An adaptive corona printing layout optimization system based on a neural network model

Abstract

The application provides a kind of adaptive corona printing layout optimization system based on neural network model, it is related to data processing technical field, including: detection module, for by confocal microscope array and laser displacement sensor, the surface of PET film after corona treatment is scanned in three dimensions, obtains surface fluctuation characteristic value set;Characteristic processing module is used to calculate the change value of surface micro-roughness before and after processing according to surface fluctuation characteristic value set, extracts the regular convex structure characteristic parameters generated by corona treatment, including the distribution of convex height, pitch density and surface real contact area increment ratio.The application accurately obtains the surface topography data of PET film through the detection module, and through the deep analysis and intelligent calculation of feature processing, dynamic prediction and other modules, combines adaptive layout optimization circuit parameters, and forms a closed loop optimization by using feedback control, effectively improves the quality and production efficiency of printed circuit, enhances the adaptive ability and reliability of the system.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of data processing, in particular to an adaptive corona printing layout optimization system based on a neural network model. BACKGROUND

[0002] The defects of the traditional fixed layout scheme in flexible substrate corona printing are obvious:

[0003] Firstly, the layout cannot be adjusted according to the changes in the micro-topography of the flexible substrate after corona treatment, resulting in poor adaptability of the printed circuit to the substrate surface. For example, after the PET film is treated by corona, micro-protrusions appear on the surface, and the traditional scheme still prints according to the original fixed mode, and the circuit is prone to breakage at the protrusion due to stress concentration.

[0004] Secondly, it is easy to cause circuit breakage or resistance fluctuation problems, and cannot meet the quality requirements of high-precision circuits. For example, when printing fine sensor circuits, the traditional scheme does not consider the surface topography, and the resistance of part of the circuit fluctuates greatly, resulting in inaccurate sensor data acquisition.

[0005] In addition, the electronic manufacturing industry continues to improve the quality and performance requirements of products, and the traditional scheme cannot adapt to the demand of the industry development for high-quality printed conductive circuits on flexible substrates.

[0006] For example, in the current rapid development of wearable devices, the circuits printed by the traditional scheme are difficult to support the stable operation of the equipment for a long time due to unstable quality. SUMMARY

[0007] The technical problem to be solved by the present application is to provide an adaptive corona printing layout optimization system based on a neural network model, which improves the precision and quality of the printed circuit and ensures the stable performance of electronic equipment.

[0008] To solve the above technical problems, the technical scheme of the present application is as follows:

[0009] In a first aspect, an adaptive corona printing layout optimization system based on a neural network model comprises:

[0010] A detection module for performing three-dimensional topography scanning on the surface of the PET film after corona treatment by a confocal microscope array and a laser displacement sensor, and obtaining a set of surface fluctuation characteristic values;

[0011] A feature processing module for calculating the change in surface micro-roughness before and after processing according to the set of surface fluctuation characteristic values, and extracting regular protrusion structure feature parameters generated by corona treatment, including protrusion height distribution, spacing density and surface real contact area increment ratio;

[0012] The dynamic prediction module is configured to analyze the surface topography parameters and preset circuit layout information by using a double-channel convolutional neural network, and output each side fracture probability and resistance fluctuation coefficient.

[0013] The adaptive layout module is configured to dynamically adjust the circuit parameters by using a path planner including a group incremental learning algorithm according to the fracture probability and the resistance fluctuation coefficient.

[0014] The feedback control module is configured to monitor the resistance value of the printed circuit in real time by using an online impedance measuring instrument, and trigger a parameter feedback mechanism when the relative deviation between the actual value and the theoretical value of the resistance is greater than or equal to 5%, that is, the coordinates of the abnormal area are fed back to the detection module for local re-measurement, and the re-measurement data are input into the dynamic prediction module to update the weight matrix.

[0015] Further, the surface micro-roughness change value before and after processing is calculated based on the surface relief characteristic value set, and the regular protrusion structure characteristic parameters generated by the corona treatment are extracted, including the protrusion height distribution, the spacing density and the surface real contact area increment ratio, including:

[0016] The surface relief characteristic value set obtained after the corona treatment is integrated with the PET film surface three-dimensional topography data before the corona treatment and the corresponding characteristic value set to form a data comparison group.

[0017] The surface height data before and after processing in the data comparison group are analyzed, and the root mean square roughness of the surface height before and after processing is calculated, respectively, to obtain the surface micro-roughness change value.

[0018] According to the surface micro-roughness change value, the three-dimensional topography image constructed based on the surface relief characteristic value set is subjected to image morphological processing operation, and a threshold value is set according to the height difference between the protrusion and the substrate to separate the protrusion structure from the surface topography to form an independent protrusion structure region.

[0019] Each protrusion structure is traversed, and the vertical distance between the highest point and the substrate plane, that is, the protrusion height, is measured. The height values of each protrusion structure are summarized, grouped and counted according to certain height intervals, and the proportion of the number of protrusions in each interval to the total number of protrusions is calculated to generate a protrusion height distribution table.

[0020] An effective distance threshold is set, the number of adjacent protrusions of each protrusion structure within the threshold range is counted, and the average spacing density is obtained.

[0021] According to the average spacing density and the surface relief characteristic value set, three-dimensional models of the PET film surfaces before and after processing are respectively constructed by using a three-dimensional modeling software, and the surface real contact area increment ratio before and after processing is calculated by comparing the surface real contact areas in the three-dimensional models.

[0022] Further, the surface topography parameters and the preset circuit layout information are analyzed by using a double-channel convolutional neural network to output the edge fracture probability and the resistance fluctuation coefficient, including:

[0023] The surface topography parameters and the preset circuit layout information are collected, and the surface topography parameters are converted into a tensor form and the circuit layout information is converted into a two-dimensional matrix;

[0024] The converted surface topography parameters and the preset circuit layout information are normalized to map the converted surface topography parameters and the preset circuit layout information to the [0, 1] interval to obtain normalized data;

[0025] According to the normalized data, a double-channel convolutional neural network is constructed, and the normalized data is divided into a training set and a validation set; the training set is input into the double-channel convolutional neural network to extract multi-scale features of the surface topography parameters and local features of the circuit connection;

[0026] The multi-scale features of the surface topography parameters and the local features of the circuit connection are spliced to obtain combined features, and the combined features are processed through a fully connected layer to learn the relationship between the features;

[0027] The combined features processed through the fully connected layer are input into an output layer of the double-channel convolutional neural network, one neuron is used to calculate the edge fracture probability of the printed circuit by combining a normalized exponential function; the edge fracture probability output value is in the [0, 1] interval and the sum of all edge probabilities is 1; another neuron is used to obtain the resistance fluctuation coefficient by combining a linear activation function;

[0028] Cross-entropy and mean square error loss functions are defined, and the gradients of the loss functions with respect to the parameters of the double-channel convolutional neural network are calculated, and the parameters of the double-channel convolutional neural network are updated at the same time;

[0029] The parameters of the double-channel convolutional neural network are evaluated using the validation set, and the structure and hyperparameters of the double-channel convolutional neural network are adjusted according to the evaluation results until a preset iteration number is reached to obtain a trained and optimized double-channel convolutional neural network;

[0030] The normalized surface topography parameters and the preset circuit layout information are input into the trained and optimized double-channel convolutional neural network to output the edge fracture probability and the resistance fluctuation coefficient.

[0031] Further, the double-channel convolutional neural network includes:

[0032] The first channel: input the surface topography parameter tensor in the training set, extract the multi-scale features of the surface topography parameters through a 3-layer convolutional network;

[0033] The second channel: input a two-dimensional matrix of preset line layout information, aggregate node neighborhood information through a graph attention network, and generate an edge feature vector.

[0034] Further, according to the fracture probability and the resistance fluctuation coefficient, the line parameters are dynamically adjusted by a path planner containing a group incremental learning algorithm, including:

[0035] A path planner based on a group incremental learning algorithm is constructed, and parameters of the group incremental learning algorithm are set, including the group size, the learning rate, and the step length of incremental learning.

[0036] The preset line layout information is sorted to determine the connection relationship and node position of the line, and the adjustment range of the line parameters, including the minimum and maximum values of the width, thickness, and spacing parameters of the line, is determined.

[0037] According to the preset line layout information and the adjustment range of the line parameters, an initial line parameter combination group is randomly generated, and each combination represents a line parameter setting scheme.

[0038] For each line parameter combination in the group, the corresponding fitness index is calculated in combination with the fracture probability and the resistance fluctuation coefficient.

[0039] At the beginning of each iteration, new line performance data is obtained from the actual printing process, and the new line performance data and the line parameter combinations in the current group and the corresponding fitness indexes are input into the path planner, the path planner is updated, and an updated path planner is obtained.

[0040] According to the updated path planner, the line parameter combinations in the group are adjusted to obtain adjusted line parameter combinations, and the fitness indexes of the adjusted line parameter combinations are recalculated.

[0041] The process of obtaining new line performance data, updating the path planner, adjusting the line parameter combinations, and recalculating the fitness indexes is repeated until a preset maximum number of iterations is reached, and the final line parameter combination is determined from the group according to the fitness indexes of the line parameter combinations.

[0042] According to the final line parameter combination, the actual parameters of the printed line are dynamically adjusted.

[0043] Further, the line parameter combination includes line width, line thickness, line spacing, line length, conductor material, insulating material, solder joint quality, and via parameters.

