Method and device for detecting particle size of green silicon carbide micro powder

Abstract

The invention relates to the technical field of particle size detection, and discloses a green silicon carbide micro-powder particle size detection method and device.The method comprises the steps that particle size distribution data is collected, and a two-dimensional detection data set is obtained; a particle size distribution data set is obtained after preprocessing; performing dimension reduction analysis on the particle size distribution data set to form a particle size distribution feature vector; obtaining a matching parameter time sequence of the silicon content and the carbon content; calculating a correlation coefficient between the particle size distribution feature vector and a matching parameter, and establishing a dynamic particle size prediction model; inputting real-time silicon and carbon content data into the dynamic granularity prediction model to obtain granularity distribution characteristics; when the main particle size interval deviates from a preset target particle size range, adjusting the silicon and carbon contents to obtain an optimized proportioning parameter time sequence; and updating the model based on the optimized ratio parameter time sequence to obtain an adjusted granularity prediction model, and performing repeated prediction to obtain optimized granularity distribution characteristics. According to the method, accurate prediction and dynamic adjustment of particle size distribution can be realized, and the product particle size stability is improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of micro-powder particle size detection, and in particular to a green silicon carbide micro-powder particle size detection method and device. BACKGROUND

[0002] Green silicon carbide micro-powder, as a key component of high-end abrasives and functional materials, plays an irreplaceable role in precision machining, electronic device manufacturing, and new energy fields. Its particle size distribution directly affects product quality and application performance. The uniformity of particle size distribution and the precise control of the main particle size range are the core prerequisites for ensuring that the micro-powder meets the design requirements in terms of polishing accuracy and device performance.

[0003] In the prior art, green silicon carbide micro-powder particle size detection relies on single devices such as laser particle size analyzers and sedimentation instruments. Particle size distribution data is obtained through single or batch detection. In terms of production control, process parameters such as raw material ratio and grinding time are adjusted empirically, and particle size distribution is optimized based on offline detection results. In this mode, the correlation analysis of detection data and process parameters only stays on the surface, neither systematically aggregates multiple batches of data to uncover the rules nor quantifies the internal relationship between raw material ratio and particle size output through mathematical modeling, making it difficult to achieve active prediction and dynamic optimization of particle size characteristics.

[0004] In summary, the prior art lacks dynamic correlation between raw material ratio and final product particle size output, and there is a problem of insufficient stability of particle size distribution characteristics. SUMMARY

[0005] The present application provides a green silicon carbide micro-powder particle size detection method and device to improve the stability of particle size distribution and production efficiency, and reduce quality deviations caused by equipment jitter or ratio fluctuations.

[0006] In a first aspect, to solve the above technical problems, the present application provides a green silicon carbide micro-powder particle size detection method, comprising: Collecting and analyzing multiple batches of particle size distribution data to obtain a two-dimensional detection data set containing particle size range and distribution ratio; Preprocessing the two-dimensional detection data set to obtain a final particle size distribution data set; Performing dimensionality reduction analysis on the particle size distribution data set to determine the main particle size range and the corresponding particle volume proportion, and linearly transforming to form a particle size distribution feature vector; Obtaining raw material ratio change data containing silicon content and carbon content ratio from production records, and aggregating the raw material ratio change data in time sequence to obtain a ratio parameter time series; Correlation coefficients between the particle size distribution feature vector and the proportioning parameter time series are calculated, and when the correlation coefficients are greater than a preset correlation threshold, a nonlinear mapping relationship is established to obtain a dynamic particle size prediction model; Real-time silicon content and carbon content proportion data are obtained and input into the dynamic particle size prediction model to obtain a predicted particle size distribution feature; When a main particle size interval in the particle size distribution feature deviates from a preset target particle size range, the silicon content and carbon content proportion are adjusted to obtain an optimized proportioning parameter time series; The dynamic particle size prediction model is updated based on the optimized proportioning parameter time series to obtain an adjusted particle size prediction model, and the main particle size interval and distribution proportion are repeatedly predicted based on the adjusted particle size prediction model to obtain an optimized particle size distribution feature.

[0007] In an optional implementation, the pre-processing of the two-dimensional detection data set to obtain a final particle size distribution data set includes: The deviation absolute value of the volume proportion value of each particle size interval in the two-dimensional detection data set from the mean value of the volume proportion in the data set is calculated, and when the deviation absolute value is greater than a preset deviation threshold, the abnormal data points are marked and removed to obtain a first data set; The relative position of each data point in the first data set relative to the minimum value and the maximum value of the data set is calculated, and the numerical value is mapped to the 0-1 interval to obtain a normalized second data set; According to the second data set, the average value of the data points in a preset sliding window is calculated and the center point data value is replaced, and after a preset number of iterations, the final particle size distribution data set is obtained.

[0008] In an optional implementation, the dimensionality reduction analysis of the particle size distribution data set is performed to determine the main particle size interval and the corresponding particle volume proportion, and linear transformation is performed to form a particle size distribution feature vector, including: The original particle size range and the original distribution proportion are obtained from the particle size distribution data set, and the particle size average value is calculated; According to the principal component analysis method, the multi-dimensional features composed of the original particle size range, the original distribution proportion and the particle size average value are mapped to a low-dimensional space, and the characteristic values are calculated; The particle size intervals with characteristic values greater than a preset characteristic value threshold are screened out and determined as the main particle size interval, and the particle volume proportion corresponding to the main particle size interval is extracted; The main particle size interval and the particle volume proportion are linearly transformed and combined to form the particle size distribution feature vector.

[0009] In an optional implementation, the correlation coefficient of the particle size distribution feature vector and the proportioning parameter time sequence is calculated, when the correlation coefficient is greater than a preset correlation threshold, a nonlinear mapping relationship is established, and a dynamic particle size prediction model is obtained, including: According to the proportioning parameter time sequence, silicon content proportion data and carbon content proportion data are obtained; The correlation coefficient of the silicon content proportion data and the carbon content proportion data and the particle size distribution feature vector is calculated, and a silicon content correlation coefficient value and a carbon content correlation coefficient value are obtained; When the silicon content correlation coefficient value or the carbon content correlation coefficient value exceeds a preset correlation threshold, a nonlinear mapping relationship between silicon content and carbon content proportion and particle size distribution is fitted using a support vector regression algorithm, and main particle size interval prediction values and distribution proportion prediction values are obtained; The main particle size interval prediction values and the distribution proportion prediction values are weighted and fused with the particle size distribution feature vector to obtain the dynamic particle size prediction model.

[0010] In an optional implementation, the real-time silicon content and carbon content proportion data are obtained and input into the dynamic particle size prediction model to obtain a predicted particle size distribution feature, including: The silicon content and carbon content proportion data are collected in real time, and abnormal values exceeding a preset reasonable range are removed. Taking a time point corresponding to a missing value as the center, an average value of valid data in a preset time period is taken to fill in the vacancy, and a clean data set is obtained; The time sequence characteristics of the clean data set are extracted using principal component analysis to obtain an element proportion feature set containing silicon content and carbon content proportion change trends; The variance contribution rate of the element proportion feature set is calculated, when the variance contribution rate exceeds a preset variance contribution threshold, the element proportion feature set is input into the dynamic particle size prediction model, and the predicted particle size distribution feature is obtained.

