Deep learning-based column nail fracture and roller surface defect timely diagnosis system and deep learning-based column nail fracture and roller surface defect timely diagnosis method

Through a deep learning-based system, the roller surface image detection is performed using the improved YOLOv5s model, which solves the problem that traditional detection methods cannot identify column nail breaks and roller surface defects in real time, and realizes efficient defect diagnosis and classification, improving production efficiency.

CN118918090APending Publication Date: 2024-11-08SHAANXI UNIV OF SCI & TECH
View PDF 0 Cites 1 Cited by

Patent Information

Application Number
CN202411056270.4
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-08-02
Publication Date
2024-11-08

AI Technical Summary

Technical Problem

Traditional detection methods cannot detect column nail breaks and roller surface defects in cement roller presses in real time and effectively, resulting in low production efficiency.

Method used

采用基于深度学习的系统,通过图像采集装置实时采集辊面图像,并利用改进的YOLOv5s模型进行目标检测和缺陷分类,实现对柱钉断裂和辊面缺陷的实时诊断。

Benefits of technology

It realizes efficient identification and classification of roller surface defects, improves detection efficiency, can provide defect information in real time, supports users to repair in a timely manner, and improves production efficiency.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118918090A_ABST
    Figure CN118918090A_ABST
Patent Text Reader

Abstract

The invention discloses a stud fracture and roller surface defect timely diagnosis system and method based on deep learning. The system comprises an image acquisition device, an image processing unit and a user operation unit. The image acquisition device is used for acquiring a roll surface image in real time and transmitting the roll surface image to the image processing unit for next-step work; the user operation unit is used for performing defect detection and classification on the image acquired by the image acquisition device through a target detection algorithm; and the image acquisition unit is used for controlling the movement of an image acquisition device and observing roller surface defects. The system collects the roller surface image in real time, detects and identifies the roller surface defects through a deep learning algorithm, classifies the defects, timely displays the defect part to a user through a visual interface of a PC terminal, and has the advantages of high identification efficiency, stable operation, high efficiency, practicability and the like. The problem that roller surface defects cannot be detected in real time only by means of manual work is solved.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention belongs to the technical field of roller press column nail fracture and roller surface defect detection, and specifically relates to a column nail fracture and roller surface defect timely diagnosis system and method based on deep learning. Background Art

[0002] In recent years, the domestic cement roller press roller surface technology has been continuously innovated. Various manufacturers have developed different roller surface forms, including: cladding roller, composite roller, column nail roller, etc., and various cement companies have also selected them according to their own working conditions. Among the above different forms of roller surface, the column nail roller has a stable extrusion effect, the extruded abrasive is easier to grind, the grinding efficiency of the mill is improved, and the specific surface area inside the mill is greatly increased.

[0003] However, during the working process, the nail roller may break, material layers may form between the nails, and the roller surface may fall off, which cannot meet the production needs. In the traditional detection of nail and roller surface defects, the entire roller can only be disassembled for manual observation, which takes a long time and is inefficient. It is an urgent need for cement companies to develop a timely diagnosis method for nail breakage and roller surface defects. Summary of the invention

[0004] In order to overcome the problems existing in the above prior art, the purpose of the present invention is to provide a timely diagnosis system and method for stud fracture and roller surface defects based on deep learning. The method collects roller surface images in real time through the system, detects and identifies roller surface defects through deep learning algorithms, and classifies the defects. The defective parts are displayed to the user through the PC-side visual interface in a timely manner, and have the advantages of high recognition efficiency, stable operation, high efficiency and practicality. The problem of relying only on manual and unable to detect roller surface defects in real time is solved.

[0005] In order to achieve the above object, the technical solution adopted by the present invention is:

[0006] A timely diagnosis system for stud fracture and roller surface defects based on deep learning, comprising an image acquisition device 1, an image processing unit 2, and a user operation unit 3;

[0007] The image acquisition device 1 is used to acquire the roller surface image in real time and transmit it to the image processing unit 2 for the next step;

[0008] The image processing unit 2 is used to perform defect detection and classification on the image collected by the image acquisition device 1 through a target detection algorithm;

[0009] The user operation unit 3 is used to control the movement of the image acquisition device 1 and observe roller surface defects.

[0010] The image acquisition device 1 includes an image acquisition unit 4 and a motion control unit 5;

[0011] The image acquisition unit 4 is used to capture the image of the roller surface through the industrial camera 9, and the motion control unit 5 is used to control the movement of the industrial camera 9 so as to capture the image of the entire roller surface.

