Control method and system for non-woven fabric production line

Abstract

The invention relates to the technical field of non-woven fabric production line control, in particular to a non-woven fabric production line control method and system, and the method comprises the steps: collecting videos before and after a uniform fiber web on a working production line enters a needling machine, and obtaining a needling raw material video and a needling effect video; performing frame extraction on the acupuncture effect video according to a first time interval to obtain acupuncture effect images, and inputting the preprocessed acupuncture effect images into a pricking needle state classification model for classification; if the blunt needle production image is recognized, calculating the total number of the blunt needle images within specified time to obtain a first number; if the first number exceeds a preset threshold value, the current parameter is adjusted to be a corresponding stage optimal control parameter; and if the damaged needle production image is identified, calculating the total number of damaged needle images within a specified time to obtain a second number, and if the second number exceeds a preset threshold, replacing the needle. According to the method, the stability of the product quality is improved through precise production line control parameter regulation and control.

Description

Technical Field

[0001] The present invention relates to the technical field of non-woven fabric production line control, and in particular to a control method and system for a non-woven fabric production line. Background Art

[0002] Nonwoven fabrics are different from woven fabrics made by weaving yarns. There are no yarns in nonwoven fabrics. The fibers are combined physically or chemically. They are usually light and soft. Therefore, nonwoven fabrics are widely used in medical, sanitary and industrial fields. Needle-punched nonwoven fabrics are a type of nonwoven fabrics, which mainly combine fibers together through a needle-punching process.

[0003] The existing needle-punched nonwoven production line has basically achieved automated production, which can realize the mass production of nonwovens. The production process of needle-punched nonwovens has designed multiple process links, especially in the needle-punching process, which is significantly different from other types of nonwoven production lines.

[0004] In the existing intelligent production line, during the acupuncture process, as the acupuncture operation continues, the tip of the needle will wear and age, affecting its puncture performance. Although the existing intelligent control system has a certain degree of adaptive ability, it does not fully consider the state of the needle during the processing. Although some researchers have taken needle replacement into consideration, the current method mainly relies on a single indicator of wear degree, and there are still deficiencies in needle replacement identification, and it is difficult to optimize control parameters according to different needle states. There is an urgent need for a control method for the acupuncture process in the production line to obtain more accurate production line control parameters while ensuring the continuous and stable operation of the production line and the consistency of product quality.

[0005] To this end, a control method and system for a nonwoven fabric production line are proposed. Summary of the invention

[0006] The purpose of the present invention is to provide a control method and system for a non-woven fabric production line, which can realize real-time monitoring and dynamic control of the needle state through video acquisition and image analysis. It includes collecting videos before and after the fiber web enters the needling machine, performing quality inspection on the needling raw material image and extracting frames to process the needling effect image. The image is classified by a trained needle state classification model, the production images of blunt needles and damaged needles are identified, and the first quantity and the second quantity are calculated according to the detection results. When the first quantity exceeds a preset threshold, the current control parameters of the needling machine are adjusted to the optimal control parameters of the corresponding stage; when the second quantity exceeds the preset threshold, the needles are replaced. This method realizes the optimization of adaptive control parameters under different working conditions while ensuring the continuous and stable operation of the production line and the consistency of product quality through dynamic monitoring and production line parameter adjustment, thereby improving the stability of product quality.

[0007] To achieve the above object, the present invention provides the following technical solutions:

[0008] A control method for a nonwoven fabric production line, comprising:

[0009] Collect the video of the uniform fiber web on the production line before it enters the needle punching machine to obtain the video of the needle punching raw material;

[0010] Performing quality inspection based on the acupuncture raw material video, and triggering a production line warning if the quality inspection fails; the current parameters of the acupuncture stage of the working production line are preset standard control parameters;

[0011] If the quality inspection is passed, a video of the uniform fiber web on the working production line passing through the needling machine is collected to obtain a needling effect video;

[0012] Extracting frames from the acupuncture effect video at a first time interval to obtain a plurality of first acupuncture effect images;

[0013] Preprocessing the acupuncture effect image to obtain a standard acupuncture effect image;

[0014] Input the standard acupuncture effect image into a trained needle state classification model for classification, and classify it into a normal production image, a blunt needle production image, and a damaged needle production image; the needle state classification model is trained based on historical production data;

[0015] If the blunt needle production image is identified, a first number is calculated; the first number is the total number of the blunt needle production images identified within a specified time;

[0016] If the damaged needle production image is identified, a second number is calculated; the second number is the total number of the damaged needle production images identified within the specified time;

[0017] When the first number exceeds a first preset threshold, identifying the blunt needle stage of the blunt needle production image, and adjusting the current parameter to the corresponding stage optimal control parameter according to the blunt needle stage;

[0018] When the second number exceeds a second preset threshold, the needles of the needling machine are replaced.

[0019] Further, collecting the video of the uniform fiber web on the working production line before entering the needle loom includes:

[0020] The first high-definition camera is installed behind the web laying device and just above the conveyor belt in front of the needle loom, and the shooting direction of the first high-definition camera is vertically downward.

[0021] Furthermore, performing the quality inspection according to the acupuncture raw material video includes:

[0022] Extract frames from the acupuncture raw material video at the first time interval to obtain a plurality of first acupuncture raw material images;

[0023] Preprocessing the acupuncture raw material image to obtain a standard acupuncture raw material image;

[0024] The standard needle punched raw material image is input into a trained fiber web quality assessment model to obtain a quality inspection result; the quality inspection result includes qualified and unqualified.

[0025] Furthermore, the fiber web quality assessment model includes:

[0026] The input layer is used to receive the input image and perform preliminary feature representation processing;

[0027] A feature extraction layer, used to extract key features of the input image to obtain basic features of the fiber web, wherein the basic features of the fiber web include texture features, edge features, uniformity features, and fiber distribution features of the fiber web; the feature extraction layer includes multiple convolutional layers and multiple pooling layers;

[0028] The multi-scale feature fusion layer is used to fuse the feature information of different scales from different convolutional layers, and obtain multi-scale fusion features through convolution operations of different receptive fields;

[0029] An attention mechanism layer, used to optimize feature representation through a channel attention mechanism and a spatial attention mechanism; the channel attention mechanism is used to weight different feature channels to obtain weighted channel features, and the spatial attention mechanism is used to focus on key areas of the image to obtain weighted spatial features;

[0030] A feature fusion layer, used to fuse the weighted channel features and the weighted spatial features optimized by the channel attention mechanism and the spatial attention mechanism to obtain a final fused feature;

[0031] The anomaly detection layer is used to detect anomalies of the final fusion features; it includes an autoencoder module, which is used to reconstruct the basic features of the fiber web and calculate the reconstruction error; when the reconstruction error exceeds a preset threshold, the input image is judged to be abnormal, and an abnormal warning message is directly output; if no abnormality is detected, the quality classification is continued;

[0032] The classification layer is used to classify the final fusion features through the fully connected layer to obtain the quality inspection result.

[0033] Further, if the quality inspection is passed, collecting the video of the uniform fiber web on the working production line after passing through the needle loom includes:

[0034] A second high-definition camera is installed just above the area between the needle loom and the winding device, and the shooting direction of the second high-definition camera is vertically downward.

