Cosmetic bottle size detection system based on visual measurement

Abstract

The present application relates to the technical field of machine vision detection and packaging container quality detection, in particular to a cosmetic bottle size detection system based on visual measurement, comprising: a multi-modal data acquisition module for acquiring a conventional structured light image, an orthogonal polarized light image sequence and a preset assembly process parameter; an optical distortion decoupling module for extracting a local refractive index variation and performing reverse optical compensation on a size edge to generate a real absolute size; a stress field reconstruction module for extracting a two-dimensional photoelastic interference fringe and converting it into a three-dimensional residual stress distribution tensor graph; an assembly simulation calculation module for generating a dynamic micro-topography under an assembly stress state; a risk quantification and feedback module for comparing a sealing element model and outputting a sealing failure probability index, and then outputting rejection, release or equipment process adjustment information; the present application can simultaneously realize transparent or semi-transparent cosmetic bottle size detection and sealing risk prediction, and reduce the risk of misjudgment.

Description

Technical Field

[0001] This invention relates to the field of machine vision inspection and packaging container quality inspection technology, specifically a cosmetic bottle size inspection system based on vision measurement. Background Technology

[0002] Transparent or semi-transparent cosmetic bottles are widely used in cosmetic packaging. Production lines usually need to inspect the dimensions of the bottle opening, such as the outer diameter, inner diameter, thread pitch, and sealing surface, to determine whether they meet the assembly requirements. With the development of visual measurement technology, dimension inspection methods based on structured light imaging or ordinary machine vision are gradually being used for online inspection of cosmetic bottles. This can complete the identification of geometric features of the bottle opening without stopping the production line, thus improving inspection efficiency.

[0003] However, when inspecting transparent or semi-transparent cosmetic bottles, existing technologies mostly focus on static dimensional judgment, making it difficult to take into account residual stress inside the material, local refractive index changes, and the stress deformation after screw cap assembly. This can easily lead to situations where the apparent dimensions are qualified but the actual seal fails. Especially in mass production, relying solely on single image detection or back-end leakage sampling will not only reduce the accuracy of risk identification but also make it difficult to detect abnormalities in front-end molding and downstream assembly parameters in a timely manner. Summary of the Invention

[0004] To address the aforementioned technical problems, this invention provides a cosmetic bottle size detection system based on visual measurement. Specifically, the technical solution of this invention includes:

[0005] Multimodal data acquisition module, optical distortion decoupling module, stress field reconstruction module, assembly simulation calculation module, and risk quantification and feedback module;

[0006] The multimodal data acquisition module includes a communication interface for acquiring conventional structured light images and orthogonal polarized light image sequences of cosmetic bottles as the target detection object from external vision acquisition devices, performing spatial coordinate system registration on the conventional structured light images and orthogonal polarized light image sequences, and acquiring preset assembly process parameters through the communication interface;

[0007] The optical distortion decoupling module is used to extract the local refractive index change relative to the refractive index of the standard material based on the orthogonal polarized light image sequence, and to extract the initial size edge from the conventional structured light image. The initial size edge is then inversely optically compensated using the local refractive index change to generate the true absolute size.

[0008] The stress field reconstruction module is used to extract two-dimensional photoelastic interference fringes based on orthogonally polarized light image sequences and convert the two-dimensional photoelastic interference fringes into a three-dimensional residual stress distribution tensor map.

[0009] The assembly simulation calculation module is used to take the actual absolute size and the three-dimensional residual stress distribution tensor map as initial conditions, and combine them with preset assembly process parameters to perform simulation calculations and generate the dynamic micromorphology of the assembly under stress.

[0010] The risk quantification and feedback module is used to compare the fit between the dynamic micromorphology and the pre-stored preset sealing component model to generate a seal failure probability index.

[0011] The risk quantification and feedback module is also used to determine execution conditions: when the seal failure probability index is higher than or equal to the preset risk threshold, a rejection instruction is output and equipment process adjustment information is generated; when the seal failure probability index is lower than the preset risk threshold, a qualified release instruction is output.

[0012] Preferably, the preset assembly process parameters include standard torque parameters and preset elastic modulus of the sealing material; the conventional structured light image contains the external contour features and geometric dimensions of the target object.

[0013] Preferably, the optical distortion decoupling module includes a refractive index calculation unit, an edge extraction unit, and an inverse compensation unit. The refractive index calculation unit is used to calculate the local refractive index change inside the target object by analyzing the polarization state change in the orthogonal polarized light image sequence. The edge extraction unit is used to extract the initial size edge from the conventional structured light image. The inverse compensation unit is used to construct an optical distortion model based on the local refractive index change, input the initial size edge into the optical distortion model for coordinate offset correction, and output the true absolute size.

[0014] Preferably, the stress field reconstruction module includes a fringe analysis unit and a tensor transformation unit; the fringe analysis unit is used to perform phase unwrapping processing on the orthogonal polarized light image sequence to extract two-dimensional photoelastic interference fringes; the tensor transformation unit is used to map the two-dimensional photoelastic interference fringes into spatial principal stress differences based on the photoelastic law, and construct a three-dimensional residual stress distribution tensor map of a specific region in the target object; the specific region includes the threaded region and the sealing surface region.

[0015] Preferably, the assembly simulation calculation module includes a digital twin construction unit and a finite element solution unit; the digital twin construction unit is used to construct a target detection digital twin with the actual absolute size as the geometric boundary and the three-dimensional residual stress distribution tensor map as the internal material property; the finite element solution unit is used to apply the standard torque parameter as the boundary load to the target detection digital twin, calculate the nodal displacement data after being subjected to force through finite element analysis, and reconstruct the dynamic micromorphology based on the nodal displacement data.

[0016] Preferably, when performing fit comparison, the risk quantification and feedback module is specifically used to extract the contour of the sealing contact surface in the dynamic micromorphology; perform Boolean intersection operation on the contour of the sealing contact surface and the preset sealing component model to calculate the effective contact area ratio and the maximum local gap; obtain the negative weighting factor and positive weighting factor pre-calibrated based on historical assembly leakage test data, calculate the negative weighted value of the effective contact area ratio by combining the negative weighting factor, calculate the positive weighted value of the maximum local gap by combining the positive weighting factor, sum the negative weighted value and the positive weighted value to obtain the intermediate score, use the preset bias to linearly cut the intermediate score and limit it to the interval between zero and one, and output the seal failure probability index.

[0017] Preferably, when outputting equipment process adjustment information, the risk quantification and feedback module is specifically used to analyze the location of stress concentration areas in the three-dimensional residual stress distribution tensor map; generate a cooling temperature adjustment command for the front-end injection molding equipment based on the location of the stress concentration areas, and send the cooling temperature adjustment command for the front-end injection molding equipment to the external production control terminal through a preset communication bus; generate torque compensation parameters for the downstream capping equipment based on the seal failure probability index, and send the torque compensation parameters for the downstream capping equipment to the external downstream capping equipment.

[0018] Preferably, generating torque compensation parameters for the downstream capping equipment based on the seal failure probability index specifically includes: calculating the global average stress value of the three-dimensional residual stress distribution tensor map; when the global average stress value is higher than a preset stress threshold and the seal failure probability index is lower than a preset risk threshold, generating torque compensation parameters for the downstream capping equipment to reduce the standard torque parameters by a first preset ratio; when the global average stress value is higher than the preset stress threshold and the seal failure probability index is higher than or equal to the preset risk threshold, generating torque compensation parameters for the downstream capping equipment to reduce the standard torque parameters by a second preset ratio, wherein the second preset ratio is greater than the first preset ratio; and when the global average stress value is lower than or equal to the preset stress threshold, maintaining the standard torque parameters unchanged.

[0019] Compared with the prior art, the present invention has the following beneficial effects:

[0020] 1. This invention uses a multimodal data acquisition module to simultaneously acquire conventional structured light images, orthogonally polarized light image sequences, and preset assembly process parameters, integrating the geometric detection information and assembly conditions of transparent or semi-transparent cosmetic bottles into the same detection link, thus avoiding the problem that existing technologies rely solely on static size determination and cannot identify sealing risks in a timely manner.

