Method for adjusting parameters of melting equipment for quartz crucible

Abstract

The present application relates to a method for adjusting the parameters of a melting device for a quartz crucible, and relates to the field of semiconductor material preparation. The method comprises: obtaining observed process parameters and a multi-process observed characteristic matrix during the melting process of multiple batches of quartz crucibles; using a density estimation method to determine the probability density distribution of the multi-process observed characteristic matrix, and screening out S valid batches from B batches; determining a first characteristic matrix according to the observed process parameters of the valid batches, determining a quality matrix according to the quality parameters of the quartz crucibles produced in the valid batches, and determining a first correlation coefficient matrix; determining the target parameters of the melting device from the candidate parameters based on the values ​​of each coefficient in the first correlation coefficient matrix, and determining the target time; inputting the target parameters at the target time into a pre-trained quality prediction model, obtaining a quality prediction result of the quartz crucible to be melted, and adjusting the parameters of the melting device. The use of this method can improve the accuracy and efficiency of parameter adjustment of the quartz crucible melting device.

Description

Technical Field

[0001] The invention relates to the field of semiconductor material preparation, and in particular to a method for adjusting parameters of melting equipment of a quartz crucible. Background Art

[0002] In the semiconductor industry, quartz crucibles play an important role in the production process of refining crystalline silicon due to their high purity and temperature resistance, and are key consumables in the production process of crystalline silicon. The arc method is the mainstream technology for preparing quartz crucibles. The production process includes feeding, vacuum arc melting, cooling and demolding, initial inspection, sandblasting cutting, secondary inspection, cleaning and packaging. Among them, the melting process is the core of the production process, which usually relies on a crucible melting machine to achieve.

[0003] In the related art, although automated means such as mechanical data recording and image acquisition are used to monitor the melting state of quartz sand, these methods still have shortcomings. Mainly due to the direct operation of mechanically recorded data, the lack of in-depth analysis of the data and real-time feedback makes it impossible to accurately adjust the arc size and control the process duration, making it difficult to achieve refined control of the melting process. This limitation affects the quality stability and consistency of the quartz crucible, limiting the efficiency and quality of crystalline silicon production. Therefore, the existing technology cannot meet the strict requirements of the semiconductor industry for high-purity and high-performance quartz crucibles, especially in improving the accuracy and efficiency of adjusting the parameters of the melting equipment.

[0004] In view of this, the present invention proposes a method for adjusting parameters of a melting device for a quartz crucible. Summary of the invention

[0005] Based on this, it is necessary to provide a method for adjusting parameters of a quartz crucible melting device in order to solve the above problems. The method can improve the accuracy and efficiency of parameter adjustment of the quartz crucible melting device.

[0006] In a first aspect, the present application provides a method for adjusting parameters of a melting device of a quartz crucible, the method comprising:

[0007] Obtain the observed process parameters in the multi-batch melting process of the quartz crucible, and construct a multi-process observation feature matrix of the shape of T×B according to the observed process parameters; wherein the observed process parameters include multiple candidate parameters of the melting equipment in the melting process, B is the batch number of the melting process, and T is the arc ignition time of each melting process;

[0008] The density estimation method is used to determine the probability density distribution of the multi-process observation feature matrix, and S valid batches are selected from B batches according to the probability density distribution of the multi-process observation feature matrix;

[0009] Determine a first characteristic matrix according to the observed process parameters of the S valid batches, determine a quality matrix according to the quality parameters of the quartz crucibles produced by the S valid batches, and determine a first correlation coefficient matrix based on the first characteristic matrix and the quality matrix;

[0010] Based on the coefficient values ​​in the first correlation coefficient matrix, the target parameters of the melting equipment are determined from the candidate parameters, and the target time is determined according to the observed process parameters of the S valid batches and the quality parameters of the quartz crucibles produced by the S valid batches; wherein the target time belongs to the arc ignition time t of the arc ignition time T, and the target parameter is a candidate parameter whose correlation with the quality parameter of the quartz crucible is greater than a preset value;

[0011] The target parameters at the target time are input into the pre-trained quality prediction model to obtain the quality prediction result of the quartz crucible to be melted, and the parameters of the melting equipment are adjusted based on the quality prediction result.

[0012] In one embodiment, a density estimation method is used to determine the probability density distribution of a multi-process observation feature matrix, including:

[0013] Slide along the direction of increasing arc ignition time T with a time window of set length to obtain multiple time windows;

[0014] For each time window, kernel density calculation is performed on each observation sample point in the time window to obtain the probability density of each observation sample point in the time window. Based on the probability density of each observation sample point in the time window, the density matrix of the time window is determined; wherein the observation sample point is any candidate parameter of any batch at any arc ignition time t.

[0015] The density matrices of all time windows are connected end to end to obtain the probability density distribution of the multi-process observation feature matrix.

[0016] In one embodiment, the arc ignition time T includes multiple arc ignition moments t; according to the probability density distribution of the multi-process observation feature matrix, S valid batches are screened out from B batches, including:

[0017] Based on the probability density distribution of the multi-process observation feature matrix, along the direction of increasing arc ignition time T, for each arc ignition time t, the target position where the probability density maximum value is located is located;

[0018] Connect the target positions of multiple arc ignition times t into a line to obtain the center line of the consistency test;

[0019] The screening limit is determined based on the center line, and the batches beyond the screening limit are identified as abnormal batches. The abnormal batches are removed from B batches to obtain S valid batches.

[0020] In one embodiment, determining the screening limit based on the center line includes:

[0021] Determine the standard deviation of the center line, and use the preset multiple standard deviations above and below the center line as the boundaries to determine the upper and lower screening limits;

[0022] Determine the screening limits based on the upper and lower screening limits.

[0023] In one embodiment, a global average is calculated for each candidate parameter p in each batch s. , we get the S×P feature matrix, where s belongs to S, p belongs to P, P is the number of candidate parameters, and the S×P feature matrix is ​​the first feature matrix; the quality matrix is ​​the S×L feature matrix, and L is the number of quality indicators;

[0024] Based on the first characteristic matrix and the quality matrix, a first correlation coefficient matrix is ​​determined, including:

[0025] For the pth candidate parameter mean and the first quality inspection index , calculate the Pearson correlation coefficient;

[0026] The correlation coefficient calculation results of the P candidate parameters and the L quality inspection indicators are combined to obtain a first correlation coefficient matrix with a shape of P×L.

[0027] In one embodiment, for the pth candidate parameter mean and the first quality inspection index , the calculation formula for Pearson correlation coefficient is ,in is the mean of the pth candidate parameter mean sequence, is the mean of the lth quality inspection index sequence.

[0028] In one embodiment, determining the target time according to observed process parameters of S valid batches and quality parameters of quartz crucibles produced by the S valid batches includes:

[0029] For each batch s, the correlation coefficient is calculated based on the candidate parameter p and the quality parameter at each arc ignition time t in the batch s, and a correlation coefficient sequence of P bars with a length of T is obtained;

[0030] For each candidate parameter p, the target time is determined through the correlation coefficient sequence.

[0031] In one embodiment, a multi-process observation feature matrix with a shape of T×B is constructed according to the observation process parameters, including:

[0032] With the batch number B as the column vector and the observed process parameters of each batch b as the row vector, they are aligned according to the arc ignition time T of the melting process to construct a multi-process observation feature matrix with a shape of T×B.

