A method and system for defect data acquisition and reliable transmission for industrial inspection

Abstract

This invention provides a method and system for defect data acquisition and reliable transmission in industrial inspection, belonging to the field of industrial manufacturing and data analysis technology. The method includes: analyzing and processing each associated defect data acquired at the edge of the equipment to identify risk data; uploading each identified risk data to the cloud; performing defect identification on the defect images in each risk data in the cloud to filter out valid defect data and form a valid defect data set; and performing data analysis on the valid defect data set to optimize pre-associated process parameters. This invention effectively reduces data transmission volume and cloud computing load by combining edge-end pre-screening with precise cloud identification, establishing a closed loop between defect analysis and process parameter optimization, and achieving data-driven yield management.

Description

Technical Field

[0001] This invention relates to the fields of industrial manufacturing and data analysis technology, and in particular to a method and system for defect data acquisition and reliable transmission for industrial inspection. Background Technology

[0002] As industrial manufacturing moves towards intelligence and precision, the requirements for product yield in high-end manufacturing fields such as display panels and semiconductors are increasing. Production lines are equipped with a large number of visual inspection devices (such as automated optical inspection equipment and Demura repair equipment) for defect detection and repair. These devices generate massive amounts of defect images and related data during daily operation. Efficiently analyzing and utilizing this data is crucial for achieving factory-wide yield management.

[0003] Currently, the industry generally adopts an "edge-cloud" architecture for defect data collection and analysis. This involves various inspection devices deployed on the production line collecting raw defect data, performing preliminary processing via edge computing nodes, and then uploading it to a cloud server for in-depth analysis and decision-making. However, existing technical solutions still have the following problems in practical applications: First, the collaboration efficiency between the edge and the cloud is low. In existing solutions, the edge typically only handles simple data aggregation and forwarding. A large amount of raw data needs to be uploaded to the cloud for unified analysis, resulting in excessive cloud computing load and massive data transmission, making it difficult to quickly filter out truly effective defect data. While some solutions attempt preliminary detection at the edge, they lack credibility assessment of the detection results and cannot effectively distinguish between "confirmed defects," "suspected defects," and "normal product" data. This forces the cloud to still conduct a comprehensive review of all data, failing to truly leverage the advantages of cloud-edge collaboration. Furthermore, in scenarios with large amounts of concurrent data uploads, existing solutions lack effective management of data transmission priorities. High-value data competes with ordinary data for network resources, further reducing the overall efficiency of cloud-edge collaboration.

[0004] Second, the reliability of data transmission needs improvement. Industrial network environments are complex, and transmission links may experience packet loss or interruption due to equipment failure, electromagnetic interference, or other factors. Existing solutions often use a single transmission channel for data uploading, lacking redundancy mechanisms. When a link malfunctions, critical defect data may be lost or delayed, affecting the timeliness and accuracy of subsequent analysis and decision-making.

[0005] Third, the closed loop between defect data analysis and process parameter optimization has not yet been established. In existing solutions, defect detection and process parameter adjustment are usually two separate processes. Defect data is collected and analyzed by the quality department, while process parameters are adjusted by the production department based on experience. There is a lack of a data-driven, automated linkage mechanism between the two. Even if defect types and distribution patterns can be identified from defect data, it is difficult to trace back to the corresponding process production parameters, failing to form an iterative closed loop of "defect analysis, parameter optimization, and production verification," resulting in low yield management efficiency.

[0006] Therefore, there is an urgent need for a method and system that can achieve efficient collaboration between the edge and the cloud, accurately filter out effective data from massive amounts of data and reliably transmit it to the cloud, so as to provide a data foundation for data-driven yield management across the entire factory. Summary of the Invention

[0007] This invention provides a method and system for defect data acquisition and reliable transmission for industrial inspection, in order to solve at least one defect existing in the prior art.

[0008] In a first aspect, the present invention provides a method for defect data acquisition and reliable transmission for industrial inspection, comprising: analyzing and processing each associated defect data acquired at the edge of the device to identify risk data; wherein each associated defect data includes original defect data and corresponding process parameter data, the original defect data including at least a defect image and a corresponding original defect label; uploading each identified risk data to the cloud, performing defect identification on the defect image in each risk data in the cloud to filter out valid defect data from the risk data to form a valid defect data set; performing data analysis on the valid defect data set to optimize pre-associated process production parameters.

[0009] According to the defect data acquisition and reliable transmission method for industrial inspection provided by the present invention, each associated defect data acquired at the device end is analyzed and processed at the edge end to identify risk data, including: performing defect detection on the defect image in each associated defect data at the edge end; determining risk data from the associated defect data based on the defect detection results at the edge end; wherein the defect detection results include detection result confidence and defect size.