[0021] 2. This invention utilizes an optical distortion decoupling module to analyze polarization state changes to obtain local refractive index changes, and performs reverse optical compensation on the initial size edges in the structured light image. This can separate the optical edge offset caused by internal stress of transparent materials from the real entity boundary, thereby improving the reliability of detecting the true absolute dimensions such as the outer diameter, inner diameter, thread pitch, and sealing surface of the bottle opening.

[0022] 3. This invention extracts two-dimensional photoelastic interference fringes based on orthogonal polarized light image sequences and reconstructs three-dimensional residual stress distribution tensors of threaded and sealing surface regions, transforming the stress visual texture, which was originally difficult to use directly for production line determination, into spatial stress input that can participate in subsequent mechanical calculations, thereby making up for the shortcomings of existing technologies in taking into account the influence of residual stress inside the material.

[0023] 4. This invention obtains the nodal displacement data and dynamic micromorphology after the cap is subjected to force through finite element analysis. It can predict the deformation trend of the sealing surface after assembly, such as micron-level warping, local upturning or sinking, before the bottle enters the downstream capping station, thus avoiding misjudgment that the apparent dimensions are qualified but leakage occurs after actual capping. Attached Figure Description

[0024] The present invention will be further explained below with reference to the accompanying drawings and embodiments:

[0025] Figure 1 This is a schematic diagram of a visual measurement-based cosmetic bottle size detection system provided in an embodiment of this application. Detailed Implementation

[0026] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to specific embodiments.

[0027] A vision-based cosmetic bottle size detection system, where the cosmetic bottle is a packaging container made of transparent or semi-transparent material, includes:

[0028] The multimodal data acquisition module includes a communication interface for acquiring conventional structured light images and orthogonal polarized light image sequences of a cosmetic bottle as the target object from an external vision acquisition device, performing spatial coordinate system registration on the conventional structured light images and orthogonal polarized light image sequences, and acquiring preset assembly process parameters through the communication interface.

[0029] The optical distortion decoupling module is used to extract the local refractive index change relative to the refractive index of the standard material based on the orthogonal polarized light image sequence, and to extract the initial size edge from the conventional structured light image. The initial size edge is then inversely optically compensated using the local refractive index change to generate the true absolute size.

[0030] The stress field reconstruction module is used to extract two-dimensional photoelastic interference fringes based on orthogonally polarized light image sequences and convert the two-dimensional photoelastic interference fringes into a three-dimensional residual stress distribution tensor map.

[0031] The assembly simulation calculation module is used to take the actual absolute size and the three-dimensional residual stress distribution tensor map as initial conditions, and combine them with preset assembly process parameters to perform simulation calculations and generate the dynamic micromorphology of the assembly under stress.

[0032] The risk quantification and feedback module is used to compare the fit between the dynamic microstructure and a pre-stored preset sealing component model to generate a seal failure probability index; the risk quantification and feedback module is also used for execution condition determination.

[0033] When the seal failure probability index is higher than or equal to the preset risk threshold, a rejection instruction is output and equipment process adjustment information is generated; when the seal failure probability index is lower than the preset risk threshold, a qualified release instruction is output.

[0034] This embodiment provides a visual measurement and dynamic sealing risk prediction mechanism for transparent or semi-transparent cosmetic bottle production lines, such as... Figure 1 As shown; specifically, this embodiment takes the automatic inspection production line of transparent thick-walled bottles of polyethylene terephthalate copolymer for essence as the main scenario: after injection molding, cooling and conveying, the bottle enters the bottle mouth inspection station. A structured light camera is arranged above the inspection station, and an orthogonal polarization imaging component is arranged laterally. The industrial control computer receives the images and capping process parameters through Ethernet or fieldbus, and gives release, rejection or process adjustment instructions before the bottle enters the downstream capping station;

[0035] In this scenario, if only static dimensions such as the outer diameter, inner diameter, and thread pitch of the bottle opening are detected, leakage may still occur even if the static dimensions are within tolerance. The reason for this problem is that residual stress inside the transparent bottle opening can cause local refractive index changes, resulting in edge shifts in the structured light image. At the same time, under the action of the capping torque, residual stress can be released or redistributed, causing micron-level warping of the sealing surface. Therefore, this embodiment inputs the structured light image, the orthogonal polarized light image sequence, and the assembly process parameters into the same detection link, instead of separating the size detection and sealing risk assessment.

[0036] To maintain consistency in terminology, the terms "capping torque," "capping machine target torque," or "standard capping torque" used below, unless otherwise specified as instantaneous deviation values ​​of the equipment, shall be understood as specific instance values ​​or reference values ​​of the standard torque parameters in the embodiments.

[0037] Specifically, after the bottle arrives at the detection station and triggers the photoelectric sensor, the multimodal data acquisition module acquires one or more frames of conventional structured light images and simultaneously acquires a sequence of multiple orthogonal polarized light images. The conventional structured light images are used to present the outer contour, inner contour, and thread geometric boundaries of the bottle mouth. The orthogonal polarized light image sequence is used to present the changes in light intensity, color, or phase caused by stress inside the transparent material. The communication interface also reads the current batch assembly process parameters from the production line control system, such as the target torque of the capping machine, the material parameters of the sealing gasket, the bottle type number, and the preset risk threshold.

[0038] The optical distortion decoupling module performs polarization state change analysis on the orthogonally polarized light image sequence to obtain the local refractive index change relative to the refractive index of the standard material. For ease of explanation, it is assumed that the standard refractive index of the standard polyethylene terephthalate copolymer material in the current wavelength band is 1.570, and the bottle mouth detection area is divided into... Each micro-block corresponds to a local area of ​​the bottle neck sealing surface;

[0039] If the orthogonal polarization image shows that the phase lag of the central block is greater than the preset phase difference threshold, the local refractive index of the block can be estimated as 1.573, and its change in local refractive index relative to the standard value is +0.003; if the upper right block is estimated as 1.568, the change is -0.002; when the phase lag of the block is lower than or equal to the preset phase difference threshold, the change in local refractive index of the block is determined to be zero, and the standard refractive index remains unchanged; the system does not need to convert every pixel into an accurate material property value, but can map the micro-block level change to the edge coordinate compensation amount;

[0040] The optical distortion decoupling module extracts the bottle mouth edge from the structured light image. For example, if the coordinates of the outer edge of the bottle mouth are detected at a certain cross section in the structured light image as pixels 120.4 and 420.7, the apparent outer diameter is calculated to be 30.03 mm according to the calibration ratio. If the cross section passes through a region with a local refractive index change of +0.003, the system determines that there is a visual offset of +0.015 mm on the left edge and a visual offset of -0.010 mm on the right edge according to the pre-calibrated optical distortion model. Then, the true outer diameter after inverse compensation is no longer directly taken as 30.03 mm, but is recalculated according to the corrected physical coordinates of the two edges to obtain, for example, a true absolute size of 30.005 mm.

[0041] The stress field reconstruction module extracts two-dimensional photoelastic interference fringes based on orthogonal polarized light image sequences and converts these fringes into a three-dimensional residual stress distribution tensor map. For ease of understanding, the bottle mouth region can be simplified into three layers: upper, middle, and lower, each containing three stress sampling points on the left, middle, and right. If the two-dimensional fringes appear as higher-order fringes at the root of the thread on the left side and as lower-order fringes on the sealing surface on the right side, then in the converted three-dimensional residual stress distribution tensor map, the principal stress difference corresponding to the root of the thread on the left side is greater than the first preset stress value, while the difference is less than the second preset stress value on the sealing surface on the right side. Each sampling point can store stress components in the corresponding spatial direction, such as stress information in the radial, circumferential, and axial directions. This tensor map serves as the internal material state input for subsequent assembly deformation calculations.

[0042] The assembly simulation calculation module uses the actual absolute size as the geometric boundary of the target detection digital twin, the three-dimensional residual stress distribution tensor map as the initial state of the material, and combines it with assembly process parameters such as the screw cap torque for simulation calculation; the simulation can use a lightweight finite element mesh or a pre-calibrated equivalent mechanical response table;

[0043] Taking a simplified mesh as an example, the bottle neck sealing surface is divided into 8 circumferential nodes. After applying the standard capping torque, the system calculates the displacement of each node. If the axial sinking of node 1 and node 2 is 2 micrometers and 3 micrometers respectively, while the axial tilting of node 5 is 8 micrometers, then the sealing surface under assembly stress is no longer the ideal plane in static measurement, but forms an asymmetric dynamic micromorphology.

