A method for extracting feature from workpiece shape measurement point sets based on pre-positioning

Abstract

In order to solve the problem that the amount of raw data measured during the contour detection of thin-walled parts with complex cross-sections is large and there is noise interference, it is usually necessary to preprocess the raw data first, and then extract features from the preprocessed data to reduce the calculation cost. However, since the raw measurement information will be lost after preprocessing, the final extracted features will also have information missing, and accurate detection cannot be achieved. The present invention provides a method for extracting features from workpiece contour measurement point sets based on pre-positioning, firstly, the raw measurement data is subjected to noise reduction filtering and downsampling preprocessing; then, the preprocessed data point set is used as the control point, and spline curve fitting is adopted to form the contour curve of the workpiece to be measured; finally, the key position on the workpiece is located according to the detection purpose and process requirements, and the raw measurement data at the key position is extracted. The present invention has the advantages of small calculation amount and no loss of measurement information at the key position, and can achieve quasi-accurate detection.

Description

Technical Field

[0001] The invention relates to a method for extracting features of a workpiece shape measurement point set based on pre-positioning, and is particularly suitable for extracting key position features of shape profile measurement data of a thin-walled workpiece with a complex cross section. Background Art

[0002] In recent years, the digitalization, intelligence and automation level of the manufacturing industry have been continuously developing. As the requirements for workpiece processing and forming accuracy continue to increase, higher requirements are placed on the efficiency and accuracy of workpiece processing and forming quality monitoring technology. Take the laser scanning measurement and detection of rolled flexible formed workpieces represented by metal sealing rings as an example:

[0003] As a new type of sealing structure, metal sealing rings have the advantages of good high vibration tracking, high vibration absorption and wear resistance, and are mainly used for sealing key parts such as aircraft engine turbines and casings. However, the cross-sectional features of such parts are complex, and the shape and size accuracy of the processing are required to be high. Therefore, it is necessary to detect the forming quality of the workpiece during the metal sealing ring rolling process.

[0004] The shape accuracy of the metal sealing ring cross section is an important indicator of its quality, but its cross-sectional shape is complex and the surface is bright. The measurement data of some positions on the workpiece is easy to be missing in the scanning results, which affects the final judgment of the forming quality. At the same time, the original measurement data point set obtained by high-precision measuring instruments is too dense and has noise interference. If it is directly used for fitting the contour of the workpiece, it will not only lead to overfitting, but also easily amplify local features, and the calculation will take a long time, greatly affecting the detection efficiency and reliability. Therefore, it is usually necessary to preprocess the original measurement data before extracting features, such as noise reduction, low-pass filtering and smoothing, and then extract features from the preprocessed data. However, feature extraction based on preprocessed data also has its limitations: in the process of preprocessing the data, the original measurement information is lost, so the extracted features are based on incomplete measurement data, so the final extracted features will also have information missing, and accurate detection cannot be achieved.

[0005] In summary, data processing and feature extraction of the measurement results of the workpiece contour can achieve efficient and accurate detection, which has important practical significance and practical value. Summary of the invention

[0006] In order to solve the problem that the amount of raw data measured is large and there is noise interference during the shape contour detection process of complex cross-section thin-walled parts, it is usually necessary to preprocess the raw data first, and then perform feature extraction on the preprocessed data to reduce the calculation cost. However, since the raw measurement information will be lost after preprocessing, the final extracted features will also have information missing, and accurate detection cannot be achieved. The present invention provides a workpiece shape measurement point set feature extraction method based on pre-positioning, which has the advantages of small calculation amount and no loss of measurement information at key positions, and can achieve quasi-accurate detection.

[0007] The inventive concept of the present invention is:

[0008] First, the original measurement data is preprocessed by denoising filtering and downsampling; then, the preprocessed data point set is used as the control point, and spline curve fitting is adopted to form the contour curve of the workpiece to be measured; finally, the key position on the workpiece is located according to the detection purpose and process requirements, and the feature extraction of the original measurement data at the key position is performed.

