Thread taper simulation measurement method based on machine vision

Abstract

The present application relates to a kind of based on machine vision thread taper simulation measurement method, comprising: image pre-processing: to thread profile original image is carried out image filtering, realize denoising and edge keeping;Sub-pixel profile enhancement: by analyzing the gradient information of pixel point to locate the area where intensity change is remarkable in image, and by interpolation technique to refine edge position, obtain sub-pixel level edge profile;Thread axial calculation: according to the feature point of the two sides of thread fitting sub-pixel profile, according to the thread straight line direction of two sides obtains thread axial direction;Taper parameter calculation: according to the feature point of the two sides of step three thread realizes the measurement of taper.For different thread tooth type establishes adaptive measurement mechanism, improves the generalization ability of method.Through sliding window fitting optimization profile point extraction process, realize to the enhancement analysis of local feature.Combined with virtual-real fusion calibration mechanism, realize the accurate identification of thread axial and radial feature, ensure the precision and stability of taper measurement.

Description

Technical Field

[0001] This invention belongs to the field of optical measurement technology, and in particular relates to a method for simulating and measuring thread taper based on machine vision. Background Technology

[0002] The mechanical properties of threaded connections are significantly correlated with the geometric accuracy of their taper parameters. Studies have shown that thread taper deviations can cause abnormal stress distribution on the contact surface, leading to localized stress concentration effects. This not only significantly reduces the fatigue life of the connection structure but may also cause phenomena such as preload decay, ultimately affecting the reliability of the mechanical system. In the field of taper detection, existing measurement technologies can be divided into two main categories based on physical contact characteristics: contact and non-contact. Their technical characteristics differ significantly in terms of measurement accuracy, operational efficiency, and adaptability to different working conditions.

[0003] Contact measurement technology is primarily based on the physical contact principle of mechanical probes, employing a dedicated thread taper detector for manual operation. This method, leveraging a mature mechanical positioning system, offers advantages in repeatability and adaptability to standard threads, meeting the testing requirements of various standard systems. However, limited by its point-to-point contact measurement mechanism, it suffers from inherent drawbacks such as low testing efficiency, high operator skill dependence, and limited measurement capabilities for micro-threads and irregularly shaped threads. Non-contact measurement technology is mainly built upon machine vision principles, and can be categorized into two technical routes based on the spatial dimension of the measurement data: two-dimensional image analysis and three-dimensional point cloud reconstruction. While the three-dimensional measurement method can acquire complete thread morphology information, its point cloud reconstruction algorithm still needs improvement in stability under complex conditions commonly encountered in industrial settings, such as mechanical vibration and ambient light interference. In contrast, high-precision two-dimensional vision systems based on CMOS image sensors, through optimized digital image processing algorithms, achieve real-time detection speeds while improving the signal-to-noise ratio, making them particularly suitable for batch inspection scenarios. Summary of the Invention

[0004] This invention addresses the shortcomings of current machine vision-based thread taper measurement methods in areas such as adaptability to complex working conditions, local feature representation, measurement model generalization, and the balance between real-time performance and accuracy. It proposes a machine vision-based simulation measurement method for thread taper. This method first starts with a high-resolution thread image and achieves high-precision sub-pixel contour extraction by analyzing pixel neighborhood gradient changes. Then, corresponding measurement models are constructed for different thread profiles, and a sliding window method is used to fit the sub-pixel contour points to extract key thread profile feature points and determine the thread axis direction. Finally, the thread taper is calculated by determining the minor diameter difference of corresponding thread profile feature points at different positions on both sides of the thread and their axial distance.

[0005] The technical solution adopted by this invention to solve the above problems is: a machine vision-based thread taper simulation measurement method, including...

[0006] Step 1: Image preprocessing: Image filtering is performed on the acquired original image of the thread contour. Bilateral filtering is used to achieve noise reduction and edge preservation.

[0007] Step 2, Subpixel Contour Enhancement: By analyzing the gradient information of pixels, regions with significant intensity changes in the image are located, and edge positions are refined using interpolation techniques to obtain subpixel-level edge contours;

[0008] Step 3: Thread axial direction calculation: Based on the feature points on both sides of the thread, fit the sub-pixel profile, and determine the thread axis direction based on the straight line direction of the thread on both sides.

