Production process parameter management system and method, and preparation method of putty powder coarse base

Abstract

The application discloses a production process parameter management system and method and a preparation method of putty powder coarse bottom, relates to the parameter management field, and converts historical production data into reusable process knowledge base through the application, can accurately predict the quality risk under different parameter combinations through the established defective rate mapping model, and significantly reduces the resource waste caused by the traditional trial and error method; through high-precision spectral analysis, the micro differences of material components are captured, image processing technology is combined to quickly locate surface defects, and finally, millisecond-level automatic sorting of the qualified products and defective products is realized; not only the misjudgment problem of a single detection method is solved, but also the detection model can be reversely optimized through continuously accumulated sample data; when facing new raw materials or formula changes, the system can quickly migrate historical learning experience, ensures dynamic adaptation of quality control standards, and finally realizes production mode transformation from passive quality inspection to active prevention and from experience driving to data driving.

Description

Technical Field

[0001] This invention belongs to the field of parameter management, and specifically relates to a production process parameter management system, method, and method for preparing coarse putty powder base. Background Technology

[0002] The traditional coarse putty powder preparation process relies on trial and error based on human experience, which can easily lead to batch quality accidents due to delayed parameter adjustments. The traditional preparation process usually uses a single detection method, which can easily lead to problems such as spectral analysis missing internal defects or image recognition misjudging component deviations. In addition, when the characteristics of raw materials fluctuate or the formula is updated, traditional quality inspection standards often become ineffective and have poor adaptability. Summary of the Invention

[0003] In response to the problems in related technologies, this invention proposes a production process parameter management system, a method, and a method for preparing coarse putty powder, in order to overcome the aforementioned technical problems existing in the existing related technologies.

[0004] To solve the above-mentioned technical problems, the present invention is achieved through the following technical solution:

[0005] This invention relates to a method for managing production process parameters, comprising the following steps:

[0006] S1. Collect process parameter data and corresponding finished product defect rates for several product preparation and production processes in history.

[0007] S2. Construct a final finished product defect rate mapping model based on the data collected in S1; collect the process parameter data corresponding to the product to be prepared and input it into the final finished product defect rate mapping model for mapping, and adjust the process parameter data of the product to be prepared;

[0008] S3. Using the adjustment results of S2, the product to be prepared is prepared and produced to obtain the finished product set to be sorted;

[0009] S4. Collect historical finished product samples from different receipt dates to establish a sample set; collect and save spectral data of the samples in the sample set; process the spectral data of the samples to obtain processed spectral data;

[0010] S5. Collect images of finished products in the sample set, and train the recognition module based on the image collection results to obtain the trained intelligent recognition module.

[0011] S6. After training is completed, the spectral data processed in S4 is set as the reference spectrum. The finished product set to be sorted is divided based on the reference spectrum. The division results include: good finished products and defective finished products.

[0012] S7. After the division is completed, the finished product set to be sorted is input into the intelligent recognition module for recognition, and the sorting is controlled by the intelligent recognition module.

[0013] Preferably, step S1 includes the following steps:

[0014] S11. Obtain the product to be prepared and several types of process parameters corresponding to the production process of the product to be prepared, and obtain the current production process parameter type set;

[0015] S12. Based on the product to be prepared and the current set of production process parameters, collect process parameter data and corresponding finished product defect rate data from several historical production processes of products of the same type as the product to be prepared, to obtain a historical production process parameter dataset and a historical finished product defect rate dataset;

[0016] By collecting and analyzing historical data on the correlation between production process parameters and finished product defect rates, data support is provided for the subsequent construction of a mapping model between production process parameters and finished product defect rates. It can automatically identify implicit correlations between various parameters and quality indicators, dynamically optimize the current combination of production parameters, and avoid the resource waste caused by traditional trial-and-error adjustments. Combined with historical matching records of grinding fineness and construction smoothness, it can adaptively adjust the parameters of screening equipment. This data-driven closed-loop control not only reduces reliance on human experience but also enables rapid response to changes in product type. Furthermore, by continuously accumulating abnormal operating condition data, it gradually builds process tolerance boundaries, achieving an upgrade in quality control from passive error correction to proactive prevention.

[0017] Preferably, step S2 includes the following steps:

[0018] S21. Based on the historical production process parameter dataset and the historical finished product defect rate dataset, construct a mapping model between the product's production process parameters and the corresponding finished product defect rate data to obtain the final finished product defect rate mapping model.

[0019] S22. Set the current finished product defect rate threshold; obtain various preset process parameter data before the product to be prepared is prepared and produced according to the current production process parameter type set, and obtain the current production process parameter dataset;

[0020] S23. Input the current production process parameter dataset into the final finished product defect rate mapping model to obtain the current initial finished product defect rate data;

[0021] If the current initial finished product defect rate data is greater than or equal to the current finished product defect rate threshold, the current production process parameter dataset is repeatedly adjusted; otherwise, no adjustment is required; the current production process parameter dataset is used as the current final production process parameter dataset.

[0022] The adjusted current production process parameter dataset is input again into the final finished product defect rate mapping model to obtain the current adjusted finished product defect rate data; until the current adjusted finished product defect rate data is less than the current finished product defect rate threshold, the current final production process parameter dataset is obtained.