[0035] Furthermore, the process of obtaining the first quantity and the second quantity includes:

[0036] When the blunt needle production image is identified, a blunt needle effect video clip within the specified time is obtained according to the timestamp of the blunt needle production image; the blunt needle effect video clip is framed at a second time interval to obtain a plurality of second acupuncture effect images; the first time interval is greater than the second time interval;

[0037] Counting the number of the blunt needle production images in all the second acupuncture effect images to obtain the first number;

[0038] When the damaged needle production image is identified, a damaged needle effect video clip within the specified time is obtained according to the timestamp of the blunt needle production image;

[0039] Extracting frames from the damaged needle effect video clip at the second time interval to obtain a plurality of third acupuncture effect images;

[0040] The second number is obtained by counting the number of the blunt needle production images in all the third acupuncture effect images.

[0041] Furthermore, the acquisition of the historical production data includes:

[0042] The historical production data includes historical acupuncture effect images and historical annotation data in different states; the historical annotation data includes needle state annotations corresponding to the historical acupuncture effect images; the different states include normal operation state, blunt needle state and damaged needle state;

[0043] respectively acquiring the historical acupuncture effect images and the historical annotation data in the normal operation state, the blunt needle state and the damaged needle state from the operation and maintenance database; the operation and maintenance database records the historical needle working condition images corresponding to the historical acupuncture effect images;

[0044] The acquisition of the historical annotation data includes:

[0045] Acquire a standard working condition image of the needle, wherein the standard working condition image of the needle is acquired after the needle is replaced as a whole but before it starts working;

[0046] Calculating the average wear amount of the needles by analyzing the historical working condition image of the needles and the standard working condition image of the needles;

[0047] The historical acupuncture effect image is annotated according to the average wear amount of the needle to obtain the historical annotated data.

[0048] Furthermore, inputting the standard acupuncture effect image into the trained acupuncture state classification model for classification includes:

[0049] The standard acupuncture effect image is input into the trained needle state classification model to obtain the predicted needle wear amount, and according to the predicted needle wear amount, the standard acupuncture effect image is divided into the normal production image, the blunt needle production image and the damaged needle production image.

[0050] Furthermore, the process of obtaining the optimal control parameters in the stage includes:

[0051] According to the predicted wear amount of the needle, the blunt needle stage of the blunt needle production image is identified; the blunt needle stage includes an early stage, a middle stage and a late stage;

[0052] Obtain control parameters and final product quality indicators corresponding to the initial stage, the mid-stage and the late stage from the operation and maintenance database to obtain initial product data, mid-stage product data and late product data;

[0053] By correlation analysis, the key control parameters of the stages in the initial product data, the mid-term product data and the late product data are identified to obtain the initial key control parameters, the mid-term key control parameters and the late key control parameters;

[0054] For each stage, a regression model is constructed to analyze the relationship between the key control parameters of the stage and the quality index of the final product, and an initial regression model, a mid-term regression model and a late-term regression model are obtained;

[0055] Construct a comprehensive objective function;

[0056] The effects of different control parameter combinations are predicted by using the initial regression model, the mid-term regression model and the late-term regression model, and the optimal control parameters of each stage are searched through an optimization algorithm to obtain the initial optimal control parameters, the mid-term optimal control parameters and the late-term optimal control parameters.

[0057] A control system for a nonwoven fabric production line, comprising:

[0058] The raw material video acquisition module is used to collect the video of the uniform fiber web on the working production line before entering the needling machine to obtain the needling raw material video; the current parameters of the needling stage of the working production line are the preset standard control parameters;

[0059] A raw material quality inspection module is used to perform quality inspection based on the acupuncture raw material video, and if the raw material fails the quality inspection, a production line warning is triggered;

[0060] A needling effect video acquisition module, for acquiring a video of a uniform fiber web on the working production line after it passes the needling machine, to obtain a needling effect video if the quality inspection is passed;

[0061] An image processing module, configured to extract frames of the acupuncture effect video at a first time interval to obtain a plurality of first acupuncture effect images; and pre-process the acupuncture effect images to obtain standard acupuncture effect images;

[0062] A model classification module is used to input the standard acupuncture effect image into a trained needle state classification model for classification, and classify the image into a normal production image, a blunt needle production image, and a damaged needle production image; the needle state classification model is trained based on historical production data;

[0063] A quantity calculation module, configured to calculate a first quantity if the blunt needle production image is identified; the first quantity is the total quantity of the blunt needle production images identified within a specified time; and to calculate a second quantity if the damaged needle production image is identified; the second quantity is the total quantity of the damaged needle production images identified within the specified time;

[0064] A control adjustment module is used to identify the blunt needle stage of the blunt needle production image when the first number exceeds a first preset threshold, and adjust the current parameters to the corresponding stage optimal control parameters according to the blunt needle stage; when the second number exceeds a second preset threshold, replace the needles of the acupuncture machine.

[0065] Compared with the prior art, the present invention has the following beneficial effects:

[0066] 1. The present invention proposes a fiber web quality assessment model, which can obtain the state of a uniform fiber web before entering the needling machine through the needling raw material video in real time without affecting the normal operation of the production line. By performing real-time quality inspection on the needling raw material video, a reliable basis is provided for the subsequent standard needling effect image classification, eliminating the interference of external raw material factors on quality inspection. At the same time, the fiber web quality assessment model integrates an attention mechanism layer, and optimizes feature representation through channel attention and spatial attention mechanisms, so that the model can automatically focus on important areas and key features in the fiber web image, thereby improving the model's sensitivity and accuracy to small defects and quality problems in the fiber web. In addition, the model also introduces an anomaly detection layer, which reconstructs the basic features of the fiber web through the autoencoder and calculates the reconstruction error, can accurately identify unseen abnormal samples, output abnormal warnings in time, and prevent abnormal data from affecting subsequent classification decisions. This not only improves the accuracy of detection, but also enhances the robustness of the system, ensuring that any production problems can be quickly discovered and handled. The present invention can realize efficient raw material quality inspection before the fiber web enters the needling machine, provides effective raw material guarantee, ensures the consistency and stability of the raw materials before needling, makes subsequent parameter adjustment more accurate, and helps to improve the stability of the overall product quality.

[0067] 2. By using the first time interval for rough inspection, potential blunt needle and damaged needle problems in the production process can be quickly screened without affecting production efficiency. When a blunt needle production image or a damaged needle production image is identified during the rough inspection process, the corresponding video clip is further processed through the second time interval to obtain more inspection samples, so as to conduct a more detailed analysis of the production status during this time period. This dual time interval design ensures a balance between inspection efficiency and accuracy; during preliminary screening, rough inspection can quickly identify potential problems, and when problems occur, fine inspection can provide higher inspection accuracy and sample quantity statistics to ensure that the problems can be accurately identified and quantified. By counting the number of blunt needle and damaged needle images, the system can quickly respond to changes in the status of production equipment and provide real-time feedback for subsequent adaptive control, thereby effectively improving the adaptive control capabilities of the production line.

[0068] 3. The control method of the non-woven fabric production line proposed in the present invention predicts the amount of needle wear and refines the blunt needle stage into early, middle and late stages. This stage division makes the control of the production line more accurate and can make targeted adjustments for different needle wear states, avoiding the problem of single control strategy failure at different stages. Correlation analysis is performed on the historical data of the operation and maintenance database to identify the key control parameters of each stage, and the relationship between the control parameters and product quality is analyzed using a regression model to ensure the scientificity and rationality of the control parameter selection. In addition, by constructing a comprehensive objective function and using an optimization algorithm to search for the optimal control parameters, the balance between different control indicators is ensured and overall optimization is achieved. The accuracy of production line control and the stability of product quality are further improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0069] Figure 1 A flow chart of a control method for a nonwoven fabric production line provided by an embodiment of the present invention;

[0070] Figure 2 A schematic top view of the layout of various devices on a nonwoven fabric production line provided by an embodiment of the present invention;

[0071] Figure 3 A structural diagram of a fiber web quality assessment model provided by an embodiment of the present invention;

[0072] Figure 4 The present invention is a control system structure diagram of a non-woven fabric production line.

