An automatic polishing planning and decision method for aircraft structural parts

Abstract

The present application belongs to the field of aviation manufacturing, and specifically relates to an automatic polishing planning and decision method for an aircraft structural member, comprising the following steps: positioning of a structure member to be polished and a web area; single-web multi-angle imaging image combination judgment; identification and positioning of a transition area of a tool mark; regional refinement data scanning; final web area polishing boundary data acquisition; first web polishing area planning; spatial non-interference distance measurement; determination of remaining web to be polished based on a cyclic planning method; generation of a polishing program for polishing; polishing treatment; and polishing area re-inspection. Compared with a traditional manual polishing method, the polishing area planning strategy proposed in the present application can fully consider the positions of adjacent polishing areas and the corresponding attributes of polishing features, can obviously reduce the generation of invalid polishing paths, and can effectively improve the efficiency and quality of the whole-process polishing of an aviation structural member.

Description

TECHNICAL FIELD

[0001] The application belongs to the field of aviation manufacturing, and particularly relates to an automatic polishing planning and decision-making method for an aircraft structural member. BACKGROUND

[0002] An aircraft structural member is a core structure for composing a skeleton and a framework of an aircraft. Based on a designed model, accurate machining and manufacturing of the structural member cannot be completed without a series of core manufacturing links, which include preparation of a numerical control machining program of the structural member, rough and fine machining of the structural member based on a machine tool, and polishing of the machined structural member. However, the current polishing method of the structural member is mainly completed by manual work. With the development of society, the number of fitters willing to engage in the polishing work of the structural member is significantly reduced, which brings challenges to the batch production delivery of the aircraft. Therefore, it is particularly urgent and necessary to develop and research a technology capable of realizing automatic polishing of the structural member.

[0003] At present, in the polishing field, the technology for realizing automatic polishing is relatively mature and widely used in the automobile field. The main reason is that the automobile is a product of batch production, the corresponding structural member type is single, and the size and shape information are relatively fixed, which greatly reduces the polishing difficulty and enables batch polishing work to be completed based on a pre-set polishing path, parameter and trajectory. However, the structural member used on the aircraft has the characteristics of multiple varieties and small batches, and the conventional automobile polishing technology is difficult to be applied to this kind of scene. Meanwhile, due to the complexity of the configuration of the aircraft structural member, the polishing difficulty is increased to a certain extent. In view of the fact that the polishing technology in the automobile field is difficult to be applied to the polishing of the aircraft structural member, it is urgent to design a new and effective polishing method and technology for the aircraft structural member.

[0004] In the aircraft structural members, the frame and beam structural members account for a large proportion, and the corresponding importance is higher than that of other types of auxiliary structural members. A large number of groove features exist in the frame and beam structural members, and the work amount of polishing based on the fitter is large. Meanwhile, the polishing path planning is a key technology for realizing automatic polishing.

[0005] It can be known from the analysis of the existing polishing method of the frame and beam structural members of the aircraft structural member that manual polishing is still the mainstream polishing method. However, due to the characteristics of small batches and multiple categories of the aircraft structural member, the common polishing strategy is difficult to be applied to the scene of polishing the aircraft structural member. In order to break through the bottleneck of the automatic polishing technology of the aircraft structural member, the dependence on manual work can be greatly reduced, and automatic and intelligent polishing can be realized. Accurate polishing area identification and planning are the key to realizing automatic programming and a key technical problem to be solved in the entire aviation field. SUMMARY

[0006] In order to solve the problems existing in the artificial polishing, the application provides an automatic polishing planning and decision method for an aircraft structural member, realizes automatic identification and judgment of a polishing area of the aircraft structural member, and provides a basis for automatic planning of the polishing area, so as to facilitate implementation of automatic polishing technology, especially polishing of a web area of the structural member.

[0007] In order to achieve the above-mentioned application purposes, the technical scheme provided by the application is as follows:

[0008] An automatic polishing planning and decision method for an aircraft structural member comprises the following steps:

[0009] Step 1. Positioning of a to-be-polished structural member and a web area;

[0010] Step 2. Single-web multi-imaging angle image combination judgment;

[0011] Step 3. Identification and positioning of a tool mark transition area;

[0012] Step 4. Regional refinement data scanning;

[0013] Step 5. Acquisition of final web area polishing boundary data;

[0014] Step 6. First web polishing area planning;

[0015] Step 7. Spatial non-interference distance calculation;

[0016] Step 8. Determination of remaining to-be-polished webs based on a cyclic planning method;

[0017] Step 9. Generation of a polishing program for polishing;

[0018] Step 10. Polishing processing;

[0019] Step 11. Polishing area re-inspection.

[0020] Further, the positioning of the to-be-polished structural member and the web area comprises:

[0021] Based on a high-resolution three-channel fixed-position camera, the to-be-polished structural member is imaged, the fixed-position camera is configured with RGBD data acquisition capability, D represents depth information, and RGB represents image information, so that an original analysis image O is obtained; a global coarse positioning identification data model of the to-be-polished structural member is constructed, and positioning of the to-be-polished structural member in the original analysis image O is completed, and the global coarse positioning identification data model of the crane is as follows:

[0022]

[0023] The target of the mathematical model learning is to minimize the objective function M({p i},{t i}) denotes the log loss of target and non-target classes, denotes the prediction bounding box regression loss, N cls denotes the Minni-batch mini-batch value, N reg denotes the number of regression rectangular frames, λ denotes the balance weight, p i denotes the prediction probability that the Anchor is a target; denotes the Ground Truth bounding box coordinates and the prediction frame coordinates, respectively, i denotes the Ground Truth bounding box coordinates and the prediction frame coordinates, respectively, denotes the background and foreground corresponding values, the loss function contains two parts of the classification work background and structure, and the to-be-polished category loss and the regression loss of the coarse positioning frame.

[0024] Further, based on M({p i},{t i}) corresponding image O is coarsely identified and positioned corresponding result coordinates using {t x , t y , t w , t h} represents, t x is the normalized horizontal coordinate of the standard frame center, t y is the normalized vertical coordinate of the standard frame center, t w is the width value of the frame, t h is the height value of the frame; based on the result coordinates using {t x , t y , t w , t h} further crop and extract a new image P from image O; after smoothing processing of different weights on the processed image P, the corresponding Gaussian difference D(x, y, δ) of the adjacent weight difference image is calculated, x is the value of the image pixel point in the X axis direction, y is the value of the image pixel point in the Y direction, δ is the corresponding standard deviation, the key points K in the image after Gaussian difference processing are found, the key points are corrected by Taylor series expansion interpolation operation on the Gaussian difference D(x, y, δ), and the corrected points K-P are obtained. Based on K-P, the fine positioning of the to-be-polished area in image P is carried out, and the corresponding same image O of the same scale is denoted as Obj.

[0025] Further, the corresponding imaging camera arrangement of the structure to be polished in step 1 satisfies the following conditions: the camera imaging the structure to be polished includes a fixed position camera, a horizontally adjustable moving position camera 1, a horizontally adjustable moving position camera 2, and a horizontally adjustable moving position camera 3, the imaging axis of the fixed position camera is concentric with the center of the structure to be polished and the geometric center of the structure to be polished, there is a distance deviation <dis-o, dis-o is a set distance deviation threshold of the structure to be polished, and the fixed position camera is located directly above the structure to be polished; the horizontally adjustable moving position camera 1 satisfies moving in the imaging plane P1, the corresponding horizontal adjustable moving position camera 1 has an inclination angle α1, and there are num imaging stop points Point(i) on the corresponding imaging plane P1, that is, corresponding i ∈ [1, num], the arc length interval between adjacent imaging stop points Point(i) and Point(i+1) is equal to 360 / num, and each camera only images at the imaging stop point Point(i) with the corresponding inclination angle α1; the horizontally adjustable moving position camera 2 satisfies moving in the imaging plane P2, the corresponding horizontal adjustable moving position camera 2 has an inclination angle α2, and there are num imaging stop points Point(i) on the corresponding imaging plane P2, that is, corresponding i ∈ [1, num], the arc length interval between adjacent imaging stop points Point(i) and Point(i+1) is equal to 360 / num, and each camera only images at the imaging stop point Point(i) with the corresponding inclination angle α2; the horizontally adjustable moving position camera 3 satisfies moving in the imaging plane P3, the corresponding horizontal adjustable moving position camera 3 has an inclination angle α3, and there are num imaging stop points Point(i) on the corresponding imaging plane P3, that is, corresponding i ∈ [1, num], the arc length interval between adjacent imaging stop points Point(i) and Point(i+1) is equal to 360 / num, and each camera only images at the imaging stop point Point(i) with the corresponding inclination angle α3.

