A deep learning-based fiber impurity detection method

Abstract

This invention discloses a deep learning-based fiber impurity detection method, belonging to the field of fiber detection technology. The specific method includes: Step 1: Acquiring current fiber production process data, collecting key data based on the acquired production process data, and supplementing the current fiber production line with a sorting device based on the collected key data; Step 2: Acquiring several high-definition images of the sorting device during normal operation, processing the acquired high-definition images to obtain several sets of detection training sets; Step 3: Constructing a deep neural network for fiber impurity detection, then training and testing the deep neural network using the obtained training and test sets, and marking the successfully tested deep neural network as the impurity detection model; Step 4: Acquiring fiber images passing through the sorting device, marking them as target images, inputting the target images into the impurity detection model to obtain the corresponding detection results.

Description

Technical Field

[0001] This invention belongs to the field of fiber detection technology, specifically a fiber impurity detection method based on deep learning. Background Technology

[0002] During the production process, fibers often contain impurities due to various reasons. However, most current fiber impurity detection methods rely on weighing, which has significant errors, is complex to operate, requires a lot of manual labor, and cannot achieve automated measurement, resulting in a huge waste of human and material resources. Therefore, this invention provides a fiber impurity detection method based on deep learning to solve the problem of insufficient intelligence in current fiber detection. Summary of the Invention

[0003] To address the problems of the above-mentioned solutions, this invention provides a fiber impurity detection method based on deep learning.

[0004] The objective of this invention can be achieved through the following technical solutions:

[0005] A deep learning-based fiber impurity detection method, the specific method includes:

[0006] Step 1: Obtain the current fiber production process data, collect key data based on the obtained production process data, and supplement the current fiber production line with sorting equipment based on the collected key data;

[0007] Step 2: Acquire several high-definition images of the splitting device during normal operation, process the acquired high-definition images, and obtain several sets of detection training sets;

[0008] Step 3: Construct a deep neural network for fiber impurity detection, and then train and test the deep neural network using the obtained training and test sets. Mark the deep neural network that successfully passes the test as the impurity detection model.

[0009] Step 4: Collect fiber images passing through the column separator, mark them as target images, input the target images into the impurity detection model, and obtain the corresponding detection results; the detection results include qualified, unqualified, and the impurity area corresponding to the unqualified.

[0010] Furthermore, methods for collecting key data based on the acquired production process data include:

[0011] Identify fiber output, equipment, and production efficiency data in the production process data; analyze the equipment data to obtain equipment ratings; set the efficiency requirements for the current fiber production line based on the production efficiency data; and integrate fiber output, equipment ratings, and efficiency requirements into key data.

[0012] Furthermore, methods for supplementing the current fiber production line with sorting equipment based on the collected key data include:

[0013] Establish an equipment design library, identify the design evaluation value of each equipment design scheme in the library, and label the design evaluation value as i, where i = 1, 2, ..., n, and n is a positive integer; label the fiber output value, equipment evaluation value, and efficiency requirement value in the key data as A, B, and C, respectively, and label the fiber output value, equipment evaluation value, and efficiency requirement value in the design evaluation value as Pi, Li, and Ki, respectively; calculate the design value of the key data relative to each design evaluation value according to the design formula, obtain the corresponding equipment design scheme based on the calculated design value, design the corresponding column equipment based on the obtained equipment design scheme, and supplement the column equipment to the corresponding position in the fiber production line.

[0014] Furthermore, the design formula is as follows:

[0015] w i = [1-exp(-α1×|A-Pi|)]×[1-exp(-α2×|B-Li|]×[1-exp(-α3×|C-Ki|)], where α1, α2, and α3 are adjustment coefficients with values ​​ranging from [0 to 1].

[0016] Furthermore, construct the fitness function. Determine the optimal solutions for α1, α2, and α3, where β is a constant.

[0017] Furthermore, methods for establishing an equipment design library include:

[0018] Acquire all key data currently available in the fiber production plant, convert the key data into coordinates, mark them as key coordinates, map the key coordinates into a coordinate space, set up a point radius table and a merging model, and merge and mark the key coordinates in the coordinate space according to the point radius table and the merging model to obtain key merging areas. Set up corresponding equipment design schemes according to the key data corresponding to each key merging area, and label them with the corresponding key merging area. Establish a first database, send the equipment design schemes to the first database for storage, and mark the current first database as the equipment design library.

[0019] Furthermore, methods for processing the acquired high-resolution images include:

[0020] Label the impurity regions and qualified regions in the high-resolution image, and mark the corresponding position information on the impurity regions and qualified regions. Segment the high-resolution image based on the preset size to obtain several segmented images. Identify the label information on the segmented images and integrate them into a training set and a test set.

