Visual identification and adjustment method for modeling parameters of filter tip

Abstract

The invention discloses a filter tip modeling parameter visual identification and adjustment method, and relates to the technical field of filter tip production, a laser diameter measuring instrument and a linear array camera are used for synchronously collecting peripheral images of a filter tip, and high-precision and consistent size and gray scale double baselines are generated; a texture segmentation technology is adopted to identify a suspected recess area and generate a recess position vector, a confocal displacement sensor is utilized to carry out local depth walk-up to obtain depth distribution data, and a size compensation coefficient is generated and a size matrix is corrected through Gaussian process regression; statistical threshold analysis is conducted on the corrected size matrix, when parameters exceed the limit, the gap of a forming nozzle is adjusted, the size stability of the filter tip is ensured, the adjustment result and related data are written into a manufacturing knowledge base, follow-up batch optimization is supported, and the measurement reliability and the production line efficiency are improved; tiny recesses on the surface of the filter tip can be effectively identified and compensated, and the measurement accuracy and the assembly stability of modeling parameters are ensured.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of filter production, in particular to a filter shaping parameter visual identification and adjustment method. BACKGROUND

[0002] In the high-speed production process of modern cigarette filters, shaping parameters such as outer diameter and roundness need to be continuously monitored to ensure consistent and stable subsequent assembly. Most of the existing online detection devices use laser scanning or machine vision to give real-time readings. When the filter is subjected to mechanical vibration, fiber static adsorption and particle impact during rolling and handling, fine pits or depressions often form on the surface. Such depressions change the local light scattering and absorption characteristics, causing fluctuations in the reflection signal under conventional lighting. Single-line laser or single-view camera usually averages abnormal reflection and normal texture together, and the depression is thus hidden, and the system often underestimates the true size and does not give an alarm. At this time, the seemingly qualified batch has hidden size drift risks before entering the downstream process.

[0003] The present application focuses on the core problem of hidden interference caused by micro-pits on filter shaping parameter measurement.

[0004] In the rolling area, the surface layer of the fiber loses support after being subjected to instantaneous pulling or foreign object impact, resulting in a barely visible depression with the naked eye. This explains why the defect occurs. As the filter rotates at high speed, the scattering difference between the depression area and the surrounding smooth surface is obvious, but the traditional measurement method averages the abnormal signal with the normal signal, explaining how the error is formed. The output outer diameter and roundness are low, and the process control logic considers this phenomenon as normal fluctuation and does not adjust, resulting in loose assembly, tight packaging, and other problems, showing the direct impact of the defect. The depression area is structurally weak and is prone to further cracking during transportation and storage, causing increased resistance spread and even end face cracking, which is the extended consequence of the defect. Therefore, an identification and adjustment method is needed that can identify depressions in a visual interface and compensate for shaping measurements in real time to ensure accurate execution of automatic adjustment and maintain product consistency, which is the key problem that the present application intends to solve. SUMMARY

[0005] (I) Technical problems solved In view of the deficiencies of the prior art, the filter modeling parameter visual identification adjustment method provided by the present application synchronously collects filter outer peripheral images by a laser diameter measuring instrument and a linear array camera to generate high-precision and consistent size and grayscale double baselines; adopts a texture segmentation technique to identify suspected recessed areas and generate recessed position vectors, uses a confocal displacement sensor to perform local depth scanning to obtain depth distribution data, and generates a size compensation coefficient by Gaussian process regression to correct a size matrix; performs statistical threshold analysis on the corrected size matrix, adjusts the forming mouth gap when the parameters exceed the limit, ensures the filter size stability, and writes the adjustment results and related data into a manufacturing knowledge base; the filter surface micro recesses can be effectively identified and compensated, and the technical problems recorded in the background art are solved.

[0006] (II) Technical solutions To achieve the above object, the present application is implemented by the following technical solutions: the filter modeling parameter visual identification adjustment method comprises the following steps: synchronously collecting filter outer peripheral images and radial profile data by a laser diameter measuring instrument and a linear array camera to generate size and grayscale double baselines; A texture segmentation method combining local binary pattern feature extraction and grayscale difference analysis is adopted to divide the radial profile data stream into smooth areas and suspected recessed areas, and generate recessed position vectors to provide target areas for depth scanning; A confocal displacement sensor is used to perform local scanning on the suspected recessed areas to obtain depth distribution data, quantize radial gradient fluctuation and local depth fluctuation signal-to-noise ratio, and generate a size compensation coefficient by Gaussian process regression to correct a size matrix; Statistical threshold analysis of the mean and coefficient of variation of the corrected size matrix is performed, and if the parameters exceed the limit, a forming mouth gap fine adjustment instruction is sent and a synchronous prompt is given on the operation interface; The adjustment results, original images, depth distribution data and size compensation coefficient are written into a manufacturing knowledge base.

[0007] Further, a laser diameter measuring instrument and a linear array camera are arranged at the cigarette making outlet, synchronous acquisition of the laser diameter measuring instrument and the linear array camera is realized by high-precision clock signals or hardware triggers, the laser diameter measuring instrument measures the filter outer diameter size to generate a radial profile data stream, and the linear array camera collects filter surface grayscale images to generate a grayscale image data stream; The radial profile data stream is subjected to sliding window average processing to generate a size baseline, and the average grayscale value of each time point of the grayscale image data stream is calculated to generate a grayscale baseline.

[0008] Further, the radial profile data stream and the grayscale image data stream are subjected to Gaussian filtering to filter out noise, and data alignment of the radial profile data stream and the grayscale image data stream is realized by timestamp matching and rotation angle calibration; The local binary pattern feature is extracted from the grayscale image data stream to represent the texture, and the gray difference of adjacent pixels is calculated and marked as a suspected concave area according to a preset threshold.

[0009] Further, the local binary pattern feature is extracted from the grayscale image data stream, wherein a binary code is generated by comparing the gray value of each pixel with its eight surrounding neighborhood pixels to represent the texture feature; The gray difference of adjacent pixels in the grayscale image is calculated and compared with the preset threshold to mark the suspected concave area; Based on the local binary pattern feature value and the pre-determined smooth area feature value range, the pixels falling within the range are classified as smooth areas, and the pixels not falling within the range are classified as suspected concave areas, and the suspected concave area pixels are clustered.