[0011] In an optional implementation, when the main particle size interval in the particle size distribution feature deviates from a preset target particle size range, the silicon content and carbon content proportion are adjusted to obtain an optimized proportioning parameter time sequence, including: The main particle size interval is extracted from the particle size distribution feature, and the deviation degree of the main particle size interval from a preset target particle size range is calculated to obtain a particle size deviation degree; When the particle size deviation degree is greater than a preset deviation threshold, the silicon content and carbon content proportion are linearly regressed and adjusted to generate an optimized proportioning parameter; The optimized proportioning parameter is aggregated in time sequence to obtain the optimized proportioning parameter time sequence.

[0012] In an optional implementation, characterized in that, the updating the dynamic particle size prediction model based on the optimized proportioning parameter time sequence comprises: extracting mean value and variance features of the optimized proportioning parameter based on the optimized proportioning parameter time sequence; applying the optimized proportioning parameter to actual production, detecting the produced green silicon carbide micro powder to obtain an optimized particle size distribution feature; when the mean value and variance features meet preset mean value and variance thresholds respectively, performing linear regression fitting on the optimized proportioning parameter and the optimized particle size distribution feature, updating the dynamic particle size prediction model to obtain the adjusted particle size prediction model.

[0013] In a second aspect, the application provides a green silicon carbide micro powder particle size detection device, comprising: a data acquisition module that acquires and analyzes multiple batches of particle size distribution data to obtain a two-dimensional detection data set containing particle size range and distribution proportion; a data preprocessing module that preprocesses the two-dimensional detection data set to obtain a final particle size distribution data set; a feature extraction module that performs dimensionality reduction analysis on the particle size distribution data set to determine the main particle size interval and the corresponding particle volume proportion, and linearly transforms to form a particle size distribution feature vector; a proportioning aggregation module that obtains raw material proportioning change data containing silicon content and carbon content proportion from production records, and aggregates the raw material proportioning change data in time sequence to obtain a proportioning parameter time sequence; a model construction module that calculates the correlation coefficient of the particle size distribution feature vector and the proportioning parameter time sequence, establishes a nonlinear mapping relationship when the correlation coefficient is greater than a preset correlation threshold, and obtains a dynamic particle size prediction model; a real-time prediction module that obtains real-time silicon content and carbon content proportion data, inputs the data into the dynamic particle size prediction model, and obtains a predicted particle size distribution feature; a proportioning optimization module that adjusts the silicon content and carbon content proportion when the main particle size interval in the particle size distribution feature deviates from a preset target particle size range, and obtains an optimized proportioning parameter time sequence; a model updating module that updates the dynamic particle size prediction model based on the optimized proportioning parameter time sequence to obtain an adjusted particle size prediction model, and repeatedly predicts the main particle size interval and distribution proportion based on the adjusted particle size prediction model to obtain an optimized particle size distribution feature.

[0014] Compared with the prior art, the application has the following beneficial effects: (1) The present application obtains a two-dimensional detection data set by collecting and analyzing multiple batches of particle size distribution data, and pre-processes the two-dimensional detection data set to obtain a final particle size distribution data set. This process can integrate multiple batches of data and eliminate noise interference, comprehensively capture the overall characteristics of the particle size distribution, and more accurately reflect the true particle size state of the green silicon carbide micro powder, providing a high-quality data basis for subsequent analysis.

[0015] (2) The present application determines the main particle size interval and the corresponding particle volume ratio by performing dimensionality reduction analysis on the particle size distribution data set, and forms a particle size distribution feature vector after linear transformation. This operation can strip redundant information, focus on the core particle size characteristics that affect the performance of the micro powder, and convert complex data into a structured feature vector, which facilitates efficient mining of particle size distribution rules, thereby more directly associating raw material ratio with key particle characteristics and accurately controlling the direction.

[0016] (3) The present application calculates the correlation coefficient of the particle size distribution feature vector and the ratio parameter time series, and establishes a nonlinear mapping relationship to obtain a dynamic particle size prediction model when the coefficient exceeds a preset correlation threshold. This mechanism can quantify the internal relationship between the silicon and carbon content of the raw material ratio and the particle size distribution, break the limitations of empirical control, achieve active prediction of particle size characteristics based on raw material parameters, predict the particle size trend in advance, and then adjust the process parameters in a timely and scientific manner.

[0017] (4) The present application obtains real-time silicon-carbon ratio data, inputs the dynamic particle size prediction model to obtain a prediction result, adjusts the ratio when it deviates from the target particle size range, and updates the model to repeat optimization. This closed-loop process can dynamically optimize the ratio parameters based on real-time production data, continuously adapt to process fluctuations through model iteration, ensure that the particle size distribution is always stable within the target range, and significantly improve the quality stability and production efficiency of green silicon carbide micro powder products. BRIEF DESCRIPTION OF DRAWINGS

[0018] Figure 1 is a flowchart of an embodiment of the green silicon carbide micro powder particle size detection method provided by the present application; Figure 2 is a structural schematic diagram of an embodiment of the green silicon carbide micro powder particle size detection device provided by the present application. DETAILED DESCRIPTION

[0019] The technical solutions in the embodiments of the present application will be described in detail below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative labor are within the scope of protection of the present application.

[0020] ReferenceFigure 1 The first embodiment of the present application provides a green silicon carbide micro-powder particle size detection method, comprising steps S11 to S18: S11, collecting and analyzing multi-batch particle size distribution data to obtain a two-dimensional detection data set containing particle size range and distribution ratio; S12, preprocessing the two-dimensional detection data set to obtain a final particle size distribution data set; S13, performing dimensionality reduction analysis on the particle size distribution data set to determine the main particle size interval and the corresponding particle volume proportion, and linearly transforming to form a particle size distribution feature vector; S14, obtaining raw material ratio change data containing silicon content and carbon content ratio from production records, and aggregating the raw material ratio change data in chronological order to obtain a ratio parameter time series; S15, calculating the correlation coefficient of the particle size distribution feature vector and the ratio parameter time series, and when the correlation coefficient is greater than a preset correlation threshold, establishing a nonlinear mapping relationship to obtain a dynamic particle size prediction model; S16, obtaining real-time silicon content and carbon content ratio data and inputting it into the dynamic particle size prediction model to obtain a predicted particle size distribution feature; S17, when the main particle size interval in the particle size distribution feature deviates from a preset target particle size range, adjusting the silicon content and carbon content ratio to obtain an optimized ratio parameter time series; S18, updating the dynamic particle size prediction model based on the optimized ratio parameter time series to obtain an adjusted particle size prediction model, and repeatedly predicting the main particle size interval and distribution ratio based on the adjusted particle size prediction model to obtain an optimized particle size distribution feature.