[0009] Step 4: Taper parameter calculation: Based on the feature points on both sides of the thread in Step 3, the taper is measured.

[0010] Step one: The bilateral filtering is implemented by combining spatial domain weights and gray-level similarity weights for weighted filtering, so that the weighted average of pixels is affected not only by spatial distance but also by pixel intensity similarity.

[0011] More specifically, the mathematical expression for the weighted filtering that combines spatial domain weights and gray-level similarity weights is:

[0012] ;

[0013] ;

[0014] ;

[0015] in, and Each pixel and grayscale value, Represents all pixels within the filter window. It is spatial weight. This represents the Euclidean distance between pixels p and q. Control the size of the filtering window. It is the grayscale similarity weight. This represents the difference in grayscale between pixels p and q. The scope of influence of control intensity similarity It is the normalization factor, and e is the natural constant.

[0016] Step 2: First, the image is convolved using eight Sobel operators in different directions. The gradient magnitude of each pixel in each direction is calculated. By comparing the gradient magnitudes in each direction, the direction with the largest gradient magnitude is selected as the dominant gradient direction of that pixel, thereby achieving contour enhancement.

[0017] More specifically: the formulas for calculating the gradient magnitude and direction of a pixel are as follows:

[0018] ;

[0019] in, and These are the gradient values ​​of pixel (x, y) in the horizontal and vertical directions, respectively. After detecting edge points, the sub-pixel position is calculated using quadratic interpolation, as shown in the following expression:

[0020] ;

[0021] in, , , , These are the gradient magnitudes of the pixel (x, y) in the horizontal and vertical directions, respectively. α and β are interpolation weights set according to the gradient direction. sub With y sub This represents the coordinates of a subpixel.

[0022] Step three includes the following steps:

[0023] S3.1 Constructing the tooth root projection equation: Differentiated modeling is performed based on the geometric features of the thread type. For trapezoidal threads, the tooth profile is decomposed into several independent straight line segments, and the geometric relationship of each edge is described segment by segment by linear equation. Finally, the global analysis of the tooth profile projection is achieved by combining piecewise functions. For circular threads, a quadratic curve model is used for parametric fitting, and the parameters are calibrated by the least squares method.

[0024] S3.2 Sliding Window Fitting of Subpixel Contour: Given a set of data points for one side of the thread, each data point is projected onto the principal direction, and the projected set of points is used as the source data for the sliding window; a sliding window adapted to the thread is constructed, where k is the size of the sliding window, k = 0.75 × (t width / pix), t width Where is the compression width, and pix is ​​the image pixel size; the step size is 0.25 * t. width The target points within the window are fitted to a curve by constructing a system of equations, and an error equation is built. During the fitting process, if the error exceeds a given threshold... If the fitted curve is not found, the sliding window is used to fit and search for the next region until the contour points on one side are found and fitted.

[0025] S3.3 Feature Point and Axial Calculation: For the fitted tooth root projection equation, for one side of the thread, the feature point calculation can obtain multiple feature points of different tooth root projection equations. Based on the least squares fitting direction l, the feature points on the other side are also obtained in the same way, and the straight line direction l' is obtained. Based on l and l', the thread axis direction is obtained.

[0026] More specifically: Step 3.2, for the projection equation of the root of a circular thread, given the data point set P on one side of the thread. i (x i ,y i )Calculate the PCA direction of one side of the data Q calculates the center point of the data. , Then, after zero-mean normalization of the data, The reconstructed covariance matrix C:

[0027] ;

[0028] in, , , PCA direction It is the eigenvector corresponding to the largest eigenvalue of the covariance matrix.

[0029] Project each data point onto the principal direction Q, and obtain the projected coordinates. ,Depend on The set of points formed The sliding window serves as the source data for the sliding window; the sliding window fits the thread contour data, aiming to minimize the error between the model and the sub-pixel points of the contour. The target points within the window are used to construct a system of equations using the least squares method. To fit and solve for a, b, and c, the constructed error equation is:

[0030] ;

[0031] and Let be the coordinates of the i-th contour point under the window, and let a, b, and c be the coefficients of the quadratic function to be fitted. The prediction residual for the i-th contour point is used as the basis for summing the squared residuals of all points to obtain the overall fitting error. Finding the minimum value will give us a, b, and c.