[0033] In a second aspect, the present application further provides a device for adjusting parameters of a melting device of a quartz crucible, the device comprising:

[0034] An acquisition module is used to acquire observed process parameters during the multi-batch melting process of a quartz crucible, and to construct a multi-process observation feature matrix of a shape of T×B according to the observed process parameters; wherein the observed process parameters include multiple candidate parameters of a melting device during the melting process, B is the number of batches of the melting process, and T is the arc ignition time of each melting process;

[0035] A screening module is used to determine the probability density distribution of the multi-process observation feature matrix by using a density estimation method, and screen S valid batches from B batches according to the probability density distribution of the multi-process observation feature matrix;

[0036] A calculation module is used to determine a first characteristic matrix based on the observed process parameters of S valid batches, determine a quality matrix based on the quality parameters of the quartz crucibles produced by the S valid batches, and determine a first correlation coefficient matrix based on the first characteristic matrix and the quality matrix; the calculation module is also used to determine the target parameters of the melting equipment from the candidate parameters based on the coefficient values ​​in the first correlation coefficient matrix, and determine the target time based on the observed process parameters of the S valid batches and the quality parameters of the quartz crucibles produced by the S valid batches; wherein the target time belongs to the arc ignition time t of the arc ignition time T, and the target parameter is a candidate parameter whose correlation with the quality parameter of the quartz crucible is greater than a preset value;

[0037] The adjustment module is used to input the target parameters at the target time into the pre-trained quality prediction model, obtain the quality prediction result of the quartz crucible to be melted, and adjust the parameters of the melting equipment based on the quality prediction result.

[0038] In a third aspect, the present application further provides an electronic device, which includes a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, the method for adjusting the melting equipment parameters of the quartz crucible of the first aspect is implemented.

[0039] The above-mentioned method for adjusting the melting equipment parameters of the quartz crucible constructs a multi-process observation feature matrix by acquiring the observed process parameters of multiple batches of melting processes. The matrix has a shape of T×B, wherein T represents the arc ignition time of each melting process, and B represents the number of batches of the melting process, covering multiple candidate parameters of the melting equipment; the probability density distribution of the matrix is ​​determined by using a density estimation method, and S valid batches are screened out based on this distribution to eliminate the interference of abnormal data; the first feature matrix and the quality matrix are determined respectively according to the observed process parameters and quality parameters of the valid batches, and then the first correlation coefficient matrix is ​​calculated to reveal the correlation between the candidate parameters and the quality parameters; the target parameters and target moments are determined based on the coefficient values ​​of the first correlation coefficient matrix, wherein the target parameters are candidate parameters whose correlation with the quality parameters of the quartz crucible is greater than a preset value, and the target moment is a part of the arc ignition moment t in the arc ignition time T; the target parameters at the target moment are input into a pre-trained quality prediction model to obtain a quality prediction result, and the melting equipment parameters are adjusted accordingly, so that the method can accurately determine the target parameters and target moments that affect the quality of the quartz crucible, and dynamically adjust the equipment parameters through the prediction model to improve the accuracy and efficiency of parameter adjustment of the quartz crucible melting equipment. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] Figure 1 The present invention is a process of a method for adjusting parameters of a melting device of a quartz crucible in one embodiment;

[0041] Figure 2 A flowchart for determining a probability density distribution of a multi-process observation feature matrix in one embodiment;

[0042] Figure 3 A flowchart of selecting S valid batches from B batches according to the probability density distribution of a multi-process observation feature matrix in one embodiment;

[0043] Figure 4 A flowchart of determining a screening limit based on a center line in one embodiment;

[0044] Figure 5 A flowchart of determining a first correlation coefficient matrix based on a first feature matrix and a quality matrix in one embodiment;

[0045] Figure 6 A flowchart for determining a target time in one embodiment;

[0046] Figure 7 It is a device diagram of a device for adjusting parameters of a melting device of a quartz crucible in one embodiment;

[0047] Figure 8 FIG. 4 is a diagram showing the internal structure of an electronic device in one embodiment. DETAILED DESCRIPTION

[0048] In order to make the purpose, technical scheme and advantages of the present application clearer, the present application is further described in detail below in conjunction with the accompanying drawings and Examples. It should be understood that the specific embodiments described herein are only used to explain the present application and are not intended to limit the present application. Unless otherwise defined, the technical terms or scientific terms involved in the present application should have the general meaning understood by people with general skills in the technical field to which the present application belongs. "One", "one", "a kind of", "the", "these" and similar words in the present application do not represent quantitative restrictions, and they can be singular or plural.

[0049] In one embodiment, Figure 1 As shown, a method for adjusting parameters of a melting device of a quartz crucible is provided, the method comprising the following steps:

[0050] Step 101: obtaining observed process parameters during the multi-batch melting process of a quartz crucible, and constructing a multi-process observed feature matrix of a shape of T×B according to the observed process parameters; wherein the observed process parameters include multiple candidate parameters of a melting device during the melting process, B is the batch number of the melting process, and T is the arc ignition time of each melting process;

[0051] Specifically, the observed process parameters of multiple batches of quartz crucibles during the melting process are obtained. The observed process parameters may include but are not limited to multiple candidate parameters related to the melting equipment during the melting process, such as temperature parameters, pressure parameters, current parameters, and voltage parameters, etc. The candidate parameters can reflect the working state of the equipment and the characteristics of the melting process. After obtaining the candidate parameters, a multi-process observation feature matrix of the shape of T×B can be constructed according to the observed process parameters. In the multi-process observation feature matrix, T can represent the arc ignition time of each batch of the melting process, and B can represent the number of batches of the melting process, that is, the total number of batches of collected data.

[0052] For example, in the melting process, the observed process parameter values ​​of different batches at different time points can be directly obtained, that is, the observed process parameter values ​​directly include temperature parameter values, pressure parameter values, and current parameter values. The batch number of the melting process is used as the column vector, and the observed process parameter value of each batch is used as the row vector, which is aligned according to the arc ignition time of the melting process. A multi-process observation feature matrix with a shape of T×B is obtained, which is shown as follows:

[0053] .

[0054] in, represents the observed process parameter value of the jth batch at the i-th time point. represents the observed process parameter value of the jth batch at the i-th time point. For example It can refer to the observed process parameter value of the third batch at the second time point. It should be noted that The parameter value set of all candidate parameters, such as the temperature parameter value, the pressure parameter value, and the current parameter value, may be included in the jth batch at the i-th time point.

[0055] Step 102: using a density estimation method to determine the probability density distribution of the multi-process observation feature matrix, and selecting S valid batches from B batches according to the probability density distribution of the multi-process observation feature matrix;

[0056] Specifically, the density estimation method is used to analyze the multi-process observation feature matrix, aiming to determine the probability density distribution of the data points in the matrix. Using the probability density distribution, batches that are consistent with the statistical characteristics of the overall data set can be obtained. Furthermore, batches that are significantly different from the overall distribution are considered abnormal batches, and batches that are similar to the overall distribution can be determined as valid batches. By comparing the probability density value of each batch data point with the overall distribution, S valid batches that match the statistical characteristics of the overall data set can be screened out from B batches.

[0057] For example, the multi-process observation matrix exemplified in step 101 can be analyzed to obtain batches in which the probability density distribution of the data of the first and third batches is consistent with the statistical characteristics of the overall data set, while the data of the second batch deviates greatly from the statistical characteristics of the overall data set. Then, the valid batches S that can be screened out from the three batches can be two, that is, the first batch and the third batch are determined to be valid batches.