[0021] In step S11, multi-batch particle size distribution data is collected and analyzed to obtain a two-dimensional detection data set containing particle size range and distribution ratio.

[0022] It should be noted that the data collection work depends on a laser particle size analyzer, which works on the principle of using the scattering characteristics of particles to the laser, by measuring the scattering angle and scattering intensity of the laser irradiated to the green silicon carbide micro-powder particles, to infer the size distribution of the particles. The reason for selecting multiple batches is that single batch detection data may be affected by accidental factors (such as uneven sampling of samples, transient fluctuations of equipment, etc.), and multi-batch data can more comprehensively reflect the overall characteristics and fluctuation range of particle size distribution in the production process, thereby improving the reliability of subsequent analysis. The particle size range is expressed in terms of particle size interval, such as 0.5-1 microns, 1-5 microns, etc., which reflects the distribution span of the particle size of the micro-powder. The distribution ratio is measured by the volume proportion of the particles, indicating the percentage of the particle volume in a certain particle size range in the total volume.

[0023] In order to ensure the consistency and comparability of the data, the particle size range of the collected original data is divided into several fixed intervals according to a unified particle size division standard, for example, 0.1-100 microns is divided into 0.1-1 micron, 1-10 microns, and 10-100 microns three intervals, and then the volume proportion of each interval in each batch is calculated. Through this process, the original scattered data is integrated into a structured two-dimensional detection data set, and each dimension corresponds to a particle size range and a distribution ratio in the range.

[0024] In step S12, the two-dimensional detection data set is preprocessed to obtain a final particle size distribution data set, including: Calculate the absolute value of the deviation of the volume proportion of each particle size interval in the two-dimensional detection data set from the average value of the volume proportion in the data set. When the absolute value of the deviation is greater than a preset deviation threshold, mark it as an abnormal data point, and remove the abnormal data point to obtain a first data set; Calculate the relative position of each data point in the first data set relative to the minimum and maximum values of the data set, and map the values to the 0-1 interval to obtain a normalized second data set; According to the second data set, calculate the average value of the data points in a preset sliding window and replace the center point data value. After a preset number of iterations, the final particle size distribution data set is obtained.

[0025] It should be noted that when calculating the absolute value of the deviation, the volume proportion of all data points is first calculated to obtain the overall average, and then the absolute value of the difference between the volume proportion of each data point and the average is calculated. The preset deviation threshold is set based on production experience or statistical rules. The first data set filters out extreme values caused by accidental factors such as equipment instantaneous failure and sample contamination, and is closer to the true characteristics of the particle size distribution of the micro powder.

[0026] Next, the volume proportion of each data point is converted to a proportion relative to the minimum and maximum values of the data point volume proportion in the data set by the minimum-maximum normalization method. The calculation formula of this method is: (data point volume proportion-minimum value) / (maximum value-minimum value). The normalized value obtained is uniformly mapped to the 0-1 interval to form a second data set. For example, if the minimum value of the volume proportion in the first data set is 5% and the maximum value is 20%, the volume proportion of a certain particle size interval is 15%, then its normalized value is (15%-5%) / (20%-5%) = 0.667. This not only facilitates direct comparison of data from different batches, but also provides standardized input for subsequent model training, avoiding model bias caused by differences in numerical scales.

[0027] Further, the size of the preset sliding window can be set according to the data density (such as containing 3 or 5 consecutive particle size interval data points), and the principle is to weaken the influence of instantaneous fluctuations by local averaging. For example, if the normalized value sequence of the second data set is 0.2, 0.5, 0.8, 0.6 (corresponding to four consecutive particle size intervals), and the size of the sliding window is set to 3, the first window contains 0.2, 0.5, 0.8, and the average value is 0.5, which replaces the center point 0.5; the second window contains 0.5, 0.8, 0.6, and the average value is 0.63, which replaces the center point 0.8, obtaining the preliminary smoothed sequence: 0.2, 0.5, 0.633, 0.6. Performing a preset number of iterations will further enhance the smoothing effect, and the final particle size distribution data set is obtained. The setting of the preset number of iterations needs to be combined with the data noise intensity, the continuity requirement of the particle size distribution, and the calculation efficiency, and is usually 2-4 times. Through multiple iterations, the sudden rise or sudden drop mutation points caused by detection errors in the data set will be gradually weakened, highlighting the overall trend of the particle size distribution.

[0028] In step S13, the particle size distribution data set is subjected to dimensionality reduction analysis to determine the main particle size interval and the corresponding particle volume proportion, and linear transformation is performed to form a particle size distribution feature vector, including: The original particle size range and the original distribution ratio are obtained from the particle size distribution data set, and the particle size average value is calculated; Based on the principal component analysis method, the multi-dimensional features composed of the original particle size range, the original distribution ratio and the particle size average value are mapped to a low-dimensional space, and the characteristic values are calculated; The particle size interval with a characteristic value greater than a preset characteristic value threshold is selected as the main particle size interval, and the particle volume proportion corresponding to the main particle size interval is extracted; The main particle size interval and the particle volume proportion are subjected to linear transformation, and the particle size distribution feature vector is formed after combination.

[0029] It should be noted that the original particle size range refers to the retained consecutive particle size intervals after preprocessing, which is used to represent the distribution span of the particle size; the original distribution ratio refers to the volume proportion corresponding to each interval. The particle size average value is a single numerical value representing the particle size characteristic of a certain interval, which can be obtained by calculating the arithmetic mean of the upper and lower limits of the interval. In the particle size distribution data, the multi-dimensional features refer to the combination of the characteristic parameters of multiple particle size intervals, and there is a correlation between these features. For example, when the proportion of large particle size intervals is high, the proportion of small particle size intervals will be relatively low, which can easily lead to data redundancy. The principal component analysis method (PCA) is a statistical method for mapping high-dimensional particle size feature data to a low-dimensional space through linear transformation, which can reduce the data dimension while preserving the core information of the data (such as the key rules of the particle size distribution), and can eliminate redundant features. The main operation steps are as follows: Firstly, the original multi-dimensional features (such as the average value and volume proportion of each particle size interval) are standardized, that is, the value of each feature is subtracted by the average value of the feature and then divided by the standard deviation of the feature, so that features of different dimensions are unified to the same scale; then the covariance matrix of the standardized data is calculated to describe the strength of the correlation between features; then the characteristic equation of the covariance matrix is solved to obtain the corresponding eigenvalues and eigenvectors, wherein the eigenvector represents the direction of the principal component, and the eigenvalue represents the size of the original data variance explained by the principal component, and the larger the value is, the more original particle size data information the corresponding principal component contains; the eigenvalues are sorted from large to small, and at this time, the principal components with eigenvalues greater than a preset eigenvalue threshold are selected, since these principal components contain the main information of the original data, the particle size intervals with high weights in the corresponding eigenvectors play a leading role in the particle size distribution, and thus are determined as the main particle size intervals. The preset eigenvalue threshold is mainly determined according to the application scene requirement, production process stability and data information, and for the field of precision machining which requires high precision, more particle size details need to be retained, and the threshold needs to be set relatively high (such as 0.8-1).