[0012] The image acquisition device 1 is installed inside the roller press housing 12 and is used to acquire roller surface images in real time. The communication output end of the image acquisition unit 4 is connected to the communication input end of the image processing unit 2. The communication output end of the image processing unit 2 and the motion control unit 5 are connected to the communication input end of the user operation unit 3 placed on the PC end. The image processing unit 2 is used to perform target detection on the acquired image through a target detection algorithm to detect broken studs and roller surface defect areas. The user operation unit 3 classifies and counts the broken studs and various roller surface defects, controls the image acquisition device 1 to acquire images of different parts of the roller surface, and observes the roller surface status in real time.

[0013] The image acquisition device 1 includes an X-axis ball screw 6, a Y-axis ball screw 7, a slide 8, an industrial camera 9, a light source 10, and a device housing 11;

[0014] The industrial camera 9 and the light source 10 are installed above the slide 8, and the industrial camera 9 and the light source 10 are controlled to move along the X-axis and Y-axis directions on the slide 8 through the X-axis ball screw 6 and the Y-axis ball screw 7.

[0015] The image processing unit 2 includes an image preprocessing submodule and a roller surface defect detection and positioning submodule, and the image preprocessing submodule and the roller surface defect detection and positioning submodule are mounted on a computer and run.

[0016] The image preprocessing submodule is used to preprocess the roller surface image acquired by the image acquisition device 1, and the preprocessing includes image size scaling and image denoising;

[0017] The roller surface defect positioning submodule includes a model pre-training process and a real-time detection and positioning process.

[0018] The model pre-training process is specifically as follows:

[0019] a. When the roller surface is working normally, the image acquisition device 1 is used to collect pictures containing different defect categories, and the pictures of different defect categories include broken column nails, worn column nails, cracks on the column surface, and falling blocks;

[0020] A data set is formed including the cases where there are multiple defects in the same image and the defect area is partially blocked;

[0021] b. Label the data set: Label the categories, including normal column nails, roller surface fracture, column nail wear, column surface cracks, and block drop;

[0022] Mark the upper left corner coordinates and lower right corner coordinates of the defect location; divide the data set into a training set and a test set;

[0023] c. Perform Mosaic data enhancement on the data in the training set, and splice the four images by randomly scaling, randomly cropping, and randomly arranging them; then input them into the roller surface defect detection model for training.

[0024] d. Use the test set to detect the learning effect of the roller surface defect detection model.

[0025] The roller surface defect detection model adopts an improved YOLOv5s model, and the specific structure is as follows:

[0026] a. Add the EAC (Squeeze and Excitation) attention mechanism to the feature extraction network, and use one-dimensional convolution to achieve the interaction of local cross-channel features without dimensionality reduction, which greatly improves the network's ability to extract features and the network's detection accuracy, and reduces the interference of dust and other factors on the detection effect;

[0027] b. Use the improved BiFPN to replace the Neck module of YOLOv5s, perform multi-scale feature fusion, and improve the model's sensitivity to the target. The two fusion features of BiFPN are as follows:

[0028]

[0029]

[0030] Where: P6 td It is the intermediate feature of the sixth layer in the top-down path; P6 out It is the output feature of the 6th layer in the bottom-up path. On this basis, a deep decomposable convolution method is used, and batch normalization and activation functions are added after convolution to improve the effectiveness of the algorithm.

[0031] The real-time detection and positioning process includes inputting the real-time image captured by the image acquisition device 1 into a trained roller surface defect detection model to obtain a detection result. When the defect is detected, the position of the defect part on the image and the actual roller surface position are mapped to obtain the position coordinates of the defect on the roller surface, and the number of defects obtained from the detection result is counted.

[0032] The user operation unit 3 uses PYQT5 to write a user operation interface, including an image display area 12, a data statistics area 13, an image processing area 14, a camera control area 15, and a device control area 16;

[0033] The image display area 12 is used to display the image collected by the image acquisition device 1 in real time and to display the image after defect detection; the data statistics area 13 counts the types and quantities of various defects and records the defect information; the image processing area 14 is used to perform defect detection on the collected images and display the results in the image display area 12; the camera control area 15 is used to control the industrial camera to take pictures, as well as to turn it on and off, and the device control area 16 is used to control the movement of the image acquisition device, so as to collect images of different areas of the roller surface.