[0073] In the figure: 1. Uniform fiber web before entering the needling machine; 2. First high-definition camera; 3. Web laying device; 4. Needling machine; 5. Conveyor belt; 6. Uniform fiber web after entering the needling machine; 7. Second high-definition camera; 8. Winding device. DETAILED DESCRIPTION

[0074] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0075] At present, the non-woven fabric production lines of large factories have basically achieved automated production. The production method of needle-punched non-woven fabrics in existing factories usually adopts high-speed needle-punching process. The needle row is an important component in the non-woven fabric needle-punching process, on which a large number of needles are arranged; through the up and down movement of the needle-punching machine, the needle row punctures the fiber web at a high frequency. Due to the continuous high-frequency puncture, the needle row will inevitably accumulate wear and tear, and eventually cause the needle-punching effect to deviate from the standard.

[0076] Existing automated production lines have difficulty in achieving real-time monitoring of the status of needles, and usually require shutdown for inspection, which can significantly affect the overall production efficiency of nonwovens. Existing methods mainly rely on the length of time the needles have been in use to assess whether the needles need to be replaced. They lack an accurate assessment of the actual wear status and are unable to promptly identify needle wear or damage, which may result in reduced product quality or premature equipment replacement, increasing production costs. In addition, the length of use assessment method cannot be dynamically adjusted according to actual working conditions, and cannot cope with complex production environments and different needle wear speeds, resulting in insufficient automation and flexibility of the production line, making it difficult to ensure the continuity and stability of production.

[0077] To this end, a control method and system for a nonwoven fabric production line are proposed, which are described through the following embodiments.

[0078] Embodiment 1:

[0079] Company A is a company whose main business is the manufacturing of industrial textile products and fabric textile processing. The company has multiple production lines responsible for the processing of needle textile raw materials. In order to improve the quality of its product, needle-punched non-woven fabrics, Company A introduced a strict quality inspection process. After quality analysis and problem location by the quality inspection department, it was found that product quality problems mainly occurred in the needle punching process, such as needle wear and uneven needle punching, which affected the stability and consistency of the final product. Therefore, Company A proposed and applied a control method for non-woven fabric production lines, including:

[0080] See also Figure 1 S10 in which a video of a uniform fiber web on the working production line before entering the needling machine is collected to obtain a needling raw material video; quality inspection is performed based on the needling raw material video, and if the quality inspection fails, a production line warning is triggered to prompt a check of the raw material status and to suspend production; the current parameters of the needling stage of the working production line are preset standard control parameters; if the quality inspection passes, a video of the uniform fiber web on the working production line after passing through the needling machine is collected to obtain a needling effect video;

[0081] In this embodiment, other variables that may affect the image need to be strictly controlled during the production process of the production line. Ensure that the quality and properties of the raw materials remain consistent to eliminate the impact of material changes on the image. It is necessary to keep the environmental conditions of the production line (such as temperature, humidity and light) constant, which can be achieved by controlling environmental equipment such as thermostats, air circulation control devices and constant light source systems to prevent environmental factors from interfering with image data. In addition, the key parameters of the equipment operation should be operated in a stable state.

[0082] Further, see Figure 2 , the video of collecting the uniform fiber web 1 on the working production line before entering the needle punching machine 4 includes:

[0083] like Figure 2 As shown, the working production line runs from left to right, and the first high-definition camera 2 is installed just above the conveyor belt 5 behind the web laying device 3 and in front of the needling machine 4, and the shooting direction of the first high-definition camera 2 is vertically downward.

[0084] The first high-definition camera is installed in a fixed position and can stably collect video of the uniform fiber web before it enters the needling machine, thereby realizing accurate monitoring of the state of the uniform fiber web before it enters the needling machine, and can obtain in real time the complete state image of the fiber web before entering the needling process; providing a high-quality data basis for subsequent quality inspections.

[0085] Furthermore, performing the quality inspection according to the acupuncture raw material video includes:

[0086] Extract frames from the acupuncture raw material video at the first time interval to obtain a plurality of first acupuncture raw material images;

[0087] Preprocessing the acupuncture raw material image to obtain a standard acupuncture raw material image;

[0088] The standard needle punched raw material image is input into a trained fiber web quality assessment model to obtain a quality inspection result; the quality inspection result includes qualified and unqualified.

[0089] By extracting frames according to the first time interval, images can be extracted at fixed time intervals for quality inspection without affecting the production speed of the production line. This ensures that the fiber web before entering the needle loom can be monitored. This frame extraction method does not require frame-by-frame analysis of the entire video, greatly reducing the amount of calculation and inspection time.

[0090] Furthermore, if Figure 3 As shown, the fiber web quality assessment model includes:

[0091] The input layer is used to receive the input image and perform preliminary feature representation processing;

[0092] A feature extraction layer, used to extract key features of the input image to obtain basic features of the fiber web, wherein the basic features of the fiber web include texture features, edge features, uniformity features, and fiber distribution features of the fiber web; the feature extraction layer includes multiple convolutional layers and multiple pooling layers;

[0093] The multi-scale feature fusion layer is used to fuse the feature information of different scales from different convolutional layers, and obtain multi-scale fusion features through convolution operations of different receptive fields;

[0094] An attention mechanism layer, used to optimize feature representation through a channel attention mechanism and a spatial attention mechanism; the channel attention mechanism is used to weight different feature channels to obtain weighted channel features, and the spatial attention mechanism is used to focus on key areas of the image to obtain weighted spatial features;

[0095] A feature fusion layer, used to fuse the weighted channel features and the weighted spatial features optimized by the channel attention mechanism and the spatial attention mechanism to obtain a final fused feature;

[0096] The anomaly detection layer is used to detect anomalies of the final fusion features; it includes an autoencoder module, which is used to reconstruct the basic features of the fiber web and calculate the reconstruction error; when the reconstruction error exceeds a preset threshold, the input image is judged to be abnormal, and an abnormal warning message is directly output; if no abnormality is detected, the quality classification is continued;

[0097] Specifically, the anomaly detection layer includes an autoencoder module for reconstructing the basic features of the fiber web. First, the autoencoder module consists of an encoder and a decoder. The encoder compresses the basic features of the fiber web into a low-dimensional feature representation, and the decoder reconstructs the compressed low-dimensional features to generate reconstructed fiber web features. By calculating the reconstruction error between the original fiber web basic features and the reconstructed fiber web features, if the reconstruction error exceeds the preset anomaly threshold, the sample is judged to be abnormal and the abnormal warning information is directly output; if the reconstruction error does not exceed the preset anomaly threshold, the sample is considered normal, and its features are further input into the classification layer for quality classification.

[0098] The classification layer is used to classify the final fusion features through the fully connected layer to obtain the quality inspection result.

[0099] Table 1. Quality inspection comparison experiment data table

[0100]

[0101] In order to verify the quality inspection effect of raw materials (uniform fiber web) based on the fiber web quality assessment model proposed in the present invention, a comparative experiment was designed to compare and analyze the defective rate in the production of needle-punched non-woven fabrics and the accuracy of raw material quality inspection through different raw material inspection methods. The experiment is divided into three groups. Experiment 1 uses the fiber web quality assessment model proposed in the present invention for quality inspection; Experiment 2 uses the traditional CNN binary classification model to detect the fiber web raw materials. The CNN binary classification model includes multiple convolutional layers, multiple pooling layers and a fully connected layer, and finally outputs a binary classification result (qualified or unqualified); Experiment 3 is a group that did not undergo quality inspection and is used for comparison. The subsequent processes of the three groups of experiments remain consistent, refer to Figure 1Among S20 to S80, only the raw material inspection method in S10 is different.