[0044] The risk quantification and feedback module compares the fit between the dynamic micro-morphology and the pre-stored sealing model and outputs the seal failure probability index. The sealing model can be a three-dimensional contour model of the gasket, inner plug, or inner sealing ring corresponding to the current bottle type. When generating the seal failure probability index, the system uses the effective contact area ratio and the maximum local gap as the core calculation inputs, and outputs an index between 0 and 1 according to the negative weighting, positive weighting, and interval mapping rules described later.

[0045] As for information such as whether the stress concentration location falls in the critical sealing area and whether the contact continuity is interrupted, it is given priority as an explanatory label, a re-inspection trigger condition, or a basis for equipment process feedback, and does not replace the scoring status of the aforementioned two core quantities; if the index is 0.82 and the preset risk threshold is 0.70, the system determines that there is a risk of sealing failure that exceeds the acceptable range, outputs a rejection instruction to the rejection mechanism, and generates equipment process adjustment information; if the index is 0.35, a qualified release instruction is output, allowing the bottle to enter the downstream capping station;

[0046] Furthermore, when overexposure, underexposure, or missing stripes exist in the orthogonal polarized light image sequence, the system can perform anomaly handling: if the refractive index change cannot be calculated in a local area, the interpolation result of the adjacent area is used and the confidence level is marked; if the confidence level of multiple consecutive key areas is lower than the preset value, a qualified release judgment is not made directly, but a re-inspection instruction or a conservative rejection instruction is output; if the communication interface fails to read the assembly process parameters of the current batch, the system calls the default parameters of the bottle type in the local cache and marks the result as a parameter degradation state; if the local cache also does not exist, the risk index output is stopped, only the size detection result is retained, and a parameter missing alarm is sent to the production line control terminal.

[0047] On a production line that inspects 180 transparent serum bottles per minute, a structured light image of a certain bottle shows that the outer diameter of the bottle mouth and the flatness of the sealing surface are within the static tolerance. However, the orthogonal polarized light image sequence shows obvious photoelastic fringes at the beginning of the thread. After the system confirms that the actual absolute size is still qualified after reverse optical compensation, the three-dimensional residual stress distribution tensor map shows stress concentration at the root of the thread.

[0048] Simulation calculations revealed that after the standard capping torque was applied, a slight upward tilt occurred on one side of the sealing surface, resulting in a decrease in the seal fit and a final seal failure probability index of 0.78. Since this index was higher than the threshold of 0.70, the system output a rejection command and sent the stress concentration location and batch information corresponding to the bottle to the production control terminal.

[0049] The purpose of this mechanism is to integrate the apparent size detection, optical refraction compensation, residual stress identification, and post-assembly sealing risk assessment of transparent cosmetic bottles into a single closed loop. This allows the system to not only determine whether the current size is up to standard, but also to predict whether it can still reliably seal under actual capping conditions, thereby reducing the risk of misjudgment caused by latent stress and optical distortion in transparent bottles.

[0050] Furthermore, the preset assembly process parameters include standard torque parameters and preset elastic modulus of the sealing material; conventional structured light images contain the external contour features and geometric dimensions of the target object.

[0051] This embodiment provides a mechanism for the collaborative input of assembly process parameters and structured light geometric features. Specifically, in the aforementioned polyethylene terephthalate copolymer transparent essence bottle production line, if the simulation calculation relies only on the bottle mouth geometry without considering the capping torque and the elastic modulus of the sealing element, the same bottle mouth shape will result in the same risk assessment under different cap types and different gasket materials, which is inconsistent with the actual assembly results.

[0052] For example, a silicone gasket with an elastic modulus lower than the first preset value can absorb a certain amount of minor warping of the sealing surface, while a polyethylene gasket with an elastic modulus higher than the second preset value is more sensitive to local gaps; a torque greater than the preset nominal value may improve local fit, or it may amplify the deformation caused by stress release at the bottle mouth; therefore, this embodiment introduces standard torque parameters and the elastic modulus of the sealing material to make the dynamic micromorphology calculation consistent with the actual assembly conditions.

[0053] Specifically, the communication interface reads the standard torque parameter corresponding to the current bottle type from the capping equipment or production control terminal. This parameter can be a single target value, such as 0.45 N·m, or a parameter range including upper and lower limits, such as 0.43 to 0.47 N·m. The system also reads the elastic modulus of the preset sealing material, for example, the elastic modulus of an inner stopper seal in a certain batch is 3.2 MPa, and the elastic modulus of a hard gasket in a certain batch is 18 MPa. The above parameters are bound and stored with the bottle type number, cap type number, and material batch number to avoid using incorrect parameters when switching between different products.

[0054] To maintain consistent terminology throughout the text, in this embodiment and subsequent embodiments, any expressions such as target torque of the capping device, standard capping torque, or currently invoked torque value shall be regarded as specific configuration forms, instance values, or execution values ​​derived from the standard torque parameter, rather than new parameter names listed alongside the standard torque parameter.

[0055] Conventional structured light imaging is used to extract the external contour features and geometric dimensions of the target object. External contour features may include the outer edge of the bottle mouth, the transition area of ​​the bottle shoulder, the outer contour of the thread, and the positioning reference surface. Geometric dimensions may include the outer diameter, inner diameter, bottle mouth height, thread pitch, thread height, and sealing surface width.

[0056] For ease of explanation, it is assumed that the structured light image, after calibration, yields a sequence of points representing the cross-sectional profile of the bottle mouth, with the left outer edge being coordinate point A, the right outer edge being coordinate point B, and the inner edges being coordinate points C and D. The system uses the distance from A to B to calculate the outer diameter, the distance from C to D to calculate the inner diameter, and the axial distance between adjacent thread peaks to calculate the pitch. These features serve as the geometric boundaries of the simulation model.

[0057] Furthermore, as an anomaly handling mechanism, when a production line changes its model but the production control terminal fails to issue new torque or sealing parameters in a timely manner, the system will cross-validate the bottle type number with the image recognition results. For example, if the structured light image identifies the current bottle opening outer diameter as approximately 30 mm, but the process parameters transmitted through the communication interface correspond to a small bottle type with an outer diameter of 24 mm, the system will not use this parameter for direct calculation. Instead, it will generate a parameter mismatch alarm and trigger manual confirmation or match the closest bottle type from the local parameter library.

[0058] If the elastic modulus of the seal is missing, the system can call the preset safety lower limit parameter of the material type to perform a risk assessment and mark the output result as a material parameter downgrade to prevent low confidence results from being used as the final release basis.

[0059] In the main production line, the same batch of transparent essence bottles were first tested using soft silicone sealing rings, and then switched to hard anti-volatile sealing sheets. The actual geometric dimensions of the bottle openings extracted by structured light images were basically the same, but the elastic modulus of the seal changed from 3.2 MPa to 18 MPa when read by the communication interface. When the system calculated the fit, the hard sealing sheet had a lower ability to compensate for local micro gaps, and therefore output a higher seal failure probability index for the same dynamic micromorphology.

[0060] The purpose of this mechanism is to ensure that visual measurement results are not divorced from actual assembly conditions, and to avoid using static geometric dimensions in isolation as the basis for release, thereby improving the consistency of risk assessment under different cover types, different gasket materials, and different torque settings.

[0061] Furthermore, the optical distortion decoupling module includes: a refractive index calculation unit, used to calculate the local refractive index change inside the target object by analyzing the polarization state change in the orthogonal polarized light image sequence; an edge extraction unit, used to extract the initial size edge from the conventional structured light image; and an inverse compensation unit, used to construct an optical distortion model based on the local refractive index change, input the initial size edge into the optical distortion model for coordinate offset correction, and output the true absolute size.

[0062] This embodiment provides an optical distortion decoupling mechanism in the measurement of the edge of a transparent bottle; specifically, in the aforementioned detection process of a transparent essence bottle made of polyethylene terephthalate copolymer, if the bright and dark boundaries in the structured light image are directly used as the physical boundaries of the bottle opening, the refractive index change generated in the high-stress area may cause the edge position to shift.

[0063] This offset is not obvious in the inspection of ordinary opaque parts, but it can cause the system to mistake the refracted optical edge for the real physical edge in the mouth of a thick-walled transparent bottle. Therefore, this embodiment separates the polarization state change analysis, initial edge extraction and reverse coordinate compensation into continuous processing steps.