[0010] Further, according to the concave position vector, a local search is performed on the suspected concave area using a confocal displacement sensor to collect depth distribution data; based on the depth distribution data, the gradient along the radial angle is calculated, the radial gradient fluctuation is quantified by the root mean square value, and the dimensionless radial gradient fluctuation feature is generated by dividing it by the reference value.

[0011] Further, the difference between the maximum value and the minimum value of the depth distribution data in the suspected concave area is calculated to determine the local depth fluctuation amplitude, and then the signal-to-noise ratio is obtained by dividing it by the sensor measurement noise level; The radial gradient fluctuation feature and the signal-to-noise ratio are taken as inputs, and a Gaussian process regression is applied to calculate a size compensation coefficient; the size compensation coefficient is applied to the radial profile data of the suspected concave area to generate a corrected size matrix.

[0012] Further, the arithmetic mean and the coefficient of variation are calculated from the corrected size matrix; the arithmetic mean is compared with the preset ideal mean and the allowable deviation value, and the coefficient of variation is compared with the preset upper limit of the coefficient of variation; When the absolute value of the difference between the arithmetic mean and the ideal mean exceeds the allowable deviation value or the coefficient of variation exceeds the upper limit of the coefficient of variation, it is determined to be out of limit.

[0013] Further, for the out-of-limit case, the difference between the arithmetic mean and the ideal mean is calculated as the mean deviation, and the forming die gap adjustment amount is calculated according to the mean deviation and the pre-calibrated adjustment coefficient; The forming die gap adjustment amount is sent to the forming die control unit to perform fine adjustment, and the out-of-limit warning and adjustment suggestion are displayed on the operation interface.

[0014] Further, the forming nozzle gap adjustment amount, the adjusted filter size statistical index, the gray image data stream, the depth distribution data and the size compensation coefficient are bound with the production batch number, the production time and the equipment number to form a structured data record; the gray image data stream and the depth distribution data are compressed and stored after compression processing; The data records are written into the batch information table, the adjustment record table, the image data table, the depth data table and the compensation coefficient table of the manufacturing knowledge base in the order of production batch number and production time, and a transaction processing mechanism is used to ensure synchronous updating of the data.

[0015] Further, before new batch production, historical data is retrieved from the manufacturing knowledge base to match the adjustment results and compensation coefficients under similar production conditions; key factors affecting filter size are identified by cluster analysis, quality trends are monitored by time series analysis, and adjustment thresholds are optimized by regression analysis; The gray image data stream, the depth distribution data, the size compensation coefficient and the forming nozzle gap adjustment amount are connected in series according to the production batch number to form an information chain.

[0016] (Three) beneficial effects The present application provides a filter modeling parameter visual identification adjustment method, which has the following beneficial effects: The laser diameter measuring instrument and the line array camera are used to synchronously collect the filter outer peripheral image, generate the radial profile data stream and the gray image data stream, and establish a size and gray double baseline, which not only significantly improves the data collection efficiency, but also lays a solid data foundation for subsequent recess identification and size compensation. Through the sliding window average method and the Gaussian filter processing, random noise caused by mechanical vibration and environmental light is filtered out, ensuring the high precision and consistency of the radial profile data stream and the gray image data stream.

[0017] A texture segmentation method driven by scattering difference is used to divide the radial profile data stream into smooth areas and suspected recess areas, and output the recess position vector. By extracting the local binary pattern feature and analyzing the gray difference, the small recess areas on the filter surface are accurately identified, and the data processing amount of irrelevant areas is reduced, improving the response speed.

[0018] The confocal displacement sensor performs local scanning according to the recess position vector, acquires the depth distribution data, and generates the size compensation coefficient by quantifying the radial gradient fluctuation and the local depth fluctuation signal-to-noise ratio, and combining the Gaussian process regression to correct the current size matrix. Through high-precision depth measurement and scientific quantification means, the interference of the recess area on the size measurement is effectively eliminated, ensuring the accuracy of the size matrix.

[0019] The statistical threshold analysis is performed on the corrected size matrix to determine whether the filter size parameters are within an acceptable range. If the parameters exceed the limit, a forming filter gap fine-tuning instruction is sent and a synchronous prompt interface is provided to ensure the stability and consistency of subsequent assembly. Through real-time adjustment and visual prompt, the transparency and controllability of the production process are significantly improved, and human error is reduced.

[0020] The adjustment results, original images, depth distribution data, and size compensation coefficients are written into a manufacturing knowledge base to form an information chain, providing data reference and analysis support for subsequent batches. Through cluster analysis, time series analysis, and regression analysis, the method realizes intelligent prediction and optimization of the production process, and accumulates reusable data assets for flexible production. BRIEF DESCRIPTION OF DRAWINGS

[0021] Figure 1 A flowchart of the filter modeling parameter visual recognition adjustment method of the present application. DETAILED DESCRIPTION

[0022] The technical solutions in the embodiments of the present application will be clearly and completely described 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, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the present application.

[0023] Please refer to Figure 1 The present application provides a filter modeling parameter visual recognition adjustment method, which comprises, Step one, a laser diameter measuring instrument and a line array camera are arranged at the filter wrapping outlet, synchronous acquisition of the laser diameter measuring instrument and the line array camera is realized through a high-precision clock signal or a hardware trigger, the laser diameter measuring instrument measures the outer diameter size of the filter to generate a radial profile data stream, and the line array camera acquires a filter surface gray-scale image to generate a gray-scale image data stream; the radial profile data stream is subjected to sliding window average processing to generate a size baseline, and the gray-scale image data stream is subjected to calculation of the average gray-scale value at each time point to generate a gray-scale baseline; the radial profile data stream and the gray-scale image data stream are subjected to Gaussian filter processing to filter out noise, and data alignment of the radial profile data stream and the gray-scale image data stream is realized through timestamp matching and rotation angle calibration; local binary pattern features are extracted from the gray-scale image data stream to represent texture, and the gray-scale difference value of adjacent pixels is calculated and a suspected concave area is marked according to a preset threshold.

[0024] The step one comprises the following contents: Step 101, device configuration and synchronous acquisition A laser diameter measuring instrument and a line array camera are arranged at the filter wrapping outlet, and synchronous acquisition of the two in time and space is realized through a high-precision clock signal or a hardware trigger.

