Train wheel full-surface defect detection method, equipment and medium

Abstract

The invention discloses a train wheel full-surface defect detection method and device and a medium, and the method comprises the steps: combining a dome light source and a coaxial light source into a side light source, installing the side light source and a side industrial camera on an industrial robot end effector, installing a reflection type coaxial light source and a bottom industrial camera on a stepping motor, controlling a side industrial camera to shoot the side surface of the train wheel under the uniform irradiation of the side light source, and controlling a bottom industrial camera to shoot the axle hole of the train wheel under the uniform irradiation of the reflective coaxial light source so as to obtain a train wheel surface image comprising a hub, a radial plate, a rim and the axle hole, the train wheel surface image is combined with the hybrid supervised convolutional neural network model to carry out surface defect visual identification so as to find all defects on the train wheel surface and evaluate the overall quality of the train wheel, so that the overall fine defect detection on the train wheel surface is realized, the train wheel surface defect detection difficulty is reduced, and the detection efficiency is improved.

Description

Technical Field

[0001] The present invention relates to the application fields of machine vision, defect detection, image data and train wheel detection, and in particular to a method, equipment and medium for detecting full-surface defects of train wheels. Background Art

[0002] Train wheel surface defect detection technology is an important part of railway transportation safety. It aims to identify and evaluate wheel surface damage or defects through effective means to prevent accidents caused by wheel failure. With the rapid development of high-speed railways and the increase in operating frequency, traditional manual inspection methods can no longer meet the needs of efficiency and accuracy. Therefore, the use of advanced detection technology has become an urgent need for the industry. At present, common train wheel defect detection technologies include ultrasonic testing, magnetic particle testing, eddy current testing, and visual image processing technology. Ultrasonic testing can penetrate deep into the wheel to find hidden cracks; magnetic particle testing is mainly used to detect tiny cracks on the wheel surface; eddy current testing technology is suitable for non-destructive testing of high-precision surface defects; and visual image processing technology uses high-speed cameras and image processing algorithms to monitor the surface status of the wheel in real time and automatically identify possible defects.

[0003] These technologies, combined with artificial intelligence and automation systems, can achieve all-weather, all-round wheel monitoring without interfering with normal operations, effectively improving the accuracy and efficiency of wheel defect detection and ensuring the safety of railway transportation.

[0004] Due to the variety of train wheel models, the surface conditions of train wheels are complex. Currently, surface defects of train wheels are mainly judged by the naked eye. General visual inspection methods are difficult to meet the requirements of surface defect detection of all models of train wheels. Summary of the invention

[0005] In order to overcome the shortcomings of the prior art, the present invention provides a method, equipment and medium for detecting defects on the entire surface of a train wheel, so as to completely detect defects on the entire surface of the train wheel, eliminate the reflective effect caused by the smooth surface of the train wheel, solve the problems of unclear images and blind spots in shooting caused by changes in train wheel models, and achieve full-surface image shooting, detection and analysis of train wheels, thereby reducing the false detection rate of train wheel surface defects and improving the accuracy of defect detection.

[0006] In order to achieve the above-mentioned purpose, the present invention adopts the following technical scheme:

[0007] The present invention provides a method for detecting defects on the entire surface of a railway wheel, which comprises the following steps:

[0008] Step 1, placing the train wheels on the rubber pad position above the rotating base, an industrial robot and a stepper motor are respectively arranged on the left side and the bottom of the rotating base, a side industrial camera, a coaxial light source and a dome light source are installed on the end effector of the industrial robot, the coaxial light source and the dome light source are combined into a side light source, and a reflective coaxial light source, a servo motor and a bottom industrial camera are arranged on the stepper motor;

[0009] Step 2: The clamping devices on both sides of the rotating base push the train wheel to the detection position, and the rotating base drives the train wheel to rotate at a uniform speed. The host computer calculates the detection motion path of the industrial robot according to the CAD processing file of the train wheel, and when it is detected that the industrial robot reaches the detection position, controls the side industrial camera to shoot the side of the train wheel under the uniform illumination of the side light source, so as to obtain a side image of the train wheel including the hub, the spoke plate and the rim;

