Laminated film pressing system of flexible dielectric material sheet and visual neural network positioning method of laminated film pressing system

Abstract

The invention belongs to the technical field of ceramic capacitor production, and particularly relates to a laminated film pressing system of a flexible dielectric material sheet and an optic neural network positioning method of the laminated film pressing system. The method is applied to alternate alignment and hot pressing of the material sheets, a findContours recognition circle and a TensorFlow neural network are combined to accurately recognize and optimize the material sheets, the circle center of a circumcircle is connected with the circle centers of two circles, the circle centers of which are closest to and farthest from the starting point of coordinates, and the bisector angle between the line and the transverse axis and the bisector angle between the line and the longitudinal axis are calculated to serve as the eccentric angle, so that alternate stacking of the flexible dielectric material sheets is achieved. Certain pressure is applied to the alternate flexible dielectric material sheets by operating related mechanisms and the flexible dielectric material sheets are heated, so that tight combination of the flexible dielectric material sheets is realized, and the problems of complex working procedures, inaccurate electrode alignment, low available alignment area efficiency and inconstant pressure existing in the prior art are effectively solved; the method can be applied to the field of dielectric chip type multilayer ceramic capacitor lamination, and solves the problem of internal stress caused by inaccurate electrode utilization rate and alignment.

Description

Technical Field

[0001] The invention belongs to the technical field of ceramic capacitor production, and in particular relates to a lamination and lamination system of a flexible dielectric sheet and a visual neural network positioning method thereof. Background Art

[0002] Flexible dielectric sheet is one of the necessary materials in the production process of dielectric sheet multilayer ceramic capacitors (MLCC, Multilayer Ceramic Capacitor). It is a processed dielectric film that is coated with rectangular conductive slurry electrodes and dried.

[0003] In the production process of dielectric chip multilayer ceramic capacitors, one process is to align the dielectric film material coated with rectangular electrodes in the order of electrode crossing, and another process is to compact the aligned film body.

[0004] Currently, most people use manual pressure film alignment, which has a complicated working procedure, inaccurate electrode alignment, low efficiency of available alignment area, unstable pressure, and long-term repeated alignment of electrodes, which can easily cause certain harm to workers' health.

[0005] In view of this, the inventor hopes to provide a lamination and pressing system for flexible dielectric sheets and a visual neural network positioning method thereof, which can solve the problems of low efficiency of manual alignment and unstable pressure. Summary of the invention

[0006] The purpose of the present invention is to overcome the above-mentioned problems existing in the conventional technology and to provide a lamination system for a flexible dielectric sheet and a visual neural network positioning method thereof.

[0007] In order to achieve the above technical objectives and the above technical effects, the present invention is implemented through the following technical solutions:

[0008] The present invention provides a lamination and film pressing system for flexible dielectric sheets, comprising a workbench and a controller, wherein a flexible dielectric sheet placement area and a flexible dielectric sheet stacking area are arranged near the center of the workbench, two lateral displacement drive components are symmetrically installed on both sides of the workbench, and the movable parts of the two lateral displacement drive components jointly support a longitudinal displacement drive component, and the movable part of the longitudinal displacement drive component is installed with a vertical displacement drive component, and the movable end of the vertical displacement drive component is installed with a rotation drive device, and the movable part of the rotation drive device is connected with a flexible dielectric sheet adsorption plate, and a CCD camera is also installed on the workbench, and a heating device capable of heating the flexible dielectric sheet stacking area is arranged on the workbench, and the lateral displacement drive component, the longitudinal displacement drive component, the vertical displacement drive component, the rotation drive device, the flexible dielectric sheet adsorption plate, the CCD camera, and the heating device are respectively connected to the controller.

[0009] Furthermore, in the above-mentioned lamination and film pressing system, the controller is capable of receiving or calling available data output by the visual area, controlling the actions of the lateral displacement drive component, the longitudinal displacement drive component, the vertical displacement drive component, and the rotation drive device, and moving the flexible dielectric sheet adsorption plate to the flexible dielectric sheet placement area. The flexible dielectric sheet adsorption plate can suck up the flexible dielectric sheet, transport it to the stacking area, and stack it alternately.

[0010] Furthermore, in the above-mentioned lamination and film pressing system, after the flexible dielectric sheet stacking area is heated by the heating device, the flexible dielectric sheets stacked thereon can be more tightly bonded.