[0025] The laser diameter gauge continuously measures the outer diameter size of the filter, generating a radial profile data stream representing the diameter value of the filter at each time point; the line array camera continuously captures the gray scale image of the filter surface, recording the texture features and light scattering characteristics of the filter surface. The implementation of synchronous acquisition is as follows: taking a unified time signal as the reference, the laser diameter gauge and the line array camera perform data acquisition on the same filter position at the same time, and through the accurate calibration of the installation position of the equipment, the spatial correspondence of the measurement of the two is ensured. The size and surface characteristics of the filter need to be obtained under consistent time and space conditions to ensure the matching of the data in subsequent analysis, improve the reliability and consistency of the data, and avoid analysis errors caused by time or space deviation.

[0026] Step 102, data stream generation The laser diameter gauge generates a radial profile data stream according to the measurement results, specifically a sequence of outer diameter size values of the filter at each time point, representing the change of the filter diameter over time.

[0027] The line array camera generates a gray scale image data stream, specifically a sequence of gray scale values of the filter at each time point and each circumferential angle position, representing the change of the filter surface brightness over time and angle. The process of data stream generation is as follows: the laser diameter gauge scans the filter surface through the laser beam and records the diameter value calculated from the reflected signal; the line array camera collects the intensity of the reflected light from the filter surface through the optical imaging system and converts it into a gray scale value. The radial profile data stream reflects the size characteristics of the filter, and the gray scale image data stream reflects the surface texture characteristics, and the combination of the two can fully characterize the quality state of the filter, providing the original data required for subsequent processing and ensuring the comprehensiveness and accuracy of the analysis.

[0028] Step 103, double baseline establishment The radial profile data stream is processed by sliding window averaging, specifically: selecting the radial profile data within a certain time period, calculating the average value of all outer diameter size values within this time period as the average outer diameter size at this time point, and generating a continuous sequence of average outer diameter sizes by continuously moving the time window, called the size baseline. The size of the sliding window is determined according to the production speed of the filter and the data acquisition frequency to cover enough data points to achieve smoothing effect. The outer diameter size of the filter may change due to short-term fluctuations in the production process, and the average processing can eliminate these random disturbances and provide stable size reference values. The size baseline can reflect the long-term trend of the filter size, facilitating subsequent anomaly detection.

[0029] The gray scale image data stream is processed, specifically: at each time point, the gray scale values of all circumferential angle positions on the filter surface are collected, the average of these gray scale values is calculated as the average gray scale value at that time point. This process is repeated to form a continuous average gray scale value sequence, called the gray scale baseline, the light scattering characteristics of the filter surface are affected by the texture and material, the average gray scale value can represent the overall reflection characteristics of the surface, reducing the influence of local noise, the gray scale baseline provides a stable reference for the surface characteristics, which is helpful for subsequent texture analysis and identification of abnormal areas.

[0030] Step 104, data preprocessing The radial profile data stream and the gray scale image data stream are respectively subjected to Gaussian filtering processing. The processing process is: for each data point, combine the values of several data points before and after it, and perform weighted average according to the weight of Gaussian distribution to generate a new smooth data sequence. The Gaussian distribution weight is determined based on the distance between the data point and the center point, the closer the distance, the higher the weight. Random noise may be introduced by mechanical vibration or environmental light changes in the production environment, and Gaussian filtering can weaken these disturbances through smoothing processing, improve the signal-to-noise ratio of the data, and make the subsequent analysis results more accurate.

[0031] Through timestamp matching and rotation angle calibration, the radial profile data stream and the gray scale image data stream are aligned. The timestamp matching process is: comparing the time marks of the two groups of data, adjusting the correspondence of the sampling points to make them consistent in time; the rotation angle calibration process is: according to the geometric relationship of the circumferential position of the filter, adjusting the circumferential angle sequence of the gray scale image data stream to correspond to the measurement position of the radial profile data stream, and synchronously collecting may exist slight deviation, the alignment processing ensures that the two groups of data are completely matched in time and space, guarantees the accuracy of data fusion, and improves the reliability of analysis.

[0032] Step 105, abnormality detection preparation Local binary pattern features are extracted from the gray scale image data stream, specifically: for each gray scale value pixel, compare its gray scale with that of the surrounding neighborhood pixels, if the neighborhood pixel gray scale value is greater than the center pixel, then it is recorded as 1, otherwise it is recorded as 0, all comparison results are sequentially combined into binary code, and then converted into decimal number as the feature value of the pixel, the subtle changes of the filter surface texture affect the gray scale distribution, the local binary pattern can capture these local differences, generate texture feature data, and provide a method for quantifying texture information, which is convenient for subsequent identification of abnormal areas.

[0033] The difference value of the gray scale value of the adjacent pixels in the gray scale image data stream is calculated, specifically: for each pixel, the gray scale value of its adjacent pixel is subtracted to obtain a difference value sequence; if the difference value exceeds a pre-set threshold value, the corresponding region is marked as a suspected recessed area. The threshold value is determined according to the light scattering characteristic experiment of the normal surface of the filter rod. The recessed area changes the light scattering characteristic, which is usually manifested as a significant difference in the gray scale value. The potential anomaly is located by using this characteristic, and the suspected recessed area is quickly screened out, thereby improving the pertinence and efficiency of the subsequent processing.

[0034] Through the above steps, step one completes the generation of the filter rod radial profile data stream and the gray scale image data stream, and establishes the size baseline and the gray scale baseline; the data preprocessing ensures the high precision and consistency of the data, and the anomaly detection preparation provides the marking information of the suspected recessed area through the texture feature extraction and the scattering difference analysis. These processes can jointly ensure the accuracy and reliability of the subsequent recessed recognition and size compensation.

