Multi-mode AI vision-based intelligent quality inspection system for laminated boards

Abstract

The invention provides a multi-mode AI vision-based intelligent quality inspection system for a laminated board, which is applied to the technical field of industrial vision inspection and comprises a hardware unit and a software unit, the hardware unit comprises a camera array, a multi-light source module, a PLC linkage module, a sorting control module and an industrial personal computer, and the software unit comprises an image preprocessing module, an AI reasoning module, a quality data management module and a process feedback module. According to the detection system, a multispectral fusion detection technology is adopted, visible light, infrared and ultraviolet multi-mode image information is integrated, detection logic is designed according to the hot-pressing characteristic of a pressed plate, fine defects such as color gradient and texture abnormity can be accurately recognized, blindness of general AI detection is avoided, and the detection efficiency is improved. The effects of comprehensively covering multiple types of defects and unifying detection and judgment standards are achieved, the problems that traditional detection is not sensitive to fine defects, the misinformation and omission ratio is high, and manual detection standards are different are solved, and a reliable detection basis is provided for product quality stability.

Description

TECHNICAL FIELD

[0001] The application relates to the technical field of industrial visual inspection, and particularly relates to a smart quality inspection system for pressure-bonded board based on multi-modal AI vision, which is suitable for production line defect detection of pressure-bonded board. BACKGROUND

[0002] The pressure-bonded board is a kind of composite board made by combining decorative materials (such as wood veneer, paper, plastic film, etc.) with base materials (such as shaving board, medium-density board, etc.). During the production process of the pressure-bonded board, its defects need to be detected. Generally, industrial visual inspection is used. Industrial visual inspection is an industrial solution for automatic quality control using machine vision technology. It replaces manual visual inspection through image acquisition and algorithm analysis to improve detection efficiency, accuracy and consistency. The core principle of industrial visual inspection is to use industrial cameras, lenses and light sources to obtain product images, and then use image processing systems to digitally analyze pixel distribution, brightness, color and other information to complete defect recognition, size measurement, positioning guidance and other tasks.

[0003] At present, the traditional pressure-bonded board quality inspection relies on manual work, which is low in efficiency, high in missed detection rate and has different standards. Traditional visual inspection is not sensitive to subtle defects such as color gradient and texture abnormalities. General AI detection is not optimized for hot-pressing characteristics, resulting in high false positive rate of detection. In addition, high-end detection equipment is expensive and has high maintenance cost. Therefore, there is a lack of intelligent solutions that can adapt to hot-pressing production scenarios, accurately detect multiple types of defects and have controllable costs, resulting in high labor costs in the quality inspection process of enterprises, difficulty in stable improvement of product qualification rate, and difficulty in optimizing subsequent production processes through the correlation of detection quality data and production process parameters, which restricts the quality improvement and efficiency increase and scale development of pressure-bonded board production. SUMMARY

[0004] The application provides a smart quality inspection system for pressure-bonded board based on multi-modal AI vision, which can accurately detect defects of pressure-bonded board during production, and clearly identify defect positions. At the same time, data feedback is used to make corresponding corrections to the parameters of subsequent production processes, reducing the output of subsequent pressure-bonded board defective products.

