Wheel rolling profile online detection and self-adaptive correction method and system based on machine vision

Abstract

The present application relates to the field of wheel rolling deviation correction, and particularly relates to a wheel rolling profile online detection and self-adaptive deviation correction method and system based on machine vision. The method first performs edge detection on a gray image of a wheel to be measured in a rolling process to obtain initial edge points, performs region growing on a thermal infrared image to obtain a binary mask of pixel points in the thermal infrared image, obtains an edge feature vector of the initial edge points according to temperature values, gradient values, the binary mask of the initial edge points in the thermal infrared image, and a distribution of local gray values of the initial edge points in the gray image, trains a classifier using the edge feature vector, screens out actual profile points through a profile probability value obtained, and then performs registration on actual profile points of the wheel to be measured and a standard profile model, and performs deviation correction using a radial deviation amount. The present application can avoid interference of iron oxide scales on a wheel surface in the rolling process on wheel profile detection, and accurately realizes deviation correction of the wheel profile.

Description

Technical Field

[0001] This invention relates to the field of wheel rolling correction, specifically to a machine vision-based online detection and adaptive correction method and system for wheel rolling profiles. Background Technology

[0002] Wheel rolling is a key process in the manufacturing of rail transit equipment. It is an advanced metal plastic forming process, specifically referring to the precision forging and rolling technology used to manufacture wheels for rail transit vehicles such as trains, high-speed trains, and subways. However, there is a risk of wheel misalignment (eccentricity) during the rolling process. This is mainly caused by fluctuations in materials, equipment, and process parameters. This misalignment is related to the core quality indicators of wheel safety and service life. Therefore, it is necessary to eliminate misalignment to the greatest extent possible during the wheel rolling process to produce wheels with extremely high geometric precision.

[0003] In related technologies, optical or laser-based non-contact detection equipment is typically used to perform online contour detection of wheels during the rolling process. Based on the deviation between the detected contour and the standard contour, the control system parameters of wheel rolling (such as roll reduction) are adjusted to correct the wheel contour. However, since wheel rolling is usually carried out in a high-temperature environment, the formation of iron oxide scale is an unavoidable physicochemical phenomenon. The presence of iron oxide scale can be misidentified as wheel contour, resulting in low accuracy of wheel contour detection during the rolling process, and thus failing to accurately correct the wheel contour. Summary of the Invention

[0004] To address the technical problem that iron oxide scale during wheel rolling can be misidentified as wheel contour, leading to low wheel contour detection accuracy and thus hindering precise wheel contour correction, this invention aims to provide a machine vision-based online detection and adaptive correction method and system for wheel rolling contours. The specific technical solution adopted is as follows:

[0005] This invention proposes a machine vision-based online detection and adaptive correction method for wheel rolling profiles, the method comprising:

[0006] Real-time acquisition of grayscale and thermal infrared images of the wheel under test during the rolling process;

[0007] Edge detection is performed on the grayscale image to obtain initial edge points. Based on the temperature value distribution of pixels in different regions of the thermal infrared image, region growing is performed on the thermal infrared image to obtain a binary mask of pixels in the thermal infrared image. The initial edge points in the grayscale image are mapped to the thermal infrared image, and the edge feature vector of each initial edge point is obtained based on the temperature value, gradient value, binary mask of the initial edge point in the thermal infrared image, and the distribution of grayscale values ​​of pixels in the preset neighborhood of the initial edge point in the grayscale image.

[0008] Using samples of wheel rolling, a classifier is trained, and the edge feature vector of each initial edge point is input into the trained classifier to obtain the contour probability value of each initial edge point; based on the contour probability value, the actual contour points of the wheel to be tested are selected from all initial edge points.

[0009] The actual contour points of the wheel under test are registered with the standard contour model to obtain the radial deviation of the wheel under test; based on the radial deviation, the rollers are adjusted to correct the deviation.

[0010] Furthermore, obtaining the binary mask for pixels in the thermal infrared image includes:

[0011] Calculate the average and standard deviation of the temperature values ​​of all pixels in the thermal infrared image, and use the sum of the average and three times the standard deviation as the high-temperature oxidation threshold;

[0012] The thermal infrared image is evenly divided into multiple blocks, and the average temperature value of all pixels in each block is taken as the local temperature value of each block; the block with the local temperature value being less than the high-temperature oxidation threshold is taken as the base block of the wheel under test.