[0058] Step 103: determining a first characteristic matrix according to the observed process parameters of the S valid batches, determining a quality matrix according to the quality parameters of the quartz crucibles produced by the S valid batches, and determining a first correlation coefficient matrix based on the first characteristic matrix and the quality matrix;

[0059] Specifically, the first characteristic matrix is ​​determined based on the observed process parameters of the S valid batches that have been screened, and the first characteristic matrix is ​​constructed based on the observed process parameters in the S valid batches. The quality matrix can be a quality matrix determined based on the quality parameters of the quartz crucibles produced in the S valid batches. After obtaining the first characteristic matrix and the quality matrix, the first correlation coefficient matrix can be constructed by calculating the correlation between each candidate parameter in the first characteristic matrix and each quality parameter in the quality matrix.

[0060] For example, the first characteristic matrix is ​​determined for the observed process parameters of S valid batches. Assume that S=2 and there are three observed process parameters, namely temperature parameter, pressure parameter, and current parameter. Temperature parameter T1=1200°C, pressure parameter P1=0.08MPa, and current parameter I1=80A. In the second valid batch, temperature parameter T2=1230°C, pressure parameter P2=0.07MPa, and current parameter I2=85A. The first characteristic matrix can be obtained as:

[0061] .

[0062] The quality matrix is ​​determined based on the quality parameters of quartz crucibles produced in S=2 valid batches. Assume that there are two quality parameters, namely the crucible outer diameter D and thickness H. The first valid batch of crucibles has an outer diameter D1=20cm and a thickness H1=2cm; the second valid batch of crucibles has an outer diameter D2=21cm and a thickness H2=1.8cm. The quality matrix can be obtained as:

[0063] .

[0064] Furthermore, the first correlation coefficient matrix is ​​determined based on the first feature matrix and the quality matrix. The correlation coefficient between each row element in the first feature matrix and each column element in the quality matrix can be calculated by a correlation analysis method, such as a Pearson correlation coefficient calculation method. The first correlation coefficient matrix may be:

[0065] .

[0066] in, It can express the correlation coefficient between the temperature parameter and the outer diameter of the crucible, The correlation coefficient between the temperature parameter and the crucible thickness can be expressed as It can express the relationship between pressure parameters and crucible outer diameter. The correlation coefficient can reflect the degree of correlation between the observed process parameters and the quartz crucible quality parameters.

[0067] Step 104: determining target parameters of the melting equipment from candidate parameters based on the coefficient values ​​in the first correlation coefficient matrix, and determining target time according to the observed process parameters of the S valid batches and the quality parameters of the quartz crucibles produced by the S valid batches; wherein the target time belongs to the arc ignition time t of the arc ignition time T, and the target parameter is a candidate parameter whose correlation with the quality parameter of the quartz crucible is greater than a preset value;

[0068] Specifically, the target parameter is determined based on each coefficient value in the first correlation coefficient matrix. The target parameter is a candidate parameter whose correlation with the quality parameter of the quartz crucible is greater than a preset value, and the target parameter can have a significant impact on the product quality. By analyzing the observed process parameters of S valid batches and the corresponding quartz crucible quality parameters, the target moment can be determined. The target moment can belong to a specific part of the arc ignition time t within the arc ignition time T. The target moment can represent a time period that needs special attention in the melting process. At the target moment, the change of the target parameter has a decisive influence on the quality of the quartz crucible.

[0069] Step 105: inputting the target parameters at the target time into the pre-trained quality prediction model to obtain the quality prediction result of the quartz crucible to be melted, and adjusting the parameters of the melting equipment based on the quality prediction result.

[0070] Specifically, the target parameters at the determined target time are used as inputs and input into a pre-trained quality prediction model, wherein the quality prediction model can be trained based on historical data and machine learning technology, and can predict the quality characteristics of the quartz crucible according to the input parameters. After the quality prediction model processes the input data, it outputs a prediction result, which is a prediction of the quality of the quartz crucible to be melted. According to the quality prediction result, the parameters of the melting equipment, such as temperature parameters, pressure parameters, etc., are adjusted to ensure that the quartz crucible actually produced can meet the expected quality standards.

[0071] In this embodiment, a multi-process observation feature matrix is ​​constructed by acquiring observed process parameters of multiple batches of melting processes. The matrix has a shape of T×B, where T represents the arc ignition time of each melting process, and B represents the number of batches of the melting process, covering multiple candidate parameters of the melting equipment; a probability density distribution of the matrix is ​​determined by a density estimation method, and S valid batches are screened out based on this distribution to eliminate the interference of abnormal data; a first feature matrix and a quality matrix are determined respectively according to the observed process parameters and quality parameters of the valid batches, and then a first correlation coefficient matrix is ​​calculated to reveal the correlation between the candidate parameters and the quality parameters; a target parameter and a target time are determined based on the coefficient value of the first correlation coefficient matrix, where the target parameter is a candidate parameter whose correlation with the quartz crucible quality parameter is greater than a preset value, and the target time is a part of the arc ignition time t in the arc ignition time T; the target parameter at the target time is input into a pre-trained quality prediction model to obtain a quality prediction result, and the melting equipment parameters are adjusted accordingly, so that the method can accurately determine the target parameters and target time that affect the quality of the quartz crucible, and dynamically adjust the equipment parameters through the prediction model to improve the accuracy and efficiency of parameter adjustment of the quartz crucible melting equipment.

[0072] In one embodiment, Figure 2As shown, the density estimation method is used to determine the probability density distribution of the multi-process observation feature matrix, including the following steps:

[0073] Step 201: sliding a time window of a set length along a direction in which the arc ignition time T increases to obtain a plurality of time windows;

[0074] Specifically, a time window of fixed length can be set along the direction of increasing arc ignition time T, that is, the time sequence from the beginning to the end of the melting process, and the time window can be slid over the entire arc ignition time T. Exemplarily, the process is similar to measuring from left to right on the time axis with a ruler of fixed size, and the area covered by each ruler is a time window. Further, multiple time windows can be obtained, and each time window can contain observation data within a certain time range.

[0075] Step 202: For each time window, perform kernel density calculation on each observation sample point in the time window to obtain the probability density of each observation sample point in the time window, and determine the density matrix of the time window based on the probability density of each observation sample point in the time window; wherein the observation sample point is any candidate parameter of any batch at any arc ignition time t.

[0076] Specifically, in this embodiment, the density estimation method can adopt a sliding time window local normalization kernel density estimation method to improve the problem that the traditional kernel density estimation algorithm is slow when processing big data. The method specifically includes: using a time window of a specified length window_len (generally 3) in time T Slide in the increasing direction, and calculate the probability density of each observation sample point for each window by the following formula:

[0077] ,

[0078] in, t Indicates time, b Indicates batch, Indicated in t Moment b Kernel density estimation of batch observation points; n is the total number of observed samples; h t and h b Represents the kernel function K exist t and b The bandwidth parameter in the direction is used to control the extension range of the kernel function, affecting the smoothness and accuracy of the estimate. It is expressed by the formula Calculated, is the standard deviation; Indicates the estimated standard deviation between the observed data point and the time mean in the time dimension, Represents the estimated standard deviation between the observed data points and the batch mean in the batch dimension. ( x i , y i ) indicates the i Sample points, and Represent the values ​​of the observed sample points in the time dimension and batch dimension respectively. and Represent the value of the i-th observation sample point in the time dimension t and the batch dimension b respectively. K (.) is the kernel function, usually a Gaussian kernel function is chosen, such as .