[0064] Specifically, the refractive index calculation unit reads the orthogonally polarized light image sequence; the image sequence may include multiple frames of images acquired at different polarization angles, such as images corresponding to polarization directions of 0 degrees, 45 degrees, 90 degrees and 135 degrees; the system compares the gray level or phase difference of the same area under different polarization directions to obtain the intensity of polarization state change in the area, and converts it into the amount of local refractive index change according to the calibration curve;

[0065] For ease of explanation, the bottle opening detection area is divided into: The local refractive index changes of the nine blocks can be represented by a numerical distribution from small to large, for example, +0.001 for the upper left block, +0.003 for the central block, and -0.001 for the lower right block; this distribution reflects the degree of disturbance to the light propagation path inside the transparent material.

[0066] The edge extraction unit extracts the initial size edges from the conventional structured light image. During processing, camera calibration and distortion correction can be performed first, and then edge point sequences can be obtained based on the gray-level gradient of the structured light stripes, sub-pixel edge detection, or contour fitting algorithms. For example, on a cross-section of a bottle opening, the system scans from left to right to obtain the gray-level gradient peak points, selects the first stable peak as the left outer edge point, and selects the second stable peak as the right outer edge point. If the calibration ratio is 0.02 mm per pixel, the left outer edge is 100.25 pixels, and the right outer edge is 1600.75 pixels, then the initial apparent outer diameter is 30.01 mm.

[0067] The reverse compensation unit constructs an optical distortion model based on the local refractive index change. This model can be a lookup table model, a linear compensation model, or a nonlinear mapping model established by calibrating a sample. For example, the system is calibrated in advance using a transparent standard with known real size to determine that when the local refractive index change is +0.003 and the incident angle of light is within a certain range, the edge will shift outward by about 0.012 mm.

[0068] When the change is -0.001, the edge will shift inward by about 0.004 mm. During actual detection, the initial edge point falls near the central block, and the system applies the corresponding compensation to the edge coordinates. If the left edge needs to be corrected to the right by 0.012 mm and the right edge needs to be corrected to the left by 0.006 mm, the outer diameter calculation will deduct the offset on both sides and output the corrected true absolute size.

[0069] Furthermore, when an edge point spans multiple refractive index blocks, the system can perform a weighted average based on the refractive index changes in the surrounding neighborhood of the edge point to avoid abrupt compensation changes due to block boundaries. If a region in the polarization image has an abnormally high local refractive index change due to reflection from the bottle surface, the system will perform consistency verification between this region and its adjacent regions. If the difference exceeds a preset range and no corresponding fringe changes appear in adjacent regions, the region will be marked as reflective interference, and its compensation weight will be reduced. If structured light edge extraction fails, for example, due to excessive thread reflection causing unstable gradient peaks, the system will retain the polarization measurement results but will not output the true absolute size, triggering a re-shoot or re-inspection process.

[0070] In the main production line, a structured light image of a transparent essence bottle showed that the outer diameter of the bottle opening was 30.04 mm, close to the upper limit tolerance; the polarization image sequence showed that there was a local increase in refractive index on the left sealing surface, and the refractive index calculation unit obtained a change of +0.003; after reverse compensation, the system confirmed that the actual position of the left edge was further inward than the position shown in the visual image, and the true outer diameter was corrected to 30.018 mm, thus avoiding misjudging the apparent deviation caused by optical refraction as the actual dimensional deviation.

[0071] The purpose of this mechanism is to separate the optical path offset caused by the internal stress of transparent materials from the geometric dimension measurement, thereby improving the reliability of the actual size detection of transparent or semi-transparent cosmetic bottles and providing a geometric boundary that is closer to the physical entity for subsequent assembly simulation.

[0072] Furthermore, the stress field reconstruction module includes: a fringe analysis unit, used to perform phase unwrapping processing on the orthogonal polarized light image sequence to extract two-dimensional photoelastic interference fringes; and a tensor transformation unit, used to map the two-dimensional photoelastic interference fringes into spatial principal stress differences based on the photoelastic law, and to construct a three-dimensional residual stress distribution tensor map of a specific region in the target object; wherein, the specific region includes the threaded region and the sealing surface region.

[0073] This embodiment provides a mechanism for reconstructing the residual stress distribution in the key area of ​​the bottle mouth from two-dimensional photoelastic fringes; specifically, in the aforementioned detection link, if only the local refractive index change is used to correct the size edge, it is still impossible to determine whether asymmetric deformation will occur after the cap is screwed on.

[0074] The reason is that changes in refractive index can reflect the degree of optical distortion, but sealing failures often occur in the threaded area and the sealing surface area: the threaded area bears the meshing force of the cap, and the sealing surface area is in direct contact with the sealing element; therefore, in this embodiment, the photoelastic interference fringes in the orthogonal polarized light image sequence are converted into spatial stress information, focusing on covering the threaded area and the sealing surface area.

[0075] Specifically, the fringe resolution unit performs phase unwrapping processing on the orthogonal polarized light image sequence; in orthogonal polarized images, regions with high residual stress will form alternating bright and dark stripes or color-changing stripes; since the phase usually has periodic jumps, for example from close to If the position suddenly returns to 0, directly using the original phase will lead to errors in stripe order recognition; the role of phase unwrapping is to unfold this periodic jump into a continuous phase distribution;

[0076] Taking simplified data as an example, the original phase of a certain row of pixels can be represented as 0.2, 1.8, 3.0, 0.4, 1.6, where a periodic wrap-around occurs after the third number; after processing, it can be obtained as 0.2, 1.8, 3.0, 6.7, 7.9, enabling the system to recognize that the fringe order at this position is continuously increasing, rather than mistakenly believing that the stress suddenly decreases;

[0077] Tensor transformation unit maps two-dimensional photoelastic interference fringes to spatial principal stress differences based on the photoelastic law; in engineering implementation, the fringe order or phase delay can be converted into principal stress differences by using the photoelastic constant of the bottle mouth material, the sample thickness, or the equivalent optical path length.

[0078] For ease of explanation, it is assumed that the sealing surface is divided into three sampling areas: left, middle, and right. The order of the stripes after unwrapping is 1, 2, and 1, respectively; the order of the corresponding area at the root of the thread is 3, 4, and 2, respectively. The system can determine that the principal stress difference is the highest in the middle area at the root of the thread. The system combines the three-dimensional geometric model of the bottle mouth and maps the stripe information on the two-dimensional image plane to the circumferential, radial, and axial positions of the bottle mouth to form a three-dimensional residual stress distribution tensor map.

[0079] The three-dimensional residual stress distribution tensor map can be stored using mesh nodes or voxel elements; each node or voxel records the stress components associated with that location, at least characterizing the local principal stress difference and its directional features; for example, the stress value of a voxel in the bottle neck thread region is marked in the first stress interval, the stress value of a voxel at the edge of the sealing surface is marked in the second stress interval, and the stress value of a voxel at the center of the sealing surface is marked in the third stress interval; this map does not require laboratory-level absolute stress measurement accuracy in every tiny area, but is used to support the calculation of deformation trends under assembly stress conditions;

[0080] Furthermore, it should be noted that the three-dimensional residual stress distribution tensor map in this embodiment is an engineering equivalent tensor map for production line prediction purposes, rather than the material's full six-component true three-dimensional stress tensor which must be directly obtained from a single two-dimensional image; its construction process can adopt a joint reconstruction method of two-dimensional fringe principal stress difference + bottle mouth local thickness information + bottle mouth axisymmetric geometric constraints + key area stress prior: first, the principal stress difference distribution at each sampling position is obtained from two-dimensional photoelastic fringes, and then a layered mapping in the thickness direction is established based on the bottle mouth local thickness, thread direction and sealing surface normal, and the equivalent stress components in the radial, circumferential and axial directions are given in the thread area and sealing surface area respectively;

[0081] For locations where the thickness direction is difficult to observe directly, the system prioritizes outputting the equivalent principal stress difference and its principal direction, and expands it into a three-dimensional voxel field through pre-calibrated interlayer distribution rules to ensure that the subsequent assembly simulation uses engineering inputs with consistent spatial distribution, rather than arbitrary extrapolation that is detached from physical constraints.

