A quality management control method and system based on cosmetics production process

Abstract

The present invention discloses a quality management control method and system based on the cosmetic production process, which relates to the technical field of cosmetic quality management, and includes the following steps: first, the conveyor belt conveys the product to the inspection area at a preset speed, and the visual inspection system captures high-definition images of the product seal through an industrial camera; the captured image data is used to establish an orderly and retrievable image data set, and key features reflecting the abnormality of the product seal are extracted from the data set; the extracted key features are further analyzed and processed under the inspection window. The present invention introduces a deep learning model to intelligently evaluate the sealing light spot and heat distribution, generate a sealing abnormality coefficient, accurately distinguish between pollution and defects, reduce the misjudgment rate, and reduce resource waste. Real-time classification features and dynamic adjustment of the conveying speed take into account both detection accuracy and production efficiency, avoid system overload, and provide efficient and intelligent quality control for cosmetic production.

Description

Technical Field

[0001] The present invention relates to the technical field of cosmetics quality management, and in particular to a quality management control method and system based on a cosmetics production process. Background Art

[0002] Quality management and control based on the cosmetics production process refers to comprehensive quality control and optimization management of raw material procurement, production process, packaging, storage and factory inspection through systematic management methods, technical means and standard specifications in all aspects of cosmetics production. This process usually combines advanced detection technologies (such as spectral analysis, microscopic detection, etc.) and intelligent systems (such as machine vision, AI algorithms) to monitor and evaluate product quality in real time, identify and deal with potential quality problems in a timely manner, and ensure product safety, consistency and compliance with relevant regulations and consumer expectations. Through this system, it is possible to improve production efficiency, reduce the defective rate and maintain brand reputation.

[0003] The prior art has the following deficiencies:

[0004] In the prior art, when inspecting the seal of cosmetics, in order to ensure efficiency, the conveyor belt usually transports the product to the inspection area at a high speed, and the cosmetic seal is inspected by the visual inspection system to determine whether the seal has defects. During the sealing process, cosmetics may overflow or foreign matter (such as grease, dust) may adhere to the sealing surface. This type of pollution will cause abnormal texture features in the sealing area, which is easily misidentified as a defect by the visual inspection system, thereby increasing the time for feature comparison and error analysis. Since the features of contaminants are similar to those of real sealing defects, it is difficult for the system to quickly distinguish them, resulting in extended processing time and affecting the inspection efficiency. If the conveyor belt continues to transport products at a high speed during the cosmetic seal inspection process, and the visual inspection system prolongs the inspection time due to seal contamination (such as cosmetics overflow or foreign matter adhesion), serious consequences may occur. When the visual inspection system does not fully complete the feature comparison, it may misjudge the contaminated qualified products as unqualified products and reject them. This misjudgment will cause a large number of qualified products to be wasted, significantly increase the scrap rate and production loss, and cause waste of raw materials and packaging resources. In addition, products that are mistakenly rejected need to be re-tested or reworked, which further increases production costs and time consumption, affecting overall production efficiency.

[0005] The above information disclosed in this Background section is only for enhancement of understanding of the background of the present disclosure and therefore it may contain information that does not constitute the prior art that is already known to one of ordinary skill in the art. Summary of the invention

[0006] The purpose of the present invention is to provide a quality management control method and system based on the cosmetic production process, which introduces a deep learning model to intelligently evaluate the light spot distribution and heat distribution in the sealing area, generates a sealing abnormality coefficient, accurately distinguishes contamination from true sealing defects, reduces the misjudgment rate, and reduces resource waste. At the same time, by real-time classification of sealing features and dynamic adjustment of the conveyor belt speed, while ensuring detection accuracy, optimizing production efficiency and stability, avoiding efficiency decline and system overload problems, and providing an efficient and intelligent quality control solution for cosmetic production to solve the problems in the above-mentioned background technology.

[0007] In order to achieve the above object, the present invention provides the following technical solution: a quality management and control method based on the cosmetics production process, comprising the following steps:

[0008] First, the conveyor belt transports the product to the inspection area at a preset speed, and the visual inspection system captures high-definition images of the product seal through an industrial camera;

[0009] Establish an ordered and searchable image data set from the captured image data, and extract key features reflecting product sealing abnormalities from the data set;

[0010] The extracted key features are further analyzed and processed under the detection window, and the analyzed key features are input into the pre-trained deep learning model, which is used to intelligently evaluate the sealing features of the product being detected.

[0011] After the deep learning model evaluates the sealing features, it outputs the classification results and divides the sealing features into two categories: conventional feature analysis and non-standard feature evaluation;

[0012] For routine feature analysis, no further processing is required and the conveyor belt continues to transport the product to the next inspection area at a preset high speed;

[0013] For non-standard feature evaluation, based on the deep learning model evaluation results, the conveyor belt speed is reduced based on the preset conveyor belt speed to ensure that contaminants and true sealing defects can be accurately distinguished.

[0014] Preferably, key features reflecting product sealing abnormalities are extracted from the data set, the extracted features including light spot distribution in the sealing area and thermal distribution in the sealing area. Under the detection window, the extracted light spot distribution in the sealing area and thermal distribution in the sealing area are analyzed to generate a non-uniform light spot factor and a thermal distribution discrete factor, respectively. The non-uniform light spot index quantifies the uniformity of the light spot distribution in the sealing area, and reflects potential abnormal conditions by analyzing the non-uniformity of the optical properties of the sealing surface. The thermal distribution discrete index quantifies the uniformity of the thermal distribution in the sealing area, and identifies heat sealing abnormalities caused by spillage or contamination by analyzing temperature distribution fluctuations in thermal imaging images.