[0079] Normalize the probability density matrix calculated in each sliding time window, that is, i The normalized result of the points is ,in, Represents the original data points The standardized or normalized value, x min is the minimum value of all density values, x max is the maximum value of all density values.

[0080] After completing the calculation of all sample points, the probability density values ​​of the sample points in each time window are obtained, and the probability density values ​​form the density matrix of the time window. The density matrix can record the probability density of each sample point in the time window, which can reflect the density of the parameter distribution in the time window.

[0081] Step 203: Connect the density matrices of all time windows end to end to obtain the probability density distribution of the multi-process observation feature matrix.

[0082] Specifically, based on a density matrix that has been calculated in step 202, the density matrix can contain the probability density values ​​of all observed sample points in the time window. The density matrices of the time windows are connected end to end in chronological order, that is, the density matrix of the first time window is placed immediately after the density matrix of the second time window, and so on, until the entire arc ignition time T is covered. It should be noted that this process is similar to splicing a series of pictures into a video in sequence, and each picture corresponds to the density information of a time window. Furthermore, a probability density distribution of a multi-process observation feature matrix can be obtained, which can describe in detail the probability density changes of candidate parameters at different time points and different batches during the entire melting process.

[0083] In this embodiment, a time window of a set length is slid in the direction of the arc ignition time T to cover the entire melting process and generate multiple time windows; kernel density estimation is performed on the observed sample points in each time window, and the sample points represent any candidate parameter of any batch at any arc ignition time t, thereby obtaining the probability density of the sample points in each time window; the density matrix of each time window is constructed according to the probability density value; the density matrices of all time windows are connected end to end to form a probability density distribution diagram of a complete multi-process observation feature matrix. This method can provide a comprehensive probabilistic framework to capture and analyze the statistical characteristics of parameter changes in the melting process.

[0084] In one embodiment, Figure 3 As shown, the arc ignition time T includes multiple arc ignition moments t; according to the probability density distribution of the multi-process observation feature matrix, S valid batches are screened out from B batches, including the following steps:

[0085] Step 301: based on the probability density distribution of the multi-process observation feature matrix, along the direction of increasing arc ignition time T, for each arc ignition time t, locate the target position where the probability density maximum value is located;

[0086] Specifically, the arc ignition time T is divided into multiple arc ignition moments t, and the probability density distribution of the multi-process observation feature matrix provides the distribution of each parameter in different batches and time points. At each arc ignition moment t of the arc ignition time T, the target position with the maximum probability density is found. For each specific arc ignition moment t, it is necessary to check the data of all batches corresponding to that moment in the multi-process observation feature matrix, and determine which position (that is, which parameter of which batch) has the highest probability density value in all batches of data.

[0087] Step 302: Connect the target positions of multiple arc ignition times t into a line to obtain a center line for consistency inspection;

[0088] Specifically, in step 301, the target position at each arc ignition time t has been determined. The target position may refer to the point where the probability density is the maximum, representing the parameter value that is most likely to affect the quality of the quartz crucible at each time. The target positions scattered at different arc ignition times t are connected in series in chronological order to obtain a continuous center line. Each target position corresponds to a specific arc ignition time t and represents the position of the key parameter at that time.

[0089] Step 303: Determine the screening limit based on the center line, determine the batches beyond the screening limit as abnormal batches, and remove the abnormal batches from the B batches to obtain S valid batches.

[0090] Specifically, the screening limit is determined based on the consistency test center line determined in step 302. The screening limit can be set based on a certain range of fluctuations of the center line, that is, any batch beyond the range is considered abnormal. After determining the screening limit, the data of all B batches are compared with the center line to determine which batches have data points near the center line, that is, within the screening limit. Batches beyond the screening limit are considered abnormal batches, that is, they show greater inconsistency compared with most batches, representing abnormal conditions or deviations in the production process. The abnormal batches are removed from the B batches, and S valid batches that maintain good consistency with the center line are screened.

[0091] In this embodiment, the arc-starting time T is divided into a plurality of arc-starting moments t, and based on the probability density distribution of the multi-process observation characteristic matrix, the target position where the probability density maximum value is located is located for each arc-starting moment t; the target positions are connected into a line to form the center line of the consistency test; the screening limit is determined based on the center line, and the abnormal batches exceeding the limit are identified, and finally the abnormal batches are removed from the B batches to obtain S valid batches. This method improves the production consistency and reliability of the quartz crucible melting equipment by accurately identifying and screening out the abnormal batches in the production process, and by locating the target position where the probability density maximum value is located and constructing the center line, after the valid batches are screened out, the parameters of the batches can be adjusted in a targeted manner to improve the production efficiency.

[0092] In one embodiment, Figure 4 As shown, determining the screening limit based on the center line includes the following steps:

[0093] Step 401: determine the standard deviation of the center line, and determine the upper and lower screening limits based on the preset multiple standard deviations above and below the center line;

[0094] Specifically, the screening limit can be set based on the standard deviation of the consistency test center line. The standard deviation of the center line is calculated. The standard deviation is a statistical measure that can indicate the degree of dispersion of the distribution of the values ​​of each point on the center line. After determining the standard deviation, the upper and lower limits of the screening can be determined according to a preset multiple (for example, the commonly used 3 times the standard deviation). The upper limit of the screening can be the standard deviation of the center line plus the preset multiple, and the lower limit of the screening can be the standard deviation of the center line minus the preset multiple.

[0095] Step 402: Determine the screening limit based on the screening upper limit and the screening lower limit.

[0096] Specifically, in step 401, the standard deviation of the center line has been calculated, and the upper and lower limits of the screening are determined based on the standard deviation and a preset multiple (such as 3 times). The screening limit includes two parts: the upper screening limit and the lower screening limit. The upper screening limit is the center line value plus the standard deviation of the preset multiple, and the lower screening limit is the center line value minus the standard deviation of the preset multiple. Any batch exceeding the screening limit will be marked as abnormal. The parameter value of each batch can be compared with the screening limit. If the parameter value of the batch falls between the upper screening limit and the lower screening limit, the batch is considered normal, that is, it belongs to a valid batch; if the parameter value of the batch exceeds the range of the screening limit, the batch is considered abnormal and will be excluded from the valid batch.

[0097] In this embodiment, the method first determines the standard deviation of the consistency test center line, and then determines the upper and lower limits of the screening based on the standard deviation of the preset multiples (such as 3 times) above and below the center line, thereby forming the screening limit. Based on the upper and lower limits, it is clearly defined which batches of data fall within the normal range, thereby identifying valid batches, and those batches that exceed the range of the screening limit are regarded as abnormal batches and are excluded. This method provides a scientific and systematic way to identify variations and anomalies in the production process, thereby ensuring that only those batches with stable quality are selected as valid batches, which can improve the production consistency and reliability of quartz crucibles, reduce quality risks caused by abnormal batches, and optimize production efficiency.

