Product design closed-loop optimization method and system in combination with defect detection feedback
By building a closed-loop feedback mechanism during the product design phase and using deep learning models to identify and optimize design features that are prone to false detection and missed detection, the problem of lack of closed-loop feedback in existing technologies is solved, achieving high-precision detection and improved production efficiency.
Patent Information
- Application Number
- CN202510794362.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-06-13
- Publication Date
- 2025-09-26
AI Technical Summary
The existing defect detection system lacks an effective closed-loop feedback mechanism between the product design process, resulting in frequent false detections and missed detections in deep learning detection models during production, affecting the accuracy of quality control and production efficiency.
By building defect detection datasets, training deep learning models, analyzing detection results, identifying design features that lead to false detections and missed detections, and optimizing them during the design phase, a feedback mechanism is established, including data collection, defect detection, data analysis, auxiliary design model training, and design iteration, to optimize product design to reduce false detections and missed detections.
It improves detection accuracy, reduces false detection and missed detection, shortens design and production cycles, reduces production costs, and improves product consistency and quality.
Smart Images
Figure CN120706236A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to product quality control and design optimization. It utilizes deep learning and computer vision technologies to detect product defects. By analyzing the correlation between detection results and product design features, a feedback mechanism is established to optimize product design. This invention has broad application in the manufacturing of flat materials (such as PVC sheets, metal sheets, and glass sheets), textiles (such as fabrics and carpets), electronics (such as printed circuit boards and display screens), automotive parts (such as body panels), and packaging materials. Background Art
[0002] In modern manufacturing, product design and production quality control are often independent processes. While artificial intelligence and deep learning technologies, particularly object detection algorithms, have made significant progress in identifying various product defects during the production process, existing defect detection systems often lack an effective, closed-loop feedback mechanism with the product design process. This separation prevents the system from fully utilizing actual inspection data to guide and optimize the initial product design. As a result, certain product designs may contain features that are visually similar to common production defects (such as specific textures, patterns, or color combinations). This leads to frequent false positives (identifying normal features as defects) and missed detections (failing to identify real defects) in deep learning inspection models during subsequent production. This not only affects the accuracy of quality control but also reduces overall production efficiency. Summary of the Invention
[0003] In view of this, the present invention proposes a closed-loop optimization method for product design combined with defect detection feedback, comprising the following steps: Data collection: Collect product images and annotate various types of defects to build a defect detection dataset; Defect detection model training: training a first deep learning model on the defect detection dataset; Data analysis: Use the defect detection model to inspect product images, analyze the inspection results, identify the types of defects that are frequently falsely detected and missed, and analyze the specific design features that lead to these false detections and missed detections by combining them with product design drawings or texture files; Training an assisted design model: Based on the results of the data analysis, the design features that cause false detection or missed detection are annotated on the product design drawings or texture files to construct an assisted design dataset. A second deep learning model is trained on the assisted design dataset to obtain an assisted design model capable of identifying falsely detected or missed design features. Design iteration: Based on the recognition results of the auxiliary design model or the statistical results of the data analysis, design features in the design drawings or texture files that are prone to false detection or missed detection are modified or optimized through design software (such as CAD tools).
[0004] Specifically, in the data collection, a high-resolution industrial camera is used to capture product surface images, and labeling tools such as LabelImg or RectLabel are used to label common defects.
[0005] Through this technical solution, high-resolution industrial cameras are used to capture images of product surfaces, ensuring that even the smallest defects can be captured. The cameras must be able to adapt to varying lighting conditions in diverse production environments and be equipped with an automatically adjustable lighting system to minimize the impact of shadows and reflections on image quality.
[0006] Specifically, the defect detection model is trained using the YOLO target detection framework: YOLOv5 or YOLOv8. The model can identify common defect types (such as bubbles, stains, scratches, color difference, etc.), providing a basis for subsequent optimization.
[0007] The training method is specifically as follows: Divide the labeled dataset into a training set and a validation set; By setting hyperparameters during training, the ability to detect small defects can be improved; When the model performs stably on the validation set and the indicators meet expectations, the final model weights are output.
[0008] Specifically, the data analysis is specifically false detection and missed detection analysis, including: Use the trained defect detection model to inspect product images and record the types and locations of detected defects; Statistically analyze false positives and missed positives, focusing on the types of false positives and missed positives that occur frequently; The false detection is misjudging a normal design feature as a defect; The missed detection refers to a real defect that cannot be identified; In combination with product design drawings or texture files, design features associated with the high-frequency false detections or missed detections are analyzed to determine the design reasons that lead to the false detections or missed detections.