[0005] To achieve the above-mentioned purpose, the application adopts the following technical solutions: The application provides a smart quality inspection system for pressure-bonded board based on multi-modal AI vision, which is characterized in that the quality inspection system comprises a hardware unit and a software unit. The hardware unit comprises a camera array, a multi-light source module, a PLC linkage module, a sorting control module and an industrial computer. The camera array comprises a downward-facing camera, a side-view camera and an infrared camera, which are used for multi-angle shooting detection of the pressure-bonded board. The multi-light source module is configured to provide stable light for the detection process. The PLC linkage module is configured to realize linkage control of the quality inspection system and the production line equipment. The sorting control module is configured to sort qualified and unqualified plates according to the detection results. The industrial computer is configured to carry the software unit running and the data interaction and control instruction issuing of the hardware unit. The software unit includes an image preprocessing module, an AI inference module, a quality data management module, and a process feedback module. The image preprocessing module is configured to perform noise elimination and feature enhancement processing on the collected images. The AI inference module integrates a multi-spectrum fusion detection module and a hot-pressing process detection module. The multi-spectrum fusion detection module is configured to receive visible light images, infrared images, and ultraviolet images, and output defect scores through multi-feature fusion decision. The hot-pressing process detection module is configured to predict quality changes in the hot-pressing process based on an LSTM network, adjust detection thresholds according to temperature and pressure, and learn quality characteristics of material formulations. The quality data management module is configured to realize storage, summarization, and deep analysis of quality data, and establish an association between quality data and process parameters. The process feedback module is configured to feed back detection results to the production line to provide data for process parameter adjustment.

[0006] In some possible implementations, the overhead camera is an industrial camera with a frame rate ≥ 30 fps, and is installed at a height of 1.2 meters from the surface of the pressed plate using a multi-axis support for precise collection of high-definition images of the surface of the pressed plate. The frame rate of the side-view camera and the infrared camera is ≥ 25 fps. The side-view camera is configured to capture the edge shape of the pressed plate, and the infrared camera is configured to capture the internal glue distribution image of the pressed plate.

[0007] In some possible implementations, the multi-spectrum fusion detection module assigns fusion weights of ultraviolet, infrared, and visible light spectrum data according to the detected defect types of the pressed plate. In the crack defect detection, the weight ratio of the ultraviolet spectrum is ≥ 60%, in the glue distribution anomaly detection, the weight ratio of the infrared spectrum is ≥ 70%, and in the color difference defect detection, the weight ratio of the visible light spectrum is ≥ 80%, so as to improve the detection sensitivity of different types of defects.

[0008] In some possible implementation manners, the hot-pressing process detection module integrates a reverse correction model of process parameters, the reverse correction model is used to input the detected defect data into a GRU network, reversely deduce a temperature compensation value, a pressure adjustment range and a pressure holding time correction amount in the hot-pressing process, and feed back to a control unit of a production line hot-pressing machine through the process feedback module.

[0009] In some possible implementation manners, the image preprocessing module adopts a combination of Gaussian filtering and histogram equalization to eliminate image noise and enhance the contrast of flaw features. The image preprocessing module integrates an ambient light compensation submodule, the ambient light compensation submodule is used to collect ambient light parameters through a color temperature sensor, correct image color data through an RGB three-color channel compensation model, and eliminate detection errors caused by changes in ambient light.

[0010] In some possible implementation manners, the quality data management module integrates a defect traceability submodule and a warning submodule based on digital twinning; The defect traceability submodule is used to construct a digital twinning model of the whole production process of the hot-pressing, access defect data and production line process parameters, and accurately locate the position where the defect of the hot-pressed plate is generated. The warning submodule is used to set three-level warning thresholds, trigger audible and light alarms when the defect rate is greater than or equal to 2%, start the production line to slow down when the defect rate is greater than or equal to 5%, trigger the production line to pause when the defect rate is greater than or equal to 10%, and push process optimization suggestions to an operation terminal.

[0011] In some possible implementation manners, the software unit further includes a color uniformity detection module, which is used to divide the hot-pressed plate into a 5x5 grid for regional detection, process and identify the color difference between the grid area of the hot-pressed plate and the standard color plate by using a CIEDE2000 color difference formula, and simultaneously perform cross-batch color consistency comparison, and automatically warn when the color difference deviation between the batch and the historical qualified batch exceeds a preset threshold.

[0012] In some possible implementation manners, the AI inference module further includes a flaw classification submodule and a quality rating submodule. The flaw classification submodule adopts an SVM classifier, which is used to distinguish true and false flaws in combination with 3D point cloud data obtained by the camera array; The quality rating submodule is used to perform qualified judgment according to the flaw size, and determine the qualified and unqualified according to the preset flaw size value of the flaw type of the hot-pressed plate, when the unqualified is determined, classify through the sorting control module, and mark the defect position on the surface of the unqualified hot-pressed plate.