[0010] According to the defect data acquisition and reliable transmission method for industrial inspection provided by the present invention, before analyzing and processing each associated defect data acquired by the equipment end at the edge end to identify risk data, the method further includes: associating the original defect data acquired by the equipment end with the corresponding process production parameters to form associated defect data; and determining the transmission priority of the associated defect data based on the data attributes of the associated defect data.

[0011] According to the defect data acquisition and reliable transmission method for industrial inspection provided by the present invention, the transmission priority of the associated defect data is determined based on the data attributes of the associated defect data, including: acquiring the data attributes of the associated defect data; inputting the data attributes into a pre-built regression model and outputting a predicted loss amount; wherein the regression model is pre-trained based on data attribute samples and corresponding loss amount samples; determining the transmission priority of the associated defect data based on the predicted loss amount; wherein the transmission priority of the associated defect data is positively correlated with the predicted loss amount.

[0012] The defect data acquisition and reliable transmission method for industrial inspection provided by the present invention further includes: determining a quantified value of the transmission priority of the associated defect data; when the quantified value is greater than or equal to a preset threshold, transmitting the data using a dual-channel redundant transmission mode; wherein the dual channels include a timeliness channel and a reliability channel; when the quantified value is less than the preset threshold, determining the relative magnitude of the timeliness and reliability requirements of the associated defect data based on the data attributes of the associated defect data, so as to select a corresponding channel from the dual channels for data transmission.

[0013] According to the defect data acquisition and reliable transmission method for industrial inspection provided by the present invention, defect identification is performed on defect images in each risk data in the cloud to filter out valid defect data. The method includes: using multiple heterogeneous defect detection models to identify the defect images; if the detection results of the multiple heterogeneous defect detection models are consistent, the detection results of the multiple heterogeneous defect detection models are taken as the final detection result to filter out valid defect data; if the detection results of the multiple heterogeneous defect detection models are inconsistent, a preset defect detection model is used to identify the defect images; if the detection result of the preset defect detection model is consistent with the original defect label, the detection result of the preset defect detection model is taken as the final detection result to filter out valid defect data; if the detection result of the preset defect detection model is inconsistent with the original defect label, a manual review method is used to filter out valid defect data.

[0014] According to the defect data acquisition and reliable transmission method for industrial inspection provided by the present invention, data analysis is performed on the effective defect data set, and pre-associated process production parameters are optimized, including: acquiring OK product data and the corresponding process production parameters of the OK products, and constructing an OK product association data set; based on the OK product association data set and the effective defect data set, mining the correlation patterns between process parameters and defects, and forming optimized process production parameters for different defects; updating the optimized process production parameters to the corresponding production equipment for production.

[0015] According to the defect data acquisition and reliable transmission method for industrial inspection provided by the present invention, based on the OK product association data set and the effective defect data set, the method mines the correlation patterns between process parameters and defects, and forms optimized process production parameters for different defects. This includes: comparing and analyzing the value distribution of defective product samples and OK product samples on various process parameters to analyze the significance of the differences and determine the key influencing parameters associated with specific defect types; determining the optimal value range of the key influencing parameters based on the parameter distribution of normal products in the OK product association data set; and adjusting the current value of the key influencing parameters to the optimal value range to form the optimized process production parameters.

[0016] Secondly, the present invention also provides a defect data acquisition and reliable transmission system for industrial inspection, comprising: The first processing module is used to analyze and process each associated defect data collected by the device at the edge to identify risk data; wherein each associated defect data includes original defect data and corresponding process parameter data, and the original defect data includes at least a defect image and a corresponding original defect label; The second processing module is used to upload each identified risk data to the cloud, and to perform defect identification on the defect image in each risk data in the cloud, so as to filter out the valid defect data from the risk data to form a set of valid defect data. The third processing module is used to perform data analysis on the effective defect data set and optimize the pre-associated process production parameters.

[0017] Thirdly, the present invention provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the steps of the defect data acquisition and reliable transmission method for industrial inspection as described above.

[0018] The defect data acquisition and reliable transmission method and system for industrial inspection provided by this invention have the following advantages compared with the prior art: (1) This invention analyzes and processes associated defect data at the edge to pre-identify risk data, and only uploads the filtered risk data to the cloud for accurate defect identification, rather than uploading all the original data. This mechanism effectively filters a large amount of normal product data, significantly reduces the amount of data transmitted between the edge and the cloud, and alleviates the computational pressure on the cloud to repeatedly analyze the full amount of data, truly leveraging the advantages of the cloud-edge collaborative architecture.

