PVC Hard Sheet High-Speed Slitting Quality Monitoring System Based on AI Vision Detection

Abstract

The present invention discloses a high-speed slitting quality monitoring system for PVC hard sheets based on AI vision detection, which relates to the field of industrial automation technology. The system consists of several functional modules, including: a data acquisition module, which, when the PVC hard sheet enters the detection area, controls the camera and the thermal imager through the PLC to synchronously acquire image data and thermal imaging data; a three-dimensional graphics synthesis module, which, based on the image data, superimposes contour recognition to generate a three-dimensional model; based on the thermal imaging data, uses affine transformation to obtain the position information of the thermal imaging data, and performs spatial registration with the three-dimensional model to generate a temperature gradient field; a defect detection and recognition module, which evenly divides the surface of the three-dimensional model, numbers the evenly divided regions, and based on the connected domain analysis of morphological operations, superimposes multi-scale feature fusion to perform defect recognition; according to the temperature gradient field after region segmentation, performs temperature uniformity evaluation and calculates the temperature standard deviation between regions.

Description

Technical Field

[0001] The present invention relates to the field of industrial automation technology, and specifically to a high-speed slitting quality monitoring system for PVC hard sheets based on AI vision detection. Background Art

[0002] As a basic industrial material, PVC hard sheets are widely used in fields such as packaging, construction, and medical treatment; with the increasing requirements for precision and consistency in downstream industries; PVC hard sheets are high-strength and corrosion-resistant rigid plates mainly made of polyvinyl chloride (PVC) resin, and are widely used in the fields of construction, electronics, and packaging. The laminating process eliminates internal stress, improves density and surface smoothness through high-temperature calendering, and endows moisture-proof, antibacterial and other functional characteristics. The adhesive-free lamination uses corona treatment, co-extrusion technology and high-temperature physical bonding (such as hot pressing an EVA layer and an aluminized film at 108 - 115 °C), combines biaxial stretching to enhance mechanical properties, and ensures uniform lamination through drum water cooling and AI vision detection, which not only avoids the pollution problem of traditional adhesives, but also reduces production costs, and can meet the high-standard environmental protection and safety requirements of food packaging, medical equipment, etc.

[0003] Traditional optical detection methods have significant limitations in identifying defects such as scratches and bubbles on the surface of transparent materials; a single light source is difficult to capture the three-dimensional morphological characteristics of defects. Although dark-field microscopy can enhance the scattering signal, it cannot distinguish the types of defects and is easily interfered by the light-transmitting characteristics of the material; existing algorithms are mostly based on static image analysis and cannot meet the real-time detection requirements under high-speed production lines; traditional temperature control relies on thermocouple point measurement and cannot real-time sense the temperature field distribution on the material surface. At critical speeds, the heat exchange between the roller and the material is unbalanced, resulting in uneven shrinkage stress after lamination; the existing PID algorithm adjusts the heating power based on historical data and is difficult to respond in time to transient temperature fluctuations (such as an abnormal area of ±2 °C), causing delamination or warping of the hard sheet. Summary of the Invention

[0004] (I) Technical Problems to be Solved

[0005] In view of the deficiencies of the prior art, the present invention provides a high-speed slitting quality monitoring system for PVC hard sheets based on AI vision detection, which solves the problems raised in the background art.

[0006] (II) Technical Solutions

[0007] To achieve the above objectives, the present invention is realized through the following technical solutions:

[0008] A high-speed slitting quality monitoring system for PVC hard sheets based on AI vision detection, comprising:

[0009] A data acquisition module, when the PVC hard sheet enters the detection area, controls the camera and the thermal imager through the PLC to synchronously collect image data and thermal imaging data;

[0010] The 3D graphics synthesis module generates a 3D model by overlaying contour recognition based on image data; based on thermal imaging data, it uses affine transformation to obtain the position information of the thermal imaging data and performs spatial registration with the 3D model to generate a temperature gradient field.

