Injection mold production quality supervision method and system

Abstract

The invention belongs to the technical field of image processing, and particularly relates to an injection mold production quality supervision method and system, and the method comprises the steps: carrying out the two-dimensional discrete wavelet transformation of a standard image of a defect-free product and a target image corresponding to a to-be-detected product, and obtaining the sub-band energy of each level; calculating an energy deviation characteristic value of the target image relative to the standard image; selecting the sub-band with the maximum energy deviation characteristic value to construct a wavelet coefficient deviation graph, dividing the sub-band into a plurality of areas according to the key physical position of the mold, and calculating the degradation concentration degree reflecting the degradation space concentration degree; analyzing the time sequence change of the degradation concentration degree to determine the degradation tendency degree; and carrying out normalization fusion on the sensitive sub-band energy deviation, the degradation concentration degree and the degradation tendency degree to obtain a comprehensive early warning score for early warning. According to the method, the degradation trend can be identified before the defect forms a visible form, the quality control gateway is moved from post-screening to beforehand intervention, the rejection rate can be reduced, and the production stability can be improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of image processing. More particularly, the present application relates to a method and system for monitoring the production quality of an injection mold. BACKGROUND

[0002] In the field of injection molding production, the state of the mold has a decisive influence on the quality of the product. As the production process continues, the mold inevitably undergoes progressive wear, resulting in gradual changes in the surface microtexture, increased roughness, or decreased glossiness of the product, among other gradual defects. These defects, unlike obvious structural defects, develop slowly and are hidden, often difficult to detect by the naked eye or conventional detection methods in the early stages of the defect, but as the number of productions accumulates, they can eventually lead to batch quality problems.

[0003] Currently, the industry mainly relies on manual sampling inspection or automated visual inspection systems for quality inspection of finished products. Manual sampling inspection relies on subjective experience and judgment, which is not only inefficient but also unable to capture surface changes at the microscopic level. While automated visual inspection systems have achieved batch screening, their core logic still focuses on the macro appearance features of the final product, lacking the ability to sensitively capture micro changes on the surface of the product, and are unable to identify minor deterioration of the mold in the early stages. They can only conduct post-screening after the defect has taken a visible form. This post-screening model has a serious lag, and when the detection system detects obvious defects, hundreds or even thousands of substandard products have already been produced, severely affecting product quality and production efficiency.

[0004] Therefore, there is an urgent need for an injection mold production quality monitoring method that can achieve real-time monitoring and early warning of the mold state during the production process. SUMMARY

[0005] To solve the above technical problems of existing detection techniques lacking the ability to sensitively capture minor deterioration of the mold in the early stages, being unable to provide early warning before the gradual defects take a visible form, and only being able to conduct post-screening when the defects are obvious, resulting in batch substandard products, the present application provides solutions in the following aspects.

[0006] In a first aspect, the present application provides an injection mold production quality monitoring method, comprising: The most representative standard image is selected from the images of non-defective qualified products, and the image of the product to be detected is collected in real time as a target image; the standard image and the target image are respectively subjected to two-dimensional discrete wavelet transform, and the energy of each subband of each level of the standard image and the target image is obtained; the energy deviation characteristic value of each subband of the target image is determined according to the difference between the energy of each subband of each level of the target image and the energy of the corresponding subband of the corresponding level of the standard image; the sensitive subband is screened according to the size of the energy deviation characteristic value, the wavelet coefficient deviation map is constructed according to the difference between the target image and the standard image on the sensitive subband, the wavelet coefficient deviation map is divided into a plurality of key regions, and the degradation concentration degree is determined according to the wavelet coefficient deviation of the key region; the degradation trend degree is determined according to the time sequence change of the degradation concentration degree, and the comprehensive early warning score is obtained according to the energy deviation characteristic value of the target image on the sensitive subband, the degradation concentration degree of the target image and the degradation trend degree; and early warning is performed according to the size of the comprehensive early warning score.

