CN104484508B - Optimizing method for noncontact three-dimensional matching detection of complex curved-surface part - Google Patents

Abstract

The invention discloses a method for non-contact three-dimensional matching detection and optimization of complex curved surface parts, which includes: sampling the CAD model of the complex curved surface part to be detected and the scanning model; aligning the center of gravity between the CAD model of the complex curved surface part and the scanning model; calculating the complex curved surface Corresponding point pairs between the CAD model and the scanned model; the conversion matrix is obtained based on the cuckoo algorithm; the scanned model of the complex curved surface part is updated according to the conversion matrix; the matching error calculation is performed between the scanned model of the complex curved surface part and the CAD model, and an iterative update is performed . The present invention can effectively solve the problems of slow calculation convergence speed and easy to fall into local optimum in the prior art, and is especially suitable for high-precision quality inspection of large complex curved surface parts such as aeroengine blades.

Description

Optimization method for non-contact 3D matching detection of complex curved surface parts

technical field

The invention belongs to the technical field of part processing and detection, and more specifically relates to a non-contact three-dimensional matching detection optimization method for complex curved surface parts.

Background technique

At present, there are mainly two types of methods for the detection of complex curved surface parts with higher precision: contact measurement method and non-contact measurement method. The contact measurement adopts the traditional three-coordinate measuring machine to sample the parts point by point, and the measurement accuracy is high, but the measurement efficiency of this method is low, and there are special requirements for the material of the measured parts. With the continuous development of computer vision and pattern recognition technology, non-contact measurement technology has been more and more widely used in the field of detection of complex curved surface parts due to its high measurement speed and accuracy.

Non-contact measurement uses an optical scanner to obtain scanning point sets at different angles, and then stitches these point sets together to obtain a complete part scanning point set, and obtains the error information of the part through the matching analysis of the scanning point set and the CAD model point set. In the standard detection process, the processing error is generally considered, but the influence of the measurement error on the detection result is ignored; in fact, if the error of the matching itself does not reach the ideal order of magnitude, the wrong measurement result will be obtained, especially at high In the matching detection process of precision complex curved surface parts, it is necessary to minimize the matching error.

The most widely used matching detection technology in the prior art is the ICP (Iterative Closest Point) algorithm proposed by Besl et al. in 1992. The famous Geomagic Studio software of the American Geomagic Company and the Scanalyze software developed by Stanford University are all based on the ICP algorithm. Point set matching. However, the ICP algorithm has requirements for the initial positions of the two point clouds when performing the matching, that is, it needs to perform a rough matching operation on the two point sets before the algorithm is executed, and the algorithm uses singular value decomposition or quaternion method to obtain the rigid body transformation matrix, it is easy to fall into local optimum, and it is not guaranteed to get the matching optimal solution for some complex models. CN101847262A discloses a fast three-dimensional point cloud search and matching method. The matching process includes low-precision search and high-precision search, which increases the complexity of the algorithm accordingly; CN102034104A discloses a feature line detection based on random sampling consistency in a three-dimensional point cloud method, the key link of this method is to obtain the feature line of the point cloud, but for some parts whose sharp features are not obvious or the curvature changes are relatively smooth, this detection method also has a large error.

Contents of the invention

Aiming at the above defects or improvement needs of the prior art, the present invention provides a non-contact three-dimensional matching detection optimization method for complex curved surface parts, in which the optimized cuckoo algorithm is used to design the complex curved surface parts by combining their own structural characteristics The three-dimensional matching detection process improves its key process at the same time. The test shows that it can effectively solve the problems of slow convergence and easy to fall into local optimum in the existing technology, and is especially suitable for high precision of large complex curved surface parts such as aeroengine blades. Quality testing purposes.