[0011] The defect detection and recognition module evenly divides the surface of the 3D model, numbers the divided regions, performs defect recognition by overlaying multi-scale feature fusion based on connected component analysis of morphological operations; based on the temperature gradient field after region segmentation, it evaluates the temperature uniformity, calculates the temperature standard deviation between regions, and uses the LSTM network to predict the heat conduction trend.

[0012] The feedback control module dynamically adjusts the power of the heating plate according to the heat conduction trend in combination with the temperature difference compensation algorithm.

[0013] Furthermore, the acquisition of image data and thermal imaging data by humans specifically includes:

[0014] S101: The PLC synchronously triggers to control the camera and the thermal imager.

[0015] S102: Under the trigger of the PLC, the camera irradiates the PVC surface through multi-angle light sources to capture multi-view images; it measures the surface temperature distribution of the PVC sheet in real time.

[0016] S103: Use edge computing data to preprocess the image data and thermal data.

[0017] Furthermore, the generation of the 3D model specifically includes:

[0018] The PVC surface is irradiated by multi-angle light sources to capture multi-light-source images, generate a depth map, generate a detailed image of the PVC surface according to the reflection characteristics of the multi-view images, and construct a 3D model.

[0019] Furthermore, the spatial registration specifically includes:

[0020] Extract corresponding feature points from the optical image and thermal imaging data, calculate the registration parameters, and minimize the cross-error of the feature points; convert the pixel coordinates of the thermal imaging data to the coordinate system of the 3D model and perform dynamic calibration. If it is detected that the corner points of the registered thermal imaging map do not match the corner points of the 3D model, the pose of the thermal imager or the camera is adjusted through the PLC; otherwise, no adjustment is made.

[0021] Furthermore, the numbering of the divided regions specifically includes:

[0022] Using the uniform grid division algorithm, the surface of the 3D model is divided into N×N uniform regions, and a unique number is assigned to each region. A mapping relationship between the 3D model coordinates and the region numbers is established, and the temperature gradient field data is corresponding to the 3D region numbers.

[0023] Further, defect identification specifically includes:

[0024] S301: For the surface image of each region, perform binarization from the slices or projections of the 3D model; use morphological operations to remove noise;

[0025] S302: Use the connected component algorithm to identify the defect regions; filter small regions and label them as defects;

[0026] S303: Combine multi-scale features, combine shallow edge features and deep semantic features; extract details through Canny edge detection, use YOLOv8 to extract defect context features, and introduce channel attention.

[0027] Further, predicting the heat conduction trend specifically includes:

[0028] Collect the temperature distribution of each region on the surface of the PVC hard sheet in real time through an infrared thermal imager to form time series data; record the key parameters during the film laminating process, align them according to the time stamp, denoise the temperature data, select the temperature values of each region in the past period as the time series input; the current heater power and conveyor belt speed parameters are used as static feature inputs to predict the temperature changes of each region in the next few seconds, directly guiding temperature control; combine historical deformation data to output the deformation risk level.

[0029] Further, dynamically adjusting the power of the heating plate specifically includes:

[0030] Dynamically adjust the power of the heating plate according to the difference between the current temperature and the target temperature. For each region, calculate the difference between the current average temperature and the target temperature; associate the defect position coordinates with the heating plate region number, and remove the defective PVC hard sheet according to the defect position; after removal, update the qualification rate statistics and feedback the result to the production management system.

[0031] (III) Beneficial effects

[0032] The present invention provides a high-speed slitting quality monitoring system for PVC hard sheets based on AI vision detection, which has the following beneficial effects:

[0033] (1) By deploying light sources at different angles in the present invention, the transparent PVC hard sheet is irradiated from different angles. Under different light sources, the projection morphology, light and dark contrast, and deformation characteristics of defects (such as scratches, bubbles, and indentations) will have significant differences, which can highlight the direction and depth of surface scratches, locate fine cracks through shadow contrast, input the two-dimensional image data under multiple-angle light sources into a multi-dimensional feature fusion network, combine the complementarity of multiple light sources, preferentially select the light source images with significant differences for modeling, and retain the side light data for modeling to reduce computational redundancy.

