CN-121998904-A - Plastic injection molding defect real-time detection method, system and storage medium

CN121998904ACN 121998904 ACN121998904 ACN 121998904ACN-121998904-A

Abstract

The application relates to the field of industrial detection, and discloses a method, a system and a storage medium for detecting plastic injection molding defects in real time. The method comprises the steps of synchronously imaging a defect area of a plastic injection molding piece through a multi-angle camera array, generating an enhanced defect candidate point characteristic set through preprocessing, generating a defect type set through three-dimensional reconstruction through a stereoscopic vision algorithm and classification by combining a convolutional neural network, generating a defect cause parameter identification set through curvature analysis and surface texture gradient analysis and fusion of a light shadow texture analysis and texture roughness calculation result, generating a defect spatial position evolution trend information set through a dynamic tracking algorithm, and obtaining a mold design improvement data set through fusion of defect type set optimization injection molding process parameters. The application solves the problems of low detection precision, weak interference resistance and no closed-loop optimization of the prior art by fusing multiple technologies such as multi-angle imaging, stereoscopic vision, convolutional neural network, dynamic tracking and the like, and improves the detection efficiency of injection molding defects and the scientificity of process improvement.

Inventors

LUO WEI
Tang Yuandan
LI YONGQIANG
WANG LIUNING
WANG ZHAOXIA
WANG KE
LI JIANYI
ZHAO SHAOHUI
XIE YANTAO

Assignees

洛阳双正塑业有限公司

Dates

Publication Date: 20260508
Application Date: 20251222

Claims (10)