[0032] First, for the data points The x-coordinate is centered to make the calculation more stable. The centering method is as follows:

[0033] ;

[0034] ;

[0035] in, For contour points The mean, Given the changed x-coordinates, according to the least squares method, the parabola parameters a, b, c satisfy:

[0036] ;

[0037] The parabola parameters a, b, and c can be calculated using the above formula. The error of the fitted curve is defined as the mean square error.

[0038] ;

[0039] During the fitting process, if the error is greater than , that is, error> If the fitted curve is not found, discard it and slide the window to fit and search for the next region until the contour points on one side are found and fitted.

[0040] Step four includes the following steps:

[0041] S4.1 Initial Position Determination: Determine the initial position for taper measurement, for feature points on both sides of the thread. and Calculate the distance from the feature point to the pipe opening MN respectively: and , making The initial value can be determined. and Location, The expression for the distance from a point to a line is a positive value;

[0042] S4.2 Taper Calculation: Thread taper is the change in diameter per unit length along the thread axis.

[0043] Advantages and beneficial effects of the present invention:

[0044] This system achieves high-precision measurement of thread taper using non-contact, high-resolution 2D imaging technology, avoiding potential workpiece surface damage caused by traditional contact measurements. It also improves adaptability to threads of different sizes and effectively enhances the system's robustness against workpiece vibration and jitter, thus ensuring a balance between measurement accuracy and real-time performance. By constructing a parametric simulation model, an adaptive measurement mechanism is established for different thread profiles (trapezoidal threads, round threads), improving the generalization ability of the measurement method and making it applicable to complex working conditions. Furthermore, a dynamically configurable measurement strategy is adopted, optimizing the contour point extraction process through sliding window fitting to achieve enhanced analysis of local features. Combined with a virtual-real fusion calibration mechanism, the system's ability to accurately identify axial and radial features of the thread is improved, ensuring high accuracy and stability in taper measurement. Attached Figure Description

[0045] Figure 1 This is a flowchart of the thread taper simulation measurement method of the present invention;

[0046] Figure 2 This is a schematic diagram of the contour fitting of the moving window in this invention, where k is the size of the sliding window and v is the feature point;

[0047] Figure 3 This is a schematic diagram for taper calculation, where A represents the axis direction, v is the feature point, d represents the distance between relative feature points, and L represents the spacing between adjacent feature points. Detailed Implementation

[0048] The present invention will be further described in detail below with reference to the embodiments. The embodiments are exemplary and intended to explain the present invention, but should not be construed as limiting the present invention.

[0049] This embodiment discloses a machine vision-based method for simulating and measuring thread taper, the process of which is as follows: Figure 1 As shown, the original image is acquired and image filtering is performed on the original image. Then, subpixel contour enhancement is performed on the filtered image. After enhancement processing, the thread axis is accurately calculated, and the multi-section taper position is determined. After comprehensive processing, the taper parameters are accurately calculated, and finally the taper parameters are determined.

[0050] Follow these steps:

[0051] Step 1: Image preprocessing optimization.

[0052] In the precise measurement of thread profiles, it is necessary to simultaneously meet the requirements of noise reduction and edge preservation. Bilateral filtering is a nonlinear, edge-preserving smoothing filter. Its core idea is to preserve as many significant edge features as possible in the image while removing noise. It is implemented by combining spatial domain weights and gray-level similarity weights for weighted filtering, ensuring that the weighted average of pixels is influenced not only by spatial distance but also by pixel intensity similarity, thus avoiding edge blurring. Its mathematical expression is:

[0053] ;

[0054] ;

[0055] ;

[0056] in, and These are the grayscale values ​​of pixels p and q, respectively. Represents all pixels within the filter window. It is spatial weight. This represents the Euclidean distance between pixels p and q. Control the size of the filtering window. It is the grayscale similarity weight. Indicates pixel grayscale difference, The scope of influence of control intensity similarity. It is a normalization factor that ensures that the filtered pixel values ​​remain within a reasonable range.

[0057] Step 2: Subpixel contour enhancement.