[0013] In some possible implementation manners, the quality inspection system adopts a production line edge computing architecture. The edge computing architecture is equipped with a double model iteration sub-module for offline training of new defect samples and a light YOLOv5 that updates its parameters by distilling offline training knowledge, without interrupting the production line to complete the learning iteration of the double model iteration sub-module.

[0014] In some possible implementation manners, the multi-light source module comprises: a ring-shaped LED light source, a backlight light source, and a light source controller.

[0015] From the above technical solution, the present application has the following beneficial effects: 1. The detection system adopts multi-spectral fusion detection technology, integrates visible light, infrared and ultraviolet multi-modal image information, designs detection logic according to the hot-pressing characteristics of the pressed board, can accurately identify subtle defects such as color gradient and texture abnormalities, avoids the blindness of general AI detection, realizes the effect of comprehensive coverage of multiple types of defects and unified detection judgment standard, effectively reduces the problems of traditional detection not sensitive to subtle defects, high false alarm and missed detection rate and different artificial detection standards, and provides reliable detection basis for product quality stability; 2. The light AI model is optimized to adapt to high-speed production line operation, without relying on expensive high-end equipment, while realizing automatic operation of the quality inspection process, replacing the traditional inefficient manual detection mode, greatly improving the detection efficiency and controlling the equipment procurement and maintenance cost, significantly reducing the labor cost burden of the enterprise in the quality inspection link, avoiding the high investment pressure brought by high-end equipment, adapting to the actual needs of the large-scale production of pressed boards, and improving the overall production and operation efficiency of the enterprise; 3. The detection quality data management function is constructed, the effective association between the quality detection data and the production process parameters is established, the whole process traceability and deep mining of the detection data are realized, the direction for production optimization is provided through the hot-pressing process algorithm, the effect of detection data and production process feedback optimization is realized, the problem that traditional detection cannot support process optimization is reduced, the key influencing factors in the production process are accurately located, the process parameters are adjusted in a targeted manner, the production process is continuously improved, the quality and efficiency of pressed board production are improved, and the industry is upgraded. BRIEF DESCRIPTION OF DRAWINGS

[0016] The present application will be further described below with reference to the accompanying drawings.

[0017] Figure 1 The intelligent quality inspection system provided by the present application is shown in the schematic diagram; Figure 2 The quality inspection process provided by the present application is shown in the schematic diagram. DETAILED DESCRIPTION

[0018] The terms "first", "second", and "third" and the like in the specification of the present application, claims and drawings are used to distinguish different objects, and are not used to limit a specific order.

[0019] In the embodiments of the present application, the words such as "exemplary" or "for example" are used to mean serving as an example, instance, or illustration, and should not be necessarily construed as a preference or a benefit. Rather, use of the words such as "exemplary" or "for example" is intended to present concepts in a concrete manner.