[0009] The technical solution of the present invention is:

[0010] A method for extracting feature of workpiece shape measurement point set based on pre-positioning, which is special in that it includes the following steps:

[0011] Step 1, obtaining the original measurement data of the cross-sectional profile of the workpiece to be measured;

[0012] Step 2: preprocessing the acquired raw measurement data to reduce high-frequency noise and data redundancy;

[0013] Step 3, based on the data preprocessed in step 2, fitting a fitting curve of the workpiece contour;

[0014] Step 4, locating the key position on the workpiece contour curve;

[0015] 4.1 According to the specific processing requirements of the workpiece, extract the relevant geometric features of each position of the workpiece contour fitting curve obtained in step 3, and set the threshold of the relevant geometric features;

[0016] 4.2 Compare the distribution of relevant geometric features of the workpiece contour fitting curve with the set threshold value to obtain the positions where the relevant feature values ​​on the workpiece contour fitting curve exceed the threshold value, and these positions are the key positions;

[0017] Step 5: Based on the key positions located in step 4, the corresponding original measurement data is extracted to solve the workpiece shape and size characteristics.

[0018] Furthermore, the preprocessing in step 2 includes data smoothing and downsampling performed in sequence.

[0019] Furthermore, step 3 specifically uses the data point set obtained after preprocessing in step 2 as control points, and adopts non-uniform rational B-spline curve fitting to obtain the workpiece contour fitting curve.

[0020] Furthermore, the relevant geometric features described in step 4.1 include the curvature distribution of the workpiece contour curve, the wall thickness distribution, the width, height, crest circle radius, trough circle radius, transition circle radius and / or trough center distance of the workpiece cross section.

[0021] Furthermore, in step 5, an algebraic circle fitting method is used to solve the workpiece shape dimensional characteristics.

[0022] Furthermore, in step 5, the Hyper fitting method is used to solve the workpiece shape and size characteristics.

[0023] The advantages of the present invention are:

[0024] 1. The present invention pre-positions the key positions on the workpiece contour curve that do not meet the processing design requirements, and only extracts features from the original measurement data of the key positions. Compared with the method of directly using all the original measurement data for feature extraction, the consumption of computing resources is greatly reduced.

[0025] 2. For key positions on the workpiece that do not meet the processing design requirements, the present invention directly uses the original measurement data to extract features. Since the feature extraction results at the key positions on the workpiece are based on all the measurement data, the information of the original measurement data is more fully utilized, thereby ensuring the accuracy of workpiece detection.

[0026] 3. The features extracted by the present invention can be used to evaluate the shape accuracy of the workpiece, optimize the processing parameters or analyze the processing process. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] Figure 1 The figure is a flow chart of the method of the present invention.

[0028] Figure 2 The following is a flow chart of a method for determining the quality of the outer contour of a metal sealing ring using the method of the present invention.

[0029] Figure 3 Schematic diagram of using two line laser profile sensors to obtain the original measurement data point set of the sealing ring contour.

[0030] Figure 4 yes Figure 3 Top view of the .

[0031] Figure 5 This is a comparison view of the original measurement data before and after smoothing, where: (a) is the data before smoothing, and (b) is the data after smoothing.

[0032] Figure 6 This is the visualization result of the uniformly downsampled data in the measurement coordinate system in the embodiment of the present invention.

[0033] Figure 7 It is a contour curve diagram obtained by non-uniform rational B-spline curve fitting in an embodiment of the present invention.

[0034] Figure 8 1 is a curvature distribution diagram of a metal sealing ring obtained in an embodiment of the present invention, wherein: (a) is the inner contour curvature distribution, and (b) is the outer contour curvature distribution.

[0035] Fig. 9 It is the extraction result of the original measurement data point set at the key position on the metal sealing ring profile fitting curve obtained in the embodiment of the present invention.

[0036] Fig.10 This is the fitting result of the original measurement data of the arc segment at the key position on the metal sealing ring obtained in the embodiment of the present invention based on the Hyper fitting method.