1. The method for detecting the plastic injection molding defects in real time is characterized by comprising the following steps of: s101, synchronously imaging a defect area of a plastic injection molding piece through a multi-angle camera array to obtain an original image set, preprocessing the original image set, extracting a clear feature set, and generating an enhanced defect candidate point feature set; Step S102, three-dimensional reconstruction is carried out on the defect candidate point feature set through a stereoscopic vision algorithm to generate a defect space position information set, a convolutional neural network is adopted to extract features from the defect space position information set, a distinguishing basis of white points and internal bubbles in the surface of the plastic injection molding defect is determined based on the extracted features, defect classification is carried out based on the distinguishing basis, and a classified defect type set is generated; step S103, respectively carrying out curvature analysis and surface texture gradient analysis on the defect type set to obtain defect curvature change detection data, and generating a defect cause parameter identification set according to a texture analysis result fused with the defect curvature change detection data, wherein the texture analysis comprises light shadow texture analysis and texture roughness calculation; step S104, updating the spatial position change of the defect by adopting a dynamic tracking algorithm according to the defect cause parameter identification set, and generating a spatial position evolution trend information set of the defect; and step 105, fusing the defect type sets according to the spatial position evolution trend information sets of the defects, and optimizing injection molding process parameters to obtain a mold design improvement data set.
2. The method for detecting plastic injection molding defects in real time according to claim 1, wherein in the step S101, the obtaining of the original image set includes: disposing the multi-angle camera array at a plastic injection molding detection station; performing preliminary processing on the image data captured by the multi-angle camera array to obtain a processed image set, and taking all images containing micron-sized white point defects in the processed image set as an original image set; Calculating the position uncertainty of the defects in the original image set, and calculating pixel deviation of the same defect in images with different view angles, wherein the position uncertainty is represented by the pixel deviation; Calculating the size variability of defects in the original image set, and calculating the size difference of the same defect in images with different visual angles, wherein the size variability is represented by the size difference; Integrating image data comprising the position uncertainty and the size variability to generate the original image set.
3. The method according to claim 2, wherein in the step S101, generating the enhanced defect candidate point feature set includes: performing gray conversion and filtering treatment on the original image set, removing dust interference and background noise in the original image set, and enhancing image contrast to generate a treated image set; Marking pixels with gray values higher than a preset gray threshold value in the processed image set as white point defect candidate points by a threshold segmentation method; calculating standard deviation of pixel intensity of the area where the white point defect candidate points are located, and determining severity fluctuation; Analyzing the spatial distribution of the white point defect candidate points through spatial cluster analysis to determine distribution non-uniformity; Extracting a clear feature set under the fluctuation of the severity degree and the uneven distribution, wherein the clear feature set comprises an edge contour and gray level distribution features of an area where a white point defect candidate point is located; And associating the clear feature set with the corresponding white point defect candidate points to form the enhanced defect candidate point feature set.
4. The method according to claim 1, wherein generating the defect spatial location information set in step S102 comprises: If the density of the enhanced defect candidate point feature set is higher than a preset density threshold, performing feature point corresponding matching on multi-view data by adopting the stereoscopic vision algorithm; Calculating parallax according to the matching result of the characteristic points, and calculating the three-dimensional coordinates of the white point defect candidate points by combining multi-view geometric constraints, wherein the multi-view geometric constraints are determined by calibration parameters of the multi-angle camera array; correcting the three-dimensional coordinate by fusing temperature change and camera shake influence, wherein the temperature change corrects offset through a material thermal expansion coefficient, and the camera shake influence corrects offset through a dynamic compensation algorithm; and integrating the three-dimensional coordinates of all the corrected white point defect candidate points to generate a defect space position information set.
5. The method according to claim 4, wherein in the step S102, the step of generating the classified defect type set includes: Normalizing the three-dimensional coordinates in the defect space position information set; The normalized defect space position information is concentrated to extract depth features and volume features by using a convolutional neural network, wherein the convolutional neural network comprises a plurality of convolutional layers and a pooling layer, the depth features are extracted by a convolution kernel of the convolutional layers, and the volume features are extracted by three-dimensional space point cloud analysis; The depth features and the volume features are fused by using a convolutional neural network to form a comprehensive feature vector, and the distinguishing basis of the surface white point and the internal bubbles in the plastic injection molding defect is determined based on the comprehensive feature vector; Classifying and judging the comprehensive feature vector through a classification layer of the convolutional neural network, and outputting class probability corresponding to each defect; And determining that the defects are surface white points or internal bubbles according to the category probability, integrating classification results of all the defects, and generating a classified defect type set.
6. The method for detecting plastic injection molding defects in real time according to claim 1, wherein in the step S103, the defect curvature change detection data is obtained, comprising: Associating the image data captured by the multi-angle camera array with a defect region corresponding to the defect type set; Performing curvature analysis on each defect in the defect type set through multi-view geometric reconstruction, and fitting a reconstruction model by adopting a curved surface fitting algorithm to generate a local curvature radius of the defect surface; calculating the surface texture gradient of the defect area by adopting a gradient operator, acquiring the amplitude and direction information of the texture gradient, and constructing the density distribution map of the defect area by calculating the pixel intensity space distribution; Compensating curvature calculation deviation caused by reflection problem through an illumination model, and correcting texture gradient distortion caused by angle deviation based on calibration parameters of the multi-angle camera array; and integrating the local curvature radius, the density distribution mapping, the compensated curvature and the corrected texture gradient analysis result of each defect to generate defect curvature change detection data.
7. The method according to claim 6, wherein generating the defect cause parameter identification set in step S103 comprises: Performing light and shadow texture analysis on the original image set through the illumination model, extracting texture features, and performing texture roughness calculation on the original image set through surface height distribution statistics; Calculating regional density quantization indexes of defects through defect point space distribution according to the shadow texture analysis result and the texture roughness calculation result; integrating the regional density quantization index, the size variability of the defects, the severity fluctuation and the corresponding shadow textures and texture roughness data, and generating a defect cause parameter identification set by combining a preset defect cause mapping rule.
8. The method for detecting plastic injection molding defects in real time according to claim 1, wherein the step S104 comprises: If the defect cause parameter identification set is matched with the real-time production data, a multi-frame image sequence is processed by adopting a dynamic tracking algorithm so as to update the spatial position change of the defect; based on humidity and temperature values in the real-time production data, correcting environmental factors for the spatial position change; carrying out data prediction and optimization on the corrected spatial position change through Kalman filtering and particle filtering to generate accurate defect spatial position change data; Arranging the accurate defect space position change data in time sequence to form a change track of the defect position; Integrating the influence of material-related randomness and injection molding process parameters on the change track to generate an influence factor set, wherein the material-related randomness is random variation of polymer particle distribution in an injection molding material, and the injection molding process parameters are injection molding temperature and pressure data; and optimizing the change track by using the influence factor set, and generating a spatial position evolution trend information set of the defect.
9. The method for detecting plastic injection molding defects in real time according to claim 1, wherein the step S105 comprises: optimizing injection molding process parameters and injection molding temperature by adopting a genetic optimization algorithm according to the spatial position evolution trend information set of the defects and the defect type set and combining a thermodynamic model; Obtaining a pressure distribution equilibrium state and a material flow simulation basis in a mold cavity through hydrodynamic simulation and finite element analysis according to the optimized injection molding process parameters and injection molding temperature; According to the pressure distribution equilibrium state and the material flow simulation basis, adjusting cooling control parameters; Integrating the optimized injection molding process parameters, injection molding temperature, pressure distribution equilibrium state, material flow simulation basis and cooling control parameters to generate the mold design improvement data set.
10. A plastic injection defect real-time detection system, which is used for realizing the plastic injection defect real-time detection method according to any one of claims 1-9, and comprises the following steps: The imaging preprocessing module is used for synchronously imaging the defect area of the plastic injection molding piece through the multi-angle camera array to obtain an original image set, preprocessing the original image set to extract a clear characteristic set, and generating an enhanced defect candidate point characteristic set; The defect classification module is used for three-dimensionally reconstructing the defect candidate point feature set through a stereoscopic vision algorithm to generate a defect space position information set, extracting features by using a convolutional neural network to determine a distinguishing basis, and classifying to generate a defect type set; The curvature texture analysis module is used for carrying out curvature and surface texture gradient analysis on the defect type set to obtain defect curvature change detection data, and generating a defect cause parameter identification set by fusing texture analysis results; the dynamic tracking module is used for updating the spatial position change of the defect by adopting a dynamic tracking algorithm according to the defect cause parameter identification set to generate a defect spatial position evolution trend information set; And the parameter optimization module is used for fusing the defect type set according to the defect space position evolution trend information set, optimizing injection molding process parameters and obtaining a mold design improvement data set.