[0021] Furthermore, it also includes:

[0022] Step 5: Establish a fiber production line model, set up dynamic modules, and integrate the dynamic modules and the fiber production line model into a dynamic tracking model;

[0023] Step 6: Obtain the impurity areas corresponding to the non-conforming test results, input the impurity areas into the dynamic tracking model, and mark the fiber areas with impurities;

[0024] Step 7: Identify the marked areas in the dynamic tracking model in real time and sort the fibers corresponding to the marked areas.

[0025] Compared with the prior art, the beneficial effects of the present invention are:

[0026] By setting up a sorting device, a large number of fibers are sorted and diverted on the production line before fiber impurity detection, spreading the fibers out as much as possible to facilitate the acquisition of internal images of the fibers; this enables internal fiber detection, making the detection more comprehensive and solving the problem of the current problem of limited detection due to the large number of fibers on the production line; by setting up a dynamic tracking model, it is easy to track and sort fibers with impurities, realizing intelligent detection of production fibers and subsequent impurity treatment. Attached Figure Description

[0027] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0028] Figure 1 This is a flowchart of the method of the present invention. Detailed Implementation

[0029] The technical solution of the present invention will be clearly and completely described below with reference to the embodiments. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0030] like Figure 1 As shown, a deep learning-based fiber impurity detection method is described, which includes the following specific steps:

[0031] Step 1: Obtain the current fiber production process data, collect key data based on the obtained production process data, and supplement the current fiber production line with sorting equipment based on the collected key data;

[0032] Production process data refers to the data of the current fiber production line, such as equipment information and output information.

[0033] Methods for collecting key data based on acquired production process data include:

[0034] Identify fiber output values, equipment data, and production efficiency data in the production process data. Equipment data refers to the data of equipment used in the fiber production line, such as production equipment and conveying equipment. Analyze the equipment data to obtain equipment evaluation values, and set the efficiency requirements for the current fiber production line based on the production efficiency data. Integrate fiber output values, equipment evaluation values, and efficiency requirements into key data.

[0035] The method for analyzing device data is to build an intelligent model based on a CNN or DNN network, then train it by setting a training set, and finally analyze the current device data to obtain a device rating. The specific building and training process is common knowledge in this field, so it will not be described in detail.

[0036] Setting the efficiency requirement value of the current fiber production line based on production efficiency data means that the efficiency of the partitioning equipment must meet the efficiency requirements of the current fiber production line. The specific setting method for the efficiency requirement value of the partitioning equipment will be discussed and set by the expert group, or an efficiency matching table can be directly established for corresponding matching settings.

[0037] Methods for supplementing the current fiber production line with sorting equipment based on collected key data include:

[0038] Establish an equipment design library and identify the design evaluation values ​​of each equipment design scheme in the library. The design evaluation values ​​include the fiber output value, equipment evaluation value, and efficiency requirement value for the corresponding equipment design scheme. The design evaluation value is labeled as i, where i = 1, 2, ..., n, and n is a positive integer. The fiber output value, equipment evaluation value, and efficiency requirement value in the key data are labeled as A, B, and C, respectively. The fiber output value, equipment evaluation value, and efficiency requirement value in the design evaluation value are labeled as Pi, Li, and Ki, respectively. The design value of the key data relative to each design evaluation value is calculated according to the design formula. Based on the calculated design value, the corresponding equipment design scheme is obtained. The corresponding column equipment is designed according to the obtained equipment design scheme and added to the corresponding position in the fiber production line.

[0039] The design formula is:

[0040] w i =[1-exp(-α1×|A-Pi|)]×[1-exp(-α2×|B-Li|]×[1-exp(-α3×|C-Ki|)]

[0041] Where α1, α2, and α3 are adjustment coefficients, and their values ​​range from [0, 1]. The corresponding equipment design scheme is selected based on the design value.

[0042] Methods for determining the optimal solutions for α1, α2, and α3 include:

[0043] Constructing the fitness function

[0044] Where β is a constant, and different values ​​correspond to different fitness functions, such as 1, 2, 3, etc. When b = min{wi} and Fit(wij) = 0.5, a is the distance from wi to min{wi}. The initial values ​​of a and b can be set manually. Usually, the values ​​of a and b are continuously corrected through the next generation of crossover and mutation evolution using a genetic algorithm to obtain the value of wi. Substituting these values ​​into the design formula wi optimizes the initial adjustment coefficients α1, α2, and α3, thus obtaining the optimal solution for the adjustment coefficients. Alternatively, the coefficients and initial values ​​can be optimized using a genetic algorithm. Iterative calculations can be performed using the genetic algorithm toolbox built into MATLAB software.