[0023] By constructing a mapping model between production process parameters and finished product defect rate, accurate prediction and optimization of the entire process of putty powder coarse base production quality are achieved. The model trained based on historical datasets can quantitatively analyze the nonlinear impact of each parameter on finished product quality. The defect rate output by the model provides data basis for the subsequent defect sorting process, thereby comprehensively improving production stability and resource utilization. Through a closed-loop control mechanism of dynamic threshold constraints and iterative optimization, intelligent self-correction of process parameters in the putty powder production process is realized, eliminating the lag of manual trial adjustment, and quickly restoring process stability when there are batch differences in raw materials or changes in equipment status.

[0024] Preferably, the final product defect rate mapping model described in S21 adopts a hybrid model combining DNN and attention mechanism;

[0025] 1D convolution processes numerical parameters, LSTM captures temporal dependencies, and attention mechanism dynamically weights the data, comprehensively covering the complex influence of process parameters; Anti-overfitting design: Dropout layer combined with early stopping mechanism avoids overfitting to small datasets; Interpretability enhancement: Attention weight visualization can intuitively display key process parameters.

[0026] Preferably, step S3 includes the following steps:

[0027] S31. The product to be prepared is prepared and produced using the current final production process parameter dataset. After the preparation and production are completed, a finished product set to be sorted is obtained.

[0028] The final process parameters validated by the model are directly applied to the actual production line, ensuring that the production process of each batch of putty powder strictly follows the data-driven quality baseline, thereby significantly reducing quality fluctuations caused by parameter drift; eliminating the error risk of manually passing parameters in traditional production, and can quickly adapt to the optimal process combination when the characteristics of raw materials change, while providing feedback data from real production scenarios for subsequent model optimization, continuously improving prediction accuracy and process stability.

[0029] Preferably, step S4 involves acquiring and saving spectral data of samples in the sample set; processing the spectral data of the samples to obtain processed spectral data includes the following steps:

[0030] S41. Based on near-infrared spectroscopy detection technology, perform spectral acquisition on historical finished product samples in the sample set; place historical finished product samples from different dates on the infrared acquisition window of the instrument, set the infrared scanning parameters of the instrument, and install spectral detectors around the objects to be scanned.

[0031] S42. Quantitative analysis was conducted on the different absorbance of historical product samples from different dates in different wavelength bands. Based on the formula for the absorption of light by substances, multiple scans were performed, and different parts of each historical product sample were repeatedly collected. The results of multiple scans were summarized and averaged to construct near-infrared spectral data.

[0032] S43. Average the near-infrared spectral data to obtain the averaged spectral data.

[0033] S44. Denoise the averaged spectral data; after denoising, cluster the denoised spectral data.

[0034] By averaging the spectral data of historical finished product samples from different receiving dates, and then denoising the averaged spectral data using a smoothing and noise reduction method, the impact of noise on the spectral data is reduced. After denoising, the denoised spectral data is clustered to separate genuine finished products from defective finished products, thereby improving the efficiency of finished product sorting.

[0035] Preferably, S44 includes the following steps:

[0036] S441. Noise in the averaged spectral data is eliminated by setting a convolutional smoothing window; a smoothing window with a width of 2w+1 is set, and the average value of the measured values ​​at the center wavelength point j and the points w before and after it is used to replace the measured value of the center wavelength point. At the same time, j is moved from left to right until the smoothing of all points is completed; the denoised spectral data is obtained.

[0037] S442. Select a set of denoised spectral data corresponding to genuine finished products and defective finished products in the sample set as cluster centers. The distance from the center of each set represents a type. Calculate the distance between other denoised spectral data in the sample set and the cluster centers, and add the type represented by the nearest cluster center.

[0038] The calculation process involves denoising the spectral data of the data type, updating the spectral data of the cluster centers, re-clustering based on the updated spectral data of the cluster centers, and iteratively calculating until the data type converges, then outputting the clustered spectral data.

[0039] The clustered spectral data is set as the processed spectral data.

[0040] Preferably, step S5 includes the following steps:

[0041] S51. Collect images of finished products in the sample set, segment the collected image data, and obtain the processed image data.

[0042] S52. Train the recognition of the processed image data and save the training recognition result to the intelligent recognition module;

[0043] S51 includes the following steps:

[0044] S511. Set the total number of pixels in the acquired image data to N, and divide them into D classes. The center point of each class, Point Belongs to the Class, construct the objective function and constraints;

[0045] S512. Perform iterative clustering operations on pixels of the same type to separate the image target from the background;

[0046] Iterative clustering of pixels of the same type includes: calculating the distance from random points in each class to the class center point based on the initially set center point of each class; continuously adjusting the center point of each class according to the constraints, and iterating until the center points of all classes no longer change, at which point the iteration stops;

[0047] By comparing the distances in set J, when the distances in set J no longer decrease, the image target and background are separated; and the separated image is set as the processed image data.

[0048] S52 includes the following steps:

[0049] S521. Divide the processed image data into image data blocks of the same size to obtain an image data block set;

[0050] Encode the pixel data blocks in the image data block set;

[0051] The encoding of pixel data blocks in the image data block set is set as follows: ; These represent the red, green, and blue pixel values ​​of each pixel in the image data block, respectively. Encode the pixel data in the image data block; obtain the encoded image data block.

[0052] S522. Input the encoded image data block into the convolutional neural network for feature extraction; divide the convolutional neural network into channels based on the pixel data in the encoded image data block, and set three groups of channels to extract features from the pixel values ​​of red, green and blue colors in the image data block respectively.