[0011] Furthermore, in the above-mentioned lamination and film pressing system, the lateral displacement drive assembly and the longitudinal displacement drive assembly are linear drives based on screw transmission.

[0012] Furthermore, in the above-mentioned laminating and film pressing system, infrared anti-collision limiters for blocking deceleration are installed on the lateral displacement drive assembly and the longitudinal displacement drive assembly.

[0013] Furthermore, in the above lamination and film pressing system, a temperature sensor capable of detecting the temperature at the flexible dielectric sheet stacking area is arranged on the workbench, and the temperature sensor is connected to the controller.

[0014] The present invention also provides a visual neural network positioning method for a lamination film lamination system, comprising the following steps:

[0015] S1, obtaining an image of a flexible dielectric sheet of a deceleration mechanism in a visual positioning area;

[0016] S2, identifying the acquired flexible dielectric sheet image, and performing binarization processing on the image to obtain a binary image;

[0017] S3, using findContours function to detect the circular contour of the obtained binary image, and fitting a circumscribed circle based on the circular contour;

[0018] S4. Based on the convolutional neural network, the data with a circle accuracy greater than 0.99 is obtained, the coordinates of the centers of all circles fitted in findContours are saved, and the distances between the four vertices of the rectangular matrix and the four centers of the circles are compared;

[0019] S5. When stacking for the first time, take the coordinates of the center of the circumscribed circle closest to the coordinate starting point as the additional moving distance of the guide rail, connect the centers of the two circles whose centers are closest to the coordinate starting point and farthest from the coordinate starting point, and calculate the angle between this line and the bisector of the horizontal axis and the vertical axis as the eccentricity angle;

[0020] S6, inputting the additional moving distance and eccentric coordinates output by the computer into the controller, controlling the rotation and movement of the guide rail and the film pressing mechanism to the stacking area, and stacking the films alternately; wherein, in step S4, the distances between the four vertices of the rectangular array and the four centers of the circles are one pair longer and one pair shorter, and alternate stacking means, for example, selecting the longer pair as the initialization side for the first time, and selecting the shorter side as the initialization side for the second time;

[0021] S7. After the cross stacking reaches the specified number of layers, the deceleration mechanism can apply a certain pressure to the stacked film above the stacking area, and the heating device below can heat the pressurized film.

[0022] Furthermore, in step S2, the binary image data of the flexible dielectric sheet image is determined by the average value threshold method. Specifically, the circular positioning slurry image coated on the flexible dielectric sheet in step S2 is obtained by binarizing the image in step S1 and determining the threshold using the average value method.

[0023] Furthermore, in step S3, the steps of using the findContours function to determine the circular rectangle are as follows:

[0024] First, the contour of the binarized image is read to calculate the area and perimeter of each contour;

[0025] Then, the ratio of the area to the square of the perimeter is calculated, and the data whose value is greater than the set value is determined to be a circle, and input into the neural network calculation;

[0026] Finally, the remaining contours are tested for vertices, and those with a vertex value of 4 are determined to be rectangles.

[0027] It should be understood that the coordinate starting point is preferably the point with coordinates (0,0) at the upper left corner of the flexible dielectric sheet image in the visual positioning area. Other feasible starting point determination methods can set other easily determined starting points.

[0028] The visual positioning model training method provided by the present invention is to use TensorFlow to extract the feature value of the circular contour detected by the findContours function, and use Sequential connection to perform several convolution poolings so that the model can accurately identify the circular slurry area and the rectangular slurry area (dried) that have been binarized. Then, the additional moving distance and eccentric coordinates in the input controller can more accurately control the movement of the guide rail and the rotation and movement of the film pressing mechanism to the stacking area, and make the film body alternately stacked. It should be understood that the coordinate starting point is used as a calibration position, which is regarded as a coordinate point that can meet the visual positioning.