[0015] Preferably, in the detection window, the spot distribution of the sealing area is analyzed, and the specific steps of generating the non-uniform spot factor are as follows:

[0016] Under the inspection window, the image of the sealing area is captured by the visual inspection system, and the grayscale value of the image pixel is obtained. The calculation expression is as follows: , where , as well as are the red, green, and blue component values ​​of the pixel, respectively. is the gray value, is the pixel position, , as well as are the weight coefficients of the red, green, and blue component values ​​of the pixel respectively;

[0017] pass The operator extracts the image gradient component and calculates the spot gradient intensity in the sealing area. The calculation expression is as follows: , , where The pixel is The grayscale change rate in the direction, The pixel is The grayscale change rate in the direction, is the spot gradient intensity;

[0018] The gradient strength Converted to local energy to quantitatively evaluate the energy contribution of bright and dark spots. The calculation expression is as follows: , , where is the gradient strength The weight of is the gradient sensitivity parameter, is the equilibrium parameter, is the local energy;

[0019] By local energy The distribution of is subjected to high-order statistical analysis to generate a non-uniform spot factor, which is used to quantify the overall non-uniformity of the spot distribution. The calculation expression is as follows: , where is the non-uniform spot factor, is the total number of pixels in the capping area, is the nonlinear enhancement parameter, is the power adjustment parameter.

[0020] Preferably, in the detection window, the heat distribution of the sealing area is analyzed, and the specific steps of generating the heat distribution discrete factor are as follows:

[0021] Obtain the thermal distribution image of the sealing area and divide the sealing area into multiple sub-areas through the area segmentation algorithm , each sub-region contains a number of pixels, and the segmentation of the thermal distribution image is as follows: , where is the temperature value of the pixel in the sealing area, is the pixel position, The thermal distribution image sub-regions, is the total number of sub-regions, Indicates that only The pixels of the sub-region will be included in middle;

[0022] For each sub-region The temperature distribution inside is calculated by gradient to generate the thermal gradient field. The calculation expression is as follows: , where is the thermal gradient field, which represents the thermal gradient value of the pixel point. is the temperature change rate of the pixel in the horizontal direction, indicating that the temperature The gradient of the direction, is the temperature change rate of the pixel in the vertical direction, indicating that the temperature Directional gradient;

[0023] For each sub-region , calculate the thermal fluctuation aggregation degree of the thermal gradient field, the calculation expression is as follows: , where is the thermal fluctuation degree of aggregation, i.e., the sub-region Thermal fluctuation degree of polymerization, is the maximum thermal gradient value in the region, is the minimum thermal gradient value in the region, is a tiny constant that prevents the denominator from being zero;

[0024] The degree of polymerization of all sub-regions was obtained by thermal fluctuation , generate the heat distribution discrete factor, the calculation expression is as follows: , where is the heat distribution discrete factor, It is a sub-region The weight of .

[0025] Preferably, after obtaining the non-uniform spot factor and the thermal distribution discrete factor generated after analyzing the extracted key features, the non-uniform spot factor and the thermal distribution discrete factor are input into a pre-learned deep learning model, and a sealing abnormality coefficient is generated by the deep learning model. The sealing characteristics of the currently detected product are intelligently evaluated by the sealing abnormality coefficient.

[0026] Preferably, the generated sealing abnormality coefficient is compared and analyzed with a preset sealing abnormality coefficient reference threshold, and the sealing characteristics of the current product are divided, and the division steps are as follows:

[0027] If the sealing abnormality coefficient is greater than or equal to the preset sealing abnormality coefficient reference threshold, the current product sealing feature is classified as a non-standard feature evaluation;

[0028] If the sealing abnormality coefficient is less than a preset sealing abnormality coefficient reference threshold, the current product sealing feature is classified as a conventional feature analysis.

[0029] Preferably, for non-standard feature evaluation, based on the deep learning model evaluation results, the conveyor belt speed is reduced based on the preset conveyor belt speed to ensure that contaminants and true sealing defects can be accurately distinguished. The specific steps are as follows:

[0030] First, according to the sealing abnormality coefficient The dynamic adjustment factor is calculated by the ratio of the reference threshold of the sealing abnormality coefficient. The expression is as follows: , where is the reference threshold of the sealing abnormality coefficient, is the sensitivity coefficient of abnormality to speed adjustment, is the dynamic adjustment factor;

[0031] Getting the dynamic adjustment factor After that, according to the preset conveyor belt speed and Calculate the actual adjusted conveying speed, the calculation expression is as follows: , where is the actual adjusted conveyor belt speed, is the preset conveyor belt speed;

[0032] After the conveyor belt slows down, in order to ensure that the visual inspection system can fully process abnormal features, the inspection dwell time is dynamically adjusted. The calculation expression is as follows: , where is the detection dwell time, is the length of the detection area;

[0033] Finally, the adjustment effect is evaluated according to the real-time feedback mechanism of the detection system. The calculation expression is as follows: , where The optimized sensitivity coefficient is is the abnormality rate of feedback during the detection process, is the preset target abnormality rate threshold.

[0034] A quality management and control system based on the cosmetics production process, including an image capture module, a data construction and feature extraction module, a feature analysis and intelligent evaluation module, a feature classification module, a high-speed conveying module, and a dynamic deceleration and fine detection module:

[0035] Image capture module, first, the conveyor belt transports the product to the inspection area at a preset speed, and the visual inspection system captures high-definition images of the product seal through an industrial camera;

[0036] The data construction and feature extraction module builds an ordered and searchable image data set from the captured image data, and extracts key features that reflect product sealing anomalies from the data set;

[0037] Feature analysis and intelligent evaluation module: The extracted key features are further analyzed and processed under the detection window, and the analyzed key features are input into the pre-trained deep learning model, which is used to intelligently evaluate the sealing features of the product being tested.

[0038] Feature classification module: After the deep learning model evaluates the sealing features, it outputs the classification results and divides the sealing features into two categories: conventional feature analysis and non-standard feature evaluation;

[0039] High-speed conveyor module: for routine feature analysis, no further processing is required, and the conveyor belt continues to transport the product to the next inspection area at a preset high speed;

[0040] The dynamic deceleration and fine inspection module evaluates non-standard features and reduces the speed of the conveyor belt based on the results of the deep learning model, based on the preset conveyor belt speed, to ensure that contaminants and true sealing defects can be accurately distinguished.