[0098] In one embodiment, Figure 5 As shown, the global average is calculated for each candidate parameter p in each batch s , we get the S×P feature matrix, where s belongs to S, p belongs to P, P is the number of candidate parameters, and the S×P feature matrix is ​​the first feature matrix; the quality matrix is ​​the S×L feature matrix, and L is the number of quality indicators;

[0099] Determining a first correlation coefficient matrix based on the first feature matrix and the quality matrix includes the following steps:

[0100] Step 501: For the pth candidate parameter mean and the first quality inspection index , calculate the Pearson correlation coefficient;

[0101] Specifically, for batch s, the parameter value of the pth candidate parameter is measured multiple times to ensure the accuracy of the parameter value and prevent deviation. The average of the pth candidate parameter values ​​obtained by multiple measurements is calculated, that is, all the measured parameter values ​​of the pth parameter in the batch are added together and divided by the number of parameter values ​​to obtain the global average. . Global average It can represent the average value of the pth parameter in batch s.

[0102] Furthermore, there are S batches and P candidate parameters, and a feature matrix of shape S×P can be constructed, where S represents the number of valid batches, S is a row vector, P represents the number of candidate parameters, and P is a column vector. This matrix is ​​called the first feature matrix. The first feature matrix can contain the global average of all candidate parameters in all batches.

[0103] It should be noted that a quality matrix needs to be constructed. The quality matrix is ​​an S×L feature matrix, where S represents the number of valid batches and L represents the number of quality indicators. The quality matrix contains the observed values ​​of all quality indicators in all batches. In batch s, l A quality inspection indicator.

[0104] Based on the first feature matrix and the quality matrix, the Pearson correlation coefficient is calculated. For each candidate parameter mean in the first feature matrix And each quality inspection indicator in the quality matrix , calculate the mean of candidate parameters Quality inspection indicators The Pearson correlation coefficient between two variables. The Pearson correlation coefficient is a statistic that measures the degree of linear correlation between two variables. Its value is between -1 and 1. The closer the value is to 1 or -1, the stronger the linear relationship between the two variables is. The value close to 0 indicates that there is almost no linear relationship.

[0105] For example, in a batch, taking the pth candidate parameter temperature parameter as an example, the global average of the temperature parameter in the batch is For the quality inspection index, take the lth quality inspection index as an example, the quality inspection index value of this batch is Assuming that the Pearson correlation coefficient between the two is 0.8, it means that the temperature parameter in the batch is highly correlated with the lth quality inspection index.

[0106] Step 502: Combine the correlation coefficient calculation results of the P candidate parameters and the L quality inspection indicators to obtain a first correlation coefficient matrix of a shape of P×L.

[0107] Specifically, for each candidate parameter p, the mean The Pearson correlation coefficient between the candidate parameter and the quality indicator l is calculated. The Pearson correlation coefficient can reflect the linear correlation strength between the candidate parameter and the quality indicator. The scattered correlation coefficients are integrated according to the combination of the candidate parameter P and the quality indicator L. Specifically, for each candidate parameter p, the mean There is a correlation coefficient sequence corresponding to all quality indicators l. For each quality indicator l, there is also a mean value with all candidate parameters p. Corresponding correlation coefficient sequence. The sequence is integrated into a matrix of shape P×L according to the correspondence between parameters and indicators. In the first correlation coefficient matrix, the rows represent candidate parameters p, the columns represent quality indicators l, and each element in the matrix is ​​the Pearson correlation coefficient between the corresponding candidate parameters and the quality indicators.

[0108] Exemplarily, it is assumed that there are three candidate parameters, P=3, which are temperature parameter, pressure parameter and current parameter, and there are two quality inspection indicators L=2, such as the thickness and hardness of the crucible.

[0109] Based on step 501, the Pearson correlation coefficient of temperature and crucible thickness is calculated, and the result is assumed to be 0.8, indicating that there is a strong positive correlation between the temperature parameter and the crucible thickness. Further, the correlation coefficient of the temperature parameter and the crucible hardness is calculated, and the result is assumed to be 0.3, indicating that the correlation between the two is relatively weak. Based on the same method, the correlation coefficient of the pressure parameter and the crucible thickness is calculated, and it is assumed to be -0.5, indicating that there is a negative correlation between the two. The correlation coefficient of the pressure parameter and the crucible hardness is calculated, and it is assumed to be 0.1, indicating that the correlation between the two is weak. Further, the correlation coefficient of the current parameter and the crucible thickness is calculated, and it is assumed to be 0.7, indicating that there is a positive correlation between the two, and the correlation coefficient of the current parameter and the crucible hardness is assumed to be -0.2, indicating that the two are negatively correlated. The calculation results are combined to form the first correlation coefficient matrix:

[0110] .

[0111] In the first correlation coefficient matrix, the first row represents the correlation coefficients of the temperature parameter with the crucible thickness and hardness, the second row represents the correlation coefficients of the pressure parameter with the crucible thickness and hardness, and the third row represents the correlation coefficients of the current parameter with the crucible thickness and hardness.

[0112] The first correlation coefficient matrix can further analyze which candidate parameters have a greater impact on the quality indicators of the crucible and which have a smaller impact, thereby providing a basis for subsequent production process adjustment and optimization. For example, if the thickness of the crucible needs to be increased, the first correlation coefficient matrix can be used to obtain the temperature and current parameters that need to be focused on and adjusted.

[0113] In this embodiment, the method can reveal the degree and direction of the influence of different process parameters on product quality, and provide a quantitative basis for the adjustment of process parameters. By analyzing the first correlation coefficient matrix, the key parameters that have a significant impact on product quality can be identified, so that the parameters can be optimized in a targeted manner to improve the production efficiency and product quality of the quartz crucible.

[0114] In one embodiment, for the pth candidate parameter mean and the first quality inspection index , the calculation formula for Pearson correlation coefficient is

[0115] ,

[0116] in is the mean of the pth candidate parameter mean sequence, is the mean of the lth quality inspection index sequence.

[0117] Specifically, in this embodiment, the mean of the pth candidate parameter is calculated and the first quality inspection index The formula for the Pearson correlation coefficient is as follows:

[0118] .

[0119] In this formula, represents the mean of the p-th candidate parameter in the s-th batch, is the overall mean of the means of the p-th candidate parameters in all batches, is the value of the lth quality indicator in the sth batch, is the overall mean of the lth quality indicator in all batches. By taking the difference, that is, the difference between the mean of the candidate parameter in each batch and the overall mean of the parameter and the difference between the quality index in each batch and the overall mean of the index . Multiply the differences in pairs and sum them to get the numerator. The denominator is the product of the standard deviations of the two sequences (the candidate parameter mean sequence and the quality indicator sequence). Each standard deviation is obtained by calculating the square root of the sum of the squares of the differences of the respective sequences. The result of the formula r is the Pearson correlation coefficient, which can quantify the degree of linear correlation between two variables. Its value is between -1 and 1. The closer the value is to 1 or -1, the stronger the linear relationship between the two variables is. The value close to 0 indicates that there is almost no linear relationship.

[0120] By calculating the Pearson correlation coefficient between the mean of each candidate parameter and each indicator, a first correlation coefficient matrix with a shape of P×L can be obtained, as shown below:

[0121]

[0122] in, Represents the correlation coefficient between the mean of the Pth candidate parameter and the Lth indicator.

[0123] Furthermore, for each value in the correlation coefficient matrix, the degree of its influence on quality is evaluated according to the size of the coefficient. The positive correlation coefficient indicates the strength of the positive relationship between the process parameter and the quality inspection index. A value greater than 0.8 indicates a very high positive correlation, 0.6-0.8 indicates a high positive correlation, 0.4-0.6 indicates a medium positive correlation, 0.2-0.4 indicates a weak positive correlation, and 0-0.2 indicates no correlation. The negative correlation coefficient indicates a negative correlation, a value less than -0.8 indicates a very high negative correlation, -0.6--0.8 indicates a high negative correlation, -0.4--0.6 indicates a medium negative correlation, -0.2--0.4 indicates a weak negative correlation, and 0--0.2 indicates no correlation.