[0019] (2) Based on the set of valid defect data selected from risk data, this invention combines OK product data and corresponding process production parameters to automatically mine the correlation between process parameters and defects, and generates optimized process production parameters to update the production equipment. This method changes the situation where defect analysis and process parameter adjustment are separated in the traditional solution, forming an automated iterative closed loop of "defect detection, data analysis, parameter optimization, and production verification", which effectively improves the efficiency and scientific nature of yield management.

[0020] (3) This invention determines the transmission priority of associated defect data and adopts a dual-channel redundant transmission mode when the transmission priority is higher than the preset threshold. The data is transmitted through the timeliness channel and the reliability channel at the same time, which ensures the reliable delivery of high-value risk data in complex network environment and avoids the problem of key data loss due to network anomalies.

[0021] (4) This invention uses multiple heterogeneous defect detection models in the cloud to cross-validate risk data, and introduces a preset model review and manual verification mechanism when the model results are inconsistent. Through multi-level progressive judgment logic, the risk of misjudgment or omission by a single model is effectively reduced, and the accuracy and credibility of the final effective defect data are significantly improved. Attached Figure Description

[0022] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0023] Figure 1 This is a flowchart illustrating the defect data acquisition and reliable transmission method for industrial inspection provided by the present invention. Figure 2 This is a schematic diagram of the structure for defect data acquisition and reliable transmission for industrial inspection provided by the present invention; Figure 3 This is a schematic diagram of the structure of the electronic device provided by the present invention. Detailed Implementation

[0024] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this 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 this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0025] It should be noted that, in the description of the embodiments of the present invention, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element. Those skilled in the art can understand the specific meaning of the above terms in the present invention according to the specific circumstances.

[0026] The following is combined Figures 1-3 This invention describes a method and system for defect data acquisition and reliable transmission for industrial inspection, provided by embodiments of the present invention.

[0027] Figure 1 This is a flowchart illustrating the defect data acquisition and reliable transmission method for industrial inspection provided by the present invention, as shown below. Figure 1 As shown, including but not limited to the following steps: The present invention provides a defect data acquisition and reliable transmission method for industrial inspection, comprising: Step 101: Analyze and process each associated defect data collected by the device at the edge to identify risk data; wherein, each associated defect data includes original defect data and corresponding process parameter data, and the original defect data includes at least a defect image and a corresponding original defect label.

[0028] Specifically, this step is performed by edge computing nodes deployed at the edge of the production line to perform preliminary screening of massive amounts of raw data, filtering out obviously normal product data and uploading only risky data that may have defects to the cloud, thereby reducing the amount of data transmitted and the cloud computing load.

[0029] The equipment side refers to the inspection and production equipment deployed at each stage of the manufacturing process, such as automated optical inspection (AOI) equipment and microscopic hyperspectral imaging equipment. These devices continuously collect raw defect data of the products during operation, including but not limited to defect images. Simultaneously, the manufacturing equipment records process parameter data for each product at different stages of the process, such as temperature, pressure, speed, raw material ratio, and process time.

[0030] The analysis and processing of associated defect data at the edge can be carried out by a lightweight defect detection model deployed at the edge to detect defect images in real time and output detection results, including the confidence level of the detection results, defect size (such as length and width), and other information, to further identify risk data.

[0031] The original defect label refers to the preliminary judgment result (such as OK / NG mark or specific defect type) generated by the detection device's own algorithm (or the device can set an existing general defect detection model) when collecting data. This label can be used as one of the reference bases for subsequent cloud identification.

[0032] The edge pre-screening mechanism in this step can significantly reduce the data transmission scale.

[0033] Step 102: Upload each identified risk data to the cloud, and perform defect identification on the defect image in each risk data in the cloud to filter out valid defect data from the risk data to form a set of valid defect data.

[0034] Specifically, this step, executed on the cloud server, is the core process for accurately identifying and classifying risk data uploaded from the edge. The main task at the edge is to quickly filter out data suspected of being problematic, but the accuracy and complexity of its detection model are relatively limited. The cloud, on the other hand, utilizes richer computing resources and larger-scale models to perform more refined defect identification on the risk data, ultimately determining the valid defect data.

[0035] Step 103: Perform data analysis on the effective defect data set and optimize the pre-associated process production parameters.

[0036] Specifically, this step achieves a closed loop from defect data to process parameter optimization, changing the situation where defect detection and process adjustment are separated in the traditional approach.

[0037] This invention, based on a set of valid defect data selected from risk data, combined with OK product data and corresponding process production parameters, automatically mines the correlation between process parameters and defects, and generates optimized process production parameters that are then updated to the production equipment. This method changes the traditional approach where defect analysis and process parameter adjustment are separate processes, forming an automated iterative closed loop of "defect detection, data analysis, parameter optimization, and production verification," effectively improving the efficiency and scientific rigor of yield management.