[0053] S523. The extracted features are fused to obtain the fused image data block features; the features extracted from the three channels are fused through a fully connected operation to obtain the fused image data block.

[0054] S524. Save the features of the fused image data block to the intelligent recognition module and use it as a reference.

[0055] Preferably, step S7 includes the following steps:

[0056] S71. When the intelligent recognition module receives the divided finished product set to be sorted, it identifies the color characteristics of the finished product to be sorted in real time, and based on the characteristics stored in the intelligent recognition module, it performs secondary division of the finished product to be sorted in real time according to color.

[0057] When the color features of the finished product to be sorted identified in real time are not similar to the features stored in the intelligent recognition module, the color features of the finished product to be sorted in real time will be saved and the intelligent recognition module will be updated.

[0058] The production process parameter management system includes a historical product preparation production data acquisition module, a finished product defect rate mapping model construction module, a current production process parameter data mapping adjustment module, a current preparation production execution module, a finished product spectral data acquisition and processing module, a finished product image data training module, and an intelligent recognition module.

[0059] The present invention has the following beneficial effects:

[0060] 1. This invention achieves intelligent upgrading of the entire process from process optimization to finished product sorting by constructing a closed-loop quality control system of data-model-execution-feedback; it transforms historical production data into a reusable process knowledge base, and through the established defect rate mapping model, it can accurately predict quality risks under different parameter combinations, significantly reducing the resource waste caused by traditional trial and error methods; it captures the microscopic differences in material composition through high-precision spectral analysis, and quickly locates surface defects by combining image processing technology, ultimately achieving millisecond-level automatic sorting of good and bad products; it not only solves the problem of misjudgment by single detection methods, but also optimizes the detection model through continuously accumulated sample data; when faced with new raw materials or formula changes, the system can quickly transfer historical learning experience to ensure dynamic adaptation of quality control standards, ultimately realizing a production mode transformation from passive quality inspection to proactive prevention, and from experience-driven to data-driven.

[0061] 2. In this invention, historical product samples are collected and spectral data is acquired using near-infrared spectroscopy detection technology, while the spectral data of the historical product samples is saved. Then, the spectral data of the historical product samples is processed using spectral data analysis methods, and images of the historical product samples are acquired. Based on the image acquisition results, a trained intelligent recognition module is obtained. After training, the processed spectral data is set as the reference spectrum, and the real-time received products to be sorted are divided based on the reference spectrum. After the division, the real-time received products to be sorted are input into the intelligent recognition module for recognition, and the intelligent recognition module performs sorting control, thereby improving the intelligence of the sorting control of the products to be sorted.

[0062] 3. In this invention, the spectral data of historical finished product samples in the sample set are collected based on near-infrared spectroscopy detection technology, and quantitative analysis is performed on the different absorbance of the samples in different wavelength bands to construct near-infrared spectral data, thereby improving the accuracy of spectral detection of finished product samples.

[0063] Of course, any product implementing this invention does not necessarily need to achieve all of the advantages described above at the same time. Attached Figure Description

[0064] To more clearly illustrate the technical solutions of the embodiments of the invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0065] Figure 1 This is a flowchart illustrating the production process parameter management method of the present invention;

[0066] Figure 2 This is a schematic diagram of the modules of the production process parameter management system of the present invention. Detailed Implementation

[0067] The technical solutions of the embodiments of the invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the invention, and not all embodiments. Based on the embodiments of the invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the invention.

[0068] Example 1

[0069] Please see Figure 1 This embodiment describes a method for managing production process parameters, including the following steps:

[0070] S1. Collect process parameter data and corresponding finished product defect rates for several product preparation and production processes in history.

[0071] S1 includes the following steps:

[0072] S11. Obtain the product to be prepared and several types of process parameters corresponding to the production process of the product to be prepared, to obtain the current production process parameter type set; taking the preparation and production of coarse putty powder as an example, the current production process parameter type set includes the ratio of cementitious materials to fillers (the ratio of cementitious materials such as cement and gypsum to fillers such as calcium carbonate and talc needs to be adjusted according to the product type, for example, waterproof putty needs to increase the proportion of lime powder to improve water resistance), the amount of additives (the amount of additives such as cellulose ether (water retention agent) and latex powder (enhancing adhesion) needs to be precisely controlled), and mixing. Time and speed (mixing time and speed need to be precisely controlled to avoid over-stirring, which can lead to material clumping or performance degradation), raw material moisture content (if the raw material has a high moisture content, it needs to be adjusted by drying equipment (such as a drum dryer) to prevent the finished product from becoming damp and deteriorating), grinding fineness (putty powder particles have strict requirements for fineness; if they are too coarse, they will affect the smoothness of the construction), screening efficiency (the amplitude and frequency of the vibrating screen need to be adjusted to ensure the removal of large particles of impurities, while avoiding excessive loss of qualified materials), drying temperature (too high a temperature in the airflow dryer will cause the additives to become ineffective, while too low a temperature will result in incomplete drying), etc.