[0035] Step two, extracting the local binary pattern feature from the gray scale image data stream, specifically generating a binary code to represent the texture feature by comparing the gray scale value of each pixel with that of its eight surrounding neighborhood pixels; calculating the gray scale difference value of the adjacent pixels in the gray scale image and comparing it with a pre-set threshold value to mark the suspected recessed area; classifying based on the local binary pattern feature value and the pre-determined smooth area feature value range, and classifying the pixels falling within the range as smooth areas and the pixels not falling within the range as suspected recessed areas; clustering the suspected recessed area pixels; The step two includes the following contents: Step 201, texture feature extraction The local binary pattern feature is extracted from the gray scale image data stream, and the specific process is as follows: for each pixel in the gray scale image, eight neighborhood pixels around the center pixel are selected. The gray scale value of the center pixel is compared with the gray scale value of each neighborhood pixel one by one. If the gray scale value of a neighborhood pixel is greater than or equal to that of the center pixel, the comparison result is recorded as 1; if the gray scale value of a neighborhood pixel is less than that of the center pixel, the comparison result is recorded as 0. In this way, eight binary bits are obtained. The eight binary bits are arranged in a predetermined order and converted into a decimal number. The decimal number is the local binary pattern feature value of the pixel.

[0036] The local binary pattern feature value reflects the gray level change pattern of the local texture on the filter surface, and can highlight the significant difference in texture features between the smooth area and the recessed area, providing a reliable basis for subsequent area classification. The selection of the local binary pattern feature has strong robustness to changes in lighting conditions, and at the same time shows high sensitivity to local texture details, which can effectively distinguish the normal texture area and the abnormal recessed area on the filter surface, improve the accuracy and stability of the texture analysis, and ensure that the subsequent processing steps can accurately identify different areas.

[0037] Step 202, scattering difference analysis The gray difference value of the adjacent pixels in the gray image data stream is calculated, and the specific process is as follows: for each pixel in the gray image, the difference value between the gray value of the pixel and the gray value of the next pixel along the radial angle direction is calculated, and the absolute value of the difference value is taken to obtain the gray difference value.

[0038] The gray difference value is compared with the pre-set threshold value. If the gray difference value is greater than the threshold value, it is determined that the area may have a recess; if the gray difference value is less than or equal to the threshold value, it is determined that the area belongs to the normal texture area. The determination of the threshold value is based on the experimental determination of the normal surface light scattering characteristics of the filter, and by analyzing the scattering difference characteristics of the normal area and the recessed area, a value is set which can effectively distinguish the two. The recessed area usually causes a significant difference in gray value due to the change in light scattering characteristics, and this physical characteristic is used to quickly screen out suspected recessed areas. The efficiency and pertinence of the abnormal area detection are improved, so that the subsequent processing can focus on the key areas.

[0039] Step 203, texture segmentation Based on the local binary pattern feature value, the candidate area is classified, and the specific process is as follows: first, the feature value range of the smooth area is determined by statistical analysis of the local binary pattern feature value of the normal filter surface sample. For each pixel in the gray image, check whether its local binary pattern feature value falls within the feature value range of the smooth area. If it falls within the range, the pixel is classified as a smooth area; if it does not fall within the range, the pixel is classified as a suspected recessed area. Further clustering processing is performed on the pixels classified as suspected recessed areas, wherein a connectivity-based clustering technique is used to merge the suspected recessed pixels adjacent to each other into independent suspected recessed areas. The local binary pattern feature value can accurately represent the texture difference, and through the classification and clustering steps, the smooth area and the suspected recessed area are clearly divided, improving the accuracy and integrity of the area recognition, and providing a reliable area division result for the subsequent positioning of the recessed area.

[0040] Step 204, recessed position vector generation For each suspected concave area, the center position and the range are calculated, and the specific process is as follows: for all pixels in each suspected concave area, the average value of the axial position coordinate and the average value of the radial angle coordinate of the pixels are calculated respectively, and the two average values are taken as the center position of the suspected concave area. At the same time, the minimum value and the maximum value of the axial position coordinate and the minimum value and the maximum value of the radial angle coordinate of all pixels in the suspected concave area are recorded, and the minimum value and the maximum value are taken as the range of the suspected concave area. The center position and the range information of all suspected concave areas are arranged into a data set, which is called a concave position vector. The reason for generating the concave position vector is that this data set can provide accurate target area information for subsequent local depth scanning, avoid full scanning of the entire image, and thus optimize the detection process. The purpose and efficiency of obtaining the depth distribution data are ensured, and the practicability of the overall process is improved.

[0041] Through the above steps, the smooth area and the suspected concave area are divided from the filter surface radial profile data stream, and the concave position vector is generated. By fully utilizing the differences in texture features and light scattering characteristics, the accurate identification and positioning of the concave area are completed, and accurate target area information and data support are provided for the subsequent acquisition of depth distribution data and size compensation.

[0042] Step three, according to the concave position vector, the suspected concave area is locally scanned by using the confocal displacement sensor, and the depth distribution data is collected; based on the depth distribution data, the gradient along the radial angle is calculated, the radial gradient fluctuation is quantified by the root mean square value, and it is divided by the reference value to generate the dimensionless radial gradient fluctuation feature; the difference between the maximum value and the minimum value of the depth distribution data in the suspected concave area is calculated to determine the local depth fluctuation amplitude, and then it is divided by the sensor measurement noise level to obtain the signal-to-noise ratio; the radial gradient fluctuation feature and the signal-to-noise ratio are taken as the input, and the Gaussian process regression is applied to calculate the size compensation coefficient; the size compensation coefficient is applied to the radial profile data of the suspected concave area to generate the corrected size matrix.

[0043] The step three includes the following contents: Step 301, local scanning and depth distribution data acquisition According to the recess position vector provided in step two, each suspected recess area is locally searched. The specific process is: using a confocal displacement sensor to scan in the axial position range and radial angle range of the suspected recess area, recording the depth information of the filter surface, and generating depth distribution data. The depth distribution data represents the depth value of the filter surface at each axial position and radial angle. The confocal displacement sensor has sub-micron resolution, ensuring the accuracy of the depth data sufficient to capture the subtle changes of the recess area. By accurately measuring the suspected recess area, the need for comprehensive scanning of the entire filter surface is avoided, thereby reducing data processing and improving detection efficiency. The depth distribution data can directly reflect the true depth characteristics of the recess area, providing high-precision raw data support for subsequent analysis.