[0044] Further, for each line parameter combination in the group, the corresponding fitness index is calculated in combination with the fracture probability and the resistance fluctuation coefficient, including:

[0045] Obtain the actual parameter values of the breaking probability, resistance fluctuation coefficient, line width, line thickness, and line spacing of each line unit and the corresponding adjustment range;

[0046] Determine the contribution degree of the breaking probability part according to the breaking probability of each line unit, and obtain the resistance fluctuation coefficient contribution degree according to the resistance fluctuation coefficient of each line unit;

[0047] For each line unit, determine the middle position of the line width and line thickness adjustment range, and according to the middle position, determine the deviation degree of the line width from the middle position and the deviation degree of the line thickness from the middle position;

[0048] Determine the middle position of the line spacing adjustment range, and calculate the deviation degree of the actual line thickness from the middle position;

[0049] Fuse the contribution degree of the breaking probability part, the resistance fluctuation coefficient contribution degree, the deviation degree of the line width from the middle position, the deviation degree of the line thickness from the middle position, and the deviation degree of the actual line thickness from the middle position to obtain the adaptation index of the line parameter combination.

[0050] Further, through the online impedance measuring instrument, the resistance value of the printed circuit is monitored in real time, and when the relative deviation of the actual value and the theoretical value is ≥5%, the parameter feedback mechanism is triggered, that is, the coordinates of the abnormal area are fed back to the detection module for local retest, and the retest data is input into the dynamic prediction module to update the weight matrix, including:

[0051] Connect the online impedance measuring instrument with the printed circuit, and the online impedance measuring instrument measures the line resistance value in real time according to the set sampling frequency, continuously obtains the actual resistance value data of the printed circuit;

[0052] Obtain the theoretical resistance design value of the printed circuit from the database, and calculate the relative deviation value of the actual resistance value measured in real time and the theoretical value;

[0053] Compare the relative deviation value with the set threshold value 5%, if the relative deviation value <5%, continue to monitor the real-time resistance value; if the relative deviation value ≥5%, immediately trigger the parameter feedback mechanism;

[0054] When the parameter feedback mechanism is triggered, according to the layout information of the printed circuit and the measurement point position of the online impedance measuring instrument, the coordinates of the resistance value abnormal area are located;

[0055] The detection module receives the coordinates of the resistance value abnormal area, starts the local retest program, controls the detection equipment to move to the abnormal area, and retests the resistance value of the printed circuit in the abnormal area from multiple points and multiple angles; In the retest process, the environmental parameters are collected, and the retested resistance value data and environmental parameters are integrated into a retest data set;

[0056] The retest data set is input into the dynamic prediction module, the influence of the retest data on the current double-channel convolutional neural network is analyzed, and the weight matrix is updated by recalculating the weight of each parameter in the double-channel convolutional neural network, including the weight of the convolutional layer and the weight of the fully connected layer.

[0057] In a second aspect, a computing device includes:

[0058] One or more processors;

[0059] A storage device for storing one or more programs, which, when executed by the one or more processors, cause the one or more processors to implement the system.

[0060] In a third aspect, a computer-readable storage medium stores a program, which, when executed by a processor, implements the system.

[0061] The above scheme of the present application at least includes the following beneficial effects:

[0062] The detection module uses a confocal microscope array and a laser displacement sensor to perform three-dimensional topography scanning on the surface of the PET film, and can accurately obtain a surface fluctuation characteristic value set. Compared with traditional detection methods, it can capture micron-level or even nanometer-level surface structure changes, and ensure comprehensive understanding of the micro-topography of the film surface.

[0063] The feature processing module calculates the surface micro-roughness change value based on the surface fluctuation characteristic value set, and extracts key parameters such as convex height distribution, spacing density, and surface real contact area increment ratio. These quantitative indicators systematically analyze the influence of corona treatment on the film surface, provide a scientific basis for understanding the effect of corona treatment and evaluating surface quality, and enable production personnel to judge whether the film surface meets the requirements of printed circuits from multiple dimensions.

[0064] The dynamic prediction module combines the surface topography parameters with the preset circuit layout information by means of the double-channel convolutional neural network, and outputs the edge fracture probability and resistance fluctuation coefficient. This module can simulate the performance of circuits under different surface conditions and circuit layouts, predict potential fracture risks and resistance fluctuation problems in advance, and change the traditional mode of relying on experience to judge, providing forward-looking guidance for circuit layout optimization in a data-driven manner.

[0065] The adaptive layout module dynamically adjusts the circuit parameters according to the fracture probability and the resistance fluctuation coefficient, using a path planner containing a group incremental learning algorithm. This optimization method can respond in real time to the surface topography and circuit performance requirements, actively avoiding circuit quality defects caused by film surface problems, such as circuit fracture and unstable resistance, during the circuit design phase, improving the yield and reliability of printed circuits, and reducing design rework caused by unreasonable parameters, and improving production efficiency.

[0066] The feedback control module monitors the resistance value of the printed circuit in real time through an online impedance measuring instrument, triggers the parameter feedback mechanism when an abnormality occurs, realizes accurate re-measurement of the abnormal area, and uses the re-measurement data to update the weight matrix of the dynamic prediction module. The feedback mechanism forms a closed-loop system from production monitoring, problem positioning to model optimization, enabling the system to continuously learn and adapt to changes in actual production, continuously optimize the neural network model and circuit layout strategy, ensure that the quality of printed circuits remains stable and gradually improves in the long-term production process, and enhance the competitiveness of enterprises in the market. BRIEF DESCRIPTION OF DRAWINGS

[0067] Figure 1 is a schematic diagram of an adaptive corona printing layout optimization system based on a neural network model provided by an embodiment of the present application. DETAILED DESCRIPTION

[0068] Exemplary embodiments of the present disclosure will be described in greater detail below with reference to the accompanying drawings. Although exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure can be implemented in various forms and should not be limited by the embodiments set forth herein. Rather, these embodiments are provided so that the present disclosure can be more thoroughly understood and the scope of the present disclosure can be accurately conveyed to those skilled in the art.

[0069] As Figure 1 shown, an embodiment of the present application proposes an adaptive corona printing layout optimization system based on a neural network model, comprising:

[0070] The detection module 1 is used to scan the surface topography of the PET film after corona treatment by a confocal microscope array and a laser displacement sensor, and obtain a set of surface fluctuation characteristic values.

[0071] The feature processing module 2 is used to calculate the surface micro-roughness change value before and after processing according to the set of surface fluctuation characteristic values, and extract the regular protrusion structure characteristic parameters generated by corona treatment, including protrusion height distribution, pitch density and surface real contact area increment ratio.

[0072] The dynamic prediction module 3 is used to analyze the surface topography parameters and the preset circuit layout information using a double-channel convolutional neural network, and output the fracture probability and the resistance fluctuation coefficient of each side.

[0073] An adaptive layout module 4 is configured to dynamically adjust the circuit parameters through a path planner containing a group incremental learning algorithm according to the fracture probability and the resistance fluctuation coefficient;

[0074] A feedback control module 5 is configured to monitor the resistance value of the printed circuit in real time through an online impedance measuring instrument, and when the relative deviation between the actual value and the theoretical value of the resistance is greater than or equal to 5%, a parameter feedback mechanism is triggered, that is, the coordinates of the abnormal area are fed back to the detection module for local re-measurement, and the re-measurement data are input into the dynamic prediction module to update the weight matrix.

[0075] In the embodiment of the present application, the detection module uses a confocal microscope array and a laser displacement sensor to perform three-dimensional topography scanning on the surface of the PET film, which can accurately and comprehensively obtain a set of surface fluctuation characteristic values. Compared with traditional detection methods, this method can capture more subtle surface topography changes and ensure control of the surface state of the film.

[0076] The feature processing module not only calculates the surface micro-roughness change value based on the data obtained by the detection module, but also extracts the regular protruding structure characteristic parameters generated by the corona treatment, such as protrusion height distribution, spacing density and surface real contact area increment ratio. The acquisition of these parameters helps to deeply understand the influence mechanism of the corona treatment on the surface of the film, provides key basis for optimizing the printing layout, and can make targeted adjustments according to the surface characteristics after the corona treatment.

[0077] The dynamic prediction module uses a double-channel convolutional neural network to analyze the surface topography parameters and the preset circuit layout information, and outputs the fracture probability and the resistance fluctuation coefficient of each side. This module fully utilizes the powerful data analysis and prediction capability of the neural network, can quickly and accurately evaluate the potential risks of different circuit layouts under the current surface state of the film, and compared with traditional experience judgment or simple model prediction, improves the accuracy and reliability of the prediction, and provides scientific quantitative indicators for circuit optimization.

[0078] The adaptive layout module dynamically adjusts the circuit parameters through a path planner containing a group incremental learning algorithm according to the fracture probability and the resistance fluctuation coefficient output by the dynamic prediction module. This module can respond to the surface state of the film and the circuit risk evaluation result in real time, automatically optimize the circuit parameters, effectively avoid the circuit fracture and resistance fluctuation problems caused by the micro concave-convex of the film surface, and ensure that the precision of the printed circuit meets the strict requirement of ≤10 , and improves the quality and stability of the printed circuit.

[0079] The feedback control module monitors the printed circuit resistance value in real time through an online impedance measuring instrument, when the relative deviation between the actual value and the theoretical value of the resistance is greater than or equal to 5%, a parameter feedback mechanism is triggered, the abnormal area coordinates are fed back to the detection module for local re-measurement, and the re-measurement data are input into the dynamic prediction module to update the weight matrix. This closed-loop feedback mechanism realizes real-time monitoring and dynamic adjustment of the printed circuit quality, can timely discover and solve the circuit quality problem, continuously optimize the system model, can continuously adapt to various changes in the production process, guarantee the long-term stability of the printed circuit quality, and improve the production efficiency and product qualification rate.