[0073] S12. Based on the product to be prepared and the current set of production process parameters, collect process parameter data and corresponding finished product defect rate data from several historical production processes of products of the same type as the product to be prepared, to obtain a historical production process parameter dataset and a historical finished product defect rate dataset;

[0074] By collecting and analyzing historical data on the correlation between production process parameters and finished product defect rates, data support is provided for constructing a mapping model between production process parameters and finished product defect rates. Machine learning (or deep learning) models based on historical datasets can automatically identify the implicit correlations between various parameters (such as cementitious material ratios, additive dosages, and mixing processes) and quality indicators, dynamically optimizing the current combination of production parameters and avoiding the resource waste caused by traditional trial-and-error adjustments. For example, by analyzing the correlation between historical moisture content data and finished product agglomeration rate, the drying requirements of raw materials can be accurately predicted. Combined with historical matching records of grinding fineness and construction smoothness, the parameters of screening equipment can be adaptively adjusted. This data-driven closed-loop control not only reduces reliance on human experience but also enables rapid response to changes in product type (such as switching waterproof putty formulas). At the same time, by continuously accumulating abnormal operating condition data (such as cases of additive failure caused by temperature fluctuations), the process tolerance boundary is gradually built, achieving an upgrade in quality control from passive error correction to proactive prevention.

[0075] S2. Construct a final finished product defect rate mapping model based on the data collected in S1; collect the process parameter data corresponding to the product to be prepared and input it into the final finished product defect rate mapping model for mapping, and adjust the process parameter data of the product to be prepared;

[0076] S2 includes the following steps:

[0077] S21. Based on the historical production process parameter dataset and the historical finished product defect rate dataset, construct a mapping model between the product's production process parameters and the corresponding finished product defect rate data to obtain the final finished product defect rate mapping model.

[0078] The final product defect rate mapping model described in S21 adopts a hybrid model combining DNN and attention mechanism;

[0079] S21 includes the following steps:

[0080] S211. Construct an initial finished product defect rate mapping model and set the training data ratio (e.g., 8:2 or 7:3, which can be adjusted adaptively according to the actual training situation); divide the historical production process parameter dataset and the historical finished product defect rate dataset according to the training data ratio to obtain the training dataset and the test dataset.

[0081] S212. Set a training error threshold (10%~15%, which can be adjusted adaptively according to the actual training situation); input the training dataset into the initial finished product defect rate mapping model for training; during the training process, if the training error is less than the training error threshold, stop training and obtain the trained finished product defect rate mapping model; otherwise, continue training until the training error is less than the training error threshold.

[0082] S213. Set the test accuracy threshold (90%~95%, which can be adjusted adaptively according to the actual test situation); input the test dataset into the trained finished product defect rate mapping model for testing; after the test is completed, obtain the test accuracy data; if the test accuracy data is greater than or equal to the test accuracy threshold, use the trained finished product defect rate mapping model as the final finished product defect rate mapping model.

[0083] Otherwise, return to S212 to continue training the trained finished product defect rate mapping model, and repeat S213 until the test accuracy data is greater than or equal to the test accuracy threshold.

[0084] The results of the initial finished product defect rate mapping model are shown in Table 1 below:

[0085] Table 1. Example of Initial Finished Product Defect Rate Mapping Model Structure

[0086] Model Name Model type Model Structure Initial Finished Product Defect Rate Mapping Model Hybrid Model Combining DNN and Attention Mechanism Input Layer: Input Dimension: Number of corresponding process parameters (e.g., 7 types of parameters such as cementitious material ratio, moisture content, etc.), requiring standardization (Z-score standardization); Feature Extraction Module: 1D Convolutional Layer (One-Dimensional Convolution): Number of convolutional kernels: 64, size 3, stride 1, activation function is ReLU (Modified Linear Unit), used to extract the correlation features between local process parameters; Pooling Layer: Max pooling, pooling window size 2, reducing dimensionality while retaining key features; Bidirectional LSTM Layer (Long Short-Term Memory Network): Number of hidden units: 128, capturing the temporal dependencies of the process parameter sequence (e.g., the influence of stirring time on subsequent drying temperature); Attention Mechanism Layer: Self-Attention: Calculates the correlation between each process parameter... The weights of the defect rate are assigned to highlight key parameters (such as the proportion of cementitious materials, which may have a significantly higher impact than screening efficiency); Fully connected layers (densely connected layers): First layer: 256 neurons, activation function is LeakyReLU (with leakage correction linear unit, negative slope 0.2), to prevent gradient vanishing; Second layer: 128 neurons, activation function is the same as above, with an increased Dropout rate of 0.3 to reduce overfitting; Output layer: 1 neuron, using the Sigmoid function to output the defect rate probability (0~1); Optimizer and loss function: Optimizer: Adam (adaptive moment estimation), initial learning rate 0.001, with a learning rate decay strategy; Loss function: Mean squared error (MSE) or binary cross-entropy, selected according to whether the defect rate is a binary classification problem;

[0087] 1D convolution processes numerical parameters, LSTM captures temporal dependencies, and attention mechanism dynamically weights the data, comprehensively covering the complex effects of process parameters; Anti-overfitting design: Dropout layer combined with early stopping mechanism avoids overfitting to small datasets; Enhanced interpretability: Attention weight visualization can intuitively display key process parameters (such as the contribution of drying temperature to the defect rate).

