Online detection method and system for insulation performance of injection molding busbar

Abstract

The invention relates to the field of injection molding busbar quality detection, in particular to an on-line detection method and system for the insulation performance of an injection molding busbar. According to the method, after the injection molding busbar is in place, an on-line full-detection mode is started: firstly, electrical insulation data, ultrasonic scanning data and visual image data of the injection molding busbar are collected; inputting the electrical insulation data, the ultrasonic scanning data and the visual image data into corresponding defect analysis models to obtain an insulation failure confidence coefficient, an internal defect confidence coefficient and a surface defect confidence coefficient, and then based on a D-S evidence theory, performing fusion calculation on the insulation failure confidence coefficient, the internal defect confidence coefficient and the surface defect confidence coefficient to obtain a fusion result; and obtaining a comprehensive defect judgment probability, judging the quality grade of the injection molding busbar according to the comprehensive defect judgment probability and a preset confidence interval, and finally driving a corresponding sorting mechanism to circulate or reject the injection molding busbar according to the quality grade. According to the invention, the detection precision and efficiency are improved, and the situation of missing detection is avoided.

Description

Technical Field

[0001] This invention relates to the field of injection-molded busbar quality testing technology. More specifically, this invention relates to an online testing method and system for the insulation performance of injection-molded busbars. Background Technology

[0002] As a core component of power transmission in new energy vehicles, energy storage systems, and industrial control, the insulation performance of injection-molded busbars directly affects the safety and reliability of the entire power system. Injection-molded busbars are typically composed of highly conductive copper busbars and high-strength insulating material encapsulated in injection molding. During actual production, influenced by fluctuations in injection molding process parameters (such as temperature and pressure), mold precision deviations, or uneven raw material distribution, injection-molded busbars are highly susceptible to defects such as insulation layer damage, internal bubbles, microcracks, or impurity inclusions. Therefore, product quality inspection of injection-molded busbars is crucial during production. Traditional quality inspection methods typically employ offline sampling inspection, which involves randomly selecting an arbitrary number of injection-molded busbars from a batch and then performing insulation resistance testing, breakdown voltage testing, and manual visual inspection. This method has the following drawbacks: Firstly, offline testing requires transferring products from the production line to the testing station, and then returning or sorting them after testing. This process is cumbersome and cannot meet the pace requirements of large-scale production, which will lead to a longer production cycle. Therefore, the testing efficiency of this method is low.

[0003] Secondly, the sampling inspection method can only cover a portion of the products, which is prone to omissions. This increases the quality risks and after-sales costs of the products after they leave the factory.

[0004] Third, most of the above methods rely on manual implementation, which requires a high level of experience from personnel, involves a significant proportion of subjective factors, and results in low overall defect identification accuracy and standardization.

[0005] Therefore, the existing technology mainly suffers from low efficiency and poor accuracy. Summary of the Invention

[0006] To address the aforementioned technical problems of low efficiency and poor accuracy, this invention discloses an online testing method and system for the insulation performance of injection-molded busbars.

[0007] In a first aspect, the present invention discloses an online testing method for the insulation performance of injection-molded busbars, comprising: In response to the detection of the injection molding busbar, the online full inspection mode is activated, and the following steps are performed: Collect electrical insulation data, ultrasonic scanning data, and visual image data of the injection-molded busbar; Electrical insulation data, ultrasonic scanning data, and visual image data are input into the corresponding defect analysis model to obtain the confidence level of insulation failure, internal defect confidence level, and surface defect confidence level. Based on the DS evidence theory, the confidence levels of insulation failure, internal defects, and surface defects are fused together to obtain the comprehensive defect judgment probability. The quality grade of the injection-molded busbar is determined based on the comprehensive defect judgment probability and the preset confidence interval. The injection molding motherboard is transferred or rejected based on its quality grade and the corresponding sorting mechanism. Preferably, the defect analysis model has a built-in gradient electrical threshold judgment algorithm; the electrical insulation data is input into the corresponding defect analysis model to obtain the insulation failure confidence level, specifically: Input the electrical insulation data into the gradient electrical threshold judgment algorithm and perform the following steps: Identify the insulation threshold range within which the electrical insulation data falls; Based on the insulation threshold range and the preset gradient range mapping table, the insulation failure confidence level corresponding to the electrical insulation data is assigned.