[0058] After the image preprocessing optimization stage, in order to extract high-precision contour boundaries from the 2D image, especially for the high-precision perception requirements of thread contours, it is necessary to locate regions with significant intensity changes in the image by analyzing the gradient information of pixels, and further refine the edge positions through interpolation techniques to obtain sub-pixel level edge contours. Specifically, firstly, the image is convolved using eight Sobel operators in different directions to calculate the gradient magnitude of each pixel in each direction. Then, by comparing the gradient magnitudes in each direction, the direction with the largest gradient magnitude is selected as the dominant gradient direction for that pixel. The formulas for calculating gradient magnitude and direction are as follows:

[0059] ;

[0060] in, and These are the gradient values ​​of pixel (x, y) in the horizontal and vertical directions, respectively. After detecting edge points, the sub-pixel position is calculated using quadratic interpolation, as shown in the following expression:

[0061] ;

[0062] in, , , , These are the gradient magnitudes of the pixel (x, y) in the horizontal and vertical directions, respectively. and The interpolation weights are set based on the gradient direction.

[0063] Step 3: Accurate calculation of thread axial direction.

[0064] Step 3.1 Constructing the tooth root projection equation. In thread geometry measurement, the construction of the tooth root projection equation needs to be based on differentiated modeling of the geometric characteristics of the thread type. For a trapezoidal thread composed of multiple discrete straight lines, its mathematical representation depends on the construction of piecewise linear functions. This is achieved by decomposing the tooth profile into several independent straight line segments and using linear equations. The geometric relationships of each edge are described segment by segment, and the global analysis of the tooth profile projection is finally achieved through a combination of piecewise functions. Because of the continuous and smooth curvature of the tooth profile, circular threads require parametric fitting using a quadratic curve model, typically in the form of a parabolic equation. Its parameters are calibrated using the least squares method to minimize the residuals between the model and the measured geometry, thereby achieving a high-precision approximation of the tooth profile.

[0065] Step 3.2 Sliding window fitting of sub-pixel contours. Taking the projection equation of the tooth root of a circular thread as an example, as follows... Figure 2 As shown, a set of data points on one side of the thread. Calculate the PCA direction of one side of the data Q calculates the center point of the data. , Then, after normalizing the data to zero mean, The reconstructed covariance matrix C:

[0066] ;

[0067] in, , , PCA direction This is the eigenvector corresponding to the largest eigenvalue of the covariance matrix. For each data point... Project onto the principal direction Q, and obtain the projected coordinates. ,Depend on The set of points formed As the source data for the sliding window. Figure 2 The middle part represents the sliding window that adapts to the thread, and k represents the size of the window.

[0068] ;

[0069] in, This represents the approximate width of the tooth shape. The image pixel size is used. A sliding window is used to fit the thread contour data, aiming to minimize the error between the model and the sub-pixel level of the contour. The step size is set to 0.25. The target points within the window are used to construct a system of equations using the least squares method. We will use this to fit and solve for a, b, and c. The constructed error equation is:

[0070] ;

[0071] First, for the data points The x-coordinate is centered to make the calculation more stable. The centering method is as follows:

[0072] ;

[0073] ;

[0074] in, For contour points The mean, Given the changed x-coordinates, according to the least squares method, the parabola parameters a, b, c satisfy:

[0075] ;

[0076] The parabola parameters a, b, and c can be calculated using the above formula. The error of the fitted curve is defined as the mean square error.

[0077] ;

[0078] During the fitting process, if the error is greater than ,Right now If the fitted curve fails, discard it. The sliding window then proceeds to fit and search for the next region until the contour points on one side are found and fitted.

[0079] Step 3.3 Feature point and axial calculation. For the fitted tooth base projection equation, Figure 2 The area in yellow represents the range of the projection equation for each tooth base. Figure 2 middle The feature points representing different tooth base projection equations are calculated as follows:

[0080] ;

[0081] ;

[0082] Where hj represents the coordinate of the Q direction of the j-th tooth floor projection equation, and kj represents the coordinate of the Q-vertical direction of the j-th tooth floor projection equation. Figure 2 In the case of a thread on one side, the characteristic points can be calculated to obtain a set of multiple features. Feature points ( Figure 2 (The red dot in the middle), according to the least squares fitting direction l, the feature point v' on the other side. m (h m ,k m The direction of the straight line, l', can be derived in the same way. The direction of the thread axis can then be determined from l and l'. ,like Figure 3 The threaded axis in the middle.