[0007] In the application, the most representative standard image is selected from the images of non-defective qualified products as a reference, and the standard image and the target image are subjected to two-dimensional discrete wavelet transform to obtain the energy of each level subband, which can accurately quantify the texture and gloss characteristics of the product surface in different scales and directions; the energy deviation characteristic value of the corresponding subband of the target image and the standard image is calculated, the microscopic changes (such as roughness increase and gloss decrease) of the product surface can be converted into measurable numerical signals, so that early signs of degradation that are difficult to detect with the naked eye can be captured; the sensitive subband is screened according to the energy deviation characteristic value, the wavelet coefficient deviation map is constructed, the wavelet coefficient deviation map is divided into a plurality of key regions corresponding to the key physical positions of the mold, and then the degradation concentration degree is calculated, which can effectively distinguish between systematic local degradation and global random fluctuations, thereby avoiding misjudgment; the degradation trend degree is determined by analyzing the time sequence change of the degradation concentration degree, the energy deviation characteristic value of the target image on the sensitive subband, the degradation concentration degree and the degradation trend degree are fused to obtain the comprehensive early warning score, and the "instantaneous-space-time" three-in-one degradation representation is realized, so that the system can early, sensitive and reliable early warning of the gradual degradation, avoiding the lag problem that the traditional detection method can only identify the visible defects formed, realizing the forward shift of the quality control from post-screening to pre-intervention, which can significantly reduce the scrap rate and improve the stability of the production process.

[0008] Preferably, the method further comprises: respectively pre-processing the standard image and the target image, and the pre-processing includes image registration and illumination normalization, the image registration adopts a SIFT feature matching algorithm combined with a RANSAC algorithm, and the illumination normalization adopts histogram equalization.

[0009] Preferably, the energy deviation characteristic value satisfies the expression: ; in the expression, an energy deviation feature value of a first subband of a first level of the target image; an energy deviation feature value of a first subband of a first level of the target image; an energy deviation feature value of a first subband of a first level of the target image; an energy deviation feature value of a first subband of a first level of the target image; an energy deviation feature value of a first subband of a first level of the target image; an energy deviation feature value of a first subband of a first level of the target image; an energy deviation feature value of a first subband of a first level of the target image; an energy deviation feature value of a first subband of a first level of the target image; an energy deviation feature value of a first subband of a first level of the target image; an energy deviation feature value of a first subband of a first level of the target image;

[0010] The present application converts the naked-eye invisible surface deterioration into a measurable numerical signal by quantifying the relative change of wavelet energy, realizes early warning of the mold state, and avoids the hysteresis of relying on the final product appearance detection.

[0011] Preferably, the sensitive subband is selected according to the energy deviation feature value, including: selecting the subband with the largest energy deviation feature value in all subbands of all levels of the target image as the sensitive subband.

[0012] The present application selects the subband with the largest energy deviation feature value in all subbands of all levels of the target image as the sensitive subband, which can accurately capture the most significant deterioration characteristics of the current product, so that the system can automatically focus on the characteristic scale and direction that can best reflect the current main deterioration mode, and avoids the information redundancy and computational burden caused by equal processing of all subbands.

[0013] Preferably, the wavelet coefficient deviation map is constructed, including: calculating the difference between the wavelet coefficient values of the same position in the sensitive subband of the target image and the standard image as the wavelet coefficient deviation of the corresponding position in the sensitive subband of the target image and the standard image, and constructing the wavelet coefficient deviation map by the wavelet coefficient deviations of all positions in the sensitive subband of the target image and the standard image.

[0014] Preferably, the wavelet coefficient deviation map is divided into a plurality of key regions, including: dividing the wavelet coefficient deviation map into a plurality of key regions according to the geometric regions corresponding to the key physical positions of the mold.

[0015] Preferably, the deterioration concentration degree is determined according to the wavelet coefficient deviation of the key region, including: calculating the mean value of the wavelet coefficient deviation of all positions in each key region as the deviation index of each key region, and taking the variance of the deviation indexes of all key regions as the deterioration concentration degree.

