Alumina ceramic substrate and method of manufacturing the same

Abstract

The embodiment of the present specification provides an alumina ceramic substrate and a manufacturing method thereof, wherein the manufacturing method of the alumina ceramic substrate comprises the following steps: selecting a pressing plate matched with the alumina ceramic substrate, and fixedly connecting the alumina ceramic substrate and the pressing plate; placing the alumina ceramic substrate and the pressing plate in a reaction chamber; determining a set heating rate in the reaction chamber according to parameter information of the alumina ceramic substrate, and performing a sintering treatment on the alumina ceramic substrate. By using the above technical solution, the warping of the alumina ceramic substrate can be reduced, the qualified rate is improved, and the working performance of a device using the alumina ceramic substrate is improved.

Description

TECHNICAL FIELD

[0001] The embodiment of the present specification relates to the technical field of alumina ceramic substrate, and particularly relates to an alumina ceramic substrate and a manufacturing method thereof. BACKGROUND

[0002] The alumina ceramic substrate is a sheet-shaped material made of alumina with specific purity. The alumina ceramic substrate is widely used in electronic, optical and mechanical fields due to high strength, high hardness and excellent thermal stability.

[0003] In recent years, with the development of microelectronic and optoelectronic technologies, the demand for alumina ceramic substrates is increasing, which requires production technology to provide substrates with accurate size, high surface finish and no warping.

[0004] However, when manufacturing the alumina ceramic substrate, warping problems are prone to occur, especially during high-temperature sintering. The warped substrate not only affects the appearance and use performance of the product, but also reduces the production efficiency and output rate.

[0005] Under this background, how to reduce the warping problem of the alumina ceramic substrate becomes particularly important and needs to be solved by those skilled in the art. SUMMARY

[0006] Therefore, the embodiment of the present specification provides an alumina ceramic substrate and a manufacturing method thereof, which can reduce the warping of the alumina ceramic substrate, improve the qualified rate, and further improve the working performance of the device using the alumina ceramic substrate.

[0007] Firstly, the embodiment of the present specification provides a manufacturing method of an alumina ceramic substrate, comprising:

[0008] selecting a pressing plate matched with the alumina ceramic substrate, and fixedly connecting the alumina ceramic substrate and the pressing plate;

[0009] placing the alumina ceramic substrate and the pressing plate in a reaction chamber;

[0010] determining a set heating rate in the reaction chamber according to parameter information of the alumina ceramic substrate, and performing sintering treatment on the alumina ceramic substrate.

[0011] Optionally, the step of determining the set heating rate in the reaction chamber according to the parameter information of the alumina ceramic substrate, and performing sintering treatment on the alumina ceramic substrate, comprises:

[0012] determining the set heating rate in the reaction chamber according to size information and thickness information of the alumina ceramic substrate;

[0013] According to the set temperature rising rate, a temperature rising operation is performed on the reaction chamber to perform a sintering process on the alumina ceramic substrate.

[0014] Optionally, the determining the set temperature rising rate in the reaction chamber according to the parameter information of the alumina ceramic substrate further comprises:

[0015] The morphology change of the alumina ceramic substrate in the sintering process is determined, and the morphology change comprises a morphology and a morphology change rate of the alumina ceramic substrate.

[0016] According to the morphology change of the alumina ceramic substrate, the temperature rising rate in the reaction chamber is adjusted.

[0017] Optionally, the determining the morphology change of the alumina ceramic substrate in the sintering process comprises:

[0018] A preset first neural network model is used to determine the morphology change of the alumina ceramic substrate in the sintering process, wherein the first neural network model is based on a temperature distribution on a surface of the alumina ceramic substrate, and a temperature rising rate and a historical morphology change corresponding to the temperature distribution.

[0019] Optionally, the first neural network model is based on the temperature distribution on the surface of the alumina ceramic substrate, and the temperature rising rate and the historical morphology change corresponding to the temperature distribution, comprising:

[0020] The obtained temperature distribution on the surface of the alumina ceramic substrate, and the temperature rising rate and the historical morphology corresponding to the temperature distribution are preprocessed to obtain a data set;

[0021] The data set is divided into a training data set, a validation data set and a test data set;

[0022] The training data set is input into a preset first pre-training model, the first pre-training model is trained to obtain the first neural network model;

[0023] The validation data set is input into the first neural network model to obtain an actual morphology of the alumina ceramic substrate corresponding to each validation data set;

[0024] According to the actual morphology of the alumina ceramic substrate corresponding to each validation data set and the historical morphology of the alumina ceramic substrate contained in each validation data set, the first neural network model is adjusted;

[0025] The test data set is input into the adjusted first neural network model, and when the performance of the first neural network model meets a set performance, the training of the first neural network model is stopped.

[0026] Optionally, the adopting the preset first neural network model to determine the shape change of the alumina ceramic substrate in the sintering process comprises:

[0027] acquiring a current temperature and a current heating rate in the reaction chamber;

[0028] inputting the current temperature and the current heating rate in the reaction chamber into the first neural network model to determine a predicted shape and a predicted shape change rate of the alumina ceramic substrate at the current temperature and the current heating rate.

[0029] Optionally, the adjusting the heating rate in the reaction chamber according to the shape change of the alumina ceramic substrate comprises:

[0030] determining a set shape of the alumina ceramic substrate at the current temperature and the current heating rate based on the alumina ceramic substrate at the current temperature and the current heating rate by the first neural network model;

[0031] adopting a preset second neural network model to determine a set heating rate corresponding to the alumina ceramic substrate according to the set shape of the alumina ceramic substrate at the current temperature and the current heating rate, wherein the second neural network model is based on the shape change of the alumina ceramic substrate and the heating rate and the temperature in the reaction chamber corresponding to the shape change;

[0032] adjusting the heating rate in the reaction chamber according to the current heating rate and the set heating rate of the alumina ceramic substrate.

[0033] Optionally, the method for manufacturing the alumina ceramic substrate further comprises:

[0034] when it is determined that the predicted shape change of the alumina ceramic substrate at the current temperature and the current heating rate is greater than the set shape change, resetting the temperature and / or the heating rate in the reaction chamber.

[0035] Optionally, the method for manufacturing the alumina ceramic substrate further comprises:

[0036] determining the flatness of the alumina ceramic substrate after the sintering treatment.

[0037] Optionally, the determining the flatness of the alumina ceramic substrate after the sintering treatment comprises:

[0038] adopting a preset third neural network model to determine the flatness of the alumina ceramic substrate after the sintering treatment, wherein the third neural network model is based on the image information of the alumina ceramic substrate collected.

[0039] Optionally, the third neural network model is obtained based on image information of the alumina ceramic substrate, and comprises:

[0040] Obtain images of the alumina ceramic substrate from multiple perspectives after sintering treatment, and determine label information corresponding to each perspective image;

[0041] Preprocess the images from each perspective to obtain multiple image data;

[0042] Divide the multiple image data into a training image data set and a verification image data set;

[0043] In any round, output the training image data set to the third pre-trained model according to a set batch, train the third pre-trained model, and obtain the third neural network model;

[0044] Input the verification image data set into the third neural network model to obtain actual label information corresponding to each training image data;

[0045] Adjust the third neural network model according to the actual label information corresponding to the verification image data set and the label information contained in the verification image data set.

[0046] Optionally, the obtaining of the images of the alumina ceramic substrate from multiple perspectives after sintering treatment and the determination of label information corresponding to each perspective image comprise:

[0047] Generate a three-dimensional image model of the alumina ceramic substrate after sintering treatment;

[0048] Render a two-dimensional image corresponding to the three-dimensional image model from multiple perspectives to obtain the images of the alumina ceramic substrate from multiple perspectives after sintering treatment;

[0049] Determine label information corresponding to each perspective image according to the relationship between the parallelism of each perspective image and a set parallelism, wherein the label information comprises normal substrate and defective substrate.

[0050] Optionally, the determination of the flatness of the alumina ceramic substrate after sintering treatment using the preset third neural network model comprises:

[0051] Generate multiple perspective images of the alumina ceramic substrate after sintering treatment;

[0052] Input the images from each perspective into the third neural network model to obtain a flatness score corresponding to each perspective image, wherein the flatness score is used to represent the flatness of each perspective image.

[0053] Optionally, the method for manufacturing the alumina ceramic substrate further comprises:

[0054] According to the flatness of the image from each perspective and the set parallelism reference, the qualification of the alumina ceramic substrate is determined.

