A mold injection control method and system for injection molding of automobile parts

Abstract

The application discloses a kind of mould injection tube control methods and systems of automobile accessory injection molding, including obtaining current injection part and the three-dimensional point cloud of immediately preceding injection part, the global difference degree between the two is calculated by overall registration, if exceed threshold value, difference mask is generated and is covered to cavity real-time image, difference image is formed, difference image is carried out mixed clustering with known defect type database, obtain candidate defect type and membership degree, with membership degree as input, combined with bayesian inference algorithm, calculate the posterior probability contribution of each sensor node to candidate defect type, if the contribution of any sensor node exceeds threshold value, output defect root cause report, and drive machine or mould execution mechanism to correct.The application solves the defect detection and root cause analysis of automobile accessories rely on artificial visual inspection or single dimension automatic detection, there is low detection efficiency, subjective bias is big, root cause judgment lag.

Description

Technical Field

[0001] This invention relates to the field of mold injection technology, and in particular to a mold injection control method and system for injection molding of automotive parts. Background Technology

[0002] In modern industrial production, product quality inspection and defect identification are crucial for ensuring product reliability and performance. With the rapid development of automated production and intelligent manufacturing, traditional visual defect detection methods are no longer sufficient to meet the requirements of high efficiency and high accuracy. Therefore, automatic defect detection technology based on image processing and pattern recognition has emerged and is gradually becoming an important tool in the field of industrial inspection.

[0003] In automated defect detection technology, image processing techniques are used to acquire surface images of products and analyze abnormal areas in the images through algorithms to identify potential defects. However, existing defect detection methods have certain limitations when dealing with complex defect types. When faced with new and unknown defect types, traditional detection methods often struggle to accurately identify and classify them. Summary of the Invention

[0004] To address at least one of the aforementioned technical problems, this invention provides a mold injection control method and system for injection molding of automotive parts.

[0005] In a first aspect, the present invention provides a method for controlling the injection molding of molds for automotive parts, the method comprising:

[0006] At the exit station of each independent cavity, continuously acquire the 3D point cloud of the current injection molded part and the 3D point cloud of the next adjacent injection molded part.

[0007] The current point cloud is registered with the point cloud at the previous time step to obtain the difference region, and the global difference between the two is calculated.

[0008] When the global difference exceeds a preset threshold, a difference mask is generated based on the difference region, and the difference mask is overlaid on the synchronously acquired real-time image of the corresponding cavity to form a difference image;

[0009] The difference images are subjected to hybrid clustering with a database of known defect types to obtain candidate defect types and their corresponding membership degrees.

[0010] Using the membership degree as the observation input, the digital twin is invoked and combined with the Bayesian inference algorithm to calculate the posterior probability contribution of each sensing node to each candidate defect type.

[0011] If the posterior probability contribution of any sensor node exceeds the preset root cause threshold, a defect root cause report containing the cavity number, defect type, suspected sensor node number, and suggested repair actions will be output, and the machine tool or mold actuator will be driven to implement the correction.

[0012] Preferably, the step of performing hybrid clustering between the difference images and a known defect type database to obtain candidate defect types and corresponding membership degrees includes:

[0013] The difference images are used as the first image set, and standard difference images of several defect types are extracted from a known defect type database as the second image set.

[0014] The first image set and the second image set are merged to form a mixed image set; clustering is then performed on the mixed image set to obtain several subsets of categorized data.

[0015] The distribution of the number of images in the first and second image sets of each category data subset is statistically analyzed, and the candidate defect types and their corresponding membership degrees are obtained based on the distribution of the number of images.

[0016] Preferably, the step of obtaining candidate defect types and corresponding membership degrees based on the distribution of the number of images includes:

[0017] If the total number of classification data subsets is greater than the number of defect types, the subsets that do not contain any second image set are marked as unknown defect type subsets, and the first image set images in the corresponding classification data subsets are determined as unknown defect candidates.

[0018] Otherwise, based on the distribution of each defect type in each subset of the second image set, the first image set image in the classification data subset containing the corresponding known defect type image is labeled with the defect type to obtain the candidate image of the known defect type and its membership degree.

[0019] Preferably, the step of using the membership degree as the observation input, calling the digital twin and combining it with a Bayesian inference algorithm to calculate the posterior probability contribution of each sensing node to each candidate defect type includes:

[0020] The membership degree of the candidate defect type is used as the observation likelihood;

[0021] The real-time process data corresponding to the current cavity in the digital twin is called, including the pressure, temperature and flow rate of each sensor node, as input variables for the prior probability;

[0022] The prior probability and the observed likelihood are fused node by node according to Bayes' formula to calculate the contribution of each sensing node to the posterior probability of each candidate defect type.

[0023] When the posterior probability contribution of any sensing node exceeds the preset root cause threshold, the node is marked as a suspected root cause node, and the node number and its real-time process data are written into the defect root cause report.

[0024] Preferably, the step of fusing the prior probability and the observed likelihood node-by-node according to Bayes' theorem to calculate the posterior probability contribution of each sensing node to each candidate defect type includes:

[0025] Based on the process data of each sensor node recorded by the digital twin in the current period, a prior probability distribution with nodes as conditional variables is constructed.

[0026] The membership degree of candidate defect types output by the mixed clustering is used as the observation likelihood function;

[0027] For each sensing node, the prior probability distribution and the observation likelihood function are multiplied and normalized per sensing node under the Bayesian framework to obtain the posterior probability of the corresponding sensing node for each candidate defect type.