[0034] A method for using a timely diagnosis system for stud fracture and roller surface defects based on deep learning, specifically comprising the following steps:

[0035] Step 1: The image acquisition device 1 is initialized, the industrial camera 9 moves to one end of the roller surface, the industrial camera 9 and the light source 10 are adjusted to a suitable position away from the roller surface along the direction of the Y-axis ball screw 7, the industrial camera 9 and the light source 10 are turned on, the roller surface continues to rotate, and the industrial camera 9 starts to collect roller surface images. When the image acquisition within the field of view of the industrial camera 9 is completed, the industrial camera 9 moves a distance along the X-axis ball screw 6 and continues to collect roller surface images. According to the above steps, the industrial camera 9 collects images from one end of the roller surface to the other end, completing the acquisition of the entire roller surface image;

[0036] Step 2: The image acquisition unit 4 transmits the acquired image to the image processing unit 2, and uses the trained defect detection model to perform defect detection on the acquired roller surface image;

[0037] Step 3: The detected defects are classified by the defect location submodule in the image processing unit 2, and the quantity and coordinates are counted. When the number of defects reaches the value set by the system, the system generates an early warning message to remind the operator;

[0038] Step 4: Observe the detected defects in the user operation unit 3 to provide a reference for the subsequent roller surface repair work.

[0039] The step of observing defects in step 4 includes:

[0040] a. Control the industrial camera 9 to move in the X-axis direction to observe different positions of the roller surface through the user operation unit 3, control the industrial camera 9 to move in the Y-axis direction, change the distance between the camera and the roller surface, and further observe the roller surface image;

[0041] b. In step a, if the defects need to be saved, the defects are collected and saved through the user operation unit 3.

[0042] Beneficial effects of the present invention:

[0043] 1. The present invention's timely diagnosis system for stud fracture and roller surface defects based on deep learning has a simple structure, convenient installation, and easy operation. It can realize real-time acquisition of images of various parts of the roller surface, and can set manual and automatic control through the PC end.

[0044] 2. The defect data set used in the present invention can be expanded at any time through the step 2 of training the model, and the data set can be retrained to obtain a new defect detection model, that is, different types of roller surface defects can be detected.

[0045] 3. The image processing module of the present invention adopts an improved YOLOv5s network model, and adds the EAC attention mechanism to the feature network part, which greatly improves the network's ability to extract features and the network's detection accuracy.

[0046] 4. Using the improved BiFPN to replace the Neck module of YOLOv5s can make multi-scale feature fusion simpler and faster, and improve the model's sensitivity to the target. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] Figure 1 It is a structural block diagram of the present invention.

[0048] Figure 2 This is the image acquisition device designed in the present invention.

[0049] Figure 3 This is the installation position of the image acquisition device designed in the present invention.

[0050] Figure 4 This is the workflow of the image acquisition device in the present invention.

[0051] Figure 5 This is a flow chart of defect detection model training in the present invention.

[0052] Figure 6 Schematic diagram of the YOLOv5s network model structure used in the present invention.

[0053] Figure 7 Schematic diagram of the BiFPN and improved BiFPN structures provided in the present invention.

[0054] Figure 8 This is a schematic diagram of the network model structure in which the improved BiFPN in the invention replaces the Neck module of YOLOv5s. DETAILED DESCRIPTION

[0055] The present invention will be further described in detail below in conjunction with the accompanying drawings.

[0056] like Figure 1-Figure 8 As shown, a method for timely diagnosis of stud fracture and roller surface defects based on deep learning specifically includes the following steps:

[0057] Step 1, the image acquisition device 1 is initialized, the industrial camera 9 moves to one end of the roller surface, the industrial camera 9 and the light source are adjusted to a suitable position away from the roller surface along the direction of the Y guide rail, the industrial camera and the light source are turned on, the roller surface continues to rotate, and the industrial camera 9 starts to collect roller surface images. When the image acquisition within the field of view of the industrial camera 9 is completed, the industrial camera 9 moves a certain distance along the X axis and continues to collect roller surface images. According to the above steps, the industrial camera 9 collects from one end of the roller surface to the other end to complete the acquisition of the entire roller surface image.

[0058] Step 2: The image acquisition unit 4 transmits the acquired image to the image processing unit 2, and uses the trained defect detection model to perform defect detection on the acquired roller surface image.