[0026] Further, the corresponding structure to be polished placement strategy is: after the structure is placed, the area to be polished of the corresponding structure to be polished faces upwards, if the structure needs to be polished on both sides, one side is selected to face upwards for polishing first, and after polishing, the other side of the model is polished; the geometric center of the placed structure is fixed in position on the entire workbench and is coaxial with the fixed position camera to ensure the integrity of the structure area.

[0027] Further, the corresponding mathematical model M({p i},{t iIn order to prevent the influence of BX and GT position and size on regression scale, wherein BX is a model prediction area, GT is an actual result area, BX coordinates need to be normalized, that is:

[0028]

[0029] wherein p x is an x value of a prediction box center; p y is a y value of a prediction box center; is an x value of a standard box GT center; is a y value of a standard box center; b x is an x value of a potential target area box center; b y is a y value of a prediction box center; p w is a prediction box width value; p h is a prediction box height value; b w is a prediction box center width value; b h is a potential target area height value; is a standard box width value; is a standard box height value; t x , t y , t w , t h and t respectively represent the normalized center coordinates and the width and height values of the prediction box and the standard box;

[0030] i is an index of a prediction box center; p i represents a prediction probability that a prediction box center is a target; represents a value corresponding to background and foreground, wherein the background is 0 and the foreground is 1; and t i respectively represent the coordinates of GT and the prediction box, wherein t i is a coordinate vector composed of {t x , t y , t w , t h}; represents a log loss corresponding to the identification and positioning of the workbench background and the area to be polished, that is: represents a prediction box coordinate regression deviation loss, and the corresponding mathematical model can be represented as wherein R represents a loss evaluation function, and let In the formula, only foreground exists coordinate position deviation loss;

[0031] The output results of the classification model and the regression model are represented as {p i} and {t i}, respectively, and N​cls denotes the normalization processing, N reg The balance weight λ is introduced and the normalization processing is realized, N cls is the small batch value, and N reg is the number of predicted box center positions, wherein the balance weight λ corresponds to a positive integer;

[0032] The multi-classifier mathematical model is set as:

[0033]

[0034] The optimized processing mode is to multiply a constant C in the numerator and the denominator, that is:

[0035]

[0036] All values are added to 1, which can ensure that each output result can be transformed to the [0, 1] interval; i represents the number index of the class, j represents the number of classes output by the neural network, f j denotes the value corresponding to the jth class, y i denotes the probability that the neural network prediction output is the ith class, f yi denotes the value corresponding to the prediction of the ith class, and x denotes the network input data;

[0037] A mapping table Table(x, y) of the region position and the physical scale is constructed, and the actual physical distance data Dis(point(x, y), C(x, y)) from each point point(x, y) in the image O and the Obj to the geometric center / center C(x, y) of the structure to be polished is recorded.

[0038] Further, the joint correction strategy of the placement pose of the structure to be polished and the design numerical model is based on the pose information of the structure placement to correct the placement pose of the structure in the design numerical model, so that the pose of the structure in the numerical model is consistent with the pose of the structure to be polished. The specific joint correction strategy steps are as follows: in the recognition result image Obj, there are four orientations, which are vertical upward, downward, horizontal left, and right, and the reference point of the orientation is the center of the Obj, that is, (rows / 2, cols / 2), wherein row represents the number of rows of the Obj, and cols represents the number of columns of the Obj; in the design numerical model software, the numerical model is adjusted based on the features corresponding to the vertical upward, downward, horizontal left, and right of the recognition result image Obj, so that the features presented on the screen are consistent with the features in the Obj.

[0039] Further, the web area potentially requiring polishing is located, first, the identification of the web area is completed in the Obj, and the corresponding specific steps are: first, the RGB data and D data obtained by the fixed camera are registered to ensure that the coordinate systems of the two types of data obtained coincide, the collection function of the camera depth D information is started, and the D data information is obtained; the structure data and background area (non-structure) data are separated from the D information data based on the structure boundary obtained by the Obj; the background area part of the separated D information data is set to 0, and the data in the structure area remains unchanged; since the obtained D data is in the form of a matrix, the length and width of the matrix are the same as the Obj, the data Data(i,j) of each point corresponding to the structure in the D data is read, i represents the corresponding row number in the data D, j represents the corresponding column number in the data D, and all Data num (i,j) are collectively referred to as Set(Data num (i,j)), and num represents the total number of data points corresponding to the structure in the D data; each Data num (i,j) in the set Set(Data num (i,j)) is sorted to obtain the minimum value Min(Data num (i,j)) and the maximum value Max(Data num (i,j)); all Data num (i,j) in the set Set(Data num (i,j)) are interval transformed, and the corresponding transformation formula is: T(Data num (i,j)) = 255*|Min(Data num (i,j))-Data num (i,j)| / (Max(Data num (i,j))-Min(Data num (i,j))), and the data corresponding to the transformation processing is T-Set(Data num (i,j)); the data of T-Set(Data num (i,j)) is twice transformed based on the first-order difference, and the corresponding second transformation formula is: TData num (i,j) = (|TData num (i,j)-TData num (i,j+1)|+|TData num (i,j)-TData num (i+1,j)|) / 2, wherein TData num (i,j) represents the set T-Set(Data numeach point data in corresponding; set the third transformation data processing threshold T-threshold, if T-Set(Data num (i,j))≤T-threshold, the corresponding T-Set(Data num (i,j))=0, otherwise remain unchanged; the D data information after 3 times of transformation is subjected to image transformation to obtain an image Dimg, and the Dimg is a single-channel gray image;

[0040] Based on the designed random assignment, different methods are used to extract the regions with the same gray scale in the image Dimg. The specific method is as follows: the gray value corresponding to each point in the image Dimg is read and changed to DV(i,j), DV(i,j)={Rv=(Num(random)%255), Gv=(Num(random)%255)and Gv!=Rv, Bv(Num(random)%255)and Bv!=Rv and Bv!=Gv}, if Gv=Rv, then Gv is regenerated, if Bv=Rv, then Bv is regenerated, if Bv=Gv, then Bv is regenerated, until Rv, Gv, and Bv are all different, wherein Rv represents the value assigned to the newly added channel of the point (i,j) in the image Dimg, Gv represents the value assigned to the newly added channel of the point (i,j) in the image Dimg, and Bv represents the value assigned to the newly added channel of the point (i,j) in the image Dimg, Num(random) represents a random positive integer; the Rv, Gv, and Bv values corresponding to each point data in the image Dimg are read, if the values corresponding to the points separated by a distance of 1 unit distance are equal, then such data points can be classified into a class, and the region composed of the points satisfying the class relationship is extracted and input into a classification neural network for classification, and the classification network belongs to binary classification, i.e., web and non-web; the region recognized as a web by the network is the completed web area Area k (i,j) positioning step, k represents the number of successfully recognized webs, and (i,j) represents the center coordinates of the corresponding web.

[0041] Further, the single-web multi-imaging angle image combination judgment is as follows: the horizontal adjustable moving camera 1, the horizontal adjustable moving camera 2, and the horizontal adjustable moving camera 3 are started, the corresponding imaging angles are α1, α2, and α3, and the corresponding num imaging stop points Point(i) on the imaging planes P1, P2, and P3 are imaged on each Area k(i,j) are imaged, and the number of images corresponding to each k web plate that can be imaged is 3*num image data, and the corresponding images are denoted as I1(h), I2(h), and I3(h), respectively, where h∈[1, num] represents the corresponding image number; the acquired I1(h), I2(h), and I3(h) 3*num image data are respectively evaluated based on the following designed formula,

[0042]

[0043] For I1(h), the greater the Value, the more there are tool marks that need to be polished; the 3*num calculated Value values are analyzed and compared, and they are sorted from large to small, and if there are β corresponding Value values greater than the set lower limit, it indicates that the corresponding web plate region has tool mark characteristics that need to be polished, and the region that needs to be polished is marked as G(m), m represents the number of regions that need to be polished; cols and rows represent the number of columns and rows of I1(h), I2(h), and I3(h), respectively, i and j represent the subscripts of the pixel points along the row and column directions, respectively, f represents the gray value at the (i, j) position, and x and y represent the horizontal and vertical directions, respectively.