[0102] Referring to Table 1, the quality inspection method based on the fiber web quality assessment model used in this embodiment can effectively reduce the defective rate of the product, and the inspection accuracy is significantly higher than that of the traditional detection model.

[0103] The present invention realizes efficient evaluation of fiber mesh quality by introducing a fiber mesh quality evaluation model and combining it with a multi-layer structure design. The feature extraction layer obtains the basic features of the fiber mesh, such as texture, edge, uniformity and fiber distribution, through the convolution layer and the pooling layer, ensuring that the subtle information of the fiber mesh structure can be accurately extracted. The multi-scale feature fusion layer fuses information from different receptive fields through convolution operations of different scales to ensure that the model is robust when processing features of different sizes and shapes. The attention mechanism layer further optimizes the feature representation through channel attention and spatial attention mechanisms, so that the model can automatically focus on the key areas and important features in the fiber mesh image, thereby improving the sensitivity and accuracy of the model to fiber mesh defects and quality issues. The introduced anomaly detection layer reconstructs the basic features of the fiber mesh through the autoencoder module and calculates the reconstruction error for detecting potential unseen abnormal samples. When the reconstruction error exceeds the preset threshold, an abnormal warning can be issued in time to prevent abnormal samples from affecting subsequent quality classification. The proposed fiber web quality assessment model not only improves the accuracy of fiber web quality detection, but also enhances the robustness of the system, ensuring the rapid discovery and handling of fiber web quality problems, laying the foundation for the subsequent classification of needle states based on needle effect images, ensuring that the raw materials meet the standards, and preventing them from bringing in additional interference that affects subsequent analysis, thereby improving the automation level of the production line and the consistency of the products.

[0104] Further, see Figure 2 If the quality inspection is passed, the video of the uniform fiber web 6 on the working production line after passing through the needle punching machine 4 includes:

[0105] like Figure 2 As shown, the working production line runs from left to right, and the second high-definition camera 7 is installed just above the area 8 between the needling machine 4 and the winding device, and the shooting direction of the second high-definition camera 7 is vertically downward.

[0106] This video acquisition method ensures that the final needling effect of the fiber web can be monitored in real time after the needling process is completed. It helps to accurately capture the state changes of the fiber web after the needling process, especially key quality indicators such as needling strength, fiber distribution and fiber web uniformity. Placing the camera between the needling machine and the winding device can avoid interference with the fiber web during subsequent processing, allowing the system to detect potential quality problems or process anomalies in time before the fiber web is further processed or rolled up. This layout can not only improve the accuracy of quality inspection, but also issue timely warnings to prevent unqualified fiber webs from entering the next production link, thereby improving the overall quality control efficiency and automation level of the production line.

[0107] See also Figure 1 In S20, the acupuncture effect video is framed at a first time interval to obtain a plurality of first acupuncture effect images; the method can flexibly adjust the monitoring interval according to the life cycle of the needle. For example, the first time interval is often longer in the initial stage, which can save computing resources, while the first acupuncture effect images can be collected more frequently in the middle and late stages of the life cycle of the needle.

[0108] The acupuncture effect video is a continuous and uninterrupted monitoring video. Those skilled in the art can set the first time interval according to the actual situation. Generally, the first time interval is set larger in the early stage after the needle row is replaced, and the first time interval can be gradually reduced as the cumulative working time increases. In this embodiment, the cumulative working time of the needle row is close to its expected life span, and the first time interval is set to 1 hour.

[0109] See also Figure 1 S30, preprocessing the acupuncture effect image to obtain a standard acupuncture effect image;

[0110] See also Figure 1 In S40, the standard acupuncture effect image is input into a trained needle state classification model to obtain a predicted needle wear amount, and then classified according to the predicted wear amount into normal production images (0≤predicted needle wear amount<30%), blunt needle production images (30%≤predicted needle wear amount≤70%) and damaged needle production images (70%<predicted needle wear amount≤100%); the needle state classification model is trained based on historical production data;

[0111] Furthermore, the acquisition of the historical production data includes:

[0112] The historical production data includes historical acupuncture effect images and historical annotation data in different states; the historical annotation data includes needle state annotations corresponding to the historical acupuncture effect images; the different states include normal operation state, blunt needle state and damaged needle state;

[0113] respectively acquiring the historical acupuncture effect images and the historical annotation data in the normal operation state, the blunt needle state and the damaged needle state from the operation and maintenance database; the operation and maintenance database records the historical needle working condition images corresponding to the historical acupuncture effect images;

[0114] The acquisition of the historical annotation data includes:

[0115] Acquire a standard working condition image of the needle, wherein the standard working condition image of the needle is acquired after the needle is replaced as a whole but before it starts working;

[0116] Calculating the average wear amount of the needles by analyzing the historical working condition image of the needles and the standard working condition image of the needles;

[0117] The historical acupuncture effect image is annotated according to the average wear amount of the needle to obtain the historical annotated data.

[0118] The present invention can provide accurate reference and optimization basis for the real-time production process of the current production line by acquiring and analyzing historical production data under different states, including historical acupuncture effect images and corresponding historical annotation data. By extracting historical images and annotation data of normal operation state, blunt needle state and damaged needle state from the operation and maintenance database, and combining the standard working condition image after the needle is replaced, the average wear of the needle can be accurately calculated. It is helpful to make intelligent judgment and prediction of the state of the needle in the current production based on historical data, so as to identify potential equipment problems in advance and avoid the limitations of traditional maintenance that relies on fixed time. Through the annotation and analysis of such historical data, the system can predict the actual wear of the needle according to the acupuncture effect image, and adaptively adjust the current production parameters or directly replace the needle according to the predicted wear of the needle, thereby improving the real-time monitoring capability and response speed of the production line, and ensuring the stability of the acupuncture process and the consistency of product quality. At the same time, this intelligent analysis based on historical data provides an important reference for preventive maintenance, prolongs the service life of the equipment, and reduces the risk of downtime.

[0119] Furthermore, inputting the standard acupuncture effect image into the trained acupuncture state classification model for classification includes:

[0120] The standard acupuncture effect image is input into the trained needle state classification model to obtain the predicted needle wear amount, and according to the predicted needle wear amount, the standard acupuncture effect image is divided into the normal production image, the blunt needle production image and the damaged needle production image.

[0121] The present invention realizes intelligent classification and prediction of needle wear status by inputting standard acupuncture effect images into a trained needle state classification model. Through the predicted needle wear output by the model, the acupuncture effect image can be accurately divided into normal production images, blunt needle production images and damaged needle production images. Through the automatic classification of needle status, the different wear stages of the needle can be identified in time without stopping for inspection, avoiding the lag and inaccuracy of shutdown inspection and empirical judgment. When the needle wear reaches the passivation or damage state, the corresponding maintenance or replacement operation can be automatically triggered, thereby ensuring the continuity of the production process and the stability of product quality. In addition, this model-based classification method can significantly reduce the frequency of shutdown inspections, improve the degree of automation of the production line, extend the service life of the needle, and optimize the overall efficiency of the production process.