[0028] The posterior probability contribution of a node is obtained by summing the posterior probabilities of the same node for all candidate defect types.

[0029] When the posterior probability contribution of any sensor node exceeds the preset root cause threshold, the corresponding sensor node is marked as a suspected root cause node, and the corresponding number and real-time process data are written into the defect root cause report.

[0030] Secondly, the present invention also provides a mold injection control system for injection molding of automotive parts, the system comprising:

[0031] The 3D point cloud acquisition module is used to continuously acquire the 3D point cloud of the current injection molded part and the 3D point cloud of the next adjacent injection molded part at the exit station of each independent cavity.

[0032] The point cloud registration and difference analysis module is used to register the current point cloud with the point cloud at the previous time step, obtain the difference region, and calculate the global difference degree between the two.

[0033] The difference image generation module is used to generate a difference mask based on the difference region when the global difference exceeds a preset threshold, and to cover the difference mask onto the synchronously acquired real-time image of the corresponding cavity to form a difference image.

[0034] The defect identification and clustering analysis module is used to perform hybrid clustering between the difference images and the known defect type database to obtain candidate defect types and their corresponding membership degrees;

[0035] The root cause reasoning and Bayesian analysis module is used to use the membership degree as the observation input, call the digital twin and combine it with the Bayesian reasoning algorithm to calculate the posterior probability contribution of each sensing node to each candidate defect type.

[0036] The defect reporting and control response module is used to output a defect root cause report containing the cavity number, defect type, suspected sensor node number and suggested repair action if the posterior probability contribution of any sensor node exceeds the preset root cause threshold, and drive the machine tool or mold actuator to implement the correction.

[0037] Preferably, the step of performing hybrid clustering between the difference images and a known defect type database to obtain candidate defect types and corresponding membership degrees includes:

[0038] The difference images are used as the first image set, and standard difference images of several defect types are extracted from a known defect type database as the second image set.

[0039] The first image set and the second image set are merged to form a mixed image set; clustering is then performed on the mixed image set to obtain several subsets of categorized data.

[0040] The distribution of the number of images in the first and second image sets of each category data subset is statistically analyzed, and the candidate defect types and their corresponding membership degrees are obtained based on the distribution of the number of images.

[0041] Preferably, the step of using the membership degree as the observation input, calling the digital twin and combining it with a Bayesian inference algorithm to calculate the posterior probability contribution of each sensing node to each candidate defect type includes:

[0042] The membership degree of the candidate defect type is used as the observation likelihood;

[0043] The real-time process data corresponding to the current cavity in the digital twin is called, including the pressure, temperature and flow rate of each sensor node, as input variables for the prior probability;

[0044] The prior probability and the observed likelihood are fused node by node according to Bayes' formula to calculate the contribution of each sensing node to the posterior probability of each candidate defect type.

[0045] When the posterior probability contribution of any sensing node exceeds the preset root cause threshold, the node is marked as a suspected root cause node, and the node number and its real-time process data are written into the defect root cause report.

[0046] Thirdly, the present invention also provides an electronic device including a processor and a memory, the memory being used to store computer program code, the computer program code including computer instructions, wherein when the processor executes the computer instructions, the electronic device performs the method as described in the first aspect above and any possible implementation thereof.

[0047] Fourthly, the present invention also provides a computer-readable storage medium storing a computer program, the computer program including program instructions that, when executed by a processor of an electronic device, cause the processor to perform a method as described in the first aspect above and any possible implementation thereof.

[0048] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0049] This invention compares the three-dimensional morphology of injection molded parts based on a unified coordinate system. By setting a preset difference threshold, it achieves preliminary screening of morphological anomalies, providing a precise spatial coordinate reference for subsequent location of difference areas. This ensures the accurate locking of the physical location of difference areas, solving the problem of blurred or delayed location of difference areas in traditional inspections due to the lack of a three-dimensional reference. When the difference exceeds the preset threshold, a difference mask generated based on this location is fused with synchronously acquired real-time images, transforming the differences in three-dimensional space into intuitive two-dimensional image features. This transformation from three-dimensional morphological differences to two-dimensional visual features provides easily identifiable morphological features for defect type analysis. Based on this, defect type analysis can be matched with known defect types by relying on clear image features, overcoming the limitations of single three-dimensional data morphological features being unintuitive and difficult to quickly associate with defect types. From point cloud registration to achieve differential localization, to image fusion to generate visual features, and then to defect type analysis based on fused features, a complete technical chain of "precise localization - feature visualization - type recognition" is formed. This not only retains the positioning accuracy advantage of three-dimensional registration, but also leverages the feature recognition efficiency of two-dimensional images, effectively replacing the traditional method of relying on manual visual inspection or single-dimensional detection, thereby achieving timely detection and accurate identification of defects.

[0050] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and are not intended to limit this disclosure. Attached Figure Description

[0051] To more clearly illustrate the technical solutions in the embodiments of the present invention or the background art, the accompanying drawings used in the embodiments of the present invention or the background art will be described below.

[0052] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this disclosure and, together with the specification, serve to illustrate the technical solutions of this disclosure.

[0053] Figure 1 A schematic flowchart illustrating a method for controlling the injection molding of automotive parts, provided in an embodiment of the present invention.

[0054] Figure 2 This is a schematic diagram of the structure of a mold injection control system for injection molding of automotive parts, provided as an embodiment of the present invention. Detailed Implementation

[0055] To enable those skilled in the art to better understand the present invention, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0056] In this document, the term "embodiment" means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of the invention. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0057] In existing methods for detecting and analyzing defects in injection-molded automotive parts, manual inspection suffers from low efficiency, significant subjective bias, and delayed root cause assessment.