[0088] S22. Set the current finished product defect rate threshold (which can be adaptively set according to the actual process production requirements); obtain various preset process parameter data before the product to be prepared is prepared and produced according to the current production process parameter type set, and obtain the current production process parameter dataset;

[0089] S23. Input the current production process parameter dataset into the final finished product defect rate mapping model to obtain the current initial finished product defect rate data;

[0090] If the current initial finished product defect rate data is greater than or equal to the current finished product defect rate threshold, the current production process parameter dataset is repeatedly adjusted; otherwise, no adjustment is required; the current production process parameter dataset is used as the current final production process parameter dataset.

[0091] The adjusted current production process parameter dataset is input again into the final finished product defect rate mapping model to obtain the current adjusted finished product defect rate data; until the current adjusted finished product defect rate data is less than the current finished product defect rate threshold, the current final production process parameter dataset is obtained.

[0092] By constructing a mapping model between production process parameters and finished product defect rate, the entire process of accurate prediction and optimization of the coarse base putty powder production quality is achieved. The model trained based on historical datasets can quantitatively analyze the nonlinear impact of various parameters (such as cementitious material ratio, additive dosage, mixing process, etc.) on finished product quality. The defect rate output by the model provides data basis for the subsequent defect sorting process, thereby comprehensively improving production stability and resource utilization. Through a closed-loop control mechanism of dynamic threshold constraints and iterative optimization, intelligent self-correction of process parameters in the putty powder production process is achieved. For example, in response to the problem of insufficient gray calcium powder ratio that may occur in waterproof putty, the system will gradually increase the cementitious material ratio according to the gradient, and at the same time, adjust the mixing time to avoid clumping, until the adjusted defect rate prediction value drops below the threshold, eliminating the lag of manual trial adjustment, and can also quickly restore process stability when there are batch differences in raw materials (such as moisture content fluctuations) or changes in equipment status.

[0093] S3. Using the adjustment results of S2, the product to be prepared is prepared and produced to obtain the finished product set to be sorted;

[0094] S3 includes the following steps:

[0095] S31. The product to be prepared is prepared and produced using the current final production process parameter dataset. After the preparation and production are completed, a finished product set to be sorted is obtained.

[0096] The final process parameters validated by the model (such as the optimized proportion of cementitious materials and mixing time) are directly applied to the actual production line to ensure that the production process of each batch of putty powder strictly follows the data-driven quality baseline, thereby significantly reducing quality fluctuations caused by parameter drift; it eliminates the error risk of manually passing parameters in traditional production, and can quickly adapt to the optimal process combination when the characteristics of raw materials change (such as the difference in activity of new batches of gray calcium powder), while providing feedback data from real production scenarios for subsequent model optimization, continuously improving prediction accuracy and process stability;

[0097] S4. Collect historical finished product samples from different receipt dates to establish a sample set; collect and save spectral data of the samples in the sample set; process the spectral data of the samples to obtain processed spectral data;

[0098] S4 describes the process of acquiring and saving spectral data from the samples in the sample set; processing the spectral data of the samples to obtain processed spectral data includes the following steps:

[0099] S41. Based on near-infrared spectroscopy detection technology, perform spectral acquisition on historical finished product samples in the sample set; place historical finished product samples from different dates on the infrared acquisition window of the instrument, set the infrared scanning parameters of the instrument, and install spectral detectors around the objects to be scanned.

[0100] S42. Quantitative analysis was conducted on the different absorbance of historical product samples from different dates in different wavelength bands. Based on the formula for the absorption of light by substances, multiple scans were performed, and different parts of each historical product sample were repeatedly collected. The results of multiple scans were summarized and averaged to construct near-infrared spectral data.

[0101] The formula for the absorption of light by matter is shown below:

[0102] ;

[0103] Where A is the absorbance and I is the intensity of the light beam after passing through the finished product. The molar absorptivity is 1. denoted as , where c is the initial incident light intensity, c is the component concentration of the finished product under test, and b is the optical path length.

[0104] S43. Average the near-infrared spectral data to obtain the averaged spectral data.

[0105] Specifically, calculate the average spectral value for each sample in the near-infrared spectral data. ;

[0106] ;

[0107] Where n is the number of samples and k is the number of wavelength points. This represents the average spectral value with k wavelength points. This represents the spectral value of wavelength point k in the i-th sample;

[0108] S44. Denoise the averaged spectral data; after denoising, cluster the denoised spectral data.

[0109] S44 includes the following steps:

[0110] S441. Noise in the averaged spectral data is eliminated by setting a convolutional smoothing window; a smoothing window with a width of 2w+1 is set, and the average value of the measured values ​​at the center wavelength point j and the points w before and after it is used to replace the measured value of the center wavelength point. At the same time, j is moved from left to right until the smoothing of all points is completed; the denoised spectral data is obtained.

[0111] The convolution smoothing formula is as follows:

[0112] ;

[0113] in, This represents the measured value of the center wavelength point j after smoothing. H represents the average spectral value at the center wavelength point j, and H represents the normalization factor. This represents the measured values ​​at points z before and after the center point j. The smoothing coefficient for point z is represented by ; w is a randomly selected number.

[0114] S442. Select a set of denoised spectral data corresponding to genuine finished products and defective finished products in the sample set as cluster centers. The distance from the center of each set represents a type. Calculate the distance between other denoised spectral data in the sample set and the cluster centers, and add the type represented by the nearest cluster center.

[0115] The calculation process involves denoising the spectral data of the data type, updating the spectral data of the cluster centers, re-clustering based on the updated spectral data of the cluster centers, and iteratively calculating until the data type converges, then outputting the clustered spectral data.