[0037] Description of reference numerals:

[0038] 1-Line laser profile sensor; 2-Base plate; 3-Metal sealing ring, 4-Structured light plane. DETAILED DESCRIPTION

[0039] The present invention will be further described below in conjunction with the accompanying drawings.

[0040] See also Figure 1 The method for extracting features of a workpiece shape measurement point set based on pre-positioning provided by the present invention comprises the following steps:

[0041] Step 1, obtaining the original measurement data of the cross-sectional profile of the workpiece to be measured;

[0042] The original measurement data point set of the inner and outer contours of the workpiece cross section is obtained by a dimension measuring instrument such as a high-precision laser scanner.

[0043] Step 2, preprocessing the acquired raw measurement data;

[0044] 2.1 Based on the Gaussian filtering method, the original measurement data point set of the workpiece contour obtained in step 1 is smoothed;

[0045] 2.2 Downsample the data point set obtained after smoothing.

[0046] Step 3, obtaining a fitting curve of the workpiece contour;

[0047] The data point set obtained after the preprocessing in step 2 is used as the control point, and the non-uniform rational B-spline curve is used to fit the workpiece contour fitting curve.

[0048] Step 4, locating the key position on the workpiece contour curve;

[0049] 4.1 According to the specific processing requirements of the workpiece, extract the relevant geometric features of each position of the workpiece contour fitting curve obtained in step 3 (including the curvature distribution of the workpiece contour curve, the wall thickness distribution, the width and height of the workpiece cross section; if the workpiece is a metal sealing ring, the relevant geometric features include the crest circle radius, the trough circle radius, the transition circle radius and the trough center distance of the metal sealing ring), and set the threshold of the relevant geometric features according to the process requirements;

[0050] 4.2 Compare the distribution of relevant geometric features of the workpiece contour fitting curve with the set threshold value, and obtain the positions where the relevant feature values ​​on the workpiece contour fitting curve exceed the threshold value. These positions are the key positions.

[0051] Step 5: Based on the key positions located in step 4, the corresponding original measurement data is extracted to solve the workpiece shape and size characteristics.

[0052] Example:

[0053] See also Figure 2 In this embodiment, the measurement of the sealing ring profile is taken as an example. The method of extracting the features of the sealing ring profile using the method of the present invention includes the following steps:

[0054] Step 1, obtaining original measurement data of the sealing ring profile;

[0055] Two linear laser profile sensors are installed in alignment to obtain the original measurement data point set of the sealing ring profile. Figure 3 , 4 As shown, the measuring device consists of two line laser profile sensors 1 and a base plate 2. The line laser profile sensor 1 itself is provided with a positioning hole and three threaded holes, which are positioned by positioning pins and fixed to the base plate 2 using bolts. The positional accuracy between the line laser profile sensors is ensured by installation and positioning.

[0056] Step 2, preprocessing the acquired raw measurement data;

[0057] 2.1 Perform Gaussian filtering and smoothing on the original measurement data;

[0058] There is a lot of high-frequency noise in the original measurement data point set, which is not conducive to subsequent data analysis and feature extraction. Therefore, it is necessary to perform Gaussian filtering on the original measurement data point set of the sealing ring contour to reduce the high-frequency noise in the point set. Figure 5 As shown;

[0059] 2.2 Downsample the smoothed data;

[0060] The number of original data point sets is huge, and there is a certain amount of data redundancy, which leads to a long time for subsequent data analysis and easily leads to problems such as overfitting. Therefore, the data is simplified and the model features are retained as much as possible. The commonly used simplified processing method is the downsampling method, which samples the original data according to certain rules, retains the sampled data, and ignores other data. Typical downsampling methods include uniform downsampling, rate sampling, chord deviation sampling and other methods. The specific downsampling method is determined by the characteristics of the data point set itself and the subsequent data processing required. The data point set of this embodiment is basically evenly distributed along the coordinate axis, and the mixed noise is random noise. The density of information on the point set is basically uniform, and there is no area that needs to be monitored, so uniform downsampling is most appropriate, and the algorithm is the simplest, easy to use and stable. The sampling factor is mainly determined by the complexity of the algorithm for subsequent data processing and the level of computing hardware. The more complex the subsequent processing and the worse the hardware conditions, the higher the sampling factor (the fewer points sampled).