[0013] The average value of the local temperature values ​​of all the substrate blocks is taken as the substrate temperature of the wheel under test, and the sum of the substrate temperature and the preset temperature adjustment value is taken as the seed point screening threshold.

[0014] In the thermal infrared image, pixels with a temperature value greater than the seed point screening threshold are selected as seed points, and region growing is performed on each seed point to obtain a binary mask of the pixels in the thermal infrared image. The binary mask of each pixel in the thermal infrared image has a value of 0 or 1.

[0015] Further, obtaining the edge feature vector of each initial edge point includes:

[0016] The gradient operator is used to process the thermal infrared image to obtain the gradient value of each pixel in the thermal infrared image;

[0017] For each initial edge point in a grayscale image, calculate the grayscale co-occurrence matrix of each initial edge point in the preset neighborhood of the pixel, and use the entropy value of the grayscale co-occurrence matrix of each initial edge point as the local texture complexity of each initial edge point;

[0018] The four-dimensional vector composed of the temperature value, gradient value, binary mask, and local texture complexity of each initial edge point in the thermal infrared image is used as the edge feature vector of each initial edge point.

[0019] Furthermore, training the classifier includes:

[0020] Grayscale images and thermal infrared images of wheel samples, including at least two types—normal contour and iron oxide scale interference—are collected during the wheel rolling process. In the grayscale and thermal infrared images of the wheel samples, the manually marked real contour points and iron oxide scale edge points are used as sample points. The edge feature vector of each sample point is extracted as a training dataset and input into a support vector machine for training to obtain a trained support vector machine.

[0021] Further, obtaining the contour probability value for each initial edge point includes:

[0022] The edge feature vector of each initial edge point is input into the trained support vector machine, and the support vector machine outputs the contour probability value of each initial edge point, wherein the contour probability value of each initial edge point ranges from 0 to 1.

[0023] Furthermore, the step of filtering out the actual contour points of the wheel to be tested from all initial edge points includes:

[0024] The initial edge points whose contour probability values ​​are greater than a preset probability threshold are taken as the actual contour points of the wheel under test.

[0025] Furthermore, obtaining the radial deviation of the wheel under test includes:

[0026] The standard contour model of the wheel under test is discretized into a reference point cloud, and the actual contour points of the wheel under test are used as the measured point cloud. The number of measured point clouds and reference point clouds is the same.

[0027] The ICP registration algorithm is used to register the reference point cloud and the measured point cloud. The obtained optimal transformation matrix is ​​used to transform the measured point cloud to the same coordinate system as the reference point cloud, so as to obtain the registered measured point cloud corresponding to each reference point cloud. The reference point cloud and the registered measured point cloud are in one-to-one correspondence, and the measured point cloud and the registered measured point cloud are in one-to-one correspondence.

[0028] The difference between the modulus of the position coordinates of each registered measured point cloud and the modulus of the position coordinates of the corresponding reference point cloud is used as the radial deviation index of each measured point cloud.

[0029] The average value of the radial deviation indices of all measured point clouds is taken as the radial deviation of the wheel under test.

[0030] Furthermore, adjusting the rolls for correction based on the radial deviation includes:

[0031] The radial deviation of the wheel under test is input into the PID controller, which outputs a control signal to the servo hydraulic system to drive the roll pressing mechanism for correction.

[0032] Furthermore, after obtaining the grayscale image and the thermal infrared image, it is necessary to perform spatiotemporal alignment preprocessing on the grayscale image and the thermal infrared image.

[0033] The present invention also proposes a machine vision-based online detection and adaptive correction system for wheel rolling profiles. The system includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements any of the steps of a machine vision-based online detection and adaptive correction method for wheel rolling profiles.

[0034] The present invention has the following beneficial effects:

[0035] This invention addresses the issue that the presence of iron oxide scale during wheel rolling can be misidentified as the wheel contour, leading to low accuracy in wheel contour detection and hindering precise contour correction. Therefore, it first acquires grayscale and thermal infrared images of the wheel under test. These two image types are then combined to accurately identify the wheel's contour points, eliminating interference from iron oxide scale. Edge pixels are then initially extracted from the grayscale image. Considering that the initial edge points contain both the actual wheel contour points and iron oxide scale edge points due to iron oxide scale interference, and given the exothermic oxidation reaction causing the iron oxide scale to be hotter than the wheel substrate, the thermal infrared image is further processed with region growing to obtain a binary mask of the pixels. This effectively separates the wheel substrate from the iron oxide scale, while simultaneously... The initial edge points in the grayscale image are mapped to the thermal infrared image, so that the corresponding initial edge points in the two images represent the pixels at the same position of the wheel under test. Then, the features of the initial edge points in the two images are extracted to obtain edge feature vectors that can reflect the temperature value, temperature gradient, texture entropy and other dimensions of each initial edge point. Then, the obtained contour probability value reflects the probability that each initial edge point is the actual contour point of the wheel under test, thereby accurately identifying the actual contour point of the wheel under test and eliminating the interference of iron oxide scale edge points. At the same time, considering that the radial deviation in the wheel rolling process is the most important factor affecting product quality, the actual contour point of the wheel under test is registered with the standard contour model to obtain the radial deviation of the wheel under test. Based on the radial deviation, the wheel contour is accurately corrected. Attached Figure Description