[0034] (2) Through the real-time monitoring and analysis of thermal imaging data, the present invention can accurately capture the temperature differences on the surface of PVC hard plates. By combining algorithm calculations and predictions, it dynamically adjusts the heating or cooling parameters during the film pressing process to ensure the temperature uniformity of each region of the material. This technology can effectively prevent the problem of uneven thermal expansion and contraction caused by excessive local temperature differences, thereby avoiding deformation defects such as bending, twisting, or delamination of the hard plates after film pressing due to unbalanced shrinkage stress, and significantly improving the product qualification rate. At the same time, by identifying abnormal temperature regions in advance and feeding them back to the process control terminal, it can reduce manual intervention and rework costs, realize the intelligence and high stability of the production process, and ultimately ensure the high-precision application requirements of PVC hard plates in fields such as electronic packaging and packaging materials. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] Figure 1 It is a schematic diagram of the system of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0036] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0037] Embodiment 1:

[0038] Please refer to Figure 1 , this embodiment provides a high-speed slitting quality monitoring system for PVC hard sheets based on AI vision detection. The detection system includes:

[0039] A data acquisition module. When the PVC hard sheet enters the detection area, it synchronously acquires image data and thermal imaging data by controlling the camera and thermal imager through the PLC.

[0040] The PLC: As the core control unit, it is responsible for coordinating the synchronous operation of production line equipment (such as conveyor belts, cameras, thermal imagers).

[0041] An optical camera: Used to capture high-resolution images of the surface of the PVC hard sheet. A multi-angle light source system is set at the end of the production line, combined with a high-resolution camera array, to capture detailed images of the surface of the PVC hard sheet and detect defects such as scratches, bubbles, and impurities.

[0042] An infrared thermal imager: Measures the surface temperature distribution of the PVC hard sheet in real time and evaluates the heating uniformity.

[0043] S101: The PLC synchronously triggers the control of the camera and the thermal imager; detects the arrival of the PVC hard sheet at the detection area through a photoelectric sensor or an encoder to generate a trigger signal; after receiving the position signal, the PLC calculates the optimal acquisition timing according to the preset logic, such as the conveyor belt speed and the length of the detection area, and sends a synchronous trigger instruction to the camera and the thermal imager; uses hardware triggering, such as directly connecting the PLC to the device or timestamp alignment, and synchronizes through a high-precision clock to ensure the time consistency of the optical image and the thermal imaging data;

[0044] S102: Under the trigger of the PLC, the camera irradiates the PVC surface through multi-angle light sources to capture multi-view images; uses lossless compression, such as PNG or JPEG, and at the same time the resolution needs to meet the defect detection requirements, such as a camera with a resolution of ≥10MP is required for defect identification of ≥0.1mm; the thermal imager synchronously scans the PVC surface to generate a thermal imaging map, the original temperature data or the thermal map, such as PNG / BMP with temperature tags superimposed;

[0045] S103: Transmits the image data and the thermal data to the edge computing node to ensure low latency (<100ms); uses the MQTT protocol to distribute the data to different analysis modules; and performs preprocessing, performs white balance, contrast enhancement, and noise reduction processing on the optical image; based on blackbody calibration, eliminates the interference of the ambient temperature to ensure the temperature measurement accuracy (such as ±0.5°C);

[0046] The 3D graphics synthesis module, based on the image data, superimposes contour recognition to generate a 3D model; based on the thermal imaging data, uses affine transformation to obtain the position information of the thermal imaging data, and performs spatial registration between the 3D model to generate a temperature gradient field;

[0047] 3D model generation:

[0048] Irradiate the PVC surface through multi-angle light sources to capture multi-light source images, construct a 3D model, use a similar Quark algorithm, by inputting multi-view optical images, such as 3 angles, with a spacing of 30cm, to generate a hierarchical depth map in real time, describe the geometric structure of the PVC surface, and enhance the geometric details of the defect area; generate a detailed image of the PVC surface according to the reflection characteristics of the multi-view images; use GPU accelerated rendering to generate a high-resolution 3D model;

[0049] Spatial registration specifically includes:

[0050] Align the position information of the thermal imaging data with the 3D model to establish a spatial correspondence relationship;

[0051] Affine transformation alignment:

[0052] Extract corresponding feature points, such as edges and corners, from optical images and thermal imaging data; through affine transformation models, such as translation, rotation, and scaling; calculate registration parameters to minimize the cross-error of feature points; convert the pixel coordinates of thermal imaging data into the coordinate system of the 3D model, and project the temperature value of each pixel of the thermal imaging image onto the mesh vertices on the surface of the 3D model according to the coordinates after affine transformation. The formula is as follows: ;

[0053] where, are the x-axis, y-axis, and z-axis coordinates corresponding to the vertices of the 3D model respectively; are the x-axis and y-axis coordinates corresponding to the thermal imaging pixels respectively; are the x-axis and y-axis coordinates corresponding to the focal lengths respectively; are the x-axis and y-axis coordinates corresponding to the principal points respectively; and perform dynamic calibration. If it is detected that the corner points of the registered thermal imaging map do not match the corner points of the 3D model, adjust the pose of the thermal imager or camera through the PLC; otherwise, do not make adjustments;

[0054] Temperature gradient field generation:

[0055] Map the thermal imaging data to the 3D model, assign the temperature values of the thermal imaging to the corresponding grid points of the 3D model according to the coordinates after affine transformation; calculate the local temperature gradient based on the temperature difference between adjacent grid points; and generate a heat map to overlay on the surface of the 3D model to mark the temperature anomaly areas; for non-integer mapping positions, use bilinear interpolation or cubic interpolation to calculate the temperature values of the grid vertices to ensure the continuity of temperature distribution;

[0056] Defect detection and recognition module: evenly divide the surface of the 3D model, number the evenly divided areas; based on the connected component analysis of morphological operations, overlay multi-scale feature fusion to perform defect recognition; according to the temperature gradient field after region segmentation, perform temperature uniformity evaluation, calculate the temperature standard deviation between regions, and use the LSTM network to predict the heat conduction trend;

[0057] The specific process of numbering the evenly divided areas includes:

[0058] Use the uniform grid division algorithm to divide the surface of the 3D model into N×N uniform areas, such as a 100×100 grid; assign a unique number to each area, establish the mapping relationship between the 3D model coordinates and the area numbers to ensure that the temperature gradient field data corresponds one-to-one with the 3D area numbers;

[0059] The specific process of defect recognition includes:

[0060] Extract the defect areas through morphological operations and combine multi-scale feature fusion to improve the detection accuracy:

[0061] S301: Binarize the surface image of each region from the slice or projection of the three-dimensional model; remove noise using morphological operations;

[0062] S302: Use the connected domain algorithm to identify defect areas; filter small areas and mark them as defects;

[0063] By removing small noise and optimizing image quality, the image is divided into defective area and normal area by setting a threshold. If the defect is a dark bubble, a low threshold is set to mark the bubble as white, and the rest as black. A unique label is assigned to each connected area to distinguish different defects. The image is scanned, a temporary label is assigned to each foreground pixel, and the equivalent relationship of the labels of adjacent areas is recorded. The labels are unified according to the equivalent relationship to ensure that the same connected area has a unique label.