[0083] Step 4: Accurately calculate the taper parameters.

[0084] Step 4.1 Initial Position Determination. To simulate the taper measuring instrument and calculate the taper, the initial position for taper measurement needs to be determined. For the two feature points v1 and v'1, the distance from each feature point to the pipe opening MN is calculated. and , making This allows us to determine the initial positions of v1 and v'1.

[0085] Step 4.2 Taper parameter calculation. Thread taper can be defined as the change in diameter per unit length along the thread axis. For example... Figure 3 As shown, its mathematical representation can be described by the following geometric parameters:

[0086] Diameter measurement reference definition:

[0087] d1 is the thread root feature point and The Euclidean distance between them, i.e. .

[0088] d2 is the corresponding feature point outside the axial spacing ΔL. and The Euclidean distance between them, i.e. .

[0089] When two measurement sections with an axial distance of L are selected (corresponding to d1 and d2), the thread taper tpr can be characterized by the following formula:

[0090] ;

[0091] in, and The lengths of d1 and d2 in the direction perpendicular to the axis are represented as follows:

[0092] ;

[0093] ;

[0094] Therefore, the taper is expressed as:

[0095] ;

[0096] Based on the feature points on both sides of the thread calculated in step 3, the taper of the above formula can be measured. Table 1 shows the comparison between the taper experimental measurement results and manual measurement in this embodiment. To verify the reliability of the algorithm, 10 sets of thread specimens were used for comparative experiments. The absolute deviation between the taper value calculated by this method and the measurement results of the manual taper meter are all less than 0.001 mm, which meets the thread tolerance requirements. The average error of the other 10 sets of sample data is 3×10. -4 mm, standard deviation σ = 7.89 × 10 -4 mm, which also indicates that the algorithm has excellent repeatability.

[0097] Table 1

[0098]

[0099] This invention is not limited to the specific embodiments described above; its description is intended to illustrate and explain the technical solutions of this invention. Any obvious modifications or alternatives made based on the technical teachings of this invention should be considered to fall within the protection scope of this invention. The specific embodiments provided by this invention are intended to reveal best practice methods, enabling those skilled in the art to flexibly apply different implementation methods or alternatives based on the principles of this invention to achieve the same technical objectives.

Claims

1. A machine vision-based method for simulating and measuring thread taper, characterized by: steps include: Step 1: Image preprocessing: The acquired original image of the thread contour is filtered using bilateral filtering to achieve noise reduction and edge preservation. The bilateral filtering is implemented by weighted filtering by combining spatial domain weights and gray-level similarity weights. Step 2, Subpixel Contour Enhancement: Machine vision is used to analyze the gradient information of pixels to locate areas with significant intensity changes in the image, and interpolation techniques are used to refine the edge positions to obtain subpixel-level edge contours. First, eight Sobel operators in different directions are used to perform convolution operations on the image to calculate the gradient magnitude of each pixel in each direction. By comparing the gradient magnitudes in each direction, the direction with the largest gradient magnitude is selected as the dominant gradient direction of the pixel, thereby achieving contour enhancement. Step 3: Thread Axial Calculation: Based on the thread type, select feature points on both sides of the thread. Calculate the thread straight lines on both sides using sub-pixel contour fitting based on the feature points. Determine the thread axis direction based on the direction of the thread straight lines on both sides. This includes the following steps: S3.1 Constructing the tooth root projection equation: Differentiated modeling is performed based on the geometric features of the thread type. For trapezoidal threads, the tooth profile is decomposed into several independent straight line segments, and the geometric relationship of each edge is described segment by segment by linear equation. Finally, the global analysis of the tooth profile projection is achieved by combining piecewise functions. For circular threads, a quadratic curve model is used for parametric fitting, and the parameters are calibrated by the least squares method. S3.2 Sliding Window Fitting of Subpixel Contour: Given a set of data points for one side of the spiral, each data point is projected onto the principal direction, and the projected set of points is used as the source data for the sliding window; a sliding window adapted to the spiral is constructed. k The size of the sliding window. k = 0.75 × (t) width / pix) , t width This refers to the width of the molding process. pix Image pixel size; Step length step It is 0.25 t width The target points within the window are fitted to a curve by constructing a system of equations, and an error equation is constructed. During the fitting process, if the error is greater than... If the fitted curve is not found, the sliding window is used to fit and search for the next region until the contour points on one side are found and fitted. S3.3 Feature Point and Axial Calculation: For the fitted tooth root projection equation, for a single thread, the feature point calculation yields multiple feature points with different tooth root projection equations. Based on the least squares fitting direction... l The feature points on the other side are also derived in the same way, thus determining the direction of the straight line. l' ,according to l and l' Determine the direction of the thread axis; Step 4: Taper Parameter Calculation: The taper is measured based on the feature points on both sides of the thread from Step 3. This includes the following steps: S4.1 Initial Position Determination: Determine the initial position for taper measurement, for feature points on both sides of the thread. and Calculate the distance from the feature point to the pipe opening respectively. MN Distance: and , making , t width The width of the tooth profile. lth The distance from a point to a line is expressed as a positive value. S4.2 Taper Calculation: Thread taper is the change in diameter per unit length along the thread axis, and its mathematical representation is described by the following geometric parameters: Diameter measurement reference definition: For thread root feature points and The Euclidean distance between them ; Δ is the axial spacing L Corresponding feature points outside and Euclidean distance between them ; When two axial spacings are selected as L When measuring the cross section, corresponding , thread taper tpr Characterized by the following formula: in, and express , The length in the direction perpendicular to the axis is expressed as: Therefore, the taper is expressed as: in, A The axis direction vector. The vector perpendicular to the axis. L' for L Vertical distance.