[0080] In a preferred embodiment of the present application, the surface micro-roughness change value before and after processing is calculated and processed according to the surface relief feature value set, and the regular protrusion structure feature parameters generated by corona treatment are extracted, including protrusion height distribution, spacing density and surface real contact area increment ratio, which can include:

[0081] The surface relief feature value set obtained after corona treatment is integrated with the PET film surface three-dimensional topography data before corona treatment and the corresponding feature value set to form a data comparison group;

[0082] The surface height data before and after processing in the data comparison group is analyzed, and the root mean square roughness of the surface height before and after processing is calculated respectively to obtain the surface micro-roughness change value;

[0083] According to the surface micro-roughness change value, the three-dimensional topography image constructed by the surface relief feature value set is subjected to image morphological processing operation, and a threshold value is set according to the height difference between the protrusion and the substrate to separate the protrusion structure from the surface topography and form an independent protrusion structure region;

[0084] Each protrusion structure is traversed, and the vertical distance between the highest point and the substrate plane, i.e. the protrusion height, is measured; the height value of each protrusion structure is summarized, grouped and counted according to a certain height interval, and the proportion of the number of protrusions in each interval to the total number of protrusions is calculated to generate a protrusion height distribution table;

[0085] An effective distance threshold is set, the number of adjacent protrusions of each protrusion structure within the threshold range is counted, and the average spacing density is obtained;

[0086] According to the average spacing density and the surface relief feature value set, three-dimensional models of the PET film surfaces before and after processing are respectively constructed by using three-dimensional modeling software, the surface real contact areas in the three-dimensional models are compared, and the surface real contact area increment ratio before and after processing is calculated.

[0087] In the embodiments of the present application, the detection module obtains the set of surface relief characteristic values of the PET film after corona treatment through the confocal microscope array and the laser displacement sensor. The three-dimensional topography data of the PET film before corona treatment and the corresponding set of characteristic values are collected in advance using the same detection equipment and method before the film is subjected to corona treatment and are stored in the system database. The data before and after treatment are matched and merged according to the same coordinate system or measurement point position. For example, if the coordinate system is established with a certain corner of the film as the origin for both the before and after measurements, then the characteristic values of the corresponding points are associated according to the coordinate position to form a data set containing the data before and after treatment, i.e., a data comparison group.

[0088] The root mean square roughness (Rq) is an important indicator for measuring the micro-roughness of the surface, and its calculation formula is wherein is the surface height of the th measurement point, is the average value of the surface heights of all measurement points, is the total number of measurement points.

[0089] Specifically, the surface height data before and after treatment are extracted from the data comparison group. For each group of data, the average value of the surface heights of all measurement points is calculated first, and then the root mean square roughness is calculated according to the above formula. Finally, the root mean square roughness after treatment is subtracted from the root mean square roughness before treatment to obtain the surface micro-roughness change value.

[0090] The three-dimensional topography image constructed from the set of surface relief characteristic values is processed using image processing software. The operations include erosion, dilation, opening operation, and closing operation, etc. For example, the opening operation can first perform an erosion operation on the image to remove some small protrusions and noise, and then perform a dilation operation to restore the larger protrusion structure, so that the protrusion structure is clearer and more independent. According to the surface micro-roughness change value and the actual height difference between the protrusion and the base, a suitable threshold value is set. The part of the three-dimensional topography image with a height greater than the threshold value is determined as the protrusion structure, and the part with a height less than the threshold value is determined as the base. In this way, the protrusion structure is separated from the surface topography to form an independent protrusion structure region.

[0091] Each protrusion structure separated is traversed to find its highest point. With the base plane as the reference, the vertical distance between the highest point and the base plane is measured, which is the height of the protrusion structure. A suitable height interval is determined, for example, the protrusion height can be divided into 0 - 1 , 1 - 2 , 2 - 3 Intervals. After collecting the height values of all the protrusions, the number of protrusions in each interval is counted. Then the proportion of the number of protrusions in each interval to the total number of protrusions is calculated. Finally, the height intervals and the corresponding proportions are arranged in a table form, i.e. the protrusion height distribution table. According to the actual research needs and the distribution of protrusion structures, an effective distance threshold is set. For example, the threshold can be set to 5 , which means that the number of adjacent protrusions is counted within a circular region centered at a certain protrusion structure with a radius of 5 . For each protrusion structure, the number of adjacent protrusions within the effective distance threshold is counted. The total number of adjacent protrusions of all protrusion structures is added up and then divided by the total number of protrusion structures to obtain the average spacing density.

[0092] Using professional three-dimensional modeling software such as SolidWorks, AutoCAD, etc., the three-dimensional models of the PET film surfaces before and after corona treatment are constructed according to the average spacing density and the set of surface roughness values. During modeling, the height, position and distribution of protrusion structures are accurately reflected in the model. Through the measurement function of the three-dimensional modeling software, the real contact area of the surface in the three-dimensional model before and after treatment is calculated respectively. The real contact area of the surface refers to the area of the film surface actually in contact with other objects, taking into account the micro-topography of the surface. The real contact area of the surface after treatment is subtracted from the real contact area of the surface before treatment, and then divided by the real contact area of the surface before treatment to obtain the real contact area increment ratio of the surface.

[0093] Suppose there is a PET film, before corona treatment, the surface is scanned using a confocal microscope array and laser displacement sensor to obtain the surface height data containing 1000 measurement points. After calculation, the root mean square roughness of the surface height before treatment is 0.5 . After corona treatment, the film surface is scanned again to obtain a new set of surface roughness values, also containing 1000 measurement points. After integrating the data before and after treatment to form a data comparison group, the root mean square roughness of the surface height after treatment is calculated to be 0.8 , then the change value of the surface micro-roughness is 0.8−0.5=0.3 .

[0094] Traverse the 200 protrusion structures, measure the height of each protrusion, and then group the height according to 0 - 1 , 1 - 2 , 2 - 3 intervals for statistical grouping. Suppose there are 120 protrusions in the 0 - 1 interval, 1 - 2 There are 60 protrusions in the interval, 2 - 3 There are 20 protrusions in the interval. Then the proportion of the number of protrusions in each interval to the total number of protrusions is 120 / 200=60%, 60 / 200=30%, and 20 / 200=10%, respectively. The generated protrusion height distribution is [0-1 60%, 1 - 2 30%, 2-3 10%];

[0095] The effective distance threshold is set to 5 , the number of adjacent protrusions in the threshold range of each protrusion structure is counted, and the average spacing density is calculated to be 2 adjacent protrusions around each protrusion. The three-dimensional modeling software is used to construct the three-dimensional models of the PE film surfaces before and after processing, and the real contact area of the surface before processing is measured to be 100mm 2 , and the real contact area of the surface after processing is measured to be 120mm 2 . Then the real contact area increment ratio is (120−100)÷100=20%.

[0096] By integrating the data before and after processing to form a comparison group, and calculating the surface micro-roughness change value, the influence of corona treatment on the microstructure of the PET film surface can be comprehensively and accurately understood. These data provide a basis for subsequent protrusion structure feature extraction and analysis, and help to deeply study the effect and mechanism of corona treatment. Accurate extraction of regular protrusion structure feature parameters generated by corona treatment, such as protrusion height distribution, spacing density and surface real contact area increment ratio, can provide key basis for printing layout optimization. For example, the protrusion height distribution can help to judge the influence of the protrusion on the printed circuit, the spacing density can affect the layout and connection method of the circuit, and the surface real contact area increment ratio can reflect the adhesion performance of the film surface to the conductive ink, thereby providing scientific guidance for adjusting the circuit parameters.

[0097] Based on these accurate surface feature parameters, the adaptive layout module can more accurately adjust the circuit parameters to avoid circuit breakage or resistance fluctuation caused by the protrusion structure of the film surface, thereby improving the precision and stability of the printed circuit to meet the stringent requirements of electronic equipment on circuit quality. Through in-depth analysis of the surface features of the film after corona treatment, the corona treatment process parameters can be optimized to improve the treatment effect and reduce the rate of defective products caused by surface problems. At the same time, reasonable layout optimization of the circuit can reduce production cost, improve production efficiency, and enhance the competitiveness of enterprises in the market.

[0098] In a preferred embodiment of the present application, a double-channel convolutional neural network is used to analyze the surface topography parameters and the preset circuit layout information, and output the edge breakage probability and the resistance fluctuation coefficient, including:

[0099] Collecting surface topography parameters and preset circuit layout information, and converting the surface topography parameters into tensor form and the circuit layout information into a two-dimensional matrix;

[0100] Normalizing the converted surface topography parameters and the preset circuit layout information, mapping the converted surface topography parameters and the preset circuit layout information to the [0, 1] interval to obtain normalized data;

[0101] According to the normalized data, a double-channel convolutional neural network is constructed, and the normalized data is divided into a training set and a validation set; the training set is input into the double-channel convolutional neural network to extract multi-scale features of the surface topography parameters and local features of the circuit connection; the double-channel convolutional neural network comprises:

[0102] The first channel: inputting the surface topography parameter tensor in the training set, extracting multi-scale features of the surface topography parameters through a 3-layer convolutional network;

[0103] The second channel: inputting the two-dimensional matrix of the preset circuit layout information, aggregating node neighborhood information through a graph attention network to generate an edge feature vector;

[0104] The multi-scale features of the surface topography parameters and the local features of the circuit connection are spliced to obtain combined features, and the combined features are processed through a fully connected layer to learn the relationship between the features;

[0105] The combined features processed through the fully connected layer are input into the output layer of the double-channel convolutional neural network, and a neuron is used to calculate the fracture probability of each edge of the printed circuit combined with a normalized exponential function; the fracture probability output value is in the [0, 1] interval and the sum of all edge probabilities is 1; another neuron is used to obtain the resistance fluctuation coefficient with a linear activation function;

[0106] Define the cross-entropy and mean square error loss functions, and calculate the gradient of the loss function with respect to the parameters of the double-channel convolutional neural network, and update the parameters of the double-channel convolutional neural network;

[0107] The validation set is used to evaluate the parameters of the double-channel convolutional neural network, and the structure and hyperparameters of the double-channel convolutional neural network are adjusted according to the evaluation results until a preset number of iterations is reached, and a trained and optimized double-channel convolutional neural network is obtained;

[0108] The normalized surface topography parameters and the preset circuit layout information are input into the trained and optimized double-channel convolutional neural network to output the fracture probability of each edge and the resistance fluctuation coefficient.