[0020] Embodiment 1 To solve the above problems, the present application provides a multi-modal AI vision-based intelligent quality inspection system for pressed and pasted boards. Please refer to Figure 1 and Figure 2 ; In some possible implementations, a camera array and a multi-light source module are deployed at a production line detection station. The overhead camera uses a 20 million pixel industrial camera, is installed at a height of 1.2 meters from the surface of the board, is fixed through a multi-axis support, and is set to a frame rate of 30 fps to collect high-definition visible light images of the surface of the pressed and pasted board. The side-view camera and the infrared camera are both set to a frame rate of 25 fps to capture the edge shape of the board and the infrared image of the internal glue distribution, respectively. Four ring-shaped LED light sources in the multi-light source module surround the detection area to eliminate reflection interference, and the backlight source enhances the contrast of the defect profile. Through a light source controller, the light adjustment is triggered synchronously, and at the same time, through an industrial computer, the camera array, the multi-light source module, and the production line hot press are linked to collect temperature and pressure parameters in real time during the hot pressing process. The image preprocessing module is started. First, the Gaussian filter algorithm is used to eliminate image noise, and then the histogram equalization is used to enhance the contrast between the defect features and the background. The ambient light compensation submodule is started synchronously. The color temperature sensor is used to collect the workshop ambient light parameters (measurement range 2000K-10000K), the RGB three-color channel compensation model is used to correct the visible light image color data, and the detection error caused by the change of ambient light is avoided. The multispectral fusion detection module of the AI inference module receives the preprocessed visible light image, infrared image, and ultraviolet image, and performs detection according to the dynamic weight distribution rule: when detecting crack defects, the weight proportion of the ultraviolet spectrum is set to 65%, and the width of the micro crack ≥0.01mm is mainly identified; when detecting abnormal glue distribution, the weight proportion of the infrared spectrum is adjusted to 72%, and the uniformity parameter is calculated through the infrared glue distribution analysis unit; when detecting surface bubbles, the visible light spectrum is mainly used (weight proportion 80%), and the contour feature recognition is combined. The hot pressing process detection module is based on an LSTM network, and correlates the real-time collected hot pressing temperature, pressure parameters and defect detection data for analysis, to predict the quality change trend. When uneven glue distribution is detected, the defect position, unevenness degree and other data are input into the improved GRU network to reversely deduce the hot pressing temperature compensation value (accuracy ±0.5℃), pressure adjustment amplitude (accuracy ±0.1MPa) and pressure holding time correction amount (accuracy ±1s), which are fed back to the control unit of the hot press in real time through the process feedback module to dynamically adjust the detection threshold; The flaw classification submodule of the AI reasoning module calls the SVM classifier, combines the 3D point cloud data collected by the camera array, calculates the height difference of the defect area (≥0.1mm is determined as a real defect), and distinguishes between bubbles and dust, scratches and other false flaws. The quality rating submodule performs pass / fail judgment according to the preset standard: bubble diameter > 2mm, crack length > 3mm, and uneven glue distribution area ratio > 5% are determined as unqualified. The sorting control module triggers the sorting robot, and at the same time, the marking machine marks the defect position on the surface of the unqualified plate with high precision (accuracy ±1mm).

[0021] The embodiment realizes the synchronous detection of multiple types of defects through the precise data collection of the hardware unit and the multi-algorithm cooperation of the software unit. Compared with the limitation of traditional visual detection that can only identify obvious defects, the present scheme improves the detection sensitivity of different defects through dynamic spectral weight distribution, distinguishes between true and false flaws based on 3D point cloud data to reduce the false positive rate, and reversely corrects the hot pressing process parameters to form a closed loop of detection, analysis and adjustment, solving the problem of disconnection between traditional detection and production process.

[0022] Embodiment 2 In some possible implementations, the color uniformity detection module automatically divides the pressed board into a 5x5 grid, which can realize precise detection of small areas, and the positioning of flaw positions is more convenient and accurate. The color difference calculation and dynamic calibration use the CIEDE2000 color difference formula to calculate the color difference value of each grid area and the standard color plate, and simultaneously start the ambient light compensation submodule of the image preprocessing module to real-time correct the interference of ambient light on color data, so that the color difference detection accuracy reaches ΔE≤0.5. A dynamic threshold model is established based on historical quality inspection data, and the color difference allowable range is adjusted adaptively according to the color characteristics of different material boards; The quality data management module calls the color data of the historical qualified batches as a reference benchmark, and compares the color difference data of each grid area of the current batch with the benchmark data one by one. When the color difference deviation of a certain area exceeds the preset threshold, the system automatically triggers an early warning reminder. If the overall color difference fluctuation amplitude across batches is >1.0ΔE, the process optimization suggestion is pushed to the operation terminal simultaneously, prompting to adjust the hot pressing temperature or paint ratio; The color uniformity detection module generates a detailed color detection report, marks the unqualified grid area and the color difference value, and if qualified, passes through the sorting control module, if unqualified, marks the problem area through the marking machine, and feeds back the detection data to the process feedback module to provide data support for the color parameter adjustment of subsequent production.