In order to achieve the above object, according to one aspect of the present invention, a non-contact three-dimensional matching detection optimization method for parts with complex curved surfaces is provided, which is characterized in that the method includes the following steps:

(a) Perform the first scan of the complex surface part to be detected to obtain multiple 3D measurement points, and then simplify the sub-sampling of the 3D measurement points and the corresponding discrete points of the CAD model, thereby obtaining two pieces as matching comparison objects point cloud; in addition, set the point cloud corresponding to the CAD model as The measurement point cloud of the corresponding part is set as Where j=1,...,n _q , i=1,...,n _p , n _q , n _p represent the total number of points in point cloud Q and point cloud P respectively;

(b) Calculate the respective centers of gravity of the two point clouds Q and P based on the following formulas (1) and (2) with Then move the two point clouds to the same XYZ coordinate system, and coincide their respective centers of gravity with the coordinate origin, where Represent each point in the two point clouds Q and P respectively, and each point is expressed in the form of a three-dimensional coordinate value:

(one)

(two)

(c) Keep the point cloud Q fixed, traverse each point in the point cloud P in turn, find the point with the closest relative distance in the point cloud Q as the corresponding point, and then form a point pair, that is, and Corresponding;

(d) Based on the cuckoo algorithm, in the above-mentioned XYZ coordinate system, the point cloud P is rotated around the xyz three-axis and the distance is translated along the xyz three-axis, randomly generates 6 variables, and then encodes them As an individual and form a cuckoo population; then, when performing iterative calculations, decode the individual to generate a rotation matrix containing and translation matrix Two-Part Transformation Matrix where the rotation matrix Represents the rotation matrix and translation matrix of 3 rows × 3 columns obtained by performing angular rotation along the xyz three axes of the point cloud P Represents the translation vector of 3 rows × 1 column obtained by performing distance translation along the xyz three axes of the point cloud P, and then updates the point cloud P according to the following functional formula (3):

(three)

(e) Calculate the matching error value ER between point cloud P and point cloud Q based on the following formula (4):

(Four)

in, Represents the points in the point cloud Q The closest corresponding point, and each point is expressed in the form of three-dimensional coordinates; ||X|| represents the absolute value of X; and when the calculated matching error value is greater than the preset matching threshold, based on step (d ) function iteratively updates the point cloud P until the calculated matching error value is less than or equal to the matching threshold, and obtains the optimal transformation matrix under the current correspondence;

(f) update the point cloud P based on the current optimal transformation matrix obtained in step (e), and then calculate the matching error value between the updated point cloud P and point cloud Q based on the formula (4) , and when the calculated matching error value is greater than the matching threshold, return to step (c) to recalculate the corresponding point pairs of point cloud Q and point cloud P, and continue to loop until the calculated matching error value is less than or When it is equal to the matching threshold, the entire three-dimensional matching detection optimization process is thus completed.

As a further preference, in step (a), the sub-sampling is performed using approximate model theory.

As a further preference, in step (c), the K-D tree method is preferably used to speed up the search for point pairs between the two point clouds Q and P.

As a further preference, the complex curved surface part is preferably an aeroengine blade.

According to another aspect of the present invention, another non-contact three-dimensional matching detection optimization method for parts with complex curved surfaces is provided, which is characterized in that the method includes the following steps:

(i) Perform the first scan of the complex surface part to be detected to obtain multiple 3D measurement points, and then resample the 3D measurement points and the corresponding discrete points of the CAD model to simplify, thereby obtaining two pieces of points as matching comparison objects cloud; in addition, the point cloud corresponding to the CAD model is set as The measurement point cloud corresponding to the complex surface part is set as Where j=1,...,n _q , i=1,...,n _p , n _q , n _p represent the total number of points in point cloud Q and point cloud P respectively;

(ii) Calculate the respective centers of gravity of the two point clouds Q and P based on the following formulas (1) and (2) with Then move the two point clouds to the same XYZ coordinate system, and coincide their respective centers of gravity with the coordinate origin, where Represent each point in the two point clouds Q and P respectively, and each point is expressed in the form of a three-dimensional coordinate value:

(one)

(two)

(iii) Keep the point cloud Q fixed, traverse each point in the point cloud P in turn, find the point with the closest relative distance in the point cloud Q as the corresponding point, and then form a point pair, that is, and Corresponding;