[0038] Based on the above embodiments, as an optional embodiment, the defect data acquisition and reliable transmission method for industrial inspection provided by the present invention analyzes and processes each associated defect data acquired at the device end at the edge to identify risk data, including: (1) Defect detection is performed on the defect image in each associated defect data at the edge.

[0039] Specifically, this step is performed by edge computing nodes deployed at the edge. Lightweight defect detection models are deployed on these edge computing nodes for real-time inference of defect images.

[0040] It should be noted that this lightweight defect detection model can adopt existing, publicly available general-purpose target detection model architectures, such as models based on lightweight network structures like the YOLO series (e.g., YOLOv5, YOLOv8) or MobileNet-SSD. This invention does not specifically limit these models. The main function of the edge-end model is to leverage its fast inference speed and low resource consumption to perform rapid initial screening of massive defect images at the production line edge. The improvement of this invention at the edge end lies not in the specific network structure of the defect detection model, but in how to use the detection results output by the model for subsequent graded risk assessment.

[0041] Once the associated defect data reaches the edge, the edge computing nodes extract the defect images and input them into the deployed defect detection model for inference and analysis. The defect detection model analyzes the images and outputs detection results related to the defects.

[0042] (2) Based on the defect detection results at the edge, risk data is determined from the associated defect data; wherein the defect detection results include the confidence level of the detection results and the defect size.

[0043] Specifically, after completing inference, the edge defect detection model outputs the defect detection result. This result includes several quantitative indicators, at least the confidence level of the detection result and the defect size. The confidence level of the detection result refers to the reliability of the model's detection conclusion, usually expressed as a probability value; the defect size refers to the size of the detected defect region in the image, usually expressed as pixel area, bounding box length and width, etc.

[0044] Specifically, corresponding judgment thresholds are set for probability values ​​and defect length and width values, and data whose index values ​​fall within the area near the judgment threshold (such as within the range of 20% above and below the threshold) are judged as risk data in the suspected range.

[0045] Understandably, samples with probability values ​​close to the corresponding judgment threshold are not well understood by the model and have high uncertainty; samples with length and width values ​​close to the corresponding judgment threshold are in a critical state between qualified and unqualified and need further confirmation.

[0046] Through the above methods, the edge device ultimately filters out the risk data, which is then uploaded to the cloud for further, more accurate defect identification.

[0047] Based on the above embodiments, as an optional embodiment, the defect data acquisition and reliable transmission method for industrial inspection provided by the present invention further includes, before analyzing and processing each associated defect data acquired at the device end at the edge end to identify risk data: (1) Associate the original defect data collected by the equipment with the corresponding process production parameters to form associated defect data.

[0048] By associating raw defect data collected from the equipment with the corresponding process parameters, each defect record not only includes the image and label information of the defect itself, but also the production environment parameters of the product at the corresponding process stage. This association provides a data foundation for subsequent process parameter optimization.

[0049] (2) Determine the transmission priority of the associated defect data based on the data attributes of the associated defect data.

[0050] This step extracts the data attributes of the associated defect data, which can include multiple dimensions such as defect type, defect location, and product batch. Based on these data attributes, a priority determination mechanism (e.g., through matching mapping) is used to assign a corresponding priority to each piece of associated defect data. This priority determination enables differentiated transmission scheduling during subsequent data uploads based on the importance and urgency of the data, ensuring that high-value, time-sensitive critical data is transmitted to the cloud reliably and prioritized, thereby improving the overall efficiency of cloud-edge collaboration.

[0051] Based on the above embodiments, as an optional embodiment, the defect data acquisition and reliable transmission method for industrial inspection provided by the present invention determines the transmission priority of the associated defect data according to the data attributes of the associated defect data, including: (1) Obtain the data attributes of the associated defect data; Specifically, the edge processing unit extracts multi-dimensional data attributes from the associated defect data, which serve as input features for subsequent priority calculations. These data attributes include, but are not limited to: Data source type: Used to identify the type of testing or process equipment from which the data originates; Defect type describes the category to which the detected defect belongs, such as Mura defect, point defect, line defect, etc. Different types of defects have different degrees of impact on product yield; Business attributes are used to record business information related to products and customers, such as product model, customer level, delivery batch, etc.

[0052] (2) Input the data attributes into a pre-built regression model and output the predicted loss amount; wherein the regression model is pre-trained based on the data attribute samples and the corresponding loss amount samples.

[0053] As an optional implementation, the regression model can employ a Gradient Boosting Decision Tree (GBDT) model. GBDT is a regression algorithm based on ensemble learning that iteratively constructs multiple decision trees, with each new tree fitting the residuals of the preceding model, thereby gradually improving prediction accuracy.