[0010] Step 3, the host computer calculates the diameter of the axle hole of the train wheel according to the CAD processing file of the train wheel, and controls the servo motor to adjust the working distance between the bottom industrial camera and the axle hole of the train wheel according to the axle hole diameter, and when it is detected that the stepper motor moves to the detection position, controls the bottom industrial camera to shoot the axle hole of the train wheel under the uniform illumination of the reflective coaxial light source, so as to obtain an image of the axle hole of the train wheel including the inner wall of the axle hole;

[0011] The train wheel side image and the train wheel axle hole image are used to form a train wheel surface image and the image is divided into a training set and a test set, wherein the training set contains a train wheel surface image without defects and a train wheel surface image with defects;

[0012] Step 4, the training set is input into the hybrid supervised convolutional neural network model, and the local features and global features of the train wheel surface image are extracted through multi-layer convolution and pooling operations, so as to finally output the defect prediction category and compare it with the real defect category, so as to construct a loss function for training the convolutional neural network model, and obtain the trained train wheel full surface defect detection model;

[0013] The test set is input into the trained train wheel full surface defect detection model for processing, and the optimal defect prediction category is output. After the optimal defect prediction category is marked as false positive or false negative, it is used as a new training set to update the trained train wheel full surface defect detection model, thereby obtaining the optimal train wheel full surface defect detection model, which is used to identify the train wheel side image and the train wheel axle hole image to find defects on the train wheel surface.

[0014] The train wheel full surface defect detection method of the present invention is also characterized in that the coaxial light source is installed at the upper observation port of the dome light source, the side industrial camera captures the side image of the train wheel through the observation port of the coaxial light source, and the uniform illumination of the side light source is achieved according to the following process:

[0015] Step a, the side industrial camera captures a uniform white paper image through the side light source;

[0016] Step b: Calculate the brightness value deviation L using formula (2.1) D :

[0017] L D =L C -L E (2.1)

[0018] In formula (2.1), L C represents the average brightness of the central area of ​​the uniform white paper image, L E represents the average brightness of the edge area of ​​the uniform white paper image;

[0019] Step c: according to the brightness value deviation L D , setting the luminous intensity of the coaxial light source to achieve uniform illumination when photographing the side of the train wheel.

[0020] Furthermore, a through circular hole is provided at the center of the rotating base, and the reflective coaxial light source, the servo motor, the bottom industrial camera and the center of the rotating base are coaxially arranged;

[0021] The angle between the semi-transparent and semi-reflective lens of the reflective coaxial light source and the tangent line of the train wheel axle hole is set to 45 degrees, and the light panel of the reflective coaxial light source is set to be parallel to the tangent line of the train wheel axle hole.

[0022] The electronic device of the present invention includes a memory and a processor, wherein the memory is used to store a program that supports the processor to execute the train wheel full surface defect detection method, and the processor is configured to execute the program stored in the memory.

[0023] The present invention provides a computer-readable storage medium, and the computer-readable storage medium stores a computer program, which is characterized in that when the computer program is run by a processor, the steps of the train wheel full-surface defect detection method are executed.

[0024] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0025] 1. The present invention calculates the detection path of the industrial robot through the two-dimensional CAD processing file of the train wheel. The industrial robot drives the industrial camera and the side light source to collect the uniformly illuminated image of the side of the train wheel, which can effectively avoid the collision between the train wheel and the visual sensor, adapt to various specifications and models of train wheels, and effectively capture the image of the entire area of ​​the side of the train wheel, thereby achieving zero missed detection of surface defects.

[0026] 2. The present invention calculates the diameter of the train wheel axle hole through the two-dimensional CAD processing file of the train wheel, and adjusts the working distance between the industrial camera and the train wheel axle hole through a servo motor, thereby ensuring that the train wheel axle holes of different diameters are imaged within the focusing area of ​​the industrial camera. The industrial camera collects the image of the inner wall of the train wheel axle hole through a reflective coaxial light source, thereby enhancing the clarity of the image features of the inner wall of the train wheel axle hole, thereby improving the accuracy of surface defect detection of the train wheel axle hole.