[0082] For example, in the sealing surface area where the local thickness variation is small at the bottle mouth, the system can assign different weights to the principal stress difference of the same plane sampling point according to three layers: upper, middle and lower. The layer weight near the inner surface is used to characterize the constraint sensitivity caused by contact with the inner plug or sealing ring, and the layer weight near the outer surface is used to characterize the effect of cap tightening and external cooling contraction. In the threaded area, the higher-order stripe area is preferentially allocated to the circumferential and axial stress components corresponding to the thread root, based on the spatial orientation of the thread profile.

[0083] Although the tensor diagram constructed in the above manner is an equivalent reconstruction result, its spatial distribution is consistent with the bottle neck geometry, thickness and assembly path, and can stably reflect where the stress is high, which direction is more sensitive, and which side is more likely to warp when screwing on the cap.

[0084] Furthermore, as an anomaly handling mechanism, when the stripe density is too high and there are multiple possible paths for phase unwrapping, the system can introduce structured light geometric boundaries and the temporal continuity of adjacent frames for constraints; for example, the stripe order change of adjacent polarization angle images in the same region should be continuous. If a region suddenly jumps by more than two orders, local recalculation is required.

[0085] If an abnormal photoelastic stripe is caused at a certain point on the bottle neck due to bubbles, scratches, or stains, the system can mark that point as a defect interference area and reduce its weight during tensor transformation. If the interference area covers the critical contact area of ​​the sealing surface by more than a preset proportion, the system will output a re-inspection or rejection suggestion instead of continuing the assembly simulation with an unreliable stress diagram.

[0086] Furthermore, if only the principal stress difference can be obtained for the critical region and the directional components cannot be stably distinguished, the system can still generate a reduced-order three-dimensional residual stress distribution tensor map. That is, the principal stress difference is used as the core field, and the local geometric normal and thread tangent are used as directional surrogates to write into the voxel or node attributes, and the result is marked as an equivalent tensor mode. When the subsequent assembly simulation reads this mode, it uses the corresponding reduced-order material response table, and does not mistakenly regard this region as having obtained complete experimental tensor data.

[0087] In the main production line, the bottle mouth size of a certain bottle was qualified after compensation, but the polarization image showed dense stripes below the starting end of the thread. After the stripe analysis unit completed the phase unwrapping, it was found that the order of the stripes at this location was higher than that of the adjacent area. The tensor transformation unit mapped the high-order stripes to the circumferential stress concentration at the root of the thread and marked it as a high-risk area in the three-dimensional residual stress distribution tensor map. In subsequent simulations, this area produced a larger local displacement under the action of standard torque, which affected the sealing surface fit.

[0088] The purpose of this mechanism is to convert photoelastic stripes, which can only be observed as two-dimensional visual textures, into three-dimensional residual stress inputs that can participate in mechanical calculations. This allows the system to analyze the actual deformation trend of the bottle neck during the capping process, rather than relying solely on static geometric appearance.

[0089] Furthermore, the assembly simulation calculation module includes: a digital twin construction unit, used to construct a target detection digital twin with the actual absolute size as the geometric boundary and the three-dimensional residual stress distribution tensor map as the internal material property; and a finite element solution unit, used to apply standard torque parameters as boundary loads to the target detection digital twin, calculate the nodal displacement data after being subjected to force through finite element analysis, and reconstruct the dynamic micromorphology based on the nodal displacement data.

[0090] This embodiment provides a digital twin simulation mechanism for bottle necks based on actual dimensions and residual stress. Specifically, in the aforementioned production line, if only the actual absolute dimensions are used for static tolerance judgment, it is still impossible to reflect the morphological changes after the capping machine applies torque; if only the residual stress diagram is used for empirical scoring, it is difficult to reflect the constraint effect of different bottle neck geometric boundaries on deformation; therefore, this embodiment constructs geometric boundaries with actual absolute dimensions, assigns the internal state of the material with a three-dimensional residual stress distribution tensor diagram, and uses the standard torque parameter as the boundary load for finite element solution.

[0091] Specifically, the digital twin construction unit constructs a target detection digital twin based on the corrected outer diameter, inner diameter, sealing surface width, thread pitch, and bottle mouth height. To meet the real-time requirements of the production line, the system can build a local model only for the key areas of the bottle mouth, without having to model the entire bottle body. In the model, the sealing surface, thread crest, thread root, and inner wall of the bottle mouth are divided into finite element nodes and elements. The three-dimensional residual stress distribution tensor is mapped to the corresponding element, so that each element not only has a geometric position but also an initial stress state.

[0092] For ease of explanation, the sealing surface can be simplified to 8 circumferential nodes, which are denoted as 8 calculation points arranged along the circumference; the actual absolute size determines the position of these 8 nodes in the initial state; the three-dimensional residual stress distribution tensor map indicates that the circumferential residual stress near nodes 1 to 2 is greater than the first preset stress threshold, and the stress near node 5 is less than the second preset stress threshold; the digital twin building unit combines this information into a local bottle mouth model that can be subjected to stress calculation;

[0093] The finite element solution unit applies the standard torque parameter as a boundary load to the digital twin; for example, the standard torque parameter is 0.45 N·m, which the system converts into circumferential force, axial clamping force or equivalent boundary constraint of the threaded region according to the thread contact position; after solving, the displacement data of each node is obtained;

[0094] In the simplified example, the axial displacements of the eight sealing surface nodes are -2, -3, -1, +1, +8, +6, 0, and -1 micrometers, respectively. Positive values ​​indicate upward tilting relative to the sealing element, while negative values ​​indicate being compressed. The system reconstructs the dynamic micromorphology based on these node displacement data to obtain the actual contour of the sealing surface in the screw cap assembly state.

[0095] To balance computational speed, finite element analysis can employ a pre-mesh template and local parameter update method. That is, the mesh topology of the same bottle type is pre-stored, and only key dimensions, residual stress distribution, and torque parameters are updated during real-time detection. For qualified areas with small stress changes, a mesh of the first preset size can be used. For high-stress areas such as the thread root and sealing surface, a locally refined mesh can be used. This avoids production line waiting caused by full-scale high-precision simulation, while preserving the accuracy of the topography calculation of key sealing areas.

[0096] Furthermore, when reading the three-dimensional residual stress distribution tensor map, the digital twin building unit can distinguish between two types of input: the complete tensor mode and the equivalent tensor mode. For the complete tensor mode, the finite element unit directly loads the initial stress components in the corresponding direction. For the equivalent tensor mode, the principal stress difference, principal direction proxy quantity, and region type are mapped to the initial strain or equivalent preload of the finite element according to the local geometric coordinate system of the bottle mouth.

[0097] In other words, subsequent solutions do not require all elements to have complete laboratory-grade tensor data. Instead, reduced-order expressions using principal stress difference plus directional proxy are allowed in key areas such as the sealing surface and thread root, as long as the expression can drive local deformation in a way that is consistent with the geometric boundary, material properties and torque load.

[0098] For example, for the sealing surface region, the system can preferentially transform high-stress elements in the equivalent tensor mode into an axially warping-sensitive initial field; for the threaded region, the system can preferentially transform similar high-stress elements into a circumferential meshing deformation-sensitive initial field. In this way, during finite element solution, stress concentration at the root of the thread is more likely to manifest as local contact stiffness differences and uneven axial compression, while high stress at the sealing surface is more likely to manifest as node warping or local subsidence. This processing method matches the engineering equivalent tensor diagram in the aforementioned stress field reconstruction, avoiding the inconsistency in context where the upstream only obtains two-dimensional stripe information while the downstream is forcibly solved according to the complete three-dimensional accurate material tensor.

[0099] Furthermore, as an anomaly handling mechanism, when the finite element solution fails to converge, the system does not directly output a low-risk conclusion, but instead enters a degraded solution process; for example, it first relaxes the local element mass constraints of the mesh and recalculates; if it still fails to converge, it uses an equivalent stiffness response table to estimate the nodal displacements and reduces the confidence level of the results.

[0100] If the input standard torque parameter exceeds the allowable range for the bottle type, for example, if the control terminal mistakenly transmits 1.2 N·m while the allowable range for the bottle type is 0.3 to 0.6 N·m, the system will refuse to use the parameter to generate the normal risk index and will output an alarm for abnormal process parameters. If some stress voxels cannot be mapped to finite element elements, the system will use neighborhood interpolation or regional mean value substitution and record the missing proportion. When the missing proportion exceeds the preset upper limit, a re-inspection or rejection will be triggered.