[0029] Among them, the visual positioning of the circular positioning slurry (dried) coated on the flexible dielectric sheet needs to be performed in the area where the flexible dielectric sheet is placed;

[0030] Before the first visual recognition of the flexible dielectric sheet, the laminating mechanism needs to be initialized, that is, the displacement and angle of the laminating mechanism deviating from the standard of the stacking area are determined using the standard image, so that the deceleration mechanism moves to the same position each time it is used;

[0031] Furthermore, in step S4, the findContours function is used to read the contour of the binarized image, and the circular data obtained by approxPolyDP function to approximate the contour is input into the convolutional neural network to obtain the circle recognition accuracy. For images with a recognition accuracy greater than 0.99, their data is input into the controller to perform subsequent operations. The role of the convolutional neural network is to improve the accuracy of ordinary machines in recognizing circles.

[0032] The circular coating area on the flexible dielectric sheet can be other recognizable contour shapes, such as triangles. It should be understood that the replacement of the circular coating area shape does not affect contour recognition and TensorFlow convolution calculations, and the shape enclosed by the circular coating area and the internal rectangular shape replacement do not affect the alternating stacking.

[0033] Furthermore, in step S4, the model training process of the convolutional neural network is:

[0034] A camera fixed above the area where the flexible dielectric sheet is placed obtains a flexible dielectric sheet placed freely in the flexible dielectric sheet area; wherein, the flexible dielectric sheet is coated with a circular positioning slurry (dried); the image is binarized to obtain a clear binary image; it should be understood that the original colors of the circular positioning slurry area and the flexible dielectric sheet area are distributed on both sides of the average value, and after binarization, the circular positioning slurry and the flexible dielectric sheet are clearly separable, and there are only four circles located near the four corners of the flexible dielectric sheet, and the four circle centers are connected to form a square.

[0035] Classify and binarize multiple standard flexible dielectric sheets to create a data set; classification refers to reading the contour of the circular slurry area coated on the flexible dielectric sheet using the findContours function on the binarized image, and using the approxPolyDP function to approximate the contour. If the contour circularity is greater than 0.8, it is classified as qualified, and the rest are unqualified. The qualified / unqualified images are made into a data set for model training;

[0036] Then, we extract features from the images in the dataset, extract the pixels with the largest value in each area, slide multiple 3×3 convolution kernels on the image, stack the multi-layer convolution models, and use max-pooling for maximum pooling.

[0037] Finally, the Dense layer connects the data input from the Flatten layer to each neuron in the current layer for accurate classification.

[0038] It should be understood that convolutional neural networks require a large amount of data for convolution calculations to improve model accuracy.

[0039] The beneficial effects of the present invention are:

[0040] The present invention provides a lamination and lamination system of a flexible dielectric sheet and a visual neural network positioning method thereof, which are applied to the alternating alignment and hot pressing of sheets, combining findContours to identify circles with TensorFlow neural network to accurately identify preferred sheets, connecting the centers of the two circles closest and farthest from the coordinate starting point with the center of the circumscribed circle, calculating the angles of the bisectors of this line with the horizontal axis and the vertical axis as the eccentric angle, and initializing according to the distances of the four vertices of the rectangular array from the four centers of the circles, realizing the alternating stacking of the flexible dielectric sheets, applying a certain pressure to the alternating flexible dielectric sheets and heating them by manipulating the deceleration mechanism, realizing the close combination of the flexible dielectric sheets, and effectively solving the existing problems of complicated working procedures, inaccurate electrode alignment, low efficiency of available facing area, and inability to keep constant pressure. In particular, the method based on visual positioning coating can be effectively applied to the field of dielectric sheet type multilayer ceramic capacitor stacking, solving the problem of internal stress caused by inaccurate electrode utilization and alignment.

[0041] Of course, any product implementing the present invention does not necessarily need to achieve all of the above advantages at the same time. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for describing the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.

[0043] Figure 1 It is a structural schematic diagram of a lamination and film pressing system of a flexible dielectric sheet of the present invention;

[0044] Figure 2 It is a schematic diagram of the structure of the flexible dielectric sheet and the coating thereon in the present invention;

[0045] Figure 3 Schematic diagram of a convolutional neural network in the present invention;

[0046] Figure 4 The present invention uses hand-stacked, machine-stacked and standard sample partial performance diagrams;

[0047] In the accompanying drawings, the reference numerals of the various components are as follows:

[0048] 1-working table, 2-flexible dielectric sheet placement area, 3-flexible dielectric sheet stacking area, 4-lateral displacement driving assembly, 5-longitudinal displacement driving assembly, 6-vertical displacement driving assembly, 7-rotation driving device, 8-flexible dielectric sheet adsorption disk, 9-CCD camera, 10-heating device, 11-controller. DETAILED DESCRIPTION

[0049] 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.