[0041] In the above technical solution, the technical effects and advantages provided by the present invention are:

[0042] The present invention introduces a deep learning model to intelligently evaluate the spot distribution and heat distribution in the sealing area. The solution can accurately generate the sealing abnormality coefficient and comprehensively quantify the abnormality of the sealing state. The non-uniform spot factor and the heat distribution discrete factor extract features from the two dimensions of optical properties and heat distribution uniformity. The deep learning model effectively distinguishes contamination (such as cosmetics overflow, foreign matter adhesion) from real sealing defects (such as cracks, incomplete heat sealing) by learning the complex nonlinear relationship between these features. This precise classification capability significantly reduces the misjudgment rate, avoids misjudging contaminated qualified products as unqualified products, reduces scrap rate and resource waste, and saves raw materials, packaging resources, and additional costs of rework and re-testing. Compared with traditional detection methods, this solution not only improves detection accuracy, but also comprehensively optimizes the utilization efficiency of production resources.

[0043] The present invention generates a sealing abnormality coefficient in real time and compares it with a reference threshold value, thereby dividing the sealing features into two categories: conventional feature analysis and non-standard feature evaluation, thereby ensuring a dynamic balance between detection efficiency and production line operation speed. For conventional feature analysis, the conveyor belt maintains high-speed operation, and the product quickly enters the next link, thereby ensuring overall production efficiency. For non-standard feature evaluation, by intelligently adjusting the conveyor belt speed, the detection system can obtain sufficient time to perform a detailed analysis of the seal, thereby improving detection accuracy. This dynamic control mechanism not only avoids the decline in efficiency caused by deceleration of the entire line, but also prevents missed detections and performance bottlenecks caused by excessive processing load of the detection system, ultimately improving the stability of the production line and the reliability of the detection system, providing a solid guarantee for efficient and intelligent cosmetics production. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in the present invention. For ordinary technicians in this field, other drawings can also be obtained based on these drawings.

[0045] Figure 1 The present invention is a method flow chart of a quality management control method based on a cosmetics production process.

[0046] Figure 2 The module diagram of a quality management and control system based on the cosmetics production process of the present invention. DETAILED DESCRIPTION

[0047] Example embodiments will now be described more fully with reference to the accompanying drawings. However, example embodiments can be implemented in a variety of forms and should not be construed as limited to the examples set forth herein; rather, these example embodiments are provided so that the description of the present disclosure will be more comprehensive and complete, and the concept of the example embodiments will be fully conveyed to those skilled in the art.

[0048] The present invention provides Figure 1 A quality management control method based on the cosmetic production process shown includes the following steps:

[0049] First, the conveyor belt transports the product to the inspection area at a preset speed, and the visual inspection system captures high-definition images of the product seal through an industrial camera;

[0050] The conveyor belt transports the product to the inspection area at a preset speed, which means that the conveyor belt's running speed is set to a fixed or dynamically adjusted value according to the production line's operating requirements and the processing capacity of the visual inspection system, so that the product can enter the inspection area continuously and stably. This speed needs to take into account multiple factors, including the size, weight, spacing of the product, and the image acquisition and processing speed of the visual inspection system, to ensure that the product stays in the inspection area long enough to support the industrial camera to complete high-quality image capture, while avoiding image blur, detection omissions or data backlogs caused by excessive speed, thereby achieving an efficient and accurate inspection process.

[0051] Establish an ordered and searchable image data set from the captured image data, and extract key features reflecting product sealing abnormalities from the data set;

[0052] The establishment of an ordered and retrievable image data set specifically includes the following steps: first, the image data captured by the industrial camera is classified and labeled according to metadata such as timestamp, product batch, production line number, etc.; second, the image is stored using standardized file naming rules and data storage structures (such as databases or distributed file systems); then, relevant feature tags are attached to the image (such as sealing area location, texture features, color distribution, etc.); finally, an efficient retrieval mechanism is constructed using indexing technology (such as database-based indexing or image retrieval algorithms) to support rapid search according to specific conditions. The purpose of establishing such a retrievable data set is: on the one hand, it is convenient to extract key features reflecting sealing anomalies from it in the future, providing input for intelligent detection; on the other hand, it provides complete data support for production quality traceability, model optimization, fault analysis, etc., which helps to achieve long-term performance improvement and anomaly tracing of the sealing detection system.

[0053] The extracted key features are further analyzed and processed under the detection window, and the analyzed key features are input into the pre-trained deep learning model, which is used to intelligently evaluate the sealing features of the product being detected.

[0054] The key features reflecting the abnormal sealing of the product are extracted from the data set. The extracted features include the light spot distribution and the thermal distribution of the sealing area. Under the detection window, the extracted light spot distribution and the thermal distribution of the sealing area are analyzed to generate the non-uniform light spot factor and the thermal distribution discrete factor respectively. The non-uniform light spot index quantifies the uniformity of the light spot distribution in the sealing area, and reflects the potential abnormal state by analyzing the non-uniformity of the optical properties of the sealing surface. The thermal distribution discrete index quantifies the uniformity of the thermal distribution in the sealing area, and identifies the heat sealing anomaly caused by overflow or contamination by analyzing the temperature distribution fluctuation in the thermal imaging image.

[0055] The uneven distribution of light spots in the sealing area leads to enhanced or weakened local reflection, which can indeed indicate that the current cosmetic seal is in an abnormal state. This abnormality usually originates from the spillage of cosmetics or the adhesion of foreign matter (such as grease and dust) to the sealing surface during the sealing process, which changes the optical properties of the sealing area. The normal sealing surface reflection should be uniform and consistent, but spillage or foreign matter will destroy this optical balance, causing the reflection characteristics of the local area to be significantly enhanced (bright spots) or weakened (dark spots). These abnormal optical phenomena reflect that the sealing area is contaminated or disturbed by the process, which may lead to a decrease in the integrity of the seal, such as poor sealing or insufficient sealing strength. In addition, the visual inspection system can easily misidentify these optical changes as other types of defects (such as cracks or surface damage), further indicating that the abnormal distribution of light spots is a direct manifestation of abnormal sealing. Therefore, from the perspective of reflecting the sealing state, the phenomenon of uneven distribution of light spots is a key signal, indicating that there may be potential problems in the current sealing area, which requires high attention and further detection and analysis.