[0008] Preferably, the internal defect analysis model is equipped with a pre-trained vector machine classifier; the confidence level of internal defects includes the confidence level of internal bubbles, the confidence level of excessive cracks, and the confidence level of excessive impurity content.

[0009] Preferably, the ultrasonic scanning data is input into the corresponding defect analysis model to obtain the confidence level of the internal defect, specifically as follows: Feature vectors are constructed by extracting features from ultrasonic scanning data acquired through a multi-channel ultrasonic probe array. The feature vectors are input into a pre-trained vector machine classifier, which outputs the confidence scores for internal bubbles, excessive cracks, and excessive impurities.

[0010] Preferably, the defect analysis model for processing visual image data includes a backbone network, a neck network, and a prediction head; the visual image data is input into the corresponding defect analysis model to obtain the confidence level of the surface defect, specifically as follows: Input visual image data into the backbone network; In the output of each cross-stage local network module of the backbone network, the embedded convolutional block attention module is used to perform channel and spatial dimension weighting on the visual image data to extract a weighted feature map containing enhanced texture features. The weighted feature map is input into the neck network, and upsampling and splicing are performed to construct a multi-scale feature map that includes a defect detection layer. The multi-scale feature map is input into the prediction head, which performs boundary regression and class determination based on the anchor box, calculates the coordinate position of the surface defect and the corresponding probability value, and uses the probability value as the confidence level of the surface defect.

[0011] Preferably, the prediction head performs boundary regression and category determination based on the anchor frame to calculate the coordinate position of the surface defect and its corresponding probability value, including: Decode the multi-scale feature map to generate an original set of candidate boxes containing coordinate information, size information and probability information; The original candidate box set is traversed and filtered according to the defect morphology template of the motherboard, and the valid candidate boxes are output. Non-maximum suppression is applied to valid candidate boxes, and overlapping boxes are eliminated by intersection-union operation to lock the final predicted box; The base confidence score is obtained by multiplying the target probability and class probability of the final predicted bounding box. A small defect penalty term based on the defect area is introduced to weight and correct the base confidence score, and the surface defect confidence score is output.

[0012] Preferably, based on the DS evidence theory, the confidence levels of insulation failure, internal defects, and surface defects are fused together to obtain a comprehensive defect judgment probability, including: Based on the confidence levels of insulation failure, internal defects, and surface defects, basic probability assignment functions characterizing defects and normality are constructed respectively. Calculate the product of the intersection of the basic probability assignment functions using the orthogonal method; The sum of the products of the intersections of mutually exclusive propositions yields the conflict coefficient; The product of the intersection of non-mutually exclusive propositions is normalized using the conflict coefficient to generate a joint mass function after conflict elimination. The confidence and likelihood of the defective state are extracted from the joint quality function. The confidence and likelihood are weighted and summed based on a preset risk preference factor to obtain the comprehensive defect decision probability.

[0013] Preferably, the basic probability assignment function is as follows:

[0014] In the formula, Indicates the first Basic probability assignment functions for each dimension; Represents a defective state; Indicates the normal state. Indicates the first Confidence levels in each dimension.

[0015] Preferably, while the injection molding motherboard is being transferred or rejected by the corresponding sorting mechanism based on its quality grade, the method of the present invention simultaneously performs the following steps: The production batch, process parameters, quality grade, test data and defect information of injection-molded busbars are stored in the database.

[0016] Secondly, the present invention discloses an online testing system for the insulation performance of injection-molded busbars, comprising a processor and a memory, wherein the memory stores computer program instructions, and when the computer program instructions are executed by the processor, the online testing method for the insulation performance of injection-molded busbars described in the first aspect is implemented.