[0058] Please see Figure 1 , Figure 1 This is a schematic flowchart illustrating a method for controlling the injection molding of automotive parts, provided as an embodiment of the present invention. Figure 1 As shown, the method includes:

[0059] S100 continuously acquires the 3D point cloud of the current injection molded part and the 3D point cloud of the next adjacent injection molded part at the exit station of each independent cavity.

[0060] In this embodiment, a high-speed laser 3D scanner with dynamic focusing function is deployed at the exit station of each independent cavity. The scanning is triggered by an infrared through-beam sensor on the side of the cavity exit. When the injection molded part is completely detached from the cavity and is in the scanning reference position (with the cavity exit positioning pin as the coordinate origin), the scanning is started immediately. The scanner has a built-in 256-channel data acquisition module, which can simultaneously acquire the 3D point cloud of the outer surface of the injection molded part and the internal cavity structure (such as threaded holes and snap-fit ​​grooves) with a depth ≤10mm. Through the integrated edge computing unit, the point cloud coordinate normalization is completed within 50ms after the scanning is completed, and the preprocessed point cloud data of the adjacent injection molded part is retrieved through the high-speed cache interface to ensure that the point clouds before and after are in the same measurement coordinate system (coordinate deviation ≤0.01mm).

[0061] S200: Perform global registration between the current point cloud and the point cloud at the previous time step to obtain the difference region and calculate the global difference between the two.

[0062] A feature-constrained ICP registration algorithm is used to register the current point cloud with the previous point cloud. The algorithm prioritizes extracting rigid positioning features of the injection molded part (such as positioning holes with a diameter ≥ 5 mm and mounting surfaces with a flatness ≤ 0.02 mm) as registration reference points. Fast alignment is achieved through two-stage processing: coarse registration (using the SAC-IA algorithm, error ≤ 0.1 mm) + fine registration (using point-to-surface ICP optimization, error ≤ 0.02 mm). After registration, the spatial distance difference between corresponding point pairs is calculated synchronously by a parallel computing unit (16-core processor). Areas with a distance difference ≥ 0.08 mm are marked as difference areas. At the same time, a weighted average error formula (weights are allocated according to feature importance, such as 0.6 for appearance surfaces and 0.4 for structural surfaces) is used to calculate the global difference degree. Preset thresholds are divided according to defect risk level (0.1 mm for critical appearance areas and 0.2 mm for structural functional areas) to ensure the accuracy and efficiency of difference identification.

[0063] S300, when the global difference exceeds a preset threshold, a difference mask is generated according to the difference region, and the difference mask is covered onto the synchronously acquired real-time image of the corresponding cavity to form a difference image;

[0064] In this embodiment, the global difference between the current point cloud and the point cloud at the previous time step exceeds a preset threshold as the trigger condition. Voxelization downsampling is performed on the difference region to obtain a difference voxel set. The difference voxel set is orthographically projected onto the real-time image plane of the corresponding cavity to generate a binary difference mask. The region with a pixel value of the first value in the mask represents the location of the difference voxel, and the region with a pixel value of the second value represents the location of the indifferent region. Morphological closing operation is performed on the binary difference mask to fill the holes and smooth the edges to obtain an optimized difference mask. The optimized difference mask is superimposed on the synchronously acquired real-time image of the corresponding cavity using a pixel-level logical AND operation. The image pixels corresponding to the first value region of the mask are retained, and the remaining pixels are removed to form a difference image.

[0065] S400: Perform hybrid clustering on the difference images and the known defect type database to obtain candidate defect types and their corresponding membership degrees;

[0066] In this embodiment, the known defect type database adopts a hierarchical index architecture (retrieval response time ≤ 0.15 seconds), and stores it according to defect morphology (such as flash, shrinkage marks, missing material) and impact degree (fatal / serious / moderate). Each type of defect contains ≥ 2000 samples (covering typical features under different materials and process parameters). The hybrid clustering algorithm adopts density clustering, and reduces the amount of computation by reducing the dimensionality of feature vectors (retaining 95% of feature information). It performs matching calculations on the geometric features (area, perimeter, circumscribed rectangle) and grayscale features (average grayscale value, gradient change) of the difference images, outputs candidate defect types and corresponding membership degrees (value range 0-1), and automatically selects defect types with membership degrees ≥ 0.8 as highly suspicious candidates, thereby improving the accuracy of defect identification.

[0067] S500, using the membership degree as the observation input, calling the digital twin and combining it with the Bayesian inference algorithm, calculates the posterior probability contribution of each sensing node to each candidate defect type.

[0068] In this embodiment, the membership degree of candidate defect types is used as the observation input. A lightweight digital twin is invoked (model loading time ≤ 0.5 seconds). Data (covering 12 core parameters such as injection pressure, mold temperature, and melt volume) is synchronized in real time with the machine sensor network through a 5G industrial module. The Bayesian inference algorithm uses a pre-trained probability parameter library (inference time ≤ 0.3 seconds). The historical defect-sensor data association model is used as the prior probability. The posterior probability contribution of each sensor node (such as temperature sensor T1 of cavity 1 and nozzle pressure sensor P2) to the candidate defect type is calculated (calculation accuracy ± 0.01). The results are then sorted and output in descending order of contribution, prioritizing the locking of sensor nodes with high correlation.