[0056] Under the detection window, analyze the spot distribution in the sealing area and generate the non-uniform spot factor in the following specific steps:

[0057] Under the inspection window, the image of the sealing area is captured by the visual inspection system, the grayscale value (brightness) of the image pixels is obtained, and the color image is converted into a grayscale image for subsequent quantitative analysis of the brightness characteristics. The grayscale value range is usually [0,255], and the calculation expression is as follows: , where , as well as are the red, green, and blue component values ​​of the pixel, respectively. is the grayscale value, which represents the brightness intensity of the pixel, is the pixel position, which represents the two-dimensional coordinates of the pixel in the image. , as well as They are the weight coefficients of the red, green, and blue component values ​​of the pixel. These coefficients are designed based on the differences in the human eye's sensitivity to color and are in line with international standards (ITU-R BT.601). The green channel has the highest weight (0.5870) because the human eye is most sensitive to green; the red channel is second (0.2989); and the blue channel has the lowest weight (0.1140).

[0058] The grayscale image will be used to analyze the spot distribution characteristics and is the basis for the entire non-uniform spot index generation process.

[0059] pass The operator extracts the image gradient component and calculates the spot gradient intensity in the sealing area to identify the changes in bright and dark spots. The calculation expression is as follows: ,

[0060] , where The pixel is The grayscale change rate in the direction is used to capture the lateral characteristics of the light spot, such as the horizontal boundary of a bright or dark spot. The pixel is The grayscale change rate in the direction is used to capture the longitudinal characteristics of the light spot, such as the vertical boundary of a bright or dark spot. It is the intensity of the light spot gradient, which indicates the variation of the gray value. It is used to comprehensively measure the variation of the light spot boundary in the horizontal and vertical directions, and reflects the unevenness of the light spot.

[0061] Areas with higher spot gradient intensity usually correspond to the edges of bright or dark spots and are key indicators of abnormal spot distribution.

[0062] The gradient strength Converted into local energy to quantitatively evaluate the energy contribution of bright and dark spots. By applying quadratic weights to weight the gradient intensity, abnormal areas with high gradient values ​​occupy a larger proportion in the overall energy calculation, thereby highlighting the significance of these areas and reducing the interference of low gradient areas. The calculation expression is as follows: , , where is the gradient strength By nonlinearly weighting the gradient intensity, the area with high gradient value has a greater influence in the overall energy calculation, while the contribution of the area with low gradient value is appropriately weakened. is the gradient sensitivity parameter, which controls the amplification effect of the weight function on the gradient change. is a balance parameter used to control the starting point of gradient strength amplification. is the local energy, which is used to quantify the contribution of each pixel to the abnormal distribution of the light spot;

[0063] High energy areas indicate significant anomalies in the light spot distribution, such as large bright or dark spots.

[0064] By local energy The distribution of is subjected to high-order statistical analysis to generate a non-uniform spot factor, which is used to quantify the overall non-uniformity of the spot distribution. The calculation expression is as follows: , where is the non-uniform spot factor, is the total number of pixels in the capping area, is a nonlinear enhancement parameter used to adjust the local energy The nonlinear enhancement parameters are Perform nonlinear amplification to make high-energy areas (such as severe bright spots or dark spots) have a more significant impact on the results. It is a power adjustment parameter used to further amplify or compress the unevenness of the spot distribution.

[0065] Under the detection window, the larger the non-uniform spot factor value generated after analyzing the spot distribution in the sealing area, the more uneven the spot distribution in the sealing area, reflecting a significant local change in the optical properties, such as local reflection enhancement (bright spot) or weakening (dark spot). These changes are usually related to cosmetic spillage or foreign matter (such as grease, dust) adhering to the sealing surface. They are important signals of sealing abnormalities and are easily misidentified as defects by the visual inspection system. On the contrary, if the non-uniform spot factor value is small, it means that the optical properties of the sealing area are uniform and consistent, and no obvious abnormalities are detected, indicating that there are no obvious problems with the sealing of the current product.

[0066] The uneven heat distribution in the sealing area leads to abnormal local heat conduction, which can indeed indicate that the current cosmetic seal is in an abnormal state, because the uniformity of heat distribution is one of the key indicators of sealing quality. During the sealing process, cosmetic spills (such as grease) or foreign matter (such as dust) attached to the sealing surface will interfere with the heat conduction path and cause local overheating or cooling. This abnormal heat distribution may lead to incomplete welding in the sealing area, insufficient sealing strength, or damage to the sealing material due to overheating, which directly affects the functionality of the seal. In addition, abnormal heat distribution may also be a signal of unstable operation of the heat sealing equipment, further exacerbating the uncertainty of the sealing quality. Since uneven heat distribution will destroy the consistency and integrity of the seal, this phenomenon is easily misidentified as a sealing defect during detection, resulting in misjudgment. Therefore, the uneven heat distribution in the sealing area not only reflects the potential abnormal state of the sealing process, but also provides a warning signal for the sealing performance and quality control of cosmetic products, which must be taken seriously and accurately judged through intelligent means.

[0067] Under the detection window, the heat distribution of the sealing area is analyzed and the specific steps of generating the heat distribution discrete factor are as follows:

[0068] Obtain the thermal distribution image of the sealing area and divide the sealing area into multiple sub-areas through the area segmentation algorithm , each sub-region contains a number of pixels, ensuring that the thermal distribution analysis is focused on the key area of ​​the seal and avoiding background noise interference. The segmentation of the thermal distribution image is as follows: , where is the temperature value of the pixel in the sealing area, is the pixel position, is the thermal distribution image sub-regions, is the total number of sub-regions, Indicates that only The pixels of the sub-region will be included in middle;

[0069] By segmenting, we can ensure that subsequent calculations focus on the sealing-related areas, thereby improving the accuracy of thermal distribution analysis.