[0050] like Figure 1-Figure 4 As shown, this embodiment provides a lamination and film pressing system for flexible dielectric sheets, including a workbench 1 and a controller 11. A flexible dielectric sheet placement area 2 and a flexible dielectric sheet stacking area 3 are provided near the center of the workbench 1. Two lateral displacement drive assemblies 4 are symmetrically installed on both sides of the workbench 1. The movable parts of the two lateral displacement drive assemblies 4 jointly support a longitudinal displacement drive assembly 5. The movable part of the longitudinal displacement drive assembly 5 is installed with a vertical displacement drive assembly 6. The movable end of the vertical displacement drive assembly 6 is installed with a rotation drive device 7. The movable part of the rotation drive device 7 is connected to a flexible dielectric sheet adsorption plate 8. A CCD camera 9 is also installed on the workbench 1. A heating device 10 capable of heating the flexible dielectric sheet stacking area 3 is provided on the workbench 1. The lateral displacement drive assembly 4, the longitudinal displacement drive assembly 5, the vertical displacement drive assembly 6, the rotation drive device 7, the flexible dielectric sheet adsorption plate 8, the CCD camera 9, and the heating device 10 are respectively connected to the controller 11.

[0051] In this embodiment, the controller 11 can receive or call the available data output by the visual area, control the operation of the lateral displacement drive component 4, the longitudinal displacement drive component 5, the vertical displacement drive component 6, and the rotation drive device 7, and move the flexible dielectric sheet adsorption plate 8 to the flexible dielectric sheet placement area 2. The flexible dielectric sheet adsorption plate 8 can suck up the flexible dielectric sheet, transport it to the designated area and stack it alternately.

[0052] In this embodiment, after the flexible dielectric sheet stacking area 3 is heated by the heating device 10 , the flexible dielectric sheets stacked thereon can be more tightly bonded.

[0053] In this embodiment, the lateral displacement driving assembly 4 and the longitudinal displacement driving assembly 5 are linear drives based on screw transmission.

[0054] In this embodiment, infrared anti-collision limiters for preventing deceleration are installed on the lateral displacement driving assembly 4 and the longitudinal displacement driving assembly 5 .

[0055] In this embodiment, a temperature sensor capable of detecting the temperature at the flexible dielectric sheet stacking area 3 is disposed on the workbench 1 , and the temperature sensor is connected to the controller 11 .

[0056] This embodiment also provides a visual neural network positioning method for the above-mentioned lamination and film pressing system, comprising the following steps:

[0057] S1, obtaining an image of a flexible dielectric sheet of a deceleration mechanism in a visual positioning area;

[0058] S2, identifying the acquired flexible dielectric sheet image, and performing binarization processing on the image to obtain a binary image;

[0059] S3, using findContours function to detect the circular contour of the obtained binary image, and fitting a circumscribed circle based on the circular contour;

[0060] S4. Based on the convolutional neural network, the data with a circle accuracy greater than 0.99 is obtained, the coordinates of the centers of all circles fitted in findContours are saved, and the distances between the four vertices of the rectangular matrix and the four centers of the circles are compared;

[0061] S5. When stacking for the first time, take the coordinates of the center of the circumscribed circle closest to the coordinate starting point as the additional moving distance of the guide rail, connect the centers of the two circles whose centers are closest to the coordinate starting point and farthest from the coordinate starting point, and calculate the angle between this line and the bisector of the horizontal axis and the vertical axis as the eccentricity angle;

[0062] S6, inputting the additional moving distance and eccentric coordinates output by the computer into the controller, controlling the rotation and movement of the guide rail and the film pressing mechanism to the stacking area, and stacking the films alternately; wherein, in step S4, the distances between the four vertices of the rectangular array and the four centers of the circles are one pair longer and one pair shorter, and alternate stacking means, for example, selecting the longer pair as the initialization side for the first time, and selecting the shorter side as the initialization side for the second time;

[0063] S7. After the cross stacking reaches the specified number of layers, the deceleration mechanism can apply a certain pressure to the stacked film above the stacking area, and the heating device below can heat the pressurized film.

