Defect classification system

Abstract

This invention provides a defect classification system that is simple to prepare or repair and can identify the cause of defects. The invention comprises: an imaging means 11, which uses an imaging device G to photograph paper P2 after it has passed through the drying section D and obtains the photographed image data; a detection means 12, which detects defects in paper P2 from the image data; an extraction means 13, which extracts the feature quantity of the defect; a calculation means 14, which calculates the confidence level of the defect cause item based on the feature quantity of the defect using a classification model 20 with pre-set reference feature quantities; and a display means 15, which displays the confidence level. The classification model 20 is formed by learning the reference feature quantities through machine learning based on the relationship between pre-stored defect feature quantities and a plurality of defect cause items.

Description

[Technical Field]

[0001] This disclosure relates to a defect classification system, and more specifically, to a defect classification system for classifying defect information obtained from defects in paper after passing through the dry part in the papermaking process into a predetermined plurality of defect cause items that match the defect cause items. [Previous Technology]

[0002] In the manufacture of paper, the following steps are performed: the pulp preparation step, in which the dried pulp sheet is defibered and the pulp is mixed with additives such as fillers or sizing agents to form a pulp dispersion; and the papermaking step, in which a papermaking machine is used to make paper from the pulp dispersion through various parts including the wire part and the press part, the wet part, the dry part, and the reel part.

[0003] However, from a production point of view, the paper (including the wet paper before drying) is transported at extremely high speeds during the papermaking process. Therefore, when defects such as foreign matter (pitch) adhering to the paper occur, the yield will be greatly reduced.

[0003] Therefore, various techniques have been developed in the papermaking process with the aim of minimizing the generation of defects.

[0004] For example, a monitoring system is known to include: an application device for applying a liquid medicine to a part of a paper machine that contacts the paper; an operation panel for setting the application conditions of the application device; a monitoring camera for monitoring the monitored part; and a control device connected to the monitoring camera via a network; the control device includes: a calculation unit for quantifying the state of the monitored part by binarization using images captured by the monitoring camera; a display unit for displaying detection data generated by the calculation unit by quantifying the change in the binarized value during operation relative to the binarized value of a constant state; and a memory unit for storing the detection data (see, for example, Patent Document 1).

[0004] [Previous Technical Documents]

[0004] [Patent Documents]

[0005] Patent Document 1: Japanese Patent Publication No. 6697132

[0006] In the monitoring system described in the above-mentioned Patent Document 1, although it is possible to identify the monitored object part that may be the cause of the defect and prevent the paper from having a defect caused by that reason, the monitoring camera needs to be set in the monitored object part with high temperature and high humidity, which will cause problems of preparation or maintenance.

[0007] The present invention was developed in view of the above-mentioned problems, and its purpose is to provide a defect classification system that is easy to prepare or maintain and can identify the cause of defects.

[0008] In order to solve the above-mentioned problems, the inventors, after careful examination, discovered that the defects occurring on the paper after passing through the drying section have characteristics. Furthermore, they discovered that the above-mentioned problems can be solved by having an extraction means, a calculation means, and a display means, and thus completed the present invention. The extraction means uses the paper after passing through the drying section as the object of photography and extracts the characteristic quantity of the defects. The calculation means calculates the confidence level using a classification model that has undergone machine learning. The display means displays the confidence level.

[0009] (1) The present invention is a defect classification system for classifying defect information into defect cause items that match among a plurality of defect cause items derived from a pre-set cause of defect, wherein the defect information is derived from defects in the paper after passing through the drying section in the papermaking process. The defect classification system comprises: an imaging means for capturing images of the paper after passing through the drying section and obtaining image data after the images are captured; a detection means for detecting paper defects in the image data; an extraction means for extracting the feature quantity of the defect; a calculation means for calculating the confidence level of the defect cause item based on the feature quantity of the defect and a classification model with pre-set reference feature quantities; and a display means for displaying the confidence level; the classification model is formed by learning the reference feature quantities through machine learning based on the relationship between the feature quantity of the defect and the plurality of defect cause items stored in advance.

[0010] (2) The present invention is a defect classification system as described in (1) above, which calculates the credibility of each defect cause item in the calculation means. The defect classification system further has a classification means that classifies defect information into defect cause items with the highest credibility among a plurality of credibility.

[0011] (3) The present invention is a defect classification system as described in (2) above, wherein the classification model is learned by machine learning based on the relationship between the feature quantity of a defect whose maximum confidence value is below a predetermined value and the defect cause item to which the defect is classified.