[0070] For each sub-region The temperature distribution inside is calculated by gradient to generate a thermal gradient field. The thermal gradient field reflects the directionality and amplitude of temperature change. By capturing gradient anomalies, potential local overheating or cooling areas can be identified. The calculation expression is as follows: , where It is the thermal gradient field, which indicates the thermal gradient value of the pixel point, indicating the direction and intensity of temperature change. is the temperature change rate of the pixel in the horizontal direction, indicating that the temperature The gradient of the direction, is the temperature change rate of the pixel in the vertical direction, indicating that the temperature Directional gradient;

[0071] For each sub-region , calculate the thermal fluctuation aggregation degree of the thermal gradient field. This index measures the concentration of temperature fluctuations in the region. By analyzing the discreteness and aggregation of the gradient value, the abnormal distribution characteristics are further mined. The calculation expression is as follows: , where is the thermal fluctuation degree of aggregation, i.e., the sub-region The thermal fluctuation aggregation degree reflects the concentration of temperature fluctuations in the area. is the maximum thermal gradient value within a region, usually occurring at local hot spots or cold spots. is the minimum thermal gradient value in the region, usually occurring in areas with stable temperatures. is a tiny constant that prevents the denominator from being zero;

[0072] This step combines the overall intensity of thermal fluctuations with the extent of local variations, highlighting the impact of anomalous fluctuations.

[0073] The degree of polymerization of all sub-regions was obtained by thermal fluctuation , generate the heat distribution discrete factor, the calculation expression is as follows: , where is the heat distribution discrete factor, It is a sub-region The weight of represents the contribution of each sub-region to the overall thermal distribution discrete index.

[0074] Under the detection window, the larger the value of the heat distribution discrete factor generated after analyzing the heat distribution of the sealing area, the more significant the fluctuation of the heat distribution in the sealing area, reflecting that there may be abnormal conditions such as cosmetics overflow or foreign matter (such as grease, dust) adhesion during the sealing process. These abnormalities will interfere with heat conduction, causing local overheating or cooling, thereby destroying the uniformity and integrity of the seal, resulting in potential defects in the sealing quality. On the contrary, when the value of the heat distribution discrete factor is small, it means that the heat distribution in the sealing area is uniform, the heat sealing process is stable, there is no obvious interference, and there is no abnormality in the seal.

[0075] After obtaining the non-uniform spot factor and thermal distribution discrete factor generated by analyzing the extracted key features, the non-uniform spot factor and thermal distribution discrete factor are input into the pre-learned deep learning model, and the sealing abnormality coefficient is generated by the deep learning model. The sealing characteristics of the currently inspected product are intelligently evaluated through the sealing abnormality coefficient.

[0076] A pre-learned deep learning model refers to an artificial intelligence model that has been trained and optimized with a large amount of sample data and has the ability to predict and evaluate specific tasks (such as sealing anomaly detection). In this application scenario, this model focuses on learning the complex relationship between the optical properties and thermal distribution characteristics of cosmetic seals, and their association with abnormal sealing states. During the model training process, the developers used a large data set containing normal sealing states and various abnormal states. The data set includes non-uniform spot factors, thermal distribution discrete factors, and the correspondence between these factors and actual detection results. By repeatedly iteratively optimizing the model's parameters (such as weights and biases), the model can accurately capture the impact of different characteristic factors on sealing anomalies, thereby establishing a mapping relationship between sealing states and abnormalities.

[0077] This deep learning model is usually based on advanced neural network structures (such as convolutional neural networks (CNN), long short-term memory networks (LSTM), or multi-layer perceptrons (MLP)) and is trained through supervised learning. Developers input a large amount of labeled data into the model, where each data sample contains a non-uniform spot factor, a heat distribution discrete factor, and a corresponding sealing abnormality degree label. The model continuously adjusts the internal weights to minimize the error between the predicted value and the actual label, and gradually learns the weight distribution and nonlinear relationship of the characteristic factors to the sealing abnormality. The trained model can quickly evaluate new samples, generate a sealing abnormality coefficient, and provide an intelligent evaluation of the product's sealing status. The pre-learned model is essentially a knowledge base that summarizes the empirical patterns in sealing anomaly detection by learning historical data, and can identify potential sealing problems in an efficient and accurate manner in real-time detection.

[0078] The deep learning model is not limited here and can realize the non-uniform spot factor and heat distribution discrete factor Perform comprehensive analysis to generate sealing abnormality coefficient In order to realize the technical solution of the present invention, the present invention provides a specific implementation method: the sealing abnormality coefficient The generation formula is as follows: , where Non-uniform spot factor and heat distribution discrete factor The preset scaling factor of Both are greater than 0.

[0079] The preset scaling factor here ( ) refers to the non-uniform spot factor based on experience or actual needs and heat distribution discrete factor The coefficient of abnormal sealing The weights set by the contribution degree in the calculation. The role of these coefficients is to adjust the two factors to the sealing abnormality coefficient The influence strength of the factors is reflected in the evaluation of sealing anomalies, thereby reflecting their relative importance in the evaluation of sealing anomalies. By assigning a larger weight (proportional coefficient) to more critical or sensitive factors, it can ensure that the calculation results are more in line with the actual detection needs and enhance the sensitivity of the model to different abnormal characteristics. Generally speaking, the preset proportional coefficient is determined by experimental verification or large-scale data training results, which is scientific and applicable.

[0080] It can be seen from the sealing abnormality coefficient that, under the detection window, the larger the non-uniform spot factor expression value generated after analyzing the spot distribution in the sealing area, and the larger the thermal distribution discrete factor expression value generated after analyzing the thermal distribution in the sealing area, the larger the sealing abnormality coefficient expression value generated by comprehensive analysis of the extracted key features under the detection window, indicating that the seal of the currently detected product may have abnormal conditions such as cosmetics overflow or foreign matter (such as grease, dust) adhesion, otherwise it means that there is no abnormality in the current product seal.