[0109] In the embodiment of the present application, the surface topography parameters, including the protrusion height distribution, pitch density, and surface real contact area increment ratio, are obtained from the feature processing module; the preset line layout information, such as the connection relationship of the lines and the node positions, is obtained from the preset parameter library. The surface topography parameters are converted into a tensor form, which is a multi-dimensional array and can be conveniently calculated and processed in a neural network. For example, the protrusion height distribution and pitch density are combined into a tensor according to a certain dimension order. For the preset line layout information, a two-dimensional matrix is converted, and the elements in the matrix can represent the connection relationship between the line nodes, such as an adjacency matrix form, 1 representing a connection between two nodes, and 0 representing no connection.

[0110] For the converted surface topography parameter tensor and line layout information two-dimensional matrix, a normalization method is used to map them to the [0, 1] interval. Common normalization methods include the minimum-maximum normalization. For each feature in the data set (each dimension in the tensor or each element in the matrix), the formula is used for calculation, where is the original data, and are the minimum and maximum values of the feature, respectively. By this method, all data is normalized to the specified interval, so that the data has the same scale, which helps the training and convergence of the neural network. A double-channel convolutional neural network is built, which includes two independent but interrelated channels. The normalized data is divided into a training set and a validation set according to a certain proportion (such as 70% for training and 30% for verification). The training set is used to train the neural network to learn the features and rules in the data; the validation set is used to evaluate the performance of the model during training to prevent overfitting.

[0111] First channel: the surface topography parameter tensor in the training set is input into the first channel, which is composed of 3 layers of convolutional networks. The convolutional layer slides on the input tensor through the convolution kernel to perform convolution operation and extract features of different scales. For example, the first convolutional layer can extract local features of small scale, and the subsequent layers can extract larger scale and more abstract surface topography features by increasing the receptive field of the convolution kernel or stacking more convolutional layers.

[0112] Second channel: the two-dimensional matrix of the preset line layout information is input into the second channel, and a graph attention network (GAT) is used. The graph attention network calculates the attention weights between nodes to aggregate the neighborhood information of the nodes, thereby generating edge feature vectors. These edge feature vectors can reflect the local features of the line connection, such as the closeness and importance of the connection between nodes.

[0113] The multi-scale features of surface topography parameters extracted from the first channel and the local features of line connections generated from the second channel are concatenated, that is, the two feature vectors are sequentially connected into a longer vector to obtain the combined features. The combined features are then input into a fully connected layer, where each neuron is connected to all neurons in the previous layer. By performing weighted summation and activation operations on the combined features, the relationships between different features are learned, and higher-level feature representations are further extracted.

[0114] Suppose that the combined features processed by the fully connected layer are input into the output layer, corresponding to the first... The input value of the line edge is There are a total of the side of the line The Softmax function calculates the... Probability of breakage along the edge of the line The formula is ,in It is the natural constant (approximately 2.71828). Indicates all The corresponding line Summation, Based on the natural constant As the base, The result of exponential operation on an exponent. The combined features processed by the fully connected layer are input into the output layer, corresponding to the first... The numerical value of each line edge; that is, for each line edge, after processing by the preceding layers of the neural network (including fully connected layers), a corresponding numerical value will be obtained. During the calculation of the Softmax function, the result is calculated for each line edge. The value, then all line edges Add them together to get the denominator .pass satisfy ,and This clearly reflects the probability distribution of breakage at each line edge. Another neuron, in conjunction with a linear activation function, obtains the resistance fluctuation coefficient. The linear activation function directly outputs a real number, which represents the resistance fluctuation coefficient of the corresponding line edge, reflecting the degree of resistance fluctuation.

[0115] We define a cross-entropy loss function (used to calculate the difference between the predicted breakage probability and the true label, suitable for classification problems) and a mean squared error loss function (used to calculate the error between the predicted resistance fluctuation coefficient and the true value, suitable for regression problems). We use the backpropagation algorithm to calculate the gradient of the loss function with respect to the parameters of the two-channel convolutional neural network (such as the kernel weights of convolutional layers, the weights and biases of fully connected layers, etc.). We then update the network parameters using gradient descent or its variants (such as Adam, Adagrad, etc.), continuously adjusting the parameter values ​​to gradually decrease the value of the loss function, thereby improving the model's prediction accuracy.

[0116] The parameters of the dual-channel convolutional neural network (CNN) during training are evaluated using a validation set. The loss value and other evaluation metrics (such as accuracy and mean squared error) on the validation set are calculated. Based on the evaluation results, the structure of the CNN (e.g., adding or removing convolutional layers, adjusting the number of layers in the graph attention network) and hyperparameters (e.g., learning rate, kernel size, attention mechanism parameters) are adjusted. This training, evaluation, and adjustment process is repeated until a preset number of iterations (e.g., 1000 iterations) is reached, resulting in a trained and optimized CNN. Normalized surface topography parameters and preset circuit layout information are input into the trained and optimized CNN. The network outputs the breakage probability and resistance fluctuation coefficient of each side, providing a quantitative basis for subsequent circuit layout optimization.

[0117] Suppose there is a PET film, and the surface morphology parameters obtained by the feature processing module include: the protrusion height distribution is [0-1]. 60% of the total, 1-2 30%, 2-3 [Percentage 10%], spacing density is an average of 2 adjacent protrusions around each protrusion, and the actual surface contact area increment ratio is 20%. These parameters are converted into tensor form, such as... The default route layout is a simple 4-node route, and its adjacency matrix is ​​represented as follows:

[0118]

[0119] The above data is normalized, and the surface topography parameter tensor and the two-dimensional matrix of line layout information are mapped to the interval [0, 1]. Then, a dual-channel convolutional neural network is constructed, and the normalized data is divided into a training set and a validation set.

[0120] In the training process, the 3-layer convolutional network of the first channel processes the surface topography parameter tensor to extract surface features of different scales; the graph attention network of the second channel processes the line layout information two-dimensional matrix to generate edge feature vectors. After splicing the extracted features, the feature relationship is learned through the full connection layer, and then the fracture probability of each edge is calculated through the output layer, such as the fracture probability of edge 1-2 being 0.1, the fracture probability of edge 2-3 being 0.05, etc., and the resistance fluctuation coefficient of each edge, such as the resistance fluctuation coefficient of edge 1-2 being 0.2, etc. Cross-entropy and mean square error loss functions are defined to update network parameters constantly. After multiple iterations and evaluation and adjustment using the validation set, the trained and optimized dual-channel convolutional neural network is obtained for predicting the fracture probability and resistance fluctuation coefficient of each edge under new surface topography and line layout.

[0121] The dual-channel convolutional neural network designs different feature extraction methods for surface topography parameters and line layout information. The first channel uses a convolutional network to extract multi-scale features of the surface topography, and the second channel uses a graph attention network to obtain local features of the line connection, which can efficiently and comprehensively extract key information. Compared with a single network structure, the extracted features are more targeted and rich. By fusing the two types of features and using appropriate activation functions and loss functions for training, the fracture probability and resistance fluctuation coefficient of each edge can be accurately output. This provides accurate quantitative indicators for line layout optimization, enabling the adaptive layout module to more accurately adjust line parameters and effectively reduce the risk of line fracture and resistance fluctuation.

[0122] The network can be trained and optimized according to different surface topography and line layout information, and can adapt to the needs of various actual production scenarios. Whether it is a flexible substrate of different materials or a line layout of different complexity, it can be effectively analyzed and predicted by adjusting the data and network parameters. Accurate prediction results can help production enterprises to discover potential line quality problems in advance, optimize corona treatment processes and printing line layout schemes, reduce the rate of defective products caused by line faults, improve production efficiency, reduce production costs, and enhance the competitiveness of enterprises in the market.