[0044] Step 302, quantifying radial gradient fluctuation Radial gradient calculation is performed on the depth distribution data. The specific process is: for each axial position and radial angle, the rate of change of the depth value along the radial angle direction is calculated to obtain the radial gradient value. Then, the root mean square value is calculated for the radial gradient fluctuation amplitude of each axial position, where the square sum of the radial gradient values for all radial angles of the axial position is calculated, then divided by the number of radial angle sampling points, and finally the square root is taken to obtain the root mean square value. Divide this root mean square value by the radial gradient fluctuation reference value of the normal filter surface determined in advance to obtain the dimensionless radial gradient fluctuation feature. The depth change of the recess area is more dramatic, which will cause the radial gradient fluctuation amplitude to increase. This feature can effectively quantify the impact of the recess on the depth distribution, and the radial gradient fluctuation feature provides a quantitative indicator of the degree of depth change of the recess area, facilitating subsequent feature analysis and processing.

[0045] Step 303, quantifying local depth fluctuation signal-to-noise ratio For each suspected recess area, the difference between the maximum and minimum values in its depth distribution data is calculated to obtain the local depth fluctuation amplitude; divide this local depth fluctuation amplitude by the measurement noise level of the confocal displacement sensor to obtain the signal-to-noise ratio. The measurement noise level is determined by experiment, that is, the output fluctuation of the sensor is recorded and the standard value is calculated when there is no signal input. Quantifying the local depth fluctuation signal-to-noise ratio, the signal-to-noise ratio can reflect the significance of the depth change of the recess area, thereby distinguishing the real recess feature from the influence of measurement noise. The local depth fluctuation signal-to-noise ratio provides a reliability indicator for the depth feature of the recess area, ensuring that subsequent analysis is based on real and reliable data.

[0046] Step 304, generating size compensation coefficient by Gaussian process regression The radial gradient fluctuation feature and the local depth fluctuation signal-to-noise ratio are taken as input features, and a Gaussian process regression is used to establish a mapping relationship between the features and the size compensation coefficient. The Gaussian process regression is a non-parametric regression method based on Bayesian theory, which can effectively capture the nonlinear relationship between the input features and the target variable.

[0047] The training process is as follows: using filter sample data with known recess degree and corresponding size error, the hyperparameters of the Gaussian process regression model are optimized to balance the fitting accuracy of the model to the training data and the prediction ability of the new data. For each suspected recess area of the current filter, the trained Gaussian process regression model is used to predict the corresponding size compensation coefficient. The Gaussian process regression is suitable for small sample data processing and can provide uncertainty estimation of the prediction result, which is suitable for describing the complex relationship between the recess feature and the size error, and the generated size compensation coefficient can accurately reflect the influence of the recess on the size measurement, improving the compensation accuracy.

[0048] Step 305, correcting the size matrix The predicted size compensation coefficient is applied to the radial profile data of the suspected recess area.

[0049] The specific process is as follows: for each suspected recess area, the size compensation coefficient of the area is added to the corresponding radial profile data to obtain the corrected size data. After completing the compensation for all suspected recess areas, a corrected size matrix is generated. The depth change of the recess area will cause the radial profile data to be low, and by adding the size compensation coefficient, the actual size of the filter can be restored. The corrected size matrix can more accurately reflect the true size characteristics of the filter, providing reliable data support for subsequent process adjustment.

[0050] Through the above steps, step three realizes the acquisition and analysis of the depth distribution data of the suspected recess area of the filter. By quantifying the radial gradient fluctuation and the local depth fluctuation signal-to-noise ratio, and combining the Gaussian process regression to generate the size compensation coefficient, the correction of the size matrix is finally completed. The interference of the recess area on the size measurement can be effectively eliminated, and the measurement accuracy of the filter modeling parameters is ensured. Each sub-step is closely connected, the technical logic is rigorous and coherent, and the accurate compensation of the recess influence is realized together.

[0051] Step four, calculate the arithmetic mean and the coefficient of variation from the revised size matrix obtained in step three; compare the arithmetic mean with the preset ideal average value and the allowable deviation value, and compare the coefficient of variation with the preset upper limit of the coefficient of variation; when the absolute value of the difference between the arithmetic mean and the ideal average value exceeds the allowable deviation value or the coefficient of variation exceeds the upper limit of the coefficient of variation, it is determined to be out of limit; for the out-of-limit case, calculate the difference between the arithmetic mean and the ideal average value as the average deviation, and calculate the forming nozzle gap adjustment amount according to the average deviation and the pre-calibrated adjustment coefficient; send the forming nozzle gap adjustment amount to the forming nozzle control unit to perform fine adjustment, and display the out-of-limit warning and adjustment suggestion on the operation interface.

[0052] The step four includes the following contents: Step 401, threshold value analysis The revised size matrix obtained in step three contains the revised outer diameter size data of the filter at different axial positions, and the axial position is represented by the serial number of the sampling point, and the total number of sampling points is a predetermined fixed value.

[0053] First, calculate the arithmetic mean of all the revised outer diameter size data in the revised size matrix. The specific method is to add all the revised outer diameter size data, and then divide by the total number of sampling points to obtain the average value, which is used to evaluate the overall level of the filter size. Then, the coefficient of variation of the revised size matrix is calculated to evaluate the consistency of the filter size. The calculation process is as follows: first, calculate the difference between each revised outer diameter size data and the arithmetic mean, square the differences and sum them up, and then divide by the total number of sampling points to obtain the variance; then, take the square root of the variance to obtain the standard deviation; finally, divide the standard deviation by the arithmetic mean to obtain the coefficient of variation.

[0054] The arithmetic mean value reflects the central tendency of filter size, which is convenient for judging whether the overall size meets the production standard; the coefficient of variation, as the ratio of standard deviation to mean value, can quantify the relative dispersion degree of size data, which is suitable for evaluating consistency. The arithmetic mean value is convenient for quickly judging the overall eligibility of size, and the coefficient of variation provides a dimensionless index, which is convenient for comparison between different batches. According to the product specification, the ideal average value of the filter outer diameter is determined, and the allowed deviation value is set according to the process requirement. If the absolute value of the difference between the arithmetic mean value of the corrected size matrix and the ideal average value is greater than the allowed deviation value, it is determined that the average value is out of limit. According to the production consistency requirement, the upper limit of the coefficient of variation is set, and if the coefficient of variation of the corrected size matrix is greater than the upper limit, it is determined that the variability is out of limit. If either the average value is out of limit or the variability is out of limit, the corrected size matrix is out of the limited range; if neither of them is out of limit, the corrected size matrix is within the limited range. The average value control ensures that the overall filter size meets the specification, and the coefficient of variation control ensures the consistency by controlling the size fluctuation range. The dual-index analysis improves the comprehensiveness of size control, ensuring the stability and reliability of product quality.