[0030] After determining the main particle size intervals, the corresponding distribution proportions are extracted from the final particle size distribution data set to obtain the volume proportion of the particles. The volume proportion is normalized to the interval of 0-1 through linear transformation, the lower limit and the upper limit of the main particle size interval are structurally combined with the corresponding standardized volume proportion to form a particle size distribution feature vector, and the dimensionality reduction of the original high-dimensional data to a low-dimensional space is realized.

[0031] Exemplarily, the particle size distribution data set of a certain green silicon carbide powder contains 5 particle size intervals after preprocessing: 0.1-1 μm, 1-10 μm, 10-20 μm, 20-50 μm and 50-100 μm, and the corresponding original distribution proportions are 10%, 50%, 20%, 15% and 5% respectively. The average particle size of each interval is 0.55 μm, 5.5 μm, 15 μm, 35 μm and 75 μm respectively. Based on the principal component analysis method, the particle size average value and the distribution proportion of the particle size intervals are standardized, the covariance matrix is calculated, and 3 principal components are obtained by solving, the eigenvalues are 0.88, 0.09 and 0.03 respectively, and the preset eigenvalue threshold is 0.85, so the principal component corresponding to the eigenvalue 0.88 is retained, which is mainly related to the 1-10 μm interval. The feature vector of the principal component shows that the distribution proportion of the 1-10 μm interval is 50%, which is significantly higher than that of other intervals, so the 1-10 μm interval is determined as the main particle size interval. The volume proportion 50% is converted to 0.5, the lower limit 1 μm and the upper limit 10 μm are retained, and the particle size distribution feature vector [1, 10, 0.5] is formed, which clearly reflects the particle size characteristics that play a leading role in the performance of the powder.

[0032] In step S14, raw material ratio change data containing silicon content and carbon content ratio are obtained from production records, and the raw material ratio change data are data-aggregated in time sequence to obtain a ratio parameter time sequence.

[0033] It should be noted that the silicon content and the carbon content ratio refer to the percentage of the mass of silicon and carbon elements in the total mass of raw materials, which are the core raw material parameters for green silicon carbide powder production and can directly affect the particle size distribution of the generated powder. The raw material ratio change data are derived from real-time collection by production line sensors or manual records, and need to be preprocessed, such as checking whether the time stamp is complete, whether the content ratio is in a reasonable range, and whether the value is a non-negative floating point number. If there are records with missing time stamps, the time stamp of the last valid record can be filled in. If a negative value or a ratio value outside the pre-set reasonable range appears, it is determined as an abnormal value, and the average value of the normal data at the three adjacent time points before and after the abnormal value is taken as a substitute. After preprocessing, according to the time granularity requirement of the production process (such as every 4 hours, every day, etc.), a time interval is set, and the average values of the silicon content and the carbon content ratio in the interval are calculated as the representative values of the interval. In this way, discrete data are aggregated into a ratio parameter time sequence arranged in time order, each data point contains a time interval and the corresponding silicon and carbon content representative values, and the change trend of the raw material ratio with time can be clearly presented.

[0034] In step S15, a correlation coefficient between the particle size distribution feature vector and the ratio parameter time sequence is calculated, and when the correlation coefficient is greater than a pre-set correlation threshold, a non-linear mapping relationship is established to obtain a dynamic particle size prediction model, including: According to the ratio parameter time sequence, silicon content ratio data and carbon content ratio data are obtained; The correlation coefficients between the silicon content ratio data and the carbon content ratio data and the particle size distribution feature vector are calculated to obtain silicon content correlation coefficient values and carbon content correlation coefficient values; When the silicon content correlation coefficient value or the carbon content correlation coefficient value exceeds a pre-set correlation threshold, a support vector regression algorithm is used to fit the non-linear mapping relationship between the silicon content and the carbon content ratio and the particle size distribution to obtain main particle size interval prediction values and distribution ratio prediction values; The main particle size interval prediction values and the distribution ratio prediction values are weighted and fused with the particle size distribution feature vector to obtain the dynamic particle size prediction model.

[0035] It should be noted that the corresponding silicon content proportion representative value and carbon content proportion representative value are extracted from the time sequence data points of each proportioning parameter in chronological order, so as to obtain the silicon content proportion data sequence and the carbon content proportion data, respectively. The correlation coefficient is a statistical quantity for measuring the linear correlation degree between two variables, and the value range is [-1, 1]. The closer the absolute value is to 1, the stronger the correlation is. A positive value indicates a positive correlation, and a negative value indicates a negative correlation. For example, if the silicon content increases and the main particle size interval increases, it indicates a positive correlation; if the carbon content increases and the main particle size interval decreases, it indicates a negative correlation. The silicon and carbon content proportion data are one-dimensional data sequences, and the particle size distribution feature vector contains multi-dimensional data. The calculation of the correlation coefficient value needs to be performed in different dimensions. Taking the calculation of the silicon content correlation coefficient value as an example, the average value of the silicon content data and the average value of the volume proportion dimension are first calculated, and then the deviation of each data point from the corresponding average value is calculated to obtain two deviation sequences. Then, the sum of the product of the two deviation sequences is calculated as the numerator, and the square root of the sum of the squares of the two deviation sequences, i.e. the standard deviation, is calculated as the denominator. Finally, the numerator is divided by the denominator to obtain the silicon content correlation coefficient value.

[0036] In the present embodiment, the preset correlation threshold is adjusted according to the process precision requirement, and is generally set to 0.6-0.8. When the correlation coefficients of silicon or carbon do not exceed the threshold, it indicates that the correlation between the raw material proportioning and the particle size in the current data is weak, and more data needs to be supplemented. If the threshold is exceeded, it indicates that there is a significant correlation between the proportioning change of the element and the particle size distribution characteristics, and further modeling analysis is needed. The support vector regression (SVR) algorithm is a regression algorithm based on statistical learning theory, and the core is to find the optimal hyperplane to control the deviation of the training data points to the hyperplane within a preset range, so as to realize accurate prediction of continuous values. The silicon content proportion and the carbon content proportion are taken as input features, and the corresponding main particle size interval and distribution proportion are taken as output labels. After alignment by time, the training set and the validation set are divided, and standardized processing is performed to eliminate the dimension influence. Since the raw material proportioning and the particle size distribution have a nonlinear relationship, a radial basis kernel function is introduced to map the low-dimensional input to a high-dimensional space, so that the nonlinear relationship can be captured by a linear model. By optimizing the penalty coefficient for controlling the error tolerance and the kernel parameter for adjusting the mapping complexity, the model training is completed by minimizing the prediction error. When new silicon and carbon content proportions are input, the model calculates the prediction value of the main particle size interval and the prediction value of the distribution proportion through linear regression in the high-dimensional space.