[0123] In a preferred embodiment of the present application, according to the fracture probability and resistance fluctuation coefficient, the line parameters are dynamically adjusted by a path planner containing a group incremental learning algorithm, which can include:

[0124] A path planner based on a group incremental learning algorithm is constructed, and the parameters of the group incremental learning algorithm are set, including the group size, learning rate, and step size of incremental learning;

[0125] The preset line layout information is organized to determine the connection relationship and node position of the line, and to determine the adjustment range of the line parameters, including the minimum and maximum values of the width, thickness, and spacing parameters of the line;

[0126] According to the preset line layout information and the adjustment range of the line parameters, an initial line parameter combination group is randomly generated, and each combination represents a line parameter setting scheme;

[0127] For each line parameter combination in the group, a corresponding fitness index is calculated in combination with the fracture probability and the resistance fluctuation coefficient; the line parameter combination includes the line width, the line thickness, the line spacing, the line length, the conductor material, the insulating material, the solder joint quality and the via parameter;

[0128] At the beginning of each iteration, new line performance data is obtained from the actual printing process, and the new line performance data and the line parameter combination in the current group and the corresponding fitness index are input into the path planner, the path planner is updated, and an updated path planner is obtained;

[0129] According to the updated path planner, the line parameter combination in the group is adjusted to obtain an adjusted line parameter combination, and the fitness index of the adjusted line parameter combination is recalculated;

[0130] The process of obtaining new line performance data, updating the path planner, adjusting the line parameter combination and recalculating the fitness index is repeated until a preset maximum iteration number is reached, and a final line parameter combination is determined from the group according to the fitness index of the line parameter combination;

[0131] According to the final line parameter combination, the actual parameters of the printed line are dynamically adjusted.

[0132] In the embodiments of the present application, a path planner based on a population incremental learning algorithm is realized by using a programming tool (such as Python combined with a related machine learning library).

[0133] Setting parameters:

[0134] Group size: Determine the number of line parameter combinations participating in learning and optimization, for example, setting the group size to 50 means that 50 different line parameter setting schemes participate in the iterative optimization process of the algorithm.

[0135] Learning rate: Control the step size of each parameter update, for example, set the learning rate to 0.01.

[0136] Incremental learning step: Determine the amount of new data introduced for learning each time, for example, set the incremental learning step to 10, that is, perform incremental learning once every 10 groups of new line performance data.

[0137] The preset line layout information is analyzed in detail to determine the connection relationship between each node in the line, for example, by using an adjacency matrix or other graph structure representation method to record the connection between nodes. At the same time, the position coordinates of each node on the PET film are accurately determined. According to the production process and product requirements, the adjustment range of the line parameters is determined. For the line width, the minimum value is 5 , and the maximum value is 20 ; the minimum value of the line thickness is 1 , and the maximum value is 5 ; the minimum value of the line spacing is 10 , and the maximum value is 30 , etc.

[0138] According to the preset line layout information and the determined line parameter adjustment range, a random number generation method (such as a uniformly distributed random number generator) is used to randomly select values for each line parameter (line width, line thickness, line spacing, line length, conductor material, insulating material, solder joint quality, and via parameter) within its adjustment range, and combine them into a line parameter combination. Repeat this process to generate an initial line parameter combination group with a group size. For example, if the group size is 50, 50 different line parameter combinations are generated, each representing an initial line parameter setting scheme. Each line parameter combination contains information such as line width, line thickness, line spacing, line length, conductor material, insulating material, solder joint quality, and via parameter. For each line parameter combination in the group, calculate the fitness index according to the formula (such as the formula mentioned earlier that considers various factors) based on the edge breakage probability and resistance fluctuation coefficient output by the double-channel convolutional neural network. The fitness index reflects the degree of optimization of the line parameter combination considering factors such as breakage probability and resistance fluctuation coefficient.

[0139] At the beginning of each iteration, new line performance data is obtained from the actual printing process through online monitoring equipment (such as online impedance measurement instruments, microscopes, etc.), including but not limited to resistance values, line topography, etc. The newly obtained line performance data is input into the path planner along with the line parameter combinations in the current group and their corresponding fitness indices. The path planner analyzes the relationship between the new data and the existing data based on the group incremental learning algorithm, updates the internal model and parameters, and obtains an updated path planner. According to the updated path planner, the line parameter combinations in the group are adjusted. The adjustment method can be an optimization strategy based on the algorithm, such as crossover and mutation operations in genetic algorithms, or gradient descent methods, which change some parameter values in the line parameter combination to obtain adjusted line parameter combinations.

[0140] For the adjusted line parameter combination, the fitness index is recalculated according to the previous fitness index calculation formula combined with the fracture probability and resistance fluctuation coefficient to evaluate the pros and cons of the adjusted line parameter combination. The process of obtaining new line performance data, updating the path planner, adjusting the line parameter combination, and recalculating the fitness index is repeated until the maximum number of iterations (such as 100 iterations) is reached. After reaching the maximum number of iterations, the line parameter combination with the optimal fitness index (such as the smallest fitness index value) is selected as the final line parameter combination from the population according to the fitness index of the line parameter combination.

[0141] According to the determined final line parameter combination, the line width, line thickness, line spacing and other parameters in it are applied to the actual printed line production, and the actual parameters of the printed line are dynamically adjusted by adjusting the parameter settings of the printing equipment to optimize the quality and performance of the printed line.

[0142] Suppose a simple sensor line is to be printed on a PET film, and the preset line layout is a network structure containing 5 nodes and 4 edges.

[0143] A path planner is built using Python and related libraries, with a population size of 30, a learning rate of 0.005, and an incremental learning step size of 8. The line layout information is sorted to determine the connection relationship between nodes and the node position. The line parameter adjustment range is set as follows: the minimum value of the line width is 8 , the maximum value is 18 ; the minimum value of the line thickness is 2 , the maximum value is 4 ; the minimum value of the line spacing is , the maximum value is 25 . Through random number generation, 30 initial line parameter combinations are obtained, such as one combination with a line width of 12 , a line thickness of 3 , a line spacing of 18 , a line length (determined according to the layout), a conductor material (assumed to be silver), an insulating material (assumed to be polyimide), a good solder joint quality, and via parameters (such as via diameter 0.2mm). Combined with the fracture probability of each edge output by the double-channel convolutional neural network (such as edge 1-2 fracture probability 0.1, edge 2-3 fracture probability 0.08, etc.) and the resistance fluctuation coefficient (such as edge 1-2 resistance fluctuation coefficient 0.15, etc.), the fitness index of each line parameter combination is calculated.

[0144] At the beginning of each iteration, new resistance value data is obtained by the online impedance measuring instrument, and these data are input into the path planner together with the current 30 circuit parameter combinations and their adaptation indicators, and the path planner is updated. According to the updated path planner, the circuit parameter combinations are adjusted, such as adjusting the circuit width of a certain combination to 14 , and then the adaptation indicators are recalculated. After 50 iterations, the combination with the optimal adaptation indicator (e.g., the smallest adaptation indicator value) is selected from the 30 circuit parameter combinations, and it is assumed that the finally determined combination is circuit width 13 , circuit thickness 3.2 , and circuit spacing 20 . The parameters of the final circuit parameter combination are applied to the printing equipment, and the actual parameters of the printed circuit are adjusted for production.

[0145] Through the population incremental learning algorithm and the path planner, the circuit parameter combinations can be dynamically adjusted according to the continuously obtained new circuit performance data, so that the circuit parameters can adapt to the changes and requirements in the actual printing process, improving the stability and reliability of the circuit. The adaptation indicators are calculated based on the fracture probability and the resistance fluctuation coefficient, and the circuit parameters are adjusted based on this, which can effectively reduce the risk of circuit fracture and the degree of resistance fluctuation, thereby improving the quality of the printed circuit and meeting the requirements of high-precision circuits (e.g., circuit precision ≤10 ). The problem of high scrap rate and low production efficiency caused by unreasonable circuit parameters in the traditional fixed layout scheme is avoided. Through continuous iteration and optimization, the final circuit parameter combination is quickly found, reducing the debugging time and material waste in the production process and improving the production efficiency. It can adapt to different circuit layouts and production environments, and for different PET film materials, different printing process requirements, etc., the printed circuit can be optimized by adjusting the circuit parameter combination, enhancing the versatility and adaptability of the system. Based on the circuit performance data obtained in the actual printing process, the optimization process is more scientific and reasonable, reducing the subjectivity and uncertainty of human experience judgment, and improving the accuracy and reliability of decision-making.

[0146] In another preferred embodiment of the present application, for each circuit parameter combination in the population, the corresponding adaptation indicator is calculated based on the fracture probability and the resistance fluctuation coefficient, including:

[0147] Obtaining the actual parameter values of the fracture probability, the resistance fluctuation coefficient, the circuit width, the circuit thickness, and the circuit spacing of each circuit unit and the corresponding adjustment range;

[0148] Determining the contribution degree of the fracture probability part according to the fracture probability of each circuit unit, and obtaining the resistance fluctuation coefficient contribution degree according to the resistance fluctuation coefficient of each circuit unit;

[0149] For each line unit, a middle position of the line width adjustment range and the line thickness adjustment range is determined, and according to the middle position, a deviation degree of the line width from the middle position and a deviation degree of the line thickness from the middle position are determined;

[0150] A middle position of the line spacing adjustment range is determined, and a deviation degree of the actual line thickness from the middle position is calculated;

[0151] The contribution degree of the fracture probability part, the resistance fluctuation coefficient contribution degree, the deviation degree of the line width from the middle position, the deviation degree of the line thickness from the middle position, and the deviation degree of the actual line thickness from the middle position are fused to obtain an adaptation index of the line parameter combination.

[0152] In the embodiment of the present application, the line width , the line thickness , and the line spacing of each line unit are accurately measured by high-precision measuring equipment such as an optical microscope, an electron probe, etc., so as to obtain the actual parameter values thereof. These measuring equipment can meet the requirements of high-precision measurement of line parameters, and ensure that the obtained data is accurate and reliable. The adjustment range of the line parameters, i.e., the minimum value , the maximum value of the line width, the minimum value , the maximum value of the line thickness, and the minimum value , the maximum value of the line spacing, is determined by the production process requirements and the design specifications of the product.

