Intelligent steel coil inspection device adopting vision and machine learning and use method of intelligent steel coil inspection device
Intelligent steel coil inspection devices using vision and machine learning automate the detection of defects in steel coils, solving the problems of low efficiency and insufficient accuracy of manual inspection, and achieving high recognition rate and high accuracy in steel coil inspection.
Patent Information
- Application Number
- CN202511730940.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-11-24
- Publication Date
- 2026-02-24
AI Technical Summary
In existing technologies, steel coil temperature measurement, end coil number identification, and end quality inspection still rely on manual operation, resulting in low labor efficiency, large errors, inability to achieve data-driven management, and inaccurate thickness measurement, which cannot meet the needs of high-end customers.
An intelligent steel coil inspection device employing vision and machine learning combines an industrial camera, supplementary lighting, a six-axis robot, and a deep learning model to automatically detect steel coil defects. It identifies steel coil defects through image processing and deep learning, and integrates functions such as temperature measurement, width measurement, end coil number recognition, and quality inspection.
It achieves automated and precise steel coil defect identification with a high recognition rate, reduces manual intervention, lowers labor costs, improves detection accuracy and efficiency, and meets the thickness measurement requirements of high-end customers.
Smart Images
Figure CN121551405A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of measurement methods and apparatus specifically applicable to metal rolling mills, specifically to an intelligent steel coil inspection device employing vision and machine learning, and its method of use. Background Technology
[0002] Currently, a leading factory in this field still relies on manual measurement, recording, and inspection for steel coil temperature measurement, end coil number identification, and end quality inspection (tower shape, edge damage, edge notches, folds, flattened coils, loose coils, and overflow edges). Steel coil width is measured using a thickness gauge, but the gauge only provides a midpoint value. High-end customers of pickled steel sheets are highly sensitive to thickness, and the midpoint thickness cannot represent the thickness along the width. Currently, the unit manually measures the thickness at both ends of the strip, which is labor-intensive, inefficient, and prone to measurement errors, leading to inconsistent thickness and quality control. It also fails to achieve data-driven thickness detection along the strip's length. Furthermore, fluctuations in the thickness of the incoming iron oxide scale cause errors in the thickness after pickling. In summary, the automation and intelligence levels of current technologies and products need improvement. Summary of the Invention
[0003] In order to overcome the shortcomings of the existing technology and provide a rolling mill auxiliary equipment with high recognition rate and improved inspection efficiency, this invention discloses an intelligent steel coil inspection device and its usage method that adopts vision and machine learning.
[0004] The present invention achieves its objective through the following technical solution: An intelligent steel coil inspection device employing vision and machine learning includes a base, which is fixed to a plane by at least four supports. The device further includes a base adjuster, a sensor bracket, a photoelectric switch, a control cabinet, a robot, a seventh-axis device, and a vision assembly. The lower part of the support column is equipped with a base adjuster, and the bottom of the support column is equipped with a sensor bracket, on which a photoelectric switch is mounted. The control cabinet, robot, and seventh-axis device are sequentially fixed on the base. The robot's vision components are secured by clamps. The control cabinet has a built-in image processing system and is connected to the robot, the seventh axis device, the vision component, and the photoelectric switch via signal lines.
[0005] The intelligent steel coil inspection device employing vision and machine learning is characterized by: The base adjuster uses a gear-rack assembly; The controller inside the control cabinet is a microcomputer or programmable controller; A six-axis robot was selected. The vision components include an industrial camera and a supplemental light.
[0006] The method of using the intelligent steel coil inspection device employing vision and machine learning is characterized by the following steps being implemented sequentially: S1. Photography: The vision unit captures images of the steel coil surface from multiple angles and performs image acquisition; S2. Image Processing: The image processing system built into the control cabinet performs noise reduction, contrast enhancement, and edge detection on the acquired raw images to eliminate uneven lighting or environmental interference and highlight the features of defective areas in the steel coil. S3. Defect Identification: The light source design enhances the contrast between defects and the background. Image preprocessing is performed, and the morphology, texture and other features of defects are automatically learned based on a deep learning model to identify the type of steel coil defect and locate the coordinate position of the steel coil defect. S4. Handling: After receiving information about steel coil defects, the control cabinet outputs an alarm signal to notify manual handling.
[0007] The method of using the intelligent steel coil inspection device employing vision and machine learning is characterized in that: in step S3, the types of steel coil defects identified include: tower shape, edge damage, edge notch, fold, flattened coil, loose coil, and overflow edge.