[0055] Optionally, the set parallelism reference is obtained by:

[0056] A three-dimensional image model of the alumina ceramic substrate after sintering treatment is obtained.

[0057] Feature pattern information of the three-dimensional image model of the alumina ceramic substrate after sintering treatment is extracted.

[0058] At least one reference model adapted to the three-dimensional image model is selected from a preset three-dimensional model library according to the extracted feature pattern information.

[0059] A flatness label of the at least one reference model is assigned to the three-dimensional image model as the set parallelism reference of the three-dimensional image model.

[0060] Optionally, the at least one reference model adapted to the three-dimensional image model is selected from a preset three-dimensional model library according to the extracted feature pattern information, comprising:

[0061] A similarity score of the three-dimensional image model is determined by using a similarity measurement method.

[0062] At least one reference model adapted to the three-dimensional image model is selected from a preset three-dimensional model library according to the similarity score of the three-dimensional image model.

[0063] Optionally, the method for manufacturing the alumina ceramic substrate further comprises:

[0064] When it is determined that the pattern flatness corresponding to the flatness score output by the third neural network model is different from the actual flatness of the image, the actual flatness of the image is marked.

[0065] The actual flatness data of the image is taken as the training image data set, and the third neural network model is iteratively trained.

[0066] Optionally, the method for manufacturing the alumina ceramic substrate further comprises:

[0067] A flatness report of the alumina ceramic substrate after sintering treatment output by the third neural network model is generated.

[0068] The flatness report is stored.

[0069] Optionally, before the pressing plate adapted to the alumina ceramic substrate is selected, the method further comprises:

[0070] The alumina ceramic substrate is subjected to a pre-sintering process.

[0071] Optionally, after the sintering process of the alumina ceramic substrate, the method further comprises:

[0072] The reaction chamber is subjected to a cooling process until the temperature of the reaction chamber reaches a set temperature, and then the alumina ceramic substrate is removed.

[0073] The embodiments of the present specification also provide an alumina ceramic substrate, which is obtained by using the manufacturing method of the alumina ceramic substrate according to any one of the preceding embodiments.

[0074] The manufacturing scheme of the alumina ceramic substrate according to the embodiments of the present specification can improve the flatness of the alumina ceramic substrate during the sintering process, reduce warping, and improve the formation quality by selecting a pressing plate that matches the alumina ceramic substrate and fixedly connecting the alumina ceramic substrate and the pressing plate. On the other hand, the set heating rate in the reaction chamber can be determined according to the parameter information of the alumina ceramic substrate. By precisely controlling the heating rate, the alumina ceramic substrate can be uniformly heated, thereby reducing warping. Therefore, the manufacturing scheme of the alumina ceramic substrate according to the embodiments of the present specification can reduce the warping of the alumina ceramic substrate, improve the yield, and further improve the working performance of the device using the alumina ceramic substrate.

[0075] Further, the set heating rate in the reaction chamber can be determined according to the size information and thickness information of the alumina ceramic substrate. Then, the reaction chamber is subjected to a heating operation according to the set heating rate, which can uniformly heat the alumina ceramic substrate during the sintering process and improve the appearance quality of the alumina ceramic substrate.

[0076] Further, by adjusting the heating rate in the reaction chamber in real time according to the morphological changes of the alumina ceramic substrate during the sintering process, the uniformity of the surface heating of the alumina ceramic substrate during the sintering process can be further improved, and the risk of warping can be reduced.

[0077] Further, since the first neural network model is based on the temperature distribution on the surface of the alumina ceramic substrate, as well as the heating rate and historical morphological changes corresponding to the temperature distribution, the first neural network model has better generalization ability and universality. Therefore, when the first neural network model is used to determine the morphological changes of the alumina ceramic substrate during the sintering process, the accuracy of the morphological change determination can be improved, and thus an appropriate heating rate can be provided for the alumina ceramic substrate.

[0078] Further, by determining the flatness of the alumina ceramic substrate after the sintering process, the alumina ceramic substrate with warping or other defects can be identified, and the consistency of the quality of the alumina ceramic substrate can be improved.

[0079] Further, before selecting a pressing plate suitable for the alumina ceramic substrate, by pre-sintering the alumina ceramic substrate, the stress inside the alumina ceramic substrate can be released, and the probability of warping of the alumina ceramic substrate during sintering can be reduced, and the quality of the alumina ceramic substrate can be improved. BRIEF DESCRIPTION OF DRAWINGS

[0080] In order to more clearly illustrate the technical solutions in the embodiments of the present specification, the following will briefly introduce the drawings needed to be used in the embodiments of the present specification or the prior art description. Obviously, the drawings described below are only some embodiments of the present specification, and for those skilled in the art, other drawings can also be obtained from these drawings without creative labor.

[0081] Figure 1 A flow chart of a method for manufacturing an alumina ceramic substrate in an embodiment of the present specification is shown;

[0082] Figure 2 A flow chart of a method for determining a set heating rate in a reaction chamber in an embodiment of the present specification is shown;

[0083] Figure 3 A flow chart of a training process of a first neural network model in an embodiment of the present specification is shown;

[0084] Figure 4 A flow chart of a method for adjusting a heating rate in a reaction chamber in an embodiment of the present specification is shown;

[0085] Figure 5 A flow chart of a training process of a third neural network model in an embodiment of the present specification is shown;

[0086] Figure 6 A flow chart of a method for determining a set parallelism reference corresponding to each perspective image of a sintered alumina ceramic substrate in an embodiment of the present specification is shown. DETAILED DESCRIPTION

[0087] As described above, when manufacturing an alumina ceramic substrate, warping problems are prone to occur, especially during high-temperature sintering. The warped substrate not only affects the appearance and use performance of the product, but also reduces production efficiency and output rate.

[0088] To solve the above technical problems, the embodiment of the present specification provides an alumina ceramic substrate manufacturing scheme. On the one hand, by selecting a pressing plate matched with the alumina ceramic substrate and fixedly connecting the alumina ceramic substrate and the pressing plate, the flatness of the alumina ceramic substrate during sintering can be improved, the warping can be reduced, and the formation quality can be improved. On the other hand, the setting heating rate in the reaction chamber can be determined according to the parameter information of the alumina ceramic substrate. By accurately controlling the heating rate, the alumina ceramic substrate can be uniformly heated, and thus the warping can be reduced. Therefore, by using the alumina ceramic substrate manufacturing scheme in the embodiment of the present specification, the warping of the alumina ceramic substrate can be reduced, the qualified rate can be improved, and thus the working performance of the device using the alumina ceramic substrate can be improved.

[0089] The technical solutions in the embodiments of the present specification will be described clearly and completely in combination with the drawings in the embodiments of the present specification, so that those skilled in the art can better understand and implement the embodiments of the present specification.

[0090] Referring to Figure 1 The flowchart of the alumina ceramic substrate manufacturing method shown in FIG. 1 can specifically include the following steps:

[0091] S11, selecting a pressing plate matched with the alumina ceramic substrate and fixedly connecting the alumina ceramic substrate and the pressing plate.

[0092] Specifically, considering that the alumina ceramic substrate will expand at high temperature, by fixing the alumina ceramic substrate, the deformation of the alumina ceramic substrate during subsequent sintering can be reduced, and the appearance quality and product quality can be improved.

[0093] In some embodiments, a pressing plate matched with the alumina ceramic substrate can be selected, and the two can be fixed.

[0094] In some embodiments of the present specification, the size, shape, temperature resistance and the like of the alumina ceramic substrate and the pressing plate can be matched.

[0095] As an example, since a high temperature will be generated during sintering, the selected pressing plate needs to be formed of a high-temperature-resistant material. For example, the material of the pressing plate can be a high-temperature-resistant material such as ceramic, metal or carbon fiber.

[0096] In specific implementation, to avoid scratches or damage on the surface of the alumina ceramic substrate during sintering, the contact surface between the alumina ceramic substrate and the pressing plate can be smooth.

[0097] S12, placing the alumina ceramic substrate and the pressing plate in a reaction chamber.

[0098] In some embodiments, when the alumina ceramic substrate and the pressing plate are placed in the reaction chamber, the pressing plate can be kept on the alumina ceramic substrate to ensure that the alumina ceramic substrate does not warp.

[0099] In some embodiments of the present specification, the reaction chamber refers to the chamber of a high-temperature furnace with precise temperature control function.

[0100] S13, determining a set heating rate in the reaction chamber according to the parameter information of the alumina ceramic substrate, and performing sintering treatment on the alumina ceramic substrate.