[0036] To more clearly illustrate the technical solutions and advantages in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0037] Figure 1 The flowchart illustrates an online detection and adaptive correction method for wheel rolling profile based on machine vision, as provided in one embodiment of the present invention.

[0038] Figure 2 This is a schematic diagram of a machine vision-based online detection and adaptive correction system for wheel rolling profiles, provided as an embodiment of the present invention. Detailed Implementation

[0039] To further illustrate the technical means and effects adopted by the present invention to achieve its intended purpose, the following, in conjunction with the accompanying drawings and preferred embodiments, details the specific implementation, structure, features, and effects of a machine vision-based online detection and adaptive correction method and system for wheel rolling profiles proposed according to the present invention. In the following description, different "one embodiment" or "another embodiment" do not necessarily refer to the same embodiment. Furthermore, specific features, structures, or characteristics in one or more embodiments can be combined in any suitable form.

[0040] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0041] The following description, in conjunction with the accompanying drawings, details the specific scheme of the online detection and adaptive correction method and system for wheel rolling profile based on machine vision provided by this invention.

[0042] Please see Figure 1 The diagram illustrates a flowchart of an online detection and adaptive correction method for wheel rolling profile based on machine vision, according to an embodiment of the present invention. The method includes:

[0043] Step S1: Acquire grayscale and thermal infrared images of the wheel under test in real time during the rolling process.

[0044] Traditional methods typically use optical equipment to acquire grayscale or visible light images of the wheel during the rolling process. Based on the deviation between the detected profile and the standard profile, the control system parameters of the wheel rolling system (such as roll reduction) are adjusted to correct the wheel profile. Therefore, in this embodiment of the invention, an industrial camera is first installed above the rolling line, and the camera is used to acquire original images of the wheel under test in real time during the rolling process. To reduce the computational load of subsequent image processing and improve processing speed, in one embodiment of the invention, the acquired multi-channel original images are converted into grayscale images, transforming them into single-channel grayscale images. It should be noted that grayscale conversion is a technique well-known to those skilled in the art and will not be described in detail here.

[0045] Since wheel rolling is usually carried out in a high-temperature environment, the formation of iron oxide scale is an unavoidable physicochemical phenomenon. The presence of iron oxide scale can be misidentified as the wheel outline. Considering that the oxidation reaction is exothermic, the temperature of the iron oxide scale is higher than that of the wheel base. Therefore, this embodiment of the invention requires a thermal imager to be installed above the rolling line and to use the thermal imager to collect thermal infrared images of the wheel under test in real time during the wheel rolling process. Subsequently, the outline pixels of the wheel under test can be accurately identified by combining these two types of images during the rolling process, eliminating the interference of iron oxide scale edge points. In the thermal infrared image, the pixel value of the pixel generally refers to the temperature value, which is used to directly reflect the temperature of the corresponding position of the pixel.

[0046] It should be noted that after acquiring the grayscale image and the thermal infrared image, the grayscale image and the thermal infrared image need to be preprocessed for spatiotemporal alignment. That is, the grayscale image and the thermal infrared image need to be acquired synchronously at the same time, and the resolution of the two should be the same. Specifically, spatial alignment can be achieved through algorithms such as bilinear interpolation, which will not be limited or elaborated here.

[0047] Step S2: Perform edge detection on the grayscale image to obtain initial edge points in the grayscale image; perform region growing on the thermal infrared image according to the temperature value distribution of pixels in different regions in the thermal infrared image to obtain a binary mask of pixels in the thermal infrared image; map the initial edge points in the grayscale image to the thermal infrared image, and obtain the edge feature vector of each initial edge point according to the temperature value, gradient value, binary mask of the initial edge point in the thermal infrared image, and the distribution of grayscale values ​​of pixels in the preset neighborhood of the initial edge point in the grayscale image.