[0008] The method of using the intelligent steel coil inspection device employing vision and machine learning is characterized in that step S3 is implemented as follows: When the steel coil stops at a saddle-modified roller station, the walking beam system sends a signal, and the temperature sensor automatically measures the temperature of the steel coil end face and transmits the measurement result back to the main control system. The main control system matches and determines the temperature range, and then transmits the measurement data to the control network to complete the steel coil temperature measurement. The rangefinder installed at the end of the robot measures the width and transmits the data to the main control system. It then compares the data with the standard data sent by the main control system. If the error is greater than 20mm or less than 0mm, an alarm signal is issued, and the operator handles the situation to complete the steel coil width measurement. After the steel coil temperature and width measurements are completed, the fixedly installed vision component automatically takes multiple photos of the left and right ends of the steel coil to collect data and train the model. Model training requires defining a loss function and an optimizer. During training, the training data is input into the model for multiple iterations, and the model parameters are optimized through backpropagation to gradually learn effective defect features. The defect detection model can be based on a convolutional neural network (CNN) architecture or use a pre-trained model for transfer learning. Depending on the defect detection requirements, a suitable network structure needs to be designed. When building the model, an attention mechanism can be introduced to enhance the model's attention to important features. The collected data is then matched and compared with the trained defect model for analysis. After the left and right ends of the steel coil are identified, the main control system controls the operation of the modified roller and takes multiple photos of the circumference of the steel coil to ensure 360-degree quality inspection without blind spots. The vision component judges the quality of the ends and circumference of the steel coil and sequentially feeds back the type and number of defects to the industrial monitoring system. If there are no defects, a good coil signal is fed back to complete the steel coil end quality identification. The robot drives the vision component to recognize characters on the left and right ends of the steel coil in the forward direction, and compares the recognized steel coil information with the current workstation steel coil information given by the main control system. If the characters do not match the data given by the main control system, or if the characters are blurry and cannot be recognized, an alarm signal is issued for the operator to judge and handle, so as to complete the steel coil character recognition. If there are no problems with the steel coil, the main control system will automatically terminate the interlock with the walking beam system, and the walking beam system will continue to transport the next steel coil. If there is a problem with the steel coil, it is necessary to wait for the on-site operator to finish handling it and send a signal to the main control system to continue running before the walking beam system can continue to operate.
[0009] This invention comprises an industrial six-axis robot, an image acquisition system, an image processing system, a mechanical system, a PLC, a human-machine interface, and a safety protection system. It integrates functions such as temperature measurement, width measurement, automatic end coil number recognition, and end quality inspection (towering, edge damage, edge notches, folds, flattened coils, loose coils, and overflow edges). It can automatically replace manual labor to complete the following operations: measuring steel coil temperature, measuring steel coil width, identifying end coil numbers, detecting towering at the end of the steel coil, detecting edge damage at the end of the steel coil, detecting edge notches at the end of the steel coil, detecting folds at the end of the steel coil, detecting overflow edges, detecting flattened coils, and loose coils. It automatically alarms when abnormalities are detected and notifies manual handling.
[0010] This invention solves the drawbacks of traditional manual inspection of cold-rolled steel coils. It automatically detects various parameters of the steel coil using infrared and vision sensors and automatically learns defects based on a deep learning model. By statistically accumulating data, the model is optimized to further improve the recognition rate. The automatic recognition rate of non-manually handwritten steel coil numbers reaches 99%, the temperature difference of steel coil measurement is ±1℃, the width measurement error is ±2mm, the recognition rate of end quality defects (such as edge notches, folds, flattened coils, loose coils) is ≥95%, the tower shape measurement accuracy is ±10mm, and the overflow edge measurement accuracy is ±10mm.