(iv) Based on the cuckoo algorithm, in the above-mentioned XYZ coordinate system, the point cloud P is rotated around the xyz three-axis and the distance is translated along the xyz three-axis, randomly generates 6 variables, and then encodes them As an individual and form a cuckoo population; then, when performing iterative calculations, decode the individual to generate a rotation matrix containing and translation matrix Two-Part Transformation Matrix where the rotation matrix Represents the rotation matrix and translation matrix of 3 rows × 3 columns obtained by performing angular rotation along the xyz three axes of the XYZ coordinate system respectively. Represents the translation vector of 3 rows × 1 column obtained by performing distance translation along the xyz three axes of the XYZ coordinate system respectively, and then updates the point cloud P according to the following functional formula (3):

(three)

(v) Calculate the matching error value ER between point cloud P and point cloud Q based on the following formula (4):

(Four)

in, Represents the points in the point cloud Q The closest corresponding point, and each point is expressed in the form of a three-dimensional coordinate value; ||X|| represents the absolute value of X; and when the calculated matching error value is greater than the preset matching threshold, based on step (iv ) performs an iterative update on point cloud P, and returns to step (iii) to recalculate the corresponding point pairs between point cloud Q and point cloud P, and then continue to loop until the calculated matching error value is less than or equal to When the matching threshold is reached, the entire three-dimensional matching detection optimization process is thus completed.

Generally speaking, compared with the prior art, the above technical solution conceived by the present invention mainly has the following technical advantages:

1. By adopting the optimized cuckoo algorithm to design the three-dimensional matching detection process of complex surface parts, it can make full use of its characteristics of fast convergence speed, good robustness and not easy to fall into local optimum. The test shows that it can obtain high precision and high precision. 3D matching results of efficiency;

2. In particular, the design angle rotation variable and distance translation variable in the specific process flow are used as individuals to form a population, and its specific optimization strategy is designed at the same time, which can fit the structural characteristics of complex curved surface parts. At the same time, the point cloud There is no requirement for the initial position, and there is no need to perform a rough matching operation before executing the algorithm, and it can further improve the accuracy and efficiency of matching detection;

3. Since the approximate model theory is used to resample the two point sets before the matching operation, the accuracy of the matching result can be ensured while reducing the complexity of the model; in addition, the center of gravity of the two point sets is aligned before the matching operation operation, which can effectively reduce the calculation amount of calculating the translation matrix;

4. In the present invention, an optimization strategy with inner and outer loop structures is specially proposed, wherein the inner loop can obtain the optimal transformation matrix under the current corresponding situation, and when the inner loop ends, the outer loop gives a A new point pair correspondence situation, the inner loop is repeatedly executed until the outer loop meets the termination condition. The test shows that this optimization strategy can significantly minimize the difference in the spatial position of the point cloud point pair, and further improve the result of 3D matching detection. precision.

Description of drawings

Fig. 1 is the basic flowchart of the non-contact three-dimensional matching detection method conceived according to the present invention;

Fig. 2 is a flow chart of detection optimization according to the first preferred embodiment of the present invention;

Fig. 3 is a flow chart of detection optimization according to the second preferred embodiment of the present invention;

FIG. 4 is a subsampling diagram for exemplary display of initial positions of two point sets;

Fig. 5 is a schematic diagram for exemplary displaying when the centers of gravity of two point sets are aligned;

Fig. 6 is a diagram for displaying matching results obtained according to the detection and optimization process shown in Fig. 2;

FIG. 7 is a diagram for displaying matching results obtained according to the detection optimization process shown in FIG. 3 .

detailed description

In order to make the object, technical solution and advantages of the present invention clearer, the present invention will be further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described here are only used to explain the present invention, not to limit the present invention. In addition, the technical features involved in the various embodiments of the present invention described below can be combined with each other as long as they do not constitute a conflict with each other.

In the existing technology, the geometric measurement of the workpiece is often carried out when the CAD model is known. The actual element information of the workpiece is obtained through the multi-sensor measurement system, and the required geometric contour is extracted through data processing. Match detection, and finally complete the judgment of the error. Matching detection is to find an optimal rigid body transformation between two image point sets, so that each point in one image point set matches the corresponding point in the other point set, and correspondingly reduces the influence of measurement errors on the part detection results.