[0016] The application takes the variance of all key area deviation indexes as the degradation concentration degree, can effectively quantify the uneven degree of the degradation in the spatial distribution, when the degradation is a systematic local problem, the wavelet coefficient deviation will be concentrated in a few key areas, resulting in significant differences in the deviation indexes of different key areas, so that the degradation concentration degree value is larger, when the degradation is a global problem or random noise, the deviation indexes of each key area tend to be consistent, so that the degradation concentration degree value is close to 0, therefore, the application can effectively distinguish the systematic degradation and random fluctuation by calculating the degradation concentration degree, and reduce the false positive rate.

[0017] Preferably, the degradation trend degree is determined according to the time sequence change of the degradation concentration, comprising: constructing a degradation sequence by taking the degradation concentration degrees corresponding to the images of the last n products to be detected; determining the degradation trend degree of the target image according to the degradation sequence: ; wherein, represents the degradation trend degree of the target image; represents the sequence number of the degradation concentration degree in the degradation sequence; represents the degradation concentration degree in the degradation sequence; represents the window size.

[0018] The application can capture the development trend of the degradation process in the time dimension by calculating the degradation trend degree, thereby distinguishing the real systematic degradation from the random fluctuation, avoiding the false positives caused by single measurement, and improving the reliability of the mold state monitoring and the accuracy of the early warning.

[0019] Preferably, the comprehensive early warning score satisfies the expression: ; wherein, represents the comprehensive early warning score; represents the degradation concentration degree of the target image; represents the degradation trend degree of the target image; represents the energy deviation eigenvalue of the target image in the sensitive subband; represents a linear rectifier function; represents normalization.

[0020] The application integrates the energy deviation eigenvalue of the target image in the sensitive subband, the degradation concentration degree and the degradation trend degree, realizes the "instantaneous-spatial-temporal" trinity degradation representation, enables the early warning system to comprehensively evaluate the mold state, can identify the real trend in the early stage of the gradual degradation, avoids the lag problem that the traditional detection method can only identify the visible defects formed, moves the quality control link from post-screening to pre-intervention, provides a decision basis for mold maintenance, effectively reduces the scrap rate and improves the stability of the production process.

[0021] ​​In a second aspect, the present application provides an injection mold production quality supervision system, comprising a processor and a memory, wherein the memory stores computer program instructions, and the computer program instructions, when executed by the processor, implement the injection mold production quality supervision method.

[0022] By adopting the above technical solution, the injection mold production quality supervision method is generated into a computer program and stored in the memory, so as to be loaded and executed by the processor, thereby facilitating the use of the terminal equipment made of the memory and the processor.

[0023] The present application has the following advantages: In the present application, the energy of each level sub-band of the target image and the standard image can be accurately quantified to reflect the texture and gloss features of the product surface in different scales and directions; the energy deviation eigenvalue is used to quantify the micro changes of the product surface, so that the early degradation signs that cannot be detected by the naked eye can be captured; the degradation concentration degree is calculated to effectively distinguish the systematic local degradation and the global random fluctuation, thereby avoiding misjudgment; the energy deviation eigenvalue, the degradation concentration degree and the degradation trend degree are comprehensively considered to realize the "instantaneous-space-time" three-in-one degradation representation, so that the system can early, sensitively and reliably warn the gradual degradation, avoid the lag problem that the traditional detection method can only identify the visible defects, move the quality control from the post-screening to the pre-intervention, significantly reduce the waste rate and improve the stability of the production process. BRIEF DESCRIPTION OF DRAWINGS