[0044] Further, the tool mark transition region identification and positioning is specifically: determining the coordinates of the web plate region that needs to be polished based on the polished region marked as G(m); since the acquired data D is obtained through a series of transformations to obtain G(m), D and G have the same image coordinate system, that is, G(m) determines the web plate region that needs to be polished and has a one-to-one correspondence with D, and D and O are acquired based on the same position camera, so the coordinates of the web plate region that needs to be polished in O can be obtained based on G(m), and the coordinates of the web plate that needs to be polished in the numerical model are also obtained, to obtain an accurate polishing tool path.

[0045] Further, the region refinement data scanning is specifically: determining the coordinates of the web plate region that needs to be polished G(m) to complete the position information Loc(m) of the web plate that needs to be polished in the numerical model, setting the fine scanning moving track and path based on the position information Loc(m), and obtaining the accurate data PData(m) of each scanning region by using the generated path in a cyclic scanning manner; after m regions are processed, the data PData(m) is analyzed, the regions that do not need to be polished are removed from the m regions, and the regions that need to be polished are retained, to complete the secondary judgment of the web plate candidate region that needs to be polished based on G(m), and finally determine the web plate region that needs to be polished DM a (i,j) represents, a represents the number of final polishing regions, and (i,j) represents the coordinates of the corresponding region in G.

[0046] Further, the final web plate region polishing boundary data acquisition is specifically: based on the finally determined web plate region that needs to be polished DMa (i,j), the mapping of the polishing area a is completed in the digital model, and the boundary data Cdata corresponding to a is calculated in the calculation digital model.

[0047] Further, the first web polishing area planning is specifically: the distances dis from the geometric centers of the a polishing areas to be polished to the initial polishing head are calculated respectively, wherein the polishing area with the smallest distance dis to the polishing head is the first polishing area, the diameter of the surrounding area corresponding to polishing is calculated by the polishing head radius DR, and the surrounding area calculation method is

[0048]

[0049] wherein r is the actual radius of the polishing head, r r represents the radius of the corresponding polishing head ring surface, w represents the side deflection angle of the polishing head, and λ represents the front tilt angle of the polishing head; the polishing path optimization function number is designed, and the polishing area path optimization function number designed can be represented as i represents the polishing path length of the Nv polishing areas, S j represents the polishing path length of the corresponding closed area boundary, F z represents the auxiliary path length of the polishing head, Nk is the number of polishing areas, i represents the polishing path length index, j represents the closed area boundary polishing path length index, and z represents the auxiliary path length index of the polishing head; the polishing head processing track is generated based on the path optimization function, and the point data of the polishing head is generated based on the boundary data Cdata.

[0050] Further, the spatial non-interference distance measurement method is: based on the obtained first polishing area, the boundary data point Bdata(i,j) corresponding to the first polishing area is calculated, the radius direction of the polishing head is calculated, the rigid rotation area corresponding radius data GR is calculated, the point Mpoint(i,j) of the polishing head center movement is calculated, the Euclidean distance DMB between the point Mpoint(i,j) and the point Bdata(i,j) is directly calculated, and the Euclidean distance DMB between the point Mpoint(i,j) and the point Bdata(i,j) in the plane satisfies DMB≤GR, and the flag is marked as 0, and DMB>GR, and the flag is marked as 1.

[0051] According to the optimal spatial non-interference distance measurement method, the polishing path is corrected according to the workpiece interference condition, if there is spatial polishing interference flag=0, the corresponding path Mpoint(i,j) of the interference area is directly modified, if there is no interference flag=1, the Mpoint(i,j) is not modified.

[0052] ​Further, the determination of the remaining web to be polished based on the cyclic programming method is specifically as follows: taking the first polishing area as a starting point, a global a-1 polishing areas are programmed, and the corresponding programming evaluation function is

[0053]

[0054] wherein c1 represents a motion path weight coefficient, and the corresponding meaning is the importance evaluation of the polishing head motion path index, F(T k The first term represents the total length of the polishing head motion path, c2 represents a region polishing difficulty coefficient, Di represents the center clustering of adjacent two polishing regions, and Hi represents the polishing difficulty coefficient; global programming is performed, which is converted into an F(Tk) extreme value problem, and the second polishing area is obtained based on the global optimal solution, and the same is sequentially performed, h represents the subscript of the center clustering measurement of adjacent two polishing regions, and a represents the number of global polishing regions.

[0055] Further, the generation of the polishing program for polishing is specifically as follows: a numerical control program capable of controlling the motion of the polishing head is generated.

[0056] Further, the polishing treatment is specifically as follows: the polishing head is driven based on the generated polishing program to complete the polishing treatment of the a polishing regions, the first polishing of each region is completed, and the polishing region re-inspection is performed for evaluation.

[0057] Further, the polishing region re-inspection is specifically as follows: the polished region is evaluated, if the corresponding parameters of the region's knife joint mark and step difference polishing features do not meet the requirements, further polishing is required, the further polishing is selected from the a first polishing regions that are not polished well, and the process is repeated until the quality of all a regions meets the requirements.

[0058] The present application has the following advantages:

[0059] 1. Compared with the traditional manual polishing method, the present method can realize automatic identification of the polishing region, and based on the identified results, the self-adaptive planning and decision of the polishing region can be completed, thereby providing core technical support for the full-automatic polishing of the aviation structural parts, especially the complex structural parts. The polishing region planning strategy can fully consider the position of adjacent polishing regions and the corresponding attributes of polishing features, can obviously reduce the generation of invalid polishing paths, and can effectively improve the efficiency and quality of the whole process polishing of the aviation structural parts. The judgment of the polishing region combined with the analysis results of multiple perspectives can significantly reduce the deviation introduced by single perspective analysis, improve the polishing precision and accuracy, avoid quality problems caused by polishing, and ensure the reliability of automatic polishing.

[0060] 2、The application provides an effective method suitable for polishing of an aviation structural member, and proposes a polishing area planning strategy, which can fully consider the positions of adjacent polishing areas and corresponding attributes of polishing features, can obviously reduce generation of invalid polishing paths, and effectively improves the efficiency and quality of the whole process polishing of the aviation structural member. BRIEF DESCRIPTION OF DRAWINGS

[0061] Figure 1 A flowchart of the method of the application.

[0062] Figure 2 A corresponding structural member imaging camera arrangement.

[0063] Figure 3 A corresponding polishing area and polishing area polishing head path planning. DETAILED DESCRIPTION

[0064] To make the objects, technical solutions and advantages of the application clearer, the technical solutions in the embodiments of the application will be described clearly and completely below with reference to the drawings in the embodiments of the application. Obviously, the described embodiments are used to explain the application but not to limit the application. Based on the embodiments in the application, all other embodiments obtained by those skilled in the art without creative work fall within the protection scope of the application.

[0065] The specific implementation method of the application is described below in combination with the drawings and examples, and the application is not limited to the embodiments.

[0066] Embodiment 1

[0067] An automatic polishing planning and decision method of an aircraft structural member, comprising the following steps:

[0068] Step 1. Positioning of a structural member to be polished and a web area.

[0069] Step 2. Single-web multi-imaging angle image combination judgment.

[0070] Step 3. Identification and positioning of a tool joint transition area.

[0071] Step 4. Area refinement data scanning.

[0072] Step 5. Acquisition of final web area polishing boundary data.

[0073] Step 6. First web polishing area planning.

[0074] Step 7. Spatial non-interference distance calculation.

[0075] Step 8. Determination of remaining web to be polished based on a cycle planning method.

[0076] Step 9. Generation of a polishing program for polishing.

[0077] Step 10. Polishing treatment is performed;

[0078] Step 11. Polishing area reinspection.

[0079] The step 1 comprises:

[0080] Firstly, a structure to be polished is imaged based on a high-resolution three-channel fixed-position camera, the fixed-position camera is configured with RGBD data acquisition capability, D represents depth information, and RGB represents image information, to obtain an original analysis image O; a global rough positioning identification data model of the structure to be polished is constructed, and positioning of the structure to be polished in the original analysis image O is completed, and the global rough positioning identification data model of the crane worker is as follows:

[0081]

[0082] Wherein, the target of the mathematical model learning is to minimize the objective function M({p i},{t i}), represents the logarithmic loss of the target and non-target categories, represents the prediction frame regression loss, N cls represents the Minni-batch small batch value, N reg represents the number of regression rectangular frames, λ represents the balance weight, p i represents the prediction probability of Anchor being the target; and t i respectively represent the coordinates of the Ground Truth package and the coordinates of the prediction frame, represents the value corresponding to the background and foreground, and the loss function includes two parts of the classification background and the structure, and the loss of the category to be polished and the regression loss of the rough positioning frame.