[0037] Further, according to the importance of the characteristics to the particle size distribution, a higher weight (such as 0.6-0.8) is assigned to the prediction value of the main particle size interval and the prediction value of the distribution proportion, and a lower weight (such as 0.2-0.4) is assigned to the secondary characteristics in the original particle size distribution feature vector. The upper and lower limits of the interval and the distribution proportion after weighted fusion are combined to form a dynamic particle size prediction model.

[0038] For example, in the ratio parameter time series of a certain green silicon carbide production, the silicon content ratio data are 25%, 26%, 27%, and 28%, the carbon content ratio data are 15%, 14%, 13%, and 12%, and the corresponding particle size distribution feature vectors are [50, 100, 0.6], [55, 105, 0.58], [60, 110, 0.56], and [65, 115, 0.55]. The correlation coefficients of silicon content and carbon content with the particle size feature vector are 0.82 and 0.45, respectively, and the preset correlation coefficient threshold is 0.7, so the correlation of silicon content is significant. The support vector regression algorithm is used to fit the nonlinear relationship, and when the input silicon content is 26.5% and the carbon content is 13.5%, the main particle size interval prediction value is 62-112 μm and the distribution ratio prediction value is 0.57. Setting the prediction value weight to 0.7 and the closest original feature vector [60-110 μm, 0.56] weight to 0.3, the weighted fusion obtains a dynamic particle size prediction model, and the output is a main particle size interval of 61.4-111.4 μm and a distribution ratio of 0.567. This model can accurately predict the particle size distribution through real-time raw material ratio.

[0039] In step S16, real-time silicon content and carbon content ratio data are obtained and input into the dynamic particle size prediction model to obtain a predicted particle size distribution feature, including: The silicon content and carbon content ratio data are collected in real time, and outliers beyond the preset reasonable range are removed. The average value of valid data in a preset time period is taken as the center of the missing value to fill in the vacancy, and a clean data set is obtained; The time series characteristics of the clean data set are extracted by principal component analysis to obtain an element ratio feature set containing the change trend of silicon content and carbon content ratio; The variance contribution rate of the element ratio feature set is calculated, and when the variance contribution rate exceeds the preset variance contribution threshold, the element ratio feature set is input into the dynamic particle size prediction model to obtain the predicted particle size distribution feature.

[0040] It should be noted that the high-precision sensor deployed by the production line realizes real-time collection of silicon content and carbon content ratio data, and uses a timestamp as an identifier. Due to factors such as sensor fluctuations, equipment interference, etc., there are abnormal values or missing values not recorded in the original real-time data. First, abnormal values that exceed the pre-set reasonable range are removed, and for missing values, the average value of the valid data in the pre-set time period is calculated to fill in the missing values, for example, if the silicon content is missing at a certain time point, the valid data in the previous and next 30 minutes is 25%, 26%, and 24%, then the filling value is 25%, and finally an accurate and reliable clean data set is obtained. Since the clean data set contains multi-dimensional information such as element content change trend and fluctuation amplitude, directly inputting the prediction model will cause dimension redundancy, therefore, principal component analysis is used to map the multi-dimensional time series to a low-dimensional space through linear transformation, and the generated principal components are sorted in descending order according to variance contribution, each principal component is a linear combination of the original silicon and carbon content time series, and the two principal components with the largest explained variance are retained to form the element ratio feature set.

[0041] In the present embodiment, the variance contribution rate represents the proportion of the extracted principal components that cumulatively explain the variance of the original time series data, ranging from 0 to 100%. When the variance contribution rate does not exceed the pre-set variance contribution threshold, the number of principal components needs to be increased to ensure information integrity; if it exceeds the threshold, it indicates that the feature set has retained enough original data information, which can be input into the dynamic particle size prediction model. After inputting the model, the model outputs the predicted particle size distribution characteristics based on the established nonlinear mapping relationship, including the main particle size interval and the particle volume proportion, which provides a direct basis for real-time adjustment of raw material ratio and stable product particle size.

[0042] Illustratively, the silicon content data collected by a certain production line in real time is 25%, 26%, 35% (abnormal), 27% (missing), and 28%, and the carbon content data is 15%, 14%, 13%, 12%, 14%, and 13%. During processing, the abnormal value of 35% of the silicon content is removed, and the average value of 27.5% is filled in for the missing value by taking the average of the data before and after (27% and 28%), and the clean data set is obtained, with the silicon content ratio data being 25%, 26%, 27%, 27.5%, and 28%, and the carbon content ratio data being 15%, 14%, 13%, 14%, and 13%. The PCA is used to extract the time series features, and two principal components are obtained, the first principal component reflects the overall trend of the increase in silicon content and the decrease in carbon content, and the variance contribution rate is 72%; the second principal component reflects the short-term fluctuation characteristics, and the variance contribution rate is 18%, the cumulative variance contribution rate is 90%, which exceeds the pre-set threshold of 85%, forming the element ratio feature set. Inputting it into the dynamic particle size prediction model outputs the predicted particle size distribution characteristics as the main particle size interval 61-118 μm and the volume proportion 57%.

[0043] In step S17, when the main particle size interval in the particle size distribution characteristics deviates from the preset target particle size range, the silicon content and carbon content ratio is adjusted to obtain an optimized proportioning parameter time sequence, including: The main particle size interval is extracted from the particle size distribution characteristics, and the deviation degree of the main particle size interval from the preset target particle size range is calculated to obtain a particle size deviation degree; When the particle size deviation degree is greater than a preset deviation threshold, the silicon content and carbon content ratio is linearly regressed and adjusted to generate an optimized proportioning parameter; The optimized proportioning parameter is aggregated in time sequence to obtain the optimized proportioning parameter time sequence.

[0044] It should be noted that the preset target particle size range is an ideal interval set according to the product application scene, such as finer granularity required for precision grinding. When calculating the deviation degree of the main particle size interval from the preset target particle size range, the difference between the actual prediction interval and the target interval needs to be quantified, and the weighted sum of the interval center deviation and the range deviation is used to represent it: the interval center deviation is the difference between the prediction interval center value and the target interval center value, and the weight is 0.8; the range deviation is the difference between the prediction interval width and the target interval width, and the weight is 0.2. After weighting, the particle size deviation degree is obtained, and the greater the deviation degree, the more significant the deviation.

[0045] In this embodiment, based on historical production data, the linear corresponding relationship between the change of silicon content and the change of carbon content and the change of particle size is summarized. When the particle size deviation degree is greater than the preset deviation threshold, the linear corresponding relationship is used for reverse deduction to calculate the proportion data of the silicon content and the carbon content that needs to be increased or decreased, so that the particle size is regressed to the preset target particle size range. At the same time, the constraint range of silicon content and carbon content in the production process needs to be strictly followed, and the single adjustment amplitude needs to be controlled to avoid large process fluctuations, and finally the silicon and carbon content ratio that meets the constraints and can realize particle size correction is obtained, which is the optimized proportioning parameter. The optimized proportioning parameter is recorded according to the production time interval to form a sequence aligned with time, and each data point contains a time interval and a corresponding optimized silicon and carbon content ratio value, which is the optimized proportioning parameter time sequence. The time sequence can not only reflect the dynamic process of proportioning adjustment, but also provide continuous parameter execution basis for the production line, realizing closed-loop control from deviation detection to parameter optimization.