[0045] The fiber sorting equipment is used to sort and divert a large number of fibers on the production line before fiber impurity detection, spreading the fibers out as much as possible. For example, setting up multiple sorting channels can divert and spread out a large number of fibers, making it easier to collect internal images of the fibers. This enables internal fiber detection, making the detection more comprehensive and solving the problem that the detection of fibers is relatively one-sided due to the large number of fibers on the production line.

[0046] Methods for establishing an equipment design library include:

[0047] Acquire all key data currently available in the fiber production plant, convert the key data into coordinates, mark them as key coordinates, map the key coordinates into a coordinate space, set up a point radius table and a merging model, and merge and mark the key coordinates in the coordinate space according to the point radius table and the merging model to obtain key merging areas. Set up corresponding equipment design schemes according to the key data corresponding to each key merging area, and label them with the corresponding key merging area. Establish a first database, send the equipment design schemes to the first database for storage, and mark the current first database as the equipment design library.

[0048] The point radius table is a set of constraints established by the expert group to limit the merging of key coordinates. The corresponding range is set from the perspective of actually compiling the equipment design plan, and then the corresponding point radius table is set. The specific setting process is common knowledge in this field, so it will not be described in detail.

[0049] The merged model is built on a CNN or DNN network and then trained by setting a training set. The specific building and training process is common knowledge in this field, so it will not be described in detail.

[0050] The equipment design schemes for each key merged area are set up through direct discussion and configuration by the expert group based on the key data.

[0051] Step 2: Acquire several high-definition images of the splitting device during normal operation, process the acquired high-definition images, and obtain several sets of detection training sets;

[0052] Methods for processing the acquired high-resolution images include:

[0053] The impurity and qualified regions in the high-definition image are marked manually, and the corresponding location information is marked on the impurity and qualified regions, that is, the specific location in the splitting device that the location on the high-definition image corresponds to; the high-definition image is segmented based on a preset size to obtain several segmented images. The preset size refers to the size set by the expert group according to the needs of subsequent deep neural network training; the marked information on the segmented images is identified and integrated into training set and test set.

[0054] Step 3: Construct a deep neural network for fiber impurity detection, and then train and test the deep neural network using the obtained training and test sets. Mark the deep neural network that successfully passes the test as the impurity detection model.

[0055] The specifics of how to construct, train, and test deep neural networks are common knowledge in this field, and therefore will not be described in detail; the impurity detection model is used to identify the corresponding contamination areas based on the input fiber image;

[0056] Step 4: Collect fiber images passing through the column separator, mark them as target images, input the target images into the impurity detection model, and obtain the corresponding detection results. The detection results include qualified, unqualified, and the impurity area corresponding to the unqualified.

[0057] In another embodiment, it also includes:

[0058] Step 5: Establish a fiber production line model, which is a three-dimensional data model based on the fiber production line, including the column equipment; set up a dynamic module and integrate the dynamic module and the fiber production line model into a dynamic tracking model;

[0059] The dynamic module is used to dynamically represent the movement of fibers in the fiber production line model. It is implemented based on IoT devices. The specific setting method is common knowledge in this field, so it will not be described in detail. The integrated dynamic tracking model can dynamically represent the transmission of fibers and can track and mark the fibers at the corresponding positions in real time based on the received position information.

[0060] Step 6: Obtain the impurity areas corresponding to the non-conforming test results, input the impurity areas into the dynamic tracking model, and mark the fiber areas with impurities;

[0061] Step 7: Identify the marked areas in the dynamic tracking model in real time and sort the fibers corresponding to the marked areas; remove fibers with impurities from the production line and clean the fibers with impurities later.

[0062] The above formulas are all numerical calculations after removing dimensions. The formulas are obtained by software simulation based on a large amount of data and are closest to the real situation. The preset parameters and preset thresholds in the formulas are set by those skilled in the art according to the actual situation or obtained by simulation based on a large amount of data.

[0063] The working principle of this invention is as follows: Acquire current fiber production process data; identify fiber output, equipment data, and production efficiency data within the production process data; analyze the equipment data to obtain equipment evaluation; set the efficiency requirement value for the current fiber production line based on the production efficiency data; integrate the fiber output, equipment evaluation, and efficiency requirement value into key data; supplement the current fiber production line with sorting equipment based on the key data; acquire several high-definition images of the sorting equipment during normal operation; mark the impurity and qualified areas in the high-definition images; mark the corresponding position information on the impurity and qualified areas; segment the high-definition images based on a preset size to obtain several segmented images; identify the marking information on the segmented images; and integrate them into a training set. The process involves: constructing a deep neural network for fiber impurity detection using the obtained training and test sets; training and testing the deep neural network using the obtained training and test sets; marking the successfully tested deep neural network as the impurity detection model; acquiring fiber images passing through the sorting device and marking them as target images; inputting the target images into the impurity detection model to obtain the corresponding detection results; establishing a fiber production line model, setting up a dynamic module, and integrating the dynamic module and the fiber production line model into a dynamic tracking model; obtaining the impurity areas corresponding to the non-conforming detection results, inputting the impurity areas into the dynamic tracking model, and marking the fiber areas with impurities; identifying the marked areas in the dynamic tracking model in real time and sorting the fibers corresponding to the marked areas.