[0064] Start with an unmarked defective pixel, expand to the surrounding areas in a recursive or queue manner, mark all connected pixels as the same area, repeat this process until all pixels are marked, and the number of pixels contained in the area is used to filter noise. Areas with too small an area may be noise rather than defects; determine the defect type by calculating the aspect ratio, circularity, etc., such as scratches are usually long and thin strips, and bubbles are close to circles; and record the coordinates of the center of the area to locate the specific location of the defect on the product; remove noise based on the area threshold, such as an area less than 50 pixels; classify defects based on shape features, such as scratches when the aspect ratio is >5, and bubbles when the circularity is close to 1;

[0065] For example:

[0066] Collect PVC board images, mark bubbles as white through threshold segmentation, and mark all bubble areas; count the area of ​​each area and filter out noise less than the threshold; output the number and position of bubbles to determine whether the product is qualified; merge adjacent small holes through equivalent relationships to avoid misjudgment as multiple independent defects; perform edge detection on the surface image and extract high-contrast areas of scratches; use the seed filling method to mark long strip connected areas; select scratch areas according to the aspect ratio and calculate their length and position; if the scratch length exceeds the threshold (such as 5mm), trigger an alarm; otherwise, continue to identify;

[0067] S303: Combine multi-scale features, shallow edge features and deep semantic features to improve detection robustness; extract details through Canny edge detection, use YOLOv8 to extract defect context features, and introduce channel attention to enhance defect area features;

[0068] Combining shallow edge features with deep semantic features:

[0069] Shallow edge features:

[0070] Extract highly sensitive edge and detail information in the image to capture the contours or textures of tiny defects; Use the Canny edge detection algorithm for gradient calculation and threshold processing to highlight details such as boundaries, cracks, concavities, and convexities in the image; Sensitive to tiny defects such as scratches or bubble edges at the 0.1mm level, and can quickly locate potential defect areas;

[0071] Deep semantic features:

[0072] Extract high-level semantic information of the image through a deep learning model to understand the type, distribution of defects, and their relationship with the surrounding environment; Real-time detect the location and category of defects, such as scratches, holes, cracks, and is suitable for rapid positioning and classification; Through the encoder-decoder structure, while retaining spatial information, fuse multi-layer features to accurately segment the boundaries of defect areas;

[0073] Align the shallow high-resolution edge features with the deep low-resolution semantic features at different scales and fuse them into a multi-scale feature map; Through learnable weight assignment, dynamically adjust the contribution ratio of edge features and semantic features; During the training process, simultaneously optimize the accuracy of edge detection and semantic segmentation through the loss function to ensure that the fused features can take into account both details and the global situation;

[0074] Gradient calculation and threshold processing:

[0075] Input the original image, use Canny edge detection to extract edge features; By inputting the original image, use YOLOv8 to extract semantic features; Fuse the edge features and semantic features at multiple scales, and enhance the features of key areas through the SE module; Output the final defect detection results, including positioning, classification, and segmentation;

[0076] Calculate the standard deviation of temperature between regions:

[0077] Measure the temperature of temperature sensors or sampling points in each region, record the temperature values of each region, calculate the average temperature of each region by averaging all the temperature values in each region; Divide the total sum of squares by the total number of regions (if it is a complete data set) or the total number of regions minus 1 (if it is sampling data) to obtain the variance;

[0078] By means of a sliding time window, calculate the standard deviation of temperature in different time periods in real time, observe its fluctuation trend. If the standard deviation rises rapidly or remains higher than the set threshold in a short period of time, it indicates that this batch of pvc hard sheets may have progressive defects, and combine with other detection modules for comprehensive judgment to ensure accurate identification of the type and location of defects;

[0079] Predicting the trend of heat conduction specifically includes:

[0080] Collect the temperature distribution of each area on the surface of the PVC hard sheet in real time through an infrared thermal imager to form time series data; record the key parameters during the film laminating process, such as heater power, conveyor belt speed, ambient temperature and humidity; align according to the timestamp to ensure that the input features and output targets at the same moment are consistent; denoise the temperature data to eliminate environmental interference and sensor noise;