[0081] After the deep learning model evaluates the sealing features, it outputs the classification results and divides the sealing features into two categories: conventional feature analysis and non-standard feature evaluation;

[0082] The generated sealing abnormality coefficient is compared and analyzed with the preset sealing abnormality coefficient reference threshold, and the sealing characteristics of the current product are divided. The division steps are as follows:

[0083] If the sealing abnormality coefficient is greater than or equal to the preset sealing abnormality coefficient reference threshold, the current product sealing feature is classified as a non-standard feature evaluation;

[0084] If the sealing abnormality coefficient is less than a preset sealing abnormality coefficient reference threshold, the current product sealing feature is classified as a conventional feature analysis.

[0085] Conventional feature analysis refers to situations where the sealing features meet production standards and quality specifications. Such features exhibit uniform texture, smooth surface, no foreign matter adhesion or contamination, and the optical reflection and shape distribution are within the normal range. Non-standard feature evaluation refers to situations where the sealing features deviate from normal standards and exhibit abnormal patterns or potential defects, such as cosmetic spillage, foreign matter contamination such as grease or dust, irregular texture, and may even have slight cracks or deformations.

[0086] For routine feature analysis, no further processing is required and the conveyor belt continues to transport the product to the next inspection area at a preset high speed;

[0087] For products that do not require further processing based on routine feature analysis, the conveyor belt continues to transport them to the next inspection area at a preset high speed in order to maximize overall production efficiency by optimizing the inspection process. This strategy avoids unnecessary speed adjustments and inspection delays, ensuring that qualified products that meet standards are quickly identified and pass inspection directly. At the same time, the high accuracy of rapid feature recognition reduces the possibility of false rejection and missed inspections, ensuring the continued efficient operation of the production line while maintaining inspection quality, thereby reducing resource waste and production costs.

[0088] For non-standard feature evaluation, based on the deep learning model evaluation results, the conveyor belt speed is reduced based on the preset conveyor belt speed to ensure accurate differentiation between contaminants and true sealing defects;

[0089] For non-standard feature evaluation, based on the deep learning model evaluation results, the conveyor belt speed is reduced based on the preset conveyor belt speed to ensure that contaminants and true sealing defects can be accurately distinguished. The specific steps are as follows:

[0090] First, according to the sealing abnormality coefficient The dynamic adjustment factor is calculated by the ratio of the reference threshold of the sealing abnormality coefficient. The dynamic adjustment factor reflects the abnormality of the sealing of the current detected product and directly determines whether the conveying speed needs to be adjusted and the adjustment range. The expression is as follows: , In the formula, is the reference threshold of the sealing abnormality coefficient, It is the sensitivity coefficient of abnormality to speed adjustment, reflecting the system's tolerance to sealing abnormalities. It is a dynamic adjustment factor, indicating whether the conveyor belt speed needs to be reduced and the degree of reduction;

[0091] when :Indicates that the sealing abnormality coefficient is lower than the set reference threshold, and the current product sealing feature is classified as a conventional feature analysis, and no additional processing is required. At this time, the dynamic adjustment factor , the conveying speed remains unchanged.

[0092] when :Indicates that the sealing abnormality coefficient reaches or exceeds the set reference threshold, and the current product sealing feature is classified as a non-standard feature evaluation. At this time, the dynamic adjustment factor , it is necessary to slow down appropriately according to the degree of abnormality to ensure detection accuracy.

[0093] Getting the dynamic adjustment factor After that, according to the preset conveyor belt speed and Calculate the actual adjusted conveying speed. The actual adjusted speed directly determines whether the conveyor belt slows down and the deceleration amplitude. The calculation expression is as follows: , where It is the actual adjusted conveyor belt speed, which is used to dynamically adjust the conveying speed to match the detection accuracy requirements. It is the preset conveyor belt speed, indicating the conveyor speed at which the production line operates normally when there are no abnormal features;

[0094] when When the dynamic adjustment factor , conveying speed is maintained .

[0095] when hour, , conveyor belt speed will be reduced, ensuring that the inspection system has enough time to distinguish contaminants from true defects.

[0096] After the conveyor belt slows down, in order to ensure that the visual inspection system can fully process abnormal features, the inspection dwell time is dynamically adjusted, that is, the time each product stays in the inspection area. The calculation expression is as follows: , where is the detection dwell time, is the length of the detection area, which represents the effective detection length covered by the visual inspection system;

[0097] The time the product stays in the inspection area after the conveyor belt speed slows down Increase, ensuring that the visual inspection system has enough time to process non-standard features and reduce the risk of misjudgment.

[0098] Finally, the adjustment effect is evaluated according to the real-time feedback mechanism of the detection system. If misjudgment or misrejection is still found during the detection process, the conveying speed and sensitivity coefficient are re-optimized. The calculation expression is as follows: , where The optimized sensitivity coefficient is used to improve the adaptability of the dynamic adjustment factor. is the abnormality rate fed back during the detection process, that is, the proportion of abnormal sealing detected, It is the preset target abnormality rate threshold, which is used to determine whether the current detection quality meets the requirements.

[0099] Adjusting sensitivity through feedback , the dynamic adjustment factor can be further optimized , achieve more precise control of the conveying speed, gradually reduce the misjudgment rate and false rejection rate, and improve detection accuracy.

[0100] For non-standard feature evaluation, based on the evaluation results of the deep learning model and the preset conveyor belt speed, the purpose of reducing the conveyor belt speed is to ensure that the visual inspection system has enough time to accurately analyze and identify the sealing features, so as to accurately distinguish between contaminants and real sealing defects. On high-speed production lines, non-standard features (such as abnormal surface spots and thermal distribution caused by cosmetics overflow, grease, dust, etc.) often show complex and subtle differences, which are easily confused with real defects. Especially at conventional conveyor speeds, the processing time of the visual inspection system is limited, resulting in an increased risk of misjudgment or missed judgment. By dynamically reducing the conveyor speed, the product's residence time in the inspection area is increased, and the visual inspection system can more fully process and analyze abnormal features, and use the deep learning model for feature segmentation, detail identification and error analysis, thereby reducing the misjudgment rate and missed detection rate. In addition, this process effectively balances production efficiency and detection accuracy by adaptively adjusting the conveyor belt speed, avoiding the waste of resources and quality risks caused by the system's excessive misrejection of qualified products or missed detection of unqualified products, ensuring that the entire production line maintains high-precision sealing quality detection while operating efficiently.