[0027] 3. Through multiple rounds of manual fine-tuning and model iteration, the present invention realizes a hybrid supervised defect detection system, combines the advantages of manual correction and machine learning models, improves the accuracy and efficiency of train wheel surface defect detection, and provides reliable technical support for quality control on the production line. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] Figure 1 Schematic diagram of the whole surface defect detection process of a train wheel according to the present invention;

[0029] Figure 2 It is a structural schematic diagram of a train wheel side defect detection system of the present invention;

[0030] Figure 3 It is a structural schematic diagram of the inner wall defect detection system of the train wheel axle hole of the present invention;

[0031] Numbers in the figure: 1 industrial robot; 2 side industrial camera; 3 coaxial light source; 4 dome light source; 5 train wheel; 6 rotating base; 7 reflective coaxial light source; 8 stepper motor; 9 servo motor; 10 bottom industrial camera. DETAILED DESCRIPTION

[0032] In this embodiment, a method for detecting defects on the entire surface of a train wheel is implemented as follows: Figure 1 As shown, it is applied in the train wheel surface defect detection system to complete tasks such as wheel surface defect marking and wheel processing quality inspection.

[0033] Among them, the train wheel surface defect detection system is composed of a side optical system, a bottom optical system and a computer. The side optical system collects images of the train wheel hub, spoke plate and rim, and the bottom optical system collects images of the train wheel axle hole.

[0034] The host computer uses a hybrid supervised defect detection model to detect and analyze the image, automatically annotate the surface defects of the train wheels, manually check and adjust the annotation results, eliminate false positives or false negatives, and iteratively update the hybrid supervised defect detection model, so as to accurately identify and annotate the surface defects of the train wheels.

[0035] In this embodiment, a method for detecting the full surface defects of a train wheel comprises a side optical system and a bottom optical system, such as Figure 2 As shown, the side optical system is applied to a detection scene composed of an industrial robot 1, a side industrial camera 2, a side light source, a train wheel 5 and a rotating base 6. The side light source includes a coaxial light source 3 and a dome light source 4. The industrial robot 1 drives the side industrial camera 2 and the side light source to collect images of the hub, spoke and rim of the train wheel 5; Figure 3 As shown, the bottom optical system is applied to a detection scene composed of a train wheel 5, a rotating base 6, a reflective coaxial light source 7, a stepper motor 8, a servo motor 9 and a bottom industrial camera 10. The stepper motor 8 drives the bottom industrial camera 10 and the reflective coaxial light source 7 to collect the axial hole image of the train wheel 5; the images of the hub, spoke and rim of the train wheel 5 and the axial hole image of the train wheel 5 constitute the surface image of the train wheel 5, as shown in FIG. Figure 1 As shown, the train wheel full surface defect detection method is performed in the following steps:

[0036] S1. Place the train wheel 5 on the rubber pad position above the rotating base 6. An industrial robot 1 and a stepper motor 8 are respectively arranged on the left side and the bottom of the rotating base 6. A side industrial camera 2, a coaxial light source 3 and a dome light source 4 are installed on the end effector of the industrial robot 1. The coaxial light source 3 and the dome light source 4 are combined into a side light source. A reflective coaxial light source 7, a servo motor 9 and a bottom industrial camera 10 are arranged on the stepper motor 8.

[0037] A through circular hole is provided at the center of the rotating base 6, and the reflective coaxial light source 7, the servo motor 9, the bottom industrial camera 10 and the center of the rotating base 6 are coaxially arranged;

[0038] The angle between the semi-transparent and semi-reflective lens of the reflective coaxial light source 7 and the tangent line of the axial hole of the train wheel 5 is 45 degrees, and the light board of the reflective coaxial light source 7 is parallel to the tangent line of the axial hole of the train wheel 5.