[0011] This invention offers the following advantages: it enables intelligent inspection of steel coils, replacing traditional manual inspection, reducing the number of on-site personnel, lowering labor costs, and freeing workers from arduous physical labor in high-temperature, high-noise, and high-risk environments. Furthermore, this device has broad prospects for widespread application. In summary, this device possesses significant economic and social value. Attached Figure Description
[0012] Figure 1 This is the front view of the present invention. Figure 2 This is the left view of the present invention. Figure 3 This is a perspective view of the present invention. Detailed Implementation
[0013] The present invention will be further illustrated below through specific embodiments. Example
[0014] An intelligent steel coil inspection device employing vision and machine learning includes a base 1, a support column 11, a base adjuster 12, a sensor bracket 21, a photoelectric switch 22, a control cabinet 31, a robot 32, a seventh-axis device 33, and a vision component 34. Figures 1-3 As shown, the specific structure is: The base 1 is fixed to the plane by at least four supports 11 arranged in two rows, such that a row of supports 11 is as follows: Figure 2 As shown at point I, it is bolted to the tunnel frame; another row of support columns 11 is as follows. Figure 2 As shown at point II, it is connected to the ground via chemical bolts. A base adjuster 12 is located at the lower part of the support column 11, and a sensor bracket 21 is located at the bottom of the support column 11. After the base adjuster 12 is used to adjust and fix the height of the support column 11 on-site, it will be as shown... Figure 2 As shown at point III, the sensor bracket 21 is welded to the support column 11, and the sensor bracket 21 is equipped with a photoelectric switch 22. The control cabinet 31, robot 32, and seventh-axis device 33 are sequentially fixed on the base 1. The vision component 34 is fixed to the robot 32 by a clamp. The control cabinet 31 has a built-in image processing system. The control cabinet 31 is connected to the robot 32, the seventh axis device 33, the vision component 34 and the photoelectric switch 22 via signal lines.
[0015] In this embodiment: The base adjuster 12 uses a gear-rack assembly; The controller inside control cabinet 31 is a microcomputer or programmable controller; Robot 32 is a six-axis robot; Vision component 34 includes an industrial camera and a supplementary light; When using this embodiment, follow these steps in sequence: S1. Photography: The vision unit captures images of the steel coil surface from multiple angles and performs image acquisition.
[0016] S2. Image Processing: The image processing system built into the control cabinet 31 performs noise reduction, contrast enhancement and edge detection on the acquired raw images to eliminate uneven lighting or environmental interference and highlight the features of the defective areas of the steel coil.
[0017] S3. Defect Identification: The light source design enhances the contrast between defects and the background. Image preprocessing is performed, and the morphology, texture and other features of defects are automatically learned based on a deep learning model to identify the types of steel coil defects and locate their coordinate positions. The identified steel coil defect types include: tower shape, edge damage, edge notch, fold, flattened coil, loose coil and overflow edge.
[0018] The specific details of S3 are as follows: When the steel coil stops at a saddle-modified roller station, the walking beam system sends a signal, and the temperature sensor automatically measures the temperature of the steel coil end face and transmits the measurement result back to the main control system. The main control system matches and determines the temperature range, and then transmits the measurement data to the control network to complete the steel coil temperature measurement. The rangefinder installed at the end of the robot 32 measures the width and transmits the data to the main control system. It then compares the data with the standard data sent by the main control system. If the error is greater than 20mm or less than 0mm, an alarm signal is issued, and the operator handles the situation to complete the steel coil width measurement. After the steel coil temperature and width measurements are completed, the fixedly installed vision component 34 automatically takes multiple photos of the left and right ends of the steel coil to collect data and train the model. Model training requires defining a loss function and an optimizer. During the training process, the training data is input into the model for multiple iterations. The model parameters are optimized through backpropagation to gradually learn effective defect features. The defect detection model can be based on a convolutional neural network (CNN) architecture or use a pre-trained model for transfer learning. Depending on the defect detection requirements, a suitable network structure needs to be designed. When building the model, an attention mechanism can be introduced to enhance the model's attention to important features. The collected data is then matched and compared with the trained defect model for analysis. After the left and right ends of the steel coil are identified, the main control system controls the operation of the modified roller and takes multiple photos of the circumference of the steel coil to ensure 360-degree quality inspection without blind spots. The vision component 34 judges the quality of the ends and circumference of the steel coil and sequentially feeds back the defect category and number of defects to the industrial monitoring system. If there are no defects, a good coil signal is fed back to complete the steel coil end quality identification. Robot 32 drives vision component 34 to perform character recognition on the left and right end faces of the steel coil in the forward direction, and compares the recognized steel coil information with the current workstation steel coil information given by the main control system. If the characters do not match the data given by the main control system, or the characters are blurry and cannot be recognized, an alarm signal is issued for the operator to judge and handle, so as to complete the steel coil character recognition. If there are no problems with the steel coil, the main control system will automatically terminate the interlock with the walking beam system, and the walking beam system will continue to transport the next steel coil. If there is a problem with the steel coil, it is necessary to wait for the on-site operator to finish handling it and send a signal to the main control system to continue running before the walking beam system can continue to operate.
[0019] S4. Handling: After receiving the steel coil defect information, the control cabinet 31 outputs a warning signal to notify manual handling.