[0059] Step 3: The defect location submodule in the image processing unit 2 classifies and locates the detected defects, and counts the number and coordinates of the defects. When the number of defects reaches the value set by the system, the system generates an early warning message to remind the operator.

[0060] Step 4: Observe the detected defects in the user operation unit 3 to provide a reference for the subsequent roller surface repair work.

[0061] The method for training the model in step 2 includes:

[0062] a. When the roller surface is working normally, collect pictures containing different defect categories, including broken studs, worn studs, cracks on the cylinder surface, and chipping; and form a data set including the situation where there are multiple defects in the same picture and the defect area is partially blocked.

[0063] b. Label the data set: Label the categories, including normal studs, roller surface fractures, stud wear, cylinder surface cracks, and block drop. Label the upper left corner coordinates and lower right corner coordinates of the defect location; divide the data set into a training set and a test set;

[0064] c. Perform Mosaic data enhancement on the data in the training set, and splice the four images by randomly scaling, randomly cropping, and randomly arranging them; then input them into the roller surface defect detection model for training.

[0065] d. Use the test set to detect the learning effect of the roller surface defect detection model.

[0066] like Figure 6-Figure 8 As shown, the roller surface defect detection model adopts an improved YOLOv5s model, and the specific structure is:

[0067] a. Add the EAC (Squeeze and Excitation) attention mechanism to the feature extraction network, and realize the interaction of local cross-channel features without dimensionality reduction through one-dimensional convolution, which greatly improves the network's ability to extract features and the network's detection accuracy. Reduce the interference of dust and other factors on the detection effect.

[0068] b. Using the improved BiFPN to replace the Neck module of YOLOv5s can make multi-scale feature fusion simpler and faster, and improve the model's sensitivity to the target. The two fusion features of BiFPN are as follows:

[0069]

[0070] Where: P6 td It is the intermediate feature of the sixth layer in the top-down path; P6 out It is the output feature of the 6th layer in the bottom-up path. On this basis, a deep decomposable convolution method is used, and batch normalization and activation function are added after convolution, which further improves the effectiveness of the algorithm.

[0071] The step of observing defects in step 4 includes:

[0072] a. The user operating unit 3 controls the camera to move in the X-axis direction to observe different positions of the roller surface, controls the camera to move in the Y-axis direction, changes the distance between the camera and the roller surface, and realizes further observation of the roller surface image.

[0073] b. In step a, if the defects need to be saved, they can be collected and saved through the user operation unit 4.

[0074] like Figure 1 The present invention is shown to be composed of three parts, namely, an image acquisition device 1, an image processing unit 2 and a user operation unit 3. The user operation unit 3 is installed on the PC side and includes an image display area 12, an image processing area 14, a camera control area 15, a device control area 16 and a data statistics area 13. The three parts cooperate with each other to complete the tasks of roller surface image acquisition, defect detection, defect classification, statistics and the like.

[0075] like Figure 2 The image acquisition device 1 designed by the present invention is shown to be composed of an X-axis ball screw 6, a Y-axis ball screw 7, a slide 8, an industrial camera 9, a light source 10, and a device housing 11. The image of the roller surface is acquired by controlling the industrial camera 9 to move on the X-axis ball screw 6 and the Y-axis ball screw 7.

[0076] like Figure 3The image acquisition device is shown as being installed in the roller press, which is installed inside the roller press housing, in front of the feed port, so that the industrial camera 9 can acquire the roller surface image as best as possible.

[0077] like Figure 4 The figure shows the working process of the image acquisition device 1. When the working roller of the roller press rotates, the industrial camera 9 moves to one end of the device to shoot the roller surface and collect images. As the roller surface rotates, after the industrial camera 9 collects the roller surface image in the field of view, it moves along the X-axis ball screw 6 to collect the roller surface image of the next field of view until it moves to the other end of the device to complete the image collection of the entire roller surface.

[0078] like Figure 5 The defect detection model training flow chart in the present invention is shown. The image acquisition device 1 is used to collect roller surface defect images, mark the defects in the images, and make a data set. The network structure is improved using the Yolov5s framework, and after the defect detection model is obtained, it is verified whether it can meet the working requirements. If the accuracy requirements are not met, the model is improved and verified again until the roller surface defect detection model is output after the accuracy requirements are met.

[0079] like Figure 6 The figure shows the schematic diagram of the Yolov5s network structure used in the present invention, which consists of four parts: input, backbone, neck and head.