[0101] Furthermore, if a region can only provide an equivalent principal stress difference but cannot stably provide a directional surrogate, the system will not perform a full finite element prestressing load on that region. Instead, it will call a pre-calibrated local response table of the same bottle type, correspond the principal stress difference to a set of candidate node displacement corrections, and take the result that best matches the displacement continuity of the surrounding elements to participate in the dynamic micromorphological reconstruction.

[0102] In the main production line, the actual outer diameter of a transparent essence bottle made of polyethylene terephthalate copolymer is 30.005 mm and the width of the sealing surface is 1.20 mm, both within the tolerance range. The three-dimensional residual stress distribution tensor diagram shows that there is local high stress in the thread initiation section. After the digital twin construction unit generates a local model of the bottle opening, the finite element solution unit applies a standard torque of 0.45 N·m. The solution results show that the sealing surface on the opposite side of the thread initiation section has an upward warping of 6 to 8 micrometers. Based on this, the system reconstructs the non-uniform dynamic micromorphology.

[0103] The purpose of this mechanism is to unify static dimensions, internal stress, and assembly load into a single calculable model, thereby obtaining the microscopic profile of the bottle opening under actual capping stress, and providing a physical basis for subsequent judgment of sealing fit.

[0104] Furthermore, when performing fit comparison, the risk quantification and feedback module is specifically used to: extract the contour of the sealing contact surface in the dynamic micromorphology;

[0105] Perform a Boolean intersection operation between the sealing contact surface contour and the preset sealing component model to calculate the effective contact area ratio and the maximum local gap value.

[0106] Obtain the negative and positive weighting factors pre-calibrated based on historical assembly leakage test data. Combine the negative weighting factors to calculate the negative weighted value of the effective contact area ratio, and combine the positive weighting factors to calculate the positive weighted value of the maximum local gap value. Sum the negative and positive weighted values ​​to obtain the intermediate score. Use a preset bias to linearly trim the intermediate score and limit it to the range of zero to one, and output the seal failure probability index.

[0107] This embodiment provides a quantitative mechanism for sealing contact surface fit. Specifically, in the aforementioned simulation link, the dynamic micromorphology itself is only node displacement or surface contour data. If it is not further converted into sealing-related indicators, the production line cannot directly make a release or rejection judgment based on it. Especially in some cases, the sealing surface has local upturn but is still covered by soft sealing components.

[0108] In other cases, the overall deformation is not significant, but sharp gaps exist on the critical sealing path. Therefore, this embodiment quantifies the sealing failure probability index by combining the effective contact area ratio and the maximum local gap value. To maintain consistency with the full text, the core calculation method of the sealing failure probability index in this embodiment is fixed as a negative weighted average of the effective contact area ratio plus a positive weighted average of the maximum local gap value before mapping. Other information such as stress concentration location, contact continuity, and area labels are only used as auxiliary information for explanation, early warning, re-inspection, or feedback, and do not replace the above two core input quantities.

[0109] Specifically, the risk quantification and feedback module first extracts the sealing contact surface profile from the dynamic micromorphology; this profile can be the annular area where the sealing surface contacts the sealing element after the cap is subjected to force; the system discretizes this annular area into multiple patches or sampling points, for example, simplifying a sealing surface into 12 sectors, each sector recording the height and normal direction after being subjected to force; the preset sealing element model includes the contact surface shape, compressibility and material elastic parameters of the sealing ring or gasket under pressure.

[0110] The system performs Boolean intersection calculations on the contour of the sealing contact surface and the preset sealing component model. Boolean intersection means calculating the area where the bottle mouth sealing surface and the sealing component overlap or are effectively pressed together in space. If the sealing surface of a certain sector enters the compressible range of the sealing component, the sector is considered to be in effective contact. If there is still a gap between a certain sector and the sealing component, the sector is considered to be ineffective or inefficient contact.

[0111] In the simplified example, 9 out of 12 sectors form effective contact, so the effective contact area ratio is 9 / 12, or 0.75; if the maximum gap in the non-contact sectors is 0.018 mm, then the maximum local gap is 0.018 mm.

[0112] The seal failure probability index is calculated by combining the effective contact area ratio and the maximum local gap value. The higher the effective contact area ratio, the lower the risk, so a negative weighting factor is used; the larger the maximum local gap value, the higher the risk, so a positive weighting factor is used. For example, the system sets the negative weighting factor corresponding to the effective contact area ratio to -0.6, and the positive weighting factor corresponding to the normalized maximum local gap value to 0.8. If the effective contact area ratio is 0.75, the corresponding negative weighting value is -0.45.

[0113] The maximum local gap of 0.018 mm is normalized to 0.60 under the preset upper limit of 0.030 mm, corresponding to a positive weighted value of 0.48; the sum of the two is 0.03; the system then combines the sum with the basic bias or mapping function to map it to a preset range of 0 to 1; for example, after mapping, it is 0.56, which indicates a medium risk of seal failure; if the effective contact area ratio further decreases to 0.50 and the maximum gap increases to 0.025 mm, the mapping result may increase to 0.82;

[0114] Furthermore, to avoid the summation and remapping process becoming a closed scoring process that is difficult to trace, the system can adopt a pre-fixed interval mapping rule; specifically, the upper limit normalization of the maximum value of the local gap can be performed first, and the normalization result is denoted as... ,in And when the maximum local gap is greater than the preset gap upper limit, take ; Calculate the intermediate score based on the rating value:

[0115]

[0116] The intermediate scores are linearly clipped to the 0-1 range using a fixed bias, for example, by taking the seal failure probability index. Therefore, in the aforementioned examples ,but If the effective contact area ratio drops to 0.50, and the maximum local gap is 0.025 mm, then Approximately 0.83, under the same weights Approximately 0.37, after mapping Approximately 0.90; the above bias values, weighting factors and preset gap upper limits can all be pre-calibrated according to bottle type, seal type and historical leakage test results, and kept fixed or switched by batch during operation to prevent the same bottle type from using different scoring calibers at different times;

[0117] Furthermore, when performing Boolean intersection operations, the system does not only determine whether there is geometric contact, but also uses contact continuity as an auxiliary verification condition. That is, even if two separate sectors form contact respectively, as long as there is a continuous non-contact zone in between, the system will still regard the circumferential path corresponding to the non-contact zone as a potential leakage channel, and reflect it in the final score through the maximum local gap or the continuous non-contact length factor.

[0118] However, without changing the calculation core defined in the embodiment, the auxiliary verification result is preferentially used to adjust the value of the maximum value of the local gap or for re-inspection and judgment, without changing the basic structure of negative weighting of the effective contact area ratio + positive weighting of the maximum value of the local gap and summing and mapping.

[0119] Furthermore, to avoid ambiguity in terminology, the effective contact area ratio in this embodiment always refers to the ratio of the effective contact area obtained after the sealing contact surface contour and the preset sealing element model are intersected by Boolean to the total evaluated area of ​​the sealing contact surface, and will no longer be replaced by other names such as fit ratio or contact coverage; the maximum local gap always refers to the maximum local gap value in the area where no effective contact is formed in the same Boolean intersection evaluation.

[0120] Furthermore, when the Boolean intersection operation yields an effective contact area ratio of 0, the system no longer performs conventional linear scoring but directly raises the seal failure probability index to a high-risk range, such as no less than 0.90. If the maximum local gap value is 0 but the effective contact area ratio is low, the system will not simply determine it as risk-free but will check whether the contact is concentrated in a few areas to prevent discontinuity in the annular sealing path. If the sealing model is missing or does not match the current bottle type, the system stops the fit comparison and outputs a model missing alarm to the production control terminal. If the confidence level of the input dynamic micromorphology is low, the seal failure probability index can be accompanied by a confidence level label, and samples with low confidence levels and close to the threshold enter the re-inspection process.

[0121] Furthermore, when the calculated intermediate score S is close to the mapping boundary, for example, falling between -0.05 and 0.05, the system can attach an output boundary sample label to indicate that although the sample may not directly exceed the threshold, it is more sensitive to the weight, gap upper limit and sealing material, and is suitable for inclusion in random inspection or review.