[0064] In step S2, binary image data of the flexible dielectric sheet image is determined by an average value threshold method.

[0065] In step S3, the steps of using the findContours function to determine the circular rectangle are as follows:

[0066] First, the contour of the binarized image is read to calculate the area and perimeter of each contour;

[0067] Then, the ratio of the area to the square of the perimeter is calculated, and the data whose value is greater than the set value is determined to be a circle, and input into the neural network calculation;

[0068] Finally, the remaining contours are tested for vertices, and those with a vertex value of 4 are determined to be rectangles.

[0069] The specific application of this embodiment is as follows:

[0070] The CCD camera 9 is fixed on the workbench 1. The image area it acquires can cover the area from the flexible dielectric sheet placement area 2 to the flexible dielectric sheet stacking area 3. The antiferroelectric flexible dielectric sheet image it acquires is transmitted to the computer. The flexible dielectric sheet image is as follows: Figure 2 As shown. The computer loaded vision and neural network calibration method is used to accurately identify and locate the flexible dielectric sheet, and finally control the lateral displacement drive component 4, the longitudinal displacement drive component 5, the vertical displacement drive component 6, the rotation drive device 7 and the flexible dielectric sheet adsorption disk 8 to move, and then the antiferroelectric flexible sheet placed in the flexible dielectric sheet placement area 2 is neatly and alternately pressed on the flexible dielectric sheet stacking area 3. The heating device 10 continuously heats the antiferroelectric flexible sheet during the alternating pressing process, and the temperature sensor ensures that the temperature is constant at the required temperature.

[0071] Standard circular slurry areas and standard images of four circles surrounding a square are classified. It should be understood that images containing three or four standard circular areas do not affect the lamination effect and such images are qualified and usable data.

[0072] The preprocessed image is binarized and classified as above. The batch_size is defined and connected using the sequential function. Conv2D convolution and MaxPooling2D pooling are performed. Flatten is used to convert the operation result into a one-dimensional vector and the input data is connected to each neuron. The compile function is used to define the evaluation indicator "accuracy" as acc. Here, the definition of the accuracy indicator comes from the built-in indicator of the Keras function. The calculation formula is shown in formula (1-1):

[0073]

[0074] The model file is obtained by convolution calculation on a large amount of data, which is suitable for judging whether the circular slurry area / input image is qualified.

[0075] The model training process of the convolutional neural network is as follows: first, multiple standard flexible dielectric sheets are classified and binarized to produce a data set; classification refers to the circular slurry area coated on the flexible dielectric sheet using the findContours function to read the contour of the binarized image, and the approxPolyDP function is used to approximate the contour, and the contour circularity greater than 0.8 is obtained, which is classified as qualified, and the rest are unqualified, and the qualified / unqualified images are made into a data set for model training; then, feature extraction is performed on the data set images, and the pixel value in each area with the largest value is taken out, and multiple 3×3 convolution kernels are used to slide on the image, and the multi-layer convolution model is stacked, and max-pooling is used for maximum pooling; finally, the Dense layer connects the data input from the Flatten layer to each neuron in the current layer for accurate classification.

[0076] The implementation process of using findContours to detect contours and identify circular slurry coated areas and rectangular slurry areas (dried) is briefly described as follows:

[0077] Step 1: The CCD camera acquires an image of the flexible dielectric sheet placement area. At this time, the antiferroelectric flexible sheet should have been placed in the flexible dielectric sheet placement area by the user.

[0078] Step 2: Binarize the image obtained in step 1 by using the average value method. First, convert the image into a grayscale image, calculate the average grayscale of all pixels, and use the average value as the threshold for binarization.

[0079] The formulas for the mean value binarization method are shown in equations (1-2), (1-3), and (1-4), where I(i) is the grayscale value of the i-th pixel in the image, and N is the total number of pixels in the image. T is the calculated average threshold, and S is the sum of the grayscale values ​​of all pixels. N is the total number of pixels in the image (for example, the height of the image multiplied by the width), and B(i) is the value of the i-th pixel in the binarized image (255 represents white, and 0 represents black).

[0080]

[0081] Step 3: Find the contour of the binary image using the findContours function, and calculate the ratio of the contour area (A) and perimeter (P) as shown in formula (1-5):

[0082]

[0083] Step 4: The values ​​with a ratio (R) greater than a set value (accuracy greater than 90%) are identified as circles, and the remaining contours are subjected to vertex detection. If the vertex value is 4, it is a rectangle.