[0116] The clustered spectral data is defined as the processed spectral data;

[0117] By averaging the spectral data of historical finished product samples collected from different receiving dates, and then denoising the averaged spectral data using a smoothing and denoising method, the impact of noise on the spectral data is reduced. After denoising, the denoised spectral data is clustered to separate genuine finished products from defective finished products, thereby improving the efficiency of finished product sorting.

[0118] S5. Collect images of finished products in the sample set, and train the recognition module based on the image collection results to obtain the trained intelligent recognition module.

[0119] S5 includes the following steps:

[0120] S51. Collect images of finished products in the sample set, segment the collected image data, and obtain the processed image data.

[0121] S52. Train the recognition of the processed image data and save the training recognition result to the intelligent recognition module;

[0122] S51 includes the following steps:

[0123] S511. Set the total number of pixels in the acquired image data to N, and divide them into D classes. The center point of each class, Point Belongs to the The class is used to construct the objective function and constraints, as follows:

[0124] ;

[0125] ;

[0126] Where m is the fuzzy index, m>1. Indicates the first The center point of the class Indicates the number of collections One sample point, Point To the Class center point The distance, J represents the point. The set of distances to various center points;

[0127] S512. Perform iterative clustering operations on pixels of the same type to separate the image target from the background;

[0128] Iterative clustering of pixels of the same type includes: calculating the distance from random points in each class to the class center point based on the initially set center point of each class; continuously adjusting the center point of each class according to the constraints, and iterating until the center points of all classes no longer change, at which point the iteration stops;

[0129] By comparing the distances in set J, when the distances in set J no longer decrease, the image target and background are separated; and the separated image is set as the processed image data.

[0130] S52 includes the following steps:

[0131] S521. Divide the processed image data into image data blocks of the same size to obtain an image data block set;

[0132] Encode the pixel data blocks in the image data block set;

[0133] The encoding of pixel data blocks in the image data block set is set as follows: ; These represent the red, green, and blue pixel values ​​of each pixel in the image data block, respectively. Encode the pixel data in the image data block; obtain the encoded image data block.

[0134] S522. Input the encoded image data block into the convolutional neural network for feature extraction; divide the convolutional neural network into channels based on the pixel data in the encoded image data block, and set three groups of channels to extract features from the pixel values ​​of red, green and blue colors in the image data block respectively.

[0135] The formula for calculating convolution is as follows:

[0136] ;

[0137] in, This represents the input image data block. This represents the weights of the corresponding convolution kernel. Indicates the bias parameter. This represents the characteristics of the corresponding image data block in the output;

[0138] S523. The extracted features are fused to obtain the fused image data block features; the features extracted from the three channels are fused through a fully connected operation to obtain the fused image data block.

[0139] The formula for a fully connected operation is as follows:

[0140]

[0141] Where W is the weighting parameter, This represents the features of the fused image data blocks. These represent the characteristics of the pixel values ​​for red, green, and blue colors in an image data block, respectively.

[0142] S524. Save the features of the fused image data blocks to the intelligent recognition module and use them as a reference;

[0143] S6. After training is completed, the spectral data processed in S4 is set as the reference spectrum. The finished product set to be sorted is divided based on the reference spectrum. The division results include: good finished products and defective finished products.

[0144] S7. After the division is completed, the finished product set of the divided products to be sorted is input into the intelligent recognition module for recognition, and sorting control is performed through the intelligent recognition module.

[0145] S7 includes the following steps:

[0146] S71. When the intelligent recognition module receives the divided finished product set to be sorted, it identifies the color characteristics of the finished product to be sorted in real time, and based on the characteristics stored in the intelligent recognition module, it performs secondary division of the finished product to be sorted in real time according to color.

[0147] When the color features of the finished product to be sorted identified in real time are not similar to the features stored in the intelligent recognition module, the color features of the finished product to be sorted in real time will be saved and the intelligent recognition module will be updated.

[0148] Example 2

[0149] Please see Figure 2 This embodiment discloses a production process parameter management system. The system can implement the methods of the above embodiments, including a historical product preparation and production data acquisition module, a finished product defect rate mapping model construction module, a current production process parameter data mapping adjustment module, a current preparation and production execution module, a finished product spectral data acquisition and processing module, a finished product image data training module, and an intelligent recognition module.

[0150] The historical product preparation and production data acquisition module collects process parameter data and corresponding finished product defect rates for several product preparation and production processes in history.

[0151] The finished product defect rate mapping model construction module constructs the final finished product defect rate mapping model based on the data collected by S1;

[0152] The current production process parameter data mapping and adjustment module collects the process parameter data corresponding to the product to be prepared and inputs it into the final finished product defect rate mapping model. If the mapping result does not meet the corresponding threshold, the process parameter data corresponding to the product to be prepared is adjusted until the threshold is met, and the current final production process parameter dataset is obtained.

[0153] The current preparation and production execution module uses the current final production process parameter dataset to prepare and produce the product to be prepared, and obtains the finished product set to be sorted.

[0154] The finished product spectral data acquisition and processing module is used to acquire the spectral data and image data of each finished product in the finished product set to be sorted, and to process the spectral data to obtain the processed spectral data.

[0155] The image data training module for finished products to be sorted is used to select and train the image data of each finished product in the set of finished products to be sorted.