Claims
1. An intelligent steel coil inspection device employing vision and machine learning, comprising a base (1), the base (1) being fixed to a plane by at least four pillars (11), characterized in that: It also includes a base adjuster (12), a sensor bracket (21), a photoelectric switch (22), a control cabinet (31), a robot (32), a seventh-axis device (33), and a vision component (34). The lower part of the support column (11) is provided with a base adjuster (12), and the bottom of the support column (11) is provided with a sensor bracket (21). A photoelectric switch (22) is provided on the sensor bracket (21). The control cabinet (31), robot (32), and seventh-axis device (33) are fixed sequentially on the base (1). The robot (32) is fixed with a vision component (34) by a clamp. The control cabinet (31) has a built-in image processing system and is connected to the robot (32), the seventh axis device (33), the vision component (34) and the photoelectric switch (22) via signal lines.
2. The intelligent steel coil inspection device employing vision and machine learning as described in claim 1, characterized in that: The base adjuster (12) uses a gear-rack assembly; The controller inside the control cabinet (31) is a microcomputer or a programmable controller; The robot (32) is a six-axis robot; The vision component (34) includes an industrial camera and a fill light.
3. The method of using the intelligent steel coil inspection device employing vision and machine learning as described in claim 1 or 2, characterized in that: Follow these steps in sequence: S1. Photography: The vision component (34) captures images of the steel coil surface from multiple angles and performs image acquisition; S2. Image processing: The image processing system built into the control cabinet (31) performs noise reduction, contrast enhancement and edge detection operations on the acquired raw images to eliminate uneven lighting or environmental interference and highlight the features of the steel coil defect area; S3. Defect Identification: The light source design enhances the contrast between defects and the background. Image preprocessing is performed, and the morphological and texture features of defects are automatically learned based on a deep learning model to identify the type of steel coil defect and locate the coordinate position of the steel coil defect. S4. Handling: After receiving the steel coil defect information, the control cabinet (31) outputs a warning signal to notify manual handling.
4. The method of using the intelligent steel coil inspection device employing vision and machine learning as described in claim 3, characterized in that: In step S3, the identified steel coil defect types include: towering, edge damage, edge notch, fold, flattened coil, loose coil, and overflow edge.
5. The method of using the intelligent steel coil inspection device employing vision and machine learning as described in claim 4, characterized in that: Step S3 shall be performed as follows: When the steel coil stops at a saddle-modified roller station, the walking beam system sends a signal, and the temperature sensor automatically measures the temperature of the steel coil end face and transmits the measurement result back to the main control system. The main control system matches and determines the temperature range, and then transmits the measurement data to the control network to complete the steel coil temperature measurement. The rangefinder installed at the end of the robot (32) measures the width and transmits the data to the main control system. It compares the data with the standard data sent by the main control system. If the error is greater than 20mm or less than 0mm, an alarm signal is issued and the operator handles it to complete the steel coil width measurement. After the steel coil temperature and width measurement are completed, the fixedly installed vision component (34) will automatically take multiple photos of the left and right end faces of the steel coil to collect data and train the model. Model training requires defining a loss function and an optimizer. During the training process, the training data is input into the model for multiple iterations. The model parameters are optimized through backpropagation to gradually learn effective defect features. The defect detection model can be based on a convolutional neural network (CNN) architecture or a pre-trained model can be used for transfer learning. The collected data is then matched and compared with the trained defect model for analysis. After the recognition of the left and right end faces of the steel coil is completed, the main control system controls the operation of the modified roller and takes multiple photos of the circumference of the steel coil to ensure that the circumference of the steel coil is inspected without dead angles. The vision component (34) judges the quality of the end face and circumference of the steel coil and feeds back the category and number of defects of the steel coil to the industrial monitoring system in turn. If there are no defects, it feeds back a good coil signal to complete the steel coil end face quality recognition. The robot (32) drives the vision component (34) to perform character recognition on the left and right ends of the steel coil in the forward direction, and compares the recognized steel coil information with the current workstation steel coil information given by the main control system. If the characters do not match the data given by the main control system, or the characters are blurry and cannot be recognized, an alarm signal is issued for the operator to judge and handle, so as to complete the steel coil character recognition. If there are no problems with the steel coil, the main control system will automatically terminate the interlock with the walking beam system, and the walking beam system will continue to transport the next steel coil. If there is a problem with the steel coil, it is necessary to wait for the on-site operator to finish handling it and send a signal to the main control system to continue running before the walking beam system can continue to operate.