[0101] Specifically, for alumina ceramic substrates with different parameter information, the heating rate determines the morphology of the alumina ceramic substrate formed after sintering treatment. If an inappropriate heating rate (e.g. too fast or too slow) is used for the alumina ceramic substrate, the alumina ceramic substrate will warp. Therefore, the set heating rate in the reaction chamber needs to be determined according to the parameter information of the alumina ceramic substrate.

[0102] Therefore, by using the manufacturing method of the alumina ceramic substrate provided in the embodiments of the present specification, on the one hand, by selecting a pressing plate suitable for the alumina ceramic substrate and fixedly connecting the alumina ceramic substrate and the pressing plate, the flatness of the alumina ceramic substrate during sintering can be improved, the warping can be reduced, and the formation quality can be improved. On the other hand, the set heating rate in the reaction chamber can be determined according to the parameter information of the alumina ceramic substrate. By accurately controlling the heating rate, the alumina ceramic substrate can be uniformly heated, thereby reducing warping. Therefore, by using the manufacturing method of the alumina ceramic substrate in the embodiments of the present specification, the warping of the alumina ceramic substrate can be reduced, the yield can be improved, and the working performance of the device using the alumina ceramic substrate can be improved.

[0103] In order for those skilled in the art to better understand and implement the embodiments of the present specification, the concepts, schemes, principles and advantages of the embodiments of the present specification are described in detail below with reference to the drawings and through specific examples.

[0104] In specific implementations, the inventors have conducted a large number of tests and researches on factors affecting the heating rate, and found that the size information and thickness information of the alumina ceramic substrate are two factors affecting the heating rate.

[0105] Based on this, as shown in the following embodiments of the present specification, the set heating rate in the reaction chamber can be determined in the following manner: Figure 2

[0106] S21, determining a set heating rate in the reaction chamber according to the size information and thickness information of the alumina ceramic substrate.

[0107] ​Specifically, the heating rate has a corresponding relationship with the size information and the thickness information, and the set heating rate in the reaction chamber can be determined according to the size information and the thickness information of the alumina ceramic substrate.

[0108] It can be understood that the set heating rate can be adjusted according to the change of the alumina ceramic substrate in the sintering process.

[0109] In some embodiments of the present specification, the alumina ceramic substrate can be divided according to the size information and the thickness information of the alumina ceramic substrate.

[0110] For example, the alumina ceramic substrate can be divided into three size intervals of large, medium and small according to the size of the alumina ceramic substrate, and the alumina ceramic substrate can be divided into three thickness intervals of thin, medium and thick according to the thickness of the alumina ceramic substrate, and then the corresponding set heating rate of the alumina ceramic substrate can be determined according to the size interval and the thickness interval of the alumina ceramic substrate.

[0111] It should be noted that "large, medium, small, thin, medium and thick" in the embodiments of the present specification are relative concepts, and are not limited to the specific size and thickness of the alumina ceramic substrate, but are used to illustrate that the alumina ceramic substrates of different sizes and thicknesses have corresponding heating rates.

[0112] In some embodiments of the present specification, the size interval of the alumina ceramic substrate can be determined in the following manner:

[0113] If it is determined that the length and the width of the alumina ceramic substrate are both less than a preset first size threshold (for example, 50 mm), it can be considered that the size of the alumina ceramic substrate is small; if it is determined that the length and the width of the alumina ceramic substrate are both greater than the preset first size threshold and less than a preset second size threshold (for example, 150 mm), it can be considered that the size of the alumina ceramic substrate is medium; if it is determined that the length and the width of the alumina ceramic substrate are both greater than the preset second size threshold, it can be considered that the size of the alumina ceramic substrate is large.

[0114] In some embodiments of the present specification, the thickness interval of the alumina ceramic substrate can be determined in the following manner:

[0115] If it is determined that the thickness of the alumina ceramic substrate is less than a preset first thickness threshold (for example, 1 mm), it can be considered that the thickness of the alumina ceramic substrate is thin; if it is determined that the thickness of the alumina ceramic substrate is greater than the preset first thickness threshold and less than a preset second thickness threshold (for example, 5 mm), it can be considered that the thickness of the alumina ceramic substrate is medium; if it is determined that the thickness of the alumina ceramic substrate is greater than the preset second thickness threshold, it can be considered that the size of the alumina ceramic substrate is large.

[0116] Further, when the alumina ceramic substrate is a small size and thin thickness substrate, the heating rate can be set to be faster, such as 5-10℃ / min, due to the small volume and fast heat transfer.

[0117] For a medium size and medium thickness substrate, the heating rate can be set to 4-7℃ / min to ensure uniform heating inside and outside.

[0118] For a large size and thick thickness substrate, due to the large volume and slow heat transfer, in order to ensure uniform heat transfer, the heating rate should be set to 2-4℃ / min.

[0119] It can be understood that the first size threshold, the second size threshold, the first thickness threshold, the second thickness threshold in the above examples, and the heating rate corresponding to the size in different intervals are only for example, and the embodiments of the present application do not make any limitation. For different application scenarios, the above parameter values can be flexibly set.

[0120] It should be noted that in some other examples, the above heating rate can also be fine-tuned according to the specific sintering atmosphere, furnace and other equipment conditions, or fine-tuned according to historical data or previous experimental results. If the previous rate causes the alumina ceramic substrate to warp or break, the heating rate should be slowed down.

[0121] Before actual production, a test can be carried out, and the heating rate can be fine-tuned according to the debugging result to optimize the heating rate, so as to ensure that ideal results can be obtained each time the sintering is carried out.

[0122] S22, according to the set heating rate, the reaction chamber is heated to sinter the alumina ceramic substrate.

[0123] Specifically, when the set heating rate in the reaction chamber is determined by step S21, the temperature in the reaction chamber can be adjusted according to the set heating rate until the temperature in the reaction chamber reaches the set temperature.

[0124] By determining the set heating rate in the reaction chamber according to the size information and thickness information of the alumina ceramic substrate, and then heating the reaction chamber according to the set heating rate, the alumina ceramic substrate can be uniformly heated during sintering, and the appearance quality of the alumina ceramic substrate can be improved.

[0125] In specific implementation, when the reaction chamber is heated according to the heating rate determined by the foregoing examples, the heating rate may be too fast or too slow, therefore, during actual sintering, the morphological change of the alumina ceramic substrate during sintering can be observed at any time to adjust the heating rate.

[0126] As a specific example, the determining the set temperature rising rate in the reaction chamber according to the parameter information of the alumina ceramic substrate can further include:

[0127] The parameter information of the alumina ceramic substrate, the determining the set temperature rising rate in the reaction chamber further includes:

[0128] The determining the morphology change of the alumina ceramic substrate in the sintering process, the morphology change includes the morphology of the alumina ceramic substrate and the morphology change rate; the adjusting the temperature rising rate in the reaction chamber according to the morphology change of the alumina ceramic substrate.

[0129] Specifically, when the temperature in the reaction chamber gradually rises, even if the pressing plate has been fixedly connected with the alumina ceramic substrate, the morphology of the alumina ceramic substrate will also change, and this morphology change can reflect the influence of the current temperature rising rate on the alumina ceramic substrate.

[0130] Specifically, when the morphology change of the alumina ceramic substrate is in a controllable range, it indicates that the current temperature rising rate is suitable for the sintering of the alumina ceramic substrate; when the morphology change of the alumina ceramic substrate is not ideal, it indicates that the current temperature rising rate is not suitable for the sintering of the alumina ceramic substrate.

[0131] Further, the temperature rising rate in the reaction chamber can be adjusted in real time according to the obtained morphology change of the alumina ceramic substrate, so that the uniformity of the surface heating of the alumina ceramic substrate in the sintering process can be further improved, and the risk of warping can be reduced.

[0132] In some embodiments of the present specification, various methods can be used to determine the morphology change of the alumina ceramic substrate in the sintering process.

[0133] As an optional example, a preset first neural network model can be used to determine the morphology change of the alumina ceramic substrate in the sintering process. Since the first neural network model is based on the temperature distribution on the surface of the alumina ceramic substrate, and the temperature rising rate and historical morphology change corresponding to the temperature distribution, the first neural network model has better generalization ability and universality, so that the accuracy of the morphology change determination can be improved when the first neural network model is used to determine the morphology change of the alumina ceramic substrate in the sintering process, and thus a suitable temperature rising rate can be provided for the alumina ceramic substrate.