[0023] The embodiment realizes accurate identification of gradual color difference by combining grid sub-area detection with CIEDE2000 color difference formula, solves the drawbacks of traditional detection, the ambient light compensation submodule ensures the detection stability under different working conditions, and the cross-batch comparison function realizes the color consistency control of batch production, meeting the requirements for product appearance quality.

[0024] Embodiment 3 The defect traceability submodule of the quality data management module constructs a full-process digital twin model of the pressing production, covering all links such as feeding, hot pressing, cooling and detection, and binds the defect data (type, position, size) detected by the camera array with real-time process parameters (hot pressing temperature, pressure, feeding speed, glue coating amount) and equipment operation state data in space and time to form a complete data chain; For example, when a batch of glue distribution uneven defects are detected, the defect traceability submodule traces back the hot pressing process data of the batch of plates through the digital twin model, locates the specific heating plate area of the hot press and the corresponding time node, and analyzes that the defect reason is that the temperature distribution of the heating plate is uneven, and for edge crack defects, whether the edge stress is uneven due to too fast plate feeding is judged through the bound feeding speed data; The early warning submodule monitors the defect rate change according to the set three-level threshold: when the defect rate reaches 2%, a sound-light alarm is triggered to remind the operator to pay attention to the corresponding process parameters, when the defect rate rises to 5%, the production line is automatically started to slow down through the PLC linkage module, and preliminary optimization suggestions (such as adjusting the temperature distribution of the hot press and checking the glue coating roller) are pushed, and when the defect rate reaches 10%, the production line is paused, a detailed defect traceability report and process adjustment scheme are generated, and the production is resumed after the operator confirms the optimization; The process adjustment suggestions obtained through traceability analysis are pushed to the hot press control unit through the process feedback module to adjust the corresponding heating plate temperature or glue coating amount, and at the same time, the parameter model of the hot pressing process detection module is updated to dynamically optimize the detection threshold and reduce the defect recurrence rate from the source; The embodiment realizes accurate traceability of defects through digital twin technology, solves the problems that in traditional detection, only defects can be detected but the specific situation of the defects and the causes of the defects cannot be determined, and the three-level early warning mechanism realizes the transformation from passive detection to active intervention, combined with the process feedback closed loop, significantly reduces the risk of producing batch unqualified products, and improves the stability of the process of the production line.

[0025] Example 4 The double-model iterative sub-module of the edge computing architecture on the industrial computer: the double models of the double-model iterative sub-module are a teacher model and a student model. The teacher model adopts a high-precision algorithm architecture and is deployed on an edge offline computing node to process new defect sample training. The student model is a lightweight YOLOv5 optimized version and is integrated into an AI inference module to be responsible for real-time detection tasks with a detection delay controlled within 90 ms; When a new type of pressing and pasting board is introduced into the production line, the system automatically collects new defect samples (such as wood skin wrinkling and local delamination) of the material under different hot pressing parameters through a camera array and uploads them to the training data set of the teacher model. The teacher model performs feature extraction, labeling, and model training on the new defect samples offline, optimizes the defect recognition algorithm, and improves the model detection accuracy after training; The model distillation mechanism is started, and the teacher model transmits the new defect feature knowledge and recognition logic obtained by training to the student model in real time. The student model updates its parameters through real-time distillation algorithm, quickly learns the recognition ability of new defects while maintaining the lightweight characteristics, and the entire iteration process runs in the background without interrupting the production line detection process; After the model iteration is completed, the system automatically selects 1000 boards containing new defects for verification. The new defect detection rate is improved from 65% before iteration to more than 98%, and the detection delay is still maintained at 85 ms, meeting the detection needs of high-speed production lines. At the same time, the new defect feature data is stored in the quality data management module, enriching the defect sample library of the system.