[0122] In the main production line, after a bottle completes the capping simulation, the sealing surface is divided into 12 sectors. The system performs a Boolean intersection operation with the current batch of silicone sealing ring models and finds that 10 sectors are in effective contact, with a maximum gap of 0.009 mm. The system outputs a seal failure probability index of 0.32, and the bottle is released. Another bottle, although with similar static dimensions, only has 7 sectors in effective contact after simulation, and there is a 0.024 mm gap on the opposite side of the thread start section. The system outputs an index of 0.79 and triggers rejection.

[0123] The purpose of this mechanism is to transform complex dynamic micromorphology into numerical indicators directly related to the sealing result, enabling the production line to make automatic judgments based on a unified threshold, while retaining two traceable engineering interpretation quantities: effective contact area and local gap.

[0124] Furthermore, when outputting equipment process adjustment information, the risk quantification and feedback module is specifically used for: analyzing the location of stress concentration areas in the three-dimensional residual stress distribution tensor map; generating a cooling temperature adjustment command for the front-end injection molding equipment based on the location of the stress concentration areas, and sending the cooling temperature adjustment command for the front-end injection molding equipment to the external production control terminal through a preset communication bus; generating torque compensation parameters for the downstream capping equipment based on the seal failure probability index, and sending the torque compensation parameters for the downstream capping equipment to the external downstream capping equipment.

[0125] This embodiment provides a two-way process feedback mechanism for both the front-end injection molding equipment and the downstream capping equipment. Specifically, in the aforementioned production line, if the system only outputs release or rejection instructions for a single bottle, it can prevent high-risk products from flowing into subsequent stages, but it cannot solve the problem of stress concentration or capping mismatch that occurs continuously in the same batch. In particular, when the injection cooling temperature is uneven, the cooling channel on one side of the mold is abnormal, or the capping torque is too high, the abnormality will continuously affect multiple subsequent bottles. Therefore, this embodiment converts the detection results into executable process adjustment information and feeds it back to the front-end injection molding equipment and the downstream capping equipment, respectively.

[0126] Specifically, the risk quantification and feedback module analyzes the location of stress concentration areas in the three-dimensional residual stress distribution tensor map; the system can divide the stress map into regions according to the circumferential and height directions of the bottle mouth, for example, dividing the bottle mouth into four circumferential regions: 0 degrees to 90 degrees, 90 degrees to 180 degrees, 180 degrees to 270 degrees, and 270 degrees to 360 degrees, and further distinguishing the threaded region, the sealing surface region, and the inner wall region of the bottle mouth; if multiple consecutive bottles show high stress at the root of the thread between 90 degrees and 180 degrees, the system determines that there is a stable deviation in the corresponding mold or cooling channel in that region;

[0127] Based on the location of the stress concentration area, the system generates a cooling temperature adjustment command for the front-end injection molding equipment. This command can include the target mold cavity number, cooling zone number, adjustment direction, and adjustment range. For example, when the residual stress on one side of the bottle opening is detected to be persistently high, the system can generate a command to reduce the temperature of the cooling zone on the left side of the bottle opening of mold cavity 3 by 1.5 degrees Celsius or increase the cooling water flow rate by a preset ratio. This command is sent to the production control terminal via a preset communication bus, and the production control terminal executes, queues, or manually confirms the command according to the equipment permissions.

[0128] Meanwhile, the system generates torque compensation parameters for downstream capping equipment based on the seal failure probability index. If the overall stress of a batch of bottles is greater than the preset stress threshold but the sealing risk has not yet exceeded the threshold, appropriately reducing the capping torque can reduce micro-cracks or warping caused by stress release. If the sealing risk has reached the preset intervention threshold, it is necessary to increase the torque compensation ratio or stop the batch from entering the capping station.

[0129] Torque compensation parameters can be expressed as an adjustment ratio relative to the standard torque, or as a new target torque value; for example, if the standard torque is 0.45 N·m and the compensation parameter is a 2% reduction, then the target torque of the downstream capping equipment will be adjusted to approximately 0.441 N·m.

[0130] Furthermore, as an anomaly handling mechanism, the system will not output front-end equipment process adjustment instructions that exceed the preset adjustment range based on the anomaly detection results of a single target object. Instead, it can use continuous sample statistics. For example, a cooling temperature adjustment instruction will only be generated when the stress concentration areas of more than 8 bottles out of 20 bottles are located in the same mold partition and the average risk index exceeds the preset observation threshold.

[0131] If the production control terminal reports that the current equipment is in a locked, maintenance, or manual takeover state, the system retains the adjustment suggestions and stops automatically issuing execution commands; if the downstream capping equipment does not confirm receipt of torque compensation parameters, the system can transfer the affected bottles into the buffer track or issue an unconditional rejection instruction to prevent them from continuing to be assembled under unadjusted conditions.

[0132] In the main production line, the system continuously detected high residual stress near the starting section of the thread on the transparent essence bottles produced by mold cavity 2, and the position was stably concentrated around 120 degrees around the bottle mouth circumference. The risk quantification and feedback module mapped this position to the corresponding cooling zone of the injection mold, generated an adjustment suggestion to reduce the cooling temperature by 1 degree Celsius, and sent it to the production control terminal via the communication bus. At the same time, the seal failure probability index of some bottles in the current batch increased to 0.62 but had not yet reached the rejection threshold of 0.70. The system sent a compensation parameter of reducing the torque by 2% to the downstream capping equipment to reduce the risk of deformation in subsequent assembly.

[0133] The purpose of this mechanism is to expand the detection system from single-point quality judgment to production line process control nodes, and to provide feedback on front-end forming conditions through stress location and feedback on downstream assembly conditions through risk index, thereby reducing continuous defects and batch rework.

[0134] Furthermore, torque compensation parameters for the downstream capping equipment are generated based on the seal failure probability index. Specifically, this includes: calculating the global average stress value of the three-dimensional residual stress distribution tensor map; when the global average stress value is higher than a preset stress threshold and the seal failure probability index is lower than a preset risk threshold, generating torque compensation parameters for the downstream capping equipment to reduce the standard torque parameters by a first preset ratio; when the global average stress value is higher than the preset stress threshold and the seal failure probability index is higher than or equal to the preset risk threshold, generating torque compensation parameters for the downstream capping equipment to reduce the standard torque parameters by a second preset ratio, wherein the second preset ratio is greater than the first preset ratio; and when the global average stress value is lower than or equal to the preset stress threshold, maintaining the standard torque parameters unchanged.

[0135] This embodiment provides a graded torque compensation mechanism based on the global average stress and the seal failure probability index. Specifically, in the aforementioned two-way process feedback process, if the capping torque is adjusted only based on whether the seal failure probability index exceeds the threshold, a transitional state may be overlooked: some bottles may still have acceptable fit, but the global residual stress is already too high, and continuing to apply the standard torque will increase the risk of micro-cracks, long-term stress relaxation, or batch fluctuations; conversely, some bottles may have high local risks but not high global stress, and unconditionally reducing the global batch torque parameter may cause the clamping force to fall below the preset lower limit; therefore, this embodiment introduces a joint judgment of the global average stress and the seal failure probability index.

[0136] Specifically, the system first calculates the global average stress from the 3D residual stress distribution tensor map. This average can be calculated for the key area of ​​the bottle opening, without having to include the entire bottle body. For example, the system takes the principal stress difference of all effective voxels in the threaded area and the sealing surface area, removes outliers caused by stains or image interference, and then calculates the average. In a simplified example, a key area of ​​a bottle body has 5 effective stress sampling values, which are 18, 22, 25, 20, and 30 MPa, respectively. The global average stress is 23 MPa. If the preset stress threshold is 20 MPa, the bottle body is considered to have high overall stress.

[0137] The system compares the seal failure probability index with a preset risk threshold. Assuming the preset risk threshold is 0.70 and the first preset ratio is 2%, when the global average stress is higher than 20 MPa and the seal failure probability index is 0.55, it indicates that the current seal risk of the bottle or batch has not yet reached the rejection criteria, but the internal stress level of the material is too high. The system generates a compensation parameter that reduces the standard torque by 2%. If the standard torque is 0.45 N·m, then the target torque after compensation is 0.441 N·m.