[0061] According to the geometric characteristics of the metal sealing ring and the process requirements, this embodiment performs uniform downsampling with a sampling factor of 12 on the data after Gaussian filtering and smoothing. Specifically, all points in the original measurement data are arranged in sequence according to the horizontal coordinate values, and sampling is performed every 12 points. Figure 6 The sampling points obtained after uniformly downsampling the sealing ring contour measurement data are visualized in the measurement coordinate system.

[0062] Step 3, obtaining a fitting curve of the sealing ring profile;

[0063] The discrete data point set obtained after preprocessing is fitted by a uniform rational B-spline curve (a type of non-uniform rational B-spline curve, which has wider universality and flexibility, and is also more complex, while the uniform rational B-spline curve is relatively simple. For a relatively simple curve such as the sealing ring contour, the uniform rational B-spline curve fitting accuracy can meet the requirements) to obtain the sealing ring contour fitting curve;

[0064] This embodiment samples a cubic uniform rational B-spline curve for fitting. The spline curve is composed of a number of cubic uniform B-spline curve segments. The distance between the curve segments can reach C 2Continuity, that is, the connected curve segments are second-order continuous and differentiable at the connection points.

[0065] The specific calculation process for solving the fitting curve is as follows:

[0066] Assume that after downsampling in step 2, there are n points in the ordered point set used to calculate the curvature. The cubic uniform B-spline curve is composed of several cubic uniform B-spline curve segments, and the distance between the curve segments can reach C 2 Continuous. For a set of control vertex sequences {v1,v2,…,v n}, every four consecutive control vertices can determine a B-spline curve segment, and the parametric equation corresponding to the B-spline curve segment is:

[0067]

[0068] Where:

[0069] u-curve segment parameters;

[0070] A-curve segment coefficient matrix;

[0071] V i - A column vector of four consecutive control vertices;

[0072] in:

[0073]

[0074]

[0075]

[0076] Among them, the subscript i represents the i-th point in the ordered point set;

[0077] Write the B-spline curve equation as a parametric equation R i (u)=(x i (u),y i (u)) of the form, that is:

[0078]

[0079] in:

[0080] X i - A column vector consisting of the horizontal coordinates of four consecutive control vertices;

[0081] Y i - A column vector consisting of the ordinates of four consecutive control vertices;

[0082] Vector X i With vector Y i The specific mathematical expression is:

[0083] X i =[x i x i+1 x i+2 x i+3 ] T

[0084] Y i =[y i y i+1 y i+2 y i+3 ] T

[0085] Among them, x i Represents the horizontal coordinate of the i-th point in the ordered point set, y i Represents the ordinate of the i-th point in the ordered point set.

[0086] Figure 7 This is a sealing ring contour curve diagram obtained by fitting a cubic uniform rational B-spline curve based on a downsampled data point set in this embodiment.

[0087] Step 4, locating the key position on the fitting curve of the sealing ring profile;

[0088] 4.1 Calculate the curvature value of each position on the sealing ring profile fitting curve obtained in step 3, and set the curvature threshold of the sealing ring profile fitting curve according to the process requirements;

[0089] The curvature value at each position is calculated as:

[0090] The first-order derivative function and the second-order derivative function of the fitting curve parameter equation of the sealing ring profile obtained in step 3 with respect to the curve segment parameter u are respectively

[0091]

[0092]

[0093] Using the curvature formula based on parametric equations

[0094]

[0095] The curvature value of the sealing ring contour curve approximated based on the downsampled data set can be obtained. Figure 8 Graph 1 is the curvature distribution at various positions on the sealing ring profile curve fitting result obtained in this embodiment.

[0096] 4.2 Compare the calculated curvature values ​​of each position with the set curvature threshold value, and locate the positions on the sealing ring contour fitting curve where the curvature value exceeds the curvature threshold value. These positions are the key positions where further feature extraction of the original measurement data is required.