[0048] Since the embodiments of the present invention need to correct the contour of the wheel under test during the rolling process, it is necessary to first extract or identify the edge pixels representing the contour of the wheel under test in the image. The embodiments of the present invention first perform edge detection on the grayscale image and extract the initial edge points in the grayscale image. The edge detection can be performed using the existing Canny edge detection algorithm or other edge detection algorithms, which are not limited or described in detail here.

[0049] Since wheel rolling is typically carried out in a high-temperature environment, the formation of iron oxide scale is an unavoidable physicochemical phenomenon. The presence of iron oxide scale can be misidentified as the wheel contour. In other words, the initial edge points contain both the actual contour points of the wheel under test and the edge points of the iron oxide scale, making it impossible to accurately identify the actual contour pixels of the wheel under test. However, due to the exothermic oxidation reaction, the temperature of the iron oxide scale is higher than that of the wheel substrate. Therefore, this embodiment of the invention further combines the temperature value distribution of pixels in different regions of the thermal infrared image to perform region growing on the thermal infrared image, obtaining a binary mask of the pixels in the thermal infrared image. The binary mask takes the value 0 or 1. The binary mask distinguishes between the wheel substrate and the iron oxide scale in the thermal infrared image, thereby effectively separating the wheel substrate and the iron oxide scale. In one embodiment of the invention, pixels with a binary mask value of 1 can represent iron oxide scale, and pixels with a binary mask value of 0 can represent the wheel substrate. Subsequently, the binary mask can be used as a feature of the initial edge points, thereby accurately identifying the edge points representing the actual contour of the wheel under test from the initial edge points and removing the interference of the iron oxide scale.

[0050] Preferably, in one embodiment of the present invention, the method for obtaining the binary mask of pixels in a thermal infrared image specifically includes:

[0051] First, because the area of ​​iron oxide scale is relatively small, the grayscale value of the pixels in the iron oxide scale is larger than the grayscale value of the pixels of the entire wheel under test in the thermal infrared image. Therefore, the average and standard deviation of the temperature values ​​of all pixels in the thermal infrared image can be calculated. The principle is to use the sum of the mean and three times the standard deviation as the high-temperature oxidation threshold, which can then be used to screen out the region representing the substrate of the wheel to be tested.

[0052] Then, the thermal infrared image is evenly divided into multiple blocks (e.g., an 8*8 grid). The average temperature value of all pixels in each block is taken as the local temperature value of each block. Since the temperature value of the block representing the wheel substrate is relatively small, the block with a local temperature value less than the high-temperature oxidation threshold can be taken as the substrate block of the wheel under test. The average local temperature value of all substrate blocks is taken as the substrate temperature of the wheel under test. The sum of the substrate temperature and the preset temperature adjustment value is taken as the seed point screening threshold. Generally, the temperature of the iron oxide scale is 15-30°C higher than that of the substrate due to the exothermic oxidation reaction. Therefore, the preset temperature adjustment value is generally in the range of 15-20°C. In one embodiment of the present invention, the preset temperature adjustment value is set to 15°C. The specific value of the preset temperature adjustment value can also be set by the implementer according to the specific implementation scenario, and is not limited here.

[0053] Finally, in the thermal infrared image, pixels with a temperature value greater than the seed point screening threshold are selected as seed points, and the minimum spacing between seed points is limited to more than 5 pixels to avoid seed point aggregation. Then, region growing is performed on each seed point to obtain a binary mask of the pixels in the thermal infrared image. The conditions for region growing are that the temperature difference between adjacent pixels is not greater than 5°C and the area of ​​the region is not greater than 500 pixels to avoid overgrowth covering the real contour. Here, pixels representing iron oxide scale are selected as seed points. In other embodiments of the present invention, pixels representing the wheel base can also be selected as seed points, that is, pixels with a temperature value not greater than the seed point screening threshold are selected as seed points. This is not limited here.

[0054] After performing region growing in the thermal infrared image, this embodiment of the invention also needs to map the initial edge points in the grayscale image to the thermal infrared image, so that the corresponding initial edge points in the two images represent the pixels at the same position of the wheel under test. Since the two images have already undergone spatial alignment preprocessing, they can be directly mapped, that is, the pixels at the same position in the thermal infrared image are marked as initial edge points.