[0134] For better implementation and use of the solutions in the embodiments of the present specification by those skilled in the art, refer to Figure 3 for a detailed description of the obtaining process of the first neural network model in the embodiments of the present specification:

[0135] S31, pre-process the temperature distribution of the surface of the alumina ceramic substrate obtained, and the heating rate and historical morphology corresponding to the temperature distribution to obtain a data set.

[0136] Specifically, the temperature, heating rate and morphology change of the alumina ceramic substrate during each sintering process can be recorded regularly to train the model and obtain the first neural network model.

[0137] In some embodiments, a thermal imaging camera can be used to capture the temperature distribution of the surface of the alumina ceramic substrate, and then the heating condition of the surface can be determined, such as whether it is uniformly heated.

[0138] In some embodiments, a high-temperature resistant camera or sensor can be installed in the reaction chamber to capture the morphology change of the substrate in real time, or a normal camera can be set outside the furnace, and the furnace part can be made of transparent glass, so that the camera outside the furnace can observe the morphology of the alumina ceramic substrate in real time.

[0139] In some embodiments, data preprocessing is a key step in machine learning and data analysis, which can make the format of the data meet the requirements of the model, converge more quickly and effectively, and improve the performance of the model.

[0140] In specific implementations, normalization or standardization can be used to pre-process the temperature distribution of the surface of the alumina ceramic substrate, and the heating rate and historical morphology corresponding to the temperature distribution.

[0141] Normalization usually scales the feature values to between 0 and 1 by subtracting the minimum value in the feature values and dividing by the difference between the maximum and minimum values in the feature values.

[0142] Standardization involves rescaling features to have a mean of 0 and a standard deviation of 1. This is often used for algorithms that assume Gaussian distribution of data, such as linear regression, logistic regression, and linear discriminant analysis.

[0143] In specific implementations, the choice of normalization and standardization depends on the specific application and the model used. For some algorithms, such as neural networks, normalization may be a better choice, as it can ensure that all input features are on the same scale. For other algorithms that require Gaussian distribution of data, standardization may be more appropriate, and the embodiments of the present disclosure do not limit the specific preprocessing method.

[0144] S32, divide the data set into training data set, validation data set and test data set.

[0145] In some embodiments, for image data involved in the data set, the image should be normalized to make the pixel value of the image between 0 and 1.

[0146] In some embodiments, if the state label corresponding to the shape change is not in digital form, one-hot encoding can be performed.

[0147] S33, input the training data set to a preset first pre-training model, train the first pre-training model to obtain the first neural network model.

[0148] In some embodiments, the preset first pre-training model can be obtained according to the following steps:

[0149] A1) Select an initialized pre-training model.

[0150] A2) Define the input layer of the pre-training model. In some embodiments, considering that the image can be in RGB form, the temperature of the input layer of the pre-training model can be (height, width, 3).

[0151] A3) Add a convolutional layer. For example, a 3x3 filter can be used, and the number of filters can be set.

[0152] A4) Add an activation function. For example, the activation function can be ReLU.

[0153] A5) Add a pooling layer. For example, the pooling layer can be a max pooling layer to reduce the dimension of the features.

[0154] A6) Repeat steps A1) to A5) to add more convolutional layers and pooling layers.

[0155] A7) Add a fully connected layer and use an activation function again.

[0156] A8) Design the output layer. If you have four shape change states, you need to set four output units corresponding to them. And use the softmax activation function to output the probability of each class.

[0157] It should be noted that the above four shape changes are only examples for illustrating that there are several shape changes, and several output units are set accordingly, and the embodiments of the present application do not limit this.

[0158] A9) Compile the model. Specifically, it includes selecting a loss function and an optimizer, and adding an evaluation index. For example, the loss function can be categorical cross-entropy, the optimizer can be Adam, SGD, etc., and the evaluation index can be `accuracy`.

[0159] It can be understood that the embodiments of the present specification do not make any limitation on the obtaining process of the first pre-training model, as long as the first neural network model is obtained according to the temperature distribution on the surface of the aluminum oxide ceramic substrate and the heating rate and the historical morphology corresponding to the temperature distribution.

[0160] In specific implementations, the first pre-training model can be trained in the same way for the first neural network model formed in different ways.

[0161] Specifically, the amount of data for each training and the number of training epochs can be set, and the training data and the labels corresponding to the training data are input into the first pre-training model to obtain the first neural network model.

[0162] It should be noted that, in the training process, to avoid overfitting, a callback function such as ModelCheckpoint and EarlyStopping can be used to automatically save the best model weight and prevent overfitting.

[0163] S34, inputting the verification data set into the first neural network model to obtain the actual morphology of the aluminum oxide ceramic substrate corresponding to each verification data set.

[0164] Specifically, the verification data includes the temperature distribution on the surface of the aluminum oxide ceramic substrate, and the heating rate and the historical morphology corresponding to the temperature distribution, so that based on the input verification data set, the actual morphology of the aluminum oxide ceramic substrate under the current temperature condition can be determined.

[0165] S35, adjusting the first neural network model according to the actual morphology of the aluminum oxide ceramic substrate corresponding to each verification data set and the historical morphology of the aluminum oxide ceramic substrate contained in each verification data set.

[0166] Specifically, when it is determined that the historical morphology of the aluminum oxide ceramic substrate contained in each verification data set is significantly different from the actual morphology of the aluminum oxide ceramic substrate, it indicates that the first neural network model cannot correctly predict the actual morphology of the aluminum oxide ceramic substrate, and then the related parameters in the first neural network model can be adjusted to improve the accuracy of the prediction result of the first neural network model.

[0167] In some embodiments, other performance indicators of the first neural network model can also be monitored, such as confusion matrix, recall rate, precision rate, etc.

[0168] S36, inputting the test data set into the adjusted first neural network model, and stopping training the first neural network model when it is determined that the performance of the first neural network model meets the set performance.

[0169] Specifically, by using the steps S31 to S35, the first neural network model with excellent prediction performance can be obtained. By inputting the test data into the adjusted first neural network model, the predicted shape of the alumina ceramic substrate output by the first neural network model can be compared with the actual shape of the alumina ceramic substrate. When it is determined that the performance of the first neural network model meets the set performance, the training of the first neural network model can be stopped, and the first neural network model can be used in the subsequent determination process.

[0170] In some embodiments of the present specification, to improve the prediction accuracy of the first neural network model, some new data can also be used to test the first neural network model. For example, the new data can be data that is not used in training the first pre-trained model or newly obtained data, and the present specification does not make any limitation on this.

[0171] In specific implementation, if the performance of the first neural network model trained by using the steps S31 to S36 still cannot meet the test requirements, data enhancement, network structure adjustment, or other regularization techniques can be used to improve the performance of the first neural network model.

[0172] By training, verifying, and adjusting the first pre-trained model in the above manner, the first neural network model can have better generalization ability and universality, so that when the first neural network model is used to determine the shape change of the alumina ceramic substrate in the sintering process, the accuracy of the shape change determination can be improved, and thus the alumina ceramic substrate can be provided with a suitable heating rate.

[0173] In some embodiments of the present specification, the shape change of the alumina ceramic substrate in the sintering process can be determined in the following manner:

[0174] The current temperature and the current heating rate in the reaction chamber are obtained, and the current temperature and the current heating rate in the reaction chamber are input into the first neural network model to determine the predicted shape and the predicted shape change rate of the alumina ceramic substrate under the current temperature and the current heating rate.

[0175] Specifically, under the joint action of the current temperature and the current heating rate, the shape of the alumina ceramic substrate will change. Since the first neural network model has been trained and verified, the predicted shape and the predicted shape change rate of the alumina ceramic substrate can be determined according to the current temperature and the current heating rate.

[0176] For example, if it is determined according to the current temperature and the current heating rate that the predicted shape of the alumina ceramic substrate is warped upward relative to the surface of the alumina ceramic substrate, and the predicted shape change rate is warped downward relative to the surface of the alumina ceramic substrate, the current heating rate can be maintained; if it is determined according to the current temperature and the current heating rate that the predicted shape of the alumina ceramic substrate is warped upward relative to the surface of the alumina ceramic substrate, and the predicted shape change rate is warped upward relative to the surface of the alumina ceramic substrate, the current heating rate needs to be adjusted.

[0177] As an optional implementation example, as shown in Figure 4 The heating rate in the reaction chamber can be adjusted in the following manner:

[0178] S41, based on the alumina ceramic substrate being at the current temperature and the current heating rate, determining, by the first neural network model, a set shape of the alumina ceramic substrate at the current temperature and the current heating rate.