[0024] Figure 1 is a flowchart schematically showing the injection mold production quality supervision method in the present application; Figure 2 is a standard image schematic diagram; Figure 3 is a low-frequency approximation sub-band schematic diagram of the first level of the standard image; Figure 4 is a high-frequency detail sub-band schematic diagram of the horizontal direction of the first level of the standard image; Figure 5 is a high-frequency detail sub-band schematic diagram of the vertical direction of the first level of the standard image; Figure 6 is a high-frequency detail sub-band schematic diagram of the diagonal direction of the first level of the standard image; Figure 7 is a low-frequency approximation sub-band schematic diagram of the second level of the standard image; Figure 8 is a high-frequency detail sub-band schematic diagram of the horizontal direction of the second level of the standard image; Figure 9is a schematic diagram of a vertical high-frequency detail subband of the second level of the standard image; Figure 10 is a schematic diagram of a diagonal high-frequency detail subband of the second level of the standard image; Figure 11 is a schematic diagram of the target image; Figure 12 is a schematic diagram of a low-frequency approximation subband of the first level of the target image; Figure 13 is a schematic diagram of a horizontal high-frequency detail subband of the first level of the target image; Figure 14 is a schematic diagram of a vertical high-frequency detail subband of the first level of the target image; Figure 15 is a schematic diagram of a diagonal high-frequency detail subband of the first level of the target image; Figure 16 is a schematic diagram of a low-frequency approximation subband of the second level of the target image; Figure 17 is a schematic diagram of a horizontal high-frequency detail subband of the second level of the target image; Figure 18 is a schematic diagram of a vertical high-frequency detail subband of the second level of the target image; Figure 19 is a schematic diagram of a diagonal high-frequency detail subband of the second level of the target image; Figure 20 is a wavelet coefficient deviation map. DETAILED DESCRIPTION

[0025] 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 some of the embodiments of the present application, rather than all the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative work fall within the protection scope of the present application.

[0026] The specific embodiments of the present application will be described in detail below with reference to the drawings.

[0027] The embodiments of the present application disclose a method for monitoring the production quality of an injection mold, referring to Figure 1 , comprising steps S1-S5: S1, selecting the most representative standard image from the defect-free qualified product image, performing two-dimensional discrete wavelet transform on the standard image, and obtaining the energy of each subband of each level of the standard image.

[0028] Specifically, during the initial production phase after a new mold is put into use or has just completed deep maintenance, an industrial camera on the production line is used to capture a group (for example, 100 to 1000) of images of qualified products that are continuously produced and confirmed to be defect-free, forming a benchmark sample set.

[0029] Each image in the benchmark sample set is preprocessed. The preprocessing process includes image registration and illumination normalization. The image registration is used to eliminate slight changes in the position and posture of the product in the camera field of view. In this embodiment, the SIFT feature matching algorithm is combined with the RANSAC algorithm. In other embodiments, the implementer can select the image registration algorithm according to the actual implementation situation. The illumination normalization is used to reduce the impact of ambient light fluctuations on the grayscale distribution of the image. In this embodiment, histogram equalization is used for illumination normalization. In other embodiments, the implementer can select the algorithm for illumination normalization according to the actual implementation situation, such as the Retinex algorithm.

[0030] Furthermore, a most representative image is selected from the preprocessed benchmark sample set as the standard image.

[0031] In one embodiment, relevant staff selects standard images from the pre-processed benchmark sample set based on manual experience.

[0032] In another embodiment, the method for selecting the standard image is: for any image in the preprocessed benchmark sample set, the sum of differences between the image and all other images is calculated, and the image with the smallest sum of differences is used as the standard image.

[0033] Furthermore, the standard image is subjected to a two-dimensional discrete wavelet transform to decompose the standard image into Each level contains a low-frequency approximation subband and three high-frequency detail subbands in different directions (horizontal, vertical and diagonal). Represents the number of levels in wavelet transform. In the present invention, The value of can be set to 3 to 5. The implementer can adjust the number of layers according to the actual implementation situation. For example, for products with high requirements for smoothness, the number of layers can be appropriately increased to capture more subtle texture changes. Figure 2 is a standard image diagram, Figure 3 is the low-frequency approximate subband of the first level of the standard image, Figure 4 、 Figure 5 、 Figure 6 They are the high-frequency detail subbands in the horizontal, vertical, and diagonal directions of the first level of the standard image, Figure 7 is the low-frequency approximate subband of the second level of the standard image, Figure 8 、 Figure 9 、 Figure 10 They are the high-frequency detail subbands in the horizontal, vertical, and diagonal directions of the second level of the standard image.