[0153] For example, considering the precision limit of the printing process, the currently used printing equipment and technology are difficult to accurately print a line with a width less than 5 , and when the line width exceeds 20 , the integration of the line and the miniaturization design of the product will be affected; from the physical properties of the material, the conductive ink and the substrate used may not be able to form a stable conductive line when the thickness is less than 1 , and when the thickness exceeds 5 , the cost and weight of the line will increase, and the flexibility of the product may be affected; based on the requirements of the circuit performance, the line spacing less than 10 is easy to increase the risk of short circuit between lines, and the spacing exceeding 30 will increase the resistance of the circuit, affecting signal transmission. Considering these factors, a reasonable value range of each line parameter is set to ensure that the produced line can meet the quality and performance standards.

[0154] First, the surface topography parameters (such as the protrusion height distribution of the PET film surface after corona treatment, pitch density, surface real contact area increment ratio, etc.) are collected and converted into tensor form; at the same time, the preset line layout information (such as the connection relationship of the line, the node position, etc.) is collected and converted into a two-dimensional matrix. The converted data is normalized and mapped to the [0, 1] interval, and then input into the double-channel convolutional neural network. The first channel of the network extracts multi-scale features of the surface topography parameters through a 3-layer convolutional network, and the second channel uses a graph attention network to aggregate the node neighborhood information of the line layout information to generate an edge feature vector. After the features extracted by the two channels are spliced, the relationship between the features is learned through a fully connected layer, and then input into the output layer. One neuron of the output layer uses the normalized exponential function (Softmax function) to calculate the fracture probability of each line unit wherein, is the exponential operation result with the natural constant as the base number and as the exponent, is the value corresponding to the th line edge (line unit), is the natural constant (approximately equal to 2.71828), is the total number of line units, is the index variable, is the exponential operation result with the natural constant as the base number and as the exponent, is the value corresponding to the th line edge (line unit) after the combined features processed by the fully connected layer, and another neuron cooperates with the linear activation function to obtain the resistance fluctuation coefficient wherein, is the slope parameter, is the intercept parameter, and is the feature value. In this way, the fracture probability and the resistance fluctuation coefficient of the line unit are indirectly obtained by using the powerful data analysis capability of the neural network.

[0155] The fracture probabilities of all line units are added up to obtain , which is then multiplied by the weight of the fracture probability to obtain the contribution degree of the fracture probability part to the adaptation index . The weight reflects the importance of the fracture probability in the overall evaluation. The greater the weight, the more critical the fracture probability is to the line quality.

[0156] Similarly, the resistance fluctuation coefficients of all line units are added up to obtain Then multiply by the weight of the resistance fluctuation coefficient Get the contribution of the resistance fluctuation coefficient part The weight Reflects the importance of the resistance fluctuation coefficient in evaluating the line quality.

[0157] For each line unit, first calculate the middle position of the line width adjustment range, that is Subtract the actual width of the line unit Take the absolute value of the difference Divide by half the width adjustment range difference Get a relative deviation ratio Multiply this ratio by the line width impact coefficient Sum the results of all line units Finally, multiply by the weight of the line width Get the contribution value of the line width deviation from the middle position to the adaptation index The impact coefficient Indicates the difference in the impact of line width changes on line performance for different line units. Assuming there are 3 line units, the specific value is set as follows:

[0158] For line unit 1, if it is in a relatively flat area of the film surface and is a normal signal transmission line, the functional importance is generally, and can be set =0.3. This means that the line width change of this line unit has relatively small impact on the line performance.

[0159] If line unit 2 bears the key power transmission function and the film surface in the area is micro-convex with complex surface topography, set =0.7 to highlight the greater impact of line width change on performance.

[0160] If line unit 3 is an auxiliary line and the surface topography in the area is relatively stable, set 0.2.

[0161] The same as the line width calculation method, first calculate the middle position of the line thickness adjustment range Get the actual thickness The deviation ratio of the actual thickness from the middle position Multiply by the line thickness impact coefficient Sum, then multiply by the line thickness weight Get the contribution value of the line thickness deviation from the middle position to the adaptation index The line thickness impact coefficient The value can be set according to the function of the circuit in the circuit, the stress condition of the area where the circuit is located, and other factors. Assuming that there are 4 circuit units on a certain printed circuit board, the specific value is set as follows:

[0162] Circuit unit 1 is a high-frequency signal transmission circuit, which is sensitive to the change of the thickness of the circuit, and is located in an area where stress is easily concentrated. Set =0.8, which indicates that the change of the thickness of the circuit has a great influence on the performance of the circuit.

[0163] Circuit unit 2 is mainly used for power transmission, and the stability of the thickness is related to the power carrying capacity. The stress in the area is moderate, and the value is set as =0.6, which means that the change of the thickness of the circuit has a relatively obvious influence on its performance.

[0164] Circuit unit 3 is a control signal circuit, which is generally important in function and has less stress in the area. Set =0.4, which means that the change of the thickness of the circuit has a moderate influence on its performance.

[0165] Circuit unit 4 is an auxiliary circuit, which has less influence on performance and is in a stable environment in the area. Set , which means that the change of the thickness of the circuit has a weak influence on its performance.

[0166] The middle position of the adjustment range of the circuit spacing is calculated , and the half of the difference of the adjustment range of the spacing is divided by to obtain the relative deviation ratio . Multiply the ratio by the circuit spacing influence coefficient , sum up, and then multiply by the circuit spacing weight to obtain the contribution value of the deviation of the circuit spacing from the middle position to the adaptation index . The circuit spacing influence coefficient can be set according to factors such as circuit function, surrounding electromagnetic environment, and layout complexity. Assuming that the circuit of a certain flexible electronic device contains 3 circuit units, the specific value is set as follows:

[0167] Circuit unit 1: This circuit unit is responsible for high-frequency signal transmission, and is located in a dense circuit area. The change of the circuit spacing easily causes signal crosstalk and electromagnetic interference, which has a significant influence on the performance of the circuit. Set =0.8 to highlight the greater influence of the change of the spacing on the performance.

[0168] Circuit unit 2: This circuit unit is used for power transmission, and there are many sensitive elements around. The circuit spacing needs to consider insulation and electromagnetic compatibility. Set =0.6, which means that the change of the spacing has a moderate influence on the performance.

[0169] Line unit 3: As an auxiliary line, it mainly undertakes simple control signal transmission, the layout of the area is loose, the electromagnetic environment is good, the change of line spacing has little effect on performance, and the line width is set = 0.2.

[0170] Add the partial contribution of the fracture probability, the resistance fluctuation coefficient contribution, the line width deviation contribution value, the line thickness deviation contribution value, and the line spacing deviation contribution value obtained by the above calculation, that is The result obtained is the adaptation index of the line parameter combination. The lower the value, the better the line parameter combination.

[0171] Among them, Reflects the importance of line fracture probability in calculating the adaptation index. The greater the weight, the greater the impact of fracture probability on the final result when evaluating line parameter combinations. The value of w1 is between [0, 1], and satisfies . Reflects the degree of attention to line resistance fluctuation in the calculation of the adaptation index. The greater the weight value, the more significant the impact of the resistance fluctuation coefficient on the final adaptation index. Used to measure the importance of the line width deviating from the center value of its adjustment range in the adaptation index. Reflects the impact of the line thickness deviating from the center value of its adjustment range on the adaptation index. Indicates the importance of the line spacing deviating from the center value of its adjustment range in calculating the adaptation index.

[0172] The adaptation index considers fracture probability, resistance fluctuation coefficient, and the deviation of line width, thickness, and spacing, evaluates line parameter combinations from multiple dimensions, avoids the one-sidedness of single index evaluation, and can comprehensively and accurately reflect the actual quality and performance level of the line. By quantifying the contribution of each factor to the adaptation index, the impact of different parameters on line quality is clear. When optimizing line parameters, the adaptation index can be used to adjust weights and parameter values, making line parameter combinations more suitable for actual needs, effectively reducing line fracture, resistance abnormalities, and other problems, improving line precision and stability, and meeting the line precision ≤10 ​The adaptation index can adapt to the differences in the micro-topography of the film surface after corona treatment, different production process requirements and other complex situations considering the adjustment range of the line parameters and the influence coefficient of different line units. Whether the material properties change or the process parameters fluctuate, the rationality of the line parameter combination can be evaluated through the index to enhance the adaptability of the system in different production scenarios. Accurate adaptation index calculation helps to select the better line parameter combination before production, reduces the trial and error cost and material waste caused by unreasonable parameters. At the same time, the rate of defective products is reduced, the production cycle is shortened, and the production efficiency is improved, thereby reducing the production cost of enterprises and improving the competitiveness of enterprises in the market.