[0055] Step 402, sending a forming nozzle gap fine-tuning instruction If the arithmetic mean value of the corrected size matrix is out of limit, calculate the average value deviation, which is the difference between the arithmetic mean value of the corrected size matrix and the ideal average value. According to this average value deviation, calculate the adjustment amount of the forming nozzle gap, which is to multiply the average value deviation by an adjustment coefficient calibrated by experiment. This coefficient reflects the linear relationship between filter size change and forming nozzle gap. If the coefficient of variation is out of limit, analyze the uniformity or stability of the forming nozzle in combination with the production process to determine whether additional gap uniformity adjustment is needed; if so, superimpose this adjustment with the adjustment amount based on the average value deviation. The forming nozzle gap directly affects the filter forming size, and the adjustment can correct the average value deviation; the variability out of limit may be caused by the non-uniformity of the forming nozzle, which needs to be adjusted specifically; the adjustment based on the average value deviation realizes real-time correction of size, and the superimposed uniformity adjustment improves consistency, ensuring the stability of the production process.

[0056] The calculated forming nozzle gap adjustment amount is sent to the forming nozzle control unit, and the gap fine-tuning operation is performed by the control unit. This process realizes dynamic optimization of filter size through precise adjustment. Among them, the adjustment amount is converted into control instruction to ensure the executability of the adjustment operation; a closed-loop control mechanism is formed to improve the accuracy of filter size parameters and production efficiency.

[0057] Step 403, synchronously prompting the operation interface The operation interface displays the over-limit warning, including the specific indicators and degree of over-limit, for example, if the average value is over-limit, the value of the average deviation is displayed; if the variability is over-limit, the value of the coefficient of variation is displayed. At the same time, the adjustment amount of the forming nozzle gap and the expected effect after adjustment are displayed, for example, it is prompted that the average value of the filter outer diameter is expected to recover to the ideal average value after adjustment. The operation interface uses real-time data display and warning mechanism to ensure that the information is updated in real time. The over-limit warning informs the production personnel of the abnormal situation in time, and the adjustment suggestion provides specific guidance to ensure the accuracy of operation; real-time updating improves the response speed: enhances the visualization and controllability of the production process, reduces human error, and improves the efficiency of quality control.

[0058] In use, step four completes the statistical threshold analysis of the corrected size matrix, and generates a forming nozzle gap fine adjustment instruction when it is out of the limited range, while the related information is prompted synchronously on the operation interface. This process ensures the overall eligibility and consistency of the filter size through dual evaluation of arithmetic mean and coefficient of variation; realizes real-time correction of size deviation through gap adjustment; improves the controllability of the production process through interface prompts.

[0059] Step five, obtaining the forming nozzle gap adjustment amount and the filter size statistical indicators after adjustment from step four, obtaining the gray image data stream from step one, obtaining the depth distribution data and size compensation coefficient from step three; binding the forming nozzle gap adjustment amount, the filter size statistical indicators after adjustment, the gray image data stream, the depth distribution data and the size compensation coefficient with the production batch number, the production time and the equipment number to form a structured data record; compressing and storing the gray image data stream and the depth distribution data; writing the data record into the batch information table, the adjustment record table, the image data table, the depth data table and the compensation coefficient table of the manufacturing knowledge base in the order of production batch number and production time, using transaction processing mechanism to ensure synchronous data update; retrieving historical data from the manufacturing knowledge base before new batch production to match the adjustment results and compensation coefficients under similar production conditions; using cluster analysis to identify key factors affecting filter size, using time series analysis to monitor quality trends, and using regression analysis to optimize adjustment thresholds; connecting the gray image data stream, the depth distribution data, the size compensation coefficient and the forming nozzle gap adjustment amount in series according to the production batch number to form an information chain, supporting quality traceability and process improvement.

[0060] The step five includes the following contents: Step 501, data collection and arrangement The adjustment amount of the forming nozzle gap and the adjusted filter size statistical indicators, including the corrected mean value and the coefficient of variation, are obtained from step four; the gray-scale image data stream of the filter outer periphery collected by the line array camera is obtained from step one, which records the gray-scale values of the filter at different time points and angular positions; the depth distribution data of the suspected concave area collected by the confocal displacement sensor is obtained from step three, which is consistent with the gray-scale image data stream in the coordinate system; and the size compensation coefficient generated by the Gaussian process regression is obtained from step three, which is used to correct the size data of the concave area. The data collection process ensures the comprehensiveness of subsequent analysis by integrating information from different sources.

[0061] During use, the integration of multi-source data can fully reflect the key parameters in the production process, improving the accuracy of analysis. Comprehensive data support helps to optimize the control process of filter size.

[0062] The forming nozzle gap adjustment amount, the corrected mean value, the coefficient of variation, the gray-scale image data stream, the depth distribution data, and the size compensation coefficient are bound with metadata such as production batch number, production time, and equipment number to form a structured data record. Among them, these data fields are associated with production batch number, production time, equipment number, etc. to generate entries containing all fields, and the binding process establishes the connection between data and production environment through metadata.

[0063] Among them, the metadata binding enhances the traceability of the data, facilitates problem positioning and quality management, facilitates quick retrieval of data under specific production conditions, and improves the efficiency of quality control.

[0064] The gray-scale image data stream and the depth distribution data are compressed, wherein the gray-scale image data stream is stored in JPEG format and the depth distribution data is stored in PNG format to reduce storage space occupation while retaining key feature information. The compression process selects the appropriate format according to the data characteristics to ensure information integrity.

[0065] Compression processing saves storage resources while ensuring that key information required for data analysis is not lost, which can reduce storage costs and improve system efficiency.

[0066] Step 502, manufacturing knowledge base construction A relational database or NoSQL database is used to design a multi-table structure to store data, including a batch information table (records production batch number, production time, and equipment number), an adjustment record table (stores the forming nozzle gap adjustment amount, the corrected mean value, and the coefficient of variation), an image data table (stores the storage path of the gray-scale image data stream), a depth data table (stores the storage path of the depth distribution data), and a compensation coefficient table (stores the size compensation coefficient). The multi-table design modularly stores data by function.