[0012] (4) The present invention is a defect classification system as described in any one of (1) to (3) above, wherein, in addition to the characteristic quantity of the defect, the defect information of the defect also includes the coordinate data of the defect when the paper has coordinates.

[0013] (5) The present invention is a defect classification system as described in any one of (1) to (4) above, wherein the defect cause item includes at least an item that sets the attachment of foreign matter in the drying section as the cause of the defect.

[0014] In the defect classification system of the present invention, the paper after passing through the drying section in the papermaking step is taken as the object of photography, so for example, a camera device can be installed after the drying section. In this way, it is no longer necessary to invade the interior of the device that performs the papermaking step in order to install a camera device, etc., so the preparation or maintenance for using the defect classification system can be carried out very safely and simply.

[0014] In addition, in the defect classification system, the paper after passing through the drying section is taken as the object of the photograph. Therefore, the defects that occur on the paper are caused by the part before the drying section of the paper preparation step or the paper making step.

[0015] In the defect classification system of the present invention, there are detection means and extraction means, so that defects occurring on the paper can be detected and the feature quantity can be extracted.

[0015] Furthermore, since there is a calculation method, the credibility of the defect cause item can be calculated from the characteristic quantity of the defect.

[0015] Furthermore, the credibility is displayed by means of display.

[0015] By means of these methods, the reliability of a defect relative to a defect cause item can be identified in a defect classification system. As a result, the defect can be classified into a defect cause item with higher reliability, and the cause of the defect can be identified.

[0016] For example, when the defect cause item includes at least the item that the defect is caused by the attachment of foreign matter in the drying section, by identifying the credibility of the item, it will be possible to identify whether the defect is caused by the drying section or by something other than the drying section.

[0017] In the defect classification system of the present invention, when there are more classification means, it will be possible to identify the defect cause items that include the defect.

[0017] Furthermore, by changing the baseline feature quantity of the classification model based on the defect information and the defect cause items after the defect is classified, the accuracy of classification can be improved.

[0018] In the defect classification system of the present invention, the classification model can be further trained through machine learning based on the relationship between the feature quantity of a defect whose maximum confidence value is below a predetermined value and the defect cause item to which the defect is classified. In this way, the accuracy of the confidence value calculated by the classification model can be further improved.

[0019] In the defect classification system of the present invention, when the defect information includes the coordinate data of the defect when the coordinates are provided on the paper, it is possible to infer a more detailed location of the part that is the cause of the defect from the coordinate data of the defect.

[0019] For example, when defects occur repeatedly along the length of the paper, there is a risk that foreign objects may adhere to the paper-guiding rollers, felt, canvas, or other annular devices. Furthermore, the diameter of the roller can be calculated from the distance between the repeated defects, when the foreign object is attached to the roller, and the position of the foreign object in the annular device can be calculated when the foreign object is attached to the annular device. [Simplified Explanation of the Diagram]

[0020] Figure 1 is a block diagram used to illustrate the structure of the defect classification system of this embodiment.

[0020] Figure 2 is a schematic side view illustrating the paper-making steps and the location of the camera device in the defect classification system of this embodiment.

[0020] Figure 3 is a diagram showing the defect cause items in the defect classification system of this embodiment, and the profile of the defects contained in the defect cause items.

[0020] Figure 4 is a flowchart showing the defect classification method of this embodiment.

Implementation Method

[0021] Preferred embodiments of the present invention will be described in detail below with reference to the drawings as needed. Furthermore, the same symbols are used for the same elements in the drawings, and repeated descriptions are omitted. In addition, unless otherwise specified, positional relationships such as up, down, left, and right are based on the positional relationships shown in the drawings. Moreover, the scale of the drawings is not limited to the scale shown.

[0022] The defect classification system of this embodiment is a program stored in a computer and is effective for papermaking machines equipped with camera devices. In addition, the computer is not limited to one unit, and multiple computers can be connected together by wired or wireless means for use.

[0023] Here, the above-mentioned camera device may be a video camera, a line sensor camera, an area sensor camera, etc.

[0023] As for the paper machine mentioned above, if it is a device capable of performing a series of paper-making steps, it is not particularly limited, and conventionally known devices may be appropriately used.

[0023] In the defect classification system, if the paper being addressed can be manufactured by a papermaking process, there are no particular limitations. For example, the paper can be printing paper, newspaper paper, coated paper, packaging paper, tissue paper, toilet paper or tissue paper, miscellaneous paper, thick board base paper, white board paper, colored board paper, paper tube base paper, building material base paper, various lining papers, etc.