[0046] Exemplarily, the preset target particle size range is 50-100 pm, the center value is 75 pm, the main particle size interval predicted at a moment is 60-110 pm, the center value is 85 pm, the center deviation is calculated as 85 pm-75 pm=10 pm, the range deviation is 0 pm, the deviation degree is 8 pm after weighting, and the deviation degree is greater than the preset deviation threshold 5 pm. According to the historical production data, the particle size center value increases by 2 pm for each 1% increase in silicon content, and the particle size center value decreases by 1.5 pm for each 1% increase in carbon content. To reduce the center value by 8 pm, the silicon content needs to be reduced from 26% to 24%, which contributes to-4 pm; the carbon content needs to be increased from 14% to 16.7%, which contributes to-4.05 pm, and the total adjustment amount meets the requirements. The production time interval is 1 hour, so the adjusted parameters in the 10:00-12:00 time period are recorded in time sequence as ["10:00-11:00, silicon 24%, carbon 16.7%", "11:00-12:00, silicon 24%, carbon 16.7%"], forming an optimized proportioning parameter time sequence to guide real-time adjustment of the production line.

[0047] In step S18, the dynamic particle size prediction model is updated based on the optimized proportioning parameter time sequence to obtain an adjusted particle size prediction model, including: Based on the optimized proportioning parameter time sequence, the mean and variance characteristics of the optimized proportioning parameters are extracted; The optimized proportioning parameters are applied to actual production, and the green silicon carbide micro powder produced is detected to obtain optimized particle size distribution characteristics; When the mean and variance characteristics meet the preset mean threshold and preset variance threshold respectively, the optimized proportioning parameters and the optimized particle size distribution characteristics are linearly regressed and fitted to update the dynamic particle size prediction model to obtain the adjusted particle size prediction model.

[0048] It should be noted that the mean value feature of the optimized proportioning parameters refers to the arithmetic mean value of the optimized silicon and carbon content ratio in the time sequence, reflecting the overall level after the adjustment of the proportioning parameters. The variance feature refers to the average value of the square of the deviation of the parameter value at each time point from the mean value, reflecting the stability of the parameter in the time dimension. The smaller the variance, the more stable the production process after the adjustment of the proportioning. The optimized proportioning parameters are applied to actual production, and the green silicon carbide micro-powder particle size is measured by a laser particle size analyzer to obtain the optimized particle size distribution feature. The role is to form a new "input-output" data pair with the optimized proportioning parameters, and to provide the latest correlation samples for model updating. The preset mean value threshold is the silicon and carbon content reference range allowed by the process, such as the silicon content generally being 22%-28% and the carbon content generally being 12%-18%. The preset variance threshold represents the upper limit of the allowed fluctuation. When the mean value and variance features meet the corresponding preset threshold, the optimized proportioning parameters are within a reasonable range and have stability, and the corresponding particle size feature is reliable.

[0049] It should be noted that the linear regression fitting method uses the least squares method to update the mapping coefficients in the model, i.e. the influence coefficients of silicon and carbon content on particle size and distribution ratio, which are in units of μm / %, indicating the change amount of particle size center value / distribution ratio corresponding to a 1% change in silicon / carbon content. Specifically, the optimized silicon and carbon content ratio is taken as the independent variable, and the corresponding main particle size interval center value and distribution ratio are taken as the dependent variable. The least squares method is used to adjust the coefficients and constant term of the independent variable, so that the fitted linear equation can most accurately reflect the correlation between the proportioning parameters and the particle size features in the new samples. The final adjusted particle size prediction model has higher prediction accuracy and can support more accurate real-time control.

[0050] Exemplarily, in the optimized proportioning parameter time sequence, the silicon content average is 25%, which is within the preset silicon content average threshold of 22%-28%, and the variance is 0.8%2, which is less than the preset silicon content variance threshold of 1.5%2; the carbon content average is 15%, which is within the preset carbon content average threshold of 12%-18%, and the variance is 0.5%2, which is less than the preset carbon content variance threshold of 1%2. According to the optimized proportioning parameter, production is carried out, and the optimized particle size distribution characteristics are detected: the main particle size interval center value is 75 μm, and the distribution proportion is 60%. The least square method is used to fit the new sample of "silicon 25%, carbon 15%, center value 75 μm, and distribution proportion 60%", and the coefficient of silicon content in the model is adjusted from 2 μm / % to 1.9 μm / %, and the coefficient of carbon content is adjusted from-1.5 μm / % to-1.4 μm / %, to obtain the adjusted particle size prediction model. In the prediction of the subsequent silicon content of 25.5% and the carbon content of 14.8%, the particle size center value error is reduced from the original 2 μm to 0.8 μm, which significantly improves the prediction accuracy.

[0051] In summary, the present application discloses a green silicon carbide micro-powder particle size detection method, which comprises collecting and analyzing multiple batches of particle size distribution data to obtain a two-dimensional detection data set containing particle size range and distribution proportion; preprocessing the two-dimensional detection data set to obtain the final particle size distribution data set; performing dimensionality reduction analysis on the particle size distribution data set to determine the main particle size interval and the corresponding particle volume proportion, and linearly transforming to form a particle size distribution feature vector; obtaining raw material proportioning change data containing silicon content and carbon content proportion from production records, and performing data aggregation on the raw material proportioning change data in chronological order to obtain a proportioning parameter time sequence; calculating the correlation coefficient of the particle size distribution feature vector and the proportioning parameter time sequence, and when the correlation coefficient is greater than a preset correlation threshold, a nonlinear mapping relationship is established to obtain a dynamic particle size prediction model; real-time silicon content and carbon content proportion data are input into the dynamic particle size prediction model to obtain a predicted particle size distribution feature; when the main particle size interval in the particle size distribution feature deviates from a preset target particle size range, the silicon content and carbon content proportion are adjusted to obtain an optimized proportioning parameter time sequence; the dynamic particle size prediction model is updated based on the optimized proportioning parameter time sequence to obtain an adjusted particle size prediction model, and the main particle size interval and the distribution proportion are repeatedly predicted based on the adjusted particle size prediction model to obtain an optimized particle size distribution feature. The present application dynamically correlates the raw material proportioning parameters with the particle size distribution characteristics, and continuously optimizes the model parameters according to real-time data feedback, thereby realizing accurate prediction and dynamic adjustment of the particle size distribution of green silicon carbide micro-powder, and effectively improving the product particle size stability.