[0034] Get the energy of the standard image in each subband: ; in, The standard image The first level The energy of each sub-band; The standard image The first level Subband mid-coordinates The wavelet coefficient value at ; Represents the standard image The size of each subband at each level, Represents the standard image The number of rows of each sub-band at each level, Represents the standard image The number of columns of each sub-band at each level.

[0035] It should be noted that energy is the energy density of the sub-band, which reflects the intensity of local changes in the image at the corresponding level and direction. The higher the energy, the richer the texture or edge information contained in the level. When there are microscopic changes on the product surface (such as increased roughness), the energy of the high-frequency detail sub-band usually increases. When the glossiness of the product surface decreases or becomes blurred, the energy of the low-frequency approximation sub-band usually decreases.

[0036] S2. Collect an image of the product to be inspected as a target image in real time, perform a two-dimensional discrete wavelet transform on the target image, obtain the energy of each subband of each level of the target image, and determine the energy deviation characteristic value of each subband of the target image based on the difference between the energy of each subband of each level of the target image and the energy of the corresponding subband of the corresponding level of the standard image.

[0037] Specifically, while the production line is running, an image of each product to be inspected is captured in real time and preprocessed, using the same preprocessing process as in step S1. The preprocessed image of the product to be inspected is used as the target image. A two-dimensional discrete wavelet transform is performed on the target image to obtain all subbands of the target image at each level, and the energy of each subband at each level of the target image is obtained. Figure 11 is a standard image diagram, Figure 12 is the low-frequency approximate subband of the first level of the standard image, Figure 13 、 Figure 14 、 Figure 15 They are the high-frequency detail subbands in the horizontal, vertical, and diagonal directions of the first level of the standard image, Figure 16 is a low frequency approximation subband of the 2nd level of the standard image, Figure 17 , Figure 18 , Figure 19 are high frequency detail subbands of the horizontal, vertical and diagonal directions of the 2nd level of the standard image respectively.

[0038] It should be noted that when the product surface appears microscopic changes, such as product surface roughness increases, gloss decreases, etc., it will inevitably cause the redistribution of wavelet energy. Therefore, the energy deviation characteristic value of the corresponding subband is obtained by calculating the energy difference of the current product to be detected and the standard image on the corresponding subband, which is used to quantify the weak deviation of the current product to be detected state relative to the ideal healthy state.

[0039] Specifically, the energy deviation characteristic value satisfies the expression: ; In the formula, , represents the energy deviation characteristic value of the jth subband of the ith level of the target image; , represents the energy of the jth subband of the ith level of the target image; , represents the energy of the jth subband of the ith level of the standard image; , represents the energy of the jth subband of the ith level of the standard image; , represents the energy of the jth subband of the ith level of the standard image; , represents the energy of the jth subband of the ith level of the standard image; , represents the energy of the jth subband of the ith level of the standard image; , represents the energy of the jth subband of the ith level of the standard image; , represents the energy of the jth subband of the ith level of the standard image; is a hyperparameter, which is used to prevent the denominator from being 0, and in the present application , the implementer can set it according to the actual implementation, but in order to avoid the influence of the value of the hyperparameter on the energy deviation characteristic value, , it needs to be within the range of . It should be noted that the energy of the corresponding subband of the standard image is used as the denominator in the form of relative deviation, which can eliminate the influence of the absolute energy dimension.

[0040] It should be further noted that when the product surface appears microscopic texture changes, such as mold wear leading to product surface roughness increases, the energy of the high frequency detail subband will slightly rise, so that the energy deviation characteristic value of the corresponding high frequency detail subband is a small positive value; when the product surface gloss decreases or blurs, the energy of the low frequency approximation subband will decrease, so that the energy deviation characteristic value of the low frequency approximation subband presents a negative value. Therefore, the energy deviation characteristic value can reflect different types of early degradation.