[0064] The above embodiments are only used to illustrate the technical methods of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical methods of the present invention without departing from the spirit and scope of the technical methods of the present invention.

Claims

1. A deep learning-based fiber impurity detection method, characterized by, Specific methods include: Step 1: Obtain the current fiber production process data, collect key data based on the obtained production process data, and supplement the current fiber production line with sorting equipment based on the collected key data; Step 2: Acquire several high-definition images of the splitting device during normal operation, process the acquired high-definition images, and obtain several sets of detection training sets; Step 3: Construct a deep neural network for fiber impurity detection, and then train and test the deep neural network using the obtained training and test sets. Mark the deep neural network that successfully passes the test as the impurity detection model. Step 4: Acquire fiber images passing through the column separator, mark them as target images, input the target images into the impurity detection model, and obtain the corresponding detection results; the detection results include qualified, unqualified, and the impurity area corresponding to the unqualified. Methods for supplementing the current fiber production line with sorting equipment based on collected key data include: Establish an equipment design library, identify the design evaluation value of each equipment design scheme in the library, and label the design evaluation value as i, where i = 1, 2, ..., n, and n is a positive integer; label the fiber output value, equipment evaluation value, and efficiency requirement value in the key data as A, B, and C, respectively, and label the fiber output value, equipment evaluation value, and efficiency requirement value in the design evaluation value as Pi, Li, and Ki, respectively; calculate the design value of the key data relative to each design evaluation value according to the design formula, obtain the corresponding equipment design scheme based on the calculated design value, design the corresponding column equipment based on the obtained equipment design scheme, and supplement the column equipment to the corresponding position in the fiber production line; The design formula is: , a1, a2, a3 are adjustment factors, with a range of [0, 1]. 2.The fiber impurity detection method based on deep learning according to claim 1, characterized in that, Methods for collecting key data based on acquired production process data include: Identify fiber output, equipment, and production efficiency data in the production process data; analyze the equipment data to obtain equipment ratings; set the efficiency requirements for the current fiber production line based on the production efficiency data; and integrate fiber output, equipment ratings, and efficiency requirements into key data.

3. The fiber impurity detection method based on deep learning according to claim 1, characterized in that, Construct fitness function The optimal solution of α1, α2, α3 is determined, β is a constant, the initial value of a and b is artificially set, and the value of a and b is continuously corrected by the next generation of genetic algorithm crossover and mutation evolution, so as to obtain the value of w i , which is brought into the design formula w i to optimize the initial adjustment coefficients α1, α2, α3.

4. The fiber impurity detection method based on deep learning according to claim 1, characterized in that, Methods for establishing an equipment design library include: Acquire all key data currently available in the fiber production plant, convert the key data into coordinates, mark them as key coordinates, map the key coordinates into a coordinate space, set up a point radius table and a merging model, and merge and mark the key coordinates in the coordinate space according to the point radius table and the merging model to obtain key merging areas. Set up corresponding equipment design schemes according to the key data corresponding to each key merging area, and label them with the corresponding key merging area. Establish a first database, send the equipment design schemes to the first database for storage, and mark the current first database as the equipment design library.

5. The fiber impurity detection method based on deep learning according to claim 1, characterized in that, Methods for processing the acquired high-resolution images include: Label the impurity regions and qualified regions in the high-resolution image, and mark the corresponding position information on the impurity regions and qualified regions. Segment the high-resolution image based on the preset size to obtain several segmented images. Identify the label information on the segmented images and integrate them into a training set and a test set.

6. The fiber impurity detection method based on deep learning according to claim 1, characterized in that, Also includes: Step 5: Establish a fiber production line model, set up dynamic modules, and integrate the dynamic modules and the fiber production line model into a dynamic tracking model; Step 6: Obtain the impurity areas corresponding to the non-conforming test results, input the impurity areas into the dynamic tracking model, and mark the fiber areas with impurities; Step 7: Identify the marked areas in the dynamic tracking model in real time and sort the fibers corresponding to the marked areas.

Images