[0069] S600 If the posterior probability contribution of any sensor node exceeds the preset root cause threshold, a defect root cause report containing the cavity number, defect type, suspected sensor node number and suggested repair action will be output, and the machine tool or mold actuator will be driven to implement the correction.

[0070] In this embodiment, the preset root cause threshold is dynamically adjusted according to the defect type (0.6 for fatal defects and 0.7 for severe defects). When the posterior probability contribution of any sensor node exceeds the corresponding threshold, a defect root cause report is output through the templated report generation engine (report generation time ≤ 0.8 seconds). The cavity number corresponds to the specific molding cavity (e.g., "cavity 3"), the defect type is specified to the specific location (e.g., "shrinkage mark near the gate"), the suspected sensor node is marked with the specific installation location (e.g., "temperature sensor T3 on the moving mold side of cavity 3"), and the suggested repair action includes quantitative parameters (e.g., "increase the mold temperature of the area corresponding to T3 by 5±1℃"). At the same time, the machine tool execution mechanism (servo adjustment module with response time ≤ 1 second) is triggered to implement the correction, and the correction effect is pre-visualized through a digital twin. If the pre-visualization shows that the defect may still exist after the correction, a secondary warning is issued simultaneously.

[0071] In this embodiment, the three-dimensional shape of the injection molded part is compared based on a unified coordinate system. A preset difference threshold is used to initially screen for morphological anomalies, providing a precise spatial coordinate reference for subsequent location of difference areas. This ensures the accurate locking of the physical location of these areas, solving the problem of unclear or delayed location of difference areas caused by the lack of a three-dimensional reference in traditional inspection. When the difference exceeds the preset threshold, the difference mask generated based on this location is fused with the synchronously acquired real-time image, transforming the differences in three-dimensional space into intuitive two-dimensional image features. This transformation from three-dimensional morphological differences to two-dimensional visual features provides easily identifiable features for defect type analysis. Based on the state, defect type analysis can match clear image features with known defect types, overcoming the limitations of single three-dimensional data morphological features being unintuitive and difficult to quickly associate with defect types. From point cloud registration to achieve differential localization, to image fusion to generate visual features, and then to defect type analysis based on fused features, a complete technical chain of "precise localization - feature visualization - type recognition" is formed. This not only retains the positioning accuracy advantage of three-dimensional registration, but also leverages the feature recognition efficiency of two-dimensional images, effectively replacing the traditional method of relying on manual visual inspection or single-dimensional detection, thereby achieving timely detection and accurate identification of defects.

[0072] Preferably, the step of performing hybrid clustering between the difference images and a known defect type database to obtain candidate defect types and corresponding membership degrees includes:

[0073] The difference images are used as the first image set, and standard difference images of several defect types are extracted from a known defect type database as the second image set.

[0074] The first image set and the second image set are merged to form a mixed image set; clustering is then performed on the mixed image set to obtain several subsets of categorized data.

[0075] The distribution of the number of images in the first and second image sets of each category data subset is statistically analyzed, and the candidate defect types and their corresponding membership degrees are obtained based on the distribution of the number of images.

[0076] Preferably, the step of obtaining candidate defect types and corresponding membership degrees based on the distribution of the number of images includes:

[0077] If the total number of classification data subsets is greater than the number of defect types, the subsets that do not contain any second image set are marked as unknown defect type subsets, and the first image set images in the corresponding classification data subsets are determined as unknown defect candidates.

[0078] Otherwise, based on the distribution of each defect type in each subset of the second image set, the first image set image in the classification data subset containing the corresponding known defect type image is labeled with the defect type to obtain the candidate image of the known defect type and its membership degree.

[0079] In this embodiment, the difference images are used as the first image set, which includes the currently detected difference images (resolution 1920×1080, image format BMP, containing pixel coordinates and grayscale information of the difference areas). From the known defect type database, according to the injection molding type corresponding to the difference images (such as automotive dashboard clips), standard difference images of several defect types (such as flash, shrinkage marks, and missing material) corresponding to this type of injection molding are extracted as the second image set, wherein there are no less than 50 standard difference images for each defect type, and all of them have been preprocessed (including size normalization and background removal) to be consistent with the image specifications of the first image set. The first image set and the second image set are merged to form a hybrid image set. The first image set contains 1 image (the current difference image), and the total number of images in the second image set is determined according to the number of defect types (e.g., 150 images for 3 defect types). The K-means clustering algorithm is used to perform clustering processing on the hybrid image set. Before clustering, the feature vectors of each image are extracted (including the area, perimeter, aspect ratio of the bounding rectangle, and grayscale mean of the difference region). The number of clusters is set to the sum of the number of defect types and 1 (e.g., 4 clusters for 3 defect types). Through iterative calculation, the Euclidean distance of the feature vectors of images in the same subset is ≤5.0, and finally, several classification data subsets are obtained.

[0080] In one possible embodiment, the first image set (containing one current difference image) is merged with the second image set (e.g., 240 standard images corresponding to three defect types) to form a mixed image set containing 241 images. DBSCAN clustering is performed on this mixed image set. Before clustering, the core feature vectors of each image are extracted by the feature extraction module (including the area of ​​the difference region in pixels, the perimeter in pixels, the aspect ratio of the bounding rectangle, and the gray standard deviation). The clustering parameters eps are set to 8.0 (density radius based on the Euclidean distance of the feature vectors) and min_samples is set to 5 (minimum number of neighborhood samples of the core point). Through density reachability analysis, images with a feature vector Euclidean distance ≤ 8.0 are divided into the same subset, resulting in several classification data subsets (each subset contains ≥ 5 images). The set counting module counts the number of images in the first image set (denoted as A, with a value of 0 or 1) and the number of images of each defect type in the second image set (denoted as Bj, where j is the defect type number, with a value ≥ 0). Based on the statistical results, when A=1, if there is a Bj>0 in the corresponding subset, the ratio of Bj of a certain defect type in the subset to the total number of images in the second image set in the subset is calculated as the membership degree of the defect type (with a value range of 0-1). If Bj in the corresponding subset is all 0, it is marked as a potential unknown defect.