Therefore, for the three-dimensional matching detection of workpieces, especially complex curved parts, considering that the cuckoo algorithm searches the solution space based on the Lévy flight principle, it has good global optimization capabilities. The present invention starts from the geometric shape of the point set, and searches for the optimal rotation and translation matrix between two point sets through the iteration of the optimized cuckoo algorithm, which can achieve fast convergence speed and good robustness, and is suitable for high-precision complex curved surface parts. 3D matching detection and other effects.

Fig. 1 is a basic flow chart of the non-contact three-dimensional matching detection method conceived according to the present invention. Below we will take a certain type of aeroengine blade model as an example for specific description.

First, perform the first scan of the aero-engine blade model to be tested to obtain multiple 3D measurement points. In order to simplify the analysis, the approximate model theory can be used to simplify the scanning model by subsampling. For example, 1000 points are taken for both models for easy understanding. point analysis (of course, it is also feasible that the number of analysis points is not equal in the actual situation), thus obtaining two point clouds as matching comparison objects; in addition, the point cloud corresponding to the CAD model can be set as The point cloud corresponding to the complex surface part is set as Where i=1,...,1000, where j is equal to i, the simplified point set is shown in Figure 4, the solid point set in the figure is the CAD model, and the hollow circle point set is the scanning model of the complex curved surface part.

Then, the respective centers of gravity of the two point clouds Q and P can be calculated based on the following formulas (1) and (2) with Then move the two point clouds to the same XYZ coordinate system, and coincide their respective centers of gravity with the coordinate origin, where Represent each point in the two point clouds Q and P respectively, and each point is expressed in the form of a three-dimensional coordinate value:

(one)

(two)

The two point clouds after the center of gravity are aligned are shown in Fig. 5.

Then, keep the point cloud Q fixed, traverse each point in the point cloud P in turn, and find the point with the closest relative distance in the point cloud Q as the corresponding point, and then form a point pair, that is, and Corresponding;

Next, based on the cuckoo algorithm, randomly generate three variables that rotate around the xyz coordinate axis and three variables that translate around the xyz coordinate axis in the above-mentioned XYZ coordinate system, encode these six variables as individuals and form a cuckoo population; In other words, the variables are randomly obtained by performing angle rotation and distance translation on the point cloud P, and these six variables are encoded to form the individuals of the cuckoo population; then, when performing iterative calculations, the individuals are Decoding generates a rotation matrix and translation matrix Two-Part Transformation Matrix where the rotation matrix Represents the 3×3 rotation matrix and translation matrix obtained by performing angle rotation along the xyz three axes of the XYZ coordinate system respectively for the point cloud P Represents the 3×1 translation vector obtained by performing distance translation along the xyz three axes of the XYZ coordinate system respectively. For example, in order to test the matching accuracy of the present invention, for example, the scanning The model is rotated 20 degrees around the xyz coordinate axis, translated 100 mm in turn, and then matched with the CAD model. Among them, all points in P are updated, and formula (3) represents the updated representation of point cloud P:

(three)

The above formula means that the transformation matrix obtained by the cuckoo algorithm is used to update the scanning model of the complex surface part.

In the present invention, it is preferable to use the KD tree method to accelerate the search for corresponding point pairs between two point sets, assuming that after calculation, in the point cloud Q and in the point cloud P Correspondingly, the following formula (4) can be used to calculate the matching error ER:

(Four)

in, Indicates in the CAD model and in the scanned model The point is the closest corresponding point, and each point is expressed in the form of a three-dimensional coordinate value, Represents the 3×3 rotation matrix and 3×1 translation vector obtained by performing angular rotation and distance translation along the xyz three axes of the XYZ coordinate system respectively, and ||X|| represents the absolute value of X value; the loop termination condition of the present invention can be set until the calculated matching error is less than or equal to the preset matching threshold, and of course it can also be set when the number of iterations is equal to the specified number of loops.

Figure 2 and Figure 3 respectively show the optimization strategies formulated according to two preferred embodiments of the present invention.