[0081] Select the temperature values of each area in the past period of time as the time series input; the current parameters such as heater power and conveyor belt speed as the static feature input; auxiliary variables such as ambient temperature and humidity, and material thickness to enhance the model's adaptability to complex working conditions; predict the temperature changes of each area in the next few seconds to directly guide temperature control; combine historical deformation data to output the deformation risk level;

[0082] LSTM model training and prediction:

[0083] Use multiple layers of LSTM units to capture the long-term dependencies of temperature changes, introduce the attention mechanism, focus on the areas with significant temperature fluctuations, splice the temporal features output by LSTM with the current process parameters, generate the temperature prediction values of each area through a fully connected layer, calculate the local deformation risk based on the deviation between the predicted temperature and the target temperature, combined with the material thermal expansion coefficient, and use historical production data for supervised learning to optimize both the temperature prediction error and the deformation risk classification accuracy at the same time;

[0084] The feedback control module dynamically adjusts the heating plate power according to the heat conduction trend, combined with the temperature difference compensation algorithm, and eliminates defective products;

[0085] Dynamically adjust the heating plate power according to the difference between the current temperature and the target temperature to ensure temperature uniformity; for each area, calculate the difference between the current average temperature and the target temperature; associate the defect position coordinates with the heating plate area number to determine the position of the unqualified products that need to be removed; remove the defective PVC hard sheets according to the defect position; after the removal is completed, the system updates the qualified rate statistics and feeds the results back to the production management system;

[0086] During the production process of PVC hard boards, high-precision monitoring of the surface temperature of the boards through a real-time thermal imaging system, combined with intelligent algorithms to dynamically adjust the heating plate power, can effectively ensure temperature uniformity and achieve precise defect removal;

[0087] After confirming the defect position, automatically remove the unqualified products through an intelligent sorting device to prevent them from entering the subsequent processes, and at the same time synchronize the removal records to the database in real time; after the removal is completed, the real-time qualified rate statistics will be dynamically updated according to the batch output and the number of qualified products, and the data will be fed back to the production management system through an industrial communication protocol to achieve the transparency of production data and the closed-loop optimization of process parameters;

[0088] For example, if there are continuous temperature deviations or abnormal defect rates in a certain area, the system can automatically trigger an alarm and prompt the maintenance personnel to check the operating status of the corresponding heating plate or sensor, or adjust the heating curve parameters of that area, thereby preventing potential quality risks; not only significantly improving the temperature uniformity, reducing deformation problems such as warping and delamination of the sheet caused by temperature differences, but also improving the product qualification rate to over 98% through precise positioning and elimination of defects, while reducing manual intervention and rework costs, achieving the improvement of production efficiency and product quality, and providing reliable material guarantee.

[0089] The above embodiments can be implemented in whole or in part by software, hardware, firmware, or any combination thereof. When implemented using software, the above embodiments can be implemented in whole or in part in the form of a computer program product. Those of ordinary skill in the art will realize that the units and algorithm steps of the examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or in a combination of computer software and electronic hardware. Whether these functions are executed in hardware or software depends on the specific application and design constraints of the technical solution.

[0090] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units. They may be located in one place or distributed to multiple network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0091] The above is only the specific implementation manner of this application, but the protection scope of this application is not limited thereto. Any person skilled in the art within the technical scope disclosed in this application can easily think of changes or substitutions, which should all be covered within the protection scope of this application.

Claims

1. A high-speed slitting quality monitoring system for PVC hard sheets based on AI vision detection, characterized in that: The system includes: A data acquisition module, which, when the PVC hard sheet enters the detection area, controls the camera and the thermal imager through the PLC to synchronously acquire image data and thermal imaging data; A 3D graphics synthesis module, which, based on the image data, superimposes contour recognition to generate a 3D model; based on the thermal imaging data, uses affine transformation to obtain the position information of the thermal imaging data, and performs spatial registration with the 3D model to generate a temperature gradient field; A defect detection and recognition module, which evenly divides the surface of the 3D model, numbers the divided areas, performs defect recognition based on connected domain analysis of morphological operations and superimposes multi-scale feature fusion; based on the temperature gradient field after region segmentation, performs temperature uniformity evaluation, calculates the temperature standard deviation between regions, and uses the LSTM network to predict the heat conduction trend; A feedback control module, which dynamically adjusts the power of the heating plate according to the heat conduction trend and combines the temperature difference compensation algorithm.