[0080] like Figure 7 The schematic diagram of the BiFPN and improved BiFPN structure provided in the present invention is shown. Compared with the original BiFPN, a 3-layer multi-scale feature map input is designed to accurately identify the output of the YOLOv5s backbone network.

[0081] like Figure 8 The figure shows that the improved BiFPN is used to replace the Neck module of YOLOv5s in the invention.

[0082] The roller surface defect detection model adopts an improved YOLOv5s model, and the specific structure is as follows:

[0083] a. Add the EAC (Squeeze and Excitation) attention mechanism to the feature extraction network, and use one-dimensional convolution to achieve the interaction of local cross-channel features without dimensionality reduction, which greatly improves the network's ability to extract features and the network's detection accuracy, and reduces the interference of dust and other factors on the detection effect;

[0084] b. Use the improved BiFPN to replace the Neck module of YOLOv5s, perform multi-scale feature fusion, and improve the model's sensitivity to the target. The two fusion features of BiFPN are as follows:

[0085]

[0086]

[0087] Where: P6 td It is the intermediate feature of the sixth layer in the top-down path; P6 out It is the output feature of the 6th layer in the bottom-up path. On this basis, a deep decomposable convolution method is used, and batch normalization and activation functions are added after convolution to improve the effectiveness of the algorithm.

Claims

1. A timely diagnosis system for stud fracture and roller surface defects based on deep learning, characterized in that: It comprises an image acquisition device (1), an image processing unit (2), and a user operation unit (3); The image acquisition device (1) is used to acquire the roller surface image in real time and transmit it to the image processing unit (2) for further processing; The image processing unit (2) is used to perform defect detection and classification on the image collected by the image collection device (1) through a target detection algorithm; The user operation unit (3) is used to control the movement of the image acquisition device (1) and observe roller surface defects.

2. According to the deep learning-based timely diagnosis system for stud fracture and roller surface defects of claim 1, it is characterized in that: The image acquisition device (1) comprises an image acquisition unit (4) and a motion control unit (5); The image acquisition unit (4) is used to capture the image of the roller surface through an industrial camera (9), and the motion control unit (5) is used to control the movement of the industrial camera (9) so as to capture the image of the entire roller surface.

3. The deep learning-based timely diagnosis system for stud fracture and roller surface defects according to claim 1 is characterized in that: The image acquisition device (1) is installed inside the roller press housing (12) and is used to acquire roller surface images in real time. The communication output end of the image acquisition unit (4) is connected to the communication input end of the image processing unit (2). The communication output end of the image processing unit (2) and the motion control unit (5) are connected to the communication input end of the user operation unit (3) placed on the PC. The image processing unit (2) is used to perform target detection on the acquired image through a target detection algorithm to detect broken studs and roller surface defect areas. The user operation unit (3) classifies and counts broken studs and various roller surface defects, controls the image acquisition device (1) to acquire images of different parts of the roller surface, and observes the roller surface state in real time. The image acquisition device (1) comprises an X-axis ball screw (6), a Y-axis ball screw (7), a slide table (8), an industrial camera (9), a light source (10), and a device housing (11); The industrial camera (9) and the light source (10) are installed above the slide table (8), and the industrial camera (9) and the light source (10) are controlled to move along the X-axis and Y-axis directions on the slide table (8) through an X-axis ball screw (6) and a Y-axis ball screw (7).

4. The deep learning-based timely diagnosis system for stud fracture and roller surface defects according to claim 1 is characterized in that: The image processing unit (2) comprises an image preprocessing submodule and a roller surface defect detection and positioning submodule, and the image preprocessing submodule and the roller surface defect detection and positioning submodule are mounted on a computer and run.

5. The timely diagnosis system for stud fracture and roller surface defects based on deep learning according to claim 4 is characterized in that: The image preprocessing submodule is used to preprocess the roller surface image acquired by the image acquisition device (1), wherein the preprocessing includes image size scaling and image denoising; The roller surface defect positioning submodule includes a model pre-training process and a real-time detection and positioning process.