[0138] When the global average stress is still higher than 20 MPa and the seal failure probability index reaches 0.76, it indicates that the stress level is too high and the risk of seal failure has exceeded the threshold. The system generates a compensation parameter that reduces the standard torque by 5%, that is, adjusts it from 0.45 N·m to 0.4275 N·m. This compensation parameter can be used by downstream equipment to apply a reduced torque to the assembly of bottles from the same batch or mold cavity. For the current high-risk bottles, the system can still output rejection instructions based on the aforementioned judgment to avoid defective products continuing to flow into the finished product by relying solely on torque reduction compensation.

[0139] When the global average stress is less than or equal to 20 MPa, the system maintains the standard torque parameter unchanged. At this time, even if a few samples have slight local gaps, the system will not automatically reduce the torque of the entire capping station. Instead, it will prioritize handling the issue by rejecting individual bottles, re-inspecting, or checking the matching of the sealing model. This can avoid insufficient capping force for the entire batch of bottles due to a few local anomalies.

[0140] Furthermore, as an anomaly handling mechanism, when the global average stress value is exactly equal to the preset stress threshold, the system processes the stress according to the branch below or equal to the threshold, that is, it maintains the standard torque parameter unchanged, thereby avoiding frequent oscillations near the threshold boundary; if the global average stress value fluctuates around the threshold in a short period of time, the system can introduce a hysteresis interval, for example, only when multiple consecutive bottles exceed the threshold plus the safety margin will torque reduction compensation be activated, and only when multiple consecutive bottles are below the threshold minus the safety margin will the standard torque be restored; if the second preset ratio causes the target torque to be lower than the minimum tightening requirement of the seal, the system will not execute the compensation value, but will limit the target torque to above the minimum safe torque, and at the same time output a stop line or manual confirmation suggestion to the production control terminal;

[0141] In the main production line, the standard capping torque of a batch of transparent essence bottles is 0.45 N·m, the preset stress threshold is 20 MPa, and the preset risk threshold is 0.70. The average global stress of the first bottle is 23 MPa, and the seal failure probability index is 0.58. The system sends a 2% reduction in torque compensation parameters to the capping equipment. Another bottle from the same mold cavity is detected to have an average global stress of 26 MPa and a seal failure probability index of 0.74. The system outputs a rejection command for this bottle, sends a 5% reduction in batch compensation parameters to the capping equipment, and alerts the front-end injection molding control terminal to an abnormal cooling zone. If the average global stress of subsequent continuous samples recovers to below 18 MPa, the system cancels the torque reduction compensation and restores the standard torque of 0.45 N·m.

[0142] The purpose of this mechanism is to avoid excessive or insufficient torque adjustment by linking the global stress level and sealing risk results in a graded manner, so that downstream capping equipment can reduce the risk of assembly damage to high-stress bottle necks while ensuring sealing and clamping force.

[0143] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A cosmetic bottle size detection system based on visual measurement, characterized in that, The cosmetic bottle is a packaging container made of a transparent or semi-transparent material, and the system includes: A multimodal data acquisition module includes a communication interface for acquiring conventional structured light images and orthogonal polarized light image sequences of the cosmetic bottle as the target detection object from an external vision acquisition device, performing spatial coordinate system registration between the conventional structured light images and the orthogonal polarized light image sequences, and acquiring preset assembly process parameters through the communication interface; An optical distortion decoupling module is used to extract the local refractive index change relative to the refractive index of a standard material based on the orthogonal polarized light image sequence, extract the initial size edge from the conventional structured light image, and use the local refractive index change to perform inverse optical compensation on the initial size edge to generate the true absolute size. The stress field reconstruction module is used to extract two-dimensional photoelastic interference fringes based on the orthogonally polarized light image sequence and convert the two-dimensional photoelastic interference fringes into a three-dimensional residual stress distribution tensor map. The assembly simulation calculation module is used to take the actual absolute size and the three-dimensional residual stress distribution tensor map as initial conditions, and combine them with the preset assembly process parameters to perform simulation calculations and generate the dynamic micromorphology under the assembly stress state. The risk quantification and feedback module is used to compare the fit between the dynamic micromorphology and the pre-stored preset sealing component model to generate a sealing failure probability index. The risk quantification and feedback module is also used to determine the execution conditions: when the seal failure probability index is higher than or equal to the preset risk threshold, a rejection instruction is output and equipment process adjustment information is generated; when the seal failure probability index is lower than the preset risk threshold, a qualified release instruction is output.

2. The cosmetic bottle size detection system based on visual measurement according to claim 1, characterized in that, The preset assembly process parameters include standard torque parameters and preset elastic modulus of the sealing material; the conventional structured light image contains the external contour features and geometric dimensions of the target object.

3. The cosmetic bottle size detection system based on visual measurement according to claim 1, characterized in that, The optical distortion decoupling module includes: The refractive index calculation unit is used to calculate the local refractive index change inside the target object by analyzing the polarization state changes in the orthogonal polarized light image sequence. An edge extraction unit is used to extract edges of initial size from the conventional structured light image; The reverse compensation unit is used to construct an optical distortion model based on the local refractive index change, input the initial size edge into the optical distortion model for coordinate offset correction, and output the true absolute size.

4. The cosmetic bottle size detection system based on visual measurement according to claim 1, characterized in that, The stress field reconstruction module includes: The fringe resolution unit is used to perform phase unwrapping processing on the orthogonal polarized light image sequence and extract the two-dimensional photoelastic interference fringes. Tensor transformation unit is used to map the two-dimensional photoelastic interference fringes into spatial principal stress differences based on the photoelastic law, and to construct the three-dimensional residual stress distribution tensor map of a specific region in the target object; The specific area includes the threaded area and the sealing surface area.

5. The cosmetic bottle size detection system based on visual measurement according to claim 2, characterized in that, The assembly simulation calculation module includes: A digital twin construction unit is used to construct a target detection digital twin with the actual absolute size as the geometric boundary and the three-dimensional residual stress distribution tensor map as the internal material property. The finite element solution unit is used to apply the standard torque parameters as boundary loads to the target detection digital twin, calculate the nodal displacement data after being subjected to force through finite element analysis, and reconstruct the dynamic micromorphology based on the nodal displacement data.

6. The cosmetic bottle size detection system based on visual measurement according to claim 1, characterized in that, When performing the fit comparison, the risk quantification and feedback module is specifically used for: Extract the contour of the sealing contact surface from the dynamic micromorphology; Perform a Boolean intersection operation between the sealing contact surface contour and the preset sealing element model to calculate the effective contact area ratio and the maximum local gap value. Obtain the negative and positive weighting factors pre-calibrated based on historical assembly leakage test data. Combine the negative weighting factors to calculate the negative weighted value of the effective contact area ratio, and combine the positive weighting factors to calculate the positive weighted value of the maximum local gap. Sum the negative weighted value and the positive weighted value to obtain an intermediate score. Use a preset bias to linearly trim the intermediate score and limit it to the range of zero to one, and output the seal failure probability index.

7. The cosmetic bottle size detection system based on visual measurement according to claim 2, characterized in that, When outputting the equipment process adjustment information, the risk quantification and feedback module is specifically used for: Analysis of the location of stress concentration regions in the three-dimensional residual stress distribution tensor diagram; A cooling temperature adjustment command for the front-end injection molding equipment is generated based on the location of the stress concentration area, and the cooling temperature adjustment command for the front-end injection molding equipment is sent to an external production control terminal through a preset communication bus. The torque compensation parameters for the downstream capping equipment are generated based on the seal failure probability index, and the torque compensation parameters for the downstream capping equipment are sent to the external downstream capping equipment.

8. The cosmetic bottle size detection system based on visual measurement according to claim 7, characterized in that, The step of generating torque compensation parameters for the downstream capping equipment based on the seal failure probability index specifically includes: Calculate the global stress mean of the three-dimensional residual stress distribution tensor map; when the global stress mean is higher than a preset stress threshold and the seal failure probability index is lower than the preset risk threshold, generate the torque compensation parameter of the downstream capping equipment to reduce the standard torque parameter by a first preset ratio; When the global average stress is higher than the preset stress threshold and the seal failure probability index is higher than or equal to the preset risk threshold, a torque compensation parameter for the downstream capping device is generated to reduce the standard torque parameter by a second preset ratio, wherein the second preset ratio is greater than the first preset ratio. When the global average stress is lower than or equal to the preset stress threshold, the standard torque parameter remains unchanged.

Images