[0039] S2. The coaxial light source 3 is installed at the observation port above the dome light source 4. The side industrial camera 2 captures the side image of the train wheel 5 through the observation port of the coaxial light source 3, and realizes uniform illumination of the side light source according to the following process:

[0040] Step a, the side industrial camera 2 captures a uniform white paper image through a side light source;

[0041] Step b: Calculate the brightness value deviation L using formula (2.1) D :

[0042] L D =L C -L E (2.1)

[0043] In formula (2.1), L C represents the average brightness of the central area of ​​the uniform white paper image, L E represents the average brightness of the edge area of ​​the uniform white paper image;

[0044] Step c: According to the brightness value deviation L D , setting the luminous intensity of the coaxial light source 3 to achieve uniform illumination when photographing the side of the train wheel 5.

[0045] S3, the clamping devices on both sides of the rotating base 6 push the train wheel 5 to the detection position, and the rotating base 6 drives the train wheel 5 to rotate at a uniform speed, and the upper computer calculates the detection motion path of the industrial robot 1 according to the CAD processing file of the train wheel 5, and when it is detected that the industrial robot 1 reaches the detection position, the side industrial camera 2 is controlled to shoot the side of the train wheel 5 under the uniform illumination of the side light source, so as to obtain the side image of the train wheel including the hub, the spoke plate and the rim;

[0046] S4, the upper computer calculates the diameter of the shaft hole of the train wheel 5 according to the CAD processing file of the train wheel 5, and controls the servo motor 9 to adjust the working distance between the bottom industrial camera 10 and the shaft hole of the train wheel 5 according to the shaft hole diameter. When it is detected that the stepper motor 8 moves to the detection position, the bottom industrial camera 10 is controlled to shoot the shaft hole of the train wheel 5 under the uniform illumination of the reflective coaxial light source 7 to obtain the train wheel shaft hole image including the inner wall of the shaft hole;

[0047] The train wheel surface images are composed of train wheel side images and train wheel axle hole images and are divided into a training set and a test set. The training set contains defect-free train wheel surface images and defective train wheel surface images, among which the defective train wheel surface images are obtained by artificially simulating various defects on the train wheel surface and synthesizing them with conventional defect-free train wheel surface images. These defect images contain different defect types and sizes.

[0048] S5. Construct a hybrid supervised convolutional neural network model in the host computer. The model consists of detectors and supervisory signals. The detectors include: YOLO, SSD, RetinaNet and Faster RCNN. The supervisory signals include supervised signals, unsupervised signals and weakly supervised signals.

[0049] The training set is input into the mixed supervised convolutional neural network model. First, the model is pre-trained using unsupervised signals and detectors, and the local and global features of the train wheel surface image are extracted through multi-layer convolution and pooling operations, so that the model can recognize the basic features in the train wheel surface image; then the model is fine-tuned through supervised signals, detectors and pixel-level masks to achieve recognition of specific target categories based on basic features; finally, the weak supervision signal is used to replace the pixel-level mask with the border-level label to reduce the model annotation requirements, so as to finally output the defect prediction category and compare it with the real defect category, so as to construct a loss function for training the convolutional neural network model, and obtain the trained train wheel full surface defect detection model;

[0050] The test set is input into the trained train wheel full surface defect detection model for processing, and the optimal defect prediction category is output. After the optimal defect prediction category is labeled as false positive or false negative, it is used as a new training set to update the trained train wheel full surface defect detection model, thereby obtaining the optimal train wheel full surface defect detection model, which is used to identify the train wheel side image and the train wheel axle hole image to detect defects on the train wheel surface.

[0051] In this embodiment, an electronic device includes a memory and a processor, wherein the memory is used to store a program that supports the processor to execute the above method, and the processor is configured to execute the program stored in the memory.

[0052] In this embodiment, a computer-readable storage medium stores a computer program on the computer-readable storage medium, and the computer program executes the steps of the above method when executed by a processor.