2. The machine vision-based thread taper simulation measurement method according to claim 1, characterized in that: The mathematical expression for the weighted filtering that combines the spatial domain weights and gray-level similarity weights is: in, and Each pixel and grayscale value, Represents all pixels within the filter window. It is spatial weight. This represents the Euclidean distance between pixels p and q. Control the size of the filter window. It is the grayscale similarity weight. Indicates pixel grayscale difference, The scope of influence of control intensity similarity It is the normalization factor.

3. The machine vision-based thread taper simulation measurement method according to claim 2, characterized in that: The formulas for calculating the gradient magnitude and direction of a pixel are as follows: in, and These are pixels ( x , y The gradient values ​​in the horizontal and vertical directions are used to calculate the sub-pixel positions after edge point detection using quadratic interpolation, as shown in the following expression: Here, α and β are interpolation weights set according to the gradient direction.

4. The machine vision-based thread taper simulation measurement method according to claim 3, characterized in that: Step 3.2, for the projection equation of the root of a circular thread, given the data point set on one side of the thread. P i ( x i ,y i ) Calculate the data on one side PCA direction for PCA The direction is x Components on the axis, for PCA The direction is y Components on the axis, Q By calculating the center point of the data , Then, after zero-mean normalization of the data, The reconstructed covariance matrix C : in, , , , PCA direction The eigenvector corresponding to the largest eigenvalue of the covariance matrix. n The total number of data points. i For the first i Index of data points, for PCA The direction is x Components on the axis, for PCA The direction is y Components on the axis; After performing principal component analysis (PCA) on the data points, the PCA direction becomes the principal direction of the data points. PCA direction Q Projecting onto the top, the coordinates after projection ,Depend on The set of points formed As the source data for the sliding window, the first i A data point index; a sliding window is used to fit the thread contour data, aiming to minimize the error between the model and the sub-pixel points of the contour. The target points within the window are used to construct a system of equations using the least squares method. To perform fitting and solution a , b , c The constructed error equation is as follows: and The first one under the window i Coordinates of the contour points a , b , c Let be the coefficients of the quadratic function to be fitted. For the first i The predicted residuals for each contour point are summed, and the overall fitting error is obtained by summing the squared residuals for all points. Finding the minimum value will yield the result. a , b , c, N Total number of data points ； First, for the data points of x Centering the coordinates makes the calculation more stable. The centering method is as follows: in, For contour points The mean, For the changed x coordinate ，N Given the total number of data points, the parabola parameters are calculated using the least squares method. a , b , c satisfy: The parabola parameters can be obtained from the above formula. a , b , c The error of the above fitted curve is defined as the mean square error: During the fitting process, if the error exceeds a given threshold... error> If the fitted curve is not found, discard it and slide the window to fit and search for the next region until the contour points on one side are found and fitted.

Images