[0084] Among them, the maximum and minimum distances from x and y among all the vertices of the rectangle form four coordinates respectively. The difference between their values ​​and the corresponding center of the circle is the distance from the four vertices of the rectangular matrix to the four centers of the circle, which should be in two pairs, one pair is longer and the other pair is shorter. When stacking for the first time, take the coordinates of the center of the circumscribed circle closest to the coordinate starting point as the additional moving distance of the guide rail, connect the centers of the two circles closest and farthest from the coordinate starting point, and calculate the angle between this line and the bisector of the horizontal and vertical axes as the eccentricity angle.

[0085] Step 5, importing the coordinates of the center of the circle, a pair of length comparison results and the eccentric angle data into the controller, the controller controls the lateral displacement drive component 4, the longitudinal displacement drive component 5, the vertical displacement drive component 6, the rotation drive device 7 and the flexible dielectric sheet adsorption disk 8 to operate, and alternately presses the antiferroelectric flexible sheet from the placement area precisely onto the flexible dielectric sheet stacking area containing the heating device 10 and the temperature sensor.

[0086] The alternating stacking implementation process is:

[0087] The results of the comparison of the two pairs of distances are stored during the first stacking. If the distance on the left side of the image is greater than that on the right side, during the second stacking, if the distance on the left side is greater than that on the right side, the steering mechanism fixed on the deceleration mechanism rotates by an eccentric angle + 180°; if the distance on the left side is less than that on the right side, the steering mechanism fixed on the deceleration mechanism rotates by an eccentric angle during the second stacking. Stack to the number of layers required by the user according to this logic.

[0088] Among them, step 4 performs convolution calculation on the data with circular accuracy greater than 0.9, outputs the circular judgment neural network model, uses the squential function in TensorFlow to perform convolution calculation, outputs the judgment accuracy, and if it is greater than 0.99, it is a standard part, and proceeds to the subsequent steps.

[0089] Among them, the convolution calculation steps for the input binary flexible dielectric sheet image are as follows: Figure 3 As shown in the figure, feature extraction is performed on the image, the pixel with the largest value in each area is taken out, multiple 3×3 convolution kernels are used to slide on the image, multi-layer convolution models are stacked, and max-pooling is used for maximum pooling. Finally, the Dense layer can connect the data input by the Flatten layer to each neuron in the current layer for accurate classification.

[0090] In this embodiment, the performance of ten layers of antiferroelectric flexible sheets manually stacked and antiferroelectric ceramics stacked according to the present invention are compared. The data are averaged by multiple groups of experiments and compared with standard samples. Figure 4 As shown, the efficiency of machine stacking is better than that of hand stacking, wherein the same antiferroelectric flexible material is used in the preparation of hand stacking, stacking of this embodiment and standard samples.

[0091] The preferred embodiments of the present invention disclosed above are only used to help explain the present invention. The preferred embodiments do not describe all the details in detail, nor do they limit the invention to specific implementation methods. Obviously, many modifications and changes can be made according to the content of this specification. This specification selects and specifically describes these embodiments in order to better explain the principles and practical applications of the present invention, so that those skilled in the art can understand and use the present invention well. The present invention is limited only by the claims and their full scope and equivalents.

Claims

1. A lamination system for flexible dielectric sheets, comprising a workbench and a controller, characterized in that: A flexible dielectric sheet placement area and a flexible dielectric sheet stacking area are provided near the center of the workbench, two lateral displacement drive components are symmetrically installed on both sides of the workbench, the movable parts of the two lateral displacement drive components jointly support a longitudinal displacement drive component, the movable part of the longitudinal displacement drive component is installed with a vertical displacement drive component, the movable end of the vertical displacement drive component is installed with a rotation drive device, the movable part of the rotation drive device is connected with a flexible dielectric sheet adsorption plate, a CCD camera is also installed on the workbench, and a heating device capable of heating the flexible dielectric sheet stacking area is provided on the workbench, and the lateral displacement drive component, the longitudinal displacement drive component, the vertical displacement drive component, the rotation drive device, the flexible dielectric sheet adsorption plate, the CCD camera, and the heating device are respectively connected to the controller.