[0124] In this embodiment, the method can provide a quantitative measure to indicate the strength of the linear relationship between the process parameters and the product quality indicators. By calculating the Pearson correlation coefficient, it is possible to identify which process parameters have a significant impact on product quality, thereby guiding the adjustment of process parameters, optimizing the production process, and improving product quality.

[0125] In one embodiment, Figure 6 As shown, according to the observed process parameters of S valid batches and the quality parameters of the quartz crucibles produced by the S valid batches, determining the target time comprises the following steps:

[0126] Step 601: for each batch s, calculate the correlation coefficient based on the candidate parameter p and the quality parameter at each arc ignition time t in the batch s, and obtain a correlation coefficient sequence of P bars with a length of T;

[0127] Specifically, at each arc ignition time t, a candidate parameter p and a quality parameter are selected, and the correlation coefficient between the candidate parameter p and the quality parameter is calculated. Through calculation, a correlation coefficient sequence with a length of T can be generated for each candidate parameter p. Each element in the sequence corresponds to an arc ignition time t, which can reflect the strength of the correlation between the parameter p and the quality parameter at that time. P correlation coefficient sequences can be obtained for each batch s, each sequence corresponds to a candidate parameter p, and the length is T, that is, all moments within the arc ignition time T.

[0128] Step 602: For each candidate parameter p, determine the target time through the correlation coefficient sequence.

[0129] Specifically, in step 601, a correlation coefficient sequence of length T has been generated for each candidate parameter p of each batch s, and the correlation coefficient sequence can reflect the strength of correlation between the candidate parameter p and the quality parameter at each moment within the arc ignition time T. For each candidate parameter p, it is necessary to analyze its corresponding correlation coefficient sequence to determine the specific moment with the greatest correlation with the quality parameter, which is called the target moment. Check the correlation coefficient sequence of the candidate parameter p and find the moment with the largest absolute value of the correlation coefficient in the sequence. The moment indicates that the candidate parameter p has the most significant impact on the quality parameter. Since each element in the sequence corresponds to a specific arc ignition moment t, the target moment can be directly identified from the sequence. For each candidate parameter p, there may be multiple moments with the largest absolute value of the correlation coefficient, which can all be considered as target moments.

[0130] In this embodiment, the correlation coefficient of each candidate parameter p and the quality parameter in each batch s is calculated at each arc ignition time t, and a correlation coefficient sequence of P bars with a length of T is generated, where P represents the number of candidate parameters and T represents the length of the arc ignition time; for each candidate parameter p, by analyzing its correlation coefficient sequence, the moment with the greatest correlation with the quality parameter, i.e., the target moment, is found. This method can accurately identify which specific moments in the melting process have a significant impact on the quality of the quartz crucible. By determining the target moment, the parameters of the melting equipment can be adjusted more specifically to optimize the production process.

[0131] In one embodiment, a multi-process observation feature matrix with a shape of T×B is constructed based on the observed process parameters, including: taking the batch number B as the column vector, taking the observed process parameters of each batch b as the row vector, and aligning them according to the arc ignition time T of the melting process to construct a multi-process observation feature matrix with a shape of T×B.

[0132] Specifically, the construction of the multi-process observation feature matrix begins with determining the column vector, which can represent the batch number B, that is, all the batches considered. For each batch b, its observed process parameters during the melting process are obtained. The observed process parameters may include temperature parameters, pressure parameters, current parameters, etc. The observed process parameters can be arranged into row vectors. The row vectors can be aligned in the order of the arc ignition time T of the melting process. Further, each row represents the parameter value of all batches at a time point, and each column represents the parameter change of a specific batch at different time points. Furthermore, the combination of row vectors and column vectors forms a multi-process observation feature matrix with a shape of T×B, where T represents the total duration of the arc ignition time and B represents the number of batches.

[0133] Exemplarily, assuming that the number of batches B is 3, the total arc time T is 5, and the observed process parameters of each batch b are used as row vectors, the observed process parameters may include multiple parameters in the melting process such as temperature, pressure, current, etc.

[0134] For batch 1, the temperature, pressure and current parameter values ​​from the first point to the last time point can be obtained. Based on the same method, the temperature, pressure and current parameter values ​​from the first point to the fifth time point can be obtained for batches 2, 3, 4 and 5. Aligning according to the arc ignition time T of the melting process can mean arranging the observed process parameters of different batches according to the time sequence after the arc ignition starts, putting the parameters of each batch at the same time point in the same row, and constructing a multi-process observation feature matrix with a shape of T×B, as shown below:

[0135] .

[0136] in, represents the observed process parameter value of the jth batch at the i-th time point. represents the observed process parameter value of the jth batch at the i-th time point. For example It can refer to the observed process parameter value of the third batch at the second time point. It should be noted that The parameter value set of all candidate parameters, such as the temperature parameter value, the pressure parameter value, and the current parameter value, may be included in the jth batch at the i-th time point.

[0137] In this embodiment, the method involves organizing the observed process parameters according to the arc ignition time T of the melting process, with the batch number B as the column vector and the observed process parameters of each batch b as the row vector, thereby constructing a multi-process observation feature matrix with a shape of T×B. In the multi-process observation feature matrix, T represents the total duration of the arc ignition time, B represents the number of batches, each row represents the parameter value of all batches at a time point, and each column represents the parameter change of a specific batch at different time points. This method can provide a comprehensive data view to capture and analyze the changes in parameters of different batches during the melting process. Through systematic data processing, the key process parameters and key moments that affect the quality of quartz crucibles can be more accurately identified.

[0138] In one embodiment, the above-extracted process parameters with high correlation with the quality inspection index (the number of parameters is set to Q ) as input, build a quality prediction model, and output the quality prediction value. The quality prediction model adopts the ensemble learning model XGBoost (extreme gradient boosting), which is based on the ensemble learning method of decision trees. By iteratively optimizing the loss function, it gradually generates a set of decision trees to improve the prediction accuracy. The calculation formula is ,in Indicates l The predicted value of a quality inspection indicator, Indicates k The prediction function of a built-in decision tree model, K The number of numbers used to input melting process parameters x l Mapped to the corresponding output value. During the training process, the quality prediction model minimizes the weighted loss function To optimize the model parameters, the weighted loss function used in the process of training the quality prediction model is It consists of two parts: the first part is the loss of prediction error, and the second part is the regularization term of model complexity. The specific meaning of each parameter is as follows: The total loss function, representing the model parameters A function that evaluates the difference between the model's predictions and the actual values ​​and includes a regularization term to prevent overfitting. is the first part of the loss function, which means that for each batch in all batches S i The sum of the prediction errors of . It is i The first batch l The actual observed value of the quality indicator. The model is i The first batch l The predicted value of the quality index. L is the loss function (usually the mean square error ), is the second part of the loss function, which represents the sum of the regularization terms of the model complexity. K is the number of base learners (e.g., decision trees) in the model. are the model parameters of the kth base learner. It is a regularization term (usually L2 regularization) used to penalize the complexity of the model, which helps control the generalization ability of the model and avoid overfitting.