For the optimization strategy shown in Figure 2, there are inner and outer loop structures in the iterative process. The purpose of the inner loop is to obtain the optimal transformation matrix under the current corresponding situation. After the inner loop ends, the outer loop gives a A new point-to-correspondence situation, the inner loop is repeatedly executed until the outer loop meets the termination condition, specifically: when the calculated matching error value is greater than the preset matching threshold, the population is executed based on the optimization function model Update iteratively until the calculated matching error value is less than or equal to the matching threshold, and obtain the current corresponding optimal transformation matrix; then, update the point cloud P based on the obtained current optimal transformation matrix, and then similarly Calculate the matching error value between the updated point cloud P and point cloud Q based on the matching error calculation formula, and when the calculated matching error value is greater than the matching threshold, return to the point pair search step to recalculate the two points The corresponding relationship between clouds continues to perform an external loop until the calculated matching error value is less than or equal to the matching threshold, thereby completing the entire three-dimensional matching detection optimization process.

For the optimization strategy shown in Figure 3, there is only one layer of cyclic structure in the iterative process, specifically: initialize the cuckoo population, and calculate the matching error between point cloud P and point cloud Q based on the matching error calculation formula value ER, and when the calculated matching error value is greater than the preset matching threshold value, iteratively update the population based on the optimization function model, and directly return to the point to find the steps to cycle until the calculated matching error value When it is less than or equal to the matching threshold, the entire three-dimensional matching detection optimization process is thus completed.

The matching results based on the above two optimization strategies are shown in Figure 6 and Figure 7, respectively. In the figure, the CAD model is represented by the solid point set, and the scanning model is represented by the hollow circle point set. The higher the coincidence degree of each point in the two point sets, the The better the match. It can be seen from the matching results in Fig. 6 and Fig. 7 that both the two optimization strategies proposed by the present invention can obtain higher-precision matching results, and can minimize the influence of measurement errors on the part detection results; especially, the inner and outer two The optimization strategy of the layer cycle can significantly minimize the difference in the spatial position of the point cloud point pairs, and correspondingly further improve the accuracy of the 3D matching detection results.

It is easy for those skilled in the art to understand that the above descriptions are only preferred embodiments of the present invention, and are not intended to limit the present invention. Any modifications, equivalent replacements and improvements made within the spirit and principles of the present invention, All should be included within the protection scope of the present invention.

Claims (5)

1. A non-contact three-dimensional matching detection optimization method for a complex curved surface part is characterized by comprising the following steps:

(a) performing primary scanning on a complex curved surface part to be detected to obtain a plurality of three-dimensional measuring points, and then respectively performing secondary sampling on the three-dimensional measuring points and corresponding discrete points of a CAD model to simplify the three-dimensional measuring points and the corresponding discrete points of the CAD model, thereby obtaining two pieces of point clouds serving as matching comparison objects; in addition, the point cloud corresponding to the CAD model is set asThe measurement point cloud corresponding to the complex curved surface part is set asWherein j is 1_q，i＝1,...,n_p，n_q、n_pRespectively representing the total number of points of the point cloud Q and the point cloud P;

(b) the respective barycenters of the two point clouds Q, P are calculated based on the following formulas (I) and (II)Andthen, two pieces of point clouds are moved to the same XYZ coordinate system, and the respective centers of gravity of the point clouds coincide with the origin of coordinates, whereinEach point in the two-piece point cloud Q, P is represented separately, and each point is represented in the form of three-dimensional coordinate values:

(c) keeping the point cloud Q fixed, sequentially traversing each point in the point cloud P, searching the point with the closest relative distance in the point cloud Q as a corresponding point, and further forming a point pair, namely obtaining the point pairAndcorresponding;

(d) based on the cuckoo algorithm, in the XYZ coordinatesRandomly generating 6 variables by performing angle rotation around an xyz three-axis and distance translation along the xyz three-axis for the point cloud P in the system, and then coding the variables as individuals to form a cuckoo population; then, when iterative calculation is performed, the individual is decoded to generate a rotation matrixAnd translation matrixTwo-part conversion matrixWherein the rotation matrixA rotation matrix and a translation matrix representing 3 rows and × 3 columns obtained by performing angular rotations of the point cloud P along the xyz three axesA translation vector representing 3 rows × 1 columns obtained by performing distance translation of the point cloud P along the xyz three axes, respectively, and then performing update on the point cloud P according to the following functional formula (three):