[0054] As an alternative implementation, the regression model can also employ other existing regression models, such as random forest regression models, XGBoost regression models, support vector regression models, or deep neural network regression models, as long as they can output corresponding predicted loss amounts based on the input data attributes. The improvement of this invention in the regression model section lies not in the selection of a specific model structure, but in using the predicted loss amount as a quantitative basis for determining transmission priority, thus constructing a complete decision-making chain from "data attributes, potential loss, and transmission priority."

[0055] In terms of model training, the regression model is pre-trained using historically accumulated data attribute samples and corresponding actual loss amount samples to obtain the pre-constructed regression model. The training objective is: given a set of data attributes, the model can accurately predict the potential loss amount that this type of defect may cause (with an error less than a certain value). The actual loss amount can be calculated based on the enterprise's historical production data, including rework costs, material losses, and delivery delay penalties caused by this type of defect.

[0056] Regarding model updates, a dynamic update mechanism can be adopted. Specifically, the model is incrementally learned and fine-tuned using the latest collected production data at regular intervals (e.g., hourly), enabling the model to continuously adapt to changes in the production environment, such as the introduction of new product models, adjustments to process parameters, and the evolution of defect distribution patterns. This dynamic update mechanism ensures the accuracy and timeliness of loss amount prediction.

[0057] (3) Determine the transmission priority of the associated defect data based on the predicted loss amount; wherein the transmission priority of the associated defect data is positively correlated with the predicted loss amount.

[0058] Optionally, to achieve a quantitative representation of transmission priority, the predicted value can be mapped to a standardized range. The mapping method should satisfy the requirement that the transmission priority of the associated defect data is positively correlated with the predicted loss amount.

[0059] This invention enables the edge processing unit to assign an objective and dynamic transmission priority to each associated defect data point based on a data-driven quantification model, providing a basis for subsequent differentiated transmission scheduling. Compared to traditional fixed priority or manual labeling methods, this method not only considers static attributes such as data type, but also comprehensively considers factors such as the potential economic losses of defects, historical statistical patterns, and the current system environment, making priority determination more accurate and adaptable to actual production.

[0060] Based on the above embodiments, as an optional embodiment, the defect data acquisition and reliable transmission method for industrial inspection provided by the present invention further includes: (1) Determine the quantified value of the transmission priority of the associated defect data; As an optional mapping method, the predicted loss amount is compressed to the [0, 1] interval using the Sigmoid function to obtain the quantized value of the transmission priority. a The calculation formula is as follows: a = 0.5 + 0.5 × sigmoid(predicted_loss / max_loss) Where predicted_loss is the predicted loss amount output by the regression model, max_loss is the preset maximum loss threshold (e.g., 100,000 yuan), and sigmoid is the S-shaped growth curve function.

[0061] (2) When the quantized value is greater than or equal to the preset threshold, a dual-channel redundant transmission mode is used for transmission; wherein, the dual channels include a timeliness channel and a reliability channel; It should be noted that the dual-channel transmission mechanism in this invention is not limited to the transmission link from the device to the edge, but also includes the transmission link from the edge to the cloud. In the scenario of this invention, data collected by the device is first transmitted to the edge for preprocessing and risk identification. The risk data filtered out by the edge needs to be further uploaded to the cloud for precise analysis. Both of these links can use the dual-channel transmission mechanism provided by this invention.

[0062] Specifically, the preset threshold is a priority judgment threshold set according to actual business needs. When the quantified value of the associated defect data is greater than or equal to the preset threshold, it indicates that the data has extremely high business value or extremely high timeliness requirements, and is considered critical data. Its loss or delay may cause significant economic losses. In this case, a dual-channel redundant transmission mode is used for transmission.

[0063] The dual-channel system comprises two logical channels with distinct functional orientations: a timeliness channel and a reliability channel. The timeliness channel prioritizes low-latency transmission and is suitable for business data sensitive to transmission latency. It typically employs the UDP protocol combined with zero-copy technology to reduce the overhead of copying data between user space and kernel space, aiming for maximum transmission speed. The reliability channel prioritizes highly reliable transmission and is suitable for business data with strict data integrity requirements. It typically employs a transmission protocol with forward error correction coding (such as RS-FEC coding). By sending redundant checksum data, the receiving end can automatically recover the original data in the event of partial packet loss without retransmission, thus ensuring reliability while avoiding the introduction of additional latency.

[0064] (3) When the quantified value is less than the preset threshold, the relative magnitude of the timeliness requirement and reliability requirement of the associated defect data is determined according to the data attributes of the associated defect data, so as to select a corresponding channel from the dual channels for data transmission.