[0017] The beneficial effects of this invention are as follows: The method of this invention realizes automatic online full inspection in three dimensions: insulation detection, internal defect detection, and external defect detection, directly replacing traditional offline sampling inspection. On this basis, this method introduces data analysis model and multi-source data fusion technology to improve the efficiency of full inspection, thereby greatly improving the detection accuracy and efficiency, and avoiding missed detection. Attached Figure Description

[0018] The above and other objects, features, and advantages of exemplary embodiments of the present invention will become readily apparent upon reading the following detailed description with reference to the accompanying drawings. In the drawings, several embodiments of the invention are illustrated by way of example and not limitation, and like or corresponding reference numerals denote like or corresponding parts, wherein: Figure 1 This is a flowchart of the online testing method for the insulation performance of injection-molded busbars in Embodiment 1 of the present invention; Figure 2 This is a schematic diagram of the online testing system for the insulation performance of injection-molded busbars in Embodiment 2 of the present invention. Detailed Implementation

[0019] 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, not all, of the embodiments of the present invention. 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.

[0020] The specific embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0021] Example 1 like Figure 1 As shown in the figure, this embodiment discloses an online testing method for the insulation performance of injection-molded busbars, including: In response to the detection of the injection molding busbar, the online full inspection mode is activated, and the following steps are performed: S10: Collect electrical insulation data, ultrasonic scanning data, and visual image data of the injection-molded busbar.

[0022] In this embodiment, before executing step S10, a conveying unit, i.e., a high-precision conveyor belt driven by a servo motor, is used to transport the injection-molded busbar to the clamping mechanism. Positioning data is collected by a preset positioning sensor to determine whether the injection-molded busbar is in position. When the clamping mechanism has performed the clamping action and the positioning data is detected, step S10 is triggered. The clamping end of the clamping mechanism can be made of a flexible material to avoid damage to the surface or insulation layer of the injection-molded busbar, while ensuring stability during the detection process.

[0023] Once the injection busbar is detected in place, the electrical performance testing module, ultrasonic testing module, and visual testing module are activated to collect the aforementioned electrical insulation data, ultrasonic scanning data, and visual image data.

[0024] Preferably, the aforementioned electrical performance testing module includes a high-voltage insulation resistance tester and a breakdown voltage testing module, which mainly uses customized probe components to contact the conductive terminals of the injection-molded busbar. During testing, a 500V or 1000V DC voltage is first applied to test the insulation resistance, and the test range is selected as 10. 6 -10 12 Ω, then gradually increase the AC voltage to 1.2 times the rated breakdown voltage, and monitor / record the changes in leakage current. The above probe assembly adopts a floating design, which can adapt to the terminal spacing of different busbar specifications, and its contact resistance is ≤0.1Ω to ensure detection accuracy.

[0025] Preferably, the ultrasonic testing module employs a multi-channel ultrasonic probe array with 8-16 channels and a center frequency of 5-10MHz. The ultrasonic scanning range must cover the entire insulation layer area of ​​the injection-molded busbar. The ultrasonic signal is amplified by a preamplifier and then converted into a digital signal by a data acquisition card. This module can penetrate the insulation layer to detect hidden defects such as internal bubbles, cracks, or impurities, with a detection depth range of 1-50mm.

[0026] Preferably, the aforementioned visual inspection module consists of a high-resolution industrial camera (5 megapixels or higher), a ring light source, and a lens. It employs a multi-angle shooting scheme, with two cameras each on the front and side for image capture. The ring light source provides uniform illumination to avoid glare interference. The camera has a frame rate of ≥30fps, enabling real-time capture of images of the busbar surface. This module is used to detect defects such as insulation surface damage, scratches, uneven color, and edge burrs, achieving full surface coverage inspection of the busbar through image stitching technology.

[0027] It should be noted that before executing step S20, the data collected by each module needs to be denoised or normalized to improve the quality of the raw data.

[0028] For the preprocessing of electrical insulation data, the Kalman filter algorithm can be used to remove random noise during the test process and normalize the insulation resistance value and leakage current data to the [0,1] interval to eliminate data differences under different test voltages.

[0029] For the preprocessing of ultrasound scanning data, wavelet transform algorithms, such as the db4 wavelet basis function, can be used to denoise the ultrasound echo signal, extract key parameters such as the signal peak value, time domain width, and frequency domain features, and construct an ultrasound feature vector as the aforementioned ultrasound scanning data.

