1*1 convolutional layer, and an Attention mechanism.

[0106] The warm fiber impurity comparison unit extracts and identifies and compares the impurity characteristics of the first fiber characteristics and the second fiber characteristics to generate impurity anomaly data. The comparison calculation of the impurity characteristics is as follows:

[0107]

[0108] Among them, ICP is the impurity feature comparison result, IC2 is the impurity pixel size of the second fiber feature, IC1 is the impurity pixel size of the first fiber feature, ICN2 is the impurity pixel quantity of the second fiber feature, and ICN1 is the impurity pixel quantity of the first fiber feature; ICP is defaulted to be in the normal range of 0%-120%, and values exceeding 120% are abnormal values. In specific implementations, the normal range can be modified according to actual situations;

[0109] The thermal insulation fiber pattern comparison unit extracts the pattern features of the first fiber feature and the second fiber feature for recognition and comparison, and generates pattern anomaly data; the comparison calculation of the pattern features is as follows:

[0110]

[0111] Among them, PS is the pattern similarity, PT2 is the pattern feature of the second fiber feature, PT1 is the pattern feature of the first fiber feature, CS is the pattern similarity, CT2 is the pattern color feature of the second fiber feature, and CT1 is the pattern color feature of the first fiber feature; the closer PS and CS are to 0, the lower the data similarity is proved;

[0112] The thermal insulation fiber thermal insulation property comparison unit extracts the thermal insulation property features of the first fiber feature and the second fiber feature for recognition and comparison, and generates thermal insulation property anomaly data; the comparison calculation of the thermal insulation property features is as follows:

[0113]

[0114] ω1 + ω2 + ω3 + ω4 = 1;

[0115] Among them, WP is the comparison calculation result of the thermal insulation property feature, WN2 is the number of pores between fibers of the second fiber feature, WN1 is the number of pores between fibers of the first fiber feature, WKN2 is the pore pixel size between fibers of the second fiber feature, WKN1 is the pore pixel size between fibers of the first fiber feature, WKN 2j is the j-th pore pixel size between fibers of the second fiber feature, WKN 1j is the j-th pore pixel size between fibers of the first fiber feature, ω1, ω2, ω3, ω4 are their respective calculation weights; the default values of the weights are 0.25, 0.25, 0.25, and 0.25, and the specific weight values can be modified in the implementation; the closer WP is to 0, the lower the data similarity is proved;

[0116] The present invention uses some simulation data for calculation. There are a total of 500 groups of simulation data. Due to the excessive amount of data, only part of the data is shown in this table, and the calculation results are shown in Table 1:

[0117] Table 1 Calculation Results and Accuracy of the Simulation Problem Model

[0118]

[0119] The present invention calculates the analysis accuracy rate by comparing the model analysis results with the true results of the simulation data. Among the last 500 groups of data, 462 groups are correct, and the accuracy rate is 92.4%.

[0120] The thermal insulation fiber abnormal output unit outputs impurity abnormal data, pattern abnormal data, and thermal insulation abnormal data.

[0121] In order to be able to lock the specific reasons for the fiber fabrics that have not passed the quality inspection, corresponding image data is obtained through an image recognition model. Based on the image features obtained by recognition, the abnormal features that appear are calculated. By calculating the problems of the thermal insulation fibers, the corresponding process problems can be accurately obtained, the robustness of the model can be improved, which is convenient for subsequent product quality analysis, and the accuracy rate of abnormal problems of the products that have not passed the quality inspection can be improved.

[0122] The third fiber feature is impurity abnormal data, pattern abnormal data, and thermal insulation abnormal data. Therefore, in order to ensure the accuracy of problem analysis, it is analyzed from three aspects: impurity abnormality, pattern abnormality, and thermal insulation abnormality, so as to ensure that all aspects of the entire production process can be covered and the comprehensiveness of problem analysis can be improved.

[0123] Conduct attribution analysis based on the third fiber feature to obtain the production process where problems occur. Call the production machine data of the corresponding process number according to the production process and the artificial fiber fabric number, and analyze the abnormal data.