[0083] Based on M({p i},{t i}) corresponding to the image O, the corresponding result coordinates of the rough identification positioning are represented by {t x , t y , t w , t h}, t x is the normalized center horizontal coordinate of the standard frame, t y is the normalized center vertical coordinate of the standard frame, t w is the width value of the frame, and t h is the height value of the frame; based on the result coordinates, {t x , t y , t w , t h}Further cropping of the image O extracts a new image P; different weight smoothing processing is performed on the processed image P, and a Gaussian difference D(x, y, δ) corresponding to adjacent weight difference images is calculated after processing, x is the value of the image pixel point in the X axis direction, y is the value of the image pixel point in the Y direction, and δ is the corresponding standard deviation. The key point K in the image after the Gaussian difference processing is found, the Gaussian difference D(x, y, δ) is interpolated by Taylor series expansion, the key point is corrected, and the corrected point K-P is obtained. Based on K-P, the fine positioning of the area to be polished in the image P is performed, that is, the image obtained by positioning the camera and based on the designed algorithm completes the accurate positioning of the area where the structure to be polished is located. The corresponding same image O is indicated as Obj.

[0084] In step 1, the corresponding structure to be polished is imaged by the camera arrangement as described above, as shown in FIG. 2. The arrangement and imaging method meet the following conditions: the camera imaging the structure to be polished includes a fixed position camera, a horizontally adjustable moving position camera 1, a horizontally adjustable moving position camera 2, and a horizontally adjustable moving position camera 3. The imaging axis of the fixed position camera is concentric with the center of the structure to be polished and the geometric center of the structure to be polished, and there is a distance deviation <dis-o, which is a set threshold for the distance deviation between the center of the structure to be polished and the geometric center of the structure to be polished. The fixed position camera is located directly above the structure to be polished. The horizontally adjustable moving position camera 1 meets the condition of moving in the imaging plane P1. The corresponding pitch angle of the horizontally adjustable moving position camera 1 is α1, and there are num imaging docking points Point(i) on the corresponding imaging plane P1, i.e. corresponding to i∈[1, num], the arc length interval between adjacent imaging docking points Point(i) and Point(i+1) is equal to 360 / num, and each camera only images at the corresponding pitch angle α1 at the imaging docking point Point(i). The horizontally adjustable moving position camera 2 meets the condition of moving in the imaging plane P2. The corresponding pitch angle of the horizontally adjustable moving position camera 2 is α2, and there are num imaging docking points Point(i) on the corresponding imaging plane P2, i.e. corresponding to i∈[1, num], the arc length interval between adjacent imaging docking points Point(i) and Point(i+1) is equal to 360 / num, and each camera only images at the corresponding pitch angle α2 at the imaging docking point Point(i). The horizontally adjustable moving position camera 3 meets the condition of moving in the imaging plane P3. The corresponding pitch angle of the horizontally adjustable moving position camera 3 is α3, and there are num imaging docking points Point(i) on the corresponding imaging plane P3, i.e. corresponding to i∈[1, num], the arc length interval between adjacent imaging docking points Point(i) and Point(i+1) is equal to 360 / num, and each camera only images at the corresponding pitch angle α3 at the imaging docking point Point(i). The fixed position camera is used to obtain the global image of the structure to be polished, so as to realize the positioning of the boundary region of the structure. The horizontally adjustable moving position camera 1, the horizontally adjustable moving position camera 2, and the horizontally adjustable moving position camera 3 are used to obtain images of the web region to be polished at different angles, and based on the obtained images, it is determined whether the corresponding region needs to be polished. The distribution of the horizontally adjustable moving position camera 1, the horizontally adjustable moving position camera 2, and the horizontally adjustable moving position camera 3 has the characteristic of equally dividing the half-sphere latitude.

[0085] As described above, the corresponding to-be-ground structure placement strategy is that after the structure is placed, the corresponding to-be-ground region of the corresponding to-be-ground structure faces upward, if the structure needs to be ground on both sides, then one side is selected to face upward first, and after the grinding is completed, the other side of the model is ground; the geometric center of the structure after being placed is fixed at the position of the entire workbench, and is coaxial with the fixed position camera, so as to ensure the integrity of the structure region.

[0086] As described above, the corresponding mathematical model M({p i},{t i}) is as follows:

[0087]

[0088] Wherein, p x is the x value of the prediction box center; p y is the y value of the prediction box center; is the x value of the standard box GT center; is the y value of the standard box center; b x is the x value of the potential target region box center; b y is the y value of the prediction box center; p w is the prediction box width value; p h is the prediction box height value; b w is the prediction box center width value; b h is the potential target region height value; is the standard box width value; is the standard box height value; t x , t y , t w , t h and respectively represent the normalized center coordinates and the width and height values of the BX and the standard box;

[0089] i is the index of the prediction box center; p i represents the prediction probability of the prediction box center being the target; represents the value corresponding to the background and foreground, the background is 0, and the foreground is 1; and t i respectively represent the coordinates of the GT and the prediction box, wherein t i is a coordinate vector composed of {t x , t y , t w , t h}; represents the logarithmic loss corresponding to the workbench background and the to-be-ground region identification and positioning, that is: represents the regression deviation loss of the prediction box coordinates, and the corresponding mathematical model can be represented as where R represents the loss evaluation function, and let where represents only the foreground There is a coordinate position deviation loss;

[0090] The output results of the classification model and the regression model are represented as {p i} and {t i}, respectively, and N cls represents the normalization processing, N reg The normalized N cls of the cls is a small batch value, while the normalized N reg of the reg item is the number of predicted box center positions, wherein the value of the balance weight λ is a positive integer;

[0091] The mathematical model of the multi-classifier is set as:

[0092]

[0093] Since the classifier formula in the above formula is an exponential operation, if the value is very large, it may cause an exponential explosion problem. The optimized processing method is to multiply a constant C in the numerator and a constant C in the denominator, that is:

[0094]

[0095] All values are added to 1, which can ensure that each output result can be transformed to the [0, 1] interval; i represents the class index, j represents the class number of the neural network output, f j represents the value corresponding to the jth class, y i represents the probability of the neural network prediction output being the i-th class, f yi represents the value corresponding to the i-th class prediction, and x represents the network input data;

[0096] A mapping table Table(x, y) of the region position and the physical size is constructed, which records the actual physical distance data Dis(point(x, y), C(x, y)) of each point point(x, y) in the image O and the Obj to the geometric center / placement center C(x, y) of the structure to be polished. The physical length value can be used to determine the position of the polishing region and the corresponding feature in the design model, because the image O and the Obj are in different coordinate systems, and the actual physical distance data Dis(point(x, y), C(x, y)) is the only feasible way to realize the mapping of the polishing region features. The key to mapping the data is that the area and size of the image taken by the fixed-position camera are fixed, the imaging angle of the camera is fixed, and the resolution and other parameters are fixed, so that the actual spatial distance difference corresponding to each image taken by the fixed-position camera is small, which does not affect the actual feature mapping.

[0097] The placement pose of the structure to be polished is corrected in combination with the design model, which is based on the pose information of the structure placement to correct the placement pose of the structure in the design model, so that the pose of the structure in the model is consistent with the pose of the structure to be polished, which is beneficial to the subsequent positioning of the polishing region in the model and facilitates the generation of an automatic polishing program that can control the movement of the polishing head. The specific joint correction strategy steps are as follows: in the recognition result image Obj of the structure to be polished, there are four orientations, which are vertical up, vertical down, horizontal left, and horizontal right. The reference point of the orientation is the center of Obj, i.e., (rows / 2, cols / 2), where row represents the number of rows of Obj, and cols represents the number of columns of Obj. In the design model software, the features corresponding to the vertical up, vertical down, horizontal left, and horizontal right in the recognition result image are adjusted based on the features, so that the features presented on the screen are consistent with the features in Obj. The consistent explanation is that the contour features of the corresponding structure features are basically the same as the contour features in Obj, and the corresponding magnification, i.e., the size of the structure features presented in the screen direction, can be different.