[0139] Furthermore, the model can be trained with offline historical data, and iteratively optimized until the model error is minimized, and the model is saved. In actual online production, according to the predicted value of the model, the fuzzy control method is used to give process optimization suggestions based on historical results, or the control parameters are adjusted in time according to the real-time parameters and expected results to improve the yield rate.

[0140] In one embodiment, the prediction results of the quality prediction model are compared with the finished product inspection specifications to obtain a fuzzy set (low, high, and in compliance). Through the correlation analysis results, fuzzy rules are set to describe the relationship between input (quality prediction value) and output (control action). For example, if the outer diameter of the crucible is predicted to be higher than the inspection specification, the set value of the mold outlet water temperature is increased. In the production process, it is necessary to continuously optimize the fuzzy sets and fuzzy rules in the fuzzy control system according to the actual control effect and the changes in the quality of the finished product, so as to improve the quality of the finished product.

[0141] Based on the same inventive concept, the embodiment of the present application also provides a quartz crucible melting equipment parameter adjustment device for implementing the above-mentioned quartz crucible melting equipment parameter adjustment method. The implementation scheme for solving the problem provided by the device is similar to the implementation scheme recorded in the above-mentioned method, so the specific limitations of the embodiments of one or more quartz crucible melting equipment parameter adjustment devices provided below can refer to the limitations of the quartz crucible melting equipment parameter adjustment method above, and will not be repeated here.

[0142] In one embodiment, Figure 7 As shown, the embodiment of the present application also provides a device for adjusting parameters of a melting device of a quartz crucible, the device comprising:

[0143] The acquisition module 701 is used to acquire the observed process parameters in the multi-batch melting process of the quartz crucible, and construct a multi-process observation feature matrix in the shape of T×B according to the observed process parameters; wherein the observed process parameters include multiple candidate parameters of the melting equipment in the melting process, B is the batch number of the melting process, and T is the arc ignition time of each melting process;

[0144] A screening module 702 is used to determine the probability density distribution of the multi-process observation feature matrix by using a density estimation method, and screen S valid batches from B batches according to the probability density distribution of the multi-process observation feature matrix;

[0145] The calculation module 703 is used to determine the first characteristic matrix based on the observed process parameters of the S valid batches, determine the quality matrix based on the quality parameters of the quartz crucibles produced by the S valid batches, and determine the first correlation coefficient matrix based on the first characteristic matrix and the quality matrix; the calculation module is also used to determine the target parameters of the melting equipment from the candidate parameters based on the coefficient values ​​in the first correlation coefficient matrix, and determine the target time based on the observed process parameters of the S valid batches and the quality parameters of the quartz crucibles produced by the S valid batches; wherein the target time belongs to the arc ignition time t of the arc ignition time T, and the target parameter is a candidate parameter whose correlation with the quality parameter of the quartz crucible is greater than a preset value;

[0146] The adjustment module 704 is used to input the target parameters at the target time into the pre-trained quality prediction model, obtain the quality prediction result of the quartz crucible to be melted, and adjust the parameters of the melting equipment based on the quality prediction result.

[0147] In one embodiment, the screening module 702 uses a density estimation method to determine the probability density distribution of the multi-process observation feature matrix, which is specifically used to: slide a time window of a set length along the direction of increasing arc ignition time T to obtain multiple time windows; for each time window, perform kernel density calculation on each observation sample point in the time window to obtain the probability density of each observation sample point in the time window, and determine the density matrix of the time window based on the probability density of each observation sample point in the time window; wherein the observation sample point is any candidate parameter of any batch at any arc ignition time t. Connect the density matrices of all time windows end to end to obtain the probability density distribution of the multi-process observation feature matrix.

[0148] In one embodiment, the arc-starting time T of the screening module 702 includes multiple arc-starting moments t; according to the probability density distribution of the multi-process observation feature matrix, S valid batches are screened out from B batches, specifically for: based on the probability density distribution of the multi-process observation feature matrix, along the direction of increasing arc-starting time T, for each arc-starting moment t, the target position where the probability density maximum value is located is located; the target positions of multiple arc-starting moments t are connected into a line to obtain the center line of the consistency test; based on the center line, the screening limit is determined, the batches beyond the screening limit are determined as abnormal batches, and the abnormal batches are removed from the B batches to obtain S valid batches.

[0149] In one embodiment, the screening module 702 determines the screening limit based on the center line, specifically for:

[0150] Determine the standard deviation of the center line, and use the preset multiple standard deviations above and below the center line as boundaries to determine the upper and lower screening limits; determine the screening limits based on the upper and lower screening limits.

[0151] In one embodiment, the calculation module 703 calculates the global average x̅_(s,p) for each candidate parameter p in each batch s to obtain an S×P feature matrix, where s belongs to S, p belongs to P, P is the number of candidate parameters, and the S×P feature matrix is ​​the first feature matrix; the quality matrix is ​​an S×L feature matrix, and L is the number of quality indicators; based on the first feature matrix and the quality matrix, a first correlation coefficient matrix is ​​determined, which is specifically used to: calculate the Pearson correlation coefficient for the p-th candidate parameter mean x̅_(s,p) and the l-th quality inspection indicator y_(s,l); merge the correlation coefficient calculation results of the P candidate parameters and the L quality inspection indicators to obtain a first correlation coefficient matrix with a shape of P×L.

[0152] In one embodiment, the calculation module 703 calculates the mean of the pth candidate parameter and the first quality inspection index , the calculation formula for Pearson correlation coefficient is ,in is the mean of the pth candidate parameter mean sequence, is the mean of the lth quality inspection index sequence.

[0153] In one embodiment, the calculation module 703 determines the target time according to the observed process parameters of S valid batches and the quality parameters of the quartz crucibles produced by the S valid batches, and is specifically used for: for each batch s, calculating the correlation coefficient based on the candidate parameter p and the quality parameter of each arc ignition time t in the batch s, and obtaining a correlation coefficient sequence of P bars with a length of T; for each candidate parameter p, determining the target time through the correlation coefficient sequence.

[0154] In one embodiment, the acquisition module 701 constructs a multi-process observation feature matrix with a shape of T×B based on the observed process parameters, and is specifically used to: use the batch number B as the column vector, the observed process parameters of each batch b as the row vector, and align them according to the arc ignition time T of the melting process to construct a multi-process observation feature matrix with a shape of T×B.

[0155] Each module in the above-mentioned device for adjusting parameters of melting equipment of quartz crucible can be implemented in whole or in part by software, hardware and their combination. Each module can be embedded in or independent of the processor in the computer device in the form of hardware, or can be stored in the memory in the computer device in the form of software, so that the processor can call and execute the operations corresponding to each module above.

[0156] Based on the same concept, the present application also provides an electronic device, including a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, a method for adjusting parameters of a melting device for a quartz crucible is provided.

[0157] In one embodiment, an electronic device is provided, including a memory and a processor. The electronic device may be a terminal, and its internal structure diagram may be as follows: Figure 8As shown. The electronic device includes a processor, a memory, a communication interface, a display screen and an input device connected through a system bus. Among them, the processor of the electronic device is used to provide computing and control capabilities. The memory of the electronic device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The communication interface of the electronic device is used to communicate with an external terminal in a wired or wireless manner, and the wireless manner can be achieved through WIFI, a mobile cellular network, NFC (near field communication) or other technologies. When the computer program is executed by the processor, a method for adjusting the parameters of a melting device for a quartz crucible is implemented. The display screen of the electronic device can be a liquid crystal display screen or an electronic ink display screen, and the input device of the electronic device can be a touch layer covered on the display screen, or a key, trackball or touchpad set on the housing of the electronic device, or an external keyboard, touchpad or mouse, etc.