[0009] Specifically, the auxiliary design model training includes: Based on the design features that are prone to false detection or missed detection identified by the data analysis, the areas of these design features are marked with bounding boxes on the corresponding product design drawings or texture files, and labels indicating their confusability are assigned to form the auxiliary design dataset; The YOLO series target detection algorithm is used to perform training or fine-tuning on the auxiliary design dataset to obtain the auxiliary design model.
[0010] Specifically, the design iteration includes a manual design feedback mechanism: Based on the statistical results of this data analysis, designers examine new design drawings or texture files within the design software to identify known design features that could easily lead to misidentification. If such features are detected, designers manually modify or remove them, for example by adjusting the texture's shape, spacing, or color. For example, if a striped texture closely resembles a scratch, the texture's shape or spacing can be adjusted during the design phase, reducing the risk of misidentification later.
[0011] Specifically, the design iteration includes an automated design feedback mechanism: The auxiliary design model directly analyzes the design drawings or texture files, automatically identifies and marks design features that are prone to false detection or missed detection, and generates modification suggestions; the identification results or modification suggestions are fed back to the designer, or directly applied in the automated design process.
[0012] Secondly, a closed-loop product design optimization system combining defect detection feedback is provided, including: Data collection module: used to collect product images, annotate various actual product defects in the images, and build a defect detection dataset; A defect detection model training module is configured to train a first deep learning model on the defect detection dataset to obtain a defect detection model capable of identifying defects in the actual product; Data analysis module: used to use the defect detection model to detect product images, analyze the detection results, identify the types of defects that are frequently falsely detected and missed, and analyze the specific design features that cause the false detections and missed detections in combination with product design drawings or texture files; An auxiliary design model training module is configured to annotate the design features that cause false detection or missed detection on the product design drawings or texture files based on the results of the data analysis, construct an auxiliary design dataset, and train a second deep learning model on the auxiliary design dataset to obtain an auxiliary design model capable of identifying design features in the design drawings or texture files that are prone to false detection or missed detection; Design iteration module: used to modify or optimize the design features in the design drawings or texture files that are prone to false detection or missed detection through design software based on the recognition results of the auxiliary design model or the statistical results of the data analysis.
[0013] Beneficial effects of the present invention: 1. Improve detection accuracy: By optimizing key features in product design that are prone to false positives or missed detections, the accuracy of deep learning detection models can be improved, reducing false positives and missed detections during the production process.
[0014] 2. Automated Design Feedback: CAD models enable automated analysis of drawings during the design phase, reducing manual troubleshooting and improving overall design efficiency. Real-time feedback allows for early detection of potential issues, avoiding rework later in the process.
[0015] 3. Shortened design and production cycles: Since design features that may lead to defects can be corrected in a timely manner in the early stages of design, trial and error costs are greatly reduced, the product launch process is accelerated, and the overall iteration cycle is shortened.
[0016] 4. Improve product consistency and quality: Through continuous design optimization and accurate defect detection, quality issues caused by design defects are minimized, thereby ensuring product consistency and reliability in large-scale production.
[0017] 5. Reduce production costs: Reduce quality issues such as false detection, missed detection and rework, which can significantly reduce production costs and improve production efficiency and economic benefits. BRIEF DESCRIPTION OF THE DRAWINGS
[0018] The present invention has the following accompanying drawings: Figure 1 Flowchart of the method of the present invention. DETAILED DESCRIPTION
[0019] To make the objects, advantages, and features of the present invention more apparent, the present invention is further described in detail below using deep learning defect detection and design optimization in PVC board production as an example.
[0020] The following uses the production process of PVC boards as an example to explain in detail how to improve product quality and production efficiency through deep learning defect detection technology and design optimization feedback mechanisms.
[0021] S1: Data collection and image acquisition.
[0022] S11: Equipment preparation.
[0023] Select a high-resolution industrial camera (resolution not less than 4K) and install it above the PVC board production line so that it can cover the entire surface area of the PVC boards.
[0024] Configure an automatically adjustable lighting system (such as a combination of LED flat light source and ring light source) to ensure uniform lighting and minimize the impact of reflections and shadows on image quality.
[0025] S12: Image acquisition.
[0026] During the production process, every time a batch of PVC boards completes a key process, the system automatically triggers the industrial camera to capture images.