[0124] Furthermore, calculate the probability of problems occurring in each link of the production process of the thermal insulation artificial fiber fabrics that have not passed the quality inspection based on the impurity abnormal data, pattern abnormal data, and thermal insulation abnormal data. The probability calculation is as follows:

[0125]

[0126] where P is the probability, P i is the problem probability of the i-th link, max() is the maximum value, is the k-th abnormal data of the i-th link, is the k-th abnormal reference value of the i-th link; is the respective feature data recognized by the thermal insulation fiber feature recognition model, is the reference homogenization of the corresponding feature data of each link obtained based on a large amount of experimental data, such as impurity pixel size, impurity quantity, etc.;

[0127] After the image feature analysis is completed, in order to confirm the specific process, considering that during the process production, an abnormal problem may be caused by multiple processes. To ensure this situation, data of each part in the production process is obtained through probability calculation. Based on the probability calculation results, the reasons for which parts are prone to this problem are obtained, improving the comprehensiveness of problem analysis and facilitating subsequent data retrieval for anomaly analysis, thus increasing the passing rate of product quality inspection.

[0128] The abnormal data analysis is obtained based on the abnormal machine analysis model for thermal insulation fiber fabrics.

[0129] Refer to Figure 4 As shown, the abnormal machine analysis model for fiber fabrics includes a machine data input unit, a machine abnormal feature extraction unit, a machine abnormal feature classification unit, and a machine abnormal feature output unit.

[0130] The machine data input unit preprocesses the production machine data to generate machine abnormal input data.

[0131] Among them, the production machine data is sorted according to the collection time, and corresponding time series sets are generated respectively according to each data type. For example, a corresponding time series set is generated for voltage data, and a time series set of displacement data is generated for displacement data. After generation, the number of time series sets is n, and the length of each time series is 1*m.

[0132] The machine abnormal feature extraction unit extracts the time series abnormal features of the machine abnormal input data to generate machine abnormal feature data. The machine abnormal feature classification unit classifies the machine abnormal feature data to generate machine abnormal features. The machine abnormal feature output unit outputs the abnormal labels of the machine abnormal features.

[0133] The abnormal machine analysis model for thermal insulation fiber fabrics uses 4 one-dimensional convolutional layers and 4 LSTM units to obtain abnormal data. The 4 one-dimensional neural networks include 2 1*5 convolutional layers and 2 1*3 convolutional layers. The 4 one-dimensional convolutional layers are directly connected to the 4 LSTM units, and classification output is performed in the Softmax layer.

[0134] In order to obtain which situations in each part of the production process actually cause the data, during the analysis process, the data from the previous production process is stored. After problem analysis, the data can be located based on the number and the reason for abnormal analysis. Through the time series analysis of each collected data, the specific data causing each problem can be analyzed, so as to facilitate subsequent relevant personnel to optimize and adjust, improve the product production line, and increase the passing rate of quality inspection of thermal insulation fiber fabrics.

[0135] Optimize the production process based on the abnormal data.

[0136] Embodiment 2

[0137] The present invention also proposes a production quality monitoring system for fiber fabrics. Through this system, based on various methods mentioned in the method, it can realize the fully automated quality inspection process of attribution analysis for each process. At the same time, according to relevant algorithms, it can also realize the automated call and automated processing of data, eliminating a lot of data processing projects. Combining the automatic processing results of the system, relevant personnel can automatically obtain problems and take corresponding solutions for rapid processing, improving the maintenance efficiency and ensuring the product quality of the thermal insulation fiber fabric;

[0138] The system includes a process number division module, a production machine data acquisition module, a thermal insulation artificial fiber fabric number marking module, a quality inspection marked data acquisition module, a first fiber feature and second fiber feature acquisition module, a third fiber feature acquisition module, an abnormal data analysis module, and a process optimization module. Refer to Figure 5 as shown, specifically:

[0139] The process number division module acquires the production process of the thermal insulation artificial fiber fabric and divides it to generate a process number;

[0140] The production machine data acquisition module marks and stores the numbers of the production machine data according to the process number;

[0141] The thermal insulation artificial fiber fabric number marking module produces and conducts quality inspection on the thermal insulation artificial fiber fabric according to the production process; during the production process, the artificial fiber fabric is numbered and marked according to the number;

[0142] The quality inspection marked data acquisition module acquires the thermal insulation artificial fiber fabric that has passed the quality inspection and obtains the quality inspection standard data of the thermal insulation artificial fiber fabric according to the quality inspection content;

[0143] The first fiber feature and second fiber feature acquisition module acquires the first image data of the thermal insulation artificial fiber fabric, segments the first image data to generate multiple first thermal insulation fabric sub-images; obtains the first fiber feature according to the first thermal insulation fabric sub-images and the quality inspection standard data; acquires the thermal insulation artificial fiber fabric that has not passed the quality inspection and obtains the second image data, segments the second image data according to the second image data to generate multiple second thermal insulation fabric sub-images; obtains the second fiber feature according to the second thermal insulation fabric sub-images and the quality inspection standard data;

