Visual inspection tool and method for tooth surface contact spots

Abstract

The invention provides a visual detection tool and method for tooth surface contact spots. The visual inspection tool comprises a connecting seat which is connected with a gear; one end of the bracket is mounted on the connecting seat, and the mounting angle of the bracket is adjustable; the shading plate is fixed on the bracket; one end of the mounting rod is movably connected with the other end of the bracket, and the mounting rod is not coplanar and parallel to the shading plate; the camera is mounted at the other end of the mounting rod; and the image processing system is in communication connection with the camera. According to the visual detection method, an AI model is adopted to process an acquired tooth surface image, the total number of contact spot pixels and the total number of tooth surface pixels in the tooth surface image are obtained, and the proportion of contact spots is calculated. The visual inspection tool and method provided by the invention can be applied to different workpiece types, and are high in image acquisition accuracy, low in noise, high in calculation result repeatability and high in calculation efficiency.

Description

Technical Field

[0001] The present invention relates to the technical field of visual inspection, and in particular to a visual inspection tool and method for tooth surface contact spots. Background Art

[0002] The current main development direction of my country's railways is high speed and heavy load. Gears, as extremely important parts in the transmission system of rail transit trains, need to meet the requirements of impact resistance, ultra-high reliability, and ultra-long life. In order to ensure the performance and safety of gears and effectively evaluate the load distribution between gear teeth, the tooth surface contact spot test is usually used to evaluate the gear meshing quality. The larger the contact spot area, the higher the gear meshing quality.

[0003] The traditional way to record the results of the tooth surface contact spot test is to take photos, draw sketches, use transparent tape, etc. This inspection method not only requires operator training to ensure the consistency of operation efficiency, but also contains subjective judgment and low work efficiency. Therefore, automated analysis of tooth surface contact spot test results has become one of the hot topics for technical personnel in this field.

[0004] With the rapid development of computer vision, the use of visual inspection tools for tooth surface image acquisition and analysis can solve many of the above-mentioned shortcomings of manual operation, but there are also the following disadvantages:

[0005] 1. The images taken by existing visual inspection tools have poor accuracy and high noise;

[0006] 2. The existing visual inspection tooling is only applicable to gears of a certain tooth width. When analyzing the tooth surface contact spots of gears with different tooth widths, different visual inspection tooling needs to be replaced, and the disassembly and assembly process is cumbersome. Summary of the invention

[0007] The purpose of the present invention is to overcome the defects of the prior art and provide a visual inspection tool and method for tooth surface contact spots, which can solve the problems of poor image accuracy, high noise, and poor universality of images taken by visual inspection tooling in the prior art.

[0008] In order to achieve the above purpose and other purposes, the present invention is implemented by including the following technical solutions: As a first aspect, the present invention proposes a tooth surface contact spot visual detection tooling, including a connecting seat connected to the gear; a bracket, one end of which is mounted on the connecting seat, and its mounting angle is adjustable; a shading plate fixed on the bracket; a mounting rod, one end of which is movably connected to the other end of the bracket, and is non-coplanar and parallel to the shading plate; a camera mounted on the other end of the mounting rod; and an image processing system, which is communicatively connected to the camera.

[0009] The design of this solution allows the shading plate to fit approximately with the tooth surfaces of gears with different helical angles, which can effectively avoid the influence of external factors and ensure that the shooting content is more accurate and noise-free; the distance between the camera and the bracket is adjustable, which can meet the calculation requirements of the tooth surface contact spots of gears with different tooth widths.

[0010] In one embodiment, the connecting seat is magnetically connected to the gear.

[0011] The design of this solution makes the connection position between the connecting seat and the gear flexible and changeable, so that the visual inspection tool can move freely while maintaining a relative position with the tooth surface on the gear, thereby improving the convenience of inspection operation.

[0012] In one embodiment, the bracket is L-shaped, and a through hole and a locking groove are opened at the other end of the bracket. The through hole is used to install the mounting rod; the locking groove is connected to the through hole, and the loosening or locking of the locking groove is adjusted by a bolt.

[0013] The design of this solution enables the movable and locking operations of the mounting rod on the bracket to be adjusted only by rotating the bolt, and has a simple structure and convenient operation.

[0014] In one embodiment, the camera is mounted on the other end of the mounting rod via a mounting platform; the mounting angle on the mounting platform is adjustable, and the distance relative to the bracket is adjustable.

[0015] The design of this solution realizes the adjustment of the camera through the mounting platform, which can avoid problems such as camera center of gravity offset caused by direct adjustment of the camera.

[0016] In one embodiment, a through hole and a locking groove are formed on the mounting platform, wherein the through hole is used to mount the mounting rod; the locking groove is connected to the through hole, and the loosening or locking of the locking groove is adjusted by a bolt.

[0017] The design of this solution enables the optional installation, removability and locking operation of the mounting platform on the mounting rod to be adjusted only by rotating the bolt, and has a simple structure and convenient operation.

[0018] As a second aspect, the present invention also provides a method for visually detecting tooth surface contact spots, comprising the steps of:

[0019] Using the visual inspection tool as described in the first aspect to collect tooth surface images of the gear under fixed lighting conditions;

[0020] The collected tooth surface image is processed using an AI model to obtain the total number of contact spot pixels and the total number of tooth surface pixels in the tooth surface image, and the contact spot ratio is calculated.

[0021] This solution adopts the visual inspection tooling described in the first aspect, which can ensure that the shooting content is more accurate, noise-free, and has a wider applicability; by using an AI model for image processing, the results analysis of the tooth surface contact spot test can be completed automatically, accurately, and efficiently.

[0022] In one embodiment, the tooth surface image acquisition comprises the steps of:

[0023] Connecting the visual inspection tool to the gear;

[0024] Adjust the bracket and the mounting platform so that the shading plate fits the tooth surface of the gear and the camera is perpendicular to the tooth surface;

[0025] The shading is turned on, and the camera collects the tooth surface image of the gear under fixed lighting conditions.

[0026] This design scheme adjusts the bracket so that the shading plate can approximately fit the tooth surface of gears with different helical angles, which can effectively avoid the influence of external factors and ensure that the shooting content is more accurate and noise-free; the camera is in a state perpendicular to the tooth surface, reducing the inaccurate visual detection results of the contact spot caused by the deformation of the tooth surface due to the angle tilt.

[0027] In one embodiment, the method for constructing the AI ​​model comprises the steps of:

[0028] Acquire tooth surface images containing colorants and tooth surface images without colorants as pre-training data sets;

[0029] Performing denoising and cropping preprocessing on all tooth surface images in the pre-training data set;

[0030] The preprocessed tooth surface image is subjected to bit plane normalization flow processing, and the contact spot confidence map is calculated by constructing the Gaussian distribution of each bit plane feature of the tooth surface without contact spot;

[0031] The contact spot confidence map is fused with the denoised tooth surface image to visualize and compare the position and edge of the contact spot in the original tooth surface image.

[0032] This scheme can highlight the image features by reconstructing the image through normalized bit planes; and realize double mapping by constructing the Gaussian distribution of each bit plane feature of the tooth surface without contact spots, ensuring that the tooth surface contact spot image analysis can be completed automatically, accurately and efficiently when the image data is small; if there is a contact spot, the confidence of each pixel in the contact spot confidence map output by the model is higher, otherwise the confidence of each pixel in the confidence map is lower, thus realizing pixel-level unsupervised detection and positioning of the contact spot.

[0033] In one embodiment, while "performing bit plane normalization flow processing on the preprocessed tooth surface image", the preprocessed tooth surface image is binarized to obtain a tooth surface contact spot mask image as a true label of the test data set.

[0034] The design of this scheme can ensure that a calculation standard is provided for the calculation of the contact patch confidence map, thereby ensuring the accuracy of the calculation.

[0035] In one embodiment, the noise reduction is performed by filtering out multiple low-order bit planes.

[0036] The design of this solution can significantly reduce the amount of image data processing and improve the efficiency of subsequent image processing while retaining valid images.

[0037] Compared with the prior art, the beneficial effects of the present invention are as follows: the visual inspection tooling and method provided by the present invention can be applied to different types of workpieces, with high image acquisition accuracy, low noise, high repeatability of calculation results, and high calculation efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] Figure 1 Shown is a schematic diagram of the three-dimensional structure of a visual inspection tool for tooth surface contact spots of the present invention connected to a gear.

[0039] Figure 2 Display as Figure 1 Left view of a visual inspection tool for tooth surface contact spots.

[0040] Figure 3 Display as Figure 1 A top view of a visual inspection tool for tooth surface contact spots.

[0041] Figure 4 Shown is a schematic flow chart of a visual detection method for tooth surface contact spots according to the present invention. DETAILED DESCRIPTION

[0042] See also Figure 1-Figure 4 The following describes the embodiments of the present invention through specific examples. Those skilled in the art can easily understand other advantages and effects of the present invention from the contents disclosed in this specification.

[0043] It should be noted that the structures, proportions, sizes, etc. illustrated in the drawings in this specification are only used to match the contents disclosed in the specification for the technical personnel in this field to understand and read, and are not used to limit the conditions under which the present invention can be implemented. Therefore, they have no substantive technical significance. Any modification of the structure, change in the proportion relationship or adjustment of the size should still fall within the scope of the technical contents disclosed in the present invention without affecting the effects and purposes that can be achieved by the present invention.

[0044] Unless otherwise defined, the technical or scientific terms used herein shall have the usual meanings as understood by persons having ordinary skills in the field to which the present invention belongs. The terms "first", "second", etc. used in the present invention are used to distinguish different objects, rather than to describe a specific order, quantity or importance. "Include" or "comprising" and similar words mean that the elements or objects appearing before the word include the elements or objects listed after the word and their equivalents, without excluding other elements or objects. The term "connection" in the present invention includes direct and indirect connections unless otherwise specified.

[0045] Embodiment 1

[0046] like Figure 1-Figure 3 As shown, this embodiment provides a tooth surface contact spot visual detection tool, including a connecting seat 1, a bracket 2, a shading plate 3, a mounting rod 4, a mounting platform 5, a camera 6 and an image processing system.

[0047] Since railway gears are generally made of steel or iron, the connecting seat 1 can be designed as a magnetic seat, which is connected to the gear A by magnetic adsorption. This design has a flexible and changeable connection position, so that the visual inspection tooling can move freely while maintaining a relative position with the tooth surface on the gear A, that is, remain relatively still.

[0048] One end of the bracket 2 can be installed on the connecting seat 1 by bolts, that is, one end of the bracket 2 is provided with threads, which are connected to the inner hole of the connecting seat 1 by nuts; the installation angle of the bracket 2 can be adjusted by the bolts; the shading plate 3 can be fixed on the bracket 2 by gluing, etc., and the shading plate 3 is a black cardboard, which can change with the angle of the bracket 2, so that the shading plate 3 can achieve approximate fit with the tooth surface of gears with different helical angles, avoiding the influence of external factors as much as possible, and ensuring that the shooting content is more accurate and noise-free. Specifically, the bracket 2 is L-shaped, so that the shading plate 3 and the mounting rod 4 on it can be non-coplanar and parallel.

[0049] One end of the mounting rod 4 can be installed in the other end of the bracket 2 by bolts, and the other end can be installed in the mounting platform 5 by bolts. In this embodiment, a through hole and a locking groove connected to the through hole are provided at the other end of the bracket 2. The locking groove can be loosened or locked by adjusting the bolts. When loosening, the mounting rod 4 can be moved in the through hole to adjust its fixed position. The mounting platform 5 and the mounting rod 4 can be installed in the same manner as the bracket 2 and the mounting rod 4. This design can adjust the distance of the mounting platform 5 relative to the bracket 2 by adjusting the connection position of the mounting rod 4, the bracket 2 and the mounting platform 5 to meet the tooth surface contact spot calculation requirements of gears with different tooth widths.

[0050] The camera 6 is mounted on the mounting platform 5 by bolts. The mounting angle of the camera 6 is adjustable, so that the angle of the camera 6 relative to the tooth surface can be adjusted to ensure that the camera 6 takes pictures and records at an angle perpendicular to the tooth surface as much as possible.

[0051] The image processing system is in communication with the camera 6 and is used to process the tooth surface image captured by the camera 6 using AI.

[0052] If the contact spot state of the tooth surface of gears with different helical angles needs to be visually inspected, it is only necessary to adjust the angle of the bracket 2 so that the shading plate 3 fits the tooth surface, the mounting rod 4 is approximately parallel to the tooth surface, and the camera 6 is perpendicular to the tooth surface, thereby reducing the inaccuracy of the visual inspection result of the contact spot caused by the deformation of the tooth surface due to the angle tilt. If the contact spot state of the tooth surface of gears with different tooth widths needs to be visually inspected, it is only necessary to adjust the fixed position of the mounting platform 5 on the mounting rod 4.

[0053] Embodiment 2

[0054] like Figure 4 As shown, this embodiment provides a method for visually inspecting contact spots on tooth surfaces, and uses a visual inspection tool for visually inspecting contact spots on tooth surfaces provided in the first embodiment to visually inspect the contact spots on the tooth surfaces of gear A, including the following steps:

[0055] S1. Use the visual inspection tooling described in Example 1 to collect tooth surface images of gear A under fixed lighting conditions;

[0056] The specific steps include:

[0057] S11. Establishing system coordinate system: establishing gear coordinate system, tooth surface contact spot visual inspection tooling coordinate system and camera coordinate system respectively;

[0058] Specifically, the gear coordinate system is based on the first radial direction of gear A ( Figure 1 The middle left and right direction) is the x-axis, the second radial direction ( Figure 1 The tooth surface contact spot visual detection tooling coordinate system takes the length direction of the long arm of the bracket 2 as the x-axis, the length direction of the short arm of the bracket 2 as the y-axis, and the length direction of the light shielding plate 3 as the z-axis; the camera coordinate system takes the length direction of the mounting platform 5 as the x-axis, the width direction as the y-axis, and the length direction of the light shielding plate 3 as the z-axis;

[0059] S12, mounting the camera 6 on the mounting platform 5; the connecting seat 1 is magnetically attracted to the gear; hand-eye calibration can be used to align the coordinate systems;

[0060] S13, fixing the angle between the tooth surface and the z-axis of the system coordinate system; adjusting the bracket 2 and the mounting platform 5 so that the shading plate 3 fits the tooth surface of the gear A and the camera 6 is perpendicular to the tooth surface;

[0061] S14, the visual inspection tooling is turned on to shield the light, and the camera 6 collects the tooth surface image of the gear A under fixed lighting conditions;

[0062] S2. Use an AI model to process the collected tooth surface image to obtain the total number of contact spot pixels and the total number of tooth surface pixels in the tooth surface image, and calculate the contact spot ratio.

[0063] With the rapid development of computer vision, AI technology based on deep learning has been applied in the field of visual inspection. For example, the existing visual inspection methods for tooth contact spots mostly adopt the YOLO series of deep neural network technology. This technology can solve many shortcomings of manual operation in the background technology, but it also has the following disadvantages: 1. It is impossible to accurately segment the tooth contact spots when there is less image data of the tooth contact spots; 2. The shapes of contact spots are different, and the calculation of tooth contact spots has very high requirements for the degree of automation. It is impossible to use point prompts or bounding box prompts to achieve image segmentation (such as SAM-related deep neural networks).

[0064] Therefore, how to automatically, accurately and efficiently complete the result analysis of the tooth surface contact spot test when there is less image data is also a technical problem that needs to be further solved by the visual detection method of the tooth surface contact spot provided in this embodiment. The following is a detailed description of the construction method of the AI ​​model in this embodiment: The construction method of the AI ​​model includes the steps of:

[0065] S21, obtaining a tooth surface image containing a colorant and a tooth surface image without a colorant as a pre-training data set;

[0066] S22, performing noise reduction processing on the images in the pre-training data set;

[0067] Specifically, for the three-channel denoising of the tooth surface image (generally an RGB image), multiple low-order bit planes are filtered out, and the image is reconstructed using the 8th, 7th, 6th, and 5th bit planes to obtain bits_images[:,:,k] (where the first ":" represents all rows in the image, the second ":" represents all columns in the image, k represents the image channel, and k = 1, 2, 3). For example, bit_images[3000,4096,3] represents an image with a width of 3000, a length of 4096, and 3 channels. It should be noted that the number of low-order bit planes to be filtered out can be set according to the characteristics of the image itself and the purpose of visual detection, and it is not necessary to filter out 4 bit planes. For example, when the image details are not concerned and the image contour is more concerned, 5 or 6 low-order bit planes can be filtered out; when the image details are more concerned, only 1 or 2 low-order bit planes can be filtered out.

[0068] S23, cropping the denoised image to remove irrelevant background;

[0069] S24, normalize the cropped image, use a deep neural network to detect and segment the tooth surface, obtain a single tooth surface image, reconstruct the single tooth surface in bit plane to obtain a tooth surface data set, and construct a Gaussian distribution Z=f of each bit plane feature of the contact spot-free tooth surface image θ (X);

[0070] S25, binarizing the cropped image to obtain a tooth surface contact spot mask image, which is used as a true label in the test data set;

[0071] S26, constructing a feature distribution p of a normal image for the single tooth surface X (x), search from the parameter space Θ to satisfy (in is the determinant of the Jacobian matrix of the multivariate Gaussian distribution) parameter θ, X = f θ (Z), Z = f θ -1 (X), x0 represents the original input image, x represents the bit plane reconstructed image, and the best distribution of the current non-contact spot tooth surface image is obtained. Construct Bit-Plane Normalizing Flow;

[0072] Specifically, the normalized image is Where k = 1, 2, 3, mean n [i] represents the mean pixel value of the i-th row of the n-th bit plane, multiplied by 2 n-1Then calculate the weighted sum to get the mean pixel value of the i-th row of the normalized image, std n [i] represents the standard deviation of the pixel values ​​in the i-th row of the n-th bit plane. 0.1×(n-4) is the weight associated with the bit plane. Each pixel value is subtracted from the weighted sum of the means of all pixels in the corresponding row of each bit plane, and then divided by the standard deviation of the pixels in that row. After normalization, the image features can be more prominent.

[0073] S27, calculating and obtaining a contact spot confidence map; This result is used as the distance score map between the input image (normalized image) and the feature distribution of the tooth surface image without contact spots. After thresholding the distance, the contact spot confidence map image[:,:] of a single channel is obtained;

[0074] When a tooth surface image that the AI ​​model has never seen before is input, if it contains a contact spot, the confidence of each pixel in the contact spot confidence map output by the model is high; otherwise, the confidence of each pixel in the confidence map is low, thereby realizing pixel-level unsupervised detection and positioning of the contact spot.

[0075] S28, fusing the contact spot confidence map with the denoised tooth surface image, visually comparing the position and edge of the contact spot in the original tooth surface image; image[:,:]∩bits_images[:,:,k]=tooth surface contact spot[:,:,k], where k represents the image channel, such as k=1, 2, 3 for RGB images;

[0076] S29, calculating the proportion p of the number of pixels of the tooth surface contact spots = the number of pixels of the tooth surface contact spots ga / the total number of pixels of the tooth surface GP.

[0077] This embodiment uses a computer vision algorithm based on deep learning to segment the tooth surface and contact spot, and obtains the intersection of the two to obtain the contact spot area / area ratio; when using a computer vision model based on deep learning, a pre-trained model based on ImageNet is used, and the low-level features and high-level features in the pre-trained model are eliminated, and the middle-level features are extracted. The contact spot and tooth surface data sets are used to train the model, so that the model can supplement the workpiece and tooth surface feature data on the pre-trained middle-level features, and use this feature data as a high-level feature, which greatly enhances the specificity of the AI ​​model algorithm in the present invention. At the same time, this embodiment uses a digital image processing algorithm to calculate the edge of the contact spot, perform connected domain analysis, and calculate the connected domain area / area ratio.

[0078] Furthermore, a dedicated large language model can be used to input prompt words, and the contact patch and tooth surface can be segmented to calculate the contact patch area / area ratio by continuously improving the description accuracy of the contact patch.

[0079] Therefore, the present invention effectively overcomes various shortcomings in the prior art and has a high industrial utilization value. The above embodiments are only illustrative of the principles and effects of the present invention, and are not intended to limit the present invention. Anyone familiar with the technology can modify or change the above embodiments without violating the spirit and scope of the present invention. Therefore, all equivalent modifications or changes made by those with ordinary knowledge in the technical field without departing from the spirit and technical ideas disclosed by the present invention should still be covered by the claims of the present invention.

Claims

1. A visual inspection tool for tooth surface contact spots, characterized in that: include A connecting seat connected to the gear; A bracket, one end of which is mounted on the connecting seat, and the mounting angle of which is adjustable; A sunshade, fixed on the bracket; A mounting rod, one end of which is movably connected to the other end of the bracket and is not coplanar with the sunshade but parallel to the sunshade; a camera mounted on the other end of the mounting rod; An image processing system is communicatively connected with the camera.

2. The visual inspection tool according to claim 1, characterized in that: The connecting seat is magnetically connected to the gear.

3. The visual inspection tool according to claim 1, characterized in that: The bracket is L-shaped, and a through hole and a locking groove are provided at the other end of the bracket. The through hole is used to install the mounting rod; the locking groove is connected to the through hole, and the loosening or locking of the locking groove is adjusted by a bolt.

4. The visual inspection tool according to claim 1, characterized in that: The camera is mounted on the other end of the mounting rod through a mounting platform; the mounting angle on the mounting platform is adjustable, and the distance relative to the bracket is adjustable.

5. The visual inspection tool according to claim 4, characterized in that: The mounting platform is provided with a through hole and a locking groove, wherein the through hole is used for mounting the mounting rod; the locking groove is communicated with the through hole, and the loosening or locking of the locking groove is adjusted by a bolt.

6. A method for visually detecting contact spots on tooth surfaces, characterized in that: Included steps Using the visual inspection tool as described in any one of claims 1 to 5 to collect the tooth surface image of the gear under fixed lighting conditions; The collected tooth surface image is processed using an AI model to obtain the total number of contact spot pixels and the total number of tooth surface pixels in the tooth surface image, and the contact spot ratio is calculated.

7. The visual inspection method according to claim 6, characterized in that: The tooth surface image acquisition comprises the steps of: Connecting the visual inspection tool to the gear; Adjust the bracket and the mounting platform so that the shading plate fits the tooth surface of the gear and the camera is perpendicular to the tooth surface; The shading is turned on, and the camera collects the tooth surface image of the gear under fixed lighting conditions.

8. The visual inspection method according to claim 6, characterized in that: The method for constructing the AI ​​model comprises the steps of: obtaining a tooth surface image containing a colorant and a tooth surface image not containing a colorant as a pre-training data set; Performing denoising and cropping preprocessing on all tooth surface images in the pre-training data set; Performing bit plane normalization flow processing on the preprocessed tooth surface image and calculating the contact spot confidence map; The contact spot confidence map is fused with the denoised tooth surface image to visualize and compare the position and edge of the contact spot in the original tooth surface image.

9. The visual inspection method according to claim 8, characterized in that: While "performing bit plane normalization flow processing on the preprocessed tooth surface image", the preprocessed tooth surface image is binarized to obtain the tooth surface contact spot mask image as the true label of the test data set.

10. The visual inspection method according to claim 8, characterized in that: The noise reduction is performed by filtering out a plurality of low-order bit planes.

Images