[0027] To reduce errors caused by changes in shooting angle or distance, try to maintain the same shooting position and angle.
[0028] The collected images are stored in a data management system, and various defects in the images (such as bubbles, scratches, stains, color differences, etc.) are annotated to build a complete data set.
[0029] S2: Defect detection model training.
[0030] S21: Data preprocessing and annotation.
[0031] Perform basic cleaning and screening on the collected images to eliminate images that are severely blurred or do not meet the shooting specifications.
[0032] Use annotation tools (such as LabelImg or RectLabel or others) to mark the defect location and type to ensure consistency and accuracy.
[0033] Perform appropriate data augmentation (such as rotation, flipping, scaling, color change, noise addition, etc.) to improve the robustness of the model.
[0034] S22: Train a deep learning defect detection model.
[0035] Use the YOLO target detection framework and divide the labeled dataset into a training set and a validation set (for example, 8:2).
[0036] During training, by reasonably setting hyperparameters such as learning rate, batch size, and number of iterations, we focus on improving the ability to detect small defects (such as minor scratches and bubbles).
[0037] When the model performs stably on the validation set and indicators such as accuracy and recall meet expectations, the final defect detection model weights are output.
[0038] S3: Data analysis.
[0039] S31: Statistics of test results.
[0040] Using the trained defect detection model, a large number of PVC board images are inspected, and the detection results of each image are recorded, including the defect type and location.
[0041] Statistically analyze false detections (normal textures are misidentified as defects) and missed detections (failure to identify real defects), paying particular attention to high-frequency and serious false detections or missed detections.
[0042] S32: Design feature association.
[0043] Combined with the design drawings or texture files of the PVC board, analyze the design features that may lead to false detection and missed detection.
[0044] If it is found that certain dot-shaped or strip-shaped textures are very similar to the appearance of defects, or that certain color combinations are easily confused with stains, these design features need to be recorded to provide a basis for subsequent auxiliary design model training and design iteration.
[0045] S4: Assisted design model training.
[0046] S41: Assisted design dataset construction.
[0047] Key textures or patterns that are easily misjudged or missed during the "data analysis" stage need to be "targetedly marked" again at the design drawing level.
[0048] Specific approach: In the PVC board texture or CAD drawing, mark the areas of these patterns / textures with bounding boxes and give them corresponding labels (such as "dot texture that is easily confused with bubbles" or "strip texture that is easily confused with scratches").
[0049] This forms an "auxiliary design dataset", which is different from the defect detection dataset and is mainly aimed at "identification of design features" rather than "defect detection".
[0050] S42: Assisted design model training.
[0051] Use the same or similar deep learning framework as YOLO target detection and train it on the "Assisted Design Dataset".
[0052] The goal of the model is to identify features in drawings or texture files that could lead to errors in later detection (e.g., streaks that are too similar to scratch shapes).
[0053] After training, the model is able to identify "confusing" textures during the design phase, providing warnings and suggestions to designers and automated optimization systems.
[0054] S5: Design iteration and optimization.
[0055] S51: Artificially design feedback mechanism.
[0056] The types of false positives that may be caused by manual inspection by designers in CAD software or texture design tools.
[0057] Designers artificially modify or remove these textures, for example, by changing the direction, width, or color depth of the stripes to reduce their resemblance to real scratches.
[0058] S52: Automated design feedback mechanism.
[0059] In a higher-level automated production design environment, the auxiliary design model can directly scan the design drawings in batches or in real time and automatically provide modification suggestions.
[0060] When certain textures are found to be similar in appearance to common defects, the system can automatically mark and highlight them on the drawing.
[0061] Designers make design modifications based on model prompts, significantly shortening the design iteration cycle.
[0062] S6: Final effect evaluation and production implementation.
[0063] S61: Production verification.
[0064] When the design adjustment is completed, a batch of PVC board samples are produced according to the new drawings, and their surfaces are inspected using a defect detection system to calculate the false detection and missed detection rates.
[0065] Compare the test data before and after optimization. If the improvement is significant, the new design can be applied to actual production in batches.
[0066] S62: Production line deployment.
[0067] Deploy defect detection models and auxiliary design models simultaneously on the production line: Online defect detection: Real-time monitoring of defects in the PVC board production process.
[0068] Design optimization feedback: When it is detected that a certain type of defect occurs frequently or certain patterns are easily misjudged, it automatically triggers inspection and suggestions on the design drawings.
[0069] Through this closed-loop mechanism, product design is continuously optimized to improve product quality and reduce production costs.