[0179] Specifically, by using the above process, the first neural network model for predicting the shape change of the alumina ceramic substrate can be obtained, and then based on the alumina ceramic substrate being at the current temperature and the current heating rate, the set shape of the alumina ceramic substrate corresponding thereto can be determined.

[0180] S42, according to the set shape of the alumina ceramic substrate at the current temperature and the current heating rate, using a preset second neural network model to determine a set heating rate corresponding to the alumina ceramic substrate.

[0181] The second neural network model is obtained based on the shape change of the alumina ceramic substrate and the heating rate and temperature in the reaction chamber corresponding to the shape change.

[0182] In a specific implementation, corresponding to different shape changes of the alumina ceramic substrate, a corresponding heating rate needs to be set, so as to control the temperature in the reaction chamber in the same time. Since the second neural network model is obtained based on the shape change of the alumina ceramic substrate and the heating rate and temperature in the reaction chamber corresponding to the shape change, when the set shape of the alumina ceramic substrate is determined, the set heating rate corresponding to the alumina ceramic substrate can be obtained by using the second neural network model.

[0183] And since the second neural network model is trained by a large amount of data, it has high accuracy and strong adaptability, and does not need human intervention, so the set heating rate with higher precision and more stability can be obtained.

[0184] S43, adjusting the temperature ramping rate in the reaction chamber according to the current temperature ramping rate and the set temperature ramping rate of the alumina ceramic substrate.

[0185] As an example, the current temperature ramping rate can be adjusted until the temperature ramping rate is the set temperature ramping rate.

[0186] In some embodiments of the present specification, the preset second pre-trained model can also be obtained according to the following steps:

[0187] B1) Select an initialized pre-trained model (which is different from the pre-trained model in step A1).

[0188] B2) Define the input layer of the pre-trained model. In some embodiments, the dimension of the input layer can be the time step and the number of features at each step.

[0189] B3) Add an RNN (Recurrent Neural Network) layer, for example, according to the complexity of different problems, a conventional RNN, LSTM or GRU, etc. can be selected.

[0190] B4) Stack multiple RNN layers to increase the depth and complexity of the pre-trained model.

[0191] It can be understood that in some embodiments, step B4) can be omitted.

[0192] B5) Add a fully connected layer and use an appropriate activation function.

[0193] B6) Design the output layer, the dimension of which should match the expected temperature adjustment strategy output.

[0194] B7) Compile the model. Specifically, select a loss function and an optimizer, and add an evaluation index. For example, the loss function can be mean square error, the optimizer can be Adam, SGD, etc., and the evaluation index can be mean absolute error.

[0195] It can be understood that the present embodiments do not make any limitation on the obtaining process of the second pre-trained model, as long as the second neural network model is obtained according to the morphological changes of the alumina ceramic substrate and the temperature ramping rate and temperature in the reaction chamber corresponding to the morphological changes.

[0196] In specific implementations, the second pre-trained model can be trained in the same way for second neural network models formed in different ways, specifically including:

[0197] C1) Collect data and preprocess the data:

[0198] C11) Real-time monitoring of the morphology of the alumina ceramic substrate and the temperature parameters (including the current temperature and the current heating rate) in the current morphology using sensors.

[0199] C12) Constructing time series data corresponding to the morphology of the alumina ceramic substrate and the temperature parameters in the current morphology. For example, data processing techniques such as sliding windows can be used to construct the time series data.

[0200] C2) Selecting a second pre-trained model.

[0201] Specifically, a second neural network model (which can be a CNN, RNN or other model, the specific type depending on actual requirements) that has been trained can be used.

[0202] In some other embodiments, the pre-trained model in the initialization state can also be retrained, and the selection process of the second pre-trained model is not limited by the embodiments of the present specification.

[0203] It can be understood that regardless of which model is selected, it is necessary to ensure that the second pre-trained model can normally run on the device, for example, the loading speed is acceptable to meet real-time requirements.

[0204] C3) Training the second pre-trained model

[0205] C31) Inputting the preprocessed data (including the morphology of the alumina ceramic substrate and the temperature parameters in the current morphology) into the second pre-trained model and outputting the shape change of the alumina ceramic substrate.

[0206] C32) According to the output of the second pre-trained model, determining whether the temperature needs to be adjusted.

[0207] If the temperature is adjusted, the adjustment strategy of the protective body and the temperature adjustment value are determined.

[0208] In some embodiments, an automatic system can be used to adjust the temperature in the reaction chamber in real time.

[0209] C4) Real-time feedback, iterative training of the second pre-trained model:

[0210] Specifically, the actual shape change state of the alumina ceramic substrate and the prediction result of the second pre-trained model are compared, and according to the comparison result, the parameters of the second pre-trained model are corrected or fine-tuned to obtain a second neural network model that can accurately predict the morphology of the alumina ceramic substrate.

[0211] In some embodiments of the present specification, when steps C1) to C4) are adopted, a trained second neural network model is obtained, and then the second neural network model can be used to adjust the temperature in the reaction chamber.

[0212] Specifically, the second neural network model can automatically adjust the temperature or the heating rate in the reaction chamber by collecting real-time data, analyzing and inferring, and making decisions.

[0213] The following is a simplified example to describe how to achieve this process through the second neural network model:

[0214] D1) Data collection.

[0215] Specifically, the temperature parameters in the reaction chamber and the morphology of the alumina ceramic substrate are obtained, and the collected data are transmitted to the second neural network model.

[0216] As an optional example, the temperature parameters in the reaction chamber can be obtained by a temperature sensor, and the morphology of the alumina ceramic substrate can be obtained by a shape detection device (e.g., a device with camera function).

[0217] D2) Data preprocessing.

[0218] For example, the temperature parameters in the reaction chamber and the morphology of the alumina ceramic substrate can be normalized, denoised, feature extracted, and the like.

[0219] D3) Output prediction results.

[0220] Using the trained second neural network model, the analysis system analyzes the temperature parameters in the reaction chamber and the morphology of the alumina ceramic substrate, and determines whether the alumina ceramic substrate is likely to warp under the current conditions.

[0221] D4) Decision making.

[0222] Specifically, if the second neural network model predicts that the alumina ceramic substrate is likely to warp, a new heating rate will be recalculated or the temperature in the reaction chamber will be changed.

[0223] For example, if the second neural network model monitors that the temperature of the first side of the alumina ceramic substrate is higher than that of the second side, it can reduce the temperature of the first side or increase the temperature of the second side to maintain uniform heating.

[0224] D5) Real-time adjustment.

[0225] Specifically, the second neural network model sends instructions to the control system of the reaction chamber to adjust the temperature or the heating rate in the reaction chamber, and then the control system in the reaction chamber can adjust the power of the heating element or change the speed of the fan to change the temperature according to the instructions.

[0226] D6) Feedback and optimization:

[0227] Specifically, the second neural network model continuously monitors the temperature parameters in the reaction chamber and the morphology of the alumina ceramic substrate. If the morphology of the alumina ceramic substrate still does not achieve the expected effect after adjusting the heating rate in the reaction chamber, the second neural network model will continue to analyze and adjust.

[0228] Over time, the second neural network model can continuously learn from the feedback of the above information and optimize its decision algorithm, so as to more accurately adjust the temperature.

[0229] Through the above steps, the second neural network model can detect the temperature parameters in the reaction chamber and the morphology of the alumina ceramic substrate in real time, and make corresponding adjustments to ensure that the alumina ceramic substrate is uniformly heated to avoid warping. This intelligent method not only can improve production efficiency, but also can reduce the scrap rate of alumina ceramic substrates.

[0230] D7) Monitoring and alarm:

[0231] Specifically, the prediction performance of the second neural network model is continuously monitored. If it is determined that the prediction of the second neural network model deviates greatly from the actual situation, an alarm is sent to the operator or technical team for subsequent analysis and improvement.

[0232] In specific implementation, there can be some cases that the predicted morphology change of the alumina ceramic substrate under the current temperature and the current heating rate is greater than the set morphology change, at this time, the temperature and / or the heating rate in the reaction chamber may not be adjusted to change the morphology of the alumina ceramic substrate.

[0233] Based on this, the manufacturing method in the embodiments of the present specification can further include: when it is determined that the predicted morphology change of the alumina ceramic substrate under the current temperature and the current heating rate is greater than the set morphology change, resetting the temperature and / or the heating rate in the reaction chamber.