[0081] The distribution of the number of images in the first and second image sets of each category data subset is statistically analyzed. Specifically, the number of images in the first image set (denoted as N1) and the number of images of each defect type in the second image set (denoted as N2i, where i is the defect type number) are counted within each subset. Based on the distribution of the number of images, the candidate defect types and their corresponding membership degrees are obtained by calculating the proportion of N1 in the total number of images in the subset and the ratio of N2i to the total sample size of the corresponding defect type. The membership degree is the product of the proportion of N2i in the subset and the proportion of N1 in the subset.

[0082] If the total number of classification data subsets is greater than the number of defect types, for example, if the number of defect types is 3 and the number of subsets is 4, then the subsets that do not contain any second image set (i.e., N2i=0) are marked as unknown defect type subsets, and the first image set image (i.e., the current difference image) in the subset is determined as an unknown defect candidate. At the same time, the feature vector of the subset is recorded to update the database. Otherwise, if the total number of classification data subsets is equal to or less than the number of defect types, according to the distribution of each defect type in the second image set in each subset (e.g., the standard image of the "flying edge" type accounts for 80% of the total number of the second image set in a subset), the first image set image in the classification data subset containing the corresponding known defect type image is labeled with the defect type (labeled as "flying edge"), and the proportion of the standard image of the defect type in the subset (80%) is used as the corresponding membership degree.

[0083] In this embodiment, the difference image and the standard difference image of known defect types are respectively used as the first image set and the second image set, and the two are merged to form a mixed image set. This provides a comprehensive and comparative image foundation for subsequent clustering processing, enabling the difference image to more accurately find its potential defect type classification by comparing it with images of known defect types. Furthermore, clustering processing is performed on the mixed image set to obtain several classification data subsets. Utilizing the characteristics of clustering algorithms, images with similar features are grouped into one category, thus initially realizing the classification of the difference images. Then, by statistically analyzing the first image set and the second image set in each classification data subset... The distribution of the number of images in the image set is further analyzed to determine candidate defect types and their membership degrees. The analysis process not only considers the matching degree between the difference images and known defect type images, but also enhances the reliability and accuracy of the classification results by considering the distribution of the number of images. The relationship between the total number of classification data subsets and the number of defect types is judged, and unknown defect types and known defect types are processed separately. This can not only identify new unknown defect types, but also accurately label and calculate the membership degrees of known defect types, thereby achieving comprehensive and accurate classification and recognition of difference images, and improving the accuracy and adaptability of defect detection.

[0084] Preferably, the step of using the membership degree as the observation input, calling the digital twin and combining it with a Bayesian inference algorithm to calculate the posterior probability contribution of each sensing node to each candidate defect type includes:

[0085] The membership degree of the candidate defect type is used as the observation likelihood;

[0086] The real-time process data corresponding to the current cavity in the digital twin is called, including the pressure, temperature and flow rate of each sensor node, as input variables for the prior probability;

[0087] The prior probability and the observed likelihood are fused node by node according to Bayes' formula to calculate the contribution of each sensing node to the posterior probability of each candidate defect type.

[0088] When the posterior probability contribution of any sensing node exceeds the preset root cause threshold, the node is marked as a suspected root cause node, and the node number and its real-time process data are written into the defect root cause report.

[0089] Preferably, the step of fusing the prior probability and the observed likelihood node-by-node according to Bayes' theorem to calculate the posterior probability contribution of each sensing node to each candidate defect type includes:

[0090] Based on the process data of each sensor node recorded by the digital twin in the current period, a prior probability distribution with nodes as conditional variables is constructed.

[0091] The membership degree of candidate defect types output by the mixed clustering is used as the observation likelihood function;

[0092] For each sensing node, the prior probability distribution and the observation likelihood function are multiplied and normalized per sensing node under the Bayesian framework to obtain the posterior probability of the corresponding sensing node for each candidate defect type.

[0093] The posterior probability contribution of a node is obtained by summing the posterior probabilities of the same node for all candidate defect types.

[0094] When the posterior probability contribution of any sensor node exceeds the preset root cause threshold, the corresponding sensor node is marked as a suspected root cause node, and the corresponding number and real-time process data are written into the defect root cause report.

[0095] In this embodiment, the membership degree of the candidate defect type is used as the observation likelihood. Specifically, the membership degree of "edge flash" (0.82), "shrinkage" (0.15), and "material shortage" (0.03) output by the mixed clustering are used as the observation likelihood values. These likelihood values ​​are mapped to the [0,1] interval through standardization and are directly used as the probability measure of the occurrence of observed events in Bayesian inference. The system retrieves real-time process data corresponding to the current cavity (e.g., cavity 2) from the digital twin. This digital twin is synchronized in real time with the machine's sensor network via an industrial Ethernet network (transmission delay ≤10ms). The data includes real-time pressure values ​​from pressure sensing nodes (P2-1 to P2-3, measurement range 0-200bar), real-time temperature values ​​from temperature sensing nodes (T2-1 to T2-2, measurement range 20-150℃), and real-time flow rate values ​​from melt flow rate sensing nodes (F2-1, measurement range 0-100mm / s) corresponding to cavity 2. The above data is updated every 0.5 seconds and stored in the twin database as input variables for prior probabilities.