[0122] See also Figure 1 In S50, if the blunt needle production image is identified, a first number is calculated; the first number is the total number of the blunt needle production images identified within a specified time;

[0123] See also Figure 1 In S60, if the damaged needle production image is identified, a second number is calculated; the second number is the total number of the damaged needle production images identified within the specified time;

[0124] Furthermore, the process of obtaining the first quantity and the second quantity includes:

[0125] When the blunt needle production image is identified, a blunt needle effect video clip within the specified time is obtained according to the timestamp of the blunt needle production image; the blunt needle effect video clip is framed at a second time interval to obtain a plurality of second acupuncture effect images; the first time interval is greater than the second time interval;

[0126] Counting the number of the blunt needle production images in all the second acupuncture effect images to obtain the first number;

[0127] When the damaged needle production image is identified, a damaged needle effect video clip within the specified time is obtained according to the timestamp of the blunt needle production image;

[0128] Extracting frames from the damaged needle effect video clip at the second time interval to obtain a plurality of third acupuncture effect images;

[0129] The second number is obtained by counting the number of the blunt needle production images in all the third acupuncture effect images.

[0130] The present invention further obtains video clips within the corresponding time period by identifying the production images of blunt needles and damaged needles, and performs frame extraction processing according to a smaller second time interval, thereby generating more acupuncture effect images for analysis. By finely extracting frames from the key time period, more production effect images of the production process in this time period (the time period where problems may exist) can be accurately captured. Compared with the originally larger first time interval, the second time interval provides a higher time resolution, can generate a large number of test samples, improves the accuracy of problem identification, and reduces errors caused by inaccurate classification of a single sample. By increasing the number of samples, random fluctuations in the data can be effectively smoothed, reducing the impact of a certain sample abnormality or classification error on the overall test results.

[0131] See also Figure 1 In S70, when the first number exceeds a first preset threshold, identifying the blunt needle stage of the blunt needle production image, and adjusting the current parameter to the optimal control parameter of the corresponding stage according to the blunt needle stage;

[0132] In this embodiment, during the continuous monitoring of the production line on a certain day, the acupuncture effect video is obtained and frame extraction is performed on an hourly basis to obtain multiple first acupuncture effect images; among the multiple first acupuncture effect images, a certain image is identified as a blunt needle production image, and at this time, a blunt needle effect video clip within one hour before and after is obtained according to the timestamp of the image; that is, if the timestamp of the blunt needle production image is 19:00 on October 23, 2024, the time range of the video clip is from 18:00 on October 23, 2024 to 22:00 on October 23, 2024.

[0133] Those skilled in the art can set the second time interval according to specific circumstances. In this embodiment, the second time interval is 10 minutes.

[0134] The video clips are framed at 10-minute intervals to obtain 12 second acupuncture effect images;

[0135] The 12 second acupuncture effect images are sequentially input into the trained acupuncture needle state classification model for classification, and are divided into normal production images, blunt needle production images, and damaged needle production images; after statistics, 10 blunt needle production images, 1 normal production image, and 1 blunt needle production image are finally obtained. The first number is 10, and the first preset threshold in this embodiment is 5. If the first number exceeds the first preset threshold, the blunt needle stage will continue to be identified based on these 10 blunt needle production images, and the current parameters will be adjusted to the corresponding stage optimal control parameters according to the identified blunt needle stage;

[0136] Furthermore, the process of obtaining the optimal control parameters in the stage includes:

[0137] According to the predicted wear of the needle, the blunt needle stage of the blunt needle production image is identified; the blunt needle stage includes an initial stage, a middle stage and a late stage; the state of the needle is expressed as a percentage (30% to 70%), from initial wear to late wear. The blunt needle stage includes an initial stage (30% ≤ predicted wear of the needle <50%), a middle stage (50% ≤ predicted wear of the needle <60%) and a late stage (60% ≤ predicted wear of the needle ≤70%).

[0138] The control parameters and final product quality indicators corresponding to the initial stage, mid-term stage and late stage are respectively obtained from the operation and maintenance database to obtain initial product data, mid-term product data and late product data; the final product quality indicator is a comprehensive evaluation indicator of key quality parameters obtained after testing by the quality inspection department.

[0139] By correlation analysis, the key control parameters of the initial product data, the mid-term product data and the late product data are identified, the key control parameters including acupuncture speed, acupuncture pressure, acupuncture frequency, web tension and feed speed, etc., to obtain the initial key control parameters, the mid-term key control parameters and the late key control parameters;

[0140] For each stage, a regression model is constructed to analyze the relationship between the key control parameters of the stage and the quality index of the final product, and an initial regression model, a mid-term regression model and a late-term regression model are obtained;

[0141] Construct a comprehensive objective function J x :

[0142] J x =α·(q(x)-q target ) 2 +β·(t v +t p +t f +t T +t s );

[0143] Where: (q(x)-q target ) 2 is the quality index q(x) and the target product quality value q target The deviation between them, x represents the vector of control parameters, including needling speed v, needling pressure p, needling frequency f, web tension T and feeding speed s, α is the first weight, β is the second weight, t v To adjust the downtime caused by the needling speed v, t p To adjust the downtime caused by the acupuncture pressure p, t f To adjust the downtime caused by the acupuncture frequency f, t T The downtime caused by adjusting the web tension T, ts It is the downtime caused by adjusting the feed speed s; if the needling speed, needling pressure, needling frequency, web tension and feed speed do not need to be adjusted, the corresponding downtime is 0.

[0144] The following constraints are constructed to ensure that each key control parameter is within the specified range:

[0145]

[0146] Among them, v min is the lower limit of acupuncture speed, v max is the upper limit of acupuncture speed, p min is the lower limit of acupuncture pressure, p max is the upper limit of acupuncture pressure, f min is the lower limit of acupuncture frequency, f max is the upper limit of acupuncture frequency, T min is the lower limit of web tension, T max is the upper limit of web tension, s min is the lower limit of feed rate, s max is the upper limit of feed rate,

[0147] Minimize the comprehensive objective function J x .

[0148] The effects of different control parameter combinations are predicted by using the initial regression model, the mid-term regression model and the late-term regression model, and the optimal control parameters of each stage are searched through an optimization algorithm to obtain the initial optimal control parameters, the mid-term optimal control parameters and the late-term optimal control parameters.

[0149] The present invention divides the blunt needle stage into three stages: early, middle and late stages based on the prediction of needle wear, and constructs a regression model according to actual production data in each stage, thereby realizing dynamic adjustment of key control parameters (such as acupuncture speed, acupuncture pressure, etc.) at different stages. By optimizing the key control parameters of each stage, it is ensured that the needles can maintain the best process control at different wear stages during the production process, thereby ensuring the stability of product quality. In addition, the constructed comprehensive objective function not only takes into account the deviation between the product quality index and the target value, but also introduces the downtime factor related to the adjustment of the control parameters. The parameter combination of each stage is searched through the optimization algorithm, thereby improving the efficiency of obtaining the optimal parameters. This parameter optimization method can extend the service life of the needles while ensuring product quality, reduce unnecessary downtime, improve the overall automation and stability of the production line, and ultimately improve the consistency and production efficiency of the products.

[0150] See also Figure 1 In S80, when the second number exceeds a second preset threshold, the needles of the acupuncture machine are replaced.