[0156] The intelligent recognition module is used to intelligently recognize each finished product in the collected set of finished products to be sorted based on the image training results.

[0157] Example 3

[0158] A method for preparing a rough putty base includes managing production process parameters during the preparation of the rough putty base, wherein the management of production process parameters during the preparation of the rough putty base adopts the method of the above embodiment; including the following steps:

[0159] 1. Control the feeding ratio of raw material silos for lime powder (40-60 mesh), heavy calcium carbonate powder (200-325 mesh), and quartz sand (70-140 mesh); equip a laser particle size analyzer to monitor the particle size distribution of raw materials in real time and automatically compensate for mixing ratio deviations;

[0160] 2. The dual-shaft zero-gravity mixer is set to a speed of 280-320 rpm; dry mixing for 120 seconds → slow addition of water (solid content 18%~22%) → wet mixing for 180 seconds; the mixing process is monitored by a temperature sensor (control range 25±3℃).

[0161] 3. Online viscosity meter (Brookfield RV type) monitors slurry viscosity in real time (standard value 11000-15000 cP).

[0162] Near-infrared spectrometer scans the chemical composition of the finished product every 5 minutes; visual inspection system captures agglomeration rate (<0.3% is acceptable);

[0163] 4. Establish a process parameter-finished product defect rate mapping model to dynamically adjust parameter combinations: when the raw material moisture content fluctuates by ±1%, the water-cement ratio is automatically corrected to 0.15-0.2;

[0164] 5. The pneumatic sorting system automatically grades the products according to their density (1.8-2.1g / cm³); the moisture-proof aluminum film packaging machine adjusts the sealing temperature (150℃-180℃) according to the ambient humidity (RH45%~65%).

[0165] In the description of this specification, references to terms such as "an embodiment," "example," "specific example," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0166] The preferred embodiments of the invention disclosed above are merely illustrative of the invention. These preferred embodiments do not exhaustively describe all details, nor do they limit the invention to the specific implementations described. Clearly, many modifications and variations can be made based on the content of this specification. This specification selects and specifically describes these embodiments to better explain the principles and practical applications of the invention, thereby enabling those skilled in the art to better understand and utilize the invention.

Claims

1. A production process parameter management method characterized by, Comprise the following steps: S1, collect historical product preparation production process process parameter data and corresponding product yield rate number; S2, based on the data collected in S1, construct a final product yield rate mapping model; collect the process parameter data corresponding to the product to be prepared and input into the final product yield rate mapping model for mapping and adjusting the process parameter data of the product to be prepared; S3, using the adjustment result of S2 to prepare the product to be prepared, obtaining a set of products to be sorted; S4, collect historical product sample sets of different receiving dates; collect and save the spectral data of the samples in the sample set; process the spectral data of the samples to obtain processed spectral data; S5, collect images of the product samples in the sample set, and train and identify based on the image collection results to obtain a trained intelligent identification module; Specifically, S51, image data of the product samples in the sample set is collected, and the collected image data is segmented to obtain processed image data; S52, the processed image data is trained and identified, and the trained and identified results are saved to the intelligent identification module; S51 comprises the following steps: S511, set the number of total pixel points of the collected image data as N, divide it into D classes, for each class center point, representative point belongs to the first class, construct the objective function and the constraint condition; S512, the same type of pixels are iteratively clustered to separate the image target from the background; S52 comprises the following steps: S521, the processed image data is divided into image data blocks of the same size to obtain an image data block set; The pixel data blocks in the image data block set are encoded to obtain encoded image data blocks; S522, the encoded image data blocks are input into a convolutional neural network for feature extraction; S523, the extracted features are fused to obtain fused image data block features; three groups of channels are fused by full connection operation to obtain fused image data blocks; S524, the fused image data block features are saved to the intelligent identification module and used as a reference; S6, after training, the processed spectral data in S4 is set as a reference spectrum, and the product set to be sorted is divided based on the reference spectrum, and the division result includes: good product and defective product; S7, after the division, the divided product set to be sorted is input into the intelligent identification module for identification, and the intelligent identification module is used for sorting control.

2. The production process parameter management method according to Claim 1, characterized by, S1 comprises the following steps: S11, obtain the product to be prepared and the corresponding process parameters of the product to be prepared during the production process, to obtain a current production process parameter type set; S12, according to the product to be prepared and the current production process parameter type set, collect historical process parameter data and corresponding product yield rate data of products of the same type as the product to be prepared during the preparation production process, to obtain a historical production process parameter data set and a historical product yield rate data set.

3. The production process parameter management method according to claim 2, characterized by, S2 comprises the following steps: S21, constructing a mapping model between production process parameters and corresponding finished product and defective product rate data of a product according to the historical production process parameter data set and the historical finished product and defective product rate data set, to obtain a final finished product and defective product rate mapping model; S22, setting a current finished product and defective product rate threshold; obtaining various preset process parameter data before preparing and producing the product to be prepared according to the current production process parameter type set, to obtain a current production process parameter data set; S23, inputting the current production process parameter data set into the final finished product and defective product rate mapping model for mapping, to obtain current initial finished product and defective product rate data; If the current initial finished product and defective product rate data is greater than or equal to the current finished product and defective product rate threshold, repeatedly adjusting the current production process parameter data set; otherwise, no adjustment is needed; taking the current production process parameter data set as a current final production process parameter data set; Inputting the adjusted current production process parameter data set into the final finished product and defective product rate mapping model again for mapping, to obtain current adjustment finished product and defective product rate data; until the current adjustment finished product and defective product rate data is less than the current finished product and defective product rate threshold, to obtain a current final production process parameter data set.