[0024] In this specification, the term "defect" refers to the attachment of foreign matter (including the mixing of foreign matter), holes, partial paper breaks, water stains, oil stains, uneven pulp concentration, etc.

[0024] In addition, "foreign matter" includes not only adhesives, but also insects, paper dust, and residue. Furthermore, "adhesives" refers to adhesives derived from tape or paste or wood, which are inclusions in pulp, the raw material of paper.

[0024] Furthermore, the term "defect cause item" refers to an item created to distinguish matters that could be the cause of a defect. In addition, the term "multiple defect cause items" refers to a collection of defect cause items, wherein the causes of the defects in the defect cause items are different from each other.

[0024] In addition, the so-called "credibility" refers to the statistical standard (accuracy) of how credible the prediction is.

[0025] The defect classification system of this embodiment classifies the defect information obtained from the defects of the paper after passing through the drying section in the paper machine into the defect cause items that match the multiple defect cause items obtained from the pre-set defect cause items.

[0025] Figure 1 is a block diagram used to illustrate the structure of the defect classification system of this embodiment.

[0025] As shown in Figure 1, the defect classification system 10 includes: an imaging means 11, which uses an imaging device to photograph paper after it has passed through the drying section and obtains the photographed image data; a detection means 12, which is used to detect paper defects in the image data; an extraction means 13, which is used to extract the feature quantity of the defect; a calculation means 14, which calculates the confidence level of the defect cause item based on the feature quantity of the defect using a classification model with a pre-set reference feature quantity; a display means 15, which is used to display the confidence level; a classification means 16, which classifies the defect information into defect cause items with a confidence level of more than a pre-set predetermined value; and a memory means 17, which is used to store the image data.

[0025] In a defect classification system, by employing such means, the reliability of a defect relative to the defect cause item for each defect can be identified. As a result, as described below, the defect can be classified into a defect cause item with higher reliability, and at the same time, the cause of the defect can be identified.

[0026] The camera means 11 is a means of using a camera device to photograph paper after it has passed through the drying section and to obtain the photographed image data.

[0026] The camera means 11 is connected to the camera device via a wired or wireless network. Furthermore, the camera device is instructed to capture images of the paper after it has passed through the drying section, causing the camera device to begin capturing images.

[0026] On the other hand, the imaging means 11 receives image data captured by the imaging device. Furthermore, from the viewpoint of the size of the detected defect, as will be described later, the resolution of the image data is preferably 10 to 500 MHz.

[0026] Furthermore, the image data received by the camera means 11 is stored in the memory means 17.

[0027] Here, the camera device is installed on the downstream side of the paper conveying direction, which is closer to the drying section in the paper-making step.

[0027] Figure 2 is a schematic side view illustrating the paper-making steps and the location of the camera device in the defect classification system of this embodiment.

[0027] As shown in Figure 2, the papermaking process is carried out using a papermaking machine M.

[0027] The papermaking process includes: a wetting section W, which is composed of a thread section W1 and a pressing section W2. The thread section W1 is used to place a papermaking thread in which pulp is dispersed in water, so that excess water can fall off naturally. The pressing section W2 is used to pass the thread section W1 between a pair of pressing rollers and press it with the pressing rollers through a felt cloth; a drying section D, which is used to dry the paper by contacting the heated dryer roll D1 through a canvas K; a calender section C; and a reel section R.

[0027] In addition, a doctor blade D2 is abutted in the dryer drum D1 to remove foreign matter adhering to the surface of the dryer drum D1.

[0028] In the papermaking step, the pulp dispersion obtained in the paper preparation step is used to pass through the wetting section W to become the so-called wet paper P1, and after passing through the drying section D, it becomes the paper P2 after the wet paper P1 is dried.

[0028] Furthermore, after passing through the drying section D, the surface of the paper P2 is smoothed by passing through the calender section C, and the paper P2 is wound by the winding device (spool) and the like using the winding section R.

[0029] The camera device G is positioned above the paper P2 between the calender section C and the winding section R. This allows the surface of the paper P2 to be photographed by the camera device G while it is positioned between the calender section C and the winding section R.

[0029] Thus, in the defect classification system 10, it is no longer necessary to invade the interior of the device that performs the paper-making step in order to install the camera device G, etc., so preparation or maintenance can be carried out very safely and simply.

[0030] Referring back to Figure 1, the detection method 12 is a means for detecting defects occurring on paper in the captured image data.