[0041] S3. Select sensitive sub-bands according to the size of the energy deviation eigenvalue. According to the difference between the target image and the standard image in the sensitive sub-bands, construct a wavelet coefficient deviation map. Divide the wavelet coefficient deviation map into multiple key areas, and determine the degradation concentration according to the wavelet coefficient deviation of the key areas.

[0042] It's important to note that gradual degradation typically doesn't occur uniformly across the entire mold, but rather begins in specific localized areas of the mold, such as the gate, ejector pinhole, or parting edge. To effectively capture this spatial imbalance, the present invention constructs a wavelet coefficient deviation map and divides it into multiple key regions. The concentration of degradation is determined based on the wavelet coefficient deviations in these key regions.

[0043] Specifically, the subband with the largest energy deviation eigenvalue can best reflect the current main degradation mode. Therefore, among all subbands of all levels of the target image, the subband with the largest energy deviation eigenvalue is selected as the sensitive subband.

[0044] For example, Figure 11 The target image and Figure 2 The energy deviation eigenvalue of the standard image in the high-frequency detail sub-band in the diagonal direction of the first level is the largest, so the high-frequency detail sub-band in the diagonal direction of the first level is the sensitive sub-band.

[0045] Furthermore, based on the difference between the target image and the standard image in the sensitive sub-band, a wavelet coefficient deviation map is constructed: ; in, Indicates the coordinates of the target image and the standard image in the sensitive subband The deviation of wavelet coefficients at ; Indicates the coordinates of the target image in the sensitive subband The wavelet coefficient value at ; Indicates the coordinates of the standard image in the sensitive subband The wavelet coefficient value at ; Indicates the absolute value symbol. The wavelet coefficient deviations of the target image and the standard image at all coordinates in the sensitive subband form a wavelet coefficient deviation map. Figure 20 is the wavelet coefficient deviation map.

[0046] Furthermore, the wavelet coefficient deviation map is divided into several key regions based on geometric areas corresponding to key physical locations of the mold (such as the gate, ejector pin location, and parting surface edge). The mean of the wavelet coefficient deviations at all locations within each key region is calculated as the deviation index for each key region. It should be noted that the number and shape of key regions can be determined by the implementer based on the mold structure and product characteristics.

[0047] Furthermore, the degradation concentration of the target image is determined based on the deviation index of all key areas: ; in, Indicates the degradation concentration of the target image, which is used to characterize the unevenness of the spatial distribution of the degradation signal; The first part represents the wavelet coefficient deviation map Deviation index of key areas; Represents the mean of the wavelet coefficient deviations at all positions in the wavelet coefficient deviation map; Indicates the number of critical areas. If the degradation is a systematic local problem, such as when the degradation is caused by local wear of the mold, the wavelet coefficient deviation will be concentrated in a few key areas, resulting in differences in the wavelet coefficient deviations between different key areas, making the degradation concentration Large; if the degradation is a global problem (such as overall color difference caused by changes in raw material batches) or random noise, the wavelet coefficient deviations of each key area will tend to be consistent, making the degradation concentration The value is close to 0.

[0048] S4. Determine the degradation trend degree according to the temporal change of the degradation concentration, and obtain a comprehensive warning score according to the energy deviation characteristic value of the target image in the sensitive sub-band, the degradation concentration degree of the target image, and the degradation trend degree.

[0049] It's important to note that real physical degradation processes exhibit temporal continuity and inertia, and their impact accumulates with increasing production cycles. For example, mold wear gradually intensifies with increasing production volume. Therefore, this paper derives the degradation trend based on the temporal changes in degradation concentration to confirm the authenticity of degradation trends over time.

[0050] Specifically, a sliding time window of size W is constructed, and the degradation concentration corresponding to the most recent W images of the product to be inspected is combined into a one-dimensional sequence, recorded as the degradation sequence. W is the preset window size. In this embodiment, W = 50. In other embodiments, implementers can set W based on actual implementation. For example, for molds with relatively rapid degradation, a smaller window (such as W = 30) can be set to improve response speed. For molds with relatively slow degradation, a larger window (such as W = 100) can be set to improve the stability of trend estimation.