[0098] After the boundary area of the structure to be polished is positioned, the web area that needs to be polished is further positioned. First, the web area is identified in the Obj. The specific steps are as follows: first, the RGB data and D data obtained by the fixed-position camera are registered to ensure that the coordinate systems of the two types of data obtained coincide, that is, the corresponding coordinates of the RGB data and D data obtained in the two coordinate systems are consistent after registration; since the fixed-position camera is equipped with the capability of collecting RGBD data information, the depth D information collection function of the camera is started to obtain D data information; the structure data and background area (non-structure) data are separated from the D information data based on the structure boundary obtained from the Obj; the background area part of the separated D information data is set to 0, while the data in the structure area remains unchanged; since the obtained D data is in matrix form, the length and width of the matrix are the same as the Obj, the data Data(i,j) of each point in the structure in the D data is read, i represents the corresponding row number in the data D, j represents the corresponding column number in the data D, and the set of all Data num (i,j) is denoted as Set(Data num (i,j)), and num represents the total number of data points corresponding to the structure in the D data; each Data num (i,j) in the set Set(Data num (i,j) is sorted to obtain the minimum value Min(Data num (i,j) and the maximum value Max(Data num (i,j); all Data num (i,j) in the set Set(Data num (i,j) is transformed, and the corresponding transformation formula is T(Data num (i,j))=255*|Min(Data num (i,j))-Data num (i,j)| / (Max(Data num (i,j))-Min(Data num (i,j)), and the data after transformation is denoted as T-Set(Data num (i,j)); the data of T-Set(Data num (i,j) is further transformed based on the first-order difference, and the corresponding second transformation formula is TData num (i,j)=(|TData num (i,j)-TData num (i,j+1)|+|TData num (i,j)-TData num(i+1,j) |) / 2, where TData num (i,j) represents each point data in the corresponding T-Set(Data num (i,j) after transformation processing; a cubic transformation data processing threshold T-threshold is set, if T-Set(Data num (i,j) ≤ T-threshold, then T-Set(Data num (i,j) = 0, otherwise remains unchanged; the D data information after 3 times of transformation is subjected to image transformation to obtain an image Dimg, which is a single-channel grayscale image.

[0099] Based on the designed random assignment of each different method, the regions with the same grayscale in the image Dimg are extracted, and the specific method is as follows: the grayscale value corresponding to each point in the image Dimg is read and changed to DV(i,j), DV(i,j) = {Rv = (Num(random) % 255), Gv = (Num(random) % 255) and Gv ≠ Rv, Bv(Num(random) % 255) and Bv ≠ Rv and Bv ≠ Gv}, if Gv = Rv, then Gv is regenerated, if Bv = Rv, then Bv is regenerated, if Bv = Gv, then Bv is regenerated, until Rv, Gv and Bv are all different, wherein Rv represents the value assigned to the newly added channel of the point (i,j) in the image Dimg, wherein Gv represents the value assigned to the newly added channel of the point (i,j) in the image Dimg, wherein Bv represents the value assigned to the newly added channel of the point (i,j) in the image Dimg, and Num(random) represents a generated random positive integer; the Rv, Gv and Bv values corresponding to each point data in the image Dimg are read, if the values corresponding to the points with a distance of 1 unit distance are equal, then such data points can be classified into a class, and the region composed of the points satisfying the class relationship is extracted and input into a classification neural network for classification, and the classification network belongs to binary classification, i.e., web plate and non-web plate; the region recognized as a web plate by the network is the completed web plate region Area k (i,j) positioning step, k represents the number of successfully recognized web plates, and (i,j) represents the center coordinates of the corresponding web plate.

[0100] The single web multi-imaging angle image combination judgment is specifically: the purpose of this step is to further determine which webs need to be polished and which webs do not need to be polished based on the web area that has been positioned above, thereby improving the polishing efficiency of the structural member. The specific implementation method steps are: starting the horizontally adjustable mobile camera 1, the horizontally adjustable mobile camera 2, and the horizontally adjustable mobile camera 3, and the corresponding imaging angles are a1, a2, and a3, respectively, and the corresponding num imaging stops Point(i) on the imaging planes P1, P2, and P3, respectively, are imaged for each Area k Each web can obtain 3*num images, and the corresponding images are denoted as I1(h), I2(h), and I3(h), where h∈[1, num] represents the corresponding image number. Based on the following formula, the 3*num images I1(h), I2(h), and I3(h) are evaluated,

[0101]

[0102] This formula can reflect and evaluate the large differences in the image, such as the knife mark and the step difference, i.e., the high-frequency characteristic feature. In theory, the larger the value of I1(h) is, the more knife marks there are that need to be polished. By analyzing and comparing the 3*num calculated value, and sorting them from large to small, if there are β corresponding value greater than the lower limit, it indicates that the corresponding web area has features such as knife marks that need to be polished. The area that needs to be polished is marked as G(m), and m represents the number of areas that need to be polished. cols and rows represent the number of columns and rows of I1(h), I2(h), and I3(h), respectively. i and j represent the subscripts of the pixel points along the row and column directions, respectively. f represents the gray value at the (i, j) position, and x and y represent the horizontal and vertical directions, respectively.

[0103] The knife mark transition area recognition and positioning is specifically: the purpose of this step is to determine the position information of the web area that needs to be polished containing the knife mark in the polishing tool and the three-dimensional model. The position information on the polishing tool can ensure accurate positioning of the web polishing area during polishing, and the position information in the three-dimensional model can ensure accurate generation of the polishing head movement program in the model. The implementation steps are: determining the coordinates of the web area that needs to be polished based on the area marked as G(m). Since the data D is obtained through a series of transformations to obtain G(m), D and G have the same image coordinate system, i.e., G(m) determines the one-to-one correspondence between the web area that needs to be polished and D. Since D and O are obtained based on the same position camera, the coordinates of the web area that needs to be polished in O can be obtained based on G(m), and the coordinates of the web area that needs to be polished in the model are also obtained to obtain the accurate polishing tool path.

[0104] The region refinement data scanning specifically is: the purpose of this step is to determine the coordinates of the web region to be polished by G(m) to make a second judgment, while avoiding invalid scanning, the accuracy of the preliminary determination of the region to be polished can be increased, the implementation steps are: G(m) determines the coordinates of the web region to be polished, the position information Loc(m) needed to be polished in the digital model is completed, the refined scanning movement track and path are set based on the position information Loc(m), the accurate data PData(m) of each scanning region is obtained by using the cyclic scanning mode according to the generated path, after m regions are processed, the data PData(m) is analyzed, the regions that do not need to be polished are removed from the m regions, and the regions to be polished are retained, the second judgment of the candidate region to be polished by G(m) is completed, and the finally determined web region to be polished is DM a (i,j) represents, a represents the number of finally needed polishing regions, (i,j) represents the coordinates of the corresponding region in G.

[0105] The final web region polishing boundary data acquisition specifically is: based on the finally determined web region to be polished DM a (i,j), the mapping of the polishing region a is completed in the digital model, and the boundary data Cdata corresponding to a in the calculation digital model is calculated. The necessity of the boundary data Cdata calculation is that the movement boundary of the polishing head can be strictly limited based on the boundary data, so as to avoid polishing the regions that do not need to be polished, resulting in the size of the structural part being out of tolerance and being scrapped.

[0106] The first web polishing region planning specifically is: the purpose of this step is to determine the first polishing web, and the movement coordinates of the polishing head and the distance data between the structural part and the polishing head corresponding to each polishing region DM a (i,j) are calculated, the distance dis between the geometric center of the a polishing regions and the initial polishing head is calculated, wherein the polishing region with the smallest distance dis from the polishing head is the first polishing region, this planning method can ensure that the invalid travel of the polishing head is as short as possible, thereby improving the polishing efficiency. The polishing head movement mode corresponding to the polishing region is designed, the purpose is to be efficient and the total movement path of the polishing head is short, thereby reducing the polishing time consumption while ensuring the polishing quality, and improving the polishing quality, the specific implementation is: the diameter of the surrounding region corresponding to the polishing is calculated by the polishing head radius DR, the surrounding region calculation method has

[0107]

[0108] Wherein, r is the actual radius of the polishing head, r rrepresents the corresponding polishing head circular surface radius, w represents the polishing head side deflection angle, and λ represents the polishing head rake angle; the polishing path optimization row number is designed, and the polishing area path optimization function number designed can be represented as where M i represents the polishing path length of the Nv polishing areas, S j represents the corresponding closed area boundary polishing path length, F z represents the auxiliary path length walked by the polishing head; Nk is the number of polishing areas, i represents the polishing path length index, j represents the closed area boundary polishing path length index, and z represents the auxiliary path length index walked by the polishing head, the generation of the polishing head processing track is completed based on the path optimization function, and the generation of the polishing head point data is based on the boundary data Cdata to achieve.