[0026] This embodiment solves the problems of interrupting the production line for updating the traditional AI detection model and long adaptation period for new materials through the double-model architecture and real-time distillation algorithm, realizes dynamic optimization and rapid iteration of the model, significantly improves the adaptability of the system to different materials and different defect types, and reduces the cost of upgrading detection technology.

[0027] Example 5 After the production line is started, the industrial computer realizes linkage with the feeding area, hot press, cooling section, and sorting area of the production line through the PLC linkage module. When the board enters the detection station after passing through the cooling section, the industrial computer issues control instructions to synchronously trigger the camera array and the multi-light source module to start working. The light source controller of the multi-light source module automatically adjusts the brightness and spectral type of the annular LED light source and the backlight light source according to the glossiness of the board surface; The camera array synchronously collects visible light, infrared and ultraviolet three-channel images, and transmits them to an industrial computer for preprocessing and feature extraction. An AI inference module runs functions such as multispectral fusion detection, heat pressing process adaptive detection, and color uniformity detection in parallel, completes defect identification, classification and quality rating, and a quality data management module stores detection results, process parameters and defect traceability information in real time, forming a traceable quality database.

[0028] A sorting control module directly releases qualified plate materials to the next process according to the quality rating results, triggers a sorting manipulator to store unqualified plate materials, and marks the defect position and type through a marking machine. Meanwhile, a process feedback module analyzes batch defect data and corresponding process parameters, generates adjustment suggestions for heat pressing temperature, pressure and holding time, and feeds back to the heat pressing machine control unit in real time, realizing dynamic optimization of the production process.

[0029] During the entire operation process, the industrial computer monitors the working state of each hardware module (camera frame rate, light source stability, sensor data) and the running efficiency of the software module in real time. When a device failure or detection accuracy anomaly occurs, an alarm device is triggered to prompt, and a quality data management module generates a production line quality inspection report periodically, including defect type distribution, qualified rate change trend, process parameter optimization effect, etc., providing decision basis for enterprise production management.

[0030] The embodiment realizes deep cooperation between hardware units and software units, builds a full-closed-loop quality inspection process of acquisition-detection-judgment-sorting-optimization, and compared with traditional manual detection and decentralized visual detection, not only improves the detection efficiency, but also significantly improves the product qualified rate through process closed-loop optimization, reduces labor cost and raw material waste, and fully reflects the value of the system in large-scale application.

[0031] The basic principles, main features and advantages of the present application are shown and described above. Those skilled in the art should understand that the present application is not limited to the above embodiments, and the above embodiments and descriptions in the specification are only to illustrate the principles of the present application. Without departing from the spirit and scope of the present application, various changes and improvements can be made to the present application, and these changes and improvements all fall within the scope of the claimed present application. The scope of protection of the present application is defined by the appended claims and their equivalents.

Claims

1. An intelligent quality inspection system for laminated boards based on multimodal AI vision, characterized in that, The quality inspection system includes: hardware units and software units; The hardware unit includes: a camera array, a multi-light source module, a PLC linkage module, a sorting control module, and an industrial computer; The camera array includes: a top-down camera, a side-view camera, and an infrared camera, used for multi-angle shooting and detection of the pressed board material; The multi-light source module is used to provide stable illumination during the detection process; The PLC linkage module is used to realize the linkage control between the quality inspection system and the production line equipment; The sorting control module is used to sort qualified and unqualified boards according to the test results; The industrial control computer is used to carry out the operation of the software unit and the data interaction and control command issuance of the hardware unit; The software unit includes: an image preprocessing module, an AI inference module, a quality data management module, and a process feedback module; The image preprocessing module is used to perform noise reduction and feature enhancement processing on the acquired images; The AI ​​inference module integrates a multispectral fusion detection module and a hot-pressing process detection module; The multispectral fusion detection module is used to receive visible light images, infrared images, and ultraviolet images, and output a defect score through multi-feature fusion decision-making. The hot pressing process detection module predicts the quality changes during the hot pressing process based on an LSTM network. It is used to adjust the detection threshold according to temperature and pressure and learn the quality characteristics of the material formulation. The quality data management module is used to store, summarize, and perform in-depth analysis of quality data, and to establish the correlation between quality data and process parameters. The process feedback module is used to feed the test results back to the production line, providing data for adjusting process parameters.