[0144] The third fiber feature acquisition module performs feature comparison calculations according to the first fiber feature and the second fiber feature to obtain the third fiber feature;

[0145] Further, the third fiber feature attribution analysis shows that the third fiber feature is the impurity anomaly data, pattern anomaly data, and warmth retention anomaly data obtained by the first fiber feature and the second fiber feature acquisition module;

[0146] Based on the impurity anomaly data, pattern anomaly data, and warmth retention anomaly data, calculate the probability of problems occurring in each link of the production process of the warmth retention artificial fiber fabric that fails quality inspection. The probability calculation is as follows:

[0147]

[0148] where P is the probability, P i is the problem probability of the i-th link, max() is the maximum value, is the k-th anomaly data of the i-th link, is the k-th anomaly reference value of the i-th link;

[0149] In order to improve the comprehensiveness of system analysis, during the system calculation process, probability calculation is used to analyze each process to find the cause of the problem, avoiding the situation where anomalies caused by multiple processes are not analyzed. At the same time, probability calculation also enables the system to automatically call corresponding data, improving the operation efficiency of the entire system and ensuring the processing efficiency of related problems, ensuring that problems can be repaired in the shortest time and improving the quality inspection rate of the warmth retention fiber fabric;

[0150] The anomaly data analysis module performs attribution analysis based on the third fiber feature to obtain the production process with problems, and calls the production machine data corresponding to the process number according to the production process and the artificial fiber fabric number and analyzes the anomaly data;

[0151] Further, the anomaly data analysis is obtained based on the warmth retention fiber fabric anomaly machine analysis model;

[0152] The fiber fabric anomaly machine analysis model includes a machine data input unit, a machine anomaly feature extraction unit, a machine anomaly feature classification unit, and a machine anomaly feature output unit;

[0153] The machine data input unit preprocesses the production machine data to generate machine anomaly input data; the machine anomaly feature extraction unit extracts the time series anomaly features of the machine anomaly input data to generate machine anomaly feature data;

[0154] The machine anomaly feature classification unit classifies the machine anomaly feature data to generate machine anomaly features; the machine anomaly feature output unit outputs the anomaly labels of the machine anomaly features;

[0155] The present invention uses a full - process automatic processing of the analog data comparison system and comprehensively analyzes and compares the expert experience and machine results in image analysis, attribution analysis, and anomaly analysis. The differences in calculation speed and accuracy are shown in Table 2 as follows;

[0156] Table 2 Comparison of Results between Fully Automatic Analysis and Expert - combined with Machine Analysis

[0157]

[0158] It can be seen from the results in Table 2 that as the data volume increases, the system can maintain a stable accuracy rate. For the system + expert, the accuracy rate gradually decreases as the data volume increases. In terms of speed, the larger the data volume, the more advantageous the system's calculation effect;

[0159] Based on the analysis of stored time - series data, the system can accurately find the cause of problems from complex issues. At the same time, it can also trace the source of abnormal problems, improve the production quality of the entire product line, analyze anomalies without affecting the production of the product line, and perform timely processing to ensure the production quality of the product line and improve the passing rate of quality inspection of thermal fiber fabrics;

[0160] The process optimization module optimizes the production process according to the abnormal data.