4. The production process parameter management method according to claim 3, characterized by, The S3 comprises the following steps: S31, using the current final production process parameter data set to prepare and produce the product to be prepared; after the preparation and production is completed, obtaining a finished product set to be sorted.

5. The production process parameter management method according to claim 4, characterized by, The S4 comprises the following steps of collecting and saving the spectral data of the samples in the sample set, and processing the spectral data of the samples to obtain processed spectral data: S41, collecting spectral data of the historical product finished sample in the sample set based on near-infrared spectral detection technology; placing the historical product finished sample of different dates on the infrared collection window of the instrument, setting the infrared scanning parameters of the instrument, and installing a spectral detector around the object to be scanned; S42, quantitatively analyzing the different absorbances of the historical product finished sample of different dates in different wave bands, and constructing near-infrared spectral data according to the light absorption formula of the material, repeatedly collecting different parts of each historical product finished sample through multiple scans, and averaging the multiple scan results; S43, performing averaging processing on the near-infrared spectral data to obtain averaged spectral data; S44, performing noise reduction on the averaged spectral data; after the noise reduction is completed, performing clustering on the noise-reduced spectral data.

6. The production process parameter management method according to Claim 5, characterized by, The S44 comprises the following steps: S441, eliminating the noise in the averaged spectral data by setting a convolution smoothing window; setting a smoothing window with a width of 2w+1, replacing the measurement value of the center wavelength point j with the average value of the measurement values of the center wavelength point j and the previous and subsequent w points in the window, and moving j from left to right until all points are smoothed; obtaining noise-reduced spectral data; S442, respectively select a group of good products and substandard products corresponding to the denoised spectral data in the sample set as the clustering center, each distance from the center represents a type; calculate the distance between other denoised spectral data in the sample set and the clustering center, and add the type represented by the nearest clustering center; Calculate the denoised spectral data in the type, update the clustering center spectral data; re-cluster according to the updated clustering center spectral data, and iterate until the type converges, and output the clustered spectral data; Set the clustered spectral data as the processed spectral data.

7. The production process parameter management method according to Claim 1, characterized by, The iterative clustering operation on the same type of pixels in S512 includes: based on the initially set center point of each class, calculating the distance from the random point in the class to the class center point; constantly adjust the center point of each class according to the constraint condition, and constantly iterate until the point in all classes no longer changes, and the iteration stops; By comparing the distance in J, when the distance in set J no longer changes, the separation of the image target and the background is realized; and the separated image is set as the obtained processed image data; The S521 specifically includes: setting the encoding of the pixel data block in the image data block set as ; respectively represent the pixel values of red, green and blue colors of each pixel in the image data block, the encoding of the pixel data in the image data block; obtaining the encoded image data block; S522 specifically includes: based on the pixel data in the encoded image data block, the convolutional neural network is divided into channels, and three groups of channels are set to extract features of red, green and blue pixel values in the image data block.

8. The production process parameter management method according to Claim 7, characterized by, S7 includes the following steps: S71, when the intelligent identification module receives the divided product set to be sorted, the color features of the product to be sorted are identified in real time, and the product to be sorted received in real time is divided according to color based on the features saved by the intelligent identification module; When the color features of the product to be sorted are not similar to the features saved by the intelligent identification module, the color features of the product to be sorted are saved and the intelligent identification module is updated.

9. A production process parameter management system characterized by: The method for realizing the production process parameter management method as claimed in any one of claims 1-8 includes a historical product preparation production data acquisition module, a finished product substandard rate mapping model construction module, a current production process parameter data mapping adjustment module, a current preparation production execution module, a product to be sorted product spectral data acquisition and processing module, a product to be sorted product image data training module, and an intelligent identification module; The historical product preparation production data acquisition module acquires process parameter data and corresponding finished product substandard rate data of several product preparation production processes in history; The finished product substandard rate mapping model construction module constructs a final finished product substandard rate mapping model based on the data collected in S1; The current production process parameter data mapping adjustment module acquires the process parameter data corresponding to the product to be prepared and inputs it into the final finished product substandard rate mapping model for mapping; if the mapping result does not meet the corresponding threshold, the process parameter data corresponding to the product to be prepared is adjusted until the threshold is met, and the current final production process parameter data set is obtained; The current preparation production execution module uses the current final production process parameter data set to prepare and produce the product to be prepared, and obtains a product set to be sorted; The to-be-sorted product finished product spectrum data acquisition and processing module is configured to acquire spectrum data and image data of each product in the to-be-sorted product finished product set, and process the spectrum data to obtain processed spectrum data. The to-be-sorted product finished product image data training module is configured to select and train the image data of each product in the to-be-sorted product finished product set. The intelligent identification module is configured to intelligently identify each product in the to-be-sorted product finished product set according to the image training result.

10. A method for preparing a putty base, comprising management of the production process parameters during the preparation of the putty base, characterized in that: The production process parameter management in the process of preparing the stone putty coarse bottom adopts the production process parameter management method according to any one of claims 1-8.

Images