[0173] In a preferred embodiment of the present application, the resistance value of the printed circuit is monitored in real time by an online impedance measuring instrument. When the relative deviation between the actual value and the theoretical value is greater than or equal to 5%, the parameter feedback mechanism is triggered, that is, the coordinates of the abnormal area are fed back to the detection module for local re-measurement, and the re-measurement data are input into the dynamic prediction module to update the weight matrix, which can include:

[0174] The online impedance measuring instrument is connected to the printed circuit, and the online impedance measuring instrument measures the resistance value of the circuit in real time according to the set sampling frequency, continuously acquiring the actual resistance value data of the printed circuit;

[0175] The theoretical resistance design value of the printed circuit is obtained from the database, and the relative deviation value between the actual resistance value measured in real time and the theoretical value is calculated;

[0176] The relative deviation value is compared with the set threshold value 5%. If the relative deviation value is less than 5%, the real-time resistance value monitoring is continued. If the relative deviation value is greater than or equal to 5%, the parameter feedback mechanism is immediately triggered;

[0177] When the parameter feedback mechanism is triggered, the coordinates of the abnormal resistance value area are located according to the layout information of the printed circuit and the measurement point position of the online impedance measuring instrument;

[0178] The detection module receives the coordinates of the abnormal resistance value area, starts the local re-measurement program, controls the detection equipment to move to the abnormal area, and re-measures the resistance value of the printed circuit in the abnormal area at multiple points and multiple angles. During the re-measurement process, the environmental parameters are collected, and the re-measured resistance value data and the environmental parameters are integrated into a re-measurement data set;

[0179] The re-measurement data set is input into the dynamic prediction module, the influence of the re-measurement data on the current double-channel convolutional neural network is analyzed, and the weight matrix is updated by recalculating the weights of each parameter in the double-channel convolutional neural network, including the weights of the convolutional layer and the weights of the fully connected layer.

[0180] In the embodiment of the present application, the online impedance measuring instrument is electrically connected to the designated measurement point of the printed circuit through a probe or a clamp, etc., to ensure the integrity and accuracy of the measurement circuit. For example, for a complex multi-layer printed circuit board, it is necessary to accurately connect to the test pads of the target circuit layer. According to the working frequency of the printed circuit, the speed of the resistance value change, and the production efficiency requirement, a suitable sampling frequency is set in the control interface of the online impedance measuring instrument or the supporting software. For example, for a high-frequency rapidly changing circuit, the sampling frequency can be set to 100 times per second to ensure that the slight fluctuation of the resistance value can be captured in time; for a low-frequency stable circuit, the sampling frequency can be set to 10 times per second to reduce the data processing amount while ensuring the monitoring effect. The online impedance measuring instrument continuously measures the resistance value of the printed circuit according to the set sampling frequency, and transmits and stores the actual resistance value data obtained each time to the data buffer area or the database of the system in real time.

[0181] The theoretical resistance design value of the printed circuit is retrieved from the database. This value is a standard reference value obtained by simulation according to the design parameters of the circuit, such as circuit length, width, material resistivity, etc. For example, during the design stage, according to the copper foil thickness, trace length and width, combined with the resistivity of copper, the theoretical resistance value is calculated and stored in the corresponding circuit design parameter table in the database. The relative deviation value of the actual resistance value measured in real time and the theoretical value is calculated using the formula: relative deviation = (actual resistance value - theoretical resistance value) / theoretical resistance value. The calculated relative deviation value is compared with the pre-set threshold value of 5%. This process is automatically executed by the logic judgment module of the system, and is realized through conditional statements (such as if-else statements).

[0182] If the relative deviation value is <5%, it means that the resistance value of the current printed circuit is within the normal fluctuation range, and real-time resistance value monitoring is continued according to the set sampling frequency; if the relative deviation value is ≥5%, it is determined that the resistance value is abnormal, and the parameter feedback mechanism is triggered immediately to inform the system to enter the abnormal processing flow. Combined with the layout information of the printed circuit (such as the topology of the circuit, the node coordinates, the trace path, etc.) and the measurement point position information (the coordinates or number corresponding to each measurement point) of the online impedance measuring instrument, a spatial positioning algorithm is used for analysis. For example, through the coordinates of the measurement points and the connection relationship of the circuit layout, it is determined which circuit sections are associated with the abnormal measurement value. According to the analysis result, the coordinate range of the resistance value abnormal area is accurately calculated. For a regular rectangular circuit area, the coordinates of the upper left corner and the lower right corner can be determined; for an irregular area, the coordinates of multiple key points can be used to describe its boundary.

[0183] After receiving the coordinates of the abnormal resistance value area, the detection module automatically starts the local retest program. By controlling the motion control system of the detection equipment (such as high-precision microscopes, probe stations, etc.), the measurement probe or lens of the detection equipment is moved to the specified position of the abnormal area. In the abnormal area, multiple point and angle resistance value retests are performed according to the preset measurement strategy. For example, in a rectangular abnormal area, multiple measurement points are selected in a grid-like distribution, and resistance value measurement is performed from different angles (such as vertical, 45 。 inclination, etc.) at each measurement point to obtain more comprehensive resistance value data. During the retest process, environmental parameters such as temperature, humidity, and air pressure are synchronously collected. These environmental parameters may affect the resistance value, so they need to be recorded. Environmental parameters can be collected in real time through devices such as temperature and humidity sensors, air pressure sensors, etc. connected to the system. The resistance value data obtained by retesting and the collected environmental parameters are integrated to form a retest data set containing multi-dimensional information. For example, the resistance value, measurement time, measurement angle, and corresponding environmental temperature, humidity, etc. information of each measurement point are combined into a data record, and all records constitute the retest data set.

[0184] The retest data set is input into the dynamic prediction module, which first preprocesses and extracts features from the retest data, analyzes the differences between the retest data and the current double-channel convolutional neural network model prediction results, and evaluates the influence of the retest data on the model. For example, by calculating the statistical characteristics (such as mean, variance, etc.) of the retest data, and comparing them with the corresponding characteristics predicted by the model, it is determined whether the prediction accuracy of the model is affected. According to the analysis results, use optimization algorithms (such as stochastic gradient descent, Adam algorithm, etc.) to recalculate the weights of each parameter in the double-channel convolutional neural network, including the convolution kernel weights of the convolution layer and the weights of the fully connected layer. The specific calculation process is to adjust the weight value according to the gradient of the loss function (such as cross-entropy loss function, mean square error loss function, etc.) on the weight, so that the prediction error of the model on the retest data is minimized. Through continuous iterative calculation and update, the updated weight matrix is finally obtained, so that the model can better adapt to the changes in actual production and improve the accuracy and reliability of prediction.

[0185] Suppose on a flexible printed circuit production line, a printed circuit for a sensor is produced.

[0186] The online impedance measuring instrument is connected to the test points of the printed circuit, and the sampling frequency is set to 50 times per second. The online impedance measuring instrument continuously measures the resistance value of the circuit, such as measuring the actual resistance value of 12.3Ω at a certain time, and storing the data in the database. The theoretical resistance design value of the printed circuit is obtained from the database as 10Ω, and the relative deviation is calculated as 23% according to the formula. Since 23% > 5%, the parameter feedback mechanism is triggered immediately, the regular monitoring is stopped, and the abnormal processing flow is entered. Combined with the circuit layout information and the measurement point position, it is determined that the abnormal area is located in a certain section of the circuit, and the coordinate range is the upper left corner (10mm, 15mm) and the lower right corner (20mm, 25mm). After receiving the coordinates, the detection module controls the high-precision probe table to move to the abnormal area. Select 9 measurement points in this area, measure the resistance value from the vertical and 45 。 Two angles, and collect the data of the environmental temperature 25℃ and humidity 50% at the same time. Integrate all the measured resistance value data and environmental parameters to form a re-measurement data set. Input the re-measurement data set into the dynamic prediction module, and the module analysis finds that the prediction of the model in this area has a large deviation. Use the Adam algorithm to recalculate the weights of the double-channel convolutional neural network, and after multiple iterations of adjustment, update the weight matrix of the model, so that the model is more accurate in predicting the resistance value of this type of circuit.

[0187] By measuring the resistance value in real time according to the set frequency through the online impedance measuring instrument, the small changes in the resistance value of the printed circuit can be captured in time, which improves the real-time and accuracy of the monitoring compared with manual sampling inspection or regular measurement, and ensures the continuous monitoring of the circuit quality. When the relative deviation between the actual value and the theoretical value of the resistance reaches the set threshold, the parameter feedback mechanism is triggered immediately, which can quickly locate the abnormal area, avoid the problem circuit flowing into the next process, reduce the production delay and resource waste caused by the circuit quality problem, and improve the production efficiency.

[0188] According to the circuit layout and measurement point information, the abnormal area is accurately located, and multi-point and multi-angle re-measurement is performed, and the environmental parameters are collected, which can comprehensively and accurately obtain the actual situation of the abnormal area, avoid misjudgment caused by single measurement point or measurement angle, and improve the accuracy of problem diagnosis. Input the re-measurement data into the dynamic prediction module to update the weight matrix, so that the double-channel convolutional neural network can continuously learn and adapt to the actual changes in the production process, improve the accuracy and reliability of the model in predicting the resistance value of the circuit, and provide a more scientific basis for circuit layout optimization and production process adjustment, thereby fundamentally improving the quality and stability of the printed circuit. This process forms a complete closed-loop quality control system from real-time monitoring, abnormal detection, accurate positioning to model optimization, which can continuously improve the production process and continuously improve product quality, and enhance the competitiveness of enterprises in the market.

[0189] The embodiment of the present application also provides a computing device, comprising a processor, a memory storing a computer program, the computer program being executed by the processor to implement the system as described above. All implementation manners in the above system embodiment are suitable for this embodiment and can achieve the same technical effects.

[0190] The embodiment of the present application also provides a computer readable storage medium storing instructions, which, when executed on a computer, cause the computer to implement the system as described above. All implementation manners in the above system embodiment are suitable for this embodiment and can achieve the same technical effects.

[0191] The above is the preferred embodiment of the present application, it should be pointed out that, for those skilled in the art, without departing from the principles of the present application, can make a number of improvements and refinements, these improvements and refinements should also be considered as the protection scope of the present application.