Claims

1. A method for detecting defects on the entire surface of a train wheel, characterized in that: The steps include: Step 1, placing a train wheel (5) at a rubber pad position above a rotating base (6), an industrial robot (1) and a stepper motor (8) are respectively arranged on the left side and the bottom of the rotating base (6), a side industrial camera (2), a coaxial light source (3) and a dome light source (4) are installed on the end effector of the industrial robot (1), and the coaxial light source (3) and the dome light source (4) are combined into a side light source, and a reflective coaxial light source (7), a servo motor (9) and a bottom industrial camera (10) are arranged on the stepper motor (8); Step 2: The clamping devices on both sides of the rotating base (6) push the train wheel (5) to the detection position, and the rotating base (6) drives the train wheel (5) to rotate at a uniform speed. The host computer calculates the detection motion path of the industrial robot (1) according to the CAD processing file of the train wheel (5), and when it is detected that the industrial robot (1) reaches the detection position, the side industrial camera (2) is controlled to shoot the side of the train wheel (5) under the uniform illumination of the side light source, so as to obtain a side image of the train wheel including the hub, the spoke plate and the rim; Step 3, the host computer calculates the diameter of the axial hole of the train wheel (5) according to the CAD processing file of the train wheel (5), and controls the servo motor (9) to adjust the working distance between the bottom industrial camera (10) and the axial hole of the train wheel (5) according to the diameter of the axial hole. When it is detected that the stepper motor (8) moves to the detection position, the host computer controls the bottom industrial camera (10) to shoot the axial hole of the train wheel (5) uniformly illuminated by the reflective coaxial light source (7), so as to obtain an image of the train wheel axial hole including the inner wall of the axial hole; The train wheel side image and the train wheel axle hole image are used to form a train wheel surface image and the image is divided into a training set and a test set, wherein the training set contains a train wheel surface image without defects and a train wheel surface image with defects; Step 4, the training set is input into the hybrid supervised convolutional neural network model, and the local features and global features of the train wheel surface image are extracted through multi-layer convolution and pooling operations, so as to finally output the defect prediction category and compare it with the real defect category, so as to construct a loss function for training the convolutional neural network model, and obtain the trained train wheel full surface defect detection model; The test set is input into the trained train wheel full surface defect detection model for processing, and the optimal defect prediction category is output. After the optimal defect prediction category is marked as false positive or false negative, it is used as a new training set to update the trained train wheel full surface defect detection model, thereby obtaining the optimal train wheel full surface defect detection model, which is used to identify the train wheel side image and the train wheel axle hole image to find defects on the train wheel surface.

2. The method for detecting full-surface defects of railway wheels according to claim 1, characterized in that: The coaxial light source (3) is installed at the observation port above the dome light source (4), and the side industrial camera (2) captures the side image of the train wheel (5) through the observation port of the coaxial light source (3), and realizes uniform illumination of the side light source according to the following process: Step a, the side industrial camera (2) captures a uniform white paper image through the side light source; Step b: Calculate the brightness value deviation L using formula (2.1) D : L D =L C -L E (2.1) In formula (2.1), L C represents the average brightness of the central area of ​​the uniform white paper image, L E represents the average brightness of the edge area of ​​the uniform white paper image; Step c: according to the brightness value deviation L D , setting the luminous intensity of the coaxial light source (3) to achieve uniform illumination when photographing the side of the train wheel (5).

3. The method for detecting full-surface defects of railway wheels according to claim 1, characterized in that: A through circular hole is provided at the center of the rotating base (6), and the reflective coaxial light source (7), the servo motor (9), the bottom industrial camera (10) and the center of the rotating base (6) are coaxially arranged; The angle between the semi-transparent and semi-reflective lens of the reflective coaxial light source (7) and the tangent line of the axial hole of the train wheel (5) is set to be 45 degrees, and the light board of the reflective coaxial light source (7) is set to be parallel to the tangent line of the axial hole of the train wheel (5).

4. An electronic device, comprising a memory and a processor, characterized in that: The memory is used to store a program that supports the processor to execute the railway wheel full-surface defect detection method as described in any one of claims 1-3, and the processor is configured to execute the program stored in the memory.

5. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method for detecting full-surface defects of railway wheels as claimed in any one of claims 1 to 3 are executed.

Images