[0055] Because the iron oxide sheet and the test wheel substrate have different temperatures, the temperature values, gradient values, and other features of their edge pixels in the thermal infrared image are also different. At the same time, the iron oxide sheet introduces a large number of irregular and fragmented edges, and the texture features around its edge pixels are relatively complex. Based on the temperature value, gradient value, and binary mask of the initial edge point in the thermal infrared image, and the distribution of the gray values ​​of the pixels in the preset neighborhood of the initial edge point in the grayscale image, an edge feature vector that reflects the temperature value, temperature gradient, texture entropy, and other dimensional features of each initial edge point is obtained. Subsequently, a classifier can be trained based on the edge feature vector, and the contour probability value of each initial edge point can be obtained using the trained classifier, thereby accurately identifying the true contour points of the test wheel. In one embodiment of the present invention, the size of the preset neighborhood is 5*5. The specific size of the preset neighborhood can also be set by the implementer according to the specific implementation scenario, and is not limited here.

[0056] Preferably, in one embodiment of the present invention, the method for obtaining the edge feature vector of each initial edge point specifically includes:

[0057] First, the thermal infrared image is processed using a gradient operator to obtain the gradient value of each pixel in the thermal infrared image. The gradient operator can be the Sobel operator or the Roberts operator, etc., and is not limited here.

[0058] Then, for the gray values ​​of pixels in the preset neighborhood of each initial edge point in the grayscale image, the gray-level co-occurrence matrix of each initial edge point is calculated, and the entropy value of the gray-level co-occurrence matrix of each initial edge point is used as the local texture complexity of each initial edge point. Since the local texture of the iron oxide scale edge point is more complex than the local texture of the contour point of the wheel under test, the greater the local texture complexity of an initial edge point, the more likely the initial edge point is to be the contour point of the wheel under test.

[0059] Finally, a four-dimensional vector composed of the temperature value, gradient value, binary mask, and local texture complexity of each initial edge point in the thermal infrared image is used as the edge feature vector of each initial edge point. Since the edge feature vector of the iron oxide scale edge point is significantly different from the edge feature vector of the contour point of the wheel under test, the initial edge points can be classified by the edge feature vector to accurately identify the true contour point of the wheel under test.

[0060] Thus, by combining grayscale images and thermal infrared images, an edge feature vector that reflects the temperature and local texture features of each initial edge point was constructed, providing a data foundation for distinguishing the true contour of the wheel under test from the edge of the iron oxide sheet.

[0061] Step S3: Use the wheel rolling samples to train the classifier, and input the edge feature vector of each initial edge point into the trained classifier to obtain the contour probability value of each initial edge point; based on the contour probability value, select the actual contour points of the wheel to be tested from all initial edge points.

[0062] After obtaining the edge feature vector of each initial edge point through the above steps, in order to accurately identify the true contour point of the wheel under test from the initial edge points using the edge feature vector, this embodiment of the invention needs to collect samples during the wheel rolling process and train the classifier by combining the edge feature vector. Subsequently, the trained classifier can be used, combined with the edge feature vector of the initial edge points, to obtain the probability that each initial edge point is a true contour point, thereby accurately extracting the actual contour point of the wheel under test and improving the correction accuracy of the wheel under test during the rolling process.

[0063] Preferably, in one embodiment of the present invention, the method for training the classifier specifically includes:

[0064] First, grayscale and thermal infrared images of wheel samples, including at least two types—normal contour and oxide scale interference—are acquired during the wheel rolling process. In these grayscale and thermal infrared images, manually labeled real contour points and oxide scale edge points are used as sample points. Then, using the same method as obtaining the edge feature vectors of the initial edge points, the edge feature vector of each sample point is obtained. This edge feature vector is then used as a training dataset and input into a support vector machine (SVM) for training, resulting in a trained SVM. The input to the SVM is the edge feature vector, and the output is the probability that the pixel corresponding to each edge feature vector represents the contour of the wheel to be tested. The SVM is a well-known technique in the field and will not be elaborated upon here.

[0065] After obtaining the trained classifier, the edge feature vector of each initial edge point can be input into the trained classifier to obtain the contour probability value of each initial edge point. The contour probability value reflects the probability that each initial edge point is the actual contour point of the wheel under test. Subsequently, the true contour point of the wheel under test can be accurately identified based on the contour probability value, thereby avoiding the interference of the iron oxide scale edge.

[0066] Preferably, in one embodiment of the present invention, the method for obtaining the contour probability value of each initial edge point specifically includes:

[0067] The edge feature vector of each initial edge point is input into the trained support vector machine, and the support vector machine outputs the contour probability value of each initial edge point, wherein the contour probability value of each initial edge point ranges from 0 to 1.