6. A timely diagnosis system for stud fracture and roller surface defects based on deep learning according to claim 5, characterized in that: The model pre-training process is specifically as follows: a. When the roller surface is working normally, the image acquisition device (1) collects pictures containing different defect categories, and the pictures of different defect categories include broken column nails, worn column nails, column surface cracks, and falling blocks; A data set is formed including the cases where there are multiple defects in the same image and the defect area is partially blocked; b. Label the data set: Label the categories, including normal column nails, roller surface fracture, column nail wear, column surface cracks, and block drop; Mark the upper left corner coordinates and lower right corner coordinates of the defect location; Divide the dataset into training and testing sets; c. Perform Mosaic data enhancement on the data in the training set, and splice the four images by randomly scaling, randomly cropping, and randomly arranging them; then input them into the roller surface defect detection model for training. d. Use the test set to detect the learning effect of the roller surface defect detection model.

7. A timely diagnosis system for stud fracture and roller surface defects based on deep learning according to claim 6, characterized in that: The roller surface defect detection model adopts an improved YOLOv5s model, and the specific structure is as follows: a. Add the EAC (Squeeze and Excitation) attention mechanism to the feature extraction network, and use one-dimensional convolution to achieve the interaction of local cross-channel features without dimensionality reduction, which greatly improves the network's ability to extract features and the network's detection accuracy, and reduces the interference of dust and other factors on the detection effect; b. Use the improved BiFPN to replace the Neck module of YOLOv5s, perform multi-scale feature fusion, and improve the model's sensitivity to the target. The two fusion features of BiFPN are as follows: Where: P6 td It is the intermediate feature of the sixth layer in the top-down path; P6 out It is the output feature of the sixth layer in the bottom-up path. On this basis, a deep decomposable convolution method is used, and batch normalization and activation functions are added after convolution to improve the effectiveness of the algorithm. The real-time detection and positioning process includes inputting the real-time image captured by the image acquisition device 1 into a trained roller surface defect detection model to obtain a detection result. When the defect is detected, the position of the defect part on the image and the actual roller surface position are mapped to obtain the position coordinates of the defect on the roller surface, and the number of defects obtained from the detection result is counted.

8. The timely diagnosis system for stud fracture and roller surface defects based on deep learning according to claim 1 is characterized in that: The user operation unit (3) uses PYQT5 to program a user operation interface, which includes an image display area (12), a data statistics area (13), an image processing area (14), a camera control area (15), and a device control area (16); The image display area (12) is used to display in real time the image collected by the image collection device (1) and the image after defect detection; the data statistics area (13) counts the types and quantities of various defects and records the defect information; the image processing area (14) is used to perform defect detection on the collected images and display the obtained results in the image display area (12); the camera control area (15) is used to control the industrial camera to take pictures and turn it on and off; the device control area (16) is used to control the movement of the image collection device, so as to collect images of different areas of the roller surface.

9. A method for using a deep learning-based timely diagnosis system for stud fracture and roller surface defects according to any one of claims 1 to 8, characterized in that: The specific steps include: Step 1: The image acquisition device (1) is initialized, the industrial camera (9) moves to one end of the roller surface, the industrial camera (9) and the light source (10) are adjusted to a suitable position away from the roller surface along the direction of the Y-axis ball screw (7), the industrial camera (9) and the light source (10) are turned on, the roller surface continues to rotate, and the industrial camera (9) starts to acquire the roller surface image. When the image acquisition within the field of view of the industrial camera (9) is completed, the industrial camera (9) moves a certain distance along the X-axis ball screw (6) and continues to acquire the roller surface image. According to the above steps, the industrial camera (9) acquires images from one end of the roller surface to the other end, completing the acquisition of the entire roller surface image; Step 2: The image acquisition unit (4) transmits the acquired image to the image processing unit (2), and uses the trained defect detection model to perform defect detection on the acquired roller surface image; Step 3: The defect location submodule in the image processing unit (2) classifies the detected defects and counts the number and coordinates. When the number of defects reaches a value set by the system, the system generates an early warning message to remind the operator; Step 4: Observe the detected defects in the user operation unit (3) to provide a reference for the subsequent roller surface repair work.

10. The method for using the deep learning-based timely diagnosis system for stud fracture and roller surface defects according to claim 9, characterized in that: The step of observing defects in step 4 includes: a. Control the industrial camera (9) to move in the X-axis direction to observe different positions of the roller surface through the user operation unit (3), control the industrial camera (9) to move in the Y-axis direction, change the distance between the camera and the roller surface, and further observe the roller surface image; b. In step a, if the defects need to be saved, they are collected and saved through the user operation unit (3).

Citation Information

Cited By

  • Online intelligent roller surface stud detection method based on three-dimensional scanning and reference template

    CN122573869A