Description

Plastic injection molding defect real-time detection method, system and storage medium Technical Field The application relates to the technical field of industrial detection, in particular to a method, a system and a storage medium for detecting plastic injection molding defects in real time. Background The thin-wall injection molding part is an indispensable part in the modern manufacturing industry, is widely applied to the fields of automobiles, electronics, medical appliances and the like, and the quality of the thin-wall injection molding part directly influences the performance and the reliability of products. The thin-wall injection molding with high quality requires few surface and internal defects, especially white spot defects of micron order, which may lead to the decrease of the strength of the parts or the disqualification of the appearance, and becomes a key problem to be solved in the production. However, existing detection and optimization techniques are not enough to meet the requirements of high precision and intelligence in complex production environments. The existing injection molding defect detection technology has the obvious defects of multiple dimensions, namely, firstly, the complex environment has weak anti-interference capability, when the illumination is uneven, the normal surface texture of an injection molding part is easily misjudged as a white point defect by traditional imaging, or a real tiny white point is omitted due to shadow shielding, the fog of a camera lens and the reduction of image contrast are caused by humidity fluctuation, defect identification deviation is further aggravated, and when the position of a part is slightly deviated, the static detection equipment cannot dynamically adjust the imaging angle, so that the defect position is misjudged. Secondly, the space information acquisition capability is lost, and defect dimension attributes, such as tiny impurity white spots attached to the surface and bubbles wrapped inside, cannot be distinguished only by depending on the two-dimensional image, and the two are similar in characteristics in the two-dimensional image, so that defect type confusion is easy to cause, and subsequent cause analysis is influenced. Thirdly, the defects and the technological parameters are associated to fracture, the defects can only be identified in the prior art, but the defects cannot be traced to the root-cause technological problems of uneven temperature distribution of the mold through the spatial positions of the defects, fluctuation of injection molding pressure, abnormal material flow and the like, so that the technological adjustment depends on manual experience, data support is lacked, and the defects are difficult to reduce from the source. Fourth, the dynamic tracking and prediction capability is insufficient, the position evolution trend of the defect in the injection molding process cannot be monitored in real time, and the defect development cannot be predicted based on historical data and real-time environmental parameters, so that a regulation instruction lags behind the defect formation process, and timely intervention cannot be performed. Aiming at the defects, the method acquires an original image set through multi-angle imaging, preprocesses enhancement features, acquires the space position of the defects through three-dimensional reconstruction of a stereoscopic vision algorithm, classifies the types of the defects by combining a neural network, identifies the causes of the defects through multi-view curvature and texture analysis, dynamically tracks the evolution trend of the defects, optimizes process parameters based on the trend and generates mold design improvement data to form an imaging-analysis-tracing-optimizing closed loop, solves the problems of low precision, weak interference resistance and closed loop optimization of the existing detection technology, improves the detection efficiency and process improvement scientificity of the injection molding defects, and meets the quality control requirement of high-precision injection molding production. Disclosure of Invention The application provides a method, a system and a storage medium for detecting plastic injection molding defects in real time, solves the problems of low precision, weak interference resistance and no closed loop optimization in the existing detection technology, improves the detection efficiency of the injection molding defects and the scientificity of technological improvement, and meets the quality control requirement of high-precision injection molding production. In a first aspect, the application provides a method for detecting plastic injection molding defects in real time, the method comprising the following steps: s101, synchronously imaging a defect area of a plastic injection molding piece through a multi-angle camera array to obtain an original image set, preprocessing the original image set, extracting a clear feature set,