[0234] The manufacturing method of the alumina ceramic substrate in the above example can determine the set heating rate in the reaction chamber according to the parameter information of the alumina ceramic substrate, and can adjust the heating rate in the reaction chamber in real time according to the morphology change of the alumina ceramic substrate during the sintering process, so as to improve the flatness of the alumina ceramic substrate.

[0235] Based on this, in some embodiments of the present specification, continuing to refer to Figure 1 The manufacturing method of the alumina ceramic substrate in the embodiments of the present specification further comprises:

[0236] S14, determining the flatness of the alumina ceramic substrate after sintering treatment.

[0237] Specifically, the above examples describe how to adjust the heating rate in the reaction chamber to avoid warping, and for the sintering process of a large batch of alumina ceramic substrates, it is necessary to determine whether the above method can really avoid warping, and then the flatness of the alumina ceramic substrate after sintering treatment can be obtained and compared with the set flatness, thereby identifying the alumina ceramic substrate with warping or other defects and improving the consistency of the quality of the alumina ceramic substrate.

[0238] In some embodiments of the present specification, the flatness of the alumina ceramic substrate after sintering treatment can be determined in various ways.

[0239] As a specific example, the heights of different surfaces of the alumina ceramic substrate relative to the reference plane can be measured to determine the flatness of the alumina ceramic substrate after sintering treatment.

[0240] For example, when the heights of different surfaces of the alumina ceramic substrate relative to the reference plane are determined to be the same or the height difference is within a set range, it can be determined that the alumina ceramic substrate has good flatness after sintering.

[0241] In specific implementation, to realize a large batch and high-precision identification process, a preset third neural network model can be used to determine the flatness of the alumina ceramic substrate after sintering treatment, wherein the third neural network model is obtained based on the image information of the alumina ceramic substrate collected.

[0242] Specifically, since the third neural network model is obtained based on the image information of the alumina ceramic substrate collected, it has better generalization ability and universality, and can automatically determine the flatness of the alumina ceramic substrate after sintering treatment without human intervention, and the flatness of the alumina ceramic substrate after sintering treatment obtained has higher precision, and determination of the flatness of a large batch of alumina ceramic substrates can be realized.

[0243] In some embodiments of the present specification, specifically, the third neural network model can be obtained according to Figure 5

[0244] S51, obtaining images of the alumina ceramic substrate after sintering treatment from multiple perspectives, and determining the label information corresponding to each perspective image.

[0245] ​Specifically, by acquiring images of the sintered alumina ceramic substrate from multiple perspectives, the warping degree of different regions of the alumina ceramic substrate can be determined, and then the corresponding label information can be set for the images from each perspective according to the warping degree of each region.

[0246] In some embodiments of the present specification, the following method can be used to determine each perspective image and its corresponding label information:

[0247] S511, generating a three-dimensional image model of the sintered alumina ceramic substrate.

[0248] S512, rendering a two-dimensional image corresponding to the three-dimensional image model from multiple perspectives to obtain multiple perspective images of the sintered alumina ceramic substrate.

[0249] Specifically, by converting the three-dimensional image into a two-dimensional image, the processing difficulty can be reduced and the processing efficiency can be improved.

[0250] S513, determining the label information corresponding to each perspective image according to the relationship between the parallelism of each perspective image and the set parallelism, wherein the label information includes normal substrate and defective substrate.

[0251] The defective substrate can include a slightly warped substrate and a severely warped substrate.

[0252] Specifically, by using steps S511 to S512, the parallelism of each perspective image can be determined. Since the set parallelism can represent that the flatness of the sintered alumina ceramic substrate meets the requirements, the label information corresponding to each perspective image can be determined according to the relationship between the parallelism of each perspective image and the set parallelism.

[0253] For example, if the parallelism of the image is the same as the set parallelism or the difference is within an acceptable range, the label information corresponding to the image is a normal substrate; if the parallelism of the image is significantly different from the set parallelism, the label information corresponding to the image is an abnormal substrate.

[0254] S52, pre-processing the images from each perspective to obtain multiple image data.

[0255] Specifically, by pre-processing the images from each perspective, the format of each perspective image can meet the input requirements of the third neural network model.

[0256] In some embodiments of the present specification, each perspective image can be normalized, for example, the pixel value of each perspective image is mapped from [0, 255] to [0, 1] or [-1, 1].

[0257] It can be understood that the plurality of image data can also be data augmented, for example, random rotation, cropping and horizontal flip, to enhance the generalization ability of the model.

[0258] S53, divide the plurality of image data into a training image data set and a validation image data set.

[0259] In some implementations of the present specification, the training image data set and the validation image data set can be divided in a certain proportion.

[0260] For example, 80% of the image data can be used as the training image data set, and 20% of the image data can be used as the validation image data set. It can be understood that the above proportion is only an example and is used to illustrate that the amount of training image data set is greater than the amount of validation image data set, and the embodiments of the present specification do not require this.

[0261] S54, in any round, output the training image data set to the third pre-training model according to a set batch, train the third pre-training model, and obtain the third neural network model.

[0262] In a specific implementation, in order to avoid overfitting, the training image data set can be regularized during training.

[0263] In some embodiments of the present specification, the structure of the third pre-training model can be obtained in the following manner, specifically including:

[0264] 1. Define the input layer

[0265] The input layer of the third pre-training model should be able to accept a tensor with a size of Height x Width x Channels x Views, where Height and Width are the dimensions of the image, Channels is usually an RGB image or a 1 grayscale image, and Views is the number of images of different views.

[0266] 2. Determine the view feature extraction

[0267] For each view of the image, a series of convolutional layers are used to extract features. For example, a common combination of convolutional layers, pooling layers and normalization layers can be used to complete this step.

[0268] 3. Determine the feature integration

[0269] Different fusion methods can be used to determine the feature integration process according to different selections.

[0270] For example, an average fusion method can be used, that is, simply averaging all view feature maps. This method is suitable for cases where the difference between views is not large and each view is equally important.

[0271] For example, the maximum fusion method is used, which maximizes the feature maps of all views and only retains the most significant features. This method works well in cases where some views may contain noise.

[0272] For another example, the full connection fusion method is used, which integrates the features of all views through a fully connected layer, allowing the model to learn how to best combine information from different views.

[0273] 4. Determine global feature extraction

[0274] Specifically, more convolutional layers, pooling layers and normalization layers are used to further extract global information from the fused features.

[0275] 5. Set the fully connected layer

[0276] One or more fully connected layers are used to obtain a compressed feature representation.

[0277] 6. Define the output layer:

[0278] The output layer can be designed flexibly according to the needs of the task. For example, if it is a classification task, an output node with the required number of classes can be used, and a softmax activation function can be applied.

[0279] 7. Regularization and optimization:

[0280] Specifically, Dropout layers or Batch Normalization layers can be added to the network to improve performance and stability, and appropriate loss functions and optimizers can be used to train the network.

[0281] When designing the third pre-trained model, the following factors should also be considered:

[0282] Data augmentation: Since the images obtained from different views may have slight rotation, scaling or other deformation, using data augmentation methods can improve the generalization ability of the network.

[0283] Weight sharing: In the view feature extraction stage, all views can share the same weights, which can reduce the number of model parameters and increase the stability of training.

[0284] Thus, through the above process, the structure trained can effectively process and integrate 2D image information from multiple views, providing rich feature representation for subsequent tasks such as classification, regression or object detection.

[0285] S55, input the verification image data set into the third pre-trained model to obtain the actual label information corresponding to each training image data.

[0286] Specifically, by step S54, a third neural network model with flatness prediction capability can be obtained, and when the verification image dataset is input into the third neural network model, a unique actual label information corresponding thereto can be determined, which can be one of a normal substrate or an abnormal substrate.

[0287] S56, according to the actual label information corresponding to the verification image dataset and the label information contained in the verification image dataset, adjusting the third neural network model.

[0288] Specifically, each verification image dataset itself has a set label information, by comparing the actual label information of the verification image dataset with the set label information, it can be determined that the third neural network model can correctly predict the flatness of the aluminum oxide ceramic substrate, and then according to the comparison result, part of the parameters or structure in the third neural network model can be adjusted.

[0289] It should be noted that if the performance of the third neural network model on the verification image dataset no longer improves, consider terminating the training or adjusting the learning rate in advance.

[0290] Therefore, by using the above steps S51 to S56, by training the third pre-trained model, a third neural network model with the function of predicting the flatness of the aluminum oxide ceramic substrate can be obtained, and then the third neural network model can be used to accurately identify the slight warping or shape deviation of the aluminum oxide ceramic substrate after sintering treatment.