[0109] The spatial non-interference distance measurement method is: based on the obtained first polishing area, the corresponding boundary data point Bdata(i,j) of the polishing area is calculated, the radius direction of the polishing head is calculated, the corresponding radius data GR of the rigid rotation area during polishing, the point Mpoint(i,j) of the polishing head center movement, the Euclidean distance DMB between the point Mpoint(i,j) and the point Bdata(i,j) is directly calculated, and the Euclidean distance DMB between the point Mpoint(i,j) and the point Bdata(i,j) in the plane satisfies DMB≤GR, and the flag is marked as 0, and DMB>GR flag is marked as 1;

[0110] According to the optimal spatial non-interference distance measurement method, the workpiece interference condition is judged to correct the polishing path, if there is spatial polishing interference flag=0, the corresponding path Mpoint(i,j) of the interference area is directly modified, and if there is no interference flag=1, Mpoint(i,j) does not need to be modified.

[0111] The remaining web to be polished is determined based on the cyclic planning method, and the action of this step is to select the remaining web to be polished, and to ensure that the sum of the paths of the polishing head holes composed of the selected web regions is as small as possible. The specific method steps are: let the first polishing area be the starting point to plan a-1 global polishing areas, and the corresponding planning evaluation function is

[0112]

[0113] wherein c1 represents a motion path weight coefficient, and the corresponding meaning is an importance evaluation of the polishing head motion path index, F(T k) the first item represents the total length of the polishing head movement path, c2 represents the area polishing difficulty coefficient, Di represents the center clustering of adjacent two polishing areas, Hi represents the polishing difficulty coefficient; global planning is performed to convert into F(Tk) extreme value problem, and the second polishing area is obtained based on the global optimal solution, and the same is repeated in turn, h represents the subscript of the center clustering measurement of adjacent two polishing areas, and a represents the number of global polishing areas.

[0114] The generated polishing program is used for polishing, which is specifically: the role of this step is to generate a numerical control program that can control the movement of the polishing head in software according to the order of each polishing area evaluated and the polishing head movement mode corresponding to each area, so as to accurately control the polishing head to move according to the established mode. The corresponding polishing area and polishing area polishing head path planning are as shown in Figure 3 .

[0115] The polishing process is specifically: driving the polishing head to complete the polishing process of a polishing area based on the generated polishing program, completing the first polishing of each area. The first polishing may not be complete, so further polishing is needed. The polishing area reinspection is used for judgment.

[0116] The polishing area reinspection is specifically: the reinspection step uses the same technical means as described above to evaluate the polished area. If the parameters corresponding to the tool joint mark and step difference of the area to be polished do not meet the requirements, further polishing is needed. The further polishing corresponds to the selection of a first polishing area that is not polished to the required position, and the process continues until the quality of all a areas meets the requirements.

[0117] Example 2

[0118] 1. A method for identifying and positioning the boundary area of a structure to be polished, which is mainly used for positioning the boundary of the structure to be polished based on the image of the structure obtained by a camera. The specific steps are as follows: first, a high-resolution three-channel fixed-position camera is used to image the structure to be polished. The fixed-position camera is configured with RGBD data acquisition capability, where D represents depth information and RGB represents image information, to obtain an original analysis image O; a global coarse positioning identification data model of the structure to be polished is constructed to complete the positioning of the structure to be polished in the original analysis image O. The global coarse positioning identification data model is as follows:

[0119]

[0120] The goal of mathematical model learning is to minimize the objective function M({p i},{t i}), which includes two parts, i.e., the classification loss of the tool background and the structure, and the regression loss of the coarse positioning bounding box; based on M({p i}, t i}) corresponding image O coarse recognition positioning corresponding results coordinates using {t x , t y , t w , t h} represent; based on the result coordinates using {t x , t y , t w , t h} further cropping of image O extraction of new image P; different weight smoothing processing is carried out on the processed image P, the corresponding Gaussian difference D(x, y, δ) of adjacent weight difference image is calculated after processing, the key point K in the image after Gaussian difference processing is searched, the Gaussian difference D(x, y, δ) is operated by Taylor series expansion, the key point is corrected, and the corrected point K-P is obtained. Based on K-P, the fine positioning of the area to be polished in image P is carried out, that is, the image obtained by positioning the camera and based on the designed algorithm completes the accurate positioning of the area where the structure to be polished is located, and the corresponding same image O of the same scale is represented by Obj.

[0121] 1.1 As described above, the corresponding structure to be polished imaging camera arrangement, as shown in Figure 2, the arrangement and imaging method meet the following conditions: the camera imaging the structure to be polished includes a fixed position camera, a horizontally adjustable moving position camera 1, a horizontally adjustable moving position camera 2, and a horizontally adjustable moving position camera 3. The imaging axis of the fixed position camera is concentric with the center of the structure to be polished and the geometric center of the structure to be polished, and there may be a distance deviation <dis-o. The fixed position camera is located directly above the structure to be polished. The horizontally adjustable moving position camera 1 meets the moving in the imaging plane P1, and the corresponding horizontally adjustable moving position camera 1 has the pitch angle α1. There are num imaging stop points Point(i) on the corresponding imaging plane P1, that is, corresponding i∈[1,num], the circular arc interval between adjacent imaging stop points Point(i) and Point(i+1) is equal to 360 / num, and each camera only images at the imaging stop point Point(i) with the corresponding pitch angle α1. The horizontally adjustable moving position camera 2 meets the moving in the imaging plane P2, and the corresponding horizontally adjustable moving position camera 2 has the pitch angle α2. There are num imaging stop points Point(i) on the corresponding imaging plane P2, that is, corresponding i∈[1,num], the circular arc interval between adjacent imaging stop points Point(i) and Point(i+1) is equal to 360 / num, and each camera only images at the imaging stop point Point(i) with the corresponding pitch angle α2. The horizontally adjustable moving position camera 3 meets the moving in the imaging plane P3, and the corresponding horizontally adjustable moving position camera 3 has the pitch angle α3. There are num imaging stop points Point(i) on the corresponding imaging plane P3, that is, corresponding i∈[1,num], the circular arc interval between adjacent imaging stop points Point(i) and Point(i+1) is equal to 360 / num, and each camera only images at the imaging stop point Point(i) with the corresponding pitch angle α3. The function of the fixed position camera is to obtain the global image of the structure to be polished, so as to realize the positioning of the boundary region of the structure. The functions of the horizontally adjustable moving position camera 1, the horizontally adjustable moving position camera 2, and the horizontally adjustable moving position camera 3 are to obtain the images of the web regions to be polished at different angles, and to judge whether the corresponding regions need to be polished based on the obtained image fusion. The distribution of the horizontally adjustable moving position camera 1, the horizontally adjustable moving position camera 2, and the horizontally adjustable moving position camera 3 has the characteristic of evenly dividing the latitude of the hemisphere.

[0122] 1.2 As described above, the corresponding structure to be polished placement strategy is: after the structure is placed, the corresponding grinding area of the corresponding structure to be polished faces up, if the structure needs to be polished on both sides, then select one side to polish first, and polish the other side after the polishing is completed; the geometric center of the structure after being placed is fixed at the position of the entire workbench, and is coaxial with the fixed position camera, so as to ensure the integrity of the structure region.

[0123] 1.3 As described above, the corresponding mathematical model M({p i},{t i}) is:

[0124]

[0125] Wherein, p x is the x value of the prediction box center; p y is the y value of the prediction box center; is the x value of the standard box GT center; is the y value of the standard box center; b x is the x value of the potential target region box center; b y is the y value of the prediction box center; p w is the prediction box width value; p h is the prediction box height value; b w is the prediction box center width value; b h is the potential target region height value; is the standard box width value; is the standard box height value; t x , t y , t w , t h and t respectively represent the normalized center coordinates and the width and height values of the BX and the standard box. i is the index of the prediction box center; p i represents the prediction probability of the prediction box center being the target; represents the value corresponding to the background and foreground, the background is 0, and the foreground is 1; and t i respectively represent the coordinates of the GT and the prediction box, wherein t i is a coordinate vector composed of {t x , t y , t w , t h}; represents the log loss corresponding to the identification and positioning of the workbench background and the grinding region, that is: represents the prediction box coordinate regression deviation loss, and the corresponding mathematical model can be represented as where R represents a loss evaluation function, let where represents only foreground There is a coordinate position deviation loss. The output results of the classification model and the regression model are represented as {p i} and {t i}, respectively, and N cls represents the normalization processing, N reg A balance weight λ is introduced and normalization processing is implemented, and N cls for the cls is a small batch value, while N reg for the reg item is normalized, and the value is the number of predicted box center positions, wherein the value of the balance weight λ is a positive integer.

[0126] The multi-classifier mathematical model is set as:

[0127]

[0128] Since the classifier formula of the above formula is an exponential operation, if the value is very large, it may cause an exponential explosion problem. The optimized processing method is to multiply a constant C in the numerator and a constant C in the denominator, that is:

[0129]

[0130] All values are added to 1, which can ensure that each output result can be transformed to the [0, 1] interval.