[0067] Modular storage facilitates data management and expansion, supports complex query requirements, can improve data retrieval speed, and ensures system stability when processing large-scale data.

[0068] The sorted data records are written into the corresponding database tables in the order of production batch number and production time. The data is sorted according to the production batch number and production time, and then inserted into the batch information table, adjustment record table, image data table, depth data table, and compensation coefficient table. The writing process maintains the logical order of time and batch. Sequential writing facilitates time series analysis and comparison between batches, maintains data logic, and can improve the accuracy of data analysis, supporting the tracking of quality trends.

[0069] For compressed grayscale image data streams and depth distribution data, store their file paths and record the associated production batch number and location information in the metadata. Record the storage path of the compressed file in the image data table and depth data table, and establish the correspondence between the production batch number and the file path in the batch information table. Path storage optimizes the database structure.

[0070] By storing paths instead of directly storing data, the database burden is reduced, and the query efficiency is improved, which can optimize system performance and support efficient management of large-scale data.

[0071] The writing process uses a transaction processing mechanism to ensure synchronous updating of all database tables. The data writing operation is encapsulated in a transaction, and if any table in the batch information table, adjustment record table, image data table, depth data table, or compensation coefficient table fails to write, all operations are rolled back to avoid data inconsistency. The transaction mechanism ensures data integrity.

[0072] Transaction processing ensures the atomicity and consistency of data writing, prevents data loss, and can improve the reliability of the manufacturing knowledge base, ensuring data availability.

[0073] Step 503, data reference and analysis Before producing a new batch, retrieve historical data from the manufacturing knowledge base. According to the current production conditions (such as equipment number, production time period), query the historical production batch data in the batch information table and adjustment record table, extract the forming nozzle gap adjustment amount, corrected average value, coefficient of variation, and size compensation coefficient, and match the records under similar production conditions. The retrieval process extracts relevant data based on condition matching.

[0074] Historical data retrieval can identify quality patterns under similar conditions, providing a reference for current production, reducing production fluctuations, and improving the effectiveness of adjustment strategies.

[0075] The historical data is grouped by cluster analysis, wherein the K-means clustering or hierarchical clustering method is adopted, the forming nozzle gap adjustment amount in the adjustment record table, the corrected average value and the coefficient of variation are taken as characteristics, the historical production batch data is classified, the key factors affecting the filter size such as the equipment state or the environmental variable are identified, the clustering process reveals the internal relationship between the data, the potential causes of quality fluctuation are found, the process optimization is supported, and data-driven improvement basis is provided to enhance the production stability.

[0076] The trend change of the corrected average value is monitored by time series analysis, wherein the corrected average values of adjacent production batches are extracted from the adjustment record table, the difference between the two batches is calculated, and then divided by the time interval of the production time of the two batches to obtain the change rate, which is used to judge the quality drift trend. Time series analysis focuses on dynamic changes.

[0077] Monitoring the change rate can timely find the abnormal trend of the quality parameter, improve the monitoring ability, and then support real-time early warning to ensure the stability of the filter size.

[0078] The adjustment threshold is optimized by regression analysis, wherein the relationship between the forming nozzle gap adjustment amount and the corrected average value and the coefficient of variation in the adjustment record table is utilized, the linear regression or nonlinear regression method is adopted, the best range of the allowable deviation value and the upper limit of the coefficient of variation in step four is determined, the regression analysis quantifies the correlation between the parameters, the data-driven threshold optimization improves the scientificity of the adjustment strategy, reduces the misjudgment, and can improve the adjustment accuracy and reduce the risk in quality control.

[0079] Step 504, information chain formation and application The gray image data stream, the depth distribution data, the size compensation coefficient and the forming nozzle gap adjustment amount are concatenated according to the production batch number. Wherein, the production batch number is taken as the primary key, the batch information table, the adjustment record table, the image data table, the depth data table and the compensation coefficient table are associated to form the whole process record from production to adjustment. The information chain integrates all related data.

[0080] The whole process record facilitates quality traceability and problem analysis, improves data utilization, can enhance production transparency and optimize quality management efficiency.

[0081] When the filter size is out of limit or the quality fluctuates, the depth distribution data and the size compensation coefficient are retrieved through the information chain to analyze the cause of the depression. Wherein, the depth distribution data in the depth data table is queried according to the production batch number, the depth characteristics of the suspected depression area are analyzed, and the size compensation coefficient in the compensation coefficient table is combined to evaluate the influence of the depression on the size. The analysis process focuses on the root cause of the problem. Data correlation analysis can quickly locate the specific cause of the quality problem, can shorten the problem solving time and improve the production efficiency.

[0082] In the process improvement or new product development, the historical information chain data is used for simulation verification. Among them, the whole process record of historical production batch is extracted from the manufacturing knowledge base, the effect of different forming nozzle gap adjustment strategy is simulated, the adjustment scheme is optimized, and the simulation verification is based on historical data for prediction. The historical data support the verification of process optimization, reduce the actual trial and error cost, can accelerate the process development process, and improve the production adaptability.

[0083] In use, step five realizes the systematic collection, arrangement and application of production data, builds a manufacturing knowledge base for filter production, and improves the reliability and production efficiency of filter size measurement through data reference, analysis and information chain application.

[0084] Those skilled in the art can appreciate that the units and algorithm steps of the examples described in combination with the embodiments disclosed herein can be realized in electronic hardware or a combination of computer software and electronic hardware. Whether the functions are realized in hardware or software depends on the specific application and design constraints of the technical solution. Professionals can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of the present application.

[0085] Those skilled in the art can clearly understand that, for the convenience and brevity of the description, the specific working process of the above-described system, device and unit can refer to the corresponding process in the foregoing method embodiments, which will not be repeated here.

[0086] In several embodiments provided in the present application, it should be understood that the disclosed system, device and method can be implemented in other ways. For example, the device embodiments described above are only schematic, for example, the division of the units is only a logical function division, and actual implementation can have another division manner, for example, a plurality of units or components can be combined or integrated into another system, or some features can be ignored or not executed. In addition, the coupling or direct coupling or communication connection between the units shown or discussed can be indirect coupling or communication connection through some interface, device or unit, and can be electrical, mechanical or other forms.