[0101] The present invention introduces a deep learning model to intelligently evaluate the spot distribution and heat distribution in the sealing area. The solution can accurately generate the sealing abnormality coefficient and comprehensively quantify the abnormality of the sealing state. The non-uniform spot factor and the heat distribution discrete factor extract features from the two dimensions of optical properties and heat distribution uniformity. The deep learning model effectively distinguishes contamination (such as cosmetics overflow, foreign matter adhesion) from real sealing defects (such as cracks, incomplete heat sealing) by learning the complex nonlinear relationship between these features. This precise classification capability significantly reduces the misjudgment rate, avoids misjudging contaminated qualified products as unqualified products, reduces scrap rate and resource waste, and saves raw materials, packaging resources, and additional costs of rework and re-testing. Compared with traditional detection methods, this solution not only improves detection accuracy, but also comprehensively optimizes the utilization efficiency of production resources.

[0102] The present invention generates a sealing abnormality coefficient in real time and compares it with a reference threshold value, thereby dividing the sealing features into two categories: conventional feature analysis and non-standard feature evaluation, thereby ensuring a dynamic balance between detection efficiency and production line operation speed. For conventional feature analysis, the conveyor belt maintains high-speed operation, and the product quickly enters the next link, thereby ensuring overall production efficiency. For non-standard feature evaluation, by intelligently adjusting the conveyor belt speed, the detection system can obtain sufficient time to perform a detailed analysis of the seal, thereby improving detection accuracy. This dynamic control mechanism not only avoids the decline in efficiency caused by deceleration of the entire line, but also prevents missed detections and performance bottlenecks caused by excessive processing load of the detection system, ultimately improving the stability of the production line and the reliability of the detection system, providing a solid guarantee for efficient and intelligent cosmetics production.

[0103] The present invention provides Figure 2 A quality management and control system based on the cosmetics production process shown in the figure includes an image capture module, a data construction and feature extraction module, a feature analysis and intelligent evaluation module, a feature classification module, a high-speed conveying module, and a dynamic deceleration and fine detection module:

[0104] Image capture module, first, the conveyor belt transports the product to the inspection area at a preset speed, and the visual inspection system captures high-definition images of the product seal through an industrial camera;

[0105] The data construction and feature extraction module builds an ordered and searchable image data set from the captured image data, and extracts key features that reflect product sealing anomalies from the data set;

[0106] Feature analysis and intelligent evaluation module: The extracted key features are further analyzed and processed under the detection window, and the analyzed key features are input into the pre-trained deep learning model, which is used to intelligently evaluate the sealing features of the product being tested.

[0107] Feature classification module: After the deep learning model evaluates the sealing features, it outputs the classification results and divides the sealing features into two categories: conventional feature analysis and non-standard feature evaluation;

[0108] High-speed conveyor module: for routine feature analysis, no further processing is required, and the conveyor belt continues to transport the product to the next inspection area at a preset high speed;

[0109] The dynamic deceleration and fine inspection module evaluates non-standard features and reduces the speed of the conveyor belt based on the results of the deep learning model, based on the preset conveyor belt speed, to ensure that contaminants and true sealing defects can be accurately distinguished.

[0110] A quality management and control method based on a cosmetics production process provided by an embodiment of the present invention is implemented by the above-mentioned quality management and control system based on a cosmetics production process. The specific method and process of a quality management and control system based on a cosmetics production process are detailed in the embodiment of the above-mentioned quality management and control method based on a cosmetics production process, which will not be repeated here.

[0111] The above formulas are all dimensionless and numerical calculations. The formula is a formula for the most recent real situation obtained by collecting a large amount of data and performing software simulation. The preset parameters in the formula are set by technicians in this field according to actual conditions.

[0112] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art who is familiar with the present technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

[0113] The above description is only by way of illustration of certain exemplary embodiments of the present invention. It is undoubted that those skilled in the art can modify the described embodiments in various ways without departing from the spirit and scope of the present invention. Therefore, the above drawings and descriptions are illustrative in nature and should not be construed as limiting the scope of protection of the claims of the present invention.

Claims

1. A quality management and control method based on the cosmetics production process, characterized in that: The following steps are involved: First, the conveyor belt transports the product to the inspection area at a preset speed, and the visual inspection system captures high-definition images of the product seal through an industrial camera; Establish an ordered and searchable image data set from the captured image data, and extract key features reflecting product sealing abnormalities from the data set; The extracted key features are further analyzed and processed under the detection window, and the analyzed key features are input into the pre-trained deep learning model, which is used to intelligently evaluate the sealing features of the product being detected. After the deep learning model evaluates the sealing features, it outputs the classification results and divides the sealing features into two categories: conventional feature analysis and non-standard feature evaluation; For routine feature analysis, no further processing is required and the conveyor belt continues to transport the product to the next inspection area at a preset high speed; For non-standard feature evaluation, based on the deep learning model evaluation results, the conveyor belt speed is reduced based on the preset conveyor belt speed to ensure accurate differentiation between contaminants and true sealing defects; After obtaining the non-uniform spot factor and heat distribution discrete factor generated by analyzing the extracted key features, the non-uniform spot factor and heat distribution discrete factor are input into the pre-learned deep learning model, and the sealing abnormality coefficient is generated by the deep learning model. The sealing characteristics of the currently inspected product are intelligently evaluated by the sealing abnormality coefficient; The generated sealing abnormality coefficient is compared and analyzed with the preset sealing abnormality coefficient reference threshold, and the sealing characteristics of the current product are divided. The division steps are as follows: If the sealing abnormality coefficient is greater than or equal to the preset sealing abnormality coefficient reference threshold, the current product sealing feature is classified as a non-standard feature evaluation; If the sealing abnormality coefficient is less than the preset sealing abnormality coefficient reference threshold, the current product sealing feature is classified as a conventional feature analysis; For non-standard feature evaluation, based on the deep learning model evaluation results, the conveyor belt speed is reduced based on the preset conveyor belt speed to ensure that contaminants and true sealing defects can be accurately distinguished. The specific steps are as follows: First, according to the sealing abnormality coefficient The dynamic adjustment factor is calculated by the ratio of the reference threshold of the sealing abnormality coefficient. The expression is as follows: , where is the reference threshold of the sealing abnormality coefficient, is the sensitivity coefficient of abnormality to speed adjustment, is the dynamic adjustment factor; Getting the dynamic adjustment factor After that, according to the preset conveyor belt speed and Calculate the actual adjusted conveying speed, the calculation expression is as follows: , where is the actual adjusted conveyor belt speed, is the preset conveyor belt speed; After the conveyor belt slows down, in order to ensure that the visual inspection system can fully process abnormal features, the inspection dwell time is dynamically adjusted. The calculation expression is as follows: , where is the detection dwell time, is the length of the detection area; Finally, the adjustment effect is evaluated according to the real-time feedback mechanism of the detection system. The calculation expression is as follows: , where The optimized sensitivity coefficient is is the abnormality rate of feedback during the detection process, is the preset target abnormality rate threshold.