[0151] In order to verify the effectiveness of the control method of the non-woven fabric production line, several groups of comparative experiments were designed. The initial state of the needle row of each production line was the same, all of which were newly replaced needle rows. They were continuously processed and produced for the same time. Except for the needling process, the control parameters and environmental parameters (workshop temperature and humidity) of the other production line processes were the same. Referring to Table 2, Experiment A is a control method for a non-woven fabric production line proposed in this embodiment. Through real-time monitoring of the production line, the needle state is accurately identified and the parameters are adjusted adaptively by detecting the needling effect image. It can optimize the needle use cycle without affecting the production efficiency, thereby reducing downtime, extending the life of the needle row, and maximizing the production line efficiency. Experiment B relies on the fixed cumulative use time of the needle to determine the replacement cycle, without considering the actual wear condition. Although the operation is simple, it often leads to the replacement of the needle too early or too late. The results show that this method has a long downtime and a short needle row life, which affects the production efficiency. Experiment C uses computer vision technology to collect needle row images in real time and calculate the needle wear through image analysis. When the real-time wear reaches the threshold, the needle row is replaced. Although experiment C can provide a certain degree of accuracy, it requires frequent shutdown for testing, which affects the overall production efficiency. Although the life of the needle row has been improved, the too frequent downtime significantly increases the maintenance time, thereby reducing the output of standard products. Through comparative experiments, it can be concluded that the non-woven fabric production line control method proposed in the present invention is superior to the existing detection method in terms of increasing the output of standard products, reducing downtime, and extending the life of the needle row.

[0152] Table 2. Comparative experimental data of different needle state recognition methods

[0153]

[0154] Furthermore, in order to verify the effectiveness of the control parameter adjustment method, another set of comparative experiments was designed. Except for the different control parameter adjustment methods, the remaining steps of Experiment 1, Experiment 2 and Experiment 3 are the same. Figure 1 ,Experiments 1, 2, and 3 differ only in the current parameter adjustment step of S70 (different control parameter adjustment strategies).

[0155] Experiment 1 is a control method based on stage optimal control parameter optimization proposed by the present invention. It adopts staged control parameter adjustment. By identifying the blunt needle stage of the blunt needle production image, according to historical data, each blunt needle stage (early, middle and late stages) corresponds to a set of optimal parameter adjustment data corresponding to the current stage, and the production parameters can be dynamically adjusted at each stage. Experiment 1 adopts Figure 1 The complete method flow is shown.

[0156] See also Figure 1S40 in Experiment 1 is classified according to the predicted wear amount, which is divided into normal production images (0≤needle predicted wear amount<30%), blunt needle production images (30%≤needle predicted wear amount≤70%) and damaged needle production images (70%<needle predicted wear amount≤100%); the blunt needle stage in subsequent Experiments 2 and 3 is also divided according to the same wear amount standard.

[0157] Experiment 2 is a traditional method based on fixed control parameters. In the operation and maintenance database, historical production data is screened according to the specified needle wear range (30% ≤ needle wear ≤ 70%) corresponding to the blunt needle stage to obtain screened historical data; the fixed control parameter is obtained based on the best control parameter combination in the screened historical data, which is a fixed value and will not change with the current production line situation. When the first number exceeds the first preset threshold, the current parameter is adjusted to the fixed control parameter obtained based on the screened historical data.

[0158] Experiment 3 is a control method based on the optimization of the optimal control parameters of the overall blunt needle stage. The specific design process is as follows:

[0159] The control parameters and final product quality indicators (without distinguishing between the early, middle and late stages of blunt needles) corresponding to the specified range of needle wear (30% ≤ needle wear ≤ 70%) are obtained from the operation and maintenance database to obtain blunt needle product data; the final product quality indicators are comprehensive evaluation indicators of key quality parameters obtained after testing by the quality inspection department.

[0160] By correlation analysis, key control parameters in the blunt needle product data are identified, the key control parameters including needling speed, needling pressure, needling frequency, web tension and feed speed, etc., to obtain a set of key control parameters;

[0161] Constructing a regression model, analyzing the relationship between the key control parameters and the final product quality index, and obtaining a blunt needle regression model (without distinguishing between the early, middle and late stages of blunt needle);

[0162] Starting from the construction of the comprehensive objective function, the subsequent steps are the same as method 1. Specifically, the key control parameters are optimized through the constructed comprehensive objective function, the effects of different control parameter combinations are predicted using a regression model, and the optimal control parameters are searched through an optimization algorithm. Finally, when the first quantity exceeds the first preset threshold, the current parameter is adjusted to the optimal control parameter. The optimal control parameter is the optimal control parameter based on the overall blunt needle stage, rather than the optimal control parameter based on the blunt needle stage sub-stage (blunt needle early, blunt needle middle and blunt needle late).

[0163] Table 3. Control parameter adjustment comparison experimental data table

[0164]

[0165] The comparative experimental results of different control parameter adjustment methods are shown in Table 3. In Experiment 1, three parameter adjustments were made in the blunt needle stage. For each identified stage, the production line control parameters were adjusted to the initial optimal control parameters, the mid-term optimal control parameters, and the late optimal control parameters. However, Experiments 2 and 3 only made one parameter adjustment, and their product qualification rates were lower than those of Experiment 1.

[0166] The present invention collects needling raw material videos and needling effect videos in real time, and combines them with a trained needle state classification model. On the premise of ensuring that the uniform fiber web meets the quality requirements, it focuses on the key link of needling, and realizes real-time monitoring of the needle state in the needling production process. By extracting frames of the needling effect image at a first time interval and classifying them, the production images of blunt needles and damaged needles can be accurately identified, and the control parameters can be automatically adjusted or the needle replacement can be triggered based on the relationship between the first number and the second number and the preset threshold. This production line control method avoids the lag of traditional methods that rely on the accumulated use time of the needles or manual inspection, and greatly improves the automation level and response speed of the production line. By dynamically adjusting the production control parameters at the blunt needle stage, it is possible to maintain the consistency of product quality at different wear stages, reduce unnecessary downtime and loss, extend the service life of the needles, and ensure production efficiency and the stability of the final product.

[0167] Embodiment 2:

[0168] Company B is a manufacturer of non-woven fabrics. The company has an automated non-woven production line. In order to further improve the quality of product production, Company B applied a control system for the non-woven production line, such as Figure 4 Shown include:

[0169] The raw material video acquisition module is used to collect the video of the uniform fiber web on the working production line before entering the needling machine to obtain the needling raw material video; the current parameters of the needling stage of the working production line are the preset standard control parameters;

[0170] A raw material quality inspection module is used to perform quality inspection based on the acupuncture raw material video, and if the raw material fails the quality inspection, a production line warning is triggered;

[0171] A needling effect video acquisition module, for acquiring a video of a uniform fiber web on the working production line after it passes the needling machine, to obtain a needling effect video if the quality inspection is passed;

[0172] An image processing module, configured to extract frames of the acupuncture effect video at a first time interval to obtain a plurality of first acupuncture effect images; and pre-process the acupuncture effect images to obtain standard acupuncture effect images;

[0173] A model classification module is used to input the standard acupuncture effect image into a trained needle state classification model for classification, and classify the image into a normal production image, a blunt needle production image, and a damaged needle production image; the needle state classification model is trained based on historical production data;

[0174] A quantity calculation module, configured to calculate a first quantity if the blunt needle production image is identified; the first quantity is the total quantity of the blunt needle production images identified within a specified time; and to calculate a second quantity if the damaged needle production image is identified; the second quantity is the total quantity of the damaged needle production images identified within the specified time;

[0175] A control adjustment module is used to identify the blunt needle stage of the blunt needle production image when the first number exceeds a first preset threshold, and adjust the current parameters to the corresponding stage optimal control parameters according to the blunt needle stage; when the second number exceeds a second preset threshold, replace the needles of the acupuncture machine.