[0052] Referring to Figure 2 , the second embodiment of the present application provides a green silicon carbide micro-powder particle size detection device, which comprises: a data collection module, which collects and analyzes multiple batches of particle size distribution data to obtain a two-dimensional detection data set containing a particle size range and a distribution ratio; a data preprocessing module, which pre-processes the two-dimensional detection data set to obtain a final particle size distribution data set; a feature extraction module, which performs dimensionality reduction analysis on the particle size distribution data set to determine a main particle size interval and a corresponding particle volume ratio, and linearly transforms the main particle size interval and the corresponding particle volume ratio to form a particle size distribution feature vector; a ratio aggregation module, which obtains raw material ratio change data containing a silicon content ratio and a carbon content ratio from production records, and aggregates the raw material ratio change data in chronological order to obtain a ratio parameter time sequence; a model construction module, which calculates a correlation coefficient of the particle size distribution feature vector and the ratio parameter time sequence, establishes a nonlinear mapping relationship when the correlation coefficient is greater than a preset correlation threshold, and obtains a dynamic particle size prediction model; a real-time prediction module, which obtains real-time silicon content ratio and carbon content ratio data, inputs the real-time silicon content ratio and carbon content ratio data into the dynamic particle size prediction model, and obtains a predicted particle size distribution feature; a ratio optimization module, which adjusts the silicon content ratio and the carbon content ratio when the main particle size interval in the particle size distribution feature deviates from a preset target particle size range, and obtains an optimized ratio parameter time sequence; a model updating module, which updates the dynamic particle size prediction model based on the optimized ratio parameter time sequence to obtain an adjusted particle size prediction model, repeatedly predicts a main particle size interval and a distribution ratio based on the adjusted particle size prediction model, and obtains an optimized particle size distribution feature.

[0053] It should be noted that the green silicon carbide micro-powder particle size detection device provided by the embodiment of the present application is used to execute all process steps of the green silicon carbide micro-powder particle size detection method provided by the above embodiment, and the working principles and beneficial effects of the two are one-to-one correspondence, so they will not be repeated here.

[0054] The embodiment of the present application also provides an electronic device. The electronic device includes a processor, a memory, and a computer program, such as an abnormal data elimination program, stored in the memory and executable on the processor. The processor implements the steps in each of the above green silicon carbide micro-powder particle size detection method embodiments when executing the computer program, such as Figure 1 the step S11 shown. Alternatively, the processor implements the functions of each module / unit in each of the above device embodiments when executing the computer program, such as the data collection module.

[0055] For example, the computer program can be divided into one or more modules / units, which are stored in the memory and executed by the processor to complete the present application. The one or more modules / units can be a series of computer program instruction segments capable of completing a specific function, which are used to describe the execution process of the computer program in the electronic device.

[0056] The electronic device can be a computing device such as a desktop computer, a notebook computer, a palm computer, and a smart tablet. The electronic device can include, but is not limited to, a processor and a memory. Those skilled in the art can understand that the above components are only examples of the electronic device and do not constitute a limitation on the electronic device, and the electronic device can include more or fewer components than the above, or combine certain components, or different components, for example, the electronic device can also include an input / output device, a network access device, a bus, etc.

[0057] The processor can be a central processing unit (CPU), and can also be other general-purpose processors, digital signal processors (DSP), application specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gates or transistor logic components, discrete hardware components, etc. The general-purpose processor can be a microprocessor or the processor can also be any conventional processor, etc. The processor is the control center of the electronic device, which connects all parts of the electronic device through various interfaces and lines.

[0058] The memory can be used to store the computer program and / or modules, and the processor realizes various functions of the electronic device by running or executing the computer program and / or modules stored in the memory, and calling data stored in the memory. The memory can mainly include a program storage area and a data storage area, wherein the program storage area can store an operating system, at least one application required by a function (such as a sound playing function, an image playing function, etc.), and the like; and the data storage area can store data created according to the use of the mobile phone (such as audio data, a phone book, etc.), and the like. In addition, the memory can include a high-speed random access memory, and can also include a nonvolatile memory, for example, a hard disk, a memory, a plug-in hard disk, a smart media card (SMC), a secure digital (SD) card, a flash card, at least one disk storage device, a flash memory device, or other volatile solid-state memory devices.

[0059] The modules / units integrated in the electronic device, if realized in the form of a software function unit and sold or used as an independent product, can be stored in a computer readable storage medium. Based on this understanding, all or part of the processes in the above-mentioned embodiment methods can also be completed by a computer program instructing related hardware, and the computer program can be stored in a computer readable storage medium. When the processor executes the computer program, the steps of the above-mentioned various method embodiments can be realized. The computer program includes computer program code, which can be in the form of source code, object code, an executable file, or some intermediate form, etc. The computer readable medium can include any entity or device capable of carrying the computer program code, a recording medium, a U disk, a mobile hard disk, a magnetic disk, an optical disk, a computer memory, a read-only memory (ROM), a random access memory (RAM), an electrical carrier signal, a telecommunication signal, and a software distribution medium, etc. It should be noted that the contents included in the computer readable medium can be appropriately increased or decreased according to the requirements of legislation and patent practice in the jurisdiction, for example, in some jurisdictions, according to legislation and patent practice, the computer readable medium does not include electrical carrier signals and telecommunication signals.

[0060] It should be noted that the apparatus embodiments described above are merely illustrative, and the units described as separate units can or can not be physically separate, and the units displayed as units can or can not be physical units, i.e. can be located in one place, or can be distributed to multiple network units. Part or all of the modules can be selected to achieve the purpose of the embodiment scheme according to actual needs. In addition, the connection relationship between the modules in the apparatus embodiment provided by the present application indicates that there is a communication connection between them, which can be implemented as one or more communication buses or signal lines. Those skilled in the art can understand and implement it without creative labor.

[0061] The above specific embodiments further illustrate the purpose, technical scheme and beneficial effects of the present application. It should be understood that the above description is only a specific embodiment of the present application and is not intended to limit the protection scope of the present application. It is particularly pointed out that any modification, equivalent replacement, improvement, etc. made by those skilled in the art within the spirit and principles of the present application shall be included in the protection scope of the present application.

Claims

1. A method for detecting the particle size of green silicon carbide powder, characterized in that: include: Collect and analyze multiple batches of particle size distribution data to obtain a two-dimensional detection data set containing particle size range and distribution ratio; Preprocessing the two-dimensional detection data set to obtain a final particle size distribution data set; Performing dimensionality reduction analysis on the particle size distribution data set to determine the main particle size intervals and the corresponding particle volume proportions, and forming a particle size distribution feature vector after linear transformation; Obtaining raw material ratio change data including silicon content and carbon content ratio from production records, and aggregating the raw material ratio change data in chronological order to obtain a ratio parameter time series; Calculating the correlation coefficient between the particle size distribution feature vector and the time series of the ratio parameter; when the correlation coefficient is greater than a preset correlation threshold, establishing a nonlinear mapping relationship to obtain a dynamic particle size prediction model; Acquiring real-time silicon content and carbon content ratio data, inputting the data into the dynamic particle size prediction model, and obtaining predicted particle size distribution characteristics; When the main particle size interval in the particle size distribution characteristics deviates from the preset target particle size range, the ratio of silicon content to carbon content is adjusted to obtain an optimized ratio parameter time series; The dynamic particle size prediction model is updated based on the optimized ratio parameter time series to obtain an adjusted particle size prediction model, and the main particle size intervals and distribution ratios are repeatedly predicted based on the adjusted particle size prediction model to obtain optimized particle size distribution characteristics.