[0030] Furthermore, as described above, since the camera device G is located downstream of the paper conveying direction, which is further downstream than the drying section D, the image data is based on the paper P2 that has passed through the drying section D. That is, the defects detected in the image data can be said to have occurred in the part of the paper preparation step or papermaking step before the drying section D.

[0031] The detection method 12 not only detects the defect, but also sets up virtual coordinates on paper to detect the coordinate data of the defect. In this way, the more detailed location of the part that caused the defect can be deduced from the coordinates of the defect.

[0031] The coordinate data of the defect is, for example, the paper feeding direction is set as the Y-axis and the paper width direction is set as the X-axis.

[0031] Here, regarding the location of the origin, for example, the front end of the paper P2 that has reached the starting position captured by the camera device G can be set as the origin of the Y-axis (Y=0), and the left end of the paper P2 can be set as the origin of the X-axis (X=0). That is, the left end of the front end of the paper P can be set as (X, Y) = (0, 0).

[0031] Furthermore, the value of Y is set to increase as the number moves upstream along the length of paper P2. The value of X is set to increase as the number moves from the left end to the right end.

[0031] Preferably, the distance between adjacent coordinates in the X-axis direction is 0.001 to 10 mm, and the distance between adjacent coordinates in the Y-axis direction is 0.001 to 10 mm. In this case, even if a very small defect occurs on the paper P2 within the visible range, it can be reliably detected.

[0032] Furthermore, in the length direction of the paper, when the same defect occurs repeatedly in a certain period, foreign matter will adhere to the ring-shaped tools such as the paper guide roller, felt, and canvas, and it is very likely to become the cause of the paper defect.

[0032] At this time, the distance between repeated defects can be calculated from the coordinates of the defects. Furthermore, when the foreign object is attached to the roller, this distance is equivalent to the circumference of the roller, so the diameter of the roller to which the foreign object is attached can be estimated, and when the foreign object is attached to the ring-shaped tool, the position of the foreign object can be estimated from this distance.

[0033] The defect detection is performed numerically by means of brightness measurement or RGB measurement. That is, when the numerical value in a predetermined coordinate system is abnormal, it can be said that the coordinate system has a defect.

[0033] Specifically, when the image is quantified by a brightness and darkness measurer, the image is set as a black and white binary image by grayscale, and its intensity is divided into 256 stages, for example, from 0 to 255, and then quantified.

[0033] In addition, when the value is converted by an RGB measurer, the specific color (e.g., blue) can be quantified.

[0033] Thus, the detection method 12 adopts brightness measurement or RGB measurement, so the presence or absence of defects can be easily identified.

[0033] In addition, the image of the defect, the coordinate data detected by the detection means 12, and the data of the defect after being quantified are stored in the memory means 17 as defect information.

[0034] Extraction means 13 is a means of extracting the characteristic quantity of a defect by using the data of the defect quantified by the detection means 12.

[0034] In extraction method 13, as a method for extracting feature quantities, filtering, wrapper, embedding and other methods can be appropriately adopted.

[0034] In the defect classification system, as a specific example of the extracted feature quantity, the size, shape, maximum value, minimum value, density difference, and density of the defect after numericalization can be listed.

[0034] As a deep learning model, for example, a CNN (Convolutional Neural Network) model can be used. As a network model, LeNet, AlexNet, VGG (VGG16, VGG19), GoogleNet, ResNet, etc. can be used.

[0034] In addition, the characteristic quantity of the defect is stored as defect information in the memory means 17.

[0035] The calculation method 14 is a means of calculating the credibility of each defect cause item based on the defect feature quantity extracted by the extraction method 13 and a classification model with pre-set benchmark feature quantities.

[0036] Figure 3 is an explanatory diagram showing the defect cause items in the defect classification system of this embodiment, and the portrait of the defects contained in the defect cause items.

[0036] As for the "adhesive" shown in Figure 3, the adhesion of adhesive originating from the roller or the canvas in the drying section is the cause of the defect. As for the "scraper blade", the adhesion of adhesive originating from the scraper blade in the drying section is the cause of the defect. "Inclusion" refers to the inclusion of foreign matter in the paper preparation step.

[0036] In addition, regarding "hole", the hole is the cause of the defect; regarding "oil stain", the oil is the cause of the defect; regarding "insect", the insect is the cause of the defect; and regarding "damage", the paper end is the cause of the defect.

[0037] The classification model is, for example, the cause of the defect shown in Figure 3, based on the defect cause item and the feature quantity extracted from the stored defect image contained in the defect cause item, and learns the baseline feature quantity through machine learning.