[0097] Figure 8 The peak of the curvature value on the curvature distribution curve corresponds to the inflection point of the metal sealing ring profile, that is, the key position on the metal sealing ring located in this embodiment. T , extract the contour point interval where the two maximum curvature values ​​are located (P A1 ,P A2 ) and (P B1 ,P B2 ). The endpoints of the two intervals are sampling points on the spline curve.

[0098] In this embodiment, according to the model design and process requirements of the sealing ring, C is selected. T =1 as the curvature threshold. A too large curvature threshold will result in too few contour points corresponding to each arc segment being extracted, thus affecting the subsequent circle fitting results; a too small curvature threshold will result in the inability to accurately extract the contour points corresponding to the arcs of the crest and trough segments of the sealing ring. In actual production practice, the setting of the threshold is related to many factors such as processing parameters, process requirements, and production inspection equipment. Usually, the recommendations in the process manual are used as the initial value, and the process personnel make adjustments based on the processing site conditions and their own experience.

[0099] Step 5, based on the key positions obtained in step 4, extract the corresponding original measurement data, and solve the fitting circle of the original measurement contour points of the sealing ring at these key positions (which can be determined by the center coordinates and radius);

[0100] Extract the key position of the metal sealing ring contour obtained in step 4, that is, the contour point interval (P A1 ,P A2 ) and (P B1 ,P B2 ) of the original measurement data point set (such as Fig. 9 As shown), let the original measurement data point set be an ordered point set {P1,P2,…,P n}, extract from it in (P A1 ,P A2 ) and (P B1 ,P B2 ), denoted as {A1,A2,…,A na} and {B1,B2,…,B nb}, which is the original measurement point set at the key position of the metal sealing ring.

[0101] An algebraic circle fit is performed on the original set of measured data points at these key locations.

[0102] The current circle fitting algorithms can be divided into two categories: geometric fit and algebraic fit. Geometric fit has higher solution accuracy than algebraic fit, but this method relies on iterative solution, has high computational cost, and the calculation process is not stable enough. The characteristics of algebraic fit are simplicity, stability, and fast operation speed, and it is more widely used in practical problems. Therefore, this embodiment chooses to use the algebraic fitting method.

[0103] Typical methods in algebraic fitting include Fitting, Pratt fit, Taubin fit and Hyper fit. The disadvantage of this method is that it tends to fit a circle with a smaller radius during the fitting process. Pratt fitting and Taubin fitting are different by changing the parameter constraints. Al-Sharadqah and Chernov proposed Hyper fitting based on Pratt fitting and Taubin fitting. This method is better than many existing algebraic fitting methods and is almost as good as some effective geometric fitting methods. It also has a better fitting effect on data distributed along circular arcs. Based on the above factors, this embodiment adopts the Hyper fitting method in the circle fitting part selection. The fitting results are as follows: Fig.10 shown.

[0104] The specific method of performing circle fitting on the data point set at the key position extracted by the Hyper fitting method in algebraic circle fitting in this embodiment is as follows:

[0105] Assume that the data point set to be fitted is {(x i ,y i )|(i=1,2,3,...,n)}, in this embodiment, they are the original measurement data point sets {A1,A2,...,A na} and {B1,B2,…,B nb}.remember The objective function of Hyper fitting is:

[0106]

[0107] Among them, A, B, C, and D are the model parameters to be solved, and A = (A, B, C, D) T , then the objective function F H The constraints are:

[0108] A T HA=1

[0109] in:

[0110]

[0111] In the above formula:

[0112]

[0113] Write the objective function in matrix form:

[0114] F H =||XA|| 2 =A T (X T X)A

[0115] in:

[0116]

[0117] Apply the Lagrange multiplier method to construct the formula F H =||XA|| 2 =A T (X T X) The optimization function of A:

[0118] F(A,η)=A T (X T X)A-η(A T HA-1)

[0119] Where:

[0120] η - Lagrange multiplier.