(e) calculating a matching error value ER between the point cloud P and the point cloud Q based on the following formula (IV):

wherein,representing the neutralization point in the point cloud QThe corresponding point with the nearest distance is represented in the form of three-dimensional coordinate values; | X | | denotes taking the absolute value of X; when the calculated matching error value is larger than a preset matching threshold value, performing iterative updating on the point cloud P based on the function formula in the step (d) until the calculated matching error value is smaller than or equal to the matching threshold value, and obtaining an optimal conversion matrix under the current correspondence;

(f) updating the point cloud P based on the current optimal conversion matrix obtained in the step (e), then calculating a matching error value between the updated point cloud P and the point cloud Q based on the formula (IV), returning to the step (c) to recalculate the corresponding point pair of the point cloud Q and the point cloud P when the calculated matching error value is larger than the matching threshold value, and continuing to circulate until the calculated matching error value is smaller than or equal to the matching threshold value, thereby completing the whole three-dimensional matching detection optimization process.

2. The three-dimensional matching detection optimization method of claim 1, wherein in step (a), said sub-sampling is performed using an approximate model theory.

3. The three-dimensional matching detection optimization method of claim 1 or 2, wherein in step (c), a K-D tree method is used to speed up the search for the point pairs between the two point clouds Q, P.

4. The three-dimensional matching detection optimization method of claim 1, wherein the complex curved surface part is an aircraft engine blade.

5. A non-contact three-dimensional matching detection optimization method for a complex curved surface part is characterized by comprising the following steps:

(i) performing first scanning on a complex curved surface part to be detected to obtain a plurality of three-dimensional measuring points, and then performing three-dimensional scanning on the complex curved surface partRespectively carrying out secondary sampling on the dimension measuring points and the corresponding discrete points of the CAD model to simplify the dimension measuring points and the corresponding discrete points of the CAD model, thereby obtaining two pieces of point clouds serving as matching comparison objects; in addition, the point cloud corresponding to the CAD model is set asThe measurement point cloud corresponding to the complex curved surface part is set asWherein j is 1_q，i＝1,...,n_p，n_q、n_pRespectively representing the total number of points of the point cloud Q and the point cloud P;

(ii) the respective barycenters of the two point clouds Q, P are calculated based on the following formulas (I) and (II)Andthen, two pieces of point clouds are moved to the same XYZ coordinate system, and the respective centers of gravity of the point clouds coincide with the origin of coordinates, whereinEach point in the two-piece point cloud Q, P is represented separately, and each point is represented in the form of three-dimensional coordinate values:

(iii) keeping the point cloud Q fixed, sequentially traversing each point in the point cloud P, searching the point with the closest relative distance in the point cloud Q as a corresponding point, and further forming a point pair, namely obtaining the point pairAndcorresponding;

(iv) based on a cuckoo algorithm, randomly generating 6 variables by performing angle rotation on the point cloud P around the XYZ three axes and distance translation along the XYZ three axes in the XYZ coordinate system, and then coding the variables to serve as individuals to form a cuckoo population; then, when iterative calculation is performed, the individual is decoded to generate a rotation matrixAnd translation matrixTwo-part conversion matrixWherein the rotation matrixA rotation matrix and a translation matrix representing 3 rows and × 3 columns obtained by performing angular rotations of the point cloud P along the XYZ three axes of the XYZ coordinate systemTranslation vectors of 3 rows × 1 columns obtained by performing distance translation of the point cloud P along the XYZ three axes of the XYZ coordinate system, respectively, and then, updating the point cloud P is performed in accordance with the following functional expression (three):

(v) calculating a matching error value ER between the point cloud P and the point cloud Q based on the following formula (IV):

wherein,representing the neutralization point in the point cloud QThe corresponding point with the nearest distance is represented in the form of three-dimensional coordinate values; | X | | denotes taking the absolute value of X; and when the calculated matching error value is greater than a preset matching threshold value, performing iterative update on the point cloud P based on the functional formula in the step (iv), returning to the step (iii) to recalculate the corresponding point pair of the point cloud Q and the point cloud P, and then continuing to circulate until the calculated matching error value is less than or equal to the matching threshold value, thereby completing the whole three-dimensional matching detection optimization process.