[0291] As an optional example, the third neural network model is used to determine the flatness of the aluminum oxide ceramic substrate after sintering treatment, which can include:

[0292] Generating images of multiple viewing angles of the aluminum oxide ceramic substrate after sintering treatment; inputting each viewing angle image into the third neural network model to obtain a flatness score corresponding to each viewing angle image, the flatness score being used to represent the flatness of each viewing angle image.

[0293] In specific implementation, after using the third neural network model to obtain the flatness of each viewing angle image, the qualification of the aluminum oxide ceramic substrate can also be evaluated.

[0294] As an implementation example, the qualification of the aluminum oxide ceramic substrate can be determined according to the flatness of each viewing angle image and a set parallelism reference.

[0295] Specifically, images from different perspectives can characterize the flatness of different areas of the alumina ceramic substrate. Furthermore, by integrating images from various perspectives, the flatness of the images from each perspective can be used to assess whether the alumina ceramic substrate can meet production requirements.

[0296] The foregoing embodiments describe how to determine the flatness of images from various viewpoints. The following section will focus on how to obtain the set parallelism benchmark for images from various viewpoints.

[0297] As an optional example, see [reference] Figure 6 The embodiment shown in this specification illustrates a method for determining a parallelism benchmark, such as... Figure 6 As shown, the parallelism benchmark can be obtained in the following way:

[0298] S61, Obtain a three-dimensional image model of the alumina ceramic substrate after sintering.

[0299] The three-dimensional image model can characterize the actual morphology of the alumina ceramic substrate after sintering.

[0300] S62, extract the feature graphic information of the three-dimensional image model of the alumina ceramic substrate after sintering.

[0301] Specifically, since feature extraction algorithms can analyze 3D models based on local or global feature descriptions, they can characterize the 3D morphology of alumina ceramic substrates after sintering by extracting feature graphic information.

[0302] In some embodiments of this specification, multiple feature extraction methods may be used, and the results of these methods may be fused to improve the accuracy and robustness of retrieval.

[0303] S63, based on the extracted feature graphic information, select at least one reference model from the preset three-dimensional model library that is compatible with the three-dimensional image model.

[0304] Specifically, since the feature graphic information can characterize the three-dimensional morphology of the alumina ceramic substrate after sintering, at least one reference model that matches the three-dimensional image model can be selected from a preset three-dimensional model library based on the feature graphic information, so as to use at least one reference model as a model to characterize the morphology of the alumina ceramic substrate.

[0305] In some embodiments of this specification, at least one reference model adapted to the three-dimensional image model can be selected from a preset three-dimensional model library using the following method:

[0306] S631, A similarity measurement method is used to determine the similarity score of the three-dimensional image model.

[0307] In some embodiments, the similarity measurement method can include Euclidean distance or cosine similarity.

[0308] S632, selecting at least one reference model that is adapted to the three-dimensional image model from the preset three-dimensional model library according to the similarity score of the three-dimensional image model.

[0309] Specifically, using the similarity measurement method can compare the features between the three-dimensional image model and the models in the preset three-dimensional model library, and then according to the similarity score, at least one reference model that is most similar to the three-dimensional image model can be selected.

[0310] S64, assigning the flatness label of the at least one reference model to the three-dimensional image model as the set flatness reference of the three-dimensional image model.

[0311] Specifically, since the at least one reference model can represent the three-dimensional morphology of the sintered aluminum oxide ceramic substrate, it can be considered that the flatness label of the at least one reference model is the set flatness reference of the three-dimensional image model.

[0312] In some embodiments of the present specification, if one reference model that is adapted to the three-dimensional image model is determined, the flatness label of the reference model can be directly used as the set flatness reference of the three-dimensional image model.

[0313] In some other embodiments, if multiple reference models that are adapted to the three-dimensional image model are determined, a weighted average score can be calculated based on the flatness labels of the multiple reference models as the set flatness reference of the three-dimensional image model.

[0314] Therefore, by using the feature extraction method, the similarity measurement method, etc. in the above examples, the set flatness reference of the three-dimensional image model of the sintered aluminum oxide ceramic substrate can be determined, and then whether the sintered aluminum oxide ceramic substrate is qualified can be determined based on the set flatness reference.

[0315] In some embodiments, there can be some misjudgments of the flatness of the aluminum oxide ceramic substrate, and to avoid such situations, the manufacturing method of the aluminum oxide ceramic substrate in the embodiments of the present specification can further include: when it is determined that the flatness of the image corresponding to the flatness score output by the third neural network model is different from the actual flatness of the image, marking the actual flatness of the image, and using the actual flatness data of the image as the training image data set to iteratively train the third neural network model.

[0316] Specifically, by using the trained third neural network model, the flatness of the sintered alumina ceramic substrate can be output according to the input data. The output result of the third neural network model can be compared with the actual flatness of the sintered alumina ceramic substrate through artificial or machine recognition. Then, the accuracy of the output result of the third neural network model can be determined, and when the flatness score corresponding to the graphical flatness is different from the actual flatness, the actual flatness of the image can be marked. Thus, the third neural network model can be iteratively trained to further improve the accuracy of the output result of the third neural network model.

[0317] In some embodiments of the present specification, the third neural network model can be iteratively trained using new data periodically.

[0318] In specific implementation, the flatness of each sintered alumina ceramic substrate can be predicted by the third neural network model. To facilitate traceability and improve management capability, the manufacturing method of the alumina ceramic substrate in the present specification can further include generating a flatness report of the sintered alumina ceramic substrate output by the third neural network model, and storing the flatness report.

[0319] Specifically, based on the output result of the third neural network model, the alumina ceramic substrate can be automatically classified as qualified or unqualified. Then, a corresponding flatness report can be generated, and the detection result can be stored in a database for production and quality management team to query and analyze.

[0320] For example, a detailed report can be generated for the unqualified alumina ceramic substrate to show the area and degree of deviation from the flatness.

[0321] Therefore, through the above examples, the manufacturing method of the alumina ceramic substrate in the present specification can realize automatic detection of the flatness of the alumina ceramic substrate, and quickly and accurately identify the substrate with warping or other defects, thereby improving production efficiency and ensuring the consistency of the quality of the alumina ceramic substrate.

[0322] In some embodiments of the present specification, the inventors found that due to the existence of internal stress of the alumina ceramic substrate, the alumina ceramic substrate may be broken due to high temperature during sintering of the alumina ceramic substrate.

[0323] In specific implementation, before selecting the pressing plate suitable for the alumina ceramic substrate, the following steps can be further included:

[0324] Pre-sintering treatment is performed on the alumina ceramic substrate.

[0325] As a specific example, the pre-sintering process of the alumina ceramic substrate can be performed as follows:

[0326] 1) Prepare the untreated alumina ceramic substrate.

[0327] 2) Place the alumina ceramic substrate on a uniformly distributed support in a pre-sintering device (e.g. an atmosphere furnace) so that there is enough space between each alumina ceramic substrate to be heated evenly.

[0328] 3) Set the temperature of the pre-sintering device to the pre-sintering temperature. For example, the pre-sintering temperature can be between 500°C and 650°C.

[0329] 4) Maintain the pre-sintering temperature for about 1-3 hours, at which time the alumina ceramic substrate will release some internal stress and impurities.

[0330] 5) Reduce the temperature in the pre-sintering device to room temperature, and then remove the pre-sintered alumina ceramic substrate.

[0331] It should be understood that, first, the above example is only used to illustrate the pre-sintering process of the alumina ceramic substrate, and cannot be construed as a limitation on the present application; second, the parameters listed in the above example are only illustrative, and different parameters can be set for different application scenarios to perform the pre-sintering process of the alumina ceramic substrate.

[0332] By pre-sintering the alumina ceramic substrate, the internal stress of the alumina ceramic substrate can be released, thereby reducing the probability of warping of the alumina ceramic substrate during the sintering process and improving the quality of the alumina ceramic substrate.

[0333] In a specific implementation, after the alumina ceramic substrate is pre-sintered and sintered in the above manner, the reaction chamber can also be subjected to a cooling process until the temperature of the reaction chamber reaches a set temperature, and then the alumina ceramic substrate is removed.

[0334] As a specific implementation example, after sintering is complete, the cooling process begins. First, the temperature in the reaction chamber is reduced to a first temperature (e.g. 1300°C), and the corresponding cooling rate can be 5-7°C / min; second, when the temperature in the reaction chamber is reduced to a second temperature (e.g. 1000°C) or below, the cooling rate is adjusted, for example, the cooling rate can be 10-12°C / min, until the temperature in the reaction chamber is reduced to room temperature; finally, the reaction chamber is opened, and the sintered alumina ceramic substrate is removed, so that the flatness of the alumina ceramic substrate is checked to ensure that there is no warping or other defects.