2. The high-speed slitting quality monitoring system for PVC hard sheets based on AI vision detection according to claim 1, wherein: Specifically, acquiring the image data and the thermal imaging data includes: S101: The PLC synchronously triggers and controls the camera and the thermal imager; S102: The camera, under the trigger of the PLC, irradiates the PVC surface through multi-angle light sources to capture multi-view images; measures the surface temperature distribution of the PVC hard sheet in real time; S103: Preprocesses the image data and the thermal imaging data using edge computing data.

3. The high-speed slitting quality monitoring system for PVC hard sheets based on AI vision detection according to claim 2, characterized in that: Specifically, generating the 3D model includes: Irradiating the PVC surface through multi-angle light sources to capture multi-light source images, generating a depth map of layers, and generating a detailed image of the PVC surface according to the reflection characteristics of the multi-view images to construct a 3D model.

4. The high-speed slitting quality monitoring system for PVC hard sheets based on AI vision detection according to claim 3, characterized in that: Specifically, the spatial registration includes: Extracting corresponding feature points from the optical image and the thermal imaging data, calculating the registration parameters, and minimizing the cross error of the feature points; converting the pixel coordinates of the thermal imaging data into the coordinate system of the 3D model and performing dynamic calibration. If it is detected that the corner points of the registered thermal imaging map do not match the corner points of the 3D model, the pose of the thermal imager or the camera is adjusted through the PLC; otherwise, no adjustment is made.

5. The high-speed slitting quality monitoring system for PVC hard sheets based on AI vision detection according to claim 1, characterized in that: Specifically, numbering the divided areas includes: Using a uniform grid division algorithm to divide the surface of the 3D model into N×N uniform areas, assigning a unique number to each area, establishing a mapping relationship between the 3D model coordinates and the area numbers, and corresponding the temperature gradient field data with the 3D area numbers.

6. The high-speed slitting quality monitoring system for PVC hard sheets based on AI vision detection according to claim 5, wherein: Specifically, defect recognition includes: S301: Binarizing the surface image of each area by slicing or projecting from the 3D model; using morphological operations to remove noise; S302: Using the connected domain algorithm to identify the defect areas; filtering small areas and marking them as defects; S303: Combining multi-scale features, combining shallow edge features and deep semantic features; extracting details through Canny edge detection, using YOLOv8 to extract defect context features, and introducing channel attention.

7. The high-speed slitting quality monitoring system for PVC hard sheets based on AI vision detection according to claim 6, characterized in that: Specifically, predicting the heat conduction trend includes: The temperature distribution of each area on the surface of the PVC hard sheet is collected in real time through an infrared thermal imager to form time series data; record the key parameters during the film laminating process, align them according to the timestamp, denoise the temperature data, and select the temperature values of each area in the past period as the time series input; the current heater power and conveyor belt speed parameters are used as static feature inputs to predict the temperature changes of each area in the next few seconds, directly guiding the temperature control; combined with historical deformation data, the deformation risk level is output.

8. The high-speed slitting quality monitoring system for PVC hard sheets based on AI vision detection according to claim 1, characterized in that: The dynamic adjustment of the heating plate power specifically includes: According to the difference between the current temperature and the target temperature, dynamically adjust the heating plate power. For each area, calculate the difference between the current average temperature and the target temperature; associate the defect position coordinates with the heating plate area number, and remove the defective PVC hard sheet according to the defect position; after the removal is completed, update the qualification rate statistics and feedback the results to the production management system.

Images