[0070] Based on the same design concept, the product design closed-loop optimization system combined with defect detection feedback of the present invention includes: Data collection module: used to collect product images, annotate various actual product defects in the images, and build a defect detection dataset; A defect detection model training module is configured to train a first deep learning model on the defect detection dataset to obtain a defect detection model capable of identifying defects in the actual product; Data analysis module: used to use the defect detection model to detect product images, analyze the detection results, identify the types of defects that are frequently falsely detected and missed, and analyze the specific design features that cause the false detections and missed detections in combination with product design drawings or texture files; An auxiliary design model training module is configured to annotate the design features that cause false detection or missed detection on the product design drawings or texture files based on the results of the data analysis, construct an auxiliary design dataset, and train a second deep learning model on the auxiliary design dataset to obtain an auxiliary design model capable of identifying design features in the design drawings or texture files that are prone to false detection or missed detection; Design iteration module: used to modify or optimize the design features in the design drawings or texture files that are prone to false detection or missed detection through design software based on the recognition results of the auxiliary design model or the statistical results of the data analysis.
[0071] It should be pointed out that this technology can be widely used in the following fields: manufacturing of flat materials: such as PVC boards, metal boards, glass boards, etc.; textile production: such as fabrics, carpets, non-woven fabrics, etc.; electronic product manufacturing: such as printed circuit boards (PCBs), display screens, etc.; automotive parts production: such as body panels, interior trims, etc.; packaging material production: such as cartons, plastic films, etc.
[0072] It should be noted that any process or method description in the embodiments may be understood as representing a module, fragment or portion of code comprising one or more executable instructions for implementing steps of a specific logical function or process, and that the scope of the preferred embodiments of the present invention includes alternative implementations in which functions may not be performed in the order shown or discussed, including performing functions in a substantially simultaneous manner or in the reverse order depending on the functions involved, which should be understood by those skilled in the art to which the embodiments of the present invention belong.
[0073] It should be noted that the logic and / or steps described in the embodiments, for example, can be considered as a sequenced list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by an instruction execution system, device, or apparatus (e.g., a computer-based system, a system including a processor, or other system that can fetch and execute instructions from an instruction execution system, device, or apparatus), or in conjunction with such an instruction execution system, device, or apparatus. For purposes of this specification, a "computer-readable medium" can be any device that can contain, store, communicate, propagate, or transmit a program for use by an instruction execution system, device, or apparatus, or in conjunction with such an instruction execution system, device, or apparatus. More specific examples (non-exhaustive list) of computer-readable media include the following: an electrical connection portion having one or more wires (electronic device), a portable computer disk cartridge (magnetic device), random access memory (RAM), read-only memory (ROM), erasable and programmable read-only memory (EPROM or flash memory), fiber optic devices, and a portable compact disc read-only memory (CDROM). Furthermore, the computer-readable medium may even be paper or other suitable medium on which the program is printed, since the program may be obtained electronically, for example, by optically scanning the paper or other medium and then editing, interpreting or processing it in another suitable manner if necessary, and then storing it in a computer memory.
[0074] It should be understood that various parts of the present invention can be implemented using hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented using software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented using hardware, any one of the following technologies known in the art or a combination thereof can be used: a discrete logic circuit having a logic gate circuit for implementing a logic function on a data signal, an application-specific integrated circuit having a suitable combination of logic gate circuits, a programmable gate array (PGA), a field programmable gate array (FPGA), etc.
[0075] Those skilled in the art will understand that all or part of the steps of implementing the above-mentioned embodiment method can be completed by instructing related hardware through a program, and the program can be stored in a computer-readable storage medium. When the program is executed, it includes one of the steps of the method embodiment or a combination thereof.
[0076] Furthermore, the functional modules in the embodiments of the present invention may be integrated into a single processing module, each module may exist physically separately, or two or more modules may be integrated into a single module. The integrated modules may be implemented in the form of hardware or software functional modules. If the integrated modules are implemented in the form of software functional modules and sold or used as independent products, they may also be stored in a computer-readable storage medium.
[0077] The storage medium mentioned above can be a read-only memory, a magnetic disk or an optical disk, etc.
[0078] The above embodiments provide a detailed description of the technical solutions of the present invention. Obviously, the present invention is not limited to the described embodiments. Based on the embodiments of the present invention, those skilled in the art may make various modifications. However, any modifications that are equivalent to or similar to the present invention fall within the scope of protection of the present invention.