[0131] 2. A mapping table Table(x, y) of regional position and physical scale is constructed, which records the actual physical distance data Dis(point(x, y), C(x, y)) from each point point(x, y) in the image O and Obj to the geometric center / center C(x, y) of the structure to be polished. Based on the value of the physical length, the subsequent positioning of the polishing area and the corresponding features in the design model can be provided. Because the image O and Obj and the design model are in different coordinate systems, the actual physical distance data Dis(point(x, y), C(x, y)) is the only feasible way to realize the mapping of the polishing area features. The key to mapping data is that the area and size of the image obtained by the fixed position camera are fixed, the imaging angle of the camera is fixed, and the resolution and other parameters are also fixed, so that the actual spatial distance difference corresponding to each image obtained by the fixed position camera is small, which does not affect the actual feature mapping.

[0132] 3. Joint correction strategy for the placement posture of the structural component to be ground and the design model: The purpose of this step is to correct the placement posture of the structural component in the design model based on the placement posture information of the structural component, so that the posture of the structural component in the model is consistent with the posture of the structural component to be ground. This facilitates the subsequent positioning of the area to be ground in the model and also facilitates the generation of an automatic grinding program that can control the movement of the grinding head. The specific joint correction strategy steps are as follows: In the recognition result image Obj of the structural component to be ground, there are 4 orientations: vertically up, down, horizontally left, and right. The reference points for the orientations are... The center of Obj is referenced at (rows / 2, cols / 2), where row represents the number of rows in Obj and cols represents the number of columns in Obj. In the design of the digital model software, the digital model is adjusted based on the vertical upward and downward, and horizontal left and right corresponding features of Obj in the recognition result image, so that the features presented on the screen are consistent with the features in Obj. Consistency means that the contour features presented by the corresponding structural components are basically the same as the contour features in Obj. The corresponding magnification, that is, the size of the structural component features presented in the screen direction, may be different.

[0133] 4. After completing the identification and positioning of the boundary area of ​​the structural component to be polished, it is necessary to further locate the potential web area that needs to be polished. First, the identification and judgment of the web area is completed in Obj. The specific steps are as follows: First, the RGB data and D data acquired by the fixed camera are registered to ensure that the coordinate systems of the two types of data are coincident, that is, after registration, the coordinates of the RGB data and D data acquired by the same object in the two coordinate systems are consistent; since the fixed camera is equipped with the ability to acquire RGBD data information, the camera depth D information acquisition function is activated to acquire D data information; based on the structural component boundary obtained from Obj, the structural component data and the background area (non-structural component) data are separated from the D information data; the background area part of the separated D information data is set to 0, while the data in the structural component area remains unchanged; since the acquired D data is in matrix form, the length and width of the matrix are the same as those in Obj, the data Data(i,j) corresponding to each point of the structural component in the D data is read, where i represents the corresponding row number in the data D, and j represents the corresponding column number in the data D. num The set of (i,j) is denoted as Set(Data). num (i,j)), where num represents the total number of data points corresponding to the structural components in the D data; calculate the set Set(Data num Each Data in (i,j) num Sort the values ​​(i,j) to obtain the minimum value Min(Data). num (i,j) and the maximum value Max(Data) num (i,j)); Set(Data) numall Data in T-Set(Data(i,j)) num Interval transformation is performed on all Data in T-Set(Data(i,j)), and the corresponding transformation formula is T(Data(i,j)) = 255*|Min(Data(i,j))-Data(i,j)| / (Max(Data(i,j))-Min(Data(i,j))), where T-Set(Data(i,j)) represents the corresponding data in the set T-Set(Data(i,j)) after transformation processing. num num num num num num Secondary transformation processing is performed on the data T-Set(Data(i,j)) based on first-order difference, and the corresponding secondary transformation formula is TData(i,j) = (|TData(i,j)-TData(i+1,j)|+|TData(i,j)-TData(i,j+1)|) / 2, where TData(i,j) represents the corresponding each point data in the set T-Set(Data(i,j)) after transformation processing. num num num num num num num num If T-Set(Data(i,j))≤T-threshold, then T-Set(Data(i,j))=0, otherwise it remains unchanged. num num Image Dimg is obtained by image transformation of the D data information after three times of transformation, and Dimg is a single-channel grayscale image.Based on the designed random assignment method, the regions with the same grayscale in the image Dimg are extracted, and the specific method is as follows: the grayscale value of each point in the image Dimg is read and changed to DV(i,j), DV(i,j) = {Rv=(Num(random)%255), Gv=(Num(random)%255) and Gv≠Rv, Bv=(Num(random)%255) and Bv≠Gv, Num(random) represents a random number, and % represents modulus operation.

[0134] ​​​​​​​​​​​​{if Rv = Gv, then re-generate Gv, if Bv = Rv, then re-generate Bv, if Bv = Gv, then re-generate Bv, until Rv, Gv, Bv are all different, wherein Rv represents the value assigned to the newly added channel of the point (i, j) in the image Dimg, wherein Gv represents the value assigned to the newly added channel of the point (i, j) in the image Dimg, wherein Bv represents the value assigned to the newly added channel of the point (i, j) in the image Dimg, and Num(random) represents a randomly generated positive integer; read the Rv, Gv, and Bv values corresponding to each data point in the image Dimg, and if the values corresponding to points separated by a distance of 1 unit are equal, such data points can be classified into a class, extract the region composed of points that satisfy the class relationship, and input the classified neural network for classification, and the classification network is a binary classification, i.e., web and non-web; the region identified as a web by the network is the completed web area Area k (i, j) positioning step, k represents the number of successfully identified webs, and (i, j) represents the center coordinates of the corresponding web.

[0135] 5. Single web multi-imaging angle image combination judgment, the purpose of this step is to further determine which webs need to be polished and which webs do not need to be polished based on the web area that has been positioned above, thereby improving the polishing efficiency of the structural part, and the specific implementation method steps are as follows: start the horizontally adjustable mobile camera 1, the horizontally adjustable mobile camera 2, and the horizontally adjustable mobile camera 3, and the corresponding imaging angles are α1, α2, and α3, respectively, and the corresponding num imaging stop points Point(i) on the imaging planes P1, P2, and P3, respectively, image each Area k (i, j), i.e., each k corresponding web can obtain 3*num images, and let the corresponding images be I1(h), I2(h), and I3(h), wherein h∈[1, num] represents the corresponding image number; based on the following formula, evaluate the 3*num images I1(h), I2(h), and I3(h) obtained,

[0136]

[0137] The formula can reflect and evaluate the large difference, i.e. high frequency characteristic feature in the image due to the joint mark, step difference, etc. In theory, the larger the Value of I1(h) is, the more the joint mark and other features that need to be polished exist; the 3*num calculated Value values are analyzed and compared, and are sorted from large to small, if there are β corresponding Value values greater than the set lower limit, it indicates that the corresponding web area has a joint mark and other features that need to be polished, and the area that needs to be polished is marked as G(m), m represents the number of areas that need to be polished.

[0138] 6. Joint mark transition area identification and positioning, the purpose of this step is to determine the position information of the web area that needs to be polished containing the joint mark in the polishing tool and the three-dimensional numerical model. The position information on the polishing tool can ensure accurate positioning of the web polishing area during polishing, and the position information in the three-dimensional numerical model can ensure the accuracy of the polishing head movement program generated in the numerical model. The implementation steps are: determining the coordinates of the web area that needs to be polished based on the area marked as G(m). Since the obtained data D is obtained through a series of transformations to obtain G(m), D and G have the same image coordinate system, i.e. G(m) determines the web area that needs to be polished and D has a one-to-one correspondence. Since D and O are obtained based on the same position camera, the coordinates of the web area that needs to be polished in O can be obtained based on G(m), and the coordinates of the web area that needs to be polished in the numerical model are also obtained to obtain an accurate polishing tool path.

[0139] 7. Area refinement data scanning, the purpose of this step is to determine the coordinates of the web area that needs to be polished based on G(m) again to avoid invalid scanning while increasing the accuracy of the preliminary determination of the polishing area. The implementation steps are: determining the coordinates of the web area that needs to be polished based on G(m) to complete the position information Loc(m) of the web area that needs to be polished in the numerical model, setting the refined scanning movement trajectory and path based on the position information Loc(m), and obtaining the accurate data PData(m) of each scanning area by using the generated path in a cyclic scanning manner. After m areas are processed, the data PData(m) is analyzed, the areas that do not need to be polished are removed from the m areas, and the polishing areas are retained to complete the second determination of the web area that needs to be polished based on G(m), and the finally determined web area that needs to be polished is DM a (i,j) represents, a represents the number of finally determined polishing areas, (i,j) represents the coordinates of the corresponding area in G.