[0087] The units described as separate components can or can not be physically separated, and the components shown as units can or can not be physical units, that is, they can be located in one place, or can be distributed on a plurality of network units. According to actual needs, part or all of the units can be selected to achieve the purpose of the embodiment scheme.

[0088] The above merely provides the specific implementation of the present application, but the protection scope of the present application is not limited to this. Any person skilled in the art can easily think of the changes or replacements within the technical range disclosed by the present application, which should be covered in the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

Claims

1. A method for visually identifying and adjusting filter shape parameters, characterized by: include, A laser diameter gauge and a linear array camera are used to synchronously collect filter peripheral images and radial profile data to generate a dual baseline of size and grayscale. A texture segmentation method combining local binary pattern feature extraction and grayscale difference analysis is used to divide the radial contour data stream into smooth areas and suspected concave areas, and generate concave position vectors to provide target areas for depth inspection. A confocal displacement sensor is used to conduct a local scan of the suspected concave area to obtain depth distribution data, quantify the radial gradient fluctuation and the local depth fluctuation signal-to-noise ratio, and use Gaussian process regression to generate size compensation coefficients to correct the size matrix. Perform statistical threshold analysis on the mean and coefficient of variation of the corrected size matrix. If the limit is exceeded, a command for fine-tuning the gap between the forming nozzles will be sent and a prompt will be displayed on the operation interface. The adjustment results, original images, depth distribution data and size compensation coefficients are written into the manufacturing knowledge base.

2. The method for visually identifying and adjusting filter shape parameters according to claim 1, characterized in that: A laser diameter gauge and a linear array camera are installed at the winding outlet. High-precision clock signals or hardware triggers are used to synchronize data acquisition between the laser diameter gauge and the linear array camera. The laser diameter gauge measures the outer diameter of the filter tip to generate a radial profile data stream, while the linear array camera collects grayscale images of the filter tip surface to generate a grayscale image data stream. The radial profile data stream is processed by sliding window averaging to generate a size baseline, and the grayscale image data stream is processed by calculating the average grayscale value at each time point to generate a grayscale baseline.

3. The method for visually identifying and adjusting filter shape parameters according to claim 2, characterized in that: Gaussian filtering is performed on the radial profile data stream and the grayscale image data stream to remove noise, and data alignment between the radial profile data stream and the grayscale image data stream is achieved through timestamp matching and rotation angle calibration; The local binary pattern features are extracted from the grayscale image data stream to represent the texture, the grayscale difference between adjacent pixels is calculated, and the suspected concave areas are marked according to the preset threshold.

4. The method for visually identifying and adjusting filter shape parameters according to claim 3, characterized in that: Extract local binary pattern features from grayscale image data stream, where a binary code is generated to represent texture features by comparing the grayscale values ​​of each pixel with its eight neighboring pixels; Calculate the grayscale difference between adjacent pixels in the grayscale image and compare it with the preset threshold to mark the suspected concave area; Classification is performed based on the local binary pattern eigenvalue and the predetermined smooth area eigenvalue range. Pixels whose eigenvalues ​​fall within the range are classified as smooth areas, and pixels that do not fall within the range are classified as suspected concave areas. Pixels in suspected concave areas are clustered.

5. The method for visually identifying and adjusting filter shape parameters according to claim 4, characterized in that: According to the depression position vector, a confocal displacement sensor is used to perform a local scan of the suspected depression area and collect depth distribution data. The gradient along the radial angle is calculated based on the depth distribution data, and the radial gradient fluctuation is quantified by the root mean square value, which is then divided by the reference value to generate a dimensionless radial gradient fluctuation feature.

6. The method for visually identifying and adjusting filter shape parameters according to claim 5, characterized in that: Calculate the difference between the maximum and minimum values ​​of the depth distribution data in the suspected depression area to determine the local depth fluctuation amplitude, and then divide it by the sensor measurement noise level to obtain the signal-to-noise ratio; Gaussian process regression is used to calculate the size compensation coefficient using radial gradient fluctuation characteristics and signal-to-noise ratio as input. The size compensation coefficient is applied to the radial profile data of the suspected concave area to generate a corrected size matrix.

7. The method for visually identifying and adjusting filter shape parameters according to claim 6, characterized in that: Calculate the arithmetic mean and coefficient of variation from the modified size matrix; compare the arithmetic mean with a preset ideal mean and allowable deviation value, and compare the coefficient of variation with a preset upper limit of the coefficient of variation; When the absolute value of the difference between the arithmetic mean and the ideal mean exceeds the allowable deviation value or the coefficient of variation exceeds the upper limit of the coefficient of variation, it is judged to be out of limit.

8. The method for visually identifying and adjusting filter shape parameters according to claim 7, characterized in that: For over-limit situations, the difference between the arithmetic mean and the ideal mean is calculated as the mean deviation, and the forming nozzle gap adjustment amount is calculated based on the mean deviation and the pre-calibrated adjustment coefficient; The forming nozzle gap adjustment amount is sent to the forming nozzle control unit for fine-tuning, and an over-limit warning and adjustment suggestions are displayed on the operation interface.

9. The method for visually identifying and adjusting filter shape parameters according to claim 8, characterized in that: Bind the nozzle gap adjustment amount, adjusted filter size statistical indicators, grayscale image data stream, depth distribution data, and size compensation coefficient with the production batch number, production time, and equipment number to form a structured data record; The grayscale image data stream and depth distribution data are compressed and stored; The data records are written into the batch information table, adjustment record table, image data table, depth data table and compensation coefficient table of the manufacturing knowledge base in the order of production batch number and production time, and the transaction processing mechanism is used to ensure the synchronous update of data.

10. The method for visually identifying and adjusting filter shape parameters according to claim 9, characterized in that: Before a new batch is produced, historical data is retrieved from the manufacturing knowledge base to match adjustment results and compensation coefficients under similar production conditions. Cluster analysis is used to identify key factors affecting filter size, time series analysis is used to monitor quality trends, and regression analysis is used to optimize adjustment thresholds. The grayscale image data stream, depth distribution data, size compensation coefficient and forming nozzle gap adjustment amount are connected in series according to the production batch number to form an information chain.

Images