[0030] For the preprocessing of visual image data, grayscale conversion, Gaussian blur denoising, and histogram equalization can be performed to enhance the contrast between the defect area and the background. Edge detection algorithms can be used to extract image edge information to prepare for subsequent defect localization.

[0031] S20: Input electrical insulation data, ultrasonic scanning data, and visual image data into the corresponding defect analysis model to obtain the confidence level of insulation failure, the confidence level of internal defects, and the confidence level of surface defects.

[0032] The aforementioned defect analysis model includes a gradient electrical threshold judgment algorithm, a vector machine classifier, and a visual defect analysis model.

[0033] Specifically, for calculating the confidence level of insulation failure, step S20 above includes: S210: Input the electrical insulation data into the gradient electrical threshold judgment algorithm and perform the following steps: S211: Identify the insulation threshold range in which the electrical insulation data falls.

[0034] In this embodiment, the aforementioned electrical insulation data mainly includes insulation resistance value and leakage current. Correspondingly, the insulation threshold includes insulation resistance threshold and leakage current threshold. Preferably, the insulation resistance threshold is 10. 9 Ω, while the leakage current is 1mA.

[0035] S212: Assign insulation failure confidence level to electrical insulation data according to the insulation threshold range and the preset gradient range mapping table.

[0036] For example, the gradient range mapping table described above can be:

[0037] Specifically, step S212 is essentially a lookup table method. The confidence level of insulation failure and the related threshold can be configured according to the actual situation, and the corresponding confidence level of insulation failure must be in the range of 0-1.

[0038] By using the steps S210-S212 above, it is possible to more quickly and conveniently make a preliminary judgment on whether the insulation layer of the injection-molded busbar is damaged or has serious impurities.

[0039] It should be noted that the preliminary test results can be temporarily stored in memory for later use in generating test reports.

[0040] Specifically, for calculating the confidence level of internal defects, step S20 above includes: S220: Extract features from the ultrasonic scanning data acquired by the multi-channel ultrasonic probe array and construct a feature vector.

[0041] The ultrasonic echo signal is denoised using a wavelet transform algorithm. Key parameters such as peak value, time domain width, and frequency domain features are extracted to construct the aforementioned feature vector.

[0042] S221: Input the feature vector into a pre-trained vector machine classifier and output the confidence scores for internal bubbles, excessive cracks, and excessive impurities.

[0043] It should be noted that the vector machine classifier can be trained by feeding over 1000 sets of historical internal defect sample data, and instructing the classifier to classify three types of defects: internal, cracks, and impurities. Specifically, the diameter of an internal bubble must be greater than or equal to 0.5 mm, the length of a crack must be greater than or equal to 1 mm, and the volume of an internal impurity must be greater than or equal to 1 mm³. The vector machine classifier can not only determine the presence or absence of these three types of defects, but also the number of occurrences. Therefore, based on the number of defects and their types, and relying on the internal training results, the over-confidence level of these three types of defects can be determined.

[0044] Through the above steps S220-S221, various internal defects of the injection molding motherboard can be accurately identified, with an accuracy rate greater than or equal to 98%.

[0045] Similarly, the confidence levels exceeding the standard for the above three types of defects are also temporarily stored in the memory as preliminary detection results for later use in generating detection reports.

[0046] Specifically, for calculating the confidence level of surface defects, this embodiment designs a vision-based defect analysis model. This vision-based defect analysis model includes a backbone network, a neck network, and a prediction head. More specifically, the steps for calculating the confidence level of surface defects using the above defect analysis model are as follows: S230: Input visual image data into the backbone network.

[0047] It should be noted that the specific design of the backbone network can refer to the YOLOv5 object detection algorithm. Before executing step S230, the defect analysis model needs to be fed with no less than 5000 image defect / normal samples to adapt to the defect analysis scenario in this field.

[0048] S231: At the output of each cross-stage local network module of the backbone network, the embedded convolutional block attention module is used to perform channel and spatial dimension weighting on the visual image data to extract a weighted feature map containing enhanced texture features.