[0079] The contents not described in detail in this specification belong to the prior art known to those skilled in the art.
Claims
1. A closed-loop optimization method for product design combined with defect detection feedback, characterized in that: The steps include: Data collection: Collect product images and annotate various types of defects to build a defect detection dataset; Defect detection model training: training a first deep learning model on the defect detection dataset; Data analysis: Use the defect detection model to inspect product images, analyze the inspection results, identify the types of defects that are frequently falsely detected and missed, and analyze the specific design features that lead to these false detections and missed detections by combining them with product design drawings or texture files; Training an assisted design model: Based on the results of the data analysis, the design features that cause false detection or missed detection are annotated on the product design drawings or texture files to construct an assisted design dataset. A second deep learning model is trained on the assisted design dataset to obtain an assisted design model capable of identifying falsely detected or missed design features. Design iteration: Based on the recognition results of the auxiliary design model or the statistical results of the data analysis, the design features in the design drawings or texture files that are prone to false detection or missed detection are modified or optimized through design software.
2. The optimization method according to claim 1, wherein: The data collection uses a high-resolution industrial camera to capture product surface images, and uses a labeling tool to mark product defects.
3. The optimization method according to claim 1, wherein: The defect detection model is trained using a target detection framework, and the target detection framework is selected from the YOLO series model.
4. The optimization method according to claim 3, wherein: The defect detection model training method is specifically as follows: Divide the labeled dataset into a training set and a validation set; By setting hyperparameters during training, the ability to detect small defects can be improved; When the model performs stably on the validation set and the indicators meet expectations, the final model weights are output.
5. The optimization method according to claim 1, wherein: The data analysis is specifically false detection and missed detection analysis, including: Use the trained defect detection model to inspect product images and record the types and locations of detected defects; Statistically analyze false positives and missed positives, focusing on the types of false positives and missed positives that occur frequently; The false detection is misjudging a normal design feature as a defect; The missed detection refers to a real defect that cannot be identified; In combination with product design drawings or texture files, design features associated with the high-frequency false detections or missed detections are analyzed to determine the design reasons that lead to the false detections or missed detections.
6. The optimization method according to claim 1, wherein: The auxiliary design model training includes: Based on the design features that are prone to false detection or missed detection identified by the data analysis, the areas of these design features are marked with bounding boxes on the corresponding product design drawings or texture files, and labels indicating their confusability are assigned to form the auxiliary design dataset; The YOLO series target detection algorithm is used to perform training or fine-tuning on the auxiliary design dataset to obtain the auxiliary design model.
7. The optimization method according to claim 6, wherein: The goal of the auxiliary design model is to identify design features in design drawings or texture files that are visually similar to known product defects and may cause subsequent defect detection models to produce false detections or missed detections.
8. The optimization method according to claim 1, wherein: The design iteration specifically includes a manual design feedback mechanism, which specifically includes: based on the statistical results of the data analysis, the designer checks the new design drawings or texture files in the design software to find out whether there are known design features that are prone to misjudgment. When such design features are detected, the designer manually modifies or deletes these features, such as adjusting the shape, spacing or color of the texture.
9. The optimization method according to claim 1, wherein: The design iteration specifically includes an automated design feedback mechanism, specifically including: The auxiliary design model directly analyzes the design drawings or texture files, automatically identifies and marks design features that are prone to false detection or missed detection, and generates modification suggestions; the identification results or modification suggestions are fed back to the designer, or directly applied in the automated design process.
10. A closed-loop optimization system for product design combined with defect detection feedback, characterized in that: include: Data collection module: used to collect product images, annotate various actual product defects in the images, and build a defect detection dataset; A defect detection model training module is configured to train a first deep learning model on the defect detection dataset to obtain a defect detection model capable of identifying defects in the actual product; Data analysis module: used to use the defect detection model to detect product images, analyze the detection results, identify the types of defects that are frequently falsely detected and missed, and analyze the specific design features that cause the false detections and missed detections in combination with product design drawings or texture files; An auxiliary design model training module is configured to annotate the design features that cause false detection or missed detection on the product design drawings or texture files based on the results of the data analysis, construct an auxiliary design dataset, and train a second deep learning model on the auxiliary design dataset to obtain an auxiliary design model capable of identifying design features in the design drawings or texture files that are prone to false detection or missed detection; Design iteration module: used to modify or optimize the design features in the design drawings or texture files that are prone to false detection or missed detection through design software based on the recognition results of the auxiliary design model or the statistical results of the data analysis.