[0176] Furthermore, performing the quality inspection according to the acupuncture raw material video includes:

[0177] Extract frames from the acupuncture raw material video at the first time interval to obtain a plurality of first acupuncture raw material images;

[0178] Preprocessing the acupuncture raw material image to obtain a standard acupuncture raw material image;

[0179] The standard needle punched raw material image is input into a trained fiber web quality assessment model to obtain a quality inspection result; the quality inspection result includes qualified and unqualified.

[0180] Furthermore, the fiber web quality assessment model includes:

[0181] The input layer is used to receive the input image and perform preliminary feature representation processing;

[0182] A feature extraction layer, used to extract key features of the input image to obtain basic features of the fiber web, wherein the basic features of the fiber web include texture features, edge features, uniformity features, and fiber distribution features of the fiber web; the feature extraction layer includes multiple convolutional layers and multiple pooling layers;

[0183] The multi-scale feature fusion layer is used to fuse the feature information of different scales from different convolutional layers, and obtain multi-scale fusion features through convolution operations of different receptive fields;

[0184] An attention mechanism layer, used to optimize feature representation through a channel attention mechanism and a spatial attention mechanism; the channel attention mechanism is used to weight different feature channels to obtain weighted channel features, and the spatial attention mechanism is used to focus on key areas of the image to obtain weighted spatial features;

[0185] A feature fusion layer, used to fuse the weighted channel features and the weighted spatial features optimized by the channel attention mechanism and the spatial attention mechanism to obtain a final fused feature;

[0186] The anomaly detection layer is used to detect anomalies of the final fusion features; it includes an autoencoder module, which is used to reconstruct the basic features of the fiber web and calculate the reconstruction error; when the reconstruction error exceeds a preset threshold, the input image is judged to be abnormal, and an abnormal warning message is directly output; if no abnormality is detected, the quality classification is continued;

[0187] Specifically, the anomaly detection layer includes an autoencoder module for reconstructing the basic features of the fiber web. First, the autoencoder module consists of an encoder and a decoder. The encoder compresses the basic features of the fiber web into a low-dimensional feature representation, and the decoder reconstructs the compressed low-dimensional features to generate reconstructed fiber web features. By calculating the reconstruction error between the original fiber web basic features and the reconstructed fiber web features, if the reconstruction error exceeds the preset anomaly threshold, the sample is judged to be abnormal and the abnormal warning information is directly output; if the reconstruction error does not exceed the preset anomaly threshold, the sample is considered normal, and its features are further input into the classification layer for quality classification.

[0188] The classification layer is used to classify the final fusion features through the fully connected layer to obtain the quality inspection result.

[0189] Specifically, the input layer receives image data and performs preliminary processing, and then extracts the key features of the fiber mesh through three sets of convolutional layers and pooling layers. The first convolutional layer uses 64 3x3 convolution kernels and outputs a feature map through the ReLU activation function. The size is halved after the first pooling. Then the feature map is further extracted and downsampled through the second convolutional layer and pooling layer, and 128 convolution kernels are used for feature extraction. The third convolutional layer uses 256 convolution kernels. Finally, a feature map of 32×32×256 is obtained through three poolings.

[0190] Further, if the quality inspection is passed, collecting the video of the uniform fiber web on the working production line after passing through the needle loom includes:

[0191] A second high-definition camera is installed just above the area between the needle loom and the winding device, and the shooting direction of the second high-definition camera is vertically downward.

[0192] Furthermore, the process of obtaining the first quantity and the second quantity includes:

[0193] When the blunt needle production image is identified, a blunt needle effect video clip within the specified time is obtained according to the timestamp of the blunt needle production image; the blunt needle effect video clip is framed at a second time interval to obtain a plurality of second acupuncture effect images; the first time interval is greater than the second time interval;

[0194] Counting the number of the blunt needle production images in all the second acupuncture effect images to obtain the first number;

[0195] When the damaged needle production image is identified, a damaged needle effect video clip within the specified time is obtained according to the timestamp of the blunt needle production image;

[0196] Extracting frames from the damaged needle effect video clip at the second time interval to obtain a plurality of third acupuncture effect images;

[0197] The second number is obtained by counting the number of the blunt needle production images in all the third acupuncture effect images.

[0198] Furthermore, the acquisition of the historical production data includes:

[0199] The historical production data includes historical acupuncture effect images and historical annotation data in different states; the historical annotation data includes needle state annotations corresponding to the historical acupuncture effect images; the different states include normal operation state, blunt needle state and damaged needle state;

[0200] respectively acquiring the historical acupuncture effect images and the historical annotation data in the normal operation state, the blunt needle state and the damaged needle state from the operation and maintenance database; the operation and maintenance database records the historical needle working condition images corresponding to the historical acupuncture effect images;

[0201] The acquisition of the historical annotation data includes:

[0202] Acquire a standard working condition image of the needle, wherein the standard working condition image of the needle is acquired after the needle is replaced as a whole but before it starts working;

[0203] Calculating the average wear amount of the needles by analyzing the historical working condition image of the needles and the standard working condition image of the needles;

[0204] The historical acupuncture effect image is annotated according to the average wear amount of the needle to obtain the historical annotated data.

[0205] Furthermore, inputting the standard acupuncture effect image into the trained acupuncture state classification model for classification includes:

[0206] The standard acupuncture effect image is input into the trained needle state classification model to obtain the predicted needle wear amount, and according to the predicted needle wear amount, the standard acupuncture effect image is divided into the normal production image, the blunt needle production image and the damaged needle production image.

[0207] Furthermore, the process of obtaining the optimal control parameters in the stage includes:

[0208] According to the predicted wear amount of the needle, the blunt needle stage of the blunt needle production image is identified; the blunt needle stage includes an early stage, a middle stage and a late stage;

[0209] Obtain control parameters and final product quality indicators corresponding to the initial stage, the mid-stage and the late stage from the operation and maintenance database to obtain initial product data, mid-stage product data and late product data;

[0210] By correlation analysis, the key control parameters of the stages in the initial product data, the mid-term product data and the late product data are identified to obtain the initial key control parameters, the mid-term key control parameters and the late key control parameters;

[0211] For each stage, a regression model is constructed to analyze the relationship between the key control parameters of the stage and the quality index of the final product, and an initial regression model, a mid-term regression model and a late-term regression model are obtained;

[0212] Construct a comprehensive objective function;

[0213] The effects of different control parameter combinations are predicted by using the initial regression model, the mid-term regression model and the late-term regression model, and the optimal control parameters of each stage are searched through an optimization algorithm to obtain the initial optimal control parameters, the mid-term optimal control parameters and the late-term optimal control parameters.

[0214] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A control method for a nonwoven fabric production line, characterized in that: include: Collect the video of the uniform fiber web on the production line before it enters the needle punching machine to obtain the video of the needle punching raw material; Performing quality inspection based on the acupuncture raw material video, and triggering a production line warning if the quality inspection fails; If the quality inspection is passed, a video of the uniform fiber web on the working production line after passing through the needling machine is collected to obtain a needling effect video; the current parameters of the needling stage of the working production line are the preset standard control parameters; Extracting frames from the acupuncture effect video at a first time interval to obtain a plurality of first acupuncture effect images; Preprocessing the acupuncture effect image to obtain a standard acupuncture effect image; Input the standard acupuncture effect image into a trained needle state classification model for classification, and classify it into a normal production image, a blunt needle production image, and a damaged needle production image; the needle state classification model is trained based on historical production data; If the blunt needle production image is identified, a first number is calculated; the first number is the total number of the blunt needle production images identified within a specified time; If the damaged needle production image is identified, a second number is calculated; the second number is the total number of the damaged needle production images identified within the specified time; When the first number exceeds a first preset threshold, identifying the blunt needle stage of the blunt needle production image, and adjusting the current parameter to the corresponding stage optimal control parameter according to the blunt needle stage; When the second number exceeds a second preset threshold, the needles of the needling machine are replaced.