[0049] It should be explained that the above convolutional block attention module includes a channel attention submodule and a spatial attention submodule, which are executed in series. The channel attention submodule is used to enhance the response to scratch texture features and suppress interference from the natural texture of the injection molding busbar surface. The spatial attention submodule is used to accurately locate the spatial position of fine scratches.

[0050] More specifically, in the channel attention submodule, the initial feature map of the visual image data can be defined as follows: The initial feature maps are first processed by global max pooling and global average pooling to obtain two 1×1×C descriptors. These descriptors are then processed by a shared multilayer perceptron (MLP) and added together. Finally, they are processed by a sigmoid activation function to generate channel-processed feature maps.

[0051] In contrast, the spatial attention submodule receives the channel processing feature map mentioned above, and then performs max pooling and average pooling along the channel axis respectively. After averaging, it is processed by a 7×7 convolutional layer to finally generate the weighted feature map mentioned above.

[0052] S232: Input the weighted feature map into the neck network, perform upsampling and splicing, and construct a multi-scale feature map that includes a defect detection layer.

[0053] It should be explained that the original YOLOv5 outputs feature maps at three scales (80×80, 40×40, 20×20), corresponding to layers P3, P4, and P5. For a microcrack with a width of only 0.1mm on an injection molding busbar, the feature information is almost completely lost after 32x downsampling. Therefore, the method in this embodiment adds a P2 detection head with a resolution of 160×160 to the neck network. Specifically, in the feature pyramid structure, the feature map of layer 1 of the backbone network is concatenated with the upsampled feature map of layer P3 to construct a high-resolution detection layer.

[0054] S233: Input the multi-scale feature map into the prediction head, and the prediction head performs boundary regression and category determination based on the anchor box to calculate the coordinate position of the surface defect and the corresponding probability value, and use the probability value as the confidence level of the surface defect.

[0055] It should be explained that because the scratches on injection molding motherboards are typically "long and thin" or "curved," with extreme aspect ratios (e.g., 1:10 or even 1:20), while the anchor boxes in the default YOLO COCO dataset are more square, it is necessary to re-cluster the bounding boxes. The preferred clustering algorithm is K-Means. More specifically, the distance metric function is defined as follows:

[0056] In the formula, Represents the bounding box. Indicates the cluster center. Represents the intersection-union ratio function. This indicates the definition of a distance metric function.

[0057] Based on the above, the SIoU (Scylla-IoU) loss function is introduced, and an angle penalty term is added to the loss function.

[0058] Specifically, the SIoU loss function is as follows:

[0059] In the formula, This represents the value of the intersection-union function. Indicates distance cost, Indicates the cost of angle.

[0060] The formula for calculating angle cost is as follows:

[0061] In the formula, This represents the height difference between the center points of the ground truth bounding boxes (labeled boxes) and the predicted bounding boxes (clustered boxes). This represents the Euclidean distance between the center points of the two boxes.

[0062] As for the distance cost, it is calculated by reweighting the aforementioned angle cost.

[0063] After introducing the above adaptive improvement strategy, the model can fit the rotation angle of the long strip scratch more quickly in the early stage of training, which improves the fit between the final output coordinate box and the real defect by more than 15%.

[0064] Furthermore, step S233 above includes: S2331: Decode the multi-scale feature map to generate a set of original candidate boxes containing coordinate information, size information and probability information.

[0065] Specifically, the coordinate information includes the two-dimensional coordinates of the center of the bounding box, and the size information mainly refers to the length and width of the bounding box. The original candidate box set consists of multiple bounding boxes, each containing the probability of the target's presence and the probability of its class.

[0066] S2332: Based on the defect morphology template of the busbar, traverse and filter the original candidate box set, and output the valid candidate boxes.

[0067] Specifically, based on physical characteristics such as scratches / damage / burrs on the surface of injection-molded busbars, morphological constraints are preset to perform initial screening of candidate detection frames for effectiveness. The aspect ratio of the detection frames is calculated. If the aspect ratio is less than a preset threshold (e.g., 3:1), it is determined to be artifact noise and is removed.