[0161] Although the embodiments of the present invention have been shown and described, for those of ordinary skill in the art, it can be understood that various changes, modifications, substitutions, and variations can be made to these embodiments without departing from the principles and spirit of the present invention. The scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A fiber fabric production quality monitoring method, characterized in that: include: Obtaining the production process of thermal insulation artificial fiber fabrics, dividing them according to the production process, and generating process numbers; Acquire production machine data, and according to the process number, label the production machine data with corresponding numbers and store them; produce and inspect the thermal insulation artificial fiber fabric according to the production process; label the thermal insulation artificial fiber fabric number during the production process; Acquire the thermal insulation artificial fiber fabric that has passed the quality inspection, and acquire the quality inspection standard data of the thermal insulation artificial fiber fabric according to the quality inspection content; Acquire first image data of the thermal insulation artificial fiber fabric, segment the first image data, and generate a plurality of first thermal insulation fabric sub-images; acquire first fiber features according to the first thermal insulation fabric sub-images and the quality inspection standard data; Acquire the thermal insulation artificial fiber fabric that has not passed the quality inspection and obtain the second image data, perform segmentation according to the second image data, and generate a plurality of second thermal insulation fabric sub-images; acquire the second fiber feature according to the second thermal insulation fabric sub-images and the quality inspection standard data; Obtain the number of length pixels and the number of width pixels of the first image data or the second image data, obtain a first segmentation common divisor according to the number of length pixels, and obtain a second segmentation common divisor according to the number of width pixels; filter and exclude the first segmentation common divisor and the second segmentation common divisor that are smaller than a segmentation threshold; If the first common divisor and the second common divisor after exclusion have the same data, segmentation is performed based on the same data; If there is no identical data, determining the data similarity between the first common divisor and the second common divisor, and selecting the first common divisor and the second common divisor with the largest data similarity for segmentation; The first fiber feature and the second fiber feature are identified by a thermal insulation fiber feature identification model; The thermal insulation fiber feature recognition model includes a thermal insulation fiber image input unit, a thermal insulation fiber feature classification unit, a thermal insulation fiber impurity comparison unit, a thermal insulation fiber pattern comparison unit, a thermal insulation fiber warmth comparison unit and a thermal insulation fiber abnormality output unit; The thermal fiber image input unit preprocesses the first image data or the second image data to generate thermal fiber input data; The thermal insulation fiber feature classification unit obtains the thermal insulation fiber input data, identifies and classifies corresponding image features according to the quality inspection standard data, and generates the first fiber feature or the second fiber feature; The thermal insulation fiber impurity comparison unit extracts the impurity features of the first fiber feature and the second fiber feature for identification and comparison, and generates impurity abnormality data; the thermal insulation fiber pattern comparison unit extracts the pattern features of the first fiber feature and the second fiber feature for identification and comparison, and generates pattern abnormality data; the thermal insulation fiber thermal insulation comparison unit extracts the thermal insulation features of the first fiber feature and the second fiber feature for identification and comparison, and generates thermal insulation abnormality data; The thermal fiber abnormality output unit outputs the impurity abnormality data, the pattern abnormality data and the thermal insulation abnormality data; the third fiber feature is the impurity abnormality data, the pattern abnormality data and the thermal insulation abnormality data; Performing feature comparison calculation based on the first fiber feature and the second fiber feature to obtain a third fiber feature; Performing attribution analysis based on the third fiber feature to obtain the production process with the problem, calling the production machine data corresponding to the process number based on the production process and the number of the thermal insulation artificial fiber fabric, and analyzing the abnormal data; The production process is optimized according to the abnormal data.

2. A fiber fabric production quality monitoring method according to claim 1, characterized in that The production machine data is collected according to a fixed time interval T, and the production machine data with different process numbers use the same time interval T.

3. A fiber fabric production quality monitoring method according to claim 1, characterized in that: The third fiber characteristic attribution analysis includes: The probability of the thermal insulation artificial fiber fabric that has not passed the quality inspection having problems in each link of the production process is calculated based on the impurity abnormality data, the pattern abnormality data and the thermal insulation abnormality data. The probability is calculated as follows: ； in, is the probability, is the probability of the problem in the ith link, max() is the maximum value, is the kth abnormal data in the ith link, is the kth abnormal reference value of the ith link.

4. A fiber fabric production quality monitoring method according to claim 1, characterized in that: Abnormal data analysis of production machine data includes: The abnormal data analysis is obtained based on the abnormal machine analysis model of thermal fiber fabrics; The fiber fabric abnormality machine analysis model includes a machine data input unit, a machine abnormality feature extraction unit, a machine abnormality feature classification unit and a machine abnormality feature output unit; The machine data input unit pre-processes the production machine data to generate machine abnormality input data; The machine abnormality feature extraction unit extracts the time series abnormality features of the machine abnormality input data to generate machine abnormality feature data; the machine abnormality feature classification unit classifies the machine abnormality feature data to generate machine abnormality features; and the machine abnormality feature output unit outputs the abnormality label of the machine abnormality feature.