[0158] Those skilled in the art will understand that Figure 8 The structure shown in the figure is merely a block diagram of a partial structure related to the scheme of the present application, and does not constitute a limitation on the electronic device to which the scheme of the present application is applied. The specific electronic device may include more or fewer components than shown in the figure, or combine certain components, or have a different arrangement of components.

[0159] The technical features of the above embodiments may be arbitrarily combined. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0160] The above embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the present application. It should be pointed out that, for a person of ordinary skill in the art, several modifications and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the present application shall be subject to the attached claims.

Claims

1. A method for adjusting parameters of a melting device of a quartz crucible, characterized in that: The method comprises: Obtaining observed process parameters during the multi-batch melting process of the quartz crucible, and constructing a multi-process observed feature matrix of a shape of T×B according to the observed process parameters; wherein the observed process parameters include multiple candidate parameters of the melting equipment during the melting process, B is the batch number of the melting process, and T is the arc ignition time of each melting process; Using a density estimation method, determine the probability density distribution of the multi-process observation feature matrix, and screen S valid batches from B batches according to the probability density distribution of the multi-process observation feature matrix; Determine a first characteristic matrix according to observed process parameters of S valid batches, determine a quality matrix according to quality parameters of quartz crucibles produced by the S valid batches, and determine a first correlation coefficient matrix based on the first characteristic matrix and the quality matrix; Based on the coefficient values ​​in the first correlation coefficient matrix, the target parameter of the melting equipment is determined from the candidate parameters, and the target time is determined according to the observed process parameters of the S valid batches and the quality parameters of the quartz crucibles produced by the S valid batches; wherein the target time belongs to the partial arc ignition time t in the arc ignition time T, and the target parameter is a candidate parameter whose correlation with the quality parameter of the quartz crucible is greater than a preset value; The target parameters at the target time are input into a pre-trained quality prediction model to obtain a quality prediction result of the quartz crucible to be melted, and the parameters of the melting equipment are adjusted based on the quality prediction result.

2. The method for adjusting the melting equipment parameters of a quartz crucible according to claim 1, characterized in that: The probability density distribution of the multi-process observation feature matrix is ​​determined by using a density estimation method, including: Sliding a time window of a set length along the direction in which the arc ignition time T increases to obtain a plurality of time windows; For each time window, a kernel density calculation is performed on each observation sample point in the time window to obtain a probability density of each observation sample point in the time window, and a density matrix of the time window is determined based on the probability density of each observation sample point in the time window; wherein the observation sample point is any candidate parameter of any batch at any arc ignition time t; The density matrices of all time windows are connected end to end to obtain the probability density distribution of the multi-process observation feature matrix.

3. The method for adjusting parameters of a melting device of a quartz crucible according to claim 1 or 2, characterized in that: The arc ignition time T includes multiple arc ignition moments t; according to the probability density distribution of the multi-process observation feature matrix, S valid batches are screened out from B batches, including: Based on the probability density distribution of the multi-process observation characteristic matrix, along the direction in which the arc ignition time T increases, for each arc ignition time t, locate the target position where the probability density maximum value is located; Connect the target positions of multiple arc ignition times t into a line to obtain the center line of the consistency test; A screening limit is determined based on the center line, batches beyond the screening limit are determined as abnormal batches, and S valid batches are obtained by removing the abnormal batches from the B batches.

4. The method for adjusting the melting equipment parameters of a quartz crucible according to claim 3, characterized in that: Determine screening limits based on the centerline, including: Determine the standard deviation of the center line, and determine the upper and lower screening limits using the preset multiple standard deviations above and below the center line as boundaries; Based on the upper screening limit and the lower screening limit, a screening boundary is determined.

5. The method for adjusting parameters of a melting device of a quartz crucible according to claim 1, characterized in that: Calculate the global average for each candidate parameter p in each batch s , obtain an S×P feature matrix, where s belongs to S, p belongs to P, P is the number of candidate parameters, and the S×P feature matrix is ​​the first feature matrix; the quality matrix is ​​an S×L feature matrix, and L is the number of quality indicators; Determining a first correlation coefficient matrix based on the first feature matrix and the quality matrix includes: For the pth candidate parameter mean and the first quality inspection index , calculate the Pearson correlation coefficient; The correlation coefficient calculation results of the P candidate parameters and the L quality inspection indicators are combined to obtain a first correlation coefficient matrix with a shape of P×L.

6. The method for adjusting parameters of a melting device of a quartz crucible according to claim 4, characterized in that: For the pth candidate parameter mean and the first quality inspection index , the calculation formula for Pearson correlation coefficient is ,in is the mean of the pth candidate parameter mean sequence, is the mean of the lth quality inspection index sequence.

7. The method for adjusting parameters of a melting device of a quartz crucible according to claim 1, characterized in that: According to the observed process parameters of S valid batches and the quality parameters of the quartz crucibles produced by S valid batches, the target time is determined, including: For each batch s, the correlation coefficient is calculated based on the candidate parameter p at each arc ignition time t in the batch s and the quality parameter to obtain a correlation coefficient sequence of P pieces with a length of T; For each candidate parameter p, the target time is determined through the correlation coefficient sequence.

8. The method for adjusting parameters of a melting device of a quartz crucible according to claim 1, characterized in that: According to the observation process parameters, a multi-process observation feature matrix with a shape of T×B is constructed, including: With the batch number B as the column vector and the observed process parameters of each batch b as the row vector, they are aligned according to the arc ignition time T of the melting process to construct a multi-process observation feature matrix with a shape of T×B.

9. A device for adjusting parameters of a melting device for a quartz crucible, characterized in that: The device comprises: An acquisition module is used to acquire observed process parameters during the multi-batch melting process of a quartz crucible, and to construct a multi-process observation feature matrix of a shape of T×B according to the observed process parameters; wherein the observed process parameters include multiple candidate parameters of a melting device during the melting process, B is the batch number of the melting process, and T is the arc ignition time of each melting process; A screening module, used to determine the probability density distribution of the multi-process observation feature matrix by using a density estimation method, and screen S valid batches from B batches according to the probability density distribution of the multi-process observation feature matrix; A calculation module, used to determine a first characteristic matrix based on the observed process parameters of S valid batches, determine a quality matrix based on the quality parameters of the quartz crucibles produced by the S valid batches, and determine a first correlation coefficient matrix based on the first characteristic matrix and the quality matrix; the calculation module is also used to determine the target parameters of the melting equipment from the candidate parameters based on the respective coefficient values ​​in the first correlation coefficient matrix, and determine the target time based on the observed process parameters of the S valid batches and the quality parameters of the quartz crucibles produced by the S valid batches; wherein the target time belongs to a part of the arc ignition time t in the arc ignition time T, and the target parameter is a candidate parameter having a correlation with the quality parameter of the quartz crucible greater than a preset value; The adjustment module is used to input the target parameters at the target time into the pre-trained quality prediction model to obtain the quality prediction result of the quartz crucible to be melted, and adjust the parameters of the melting equipment based on the quality prediction result.

10. An electronic device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method for adjusting parameters of a melting device for a quartz crucible according to any one of claims 1 to 8 are implemented.

Images