[0068] S2333: Perform non-maximum suppression on valid candidate boxes and eliminate overlapping boxes through cross-union operation to lock the final predicted box.

[0069] S2334: Calculate the product of the target probability and the class probability of the final predicted box to obtain the basic confidence score, and introduce a small defect penalty term based on the defect area to weight and correct the basic confidence score, and output the surface defect confidence score.

[0070] Specifically, the product of the target presence probability and the class probability in the final predicted bounding box is defined as the base confidence score. Based on the size and area of ​​the final predicted bounding box, a small defect penalty term is introduced to correct the base confidence score, and the corrected value is used as the final output surface defect confidence score.

[0071] More specifically, the formula for calculating the confidence level of the aforementioned surface defects is as follows:

[0072] In the formula, Indicates the confidence level of surface defects. Indicates the base confidence level. This represents the area of ​​the final predicted bounding box. Indicates the defect area threshold. This represents a decay factor greater than 0.

[0073] S30: Based on the DS evidence theory, the confidence levels of insulation failure, internal defects, and surface defects are fused together to obtain the comprehensive defect judgment probability.

[0074] Specifically, step S30 includes: S31: Based on the confidence levels of insulation failure, internal defects, and surface defects, construct basic probability assignment functions to characterize defects and normality, respectively.

[0075] The basic probability assignment function mentioned above can be:

[0076] In the formula, Indicates the first Basic probability assignment functions for each dimension; Represents a defective state; Indicates the normal state. Indicates the first Confidence levels in each dimension.

[0077] S32: Calculate the intersection product of the basic probability assignment functions using the orthogonal method.

[0078] S33: The sum of the products of the intersections of mutually exclusive propositions is the conflict coefficient.

[0079] The formula for calculating the conflict coefficient is as follows:

[0080] In the formula, No. Each dimension of defect state This indicates the case where the intersection of the three dimensions is empty. The aforementioned conflict coefficient is used to describe the overall reliability quality of all evidence that completes the conflict.

[0081] S34: Normalize the product of the intersection of non-mutually exclusive propositions using the conflict coefficient to generate a joint quality function after conflict elimination.

[0082] The expression for the joint mass function is:

[0083] In the formula, The intersection of the three dimensions is represented as The situation.

[0084] Similarly, under normal conditions Alternatively, it can be obtained using the methods described above, which will not be elaborated upon here.

[0085] S35: Extract the confidence and likelihood for the defective state from the joint quality function, and sum the confidence and likelihood based on the preset risk preference factor to obtain the comprehensive defect decision probability.

[0086] Here, the confidence level is equivalent to the solution value of the joint quality function. And the likelihood level... The calculation formula is:

[0087] Furthermore, the probability of a comprehensive defect decision can be:

[0088] In the formula, This represents a weighting coefficient that ranges from 0 to 1.

[0089] Through the above steps S31-S35, the method of this embodiment successfully integrates the three dimensions of insulation failure, internal defects and surface defects, and the resulting comprehensive analysis results are more accurate and reliable.

[0090] S40: Determine the quality level of the injection-molded busbar based on the comprehensive defect judgment probability and the preset confidence interval.

[0091] For example, the confidence level range can be divided into three levels. The first level is greater than or equal to 0.9, which indicates the presence of a type defect and is considered a non-conforming product. The second level is 0.7-0.9, which indicates a good product that may require re-inspection. The third level is less than 0.7, which indicates the absence of a defect and is considered a superior product.

[0092] S50: Drives the corresponding sorting mechanism to transfer or reject injection molding motherboards according to the quality level.

[0093] Specifically, for superior products, the production line continues normally. For good products, a second inspection is conducted. If the quality is good or superior after the second inspection, the production line continues normally. For defective products, the sorting / diversion mechanism is activated to remove them to the storage bin.

[0094] It should be noted that during the execution of steps S10-S50 above, the method of this embodiment simultaneously stores the production batch, process parameters, quality grade, inspection data and defect information of the injection molding busbar into the database through the server, and connects with the MES system in real time.