2. The system according to claim 1, characterized in that, The overhead camera is an industrial camera with a frame rate of ≥30fps. The overhead camera is installed 1.2 meters above the surface of the pressing plate and is mounted using a multi-axis bracket. It is used to accurately capture high-definition images of the surface of the pressing plate. Both the side-view camera and the infrared camera have a frame rate of ≥25fps. The side-view camera is used to capture the edge shape of the pressed board, and the infrared camera is used to capture images of the glue distribution inside the pressed board.

3. The system according to claim 1, characterized in that, The multispectral fusion detection module assigns fusion weights to ultraviolet, infrared, and visible light spectral data according to the detected defect type of the pressed plate. Among them, the ultraviolet spectrum weighting is ≥60% for crack defect detection, the infrared spectrum weighting is ≥70% for abnormal glue distribution detection, and the visible light spectrum weighting is ≥80% for color difference defect detection, which is used to improve the detection sensitivity of different types of defects.

4. The system according to claim 1, characterized in that, The hot pressing process detection module integrates a reverse correction model of process parameters. The reverse correction model is used to input the detected defect data into the GRU network, reversely derive the temperature compensation value, pressure adjustment range and holding time correction amount in the hot pressing process, and feed it back to the control unit of the hot press in the production line through the process feedback module.

5. The system according to claim 1, characterized in that, The image preprocessing module uses a combination of Gaussian filtering and histogram equalization to eliminate image noise and enhance the contrast of blemish features. The image preprocessing module integrates an ambient light compensation submodule. The ambient light compensation module is used to collect ambient light parameters through a color temperature sensor and correct image color data through an RGB three-color channel compensation model to eliminate detection errors caused by changes in ambient light.

6. The system according to claim 1, characterized in that, The quality data management module integrates a defect tracing submodule and an early warning submodule based on digital twins; The defect tracing submodule is used to build a digital twin model of the entire lamination production process, access defect data and production line process parameters, and accurately locate the location where defects occur in the lamination board. The early warning submodule is used to set three-level early warning thresholds. When the defect rate is ≥2%, an audible and visual alarm is triggered; when the defect rate is ≥5%, the production line speed is reduced; when the defect rate is ≥10%, the production line is suspended, and process optimization suggestions are pushed to the operation terminal.

7. The system according to claim 1, characterized in that, The software unit also includes a color uniformity detection module, which divides the pressed board into a 5×5 grid for regional detection, uses the CIEDE2000 color difference formula to process and identify the color difference between the grid area of ​​the pressed board and the standard color plate; at the same time, it performs cross-batch color consistency comparison, and automatically issues an early warning when the color difference deviation between the current batch and the historical qualified batch exceeds a preset threshold.

8. The system according to claim 1, characterized in that, The AI ​​reasoning module also includes: a defect classification submodule and a quality rating submodule; The defect classification submodule uses an SVM classifier to distinguish between real and fake defects by combining the 3D point cloud data acquired by the camera array. The quality rating submodule is used to perform acceptance judgment based on the defect size. It determines acceptance or non-acceptance based on the defect type of the laminated board and presets the defect size value. When non-acceptance is determined, the sorting control module classifies the non-acceptable laminated board and marks the defect location on the surface of the non-acceptable laminated board.

9. The system according to claim 1, characterized in that, The quality inspection system adopts a production line edge computing architecture; The edge computing architecture is equipped with a dual-model iterative submodule for offline training of new defect samples, and a lightweight YOLOv5, which updates its own parameters by distilling offline training knowledge, and can complete the learning iteration of the dual-model iterative submodule without interrupting the production line.

10. The system according to claim 1, characterized in that, The multi-light source module includes: a ring LED light source, a backlight light source, and a light source controller.

Images