[0068] The higher the profile probability value of an initial edge point, the more likely that the initial edge point is to be the actual profile point of the wheel under test. Therefore, based on the profile probability value, the actual profile point of the wheel under test can be selected from all initial edge points. Subsequently, the identified actual profile point can be combined to accurately correct the profile of the wheel under test during the rolling process.

[0069] Preferably, in one embodiment of the present invention, the method for obtaining the actual contour points of the wheel to be tested specifically includes:

[0070] Initial edge points with contour probability values ​​greater than a preset probability threshold are used as the actual contour points of the wheel under test. To improve the accuracy of contour point recognition, a high-confidence preset probability threshold is required. The value range of the preset probability threshold is typically [value missing]. In one embodiment of the present invention, the preset probability threshold is set to 0.9. The specific value of the preset probability threshold can also be set by the implementer according to the specific implementation scenario, and is not limited here.

[0071] Thus, by using the edge feature vectors of the initial edge points and combining them with the trained classifier, the true contour points of the wheel under test are accurately identified, avoiding interference from the edges of the iron oxide scale during the rolling process and ensuring the accuracy of subsequent wheel correction.

[0072] Step S4: Register the actual contour points of the wheel to be tested with the standard contour model to obtain the radial deviation of the wheel to be tested; adjust the rollers to correct the deviation based on the radial deviation.

[0073] Considering that radial deviation is the most important factor affecting product quality during wheel rolling, and that the presence of radial deviation will cause the produced wheels to not reach the standard circle, after obtaining the actual contour points of the wheel to be tested through the above steps, the standard contour model of the wheel to be tested can be further combined with the model to be tested, and the two can be registered to accurately obtain the radial deviation of the wheel to be tested during the rolling process. Subsequently, precise correction can be performed based on the radial deviation. The standard contour model is a known contour model of the wheel to be tested designed by the designer before rolling.

[0074] Preferably, in one embodiment of the present invention, the method for obtaining the radial deviation of the wheel under test specifically includes:

[0075] First, the standard profile model of the wheel under test is discretized into a reference point cloud, and the actual profile points of the wheel under test are used as the measured point cloud. The number of measured point clouds and reference point clouds is the same to ensure smooth registration in the future. If the number of discretized reference point clouds is too large, the reference point clouds can be sampled at equal intervals along the circumference to obtain a reference point cloud with the same number of points as the measured point cloud. At the same time, interpolation processing can be performed on the measured point clouds to ensure that the number of measured point clouds and reference point clouds is the same.

[0076] Considering that the measured point cloud and the reference point cloud need to be transformed to the same coordinate system when calculating the deviation, the ICP registration algorithm can be used to register the reference point cloud and the measured point cloud. Then, the obtained optimal transformation matrix is ​​used to transform the measured point cloud to the same coordinate system as the reference point cloud, so as to obtain the registered measured point cloud corresponding to each reference point cloud. The reference point cloud and the registered measured point cloud are in one-to-one correspondence, and the measured point cloud and the registered measured point cloud are also in one-to-one correspondence.

[0077] Furthermore, the difference between the modulus of the position coordinates of each registered measured point cloud and the modulus of the position coordinates of the corresponding reference point cloud can be used as the radial deviation index of each measured point cloud. When the radial deviation index of a measured point cloud is greater than 0 and larger, it indicates that the measured point cloud is significantly expanded outward relative to the reference point cloud. Conversely, when the radial deviation index of a measured point cloud is less than 0 and smaller, it indicates that the measured point cloud is significantly expanded inward relative to the reference point cloud. The average value of the radial deviation indices of all measured point clouds can be used as the radial deviation of the wheel under test, which reflects the deviation of the overall contour of the wheel under test.

[0078] Once the radial deviation of the wheel under test is obtained, the wheel under test can be corrected in real time during the rolling process based on the radial deviation. This correction can usually be achieved by adjusting the rolls, thereby ensuring that the produced wheel is close to the ideal circle.

[0079] Preferably, in one embodiment of the present invention, the method for correcting the deviation of the wheel under test specifically includes:

[0080] The radial deviation of the wheel under test is input into the PID controller, which outputs a control signal to the servo hydraulic system to drive the roll pressing mechanism for correction. The PID controller can balance response speed and control accuracy through a proportional, integral and derivative strategy. The servo hydraulic system has the characteristics of large output force and fast response, which is suitable for the real-time force control requirements of the rolling process.