[0140] 8. Final web area polishing boundary data acquisition, based on the finally determined web area that needs to be polished DM a(i,j), the mapping of the polishing area a is completed in the digital model, and the boundary data Cdata corresponding to a is calculated in the calculation digital model. The necessity of calculating the boundary data Cdata is that the boundary data can strictly limit the movement boundary of the polishing head, avoid polishing the area that does not need to be polished, and cause the size of the structural part to be out of tolerance and be scrapped.

[0141] 9. The first web polishing area planning, the purpose of this step is to determine the first polishing web, and the polishing area DM a (i,j) the distance data between the movement coordinates of the polishing head and the structural part, to avoid the problem of spatial interference caused by unreasonable polishing path setting, the first polishing area planning method: respectively to a geometric center of the polishing area to the distance dis of the initial polishing head, wherein the polishing area with the smallest distance dis from the polishing head is the first polishing area, this planning method can ensure that the invalid travel of the polishing head is as short as possible, thereby improving the polishing efficiency. The polishing head movement mode corresponding to the polishing area is designed, the purpose is to be efficient, the total movement path of the polishing head is short, thereby reducing the polishing time consumption while ensuring the polishing quality, and improving the polishing quality, the specific implementation is: by the polishing head radius DR, the diameter of the surrounding area corresponding to the polishing is calculated, the surrounding area calculation method has

[0142]

[0143] wherein r is the actual radius of the polishing head, r r indicates the radius of the corresponding polishing head ring surface, w indicates the side deflection angle of the polishing head, and λ indicates the front inclination angle of the polishing head; the number of polishing path optimization rows is designed, and the polishing area path optimization function number designed can be expressed as wherein Mi represents the polishing path length of Nv polishing areas, Sj represents the boundary polishing path length of the corresponding closed area, and Fz represents the auxiliary path length of the polishing head; based on the path optimization function, the generation of the polishing head processing track is completed, and the point data of the polishing head is generated based on the boundary data Cdata.

[0144] 10. Spatial non-interference distance measurement method: based on the first polishing area, the boundary data point Bdata(i,j) corresponding to the first polishing area is calculated, the radius direction of the polishing head is calculated, the rigid rotation area corresponding radius data GR is calculated, the point Mpoint(i,j) of the polishing head center movement is calculated, the Euclidean distance DMB between the point Mpoint(i,j) and the point Bdata(i,j) is directly calculated, and in the plane, the Euclidean distance DMB between the point Mpoint(i,j) and the point Bdata(i,j) satisfies DMB≤GR, and the flag is marked as 0, and DMB>GR, and the flag is marked as 1.

[0145] 11. According to the optimal spatial non-interference distance measurement method, the interference condition of the workpiece is judged to correct the polishing path. If there is spatial polishing interference, flag = 0, the path Mpoint(i,j) corresponding to the interference area is directly modified. If there is no interference, flag = 1, Mpoint(i,j) does not need to be modified.

[0146] 12. The remaining web to be polished is determined based on the cyclic planning method. The purpose of this step is to select the remaining web to be polished and ensure that the sum of the paths of the polishing head holes formed by the selected web regions is as small as possible. The specific method steps are as follows: let the first polishing region be the starting point to plan the global a-1 polishing regions. The corresponding planning evaluation function is

[0147]

[0148] wherein c1 represents a motion path weight coefficient, corresponding to the importance evaluation of the polishing head motion path index, F(Tk) is the first term representing the total length of the polishing head motion path, c2 represents a region polishing difficulty coefficient, Di represents the center clustering of adjacent two polishing regions, and Hi represents the polishing difficulty coefficient; global planning is performed to convert it into an F(Tk) extreme value problem, and the second polishing region is obtained based on the global optimal solution, and the same is true for the subsequent steps.

[0149] 13. Generating a polishing program for polishing. The purpose of this step is to generate a numerical control program that can control the motion of the polishing head in software according to the order of the evaluated polishing regions and the corresponding polishing head motion mode of each region, so as to accurately control the polishing head to move in a predetermined manner. The corresponding polishing region and polishing region polishing head path planning are as shown in Figure 3 .

[0150] 14. Polishing processing is performed. The polishing head is driven based on the generated polishing program to complete the polishing processing of a polishing region, and the first polishing of each region is completed. The first polishing may not be complete, so further polishing is needed. The polishing region reinspection is performed to judge.

[0151] 15. Polishing region reinspection. The steps of reinspection are the same as the foregoing and use the same technical means to evaluate the polished region. If the parameters corresponding to the region's knife marks, step differences, and other polishing features do not meet the requirements, further polishing is needed. The further polishing is only selected for the a first polishing regions that are not polished in place, and the process continues until all a regions meet the requirements.

[0152] The above is only the preferred embodiment of the present application, and does not hinder the present application in any form, and any simple modification or equivalent change of the above embodiment according to the technical essence of the present application falls within the protection scope of the present application.

Claims

1. An automatic polishing plan decision method for aircraft structural components, characterized in that, It comprises the following steps: Step 1. Position the structure to be polished and the web area; Step 2. Single web multi-angle image combination judgment; Step 3. Identify and position the transition area of the tool mark; Step 4. Regional refinement data scanning; Step 5. Obtain the final web area polishing boundary data; Step 6. First web polishing area planning; Step 7. Spatial non-interference distance calculation; Step 8. Determine the remaining web to be polished based on the cycle planning method; Step 9. Generate a polishing program for polishing; Step 10. Perform polishing processing; Step 11. Polishing area re-inspection; The position of the structure to be polished and the web area comprises: Imaging of the structural component to be polished is performed using a high-resolution three-channel fixed-position camera. This camera is equipped with RGB-D acquisition capability, where D represents depth information and RGB represents image information, yielding the raw analysis image. O Construct a global coarse positioning and recognition data model for the structural component to be ground, and complete the analysis of the original image. O The global coarse positioning identification data model for the positioning of structural components to be ground is as follows: where the goal of the mathematical model learning is to minimize the objective function , denotes the log loss of the target and non-target classes, denotes the predicted bounding box regression loss, denotes the Minni-batch mini-batch value, denotes the number of regression rectangular frames, denotes the balance weight, denotes the predicted probability that the Anchor is the target; and respectively denote the coordinates of the Ground Truth bounding box and the predicted frame, denotes the value corresponding to the background and foreground, the loss function contains two parts of the classification work background and structure, and the loss of the category to be polished and the regression loss of the coarse positioning frame. Based on The corresponding original analysis image O The corresponding result coordinates are obtained by performing coarse recognition positioning on the original analysis image t x , t y , t w , t h} are represented by t x is the normalized horizontal coordinate of the center of the standard frame, t y is the normalized vertical coordinate of the center of the standard frame, t w is the width value of the frame, t h is the height value of the frame; based on the result coordinates t x , t y , t w , t h The original analysis image O is further cropped to obtain a new image P; the processed image P is subjected to smoothing processing with different weights, and the corresponding Gaussian difference D( x , y , δ ) of adjacent weights is calculated after processing x is the value of the image pixel point in the X-axis direction, y is the value of the image pixel point in the Y direction, δ is is the corresponding standard deviation, and the key point K in the image after Gaussian difference processing is found, the Gaussian difference D( x , y , δ ) is subjected to interpolation operation through Taylor series expansion, the key point is corrected to obtain the corrected point K-P, and the fine positioning of the area to be polished in the image P is performed based on K-P, and the corresponding result with the same scale as the original analysis image O is represented by Obj. In step 1, the arrangement of the imaging camera corresponding to the structure to be polished satisfies the following conditions: the camera imaging the structure to be polished includes a fixed position camera, a horizontally adjustable mobile camera 1, a horizontally adjustable mobile camera 2, and a horizontally adjustable mobile camera 3. The imaging axis of the fixed position camera is concentric with the center of the structure to be polished and the geometric center of the structure to be polished, and there is a distance deviation <dis-o, which is a set threshold for the distance deviation between the center of the structure to be polished and the geometric center of the structure to be polished. The fixed position camera is located directly above the structure to be polished; The horizontally adjustable mobile camera 1 satisfies the movement in the imaging plane P1, and the corresponding horizontally adjustable mobile camera 1 has an inclination angle α1. There are num imaging docking points Point(i) on the corresponding imