A vision-based method and system for milling wind turbine brake discs

Abstract

This invention relates to the field of image processing technology, and more specifically, to a method and system for milling wind turbine brake discs based on visual positioning. The method includes: acquiring an image of the wind turbine brake disc and converting the image to the HSV color space to separate the brightness and saturation channels; performing edge detection on the brightness channel to obtain the gradient magnitude of each pixel; and calculating the final voting weight of each pixel. This invention constructs a final voting weight. This weight integrates four dimensions: a brightness suppression factor to suppress abnormal brightness, a realism factor characterizing surface authenticity, edge intensity gradient magnitude, and neighborhood grayscale range. This multi-dimensional weighting method can intelligently distinguish and suppress false edges caused by reflections, while enhancing the contribution of the real brake disc contour edges.

Description

Technical Field

[0001] This invention relates to the field of image processing technology. More specifically, this invention relates to a method and system for milling wind turbine brake discs based on vision positioning. Background Technology

[0002] With the rapid development of modern industry towards intelligence and automation, wind power, as a core clean energy source, faces increasingly higher requirements for processing precision and production efficiency in its equipment manufacturing industry. Wind turbine brake discs are critical braking and safety components of wind turbine generators. During long-term operation, the surface of the brake disc inevitably suffers wear, pitting, and scratches due to friction, requiring periodic milling repair to restore its flatness and friction performance. Traditional manual positioning methods suffer from low efficiency and cumbersome operation, making it difficult to meet the demands of batch production and high precision. Therefore, non-contact automatic positioning technology based on machine vision has become a research hotspot. By deploying industrial cameras on milling equipment, images of the brake disc can be acquired in real time, and image processing algorithms can automatically identify the geometric features of the brake disc, thereby accurately locating the center position of the brake disc to guide the milling cutter for automated operation. The Hough transform algorithm is a classic algorithm for detecting geometric shapes in images and is often applied to positioning scenarios involving circular workpieces.

[0003] Standard Hough transform algorithms typically rely on binary edge maps extracted by upstream edge detection algorithms when locating wind turbine brake disc images. However, the surface conditions of wind turbine brake discs are extremely complex: under industrial light sources, the metal substrate of the brake disc is prone to large-area reflections or localized overexposure of highlights. These highly reflective areas form extremely bright pseudo-edges. When performing voting accumulation, the standard Hough transform algorithm treats all pixels extracted from the binary edge map as equally important, assigning them the same voting weight. This results in a large number of pseudo-edge points generated by reflective boundaries, which have the same interference ability in the voting space as real brake disc edge points. These erroneous votes from pseudo-edge points severely pollute the accumulation results in the Hough parameter space, causing the detected center position to deviate from the true value. Ultimately, this makes the visual positioning accuracy insufficient for subsequent milling operations, posing a risk of machining deviation or overcutting. Summary of the Invention

[0004] This invention provides a method and system for milling wind turbine brake discs based on visual positioning. It aims to solve the problem in the existing Hough transform algorithm that treats all edge points as equally important and assigns them the same weight for voting, which makes them extremely susceptible to false edge interference caused by strong reflections, overexposure, etc., thus seriously affecting the positioning accuracy.

[0005] In a first aspect, the present invention provides a method for milling wind turbine brake discs based on visual positioning, comprising: acquiring an image of a wind turbine brake disc and converting the image to an HSV color space to separate a brightness channel and a saturation channel; performing edge detection on the brightness channel to obtain the gradient magnitude of each pixel; calculating the final voting weight of each pixel, wherein the method for calculating the final voting weight includes: determining a brightness suppression factor based on the value of the pixel in the brightness channel, the brightness suppression factor being used to characterize the degree to which the pixel should be suppressed due to brightness abnormalities; and combining the image... The realism factor of a pixel is calculated by combining its brightness suppression factor and saturation channel value. This realism factor characterizes the credibility of the pixel belonging to the real surface of the brake disc. The realism factor, gradient magnitude, and grayscale range of its neighboring pixels are fused to obtain the final voting weight. A weighted Hough transform algorithm is used to perform circular detection on the image, where the final voting weight of each pixel is used as its accumulated value in the Hough accumulation space. The detected circular parameters are used as the positioning information of the wind turbine brake disc to guide subsequent milling. A unique voting weight is calculated for each pixel by combining its brightness, saturation, gradient, and neighborhood contrast. This weight effectively suppresses false edge interference caused by reflections and enhances the saliency of the real contour, thereby greatly improving the accuracy and robustness of brake disc center positioning.

[0006] Furthermore, the method for determining the brightness suppression factor includes: the brightness suppression factor is positively correlated with the brightness value of the pixel and negatively correlated with the absolute value of the difference between the brightness value of the pixel and the maximum brightness value of the entire image. By making the suppression factor positively correlated with the pixel brightness value and negatively correlated with the absolute value of the difference between the brightness value and the maximum brightness value of the image, a continuous and smooth suppression model is established. This approach can more precisely quantify the degree to which pixels should be suppressed due to brightness abnormalities, achieving differentiated suppression effects for reflective areas of different intensities, rather than indiscriminately eliminating them. This effectively suppresses false edges while more completely preserving effective image information, improving the precision and adaptability of the localization algorithm.

[0007] Furthermore, the verisimilitude factor of the pixel is calculated, including: exponentially decaying the brightness suppression factor to obtain a brightness confidence level; and multiplying the brightness confidence level by the pixel's saturation value to obtain the verisimilitude factor. By exponentially decaying the brightness suppression factor to obtain the brightness confidence level, and then multiplying it by the pixel's saturation value, information from both brightness and color dimensions is cleverly integrated. Since metal surfaces have high saturation, while highly reflective areas typically have low color saturation, this method can very effectively distinguish real metal surfaces from false edges (such as highlight areas), greatly improving the accuracy of the confidence assessment that a pixel belongs to the real surface of the brake disc, and further enhancing the ability to suppress false edges.

[0008] Furthermore, the product operation between the brightness confidence level and the pixel saturation value also includes: multiplying the pixel saturation value by the absolute value of the difference between the saturation value and the minimum saturation value of the entire image to obtain a saturation contribution value; the realism factor is the product of the brightness confidence level and the saturation contribution value. Compared to directly using the saturation value, this method amplifies the role of effective saturation information, resulting in a higher realism factor for pixels on realistic metal surfaces and a lower realism factor for low-saturation reflective areas, thereby achieving a stronger differentiation effect and further improving the robustness of the algorithm under complex lighting conditions.

[0009] Furthermore, obtaining the final voting weight includes: multiplying the realism factor, the gradient magnitude, and the grayscale range of its neighboring pixels to obtain the final voting weight.

[0010] Furthermore, a weighted Hough transform algorithm is used to perform circle detection on the image, including: constructing a Hough accumulator in a three-dimensional parameter space consisting of the center coordinates and the radius. Compared with some simplified two-dimensional Hough transforms (e.g., fixed radius), simultaneously searching and accumulating the center and radius in three-dimensional space allows this method to not only accurately locate the center but also accurately identify the actual size of the brake disc.

[0011] Furthermore, before employing the weighted Hough transform algorithm, the process includes setting a voting threshold and performing the weighted Hough transform algorithm only on pixels with a final voting weight greater than the threshold. By setting the threshold, only the most reliable edge points with sufficiently high final voting weights participate in the voting. This preprocessing step significantly reduces the amount of data involved in the Hough transform calculation, not only significantly improving the algorithm's operating efficiency but also further filtering out interference from low-weight noise points and uncertain pseudo-edge points, making the Hough accumulation space cleaner and the peaks more prominent, thereby improving the detection speed and the reliability of the final result.

[0012] Furthermore, the method for constructing the three-dimensional parameter space includes: setting the search range of the radius and the search range of the center coordinates based on the prior size information of the wind turbine brake disc.

[0013] Furthermore, edge detection is performed on the brightness channel, including: performing a convolution operation on the brightness channel image using the Sobel operator. The Sobel operator is a classic and computationally efficient edge detection algorithm that can effectively calculate the gradient magnitude of the image.

[0014] In a second aspect, the present invention also provides a vision-based positioning wind turbine brake disc milling system, comprising a processor and a memory, wherein the memory stores a computer program, and the processor executes the computer program to implement the vision-based positioning wind turbine brake disc milling method described in any of the preceding claims.

[0015] Beneficial Effects: By converting the image to HSV space, a final voting weight was constructed. This weight integrates four dimensions: a brightness suppression factor to suppress abnormal brightness, a realism factor to characterize surface authenticity, edge intensity gradient magnitude, and neighborhood grayscale range. This multi-dimensional weighting method can intelligently distinguish and suppress false edges caused by reflections, while enhancing the contribution of the real brake disc contour edge. This allows the subsequent weighted Hough transform to achieve significantly higher accuracy and robustness than existing technologies when performing circularity detection, thus providing a high-precision positioning reference for automated milling. Attached Figure Description

[0016] Figure 1 This is a schematic diagram illustrating the milling and positioning process of a wind turbine brake disc according to an embodiment of the present invention;

[0017] Figure 2 This is a flowchart illustrating the calculation of the final voting weight of each pixel according to an embodiment of the present invention. Detailed Implementation

[0018] The specific embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0019] like Figure 1 and Figure 2 As shown, S101: Acquire images of wind turbine brake discs and perform preprocessing.

[0020] Specifically, a vision acquisition system is deployed on the milling equipment. A high-definition industrial color camera is used, with its lens optical axis perpendicular to the surface of the wind turbine brake disc to be machined, to obtain a distortion-free top view of the brake disc in RGB digital image format.

[0021] Because the RGB color space couples color and brightness information together, it is not conducive to subsequent processing that distinguishes between changes in lighting and material colors. Therefore, it is necessary to convert the acquired RGB image to a different color space. As a preferred solution, this embodiment converts the RGB image to the HSV color space. The HSV space decomposes the image into three independent channels: hue (H), saturation (S), and lightness (V). The lightness (V) channel directly reflects the brightness information of the pixel, while the saturation (S) channel represents the purity of the color. This provides favorable conditions for subsequent differentiation between metallic substrates and high-reflection areas.

[0022] To further analyze the edge intensity information of the image, this embodiment applies an edge detection operator to the separated brightness (V) channel image. Specifically, the Sobel operator is used to obtain the gradient information of each pixel by calculating the first-order gradient of the image in the horizontal and vertical directions. The calculated gradient in the x-direction is... The gradient in the y-direction is .

[0023] S102: Analyze the color and gradient features of the brake disc image and adaptively calculate the final voting weight of each pixel.

[0024] This step aims to build a voting weighting system that can intelligently distinguish between real and pseudo edges. This process is specifically broken down into the following three sub-steps.

[0025] S1021: Analyze the characteristic performance of each pixel in the brightness channel to obtain the brightness suppression factor of each pixel.

[0026] Based on prior knowledge of the scenario, under uniform industrial lighting, the brightness value of the metal substrate of a wind turbine brake disc will be distributed within a relatively concentrated range. However, the brightness value of localized highlights or large reflective areas formed by direct reflection from the light source will be significantly higher, approaching or reaching saturation. The purpose of this sub-step is to construct a quantitative index for the preliminary identification and suppression of these artifact areas formed by reflection.

[0027] Specifically, the calculation of the brightness suppression factor for each pixel is based on the following principle: the higher the brightness value of a pixel, the more likely it is to belong to a highly reflective area, and therefore the stronger its suppression effect should be, i.e., the larger its brightness suppression factor should be. Based on this logic, the brightness suppression factor for each pixel can be calculated using the following formula: In the formula, Indicates the first Line number Brightness suppression factor for column pixels. Indicates the first Line number The value of each pixel in the brightness channel. This represents the maximum brightness value among all pixels in the entire image. For hyperparameters, Its existence is to prevent The value is 0. It is an exponential function with the natural constant e as its base.

[0028] From this formula, we can see that The larger the value, the more likely it is to be the first. Line number The higher the brightness value of a pixel, the greater the likelihood that the pixel belongs to a reflective area, and its brightness suppression factor should be greater. The smaller the value, the better. Line number A higher brightness value for a pixel indicates greater confidence, suggesting it's more likely to be a reflective area, and its brightness suppression factor should be higher. When a pixel's brightness value increases and approaches the global maximum value Vmax, the difference... The corresponding decrease, but due to the product term The overall trend of change allows the formula to output a significant suppression factor value for high-brightness pixels. Conversely, for pixels within the normal brightness range, its... Value and The difference is significant, and the calculated The value is relatively small. In this way, the brightness suppression factor can effectively quantify the degree to which pixels should be suppressed due to brightness abnormalities.

[0029] S1022: Analyze the characteristic performance of each pixel in the saturation channel, and combine it with the brightness suppression factor to obtain the realism factor of each pixel.

[0030] Understandably, relying solely on brightness information is insufficient to completely filter out all interference. For example, the brightness of some reflective areas may not reach extreme peaks, but their color saturation is often significantly lower than the inherent saturation of the brake disc's metal substrate. Therefore, this step introduces saturation information, combining it with the brightness suppression factor obtained in the previous step, to more accurately assess the credibility of pixels belonging to the real brake disc surface, i.e., the fidelity factor.

[0031] The underlying logic is as follows: the smaller the brightness suppression factor of a pixel, the more normal its brightness characteristics; and the higher its saturation value, the purer its color, consistent with the characteristics of metallic materials. Therefore, the greater the likelihood that this pixel belongs to a real brake disc area, and its realism factor should also be higher. Accordingly, the realism factor of each pixel can be calculated using the following formula: In the formula, Indicates the first Line number The realism factor of column pixels. Indicates the first Line number Brightness suppression factor for column pixels. Indicates the first Line number The saturation values ​​of each pixel. This represents the minimum saturation value among all pixels in the entire image. For hyperparameters, Its existence is to prevent The value is 0. It is an exponential function with the natural constant e as its base. This is the standard normalization function.

[0032] From this formula, we can see that The smaller the value, the better. Line number When analyzing column pixels based on brightness features, the greater the likelihood that a pixel belongs to a brake disc region with a real metal substrate, the greater its corresponding edge realism factor should be. The smaller the value, the better. Line number The less likely a column pixel is to belong to a brake disc area with a real metal substrate, the smaller its corresponding realism factor should be. The smaller the value, the better. Line number The less likely a column of pixels is to belong to a brake disc region with a real metal substrate, the higher its credibility, and the smaller its corresponding fidelity factor should be.

[0033] S1023: Calculate the final voting weight of each pixel.

[0034] The core of the Hough transform is edge detection. Therefore, the contribution of a pixel to the final center localization depends not only on whether it belongs to the real brake disc surface, but also on whether it lies on a significant edge. This step aims to fuse pixel realism information with edge strength information to calculate the voting weights used for the final weighted Hough transform.

[0035] The underlying logic is as follows: the larger the realism factor of a pixel and the larger its gradient magnitude, the more likely that pixel is the true physical edge of the brake disc, and its corresponding final voting weight should also be greater. Therefore, the final voting weight of each pixel can be calculated using the following formula: In the formula, Indicates the first Line number The final voting weight of each column pixel. Indicates the first Line number Gradient magnitude of column pixels, Indicates the first Line number The realism factor of column pixels, Indicates the first Line number The grayscale values ​​of pixels within the 8-neighborhood of a column pixel correspond to the range within the set.

[0036] As can be seen from the formula, in the formula The larger the value, the more likely it is to be the first. Line number The greater the probability that a column pixel belongs to the brake disc area, the greater its corresponding final voting weight should be. The larger the value, the more likely it is to be the first. Line number The greater the probability that a column pixel belongs to the edge of the brake disc area, the greater its corresponding final voting weight should be. The larger the value, the more likely it is to be the first. Line number The larger the gradient magnitude of a column pixel, the greater the reliability, indicating that the... Line number The greater the probability that a column pixel belongs to the edge of the brake disc area, the greater its corresponding final voting weight should be. The final voting weight is the product of three factors. This ensures that the voting weights are primarily derived from the actual brake disc area; This ensures that the weights are concentrated at the edge points; This further enhances the salience of the edges. Only when a pixel simultaneously meets the three conditions of realistic material, high edge strength, and strong local contrast can it obtain a higher final voting weight.

[0037] S103: Use the optimized weighted Hough transform algorithm to position the wind turbine brake disc and guide the milling process.

[0038] After calculating the final voting weight for each pixel, this step applies the weight to the voting accumulation process of the Hough transform to achieve accurate detection of the brake disc's center and radius.

[0039] In one embodiment, the specific implementation is as follows: Parameter space initialization: Based on the prior engineering drawing information of the brake disc, the radius search range of the circle to be detected is set. and the search range of the center As an example, the radius range can preferably be [50cm, 100cm], and the center coordinate range can be preset to [(660, 1260), (240, 840)]. Simultaneously, a three-dimensional Hough accumulator array is initialized. And set the value of all cells to zero. Weighted voting: Iterate through every pixel in the image. Set a voting threshold. In this embodiment, its The preferred empirical value is 0.6. If the pixel... The final voting weight is greater than If the condition is met, then the point is considered a valid edge point. For each valid edge point, calculate all possible edges that pass through that point and have a radius r within a certain range. Coordinates of the center of the circle within the range For each calculated accumulator unit, perform a weighted accumulation operation: Compared to the traditional Hough transform where the cumulative value is 1, this method uses the final voting weight of each pixel for accumulation, which amplifies the contribution of real edge points and effectively suppresses the contribution of pseudo edge points.

[0040] Traverse the entire 3D accumulator array to find the coordinates corresponding to the cell with the highest accumulated value. These coordinates represent the most probable center position and radius of the brake disc detected by the algorithm. The detected center coordinates and radius are used as a precise positioning reference and sent to the CNC system of the milling machine. Based on this, the CNC system executes an automatic tool setting program and plans a precise milling path, guiding the milling cutter to complete subsequent automated repair machining.

[0041] The present invention also provides a vision-based positioning-based wind turbine brake disc milling system. The system includes a processor and a memory, the memory storing computer program instructions. When the processor executes the computer program instructions, it implements the vision-based positioning-based wind turbine brake disc milling method according to the first aspect of the present invention.

[0042] The system also includes other components well known to those skilled in the art, such as communication buses and communication interfaces, the settings and functions of which are known in the art and therefore will not be described in detail here.

[0043] In this invention, the aforementioned memory can be any tangible medium containing or storing a program that can be used or combined with an instruction execution system, apparatus, or device. For example, a computer-readable storage medium can be any suitable magnetic or magneto-optical storage medium, such as Resistive Random Access Memory (RRAM), Dynamic Random Access Memory (DRAM), Static Random Access Memory (SRAM), Enhanced Dynamic Random Access Memory (EDRAM), High-Bandwidth Memory (HBM), Hybrid Memory Cube (HMC), etc., or any other medium that can be used to store desired information and can be accessed by an application, module, or both. Any such computer storage medium can be part of a device or accessible to or connected to a device. Any application or module described in this invention can be implemented using computer-readable / executable instructions stored or otherwise maintained on such a computer-readable medium.

[0044] The embodiments described above are merely examples of several implementations of the present invention, and while the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these modifications and improvements all fall within the scope of protection of the present invention.

Claims

1. A wind turbine brake disc milling method based on visual positioning, characterized in that, The method comprises the following steps: Collecting an image of a wind power brake disc, and converting the image to an HSV color space to separate a lightness channel and a saturation channel, and performing edge detection on the lightness channel to obtain gradient amplitudes of each pixel point; Calculating a final voting weight of each pixel point, wherein the calculation method of the final voting weight comprises: determining a lightness inhibition factor of the pixel point based on a value of the pixel point in the lightness channel, the lightness inhibition factor being used to represent a degree of inhibition of the pixel point due to abnormal brightness; combining the lightness inhibition factor of the pixel point and a value of the pixel point in the saturation channel to calculate a realness factor of the pixel point, the realness factor representing a credibility of the pixel point belonging to a real surface of the brake disc; and fusing the realness factor of the pixel point, the gradient amplitude, and a gray scale range of neighbor pixel points of the pixel point to obtain the final voting weight; Performing circular detection on the image by using a weighted Hough transform algorithm, wherein the final voting weight of each pixel point is used as an accumulated value of the pixel point in a Hough accumulator space, and a detected circular parameter is used as positioning information of the wind power brake disc to guide subsequent milling processing; The lightness suppression factor satisfies: ; in which, represents the lightness suppression factor of a pixel point in the i-th row and the j-th column; represents the lightness suppression factor of a pixel point in the i-th row and the j-th column; represents the lightness suppression factor of a pixel point in the i-th row and the j-th column; represents the lightness suppression factor of a pixel point in the i-th row and the j-th column; represents the lightness suppression factor of a pixel point in the i-th row and the j-th column; represents the lightness suppression factor of a pixel point in the i-th row and the j-th column; represents the maximum value of lightness in all pixel points of the whole image; is a hyperparameter; is an exponential function with the natural constant e as the base. The reality factor satisfies: ; wherein, represents a reality factor of a pixel point in the i-th row and the j-th column; represents a brightness suppression factor of a pixel point in the i-th row and the j-th column; represents a saturation value of a pixel point in the i-th row and the j-th column, represents a minimum saturation value in all pixel points of the whole image; is a standard normalization function.​​​​​​ 2. The method of claim 1, wherein, The final voting weight is obtained by: Performing product operation on the realness factor, the gradient amplitude, and a gray scale range of neighbor pixel points of the pixel point to obtain the final voting weight.

3. The visual positioning based windmill brake disc milling method according to claim 1, wherein, The circular detection on the image by using the weighted Hough transform algorithm comprises: Constructing a Hough accumulator in a three-dimensional parameter space composed of a center coordinate and a radius.

4. The method of claim 3, wherein, Before the circular detection on the image by using the weighted Hough transform algorithm, the method further comprises: Setting a voting threshold, and only performing the weighted Hough transform algorithm on pixel points with a final voting weight greater than the voting threshold.

5. The method of claim 3, wherein, The construction method of the three-dimensional parameter space comprises: According to prior size information of the wind power brake disc, setting a search range of the radius and a search range of the center coordinate.

6. The visual positioning based windmill brake disc milling method according to claim 1, wherein, The edge detection on the lightness channel comprises: Performing convolution operation on the lightness channel image by using a Sobel operator.

7. A vision positioning based windmill brake disc milling system comprising a processor and a memory, characterized in that, The memory stores a computer program, and the processor executes the computer program to implement the wind power brake disc milling method based on visual positioning according to any one of claims 1-6.

Images