5. A fiber fabric production quality monitoring system, characterized in that: The process number division module obtains and divides the production process of thermal insulation artificial fiber fabrics and generates process numbers; A production machine data acquisition module, which marks the number of the production machine data according to the process number and stores it; A module for marking numbers of warm artificial fiber fabrics, wherein the warm artificial fiber fabrics are produced and quality inspected according to the production process; and the artificial fiber fabrics are numbered and marked according to the numbers during the production process; A quality inspection mark data acquisition module is used to acquire the thermal insulation artificial fiber fabric that has passed the quality inspection, and to acquire the quality inspection standard data of the thermal insulation artificial fiber fabric according to the quality inspection content; A first fiber feature and a second fiber feature acquisition module, which acquires the first image data of the thermal insulation artificial fiber fabric, segments the first image data, and generates a plurality of first thermal insulation fabric sub-images; and acquires the first fiber feature according to the first thermal insulation fabric sub-images and the quality inspection standard data; Acquire the thermal insulation artificial fiber fabric that has not passed the quality inspection and obtain the second image data, perform segmentation according to the second image data, and generate a plurality of second thermal insulation fabric sub-images; acquire the second fiber feature according to the second thermal insulation fabric sub-images and the quality inspection standard data; Obtain the number of length pixels and the number of width pixels of the first image data or the second image data, obtain a first segmentation common divisor according to the number of length pixels, and obtain a second segmentation common divisor according to the number of width pixels; filter and exclude the first segmentation common divisor and the second segmentation common divisor that are smaller than a segmentation threshold; If the first common divisor and the second common divisor after exclusion have the same data, segmentation is performed based on the same data; If there is no identical data, determining the data similarity between the first common divisor and the second common divisor, and selecting the first common divisor and the second common divisor with the largest data similarity for segmentation; The first fiber feature and the second fiber feature are identified by a thermal insulation fiber feature identification model; The thermal insulation fiber feature recognition model includes a thermal insulation fiber image input unit, a thermal insulation fiber feature classification unit, a thermal insulation fiber impurity comparison unit, a thermal insulation fiber pattern comparison unit, a thermal insulation fiber warmth comparison unit and a thermal insulation fiber abnormality output unit; The thermal fiber image input unit preprocesses the first image data or the second image data to generate thermal fiber input data; The thermal insulation fiber feature classification unit obtains the thermal insulation fiber input data, identifies and classifies corresponding image features according to the quality inspection standard data, and generates the first fiber feature or the second fiber feature; The thermal insulation fiber impurity comparison unit extracts the impurity features of the first fiber feature and the second fiber feature for identification and comparison, and generates impurity abnormality data; the thermal insulation fiber pattern comparison unit extracts the pattern features of the first fiber feature and the second fiber feature for identification and comparison, and generates pattern abnormality data; the thermal insulation fiber thermal insulation comparison unit extracts the thermal insulation features of the first fiber feature and the second fiber feature for identification and comparison, and generates thermal insulation abnormality data; The thermal fiber abnormality output unit outputs the impurity abnormality data, the pattern abnormality data and the thermal insulation abnormality data; the third fiber feature is the impurity abnormality data, the pattern abnormality data and the thermal insulation abnormality data; A third fiber feature acquisition module performs feature comparison calculation based on the first fiber feature and the second fiber feature to acquire a third fiber feature; an abnormal data analysis module, performing attribution analysis based on the third fiber feature, obtaining the production process with the problem, calling the production machine data corresponding to the process number based on the production process and the artificial fiber fabric number, and analyzing the abnormal data; A process optimization module optimizes the production process according to the abnormal data.

6. A fiber fabric production quality monitoring system according to claim 5, characterized in that: The third fiber feature acquisition module comprises: The third fiber feature attribution analysis is that the third fiber feature is the impurity abnormality data, pattern abnormality data and warmth retention abnormality data acquired by the first fiber feature and the second fiber feature acquisition module; The probability of the thermal insulation artificial fiber fabric that has not passed the quality inspection having problems in each link of the production process is calculated based on the impurity abnormality data, the pattern abnormality data and the thermal insulation abnormality data. The probability is calculated as follows: ； in, is the probability, is the probability of the problem in the ith link, max() is the maximum value, is the kth abnormal data in the ith link, is the kth abnormal reference value of the ith link.

7. A fiber fabric production quality monitoring system according to claim 5, characterized in that: The abnormal data analysis module includes: The abnormal data analysis is obtained based on the abnormal machine analysis model of thermal fiber fabrics; The fiber fabric abnormality machine analysis model includes a machine data input unit, a machine abnormality feature extraction unit, a machine abnormality feature classification unit and a machine abnormality feature output unit; The machine data input unit pre-processes the production machine data to generate machine abnormality input data; The machine abnormality feature extraction unit extracts the time series abnormality features of the machine abnormality input data to generate machine abnormality feature data; the machine abnormality feature classification unit classifies the machine abnormality feature data to generate machine abnormality features; and the machine abnormality feature output unit outputs the abnormality label of the machine abnormality feature.

Images