[0121] Taking the partial derivative of F(A,η) with respect to A, we can obtain the first-order necessary conditions for the optimal function solution:

[0122] (X T X) A-ηHA=0

[0123] Since H is reversible, multiply both sides of the above equation by H. -1 , we can get:

[0124] H -1 (X T X)A-ηA=0

[0125] In summary:

[0126] F H =A T (X T X)A=ηA T HA=η

[0127] From the above derivation, we can get the objective function F H The minimum value η of the matrix H -1 (X T X), and A is the eigenvector corresponding to η.

[0128] According to the formula (X T X) A-ηHA=0, we can get:

[0129] det(X T X-ηH)=0

[0130] make

[0131]

[0132] in, The same goes for this.

[0133] Depend on We can get:

[0134] det(M-ηH)=0

[0135] Where X T The solution vectors of the homogeneous linear equations of X-ηH and M-ηH are both coefficient vectors of the same circle.

[0136] Because this method is rotation and translation invariant, the calculation is simplified by translating the data set so that x = y = 0. At this time, the matrix M and the matrix H are:

[0137]

[0138] Expand det(M-ηH)=0 about η, and we can get

[0139] H(η)=4η 4 +c2η 2 +c1η+c0=0

[0140] in:

[0141]

[0142]

[0143]

[0144] Afterwards, Newton's method is used to search from the initial value η=0 to find the smallest positive root of H(η).

[0145]

[0146] After obtaining η, construct a homogeneous linear equation system (M-ηH)A=0 and solve the coefficient vector A. Then, the center (a, b) and radius R are calculated based on the coefficient vector A, and the relationship between them is:

[0147]

[0148]

[0149]

[0150] Translate the center coordinates of the circle to obtain the fitting center corresponding to the original data point set

[0151] The fitting circle extracted by the above method can be subsequently used to evaluate the shape accuracy of the workpiece, optimize the processing parameters or analyze the processing process. For example, when used to determine the shape accuracy of the workpiece, the curvature of the fitting circle at the key position obtained in step 5 is compared with the relevant threshold set in step 4.1. If the curvature of the characteristic circle at all key positions is within the threshold range, it means that the workpiece contour meets the process requirements, otherwise, it means that the workpiece contour does not meet the process requirements.

Claims

1. A method for extracting feature of workpiece shape measurement point sets based on pre-positioning, characterized in that: The following steps are involved: Step 1, obtaining the original measurement data of the cross-sectional profile of the workpiece to be measured; Step 2, preprocessing the acquired raw measurement data to reduce high-frequency noise and data redundancy of the data; the preprocessing includes data smoothing and downsampling performed in sequence; Step 3, based on the data preprocessed in step 2, fitting a fitting curve of the workpiece contour; Step 4, locating the key position on the workpiece contour curve; 4.1 According to the specific processing requirements of the workpiece, extract the relevant geometric features of each position of the workpiece contour fitting curve obtained in step 3, and set the threshold of the relevant geometric features; the relevant geometric features include the curvature distribution of the workpiece contour curve, the wall thickness distribution, the width, height, peak circle radius, trough circle radius, transition circle radius and / or trough center distance of the workpiece cross section; 4.2 Compare the distribution of relevant geometric features of the workpiece contour fitting curve with the set threshold value to obtain the positions where the relevant feature values ​​on the workpiece contour fitting curve exceed the threshold value, and these positions are the key positions; Step 5: Based on the key positions located in step 4, the corresponding original measurement data is extracted to solve the shape and size characteristics of the workpiece.

2. The method for extracting feature of workpiece shape measurement point sets based on pre-positioning according to claim 1, characterized in that: Step 3 specifically uses the data point set obtained after preprocessing in step 2 as control points, and adopts non-uniform rational B-spline curve fitting to obtain the workpiece contour fitting curve.

3. The method for extracting feature of workpiece shape measurement point sets based on pre-positioning according to claim 1, characterized in that: In step 5, the algebraic circle fitting method is used to solve the circular features in the workpiece shape dimensional features.

4. The method for extracting feature of workpiece shape measurement point sets based on pre-positioning according to claim 1, characterized in that: In step 5, the Hyper fitting method is used to solve the circular features in the workpiece shape dimensional features.

Images