[0037] Furthermore, when data on the defect cause items and the feature quantities of the defects contained in those defect cause items are obtained again, the classification model can further learn from that data. That is, the baseline feature quantities can be changed. In this way, the accuracy of the reliability calculated by the classification model is further improved.

[0038] Here, in the defect classification system 10, as a defect cause item, it is preferable to include at least an item that takes the attachment of foreign matter in the drying section as the cause of the defect. At this time, by identifying the reliability of this item, it is possible to identify whether the cause of the defect is due to the drying section or due to something other than the drying section.

[0038] In addition, the adhesion of the adhesive shown in FIG3 occurs in the drying section, and the mixing occurs in the paper preparation step. Therefore, by distinguishing between these, it is possible to identify whether the defect is caused by the paper preparation step or by the drying section.

[0039] In the calculation method 14, the classification model calculates the confidence level for each defect cause item based on the feature quantity of the defect extracted by the extraction method 13.

[0039] Furthermore, the calculated confidence level is displayed by display means 15.

[0039] This will enable the identification of defect cause items with higher credibility or defect cause items with lower credibility.

[0039] In addition, as for the display means 15, a monitor, touch panel or the like can be used.

[0040] Returning to Figure 1, classification method 16 is a method for classifying a defect into the folder of the defect cause item with the highest confidence level among the calculated confidence levels of each defect cause item.

[0040] This will enable the identification of the defect cause item that contains the defect.

[0040] Furthermore, by changing the baseline feature quantity of the classification model based on the defect information and the defect cause items after the defect is classified, the accuracy of classification can be improved.

[0041] In classification method 16, it is preferable that when the confidence level of the maximum value is below a predetermined value, the classification model is further trained by machine learning based on the relationship between the feature quantity of the defect and the defect cause item to which the defect is classified. Furthermore, this classification operation can classify defects whose confidence level of the maximum value is below the predetermined value into defect cause items based on the confidence level of the maximum value, and then have a human review its accuracy. Alternatively, a human can classify defects whose confidence level of the maximum value is below the predetermined value into the correct defect cause item.

[0041] This improves the accuracy of the confidence level calculated by the classification model. In addition, the preset value can be set arbitrarily.

[0042] Next, the defect classification method using the defect classification system of this embodiment will be described.

[0042] Figure 4 is a flowchart showing the defect classification method of this embodiment.

[0042] As shown in Figure 4, the defect classification method includes the following steps: imaging step S11, which involves using the imaging device G to photograph the paper after it has passed through the drying section and obtaining the photographed image data; detection step S12, which is used to detect paper defects in the image data; extraction step S13, which is used to extract the feature quantity of the defect; calculation step S14, which involves using the classification model 20 with a pre-set reference feature quantity to calculate the confidence level of the defect cause item based on the feature quantity of the defect; and display step S15, which is used to display the confidence level; and classification step S16, which involves classifying the defect information into defect cause items whose confidence level is above a pre-set predetermined value.

[0043] In the defect classification method, at least the first computer C1 is equipped with a camera means 11 and a detection means 12, and at least the second computer C2 is equipped with an extraction means 13, a calculation means 14, a display means 15, a classification means 16, and a memory means 17. Furthermore, the first computer C1 and the second computer C2 are connected through the cloud or the like.

[0043] Therefore, in the defect classification method, the imaging step S11 and the detection step S12 are executed by the first computer C1, while the extraction step S13, the calculation step S14, the display step S15 and the classification step S16 are executed by the second computer C2.

[0043] Alternatively, multiple first computers C1 that perform camera steps S11 can be connected to this second computer C2.

[0044] According to the defect classification method, since it has the same detection step S12 and extraction step S13 as the above defect classification system, the defects that occur on the paper will be detected and their feature quantities will be extracted.

[0044] Furthermore, since step S14 is included, the confidence level relative to the cause of the defect can be calculated from the characteristic quantity of the defect.

[0044] Furthermore, the credibility is displayed by displaying step S15.

[0044] In this way, the reliability of a defect relative to a defect cause item can be identified in the defect classification method. As a result, the defect can be classified into a defect cause item with higher reliability, and at the same time, the cause of the defect can be identified.

[0044] Furthermore, since there is a classification step S16, the baseline feature quantity of the classification model can be changed according to the defect information of the defect and the defect cause item after the defect is classified, thereby improving the accuracy of classification.