[0096] The prior probability and observed likelihood are fused node-by-node using Bayesian formula. Specifically, for pressure sensor node P2-1, the historical normal probability (prior probability) corresponding to its current pressure value is extracted first. Then, the posterior probability of the node for the "edge fly-edge" is calculated using Bayesian formula (posterior probability = prior probability × observed likelihood / evidence factor) with the observed likelihood of 0.82 for the "edge fly-edge". Similarly, the above calculation is performed on all sensor nodes such as T2-1 and F2-1 to obtain the posterior probability contribution of each node to each candidate defect type. When the posterior probability contribution of any sensor node exceeds the preset root cause threshold (e.g., 0.7, determined based on the historical defect repair success rate), the node (e.g., P2-1) is marked as a suspected root cause node through the root cause marking module. Its node number "P2-1" and real-time pressure value "185 bar" (current measurement value) are written into the "Suspicious Sensor Node" field of the defect root cause report, and the historical data curves of the node for the past 10 periods are stored in association.

[0097] Based on the process data of each sensor node recorded by the digital twin in the current period (the most recent 5 minutes), a prior probability distribution with nodes as conditional variables is constructed. Specifically, for temperature sensor node T2-1, a normal prior probability distribution is constructed based on its historical normal production temperature data (mean 85℃, fluctuation range ±5℃). When the real-time temperature is 92℃, the prior probability corresponding to the deviation from the normal range is obtained through the distribution function. The membership degree of candidate defect types output by mixed clustering is used as the observation likelihood function. Specifically, the membership degree of "edge flash" (0.82) is transformed into a likelihood function value, that is, the conditional probability of the occurrence of "edge flash" defects corresponding to this membership degree. The function form is a single-point probability value mapping (the membership degree value is directly used as the likelihood function output).

[0098] For each sensing node, the prior probability distribution and the observation likelihood function are normalized per-sensor node using a Bayesian multiplication method. Specifically, for the velocity sensing node F2-1, its velocity deviation prior probability (0.3) is multiplied by the "edge flash" observation likelihood (0.82) to obtain an unnormalized probability of 0.246. This is then divided by the sum of the unnormalized probabilities corresponding to all candidate defect types (e.g., 0.246 + 0.045 + 0.009) to obtain a normalized posterior probability of 0.81, which is used as the posterior probability of this node for "edge flash". The posterior probabilities of the same node for all candidate defect types are summed to obtain the posterior probability contribution of this node. For example, the posterior probability of pressure sensing node P2-1 for "edge flash" is 0.75, for "shrinkage mark" it is 0.12, and for "material shortage" it is 0.03, and the sum gives a contribution of 0.90. When the posterior probability contribution of any sensor node exceeds the preset root cause threshold (e.g., 0.7), the corresponding sensor node (e.g., P2-1) is marked as a suspected root cause node by the report generation module, and its number "P2-1", real-time pressure value "185 bar", and pressure fluctuation curve (last 30 seconds) are written into the "Root Cause Analysis" section of the defect root cause report, along with the installation location diagram of the node.

[0099] In this embodiment, the membership degree of candidate defect types is used as the observation likelihood, providing a defect probability measure based on image features for Bayesian inference, ensuring that the inference input is directly related to the defect morphological features. Real-time process data (pressure, temperature, flow rate, etc.) from the digital twin is used as prior probability input, enabling inference to reflect the equipment operating status based on actual production parameters, avoiding abstract probability calculations detached from production reality. Node-by-node fusion of Bayesian inference dynamically combines prior probability with observation likelihood, realizing the correlation analysis between equipment operating data and defect features, overcoming the root cause judgment bias caused by relying solely on image features or sensor data. The calculation of posterior probability contribution quantifies the influence weight of each sensor node on the defect through normalization and summation, providing a clear basis for root cause localization. The marking of suspected root cause nodes and the writing of data into the report transform the inference results into directly applicable root cause information, providing precise guidance for subsequent repair actions. The combination of the above steps not only retains the advantage of image features in intuitively representing defects, but also leverages the real-time reflection of equipment status by sensor data. By using Bayesian inference to achieve the organic integration of the two, accurate and real-time localization of the root cause of defects can be realized, effectively solving the problems of inaccurate and delayed root cause identification caused by traditional reliance on experience-based judgment or analysis of a single data dimension.

[0100] In summary, the method provided in this embodiment can achieve at least the following effects:

[0101] This invention compares the three-dimensional morphology of injection molded parts based on a unified coordinate system. By setting a preset difference threshold, it achieves preliminary screening of morphological anomalies, providing a precise spatial coordinate reference for subsequent location of difference areas. This ensures the accurate locking of the physical location of difference areas, solving the problem of blurred or delayed location of difference areas in traditional inspections due to the lack of a three-dimensional reference. When the difference exceeds the preset threshold, a difference mask generated based on this location is fused with synchronously acquired real-time images, transforming the differences in three-dimensional space into intuitive two-dimensional image features. This transformation from three-dimensional morphological differences to two-dimensional visual features provides easily identifiable morphological features for defect type analysis. Based on this, defect type analysis can be matched with known defect types by relying on clear image features, overcoming the limitations of single three-dimensional data morphological features being unintuitive and difficult to quickly associate with defect types. From point cloud registration to achieve differential localization, to image fusion to generate visual features, and then to defect type analysis based on fused features, a complete technical chain of "precise localization - feature visualization - type recognition" is formed. This not only retains the positioning accuracy advantage of three-dimensional registration, but also leverages the feature recognition efficiency of two-dimensional images, effectively replacing the traditional method of relying on manual visual inspection or single-dimensional detection, thereby achieving timely detection and accurate identification of defects.