[0051] Determine the degradation trend of the target image according to the degradation sequence: ; in, Indicates the degradation trend of the target image; The serial number representing the degradation concentration in the degradation sequence; Indicates the first Degradation concentration; W represents the window size. The present invention uses a univariate linear regression slope estimation formula to obtain the degradation trend of the mold. The degradation trend of the target image is equal to the fitting slope of the degradation sequence. If the change in degradation concentration is random, the fitting slope of the degradation sequence will be close to 0. Conversely, if there is a continuous degradation process, the degradation concentration in the degradation sequence will show a monotonically increasing trend, making the fitting slope of the degradation sequence a large positive value.

[0052] It should be noted that a real and urgent degradation trend should be manifested as large instantaneous deviation, uneven spatial distribution, and continuous deterioration over time. The energy deviation eigenvalues ​​of the target image in each sub-band at each level reflect the instantaneous deviation of the target image, among which the energy deviation eigenvalues ​​of the target image in the sensitive sub-band reflect the most significant instantaneous deviation of the target image. The degradation concentration of the target image reflects the degree of concentration of the degradation in spatial distribution, and the degradation trend reflects the rate of continuous deterioration of the degradation process over time. Therefore, the present invention obtains a comprehensive warning score based on the energy deviation eigenvalues ​​of the target image in the sensitive sub-band, the degradation concentration of the target image, and the degradation trend.

[0053] Specifically, the comprehensive warning score satisfies the expression: ; in, represents the comprehensive warning score; Indicates the degradation concentration of the target image; Indicates the degradation trend of the target image; Represents the energy deviation eigenvalue of the target image in the sensitive sub-band; Represents a linear rectification function, when the degradation trend is When it is negative or 0, The value is 0, when the deterioration trend When it is positive, The value is , thereby effectively suppressing false alarms caused by random noise; Represents the normalization function. In this embodiment, based on the historical data of the first 1000 products, the energy deviation characteristic value is respectively normalized by the linear normalization method. , Deterioration Concentration And the mapping value of degradation trend Normalization is performed. In other embodiments, the implementer may select a normalization method according to the actual implementation situation. When the energy deviation characteristic value , Deterioration Concentration and degradation trend When all of the energy deviation characteristic values, the deterioration concentration degree and the deterioration trend degree are relatively large, it indicates that the target image has significant instantaneous deviation, and the deterioration is concentrated in the spatial distribution and continuously deteriorates in time, and the comprehensive early warning score is relatively large, which indicates that the mold has a serious deterioration trend. When any one of the energy deviation characteristic values, the deterioration concentration degree and the deterioration trend degree is relatively small, it indicates that the deterioration characteristics of the target image are not obvious or only random fluctuations, and the comprehensive early warning score is relatively small, which indicates that the mold is in normal state.

[0054] S5, early warning according to the size of the comprehensive early warning score.

[0055] Specifically, in response to the comprehensive early warning score being greater than a preset early warning threshold, an early warning is generated, and the early warning information includes the spatial position of the deterioration (the key region with the highest deviation index in the wavelet coefficient deviation map) and the emergency degree of the deterioration (the deterioration trend degree T). It should be noted that the early warning threshold is set by the implementer according to production experience and risk tolerance, and in this embodiment, the early warning threshold is set to 0.3. The implementer can adjust the early warning threshold according to the actual implementation, for example, for products with extremely high quality requirements, the early warning threshold can be appropriately reduced to improve sensitivity, and for products that allow a certain degree of deterioration, the early warning threshold can be appropriately increased to reduce false positives.

[0056] Further, when the early warning information is generated, the system automatically generates a maintenance work order to notify the relevant workers to check the mold according to the early warning information.

[0057] The embodiment of the application also discloses an injection mold production quality supervision system, which comprises a processor and a memory, and the memory stores computer program instructions.