[0335] It can be understood that the above describes the embodiments of the manufacturing method of the plurality of alumina ceramic substrates provided by the embodiments of the present specification, and each optional mode introduced by each embodiment can be combined with each other without conflict, cross-referenced, thereby extending a plurality of possible embodiment schemes, which can be considered as the embodiments disclosed and disclosed by the present specification.

[0336] Correspondingly, the embodiments of the present specification also provide an alumina ceramic substrate, which can be obtained by the manufacturing method of the alumina ceramic substrate described in any of the preceding embodiments.

[0337] Specifically, different sintering methods can be used to manufacture the alumina ceramic substrate, and in this process, the manufacturing method in the above embodiments can be used to reduce the probability of warping of the alumina ceramic substrate.

[0338] For example, the alumina ceramic substrate is prepared by using a liquid phase sintering method. Specifically, it includes: pre-sintering at a temperature of 600℃ for 2 hours; heating to 1500℃ at a rate of 5℃ / min, and when the temperature decreases to room temperature, the alumina ceramic substrate is taken out; the alumina ceramic substrate is pressed by using a ceramic pressing plate, and then placed into a reaction chamber, and after sintering at 1500℃ for 4 hours, cooling can be performed at a rate of 10℃ / min.

[0339] For another example, the alumina ceramic substrate is prepared by using a traditional sintering method. Specifically, it includes: pre-sintering at a temperature of 500℃ for 3 hours; heating to 1400℃ at a rate of 3℃ / min, and when the temperature decreases to room temperature, the alumina ceramic substrate is taken out; the alumina ceramic substrate is pressed by using a metal pressing plate, and then placed into a reaction chamber, and after sintering at 1400℃ for 5 hours, cooling can be performed at a rate of 8℃ / min.

[0340] For another example, the alumina ceramic substrate is prepared by using a hot pressing method. Specifically, it includes: pre-sintering at a temperature of 650℃ for 1.5 hours; heating to 1550℃ at a rate of 4℃ / min, and when the temperature decreases to room temperature, the alumina ceramic substrate is taken out; the substrate is pressed by using a carbon fiber pressing plate, and then placed into a reaction chamber, and after sintering at 1550℃ for 3.5 hours, cooling can be performed at a rate of 7℃ / min.

[0341] It should be noted that the three preparation methods listed above are only illustrative examples, and the parameters listed in various preparation methods are also illustrative examples, which are only used for understanding by those skilled in the art, and in the preparation method, the heating rate and / or cooling rate can be adjusted in real time according to the manufacturing method described in the above embodiments, so as to avoid warping of the alumina ceramic substrate and improve the quality of the alumina ceramic substrate.

[0342] It should be noted that the terms "first", "second", and so on, are used herein only for descriptive purposes, and are not to be construed as indicating or implying relative importance or an indicated number of technical features. Thus, features defined with "first", "second" and the like, can be explicitly or implicitly included one or more of such features. Furthermore, the terms "first", "second" and the like are used to distinguish similar objects, and are not necessarily used to describe a particular sequence or to indicate importance. It will be understood that the terms so used are interchangeable under appropriate circumstances such that the embodiments described herein are capable of operation in other sequences than described or illustrated herein.

[0343] Although the present embodiments have been disclosed with reference to the above examples, the present application is not limited to the above examples. Any person skilled in the art, without departing from the spirit and scope of the present application, can make various changes and modifications, and thus the scope of the present application should be defined by the scope of the claims.

Claims

1. A method of manufacturing an alumina ceramic substrate, characterized by, The method comprises the following steps: selecting a pressing plate matched with the alumina ceramic substrate, and fixedly connecting the alumina ceramic substrate and the pressing plate; placing the alumina ceramic substrate and the pressing plate in a reaction chamber; determining a set heating rate in the reaction chamber according to parameter information of the alumina ceramic substrate, and performing sintering treatment on the alumina ceramic substrate; the step of determining the set heating rate in the reaction chamber according to the parameter information of the alumina ceramic substrate, and performing sintering treatment on the alumina ceramic substrate, comprises: determining the set heating rate in the reaction chamber according to size information and thickness information of the alumina ceramic substrate; performing heating operation on the reaction chamber according to the set heating rate, so as to perform sintering treatment on the alumina ceramic substrate; determining a morphological change of the alumina ceramic substrate in the sintering process, wherein the morphological change comprises a morphology and a morphological change rate of the alumina ceramic substrate; adjusting the heating rate in the reaction chamber according to the morphological change of the alumina ceramic substrate; the step of determining the morphological change of the alumina ceramic substrate in the sintering process, comprises: determining the morphological change of the alumina ceramic substrate in the sintering process by using a preset first neural network model, wherein the first neural network model is based on temperature distribution on the surface of the alumina ceramic substrate, and a heating rate and historical morphological change corresponding to the temperature distribution; the manufacturing method further comprises: determining flatness of the alumina ceramic substrate after sintering treatment by using a preset third neural network model, wherein the third neural network model is based on image information of the alumina ceramic substrate collected.

2. The production method according to claim 1, characterized by the first neural network model is based on temperature distribution on the surface of the alumina ceramic substrate, and a heating rate and historical morphological change corresponding to the temperature distribution, comprises: preprocessing the temperature distribution on the surface of the alumina ceramic substrate, and the heating rate and historical morphological change corresponding to the temperature distribution, to obtain a data set; dividing the data set into a training data set, a validation data set and a test data set; inputting the training data set into a preset first pre-training model, training the first pre-training model, and obtaining the first neural network model; inputting the validation data set into the first neural network model, and obtaining actual morphologies of the alumina ceramic substrate corresponding to each validation data set; adjusting the first neural network model according to the actual morphologies of the alumina ceramic substrate corresponding to each validation data set and historical morphologies of the alumina ceramic substrate contained in each validation data set; inputting the test data set into the adjusted first neural network model, and stopping training the first neural network model when the performance of the first neural network model meets a set performance.

3. The production method according to claim 1, characterized by the step of determining the morphological change of the alumina ceramic substrate in the sintering process by using the preset first neural network model, comprises: obtaining a current temperature and a current heating rate in the reaction chamber; inputting the current temperature and the current temperature increasing rate in the reaction chamber into the first neural network model to determine a predicted morphology and a predicted morphology change rate of the alumina ceramic substrate at the current temperature and the current temperature increasing rate.

4. The production method according to claim 3, characterized by The adjusting the temperature increasing rate in the reaction chamber according to the morphology change of the alumina ceramic substrate comprises: determining a set morphology of the alumina ceramic substrate at the current temperature and the current temperature increasing rate by the first neural network model based on the alumina ceramic substrate at the current temperature and the current temperature increasing rate; adopting a preset second neural network model to determine a set temperature increasing rate corresponding to the alumina ceramic substrate according to the set morphology of the alumina ceramic substrate at the current temperature and the current temperature increasing rate, wherein the second neural network model is based on the morphology change of the alumina ceramic substrate and the temperature increasing rate and temperature in the reaction chamber corresponding to the morphology change; adjusting the temperature increasing rate in the reaction chamber according to the current temperature increasing rate and the set temperature increasing rate of the alumina ceramic substrate.

5. The production method according to claim 3, wherein Further comprising: when it is determined that the predicted morphology change of the alumina ceramic substrate at the current temperature and the current temperature increasing rate is greater than the set morphology change, resetting the temperature and / or the temperature increasing rate in the reaction chamber.

6. The production method according to claim 1, characterized by The third neural network model is based on the collected image information of the alumina ceramic substrate and comprises: acquiring images of the alumina ceramic substrate at multiple viewing angles after sintering treatment and determining label information corresponding to each viewing angle image; preprocessing the images at each viewing angle to obtain multiple image data; dividing the multiple image data into a training image data set and a verification image data set; in any round, outputting the training image data set to a third pre-training model according to a set batch to train the third pre-training model to obtain the third neural network model; inputting the verification image data set into the third neural network model to obtain actual label information corresponding to each training image data; adjusting the third neural network model according to the actual label information corresponding to the verification image data set and the label information contained in the verification image data set.

7. An alumina ceramic substrate characterized by, comprise: obtained by the manufacturing method of any one of claims 1 to 6.

Images