2. The lamination system according to claim 1, characterized in that: The controller can receive or call available data output by the visual area, control the actions of the lateral displacement drive component, the longitudinal displacement drive component, the vertical displacement drive component, and the rotation drive device, and move the flexible dielectric sheet adsorption plate to the flexible dielectric sheet placement area. The flexible dielectric sheet adsorption plate can suck up the flexible dielectric sheet, transport it to the stacking area and stack it alternately.

3. The lamination film system according to claim 1, characterized in that: After the flexible dielectric sheet stacking area is heated by the heating device, the flexible dielectric sheets stacked thereon can be more tightly bonded.

4. The lamination film system according to claim 1, characterized in that: The lateral displacement drive assembly and the longitudinal displacement drive assembly are linear drives based on screw transmission.

5. The lamination film system according to claim 1, characterized in that: The lateral displacement drive assembly and the longitudinal displacement drive assembly are equipped with infrared anti-collision limiters for preventing deceleration.

6. The lamination film system according to claim 1, characterized in that: The workbench is provided with a temperature sensor capable of detecting the temperature of the flexible dielectric sheet stacking area 3 , and the temperature sensor is connected to the controller.

7. The visual neural network positioning method for a lamination film lamination system according to any one of claims 1 to 6, characterized in that: The steps include: S1, obtaining an image of a flexible dielectric sheet of a deceleration mechanism in a visual positioning area; S2, identifying the acquired flexible dielectric sheet image, and performing binarization processing on the image to obtain a binary image; S3, using findContours function to detect the circular contour of the obtained binary image, and fitting a circumscribed circle based on the circular contour; S4. Based on the convolutional neural network, the data with a circle accuracy greater than 0.99 is obtained, the coordinates of the centers of all circles fitted in findContours are saved, and the distances between the four vertices of the rectangular matrix and the four centers of the circles are compared; S5. When stacking for the first time, take the coordinates of the center of the circumscribed circle closest to the coordinate starting point as the additional moving distance of the guide rail, connect the centers of the two circles whose centers are closest to the coordinate starting point and farthest from the coordinate starting point, and calculate the angle between this line and the bisector of the horizontal axis and the vertical axis as the eccentricity angle; S6, inputting the additional moving distance and eccentric coordinates output by the computer into the programmable controller, controlling the rotation and movement of the guide rail and the film pressing mechanism to the stacking area, and stacking the films alternately; wherein, in step S4, the distances between the four vertices of the rectangular array and the four centers of the circles are one pair longer and one pair shorter, and alternate stacking means, for example, selecting the longer pair as the initialization side for the first time, and selecting the shorter side as the initialization side for the second time; S7. After the cross stacking reaches the specified number of layers, the deceleration mechanism can apply a certain pressure to the stacked film above the stacking area, and the heating device below can heat the pressurized film.

8. The visual neural network positioning method according to claim 6, characterized in that: In step S2, binary image data of the flexible dielectric sheet image is determined by an average value threshold method.

9. The visual neural network positioning method according to claim 6, characterized in that: In step S3, the steps of using the findContours function to determine the circular rectangle are as follows: First, the contour of the binarized image is read to calculate the area and perimeter of each contour; Then, the ratio of the area to the square of the perimeter is calculated, and the data with a value greater than 0.071 is determined to be a circle, and input into the neural network calculation; Finally, the remaining contours are tested for vertices, and those with a vertex value of 4 are determined to be rectangles.

10. The visual neural network positioning method according to claim 6, characterized in that: In step S4, the model training process of the convolutional neural network is: First, multiple standard flexible dielectric sheets are classified and binarized to create a data set. Classification refers to the circular slurry area coated on the flexible dielectric sheet being read and judged by the findCounters function. The findContours function reads the contour of the binarized image, and the approxPolyDP function is used to approximate the contour. If the contour circularity is greater than 0.8, it is classified as qualified, and the rest are unqualified. The qualified / unqualified images are made into a data set for model training. Then, we extract features from the images in the dataset, extract the pixels with the largest value in each area, slide multiple 3×3 convolution kernels on the image, stack the multi-layer convolution models, and use max-pooling for maximum pooling. Finally, the Dense layer connects the data input from the Flatten layer to each neuron in the current layer for accurate classification.

Images