[0102] See Figure 2 In one embodiment, a mold injection control system for injection molding of automotive parts is also provided, the system comprising:

[0103] The 3D point cloud acquisition module 100 is used to continuously acquire the 3D point cloud of the current injection molded part and the 3D point cloud of the next adjacent injection molded part at the exit station of each independent cavity.

[0104] The point cloud registration and difference analysis module 200 is used to register the current point cloud with the point cloud at the previous time step, obtain the difference region, and calculate the global difference degree between the two.

[0105] The difference image generation module 300 is used to generate a difference mask according to the difference region when the global difference exceeds a preset threshold, and to cover the difference mask onto the synchronously acquired real-time image of the corresponding cavity to form a difference image.

[0106] The defect identification and clustering analysis module 400 is used to perform hybrid clustering between the difference images and the known defect type database to obtain candidate defect types and their corresponding membership degrees;

[0107] The root cause reasoning and Bayesian analysis module 500 is used to use the membership degree as the observation input, call the digital twin and combine it with the Bayesian reasoning algorithm to calculate the posterior probability contribution of each sensing node to each candidate defect type.

[0108] The defect reporting and control response module 600 is used to output a defect root cause report containing the cavity number, defect type, suspected sensor node number and suggested repair action if the posterior probability contribution of any sensor node exceeds the preset root cause threshold, and drive the machine tool or mold actuator to implement the correction.

[0109] Preferably, the step of performing hybrid clustering between the difference images and a known defect type database to obtain candidate defect types and corresponding membership degrees includes:

[0110] The difference images are used as the first image set, and standard difference images of several defect types are extracted from a known defect type database as the second image set.

[0111] The first image set and the second image set are merged to form a mixed image set; clustering is then performed on the mixed image set to obtain several subsets of categorized data.

[0112] The distribution of the number of images in the first and second image sets of each category data subset is statistically analyzed, and the candidate defect types and their corresponding membership degrees are obtained based on the distribution of the number of images.

[0113] Preferably, the step of using the membership degree as the observation input, calling the digital twin and combining it with a Bayesian inference algorithm to calculate the posterior probability contribution of each sensing node to each candidate defect type includes:

[0114] The membership degree of the candidate defect type is used as the observation likelihood;

[0115] The real-time process data corresponding to the current cavity in the digital twin is called, including the pressure, temperature and flow rate of each sensor node, as input variables for the prior probability;

[0116] The prior probability and the observed likelihood are fused node by node according to Bayes' formula to calculate the contribution of each sensing node to the posterior probability of each candidate defect type.

[0117] When the posterior probability contribution of any sensing node exceeds the preset root cause threshold, the node is marked as a suspected root cause node, and the node number and its real-time process data are written into the defect root cause report.

[0118] It is understood that the system provided in this embodiment has functions or includes modules that can be used to execute the methods described in the above method embodiments. The specific implementation can be referred to the description of the above method embodiments, and for the sake of brevity, it will not be repeated here.

[0119] The present invention also provides an electronic device including a processor and a memory, the memory being used to store computer program code, the computer program code including computer instructions, wherein when the processor executes the computer instructions, the electronic device performs a method as described in any of the above possible implementations.

[0120] The present invention also provides a computer-readable storage medium storing a computer program, the computer program including program instructions that, when executed by a processor of an electronic device, cause the processor to perform a method as described in any of the above possible implementations.

[0121] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this invention.

[0122] Those skilled in the art will readily understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here. Those skilled in the art will also readily understand that the various embodiments of the present invention have different focuses, and for the sake of convenience and brevity, the same or similar parts may not be repeated in different embodiments. Therefore, parts not described or not described in detail in one embodiment can be referred to in other embodiments.

Claims

1. A method for controlling the injection molding of automotive parts, characterized in that, The method includes: At the exit station of each independent cavity, continuously acquire the 3D point cloud of the current injection molded part and the 3D point cloud of the next adjacent injection molded part. The current point cloud is registered with the point cloud at the previous time step to obtain the difference region, and the global difference between the two is calculated. When the global difference exceeds a preset threshold, a difference mask is generated based on the difference region, and the difference mask is overlaid on the synchronously acquired real-time image of the corresponding cavity to form a difference image; The difference images are subjected to hybrid clustering with a database of known defect types to obtain candidate defect types and their corresponding membership degrees. Using the membership degree as the observation input, the digital twin is invoked and combined with the Bayesian inference algorithm to calculate the posterior probability contribution of each sensing node to each candidate defect type. If the posterior probability contribution of any sensor node exceeds the preset root cause threshold, a defect root cause report containing the cavity number, defect type, suspected sensor node number, and suggested repair actions will be output, and the machine tool or mold actuator will be driven to implement the correction.

2. The method for mold injection control in the injection molding of automotive parts according to claim 1, characterized in that, The step of performing hybrid clustering between the difference images and a known defect type database to obtain candidate defect types and their corresponding membership degrees includes: The difference images are used as the first image set, and standard difference images of several defect types are extracted from a known defect type database as the second image set. The first image set and the second image set are merged to form a mixed image set; clustering is then performed on the mixed image set to obtain several subsets of categorized data. The distribution of the number of images in the first and second image sets of each category data subset is statistically analyzed, and the candidate defect types and their corresponding membership degrees are obtained based on the distribution of the number of images.