2. The green silicon carbide powder particle size detection method according to claim 1, characterized in that: The preprocessing of the two-dimensional detection data set to obtain a final particle size distribution data set includes: Calculating the absolute value of the deviation between the volume fraction of each particle size interval in the two-dimensional detection data set and the mean volume fraction in the data set; when the absolute value of the deviation is greater than a preset deviation threshold, marking it as an abnormal data point, and removing the abnormal data point to obtain a first data set; Calculating the relative position of each data point in the first data set relative to the minimum and maximum values ​​of the data set, and mapping the values ​​to the range of 0-1 to obtain a normalized second data set; According to the second data set, the average value of the data points in the preset sliding window is calculated and the center point data value is replaced. After performing a preset number of iterations, the final particle size distribution data set is obtained.

3. The green silicon carbide powder particle size detection method according to claim 1, characterized in that: The particle size distribution data set is subjected to dimensionality reduction analysis to determine the main particle size intervals and the corresponding particle volume proportions, and a particle size distribution feature vector is formed after linear transformation, including: Obtaining the original particle size range and the original distribution ratio from the particle size distribution data set, and calculating the average particle size; Based on the principal component analysis method, the multidimensional features consisting of the original particle size range, the original distribution ratio and the average particle size are mapped to a low-dimensional space, and the characteristic value is calculated; Screening out the particle size intervals whose characteristic values ​​are greater than a preset characteristic value threshold, determining them as the main particle size intervals, and extracting the particle volume proportions corresponding to the main particle size intervals; The main particle size interval and the particle volume ratio are linearly transformed and combined to form the particle size distribution feature vector.

4. The green silicon carbide powder particle size detection method according to claim 1, characterized in that: The calculation of the correlation coefficient between the particle size distribution characteristic vector and the ratio parameter time series, and when the correlation coefficient is greater than a preset correlation threshold, establishing a nonlinear mapping relationship to obtain a dynamic particle size prediction model, includes: According to the time series of the ratio parameters, silicon content ratio data and carbon content ratio data are obtained; Calculating the correlation coefficients between the silicon content ratio data and the carbon content ratio data and the particle size distribution characteristic vector to obtain a silicon content correlation coefficient value and a carbon content correlation coefficient value; When the silicon content correlation coefficient value or the carbon content correlation coefficient value exceeds a preset correlation threshold, a support vector regression algorithm is used to fit the nonlinear mapping relationship between the silicon content and carbon content ratio and the particle size distribution to obtain a main particle size interval prediction value and a distribution ratio prediction value; The main particle size interval prediction value and the distribution ratio prediction value are weightedly fused with the particle size distribution feature vector to obtain the dynamic particle size prediction model.

5. The green silicon carbide powder particle size detection method according to claim 1, characterized in that: The real-time silicon content and carbon content ratio data are obtained and input into the dynamic particle size prediction model to obtain the predicted particle size distribution characteristics, including: The silicon content and carbon content ratio data are collected in real time, outliers outside a preset reasonable range are eliminated, and the average value of valid data within a preset time period is taken as the center, with the time point corresponding to the missing value as the center, to fill the gap and obtain a clean data set; The principal component analysis method is used to extract the time series characteristics of the clean data set to obtain an element ratio feature set including the change trend of the silicon content and carbon content ratio; The variance contribution rate of the element ratio feature set is calculated. When the variance contribution rate exceeds a preset variance contribution threshold, the element ratio feature set is input into the dynamic particle size prediction model to obtain the predicted particle size distribution characteristics.

6. The green silicon carbide powder particle size detection method according to claim 1, characterized in that: When the main particle size interval in the particle size distribution characteristics deviates from the preset target particle size range, the ratio of silicon content to carbon content is adjusted to obtain an optimized ratio parameter time series, including: Extracting a main particle size interval from the particle size distribution characteristics, calculating the deviation between the main particle size interval and a preset target particle size range to obtain a particle size deviation; When the particle size deviation is greater than a preset deviation threshold, a linear regression adjustment is performed on the ratio of silicon content to carbon content to generate an optimized ratio parameter; The optimized ratio parameters are aggregated in chronological order to obtain the optimized ratio parameter time series.

7. The method for detecting particle size of green silicon carbide powder according to claim 6, wherein: The updating of the dynamic particle size prediction model based on the optimized ratio parameter time series to obtain an adjusted particle size prediction model includes: Extracting mean and variance characteristics of the optimized matching parameters based on the optimized matching parameter time series; The optimized ratio parameters are applied to actual production, and the produced green silicon carbide micropowder is tested to obtain the optimized particle size distribution characteristics; When the mean and variance characteristics meet the preset mean threshold and the preset variance threshold respectively, the optimized ratio parameters are linearly regressed with the optimized particle size distribution characteristics to update the dynamic particle size prediction model to obtain the adjusted particle size prediction model.

8. A green silicon carbide powder particle size detection device, characterized in that: include: The data acquisition module collects and analyzes multiple batches of particle size distribution data to obtain a two-dimensional detection data set containing the particle size range and distribution ratio; A data preprocessing module preprocesses the two-dimensional detection data set to obtain a final particle size distribution data set; A feature extraction module performs dimensionality reduction analysis on the particle size distribution data set to determine the main particle size intervals and the corresponding particle volume proportions, and forms a particle size distribution feature vector after linear transformation; A ratio aggregation module obtains raw material ratio change data including silicon content and carbon content ratio from production records, aggregates the raw material ratio change data in chronological order, and obtains a ratio parameter time series; A model building module calculates the correlation coefficient between the particle size distribution feature vector and the time series of the ratio parameter, and when the correlation coefficient is greater than a preset correlation threshold, establishes a nonlinear mapping relationship to obtain a dynamic particle size prediction model; A real-time prediction module obtains real-time silicon content and carbon content ratio data, inputs the data into the dynamic particle size prediction model, and obtains predicted particle size distribution characteristics; A ratio optimization module, which adjusts the ratio of silicon content to carbon content when the main particle size interval in the particle size distribution characteristic deviates from the preset target particle size range, to obtain an optimized ratio parameter time series; The model updating module updates the dynamic particle size prediction model based on the optimized ratio parameter time series to obtain an adjusted particle size prediction model, and repeatedly predicts the main particle size intervals and distribution ratios based on the adjusted particle size prediction model to obtain optimized particle size distribution characteristics.

9. An electronic device, characterized in that: The invention comprises a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, wherein when the processor executes the computer program, the method for detecting the particle size of green silicon carbide powder according to any one of claims 1 to 7 is implemented.

10. A computer-readable storage medium, characterized in that The computer-readable storage medium includes a stored computer program, wherein when the computer program is executed, the device where the computer-readable storage medium is located is controlled to execute the green silicon carbide powder particle size detection method according to any one of claims 1 to 7.

Images