[0095] Through the above steps S10-S50, the method of this embodiment has at least the following technical advantages: Firstly, it enables online real-time full inspection of injection molding busbars, with the inspection cycle synchronized with the production line, replacing traditional offline sampling inspection, reducing manual intervention, and increasing production efficiency by more than 20%.

[0096] Secondly, by employing multi-dimensional detection and multi-source data fusion technology, it can identify various surface and internal defects with a defect identification accuracy rate of over 99%, a false detection rate of less than 0.5%, and a missed detection rate of less than 0.1%, effectively reducing the risk of defective products leaving the factory.

[0097] Third, establishing a database linking testing data and production information allows for rapid tracing of production batches and process parameters of non-conforming products, providing data support for process optimization. By analyzing the correlation between defect types and process parameters, parameters such as injection molding temperature and pressure can be adjusted in a targeted manner, reducing the defect rate by more than 15%.

[0098] Fourth, it avoids after-sales costs and brand losses caused by defective products entering the market. The reduction in defect rate brought about by process optimization further reduces raw material waste and rework costs, which can reduce production costs by 10%-20%.

[0099] Example 2 like Figure 2 As shown, this embodiment discloses an online testing system for the insulation performance of injection-molded busbars, including a processor and a memory. The memory stores computer program instructions, and when the computer program instructions are executed by the processor, the online testing method for the insulation performance of injection-molded busbars described in Embodiment 1 is implemented.

[0100] The system in this embodiment also includes other components well known to those skilled in the art, such as communication interfaces. Their settings and functions are known in the art, and therefore will not be described in detail here.

[0101] In this invention, the aforementioned memory can be any tangible medium containing or storing a program that can be used or combined with an instruction execution system, apparatus, or device. For example, a computer-readable storage medium can be any suitable magnetic or magneto-optical storage medium, such as Resistive Random Access Memory (RRAM), Dynamic Random Access Memory (DRAM), Static Random Access Memory (SRAM), Enhanced Dynamic Random Access Memory (EDRAM), High-Bandwidth Memory (HBM), Hybrid Memory Cube (HMC), etc., or any other medium that can be used to store desired information and can be accessed by an application, module, or both. Any such computer storage medium can be part of a device or accessible to or connected to a device. Any application or module described in this invention can be implemented using computer-readable / executable instructions that can be stored or otherwise maintained by such a computer-readable medium.

[0102] In the description of this specification, "multiple" means at least two, such as two, three or more, etc., unless otherwise expressly and specifically defined.

[0103] While this specification has shown and described numerous embodiments of the invention, it will be apparent to those skilled in the art that such embodiments are provided by way of example only. Many modifications, alterations, and alternatives will occur to those skilled in the art without departing from the spirit and essence of the invention. It should be understood that various alternatives to the embodiments of the invention described herein may be employed in the practice of this invention.

Claims

1. A method for online testing of the insulation performance of injection-molded busbars, characterized in that, include: In response to the detection of the injection molding busbar, the online full inspection mode is activated, and the following steps are performed: The electrical insulation data, ultrasonic scanning data, and visual image data of the injection-molded busbar are collected. The electrical insulation data, ultrasonic scanning data, and visual image data are input into the corresponding defect analysis model to obtain the confidence level of insulation failure, the confidence level of internal defects, and the confidence level of surface defects. Based on the DS evidence theory, the confidence levels of insulation failure, internal defects, and surface defects are fused together to obtain a comprehensive defect judgment probability. The quality grade of the injection-molded busbar is determined based on the comprehensive defect judgment probability and the preset confidence interval. The sorting mechanism driven by the quality grade will transfer or reject the injection molding motherboard.

2. The online testing method for the insulation performance of injection-molded busbars according to claim 1, characterized in that, The defect analysis model incorporates a gradient-based electrical threshold judgment algorithm; by inputting the electrical insulation data into the corresponding defect analysis model, the insulation failure confidence level is obtained, specifically as follows: The electrical insulation data is input into the gradient electrical threshold judgment algorithm, and the following steps are performed: Identify the insulation threshold range within which the electrical insulation data falls; Based on the insulation threshold range and the preset gradient range mapping table, the electrical insulation data is assigned an insulation failure confidence level.