[0081] This invention also provides a machine vision-based online detection and adaptive correction system for wheel rolling profiles. Please refer to [link to relevant documentation]. Figure 2The diagram illustrates a schematic of a machine vision-based online detection and adaptive correction system framework for wheel rolling profiles according to an embodiment of the present invention. The system may include: a memory 201, a processor 202, and a computer program stored in the memory 201 and executable on the processor 202. When the processor 202 executes the program, it implements the machine vision-based online detection and adaptive correction method for wheel rolling profiles provided in the above embodiment.

[0082] Furthermore, the system also includes a communication interface 203 for communication between the memory 201 and the processor 202.

[0083] The memory 201 may include high-speed RAM memory, and may also include nonvolatile memory, such as at least one disk storage.

[0084] If the memory 201, processor 202, and communication interface 203 are implemented independently, then the communication interface 203, memory 201, and processor 202 can be interconnected via a bus to complete communication between them. The bus can be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus, or an Extended Industry Standard Architecture (EISA) bus, etc. The bus can be divided into address bus, data bus, control bus, etc. For ease of representation, Figure 2 The bus is represented by a single thick line, but this does not mean that there is only one bus or one type of bus.

[0085] Optionally, in a specific implementation, if the memory 201, processor 202, and communication interface 203 are integrated on a single chip, then the memory 201, processor 202, and communication interface 203 can communicate with each other through an internal interface.

[0086] The processor 202 may be a central processing unit (CPU), an application-specific integrated circuit (ASIC), or one or more integrated circuits configured to implement embodiments of the present invention.

[0087] In summary, this embodiment of the invention combines two types of images: grayscale images and thermal infrared images. First, edge pixels are initially extracted from the grayscale image. Considering the interference from iron oxide scale, the initial edge points include both the actual contour points of the wheel under test and the edge points of the iron oxide scale. However, due to the exothermic oxidation reaction, the temperature of the iron oxide scale is higher than that of the wheel substrate. Therefore, region growing is further performed on the thermal infrared image to obtain a binary mask of the pixels in the thermal infrared image, thereby effectively separating the wheel substrate and the iron oxide scale. Simultaneously, the initial edge points in the grayscale image are mapped onto the thermal infrared image, so that corresponding initial edge points in both images represent the same location on the wheel under test. The system extracts the features of the initial edge points from the two images, obtaining edge feature vectors that reflect the temperature value, temperature gradient, texture entropy, and other dimensions of each initial edge point. Then, the obtained contour probability value reflects the probability that each initial edge point is the actual contour point of the wheel under test, thereby accurately identifying the actual contour point of the wheel under test and eliminating the interference of iron oxide scale edge points. At the same time, considering that radial deviation is the most important factor affecting product quality during wheel rolling, the actual contour point of the wheel under test is registered with the standard contour model to obtain the radial deviation of the wheel under test. Based on the radial deviation, the wheel contour is accurately corrected.

[0088] It should be noted that the order of the above embodiments of the present invention is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. The processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired result. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0089] The various embodiments in this specification are described in a progressive manner. The same or similar parts between the various embodiments can be referred to each other. Each embodiment focuses on describing the differences from other embodiments.

Claims

1. A machine vision-based online detection and adaptive correction method for wheel rolling profile, characterized in that, The method includes: Real-time acquisition of grayscale and thermal infrared images of the wheel under test during the rolling process; Edge detection is performed on the grayscale image to obtain initial edge points. Based on the temperature value distribution of pixels in different regions of the thermal infrared image, region growing is performed on the thermal infrared image to obtain a binary mask of pixels in the thermal infrared image. The initial edge points in the grayscale image are mapped to the thermal infrared image, and the edge feature vector of each initial edge point is obtained based on the temperature value, gradient value, binary mask of the initial edge point in the thermal infrared image, and the distribution of grayscale values ​​of pixels in the preset neighborhood of the initial edge point in the grayscale image. Using samples of wheel rolling, a classifier is trained, and the edge feature vector of each initial edge point is input into the trained classifier to obtain the contour probability value of each initial edge point; based on the contour probability value, the actual contour points of the wheel to be tested are selected from all initial edge points. The actual contour points of the wheel under test are registered with the standard contour model to obtain the radial deviation of the wheel under test; based on the radial deviation, the rollers are adjusted to correct the deviation.