3. The method for mold injection control in the injection molding of automotive parts according to claim 2, characterized in that, Based on the distribution of the number of images, the candidate defect types and their corresponding membership degrees are obtained, including: If the total number of classification data subsets is greater than the number of defect types, the subsets that do not contain any second image set are marked as unknown defect type subsets, and the first image set images in the corresponding classification data subsets are determined as unknown defect candidates. Otherwise, based on the distribution of each defect type in each subset of the second image set, the first image set image in the classification data subset containing the corresponding known defect type image is labeled with the defect type to obtain the candidate image of the known defect type and its membership degree.

4. The method for mold injection control in the injection molding of automotive parts according to claim 1, characterized in that, The step of using the membership degree as the observation input, calling the digital twin and combining it with the Bayesian inference algorithm, to calculate the posterior probability contribution of each sensing node to each candidate defect type includes: The membership degree of the candidate defect type is used as the observation likelihood; The real-time process data corresponding to the current cavity in the digital twin is called, including the pressure, temperature and flow rate of each sensor node, as input variables for the prior probability; The prior probability and the observed likelihood are fused node by node according to Bayes' formula to calculate the contribution of each sensing node to the posterior probability of each candidate defect type. When the posterior probability contribution of any sensing node exceeds the preset root cause threshold, the node is marked as a suspected root cause node, and the node number and its real-time process data are written into the defect root cause report.

5. The injection molding control method for automotive parts injection molding according to claim 4, characterized in that, The step of fusing prior probabilities and observed likelihoods node-by-node using Bayes' theorem to calculate the posterior probability contribution of each sensing node to each candidate defect type includes: Based on the process data of each sensor node recorded by the digital twin in the current period, a prior probability distribution with nodes as conditional variables is constructed. The membership degree of candidate defect types output by the mixed clustering is used as the observation likelihood function; For each sensing node, the prior probability distribution and the observation likelihood function are multiplied and normalized per sensing node under the Bayesian framework to obtain the posterior probability of the corresponding sensing node for each candidate defect type. The posterior probability contribution of a node is obtained by summing the posterior probabilities of the same node for all candidate defect types. When the posterior probability contribution of any sensor node exceeds the preset root cause threshold, the corresponding sensor node is marked as a suspected root cause node, and the corresponding number and real-time process data are written into the defect root cause report.

6. A mold injection control system for injection molding of automotive parts, characterized in that, The system includes: The 3D point cloud acquisition module is used to continuously acquire the 3D point cloud of the current injection molded part and the 3D point cloud of the next adjacent injection molded part at the exit station of each independent cavity. The point cloud registration and difference analysis module is used to register the current point cloud with the point cloud at the previous time step, obtain the difference region, and calculate the global difference degree between the two. The difference image generation module is used to generate a difference mask based on the difference region when the global difference exceeds a preset threshold, and to cover the difference mask onto the synchronously acquired real-time image of the corresponding cavity to form a difference image. The defect identification and clustering analysis module is used to perform hybrid clustering between the difference images and the known defect type database to obtain candidate defect types and their corresponding membership degrees; The root cause reasoning and Bayesian analysis module is used to use the membership degree as the observation input, call the digital twin and combine it with the Bayesian reasoning algorithm to calculate the posterior probability contribution of each sensing node to each candidate defect type. The defect reporting and control response module is used to output a defect root cause report containing the cavity number, defect type, suspected sensor node number and suggested repair action if the posterior probability contribution of any sensor node exceeds the preset root cause threshold, and drive the machine tool or mold actuator to implement the correction.

7. The mold injection control system for injection molding of automotive parts according to claim 6, characterized in that, The step of performing hybrid clustering between the difference images and a known defect type database to obtain candidate defect types and their corresponding membership degrees includes: The difference images are used as the first image set, and standard difference images of several defect types are extracted from a known defect type database as the second image set. The first image set and the second image set are merged to form a mixed image set; clustering is then performed on the mixed image set to obtain several subsets of categorized data. The distribution of the number of images in the first and second image sets of each category data subset is statistically analyzed, and the candidate defect types and their corresponding membership degrees are obtained based on the distribution of the number of images.

8. The mold injection control system for injection molding of automotive parts according to claim 6, characterized in that, The step of using the membership degree as the observation input, calling the digital twin and combining it with the Bayesian inference algorithm, to calculate the posterior probability contribution of each sensing node to each candidate defect type includes: The membership degree of the candidate defect type is used as the observation likelihood; The real-time process data corresponding to the current cavity in the digital twin is called, including the pressure, temperature and flow rate of each sensor node, as input variables for the prior probability; The prior probability and the observed likelihood are fused node by node according to Bayes' formula to calculate the contribution of each sensing node to the posterior probability of each candidate defect type. When the posterior probability contribution of any sensing node exceeds the preset root cause threshold, the node is marked as a suspected root cause node, and the node number and its real-time process data are written into the defect root cause report.

9. An electronic device, characterized in that, include: A processor and a memory, the memory being used to store computer program code, the computer program code including computer instructions, wherein when the processor executes the computer instructions, the electronic device performs the mold injection control method for injection molding of automotive parts as described in any one of claims 1 to 5.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program, which includes program instructions that, when executed by a processor of an electronic device, cause the processor to perform the mold injection control method for injection molding of automotive parts as described in any one of claims 1 to 5.

Images