Claims

1. A neural network model based adaptive corona printing layout optimization system, characterized in that, The method comprises the following steps: a detection module for scanning the surface of the PET film after corona treatment by a confocal microscope array and a laser displacement sensor to obtain a set of surface fluctuation characteristic values; a feature processing module for calculating the surface micro-roughness change value before and after treatment according to the set of surface fluctuation characteristic values, and extracting the regular protruding structure characteristic parameters generated by the corona treatment, including protrusion height distribution, pitch density and surface real contact area increment ratio; a dynamic prediction module for analyzing the surface topography parameters and preset circuit layout information by using a double-channel convolutional neural network to output the fracture probability of each side and the resistance fluctuation coefficient, comprising: collecting the surface topography parameters and the preset circuit layout information, and converting the surface topography parameters into tensor form and the circuit layout information into a two-dimensional matrix; normalizing the converted surface topography parameters and the preset circuit layout information, mapping the converted surface topography parameters and the preset circuit layout information to the [0, 1] interval to obtain normalized data; constructing a double-channel convolutional neural network according to the normalized data, and dividing the normalized data into a training set and a validation set; inputting the training set into the double-channel convolutional neural network to extract multi-scale features of the surface topography parameters and local features of the circuit connection; splicing the multi-scale features of the surface topography parameters and the local features of the circuit connection to obtain combined features, and processing the combined features through a fully connected layer to learn the relationship between the features; inputting the combined features processed by the fully connected layer into the output layer of the double-channel convolutional neural network, using a neuron to calculate the fracture probability of each side of the printed circuit combined with a normalized exponential function; the fracture probability output value is in the [0, 1] interval and the sum of all side probabilities is 1; using another neuron with a linear activation function to obtain the resistance fluctuation coefficient; defining cross-entropy and mean square error loss functions, and calculating the gradient of the loss function on the parameters of the double-channel convolutional neural network, and updating the parameters of the double-channel convolutional neural network; evaluating the parameters of the double-channel convolutional neural network using the validation set, and adjusting the structure and hyperparameters of the double-channel convolutional neural network according to the evaluation results until a preset number of iterations is reached to obtain a trained and optimized double-channel convolutional neural network; inputting the normalized surface topography parameters and the preset circuit layout information into the trained and optimized double-channel convolutional neural network to output the fracture probability of each side and the resistance fluctuation coefficient; an adaptive layout module for dynamically adjusting the circuit parameters through a path planner containing group incremental learning algorithm according to the fracture probability and the resistance fluctuation coefficient; a feedback control module for monitoring the resistance value of the printed circuit in real time through an online impedance measuring instrument, and triggering a parameter return mechanism when the relative deviation between the actual value and the theoretical value of the resistance is greater than or equal to 5%, that is, feeding back the coordinates of the abnormal area to the detection module for local re-measurement, and inputting the re-measurement data into the dynamic prediction module to update the weight matrix.

2. The neural network model based self-adaptive corona printing layout optimization system according to claim 1, wherein, According to the set of surface fluctuation characteristic values, the surface micro-roughness change value before and after treatment is calculated, and the regular protruding structure characteristic parameters generated by the corona treatment are extracted, including protrusion height distribution, pitch density and surface real contact area increment ratio, comprising: The surface relief characteristic value set obtained after the corona treatment is integrated with the PET film surface three-dimensional topography data before the corona treatment and the corresponding characteristic value set to form a data comparison group; The surface height data before and after treatment in the data comparison group are analyzed, and the root mean square roughness of the surface height before and after treatment is calculated to obtain the surface micro-roughness change value; According to the surface micro-roughness change value, the three-dimensional topography image constructed by the surface relief characteristic value set is subjected to image morphological processing operation, and a threshold value is set according to the height difference between the protrusions and the base to separate the protrusion structures from the surface topography to form independent protrusion structure regions; Each protrusion structure is traversed, and the vertical distance between the highest point and the base plane, i.e. the protrusion height, is measured; the height values of each protrusion structure are summarized, grouped according to certain height intervals, and the proportion of the number of protrusions in each interval to the total number of protrusions is calculated to generate a protrusion height distribution table; An effective distance threshold is set, and the number of adjacent protrusions of each protrusion structure within the threshold range is counted to obtain the average spacing density; According to the average spacing density and the surface relief characteristic value set, three-dimensional models of the PET film surfaces before and after treatment are respectively constructed by using a three-dimensional modeling software, and the surface real contact area increment ratio before and after treatment is calculated by comparing the surface real contact areas in the three-dimensional models.

3. The neural network model based self-adaptive corona printing layout optimization system according to claim 2, wherein, The double-channel convolutional neural network comprises: A first channel: inputting a surface topography parameter tensor in a training set, and extracting multi-scale features of the surface topography parameters through a 3-layer convolutional network; A second channel: inputting a two-dimensional matrix of preset line layout information, and generating an edge feature vector by aggregating node neighborhood information through a graph attention network.

4. The neural network model based self-adaptive corona printing layout optimization system according to claim 3, wherein, According to the fracture probability and the resistance fluctuation coefficient, a path planner based on a group incremental learning algorithm is constructed, and parameters of the group incremental learning algorithm are set, including a group size, a learning rate, and a step length of incremental learning. The preset line layout information is sorted to determine the connection relationship and node position of the line, and the adjustment range of the line parameters is determined, including the minimum and maximum values of the width, thickness and spacing parameters of the line. According to the preset line layout information and the adjustment range of the line parameters, an initial line parameter combination group is randomly generated, and each combination represents a line parameter setting scheme. For each line parameter combination in the group, the corresponding fitness index is calculated in combination with the fracture probability and the resistance fluctuation coefficient. At the beginning of each iteration, new line performance data are obtained from the actual printing process, and the new line performance data, the line parameter combinations in the current group and the corresponding fitness indexes are input into the path planner to update the path planner and obtain an updated path planner. According to the updated path planner, the line parameter combinations in the group are adjusted to obtain adjusted line parameter combinations, and the fitness indexes of the adjusted line parameter combinations are recalculated. ​ The process of acquiring new line performance data, updating the path planner, adjusting the line parameter combination and recalculating the fitness index is repeated until a preset maximum number of iterations is reached, and the final line parameter combination is determined from the population according to the fitness index of the line parameter combination; According to the final line parameter combination, the actual parameters of the printed circuit are dynamically adjusted.

5. The neural network model based adaptive corona printing layout optimization system according to claim 4, wherein, The line parameter combination includes line width, line thickness, line spacing, line length, conductor material, insulating material, solder joint quality and via parameters.

6. The neural network model based self-adaptive corona printing layout optimization system according to claim 5, wherein, For each line parameter combination in the population, the corresponding fitness index is calculated by combining the fracture probability and the resistance fluctuation coefficient, including: Obtain the actual parameter values of the fracture probability, resistance fluctuation coefficient, line width, line thickness, line spacing and corresponding adjustment range of each line unit; According to the fracture probability of each line unit, the contribution degree of the fracture probability part is determined; according to the resistance fluctuation coefficient of each line unit, the resistance fluctuation coefficient contribution degree is obtained; For each line unit, the middle position of the line width and line thickness adjustment range is determined, and according to the middle position, the deviation degree of the line width from the middle position and the deviation degree of the line thickness from the middle position are determined; Determine the middle position of the line spacing adjustment range, and calculate the deviation degree of the actual line thickness from the middle position; Fuse the contribution degree of the fracture probability part, the resistance fluctuation coefficient contribution degree, the deviation degree of the line width from the middle position, the deviation degree of the line thickness from the middle position, and the deviation degree of the actual line thickness from the middle position to obtain the fitness index of the line parameter combination.

7. The neural network model based adaptive corona printing layout optimization system according to claim 6, wherein, Through the online impedance measuring instrument, the resistance value of the printed circuit is monitored in real time, and when the relative deviation between the actual value and the theoretical value is ≥5%, the parameter feedback mechanism is triggered, that is, the coordinates of the abnormal area are fed back to the detection module for local retest, and the retest data are input into the dynamic prediction module to update the weight matrix, including: Connect the online impedance measuring instrument with the printed circuit, and the online impedance measuring instrument measures the line resistance value in real time according to the set sampling frequency, continuously obtains the actual resistance value data of the printed circuit; Obtain the theoretical resistance design value of the printed circuit from the database, and calculate the relative deviation value between the actual resistance value measured in real time and the theoretical value; Compare the relative deviation value with the set threshold value 5%, if the relative deviation value <5%, continue to monitor the real-time resistance value; if the relative deviation value ≥5%, immediately trigger the parameter feedback mechanism; When the parameter feedback mechanism is triggered, according to the layout information of the printed circuit and the measurement point position of the online impedance measuring instrument, the coordinates of the resistance value abnormal area are located; The detection module receives the coordinates of the resistance value abnormal area, starts the local retest program, controls the detection equipment to move to the abnormal area, and retests the resistance value of the printed circuit in the abnormal area from multiple points and multiple angles; In the retest process, environmental parameters are collected, and the retested resistance value data and environmental parameters are integrated into a retest data set; The retest data set is input into the dynamic prediction module, influence of the retest data on the current double-channel convolutional neural network is analyzed, and weight matrices are updated by recalculating weights of each parameter in the double-channel convolutional neural network, including weights of a convolutional layer and weights of a fully connected layer.

8. A computing device, comprising: Comprise: One or more processors; A storage device for storing one or more programs, when the one or more programs are executed by the one or more processors, so that the one or more processors implement the system as claimed in any one of claims 1 to 7.

9. A computer-readable storage medium, characterized in that, The computer readable storage medium stores a program, and the program is executed by the processor to implement the system as claimed in any one of claims 1 to 7.

Images