2. The control method of a nonwoven fabric production line according to claim 1, characterized in that: The video of collecting the uniform fiber web on the working production line before entering the needle loom includes: The first high-definition camera is installed behind the web laying device and just above the conveyor belt in front of the needle loom, and the shooting direction of the first high-definition camera is vertically downward.

3. The control method of a nonwoven fabric production line according to claim 1, characterized in that: The quality inspection based on the acupuncture raw material video includes: Extract frames from the acupuncture raw material video at the first time interval to obtain a plurality of first acupuncture raw material images; Preprocessing the acupuncture raw material image to obtain a standard acupuncture raw material image; The standard needle punched raw material image is input into a trained fiber web quality assessment model to obtain a quality inspection result; the quality inspection result includes qualified and unqualified.

4. The control method of a nonwoven fabric production line according to claim 3, characterized in that: The fiber web quality assessment model includes: The input layer is used to receive the input image and perform preliminary feature representation processing; A feature extraction layer, used to extract key features of the input image to obtain basic features of the fiber web, wherein the basic features of the fiber web include texture features, edge features, uniformity features, and fiber distribution features of the fiber web; the feature extraction layer includes multiple convolutional layers and multiple pooling layers; The multi-scale feature fusion layer is used to fuse the feature information of different scales from different convolutional layers, and obtain multi-scale fusion features through convolution operations of different receptive fields; An attention mechanism layer, used to optimize feature representation through a channel attention mechanism and a spatial attention mechanism; the channel attention mechanism is used to weight different feature channels to obtain weighted channel features, and the spatial attention mechanism is used to focus on key areas of the image to obtain weighted spatial features; A feature fusion layer, used to fuse the weighted channel features and the weighted spatial features optimized by the channel attention mechanism and the spatial attention mechanism to obtain a final fused feature; The anomaly detection layer is used to detect anomalies of the final fusion features; it includes an autoencoder module, which is used to reconstruct the basic features of the fiber web and calculate the reconstruction error; when the reconstruction error exceeds a preset threshold, the input image is judged to be abnormal, and an abnormal warning message is directly output; if no abnormality is detected, the quality classification is continued; The classification layer is used to classify the final fusion features through the fully connected layer to obtain the quality inspection result.

5. The control method of a nonwoven fabric production line according to claim 1, characterized in that: If the quality inspection is passed, collecting the video of the uniform fiber web on the working production line after passing through the needle loom includes: A second high-definition camera is installed just above the area between the needle loom and the winding device, and the shooting direction of the second high-definition camera is vertically downward.

6. The control method of a nonwoven fabric production line according to claim 1, characterized in that: The process of obtaining the first quantity and the second quantity includes: When the blunt needle production image is identified, a blunt needle effect video clip within the specified time is obtained according to the timestamp of the blunt needle production image; the blunt needle effect video clip is framed at a second time interval to obtain a plurality of second acupuncture effect images; the first time interval is greater than the second time interval; Counting the number of the blunt needle production images in all the second acupuncture effect images to obtain the first number; When the damaged needle production image is identified, a damaged needle effect video clip within the specified time is obtained according to the timestamp of the blunt needle production image; Extracting frames from the damaged needle effect video clip at the second time interval to obtain a plurality of third acupuncture effect images; The second number is obtained by counting the number of the blunt needle production images in all the third acupuncture effect images.

7. The control method of a nonwoven fabric production line according to claim 1, characterized in that: The acquisition of the historical production data includes: The historical production data includes historical acupuncture effect images and historical annotation data in different states; the historical annotation data includes needle state annotations corresponding to the historical acupuncture effect images; the different states include normal operation state, blunt needle state and damaged needle state; respectively acquiring the historical acupuncture effect images and the historical annotation data in the normal operation state, the blunt needle state and the damaged needle state from the operation and maintenance database; the operation and maintenance database records the historical needle working condition images corresponding to the historical acupuncture effect images; The acquisition of the historical annotation data includes: Acquire a standard working condition image of the needle, wherein the standard working condition image of the needle is acquired after the needle is replaced as a whole but before it starts working; Calculating the average wear amount of the needles by analyzing the historical working condition image of the needles and the standard working condition image of the needles; The historical acupuncture effect image is annotated according to the average wear amount of the needle to obtain the historical annotated data.

8. The control method of a nonwoven fabric production line according to claim 1, characterized in that: Inputting the standard acupuncture effect image into the trained acupuncture state classification model for classification includes: The standard acupuncture effect image is input into the trained needle state classification model to obtain the predicted needle wear amount, and according to the predicted needle wear amount, the standard acupuncture effect image is divided into the normal production image, the blunt needle production image and the damaged needle production image.

9. The control method of a nonwoven fabric production line according to claim 1, characterized in that: The process of obtaining the optimal control parameters in this stage includes: According to the predicted wear amount of the needle, the blunt needle stage of the blunt needle production image is identified; the blunt needle stage includes an early stage, a middle stage and a late stage; Obtain control parameters and final product quality indicators corresponding to the initial stage, the mid-stage and the late stage from the operation and maintenance database to obtain initial product data, mid-stage product data and late product data; By correlation analysis, the key control parameters of the stages in the initial product data, the mid-term product data and the late product data are identified to obtain the initial key control parameters, the mid-term key control parameters and the late key control parameters; For each stage, a regression model is constructed to analyze the relationship between the key control parameters of the stage and the quality index of the final product, and an initial regression model, a mid-term regression model and a late-term regression model are obtained; Construct a comprehensive objective function; The effects of different control parameter combinations are predicted by using the initial regression model, the mid-term regression model and the late-term regression model, and the optimal control parameters of each stage are searched through an optimization algorithm to obtain the initial optimal control parameters, the mid-term optimal control parameters and the late-term optimal control parameters.

10. A control system for a nonwoven fabric production line, characterized in that: include: The raw material video acquisition module is used to collect the video of the uniform fiber web on the production line before it enters the needle punching machine to obtain the video of the needle punched raw material; The current parameters of the needling stage of the working production line are preset standard control parameters; A raw material quality inspection module is used to perform quality inspection based on the acupuncture raw material video, and if the raw material fails the quality inspection, a production line warning is triggered; A needling effect video acquisition module, for acquiring a video of a uniform fiber web on the working production line after it passes the needling machine, to obtain a needling effect video if the quality inspection is passed; An image processing module, configured to extract frames of the acupuncture effect video at a first time interval to obtain a plurality of first acupuncture effect images; and pre-process the acupuncture effect images to obtain standard acupuncture effect images; A model classification module is used to input the standard acupuncture effect image into a trained needle state classification model for classification, and classify the image into a normal production image, a blunt needle production image, and a damaged needle production image; the needle state classification model is trained based on historical production data; A quantity calculation module, used for calculating a first quantity if the blunt needle production image is identified; the first quantity is the total quantity of the blunt needle production images identified within a specified time; If the damaged needle production image is identified, a second number is calculated; the second number is the total number of the damaged needle production images identified within the specified time; A control adjustment module is used to identify the blunt needle stage of the blunt needle production image when the first number exceeds a first preset threshold, and adjust the current parameters to the corresponding stage optimal control parameters according to the blunt needle stage; when the second number exceeds a second preset threshold, replace the needles of the acupuncture machine.

Images