2. A quality management and control method based on the cosmetics production process according to claim 1, characterized in that: The key features reflecting the abnormal sealing of the product are extracted from the data set. The extracted features include the light spot distribution and the thermal distribution of the sealing area. Under the detection window, the extracted light spot distribution and the thermal distribution of the sealing area are analyzed to generate the non-uniform light spot factor and the thermal distribution discrete factor respectively. The non-uniform light spot index quantifies the uniformity of the light spot distribution in the sealing area, and reflects the potential abnormal state by analyzing the non-uniformity of the optical properties of the sealing surface. The thermal distribution discrete index quantifies the uniformity of the thermal distribution in the sealing area, and identifies the heat sealing anomaly caused by overflow or contamination by analyzing the temperature distribution fluctuation in the thermal imaging image.

3. A quality management and control method based on the cosmetics production process according to claim 2, characterized in that: Under the detection window, analyze the spot distribution in the sealing area and generate the non-uniform spot factor in the following specific steps: Under the inspection window, the image of the sealing area is captured by the visual inspection system, and the grayscale value of the image pixel is obtained. The calculation expression is as follows: , where , as well as are the red, green, and blue component values ​​of the pixel, respectively. is the gray value, is the pixel position, , as well as are the weight coefficients of the red, green, and blue component values ​​of the pixel respectively; pass The operator extracts the image gradient component and calculates the spot gradient intensity in the sealing area. The calculation expression is as follows: ， , where The pixel is The grayscale change rate in the direction, The pixel is The grayscale change rate in the direction, is the spot gradient intensity; The gradient strength Converted to local energy to quantitatively evaluate the energy contribution of bright and dark spots. The calculation expression is as follows: ， , where is the gradient strength The weight of is the gradient sensitivity parameter, is the equilibrium parameter, is the local energy; By local energy The distribution of is subjected to high-order statistical analysis to generate a non-uniform spot factor, which is used to quantify the overall non-uniformity of the spot distribution. The calculation expression is as follows: , where is the non-uniform spot factor, is the total number of pixels in the capping area, is the nonlinear enhancement parameter, is the power adjustment parameter.

4. A quality management and control method based on the cosmetics production process according to claim 2, characterized in that: Under the detection window, the heat distribution of the sealing area is analyzed and the specific steps of generating the heat distribution discrete factor are as follows: Obtain the thermal distribution image of the sealing area and divide the sealing area into multiple sub-areas through the area segmentation algorithm , each sub-region contains a number of pixels, and the segmentation of the thermal distribution image is as follows: , where is the temperature value of the pixel in the sealing area, is the pixel position, is the thermal distribution image sub-regions, is the total number of sub-regions, Indicates that only The pixels of the sub-region will be included in middle; For each sub-region The temperature distribution inside is calculated by gradient to generate the thermal gradient field. The calculation expression is as follows: , where is the thermal gradient field, which represents the thermal gradient value of the pixel point. is the temperature change rate of the pixel in the horizontal direction, indicating that the temperature The gradient of the direction, is the temperature change rate of the pixel in the vertical direction, indicating that the temperature Directional gradient; For each sub-region , calculate the thermal fluctuation aggregation degree of the thermal gradient field, the calculation expression is as follows: , where is the thermal fluctuation degree of aggregation, i.e., the sub-region Thermal fluctuation degree of polymerization, is the maximum thermal gradient value in the region, is the minimum thermal gradient value in the region, is a tiny constant that prevents the denominator from being zero; The degree of polymerization of all sub-regions was obtained by thermal fluctuation , generate the heat distribution discrete factor, the calculation expression is as follows: , where is the heat distribution discrete factor, It is a sub-region The weight of .

5. A quality management and control system based on the cosmetics production process, used to implement the quality management and control method based on the cosmetics production process described in any one of claims 1 to 4, characterized in that: It includes image capture module, data construction and feature extraction module, feature analysis and intelligent evaluation module, feature classification module, high-speed conveying module and dynamic deceleration and fine detection module: Image capture module, first, the conveyor belt transports the product to the inspection area at a preset speed, and the visual inspection system captures high-definition images of the product seal through an industrial camera; The data construction and feature extraction module builds an ordered and searchable image data set from the captured image data, and extracts key features that reflect product sealing anomalies from the data set; Feature analysis and intelligent evaluation module: The extracted key features are further analyzed and processed under the detection window, and the analyzed key features are input into the pre-trained deep learning model, which is used to intelligently evaluate the sealing features of the product being tested. Feature classification module: After the deep learning model evaluates the sealing features, it outputs the classification results and divides the sealing features into two categories: conventional feature analysis and non-standard feature evaluation; High-speed conveyor module: for routine feature analysis, no further processing is required, and the conveyor belt continues to transport the product to the next inspection area at a preset high speed; The dynamic deceleration and fine inspection module evaluates non-standard features and reduces the speed of the conveyor belt based on the results of the deep learning model, based on the preset conveyor belt speed, to ensure that contaminants and true sealing defects can be accurately distinguished.

Images