2. The method for online detection and adaptive correction of wheel rolling profile based on machine vision according to claim 1, characterized in that, The method for obtaining the binary mask of pixels in the thermal infrared image includes: Calculate the average and standard deviation of the temperature values ​​of all pixels in the thermal infrared image, and use the sum of the average and three times the standard deviation as the high-temperature oxidation threshold; The thermal infrared image is evenly divided into multiple blocks, and the average temperature value of all pixels in each block is taken as the local temperature value of each block; the block with the local temperature value being less than the high-temperature oxidation threshold is taken as the base block of the wheel under test. The average value of the local temperature values ​​of all the substrate blocks is taken as the substrate temperature of the wheel under test, and the sum of the substrate temperature and the preset temperature adjustment value is taken as the seed point screening threshold. In the thermal infrared image, pixels with a temperature value greater than the seed point screening threshold are selected as seed points, and region growing is performed on each seed point to obtain a binary mask of the pixels in the thermal infrared image. The binary mask of each pixel in the thermal infrared image has a value of 0 or 1.

3. The method for online detection and adaptive correction of wheel rolling profile based on machine vision according to claim 1, characterized in that, The process of obtaining the edge feature vector for each initial edge point includes: The gradient operator is used to process the thermal infrared image to obtain the gradient value of each pixel in the thermal infrared image; For each initial edge point in a grayscale image, calculate the grayscale co-occurrence matrix of each initial edge point in the preset neighborhood of the pixel, and use the entropy value of the grayscale co-occurrence matrix of each initial edge point as the local texture complexity of each initial edge point; The four-dimensional vector composed of the temperature value, gradient value, binary mask, and local texture complexity of each initial edge point in the thermal infrared image is used as the edge feature vector of each initial edge point.

4. The method for online detection and adaptive correction of wheel rolling profile based on machine vision according to claim 1, characterized in that, Training the classifier includes: Grayscale images and thermal infrared images of wheel samples, including at least two types—normal contour and iron oxide scale interference—are collected during the wheel rolling process. In the grayscale and thermal infrared images of the wheel samples, the manually marked real contour points and iron oxide scale edge points are used as sample points. The edge feature vector of each sample point is extracted as a training dataset and input into a support vector machine for training to obtain a trained support vector machine.

5. The method for online detection and adaptive correction of wheel rolling profile based on machine vision according to claim 4, characterized in that, The process of obtaining the contour probability value for each initial edge point includes: The edge feature vector of each initial edge point is input into the trained support vector machine, and the support vector machine outputs the contour probability value of each initial edge point, wherein the contour probability value of each initial edge point ranges from 0 to 1.

6. The method for online detection and adaptive correction of wheel rolling profile based on machine vision according to claim 1, characterized in that, The process of selecting the actual contour points of the wheel to be tested from all initial edge points includes: The initial edge points whose contour probability values ​​are greater than a preset probability threshold are taken as the actual contour points of the wheel under test.

7. The method for online detection and adaptive correction of wheel rolling profile based on machine vision according to claim 1, characterized in that, The radial deviation of the wheel under test is obtained by means of: The standard contour model of the wheel under test is discretized into a reference point cloud, and the actual contour points of the wheel under test are used as the measured point cloud. The number of measured point clouds and reference point clouds is the same. The ICP registration algorithm is used to register the reference point cloud and the measured point cloud. The obtained optimal transformation matrix is ​​used to transform the measured point cloud to the same coordinate system as the reference point cloud, so as to obtain the registered measured point cloud corresponding to each reference point cloud. The reference point cloud and the registered measured point cloud are in one-to-one correspondence, and the measured point cloud and the registered measured point cloud are in one-to-one correspondence. The difference between the modulus of the position coordinates of each registered measured point cloud and the modulus of the position coordinates of the corresponding reference point cloud is used as the radial deviation index of each measured point cloud. The average value of the radial deviation indices of all measured point clouds is taken as the radial deviation of the wheel under test.

8. The method for online detection and adaptive correction of wheel rolling profile based on machine vision according to claim 1, characterized in that, The step of adjusting the rolls for correction based on the radial deviation includes: The radial deviation of the wheel under test is input into the PID controller, which outputs a control signal to the servo hydraulic system to drive the roll pressing mechanism for correction.

9. The method for online detection and adaptive correction of wheel rolling profile based on machine vision according to claim 1, characterized in that, After obtaining the grayscale image and the thermal infrared image, spatiotemporal alignment preprocessing is required.

10. A machine vision-based online detection and adaptive correction method system for wheel rolling profiles, the system comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the method as described in any one of claims 1 to 9.

Images