3. The online testing method for the insulation performance of injection-molded busbars according to claim 1, characterized in that, The internal defect analysis model is equipped with a pre-trained vector machine classifier; the confidence level of the internal defect includes the confidence level of internal bubbles, the confidence level of excessive cracks, and the confidence level of excessive impurity content.

4. The online testing method for the insulation performance of injection-molded busbars according to claim 3, characterized in that, The ultrasonic scanning data is input into the corresponding defect analysis model to obtain the confidence level of the internal defect, specifically as follows: Feature vectors are constructed by extracting features from ultrasonic scanning data acquired through a multi-channel ultrasonic probe array. The feature vector is input into a pre-trained vector machine classifier, which outputs the confidence scores for the internal bubble, the crack exceeding the standard, and the impurity content exceeding the standard.

5. The online testing method for the insulation performance of injection-molded busbars according to claim 1, characterized in that, The defect analysis model for processing the visual image data includes a backbone network, a neck network, and a prediction head; the visual image data is input into the corresponding defect analysis model to obtain the surface defect confidence score, specifically: The visual image data is input into the backbone network; In the output of each cross-stage local network module of the backbone network, the visual image data is weighted by channel and spatial dimensions using an embedded convolutional block attention module to extract a weighted feature map containing enhanced texture features. The weighted feature map is input into the neck network, and upsampling and splicing are performed to construct a multi-scale feature map including a defect detection layer. The multi-scale feature map is input into the prediction head, which performs boundary regression and category determination based on the anchor box to calculate the coordinate position of the surface defect and the corresponding probability value, and uses the probability value as the confidence level of the surface defect.

6. The online testing method for the insulation performance of injection-molded busbars according to claim 5, characterized in that, The prediction head performs boundary regression and category determination based on the anchor frame to calculate the coordinates and corresponding probability values ​​of surface defects, including: The multi-scale feature map is decoded to generate an original set of candidate boxes containing coordinate information, size information, and probability information. The original candidate box set is traversed and filtered according to the busbar defect morphology template to output valid candidate boxes; Non-maximum suppression is applied to the effective candidate boxes, and overlapping boxes are eliminated by cross-union operation to lock the final predicted box; The base confidence score is obtained by multiplying the target probability and the class probability of the final predicted box. A small defect penalty term based on the defect area is introduced to weight and correct the base confidence score, and the surface defect confidence score is output.

7. The online testing method for the insulation performance of injection-molded busbars according to claim 1, characterized in that, Based on the DS evidence theory, the confidence levels of insulation failure, internal defects, and surface defects are fused together to obtain a comprehensive defect judgment probability, including: Based on the confidence levels of insulation failure, internal defects, and surface defects, basic probability assignment functions characterizing defects and normality are constructed respectively. Calculate the product of the intersection of the basic probability assignment functions using the orthogonal method; The sum of the products of the intersections of mutually exclusive propositions yields the conflict coefficient; The conflict coefficient is used to normalize the product of the intersection of non-mutually exclusive propositions to generate a joint quality function after conflict elimination. The confidence and likelihood for the defective state are extracted from the joint quality function. The confidence and likelihood are then weighted and summed based on a preset risk preference factor to obtain the comprehensive defect decision probability.

8. The online testing method for the insulation performance of injection-molded busbars according to claim 7, characterized in that, The basic probability assignment function is as follows: In the formula, Indicates the first Basic probability assignment functions for each dimension; Represents a defective state; Indicates the normal state. Indicates the first Confidence levels in each dimension.

9. The online testing method for the insulation performance of injection-molded busbars according to claim 1, characterized in that, While the sorting mechanism drives the corresponding quality grade to transfer or reject the injection molding motherboard, the method simultaneously performs the following steps: The production batch, process parameters, quality grade, test data and defect information of the injection-molded busbar are stored in the database.

10. An online testing system for the insulation performance of injection-molded busbars, characterized in that, It includes a processor and a memory, wherein the memory stores computer program instructions, and when the computer program instructions are executed by the processor, the online detection method for the insulation performance of injection-molded busbars as described in any one of claims 1-9 is implemented.

Images