[0065] Specifically, when the quantified value of the associated defect data is less than a preset threshold, it indicates that although the data is not the highest priority extremely critical data, it still needs to be uploaded to the cloud for analysis. In this case, there is no need to enable redundant transmission; instead, the most suitable single channel is selected for transmission based on the characteristics of the data itself to save network resources.

[0066] This invention can determine defects based on preset features in the data attributes of associated defective data. These preset features can be set, such as determining the corresponding channel based on the data source type.

[0067] Based on the above embodiments, as an optional embodiment, the defect data acquisition and reliable transmission method for industrial inspection provided by the present invention performs defect identification on the defect images in each risk data in the cloud to filter out valid defect data from the risk data, including: (1) Multiple heterogeneous defect detection models are used to identify the defect image. If the detection results of multiple heterogeneous defect detection models are consistent, the detection results of multiple heterogeneous defect detection models are used as the final detection results to screen out valid defect data.

[0068] Specifically, multiple heterogeneous defect detection models are deployed in the cloud to perform parallel recognition of the same defect image. It should be noted that these multiple heterogeneous defect detection models can all adopt existing, publicly available model architectures; this invention does not limit the specific model type and structure. For example, YOLO series models based on convolutional neural networks, DETR models based on the Transformer architecture, and detection models combining traditional image processing and machine learning can be used, as long as the models differ in architecture, training data, hyperparameters, or post-processing strategies, they can be considered heterogeneous.

[0069] Multiple heterogeneous models perform inference on the input defect image, outputting their respective candidate labels and confidence scores. When the detection results of multiple models are consistent, two situations need to be further distinguished: If multiple models consistently determine that a defect exists (e.g., they all determine it to be the same defect type, and the confidence scores all exceed the corresponding thresholds), it indicates that the identification result has high credibility, and the consistent conclusion can be taken as the final result, and the data can be included in the valid defect data set; if multiple models consistently determine that there is no defect, it indicates that the risk data uploaded by the edge device is actually a normal product and belongs to the edge device's false alarm, so it is excluded from the risk data and not included in the valid defect data set.

[0070] (2) When the detection results of multiple heterogeneous defect detection models are inconsistent, the defect image is identified by using a preset defect detection model. When the detection result of the preset defect detection model is consistent with the original defect label, the detection result of the preset defect detection model is used as the final detection result to filter out valid defect data.

[0071] Specifically, when the detection results of multiple heterogeneous models differ (e.g., different models output different defect types, or some models classify a sample as defective while others classify it as normal), it indicates that there is some controversy surrounding the sample, and the output of one model cannot be directly trusted. In this case, a pre-set defect detection model (i.e., a verification model) is introduced to re-identify the controversial sample.

[0072] It should be noted that the preset defect detection model can also adopt existing, publicly available model architectures, and this invention does not limit it. This preset defect detection model is typically a model with a larger number of parameters and higher accuracy (such as a larger-scale deep neural network model or a basic model), which has stronger detection capabilities but also higher computational overhead. Therefore, it is only called on a limited number of disputed samples to balance efficiency and accuracy. The preset defect detection model outputs its judgment result and confidence level.

[0073] The original defect label refers to the preliminary judgment result (such as OK / NG mark or specific defect type) generated by the detection equipment's own algorithm when collecting data. The detection result of the preset model is compared with the original defect label: if they are consistent, it means that both the preset model and the original detection equipment agree with the conclusion, and the credibility is high. In this case, the detection result of the preset model is taken as the final result and included in the valid defect data set; if they are consistent and judged as defect-free, the data is excluded from the risk data.

[0074] (3) When the detection results of the preset defect detection model are inconsistent with the original defect labels, the effective defect data are screened out by manual review.

[0075] Specifically, when the detection results of the preset defect detection model are still inconsistent with the original defect labels, it indicates that the sample is a highly controversial sample, and automated methods are unable to provide a definitive conclusion. In this case, the data is marked as a sample requiring manual verification and handed over to professional labelers for manual review.

[0076] Based on the above embodiments, as an optional embodiment, the defect data acquisition and reliable transmission method for industrial inspection provided by the present invention performs data analysis on the effective defect data set and optimizes pre-associated process production parameters, including: (1) Obtain OK product data and the corresponding process production parameters of the OK product, and construct OK product associated data set.

[0077] Specifically, data analysis not only relies on the set of valid defect data filtered from risk data, but also requires the simultaneous acquisition of data from inspected and qualified products (i.e., OK product data) as a benchmark. OK product data refers to normal product data that has been confirmed to be free of defects or with defects within acceptable limits, including its corresponding defect inspection records and results. Simultaneously, the process parameters for these OK products at the corresponding manufacturing stages are acquired, such as temperature, pressure, speed, raw material ratio, and process time. The OK product data is then correlated with the corresponding process parameters to construct an OK product correlated data set.