[0058] The above system also includes a communication bus and a communication interface and other components familiar to those skilled in the art, the settings and functions of which are known in the art, and therefore will not be described here.​​

Claims

1. A method for supervising the production quality of injection molds, characterized in that: include: Select the most representative standard image from the defect-free qualified product images, and collect the image of the product to be inspected as the target image in real time; Perform two-dimensional discrete wavelet transform on the standard image and the target image respectively to obtain the energy of each subband at each level of the standard image and the target image; Determine the energy deviation characteristic value of each subband of the target image according to the difference between the energy of each subband of each level of the target image and the energy of the corresponding subband of the corresponding level of the standard image; Sensitive subbands are screened based on the energy deviation eigenvalues. A wavelet coefficient deviation map is constructed based on the differences between the target image and the standard image in the sensitive subbands. The wavelet coefficient deviation map is divided into multiple key areas, and the degradation concentration is determined based on the wavelet coefficient deviations in the key areas. Determine the degradation trend based on the temporal changes in degradation concentration, and obtain a comprehensive warning score based on the energy deviation characteristic value of the target image in the sensitive sub-band, the degradation concentration of the target image, and the degradation trend. Issue warnings based on the comprehensive warning score.

2. A method for supervising the production quality of injection molds according to claim 1, characterized in that: Also includes: The standard image and the target image are preprocessed respectively. The preprocessing includes image registration and illumination normalization. The image registration adopts SIFT feature matching algorithm combined with RANSAC algorithm, and the illumination normalization adopts histogram equalization.

3. The method for quality control of injection mold production according to claim 1, wherein: The energy deviation characteristic value satisfies the expression: ； Where, The target image The first level The energy deviation eigenvalue of each sub-band; The target image The first level The energy of each sub-band; The standard image The first level The energy of each sub-band; is a hyperparameter used to prevent the denominator from being zero.

4. The method for quality control of injection mold production according to claim 1, wherein: The screening of sensitive sub-bands according to the energy deviation characteristic value includes: Among all sub-bands of all levels of the target image, the sub-band with the largest energy deviation eigenvalue is selected as the sensitive sub-band.

5. The method for quality control of injection mold production according to claim 1, wherein: The constructing of the wavelet coefficient deviation map includes: The difference between the wavelet coefficient values ​​of the target image and the standard image at the same position in the sensitive sub-band is calculated as the wavelet coefficient deviation of the corresponding position between the target image and the standard image in the sensitive sub-band. The wavelet coefficient deviation of all positions between the target image and the standard image in the sensitive sub-band constitutes a wavelet coefficient deviation map.

6. The method for quality control of injection mold production according to claim 1, characterized in that: The wavelet coefficient deviation map is divided into multiple key areas, including: The wavelet coefficient deviation map is divided into several key areas according to the geometric areas corresponding to the key physical positions of the mold.

7. The method for quality control of injection mold production according to claim 1, characterized in that: The step of determining the degradation concentration according to the wavelet coefficient deviation of the key area includes: The mean of the wavelet coefficient deviations at all locations in each key region is calculated as the deviation index of each key region, and the variance of the deviation indices of all key regions is taken as the degradation concentration.

8. The method for quality control of injection mold production according to claim 1, characterized in that: Determining the degradation trend degree according to the temporal change of degradation concentration includes: will be recently The degradation concentration corresponding to the images of the product to be inspected constitutes a degradation sequence; the degradation trend of the target image is determined according to the degradation sequence: ； in, Indicates the degradation trend of the target image; The serial number representing the degradation concentration in the degradation sequence; Indicates the first Deterioration concentration; Indicates the window size.

9. The method for quality control of injection mold production according to claim 1, characterized in that: The comprehensive warning score satisfies the expression: ； in, represents the comprehensive warning score; Indicates the degradation concentration of the target image; Indicates the degradation trend of the target image; Represents the energy deviation eigenvalue of the target image in the sensitive sub-band; represents the linear rectification function; Indicates normalization.

10. An injection mold production quality supervision system, characterized in that: include: A processor and a memory, wherein the memory stores computer program instructions, and when the computer program instructions are executed by the processor, an injection mold production quality supervision method according to any one of claims 1 to 9 is implemented.

Images