[0045] Although the preferred embodiments of the present invention have been described above, the present invention is not limited to the above embodiments.

[0046] In the defect classification system 10 of this embodiment, although the camera device G is located above the paper P2 between the calender section C and the winding section R, it is not limited to this position. For example, it can be located below the paper P2, or it can be located above or below it.

[0047] In the defect classification system 10 of this embodiment, the papermaking step includes a wetting section W, a drying section D, a calender section C, and a winding section R, but is not limited to these.

[0047] For example, it may not have a calender section C and a coil section R.

[0047] In addition, a processing machine that cuts and recycles paper can be installed to replace the reel section R.

[0048] In the defect classification system 10 of this embodiment, although the camera device G is located above the paper P2 between the calender section C and the winding section R, it is not limited to that position if it is located downstream of the drying section D.

[0048] For example, it can also be disposed above the paper P2 between the drying section D and the calender section C.

[0048] Furthermore, if the camera device G is located downstream of the drying section D, other camera devices may also be provided. In this case, more detailed classification will be possible.

[0049] In the defect classification system 10 of this embodiment, although the origin is set with the front end of paper P2 as the origin of the Y-axis (Y=0) and the left end of paper P as the origin of the X-axis (X=0), it is not limited to this.

[0050] In the defect classification system 10 of this embodiment, although the detection of defects is quantified by means of brightness measurement or RGB measurement, it is not limited to this. It can also use shading processing, image retrieval (pattern search), edge detection, etc.

[0050] [Industry-specific availability]

[0051] The defect classification system of the present invention can be used as a system for classifying defect information obtained from the defects of the paper after passing through the drying section in the papermaking process into a predetermined plurality of defect cause items that match the defect cause items.

[0051] According to the defect classification system of the present invention, preparation or repair is relatively simple, and the cause of the defect can be identified.

Claims

1. A defect classification system for classifying defect information into a plurality of defect cause items corresponding to a predetermined cause of the defect, wherein the defect information is derived from defects in paper after passing through the drying section in the papermaking step following the paper preparation step. The defect classification system comprises: an imaging means for capturing images of the paper after passing through the drying section using a camera device, and obtaining image data of the captured images; a detection means for detecting paper defects in the aforementioned image data; an extraction means for extracting the feature quantity of the defect; a calculation means for calculating the confidence level of the aforementioned defect cause item based on the feature quantity of the defect using a classification model with predetermined reference feature quantities, and calculating the aforementioned confidence level for each of the aforementioned defect cause items; a display means for displaying the aforementioned confidence level; and a classification means for classifying the aforementioned defect information into the aforementioned defect cause item whose maximum confidence level is among a plurality of the aforementioned confidence levels. The aforementioned classification model is based on the relationship between pre-stored defect feature quantities and the aforementioned plurality of defect cause items, and is learned by machine learning to learn the aforementioned baseline feature quantities. The aforementioned defect detection is achieved through brightness measurement and RGB measurement. The aforementioned brightness measurement is achieved by setting a grayscale binary image as white and black, and quantifying its intensity. The aforementioned RGB measurement is achieved by quantifying specific colors in RGB. The feature quantities of the aforementioned defects are extracted at least from the aforementioned defect size, the aforementioned defect shape, and the maximum value, minimum value, intensity difference, and intensity of the defect quantified by the aforementioned brightness measurement or the aforementioned RGB measurement. The aforementioned defects include at least the following: the adhesion of adhesives originating from the rollers or canvas in the aforementioned drying section; the adhesion of adhesives originating from the doctor blade in the aforementioned drying section; the mixing of foreign matter in the aforementioned paper preparation step; holes; oil adhesion; insect adhesion; and paper end damage.

2. The defect classification system as described in claim 1, wherein, The aforementioned classification model is based on the relationship between the feature quantity of the aforementioned defect below a predetermined value with the confidence level of the aforementioned maximum value and the aforementioned defect cause item to which the defect is classified, and is further learned through machine learning.

3. A defect classification system as described in claim 1 or 2, wherein, In the defect information of the aforementioned defect, in addition to the characteristic quantity of the defect, the coordinate data of the defect when the aforementioned paper has coordinates is also included. When the conveying direction of the aforementioned paper is set as the Y-axis, the width direction of the paper is set as the X-axis, and the left end of the front end of the aforementioned paper is set as the origin, the distance between adjacent coordinates in the X-axis direction is 0.001 to 10 mm, and the distance between adjacent coordinates in the Y-axis direction is 0.001 to 10 mm.

Images