[0078] (2) Based on the OK product association data set and the valid defect data set, mine the correlation patterns between process parameters and defects, and form optimized process production parameters for different defects; optionally, including: 1) Compare and analyze the value distribution of defective product samples and OK product samples on various process parameters to analyze the significance of the differences and determine the key influencing parameters associated with specific defect types; Comparative analysis methods may include, but are not limited to: calculating the means of the two types of samples on a certain parameter and comparing the differences in their central positions; calculating the variance and comparing the differences in their dispersion; and calculating quantiles (such as the median, upper and lower quartiles, etc.) and comparing the differences in their distribution patterns. When there are significant differences in the above statistics of the two types of samples on a certain parameter, it indicates that the parameter has different distribution characteristics between defective products and OK products, and that the parameter may be associated with the occurrence of a specific defect type. The more significant the difference, the stronger the association.

[0079] After completing the significance analysis of the differences in various process parameters, the parameters are sorted according to the degree of significance of the differences, and one or more process parameters with the largest significance of differences are identified as key influencing parameters associated with specific defect types.

[0080] 2) Based on the parameter distribution of normal products in the OK product association data set, determine the optimal value range of the key influencing parameters; Specifically, for the identified key influencing parameters, the value distribution of OK product samples on these parameters is obtained. Based on this distribution, the optimal value range is determined. For example, the concentrated distribution range of OK product samples on these parameters is statistically analyzed, and the interval with the highest product pass rate is determined as the optimal value range.

[0081] 3) Adjust the current value of the key influencing parameter to the optimal value range to form the optimized process production parameters.

[0082] Specifically, by comparing the value distribution of the key influencing parameter in the defective product sample with the determined optimal value range, the adjustment direction and magnitude are determined. If the parameter distribution of the defective product sample shifts to a higher value relative to the optimal value range, the target value of the parameter is adjusted to a lower value; if it shifts to a lower value, it is adjusted to a higher value. This ultimately results in optimized process production parameters that include information such as parameter name, current value range, target value range, and adjustment direction.

[0083] This invention identifies the key influencing parameters most strongly associated with specific defect types by comparing the value distribution differences of defective product samples and acceptable product samples across various process parameters. Based on the parameter distribution of the acceptable product samples, it determines the optimal value range and adjusts the key influencing parameters to this optimal range. This method can objectively and quantitatively locate the key process parameters leading to defects, avoiding the blindness of relying solely on experience, improving the scientific rigor and accuracy of process parameter optimization, and thus more effectively reducing product defect rates and increasing yield.

[0084] (3) Update the optimized process production parameters to the corresponding production equipment for production.

[0085] The generated optimized process parameters are then updated to the corresponding production equipment, including manufacturing equipment, testing equipment, and repair equipment, through the production execution system or equipment control system. After the update is complete, the production equipment is put into the next round of production according to the new process parameters.

[0086] Figure 2 This is a schematic diagram of the defect data acquisition and reliable transmission system for industrial inspection provided by the present invention, as shown below. Figure 2 As shown, the system includes: The first processing module 210 is used to analyze and process each associated defect data collected by the device at the edge end to identify risk data; wherein each associated defect data includes original defect data and corresponding process parameter data, and the original defect data includes at least a defect image and a corresponding original defect label; The second processing module 220 is used to upload each identified risk data to the cloud, and to perform defect identification on the defect image in each risk data in the cloud, so as to filter out the valid defect data from the risk data to form a set of valid defect data. The third processing module 230 is used to perform data analysis on the effective defect data set and optimize the pre-associated process production parameters.

[0087] It should be noted that the defect data acquisition and reliable transmission system for industrial inspection provided in this embodiment of the invention can execute the defect data acquisition and reliable transmission method for industrial inspection described in any of the above embodiments during specific operation, which will not be elaborated in this embodiment.

[0088] Figure 3 This is a schematic diagram of the structure of the electronic device provided by the present invention, such as... Figure 3As shown, the electronic device may include a processor 310, a communications interface 320, a memory 330, and a communication bus 340. The processor 310, communications interface 320, and memory 330 communicate with each other via the communication bus 340. The processor 310 can call logical instructions from the memory 330 to execute a defect data acquisition and reliable transmission method for industrial inspection.

[0089] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0090] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for defect data acquisition and reliable transmission for industrial inspection, characterized in that, include: At the edge, each associated defect data collected by the equipment is analyzed and processed to identify risk data; wherein, each associated defect data includes original defect data and corresponding process parameter data, and the original defect data includes at least a defect image and a corresponding original defect label; Each identified risk data is uploaded to the cloud, where defect images in each risk data are identified to filter out valid defect data and form a set of valid defect data. Data analysis is performed on the effective defect data set to optimize the pre-associated process production parameters.

2. The defect data acquisition and reliable transmission method for industrial inspection according to claim 1, characterized in that, At the edge, each associated defect data collected from the device is analyzed and processed to identify risk data, including: Defect detection is performed on the defect image in each associated defect data at the edge. Based on the defect detection results at the edge, risk data is determined from the associated defect data; The defect detection results include the confidence level of the detection results and the size of the defect.

3. The defect data acquisition and reliable transmission method for industrial inspection according to claim 1, characterized in that, Before analyzing and processing each associated defect data collected from the device at the edge to identify risk data, the process also includes: The raw defect data collected by the equipment is correlated with the corresponding process production parameters to form correlated defect data. The transmission priority of the associated defect data is determined based on its data attributes.

4. The defect data acquisition and reliable transmission method for industrial inspection according to claim 3, characterized in that, Determining the transmission priority of the associated defect data based on its data attributes includes: Obtain the data attributes of the associated defect data; The data attributes are input into a pre-built regression model, which outputs a predicted loss amount; wherein the regression model is pre-trained based on data attribute samples and corresponding loss amount samples. Based on the predicted loss amount, the transmission priority of the associated defect data is determined; wherein, the transmission priority of the associated defect data is positively correlated with the predicted loss amount.

5. The defect data acquisition and reliable transmission method for industrial inspection according to claim 4, characterized in that, Also includes: A quantified value is used to determine the transmission priority of the associated defect data; When the quantized value is greater than or equal to a preset threshold, a dual-channel redundant transmission mode is used for transmission; wherein, the dual channels include a timeliness channel and a reliability channel; If the quantified value is less than a preset threshold, the relative magnitudes of the timeliness and reliability requirements of the associated defect data are determined based on the data attributes of the associated defect data, so as to select a corresponding channel from the dual channels for data transmission.

6. The defect data acquisition and reliable transmission method for industrial inspection according to claim 1, characterized in that, The cloud platform performs defect identification on the defect images in each risk data set to filter out valid defect data from the risk data, including: Multiple heterogeneous defect detection models are used to identify the defect images. If the detection results of the multiple heterogeneous defect detection models are consistent, the detection results of the multiple heterogeneous defect detection models are used as the final detection result to filter out valid defect data. When the detection results of multiple heterogeneous defect detection models are inconsistent, a preset defect detection model is used to identify defects in the defect image. If the detection result of the preset defect detection model is consistent with the original defect label, the detection result of the preset defect detection model is taken as the final detection result to filter out valid defect data. When the detection results of the preset defect detection model are inconsistent with the original defect labels, a manual review method is used to screen out the valid defect data.

7. The defect data acquisition and reliable transmission method for industrial inspection according to claim 1, characterized in that, Data analysis is performed on the aforementioned valid defect data set, and pre-associated process production parameters are optimized, including: Obtain OK product data and corresponding process production parameters for the OK products, and construct an OK product associated data set; Based on the OK product association data set and the valid defect data set, the correlation patterns between process parameters and defects are mined, and optimized process production parameters are formed for different defects. The optimized process parameters are then updated in the corresponding production equipment for production.

8. The defect data acquisition and reliable transmission method for industrial inspection according to claim 7, characterized in that, Based on the OK product association data set and the valid defect data set, the correlation patterns between process parameters and defects are mined, and optimized process production parameters are formed for different defects, including: A comparative analysis was conducted on the value distribution of defective product samples and OK product samples across various process parameters to analyze the significance of the differences and identify the key influencing parameters associated with specific defect types. Based on the parameter distribution of normal products in the OK product association data set, the optimal value range of the key influencing parameters is determined; The current values ​​of the key influencing parameters are adjusted to the optimal value range to form the optimized process production parameters.

9. A defect data acquisition and reliable transmission system for industrial inspection, characterized in that, include: The first processing module is used to analyze and process each associated defect data collected by the device at the edge to identify risk data; wherein each associated defect data includes original defect data and corresponding process parameter data, and the original defect data includes at least a defect image and a corresponding original defect label; The second processing module is used to upload each identified risk data to the cloud, and to perform defect identification on the defect image in each risk data in the cloud, so as to filter out the valid defect data from the risk data to form a set of valid defect data. The third processing module is used to perform data analysis on the effective defect data set and optimize the pre-associated process production parameters.

10. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the defect data acquisition and reliable transmission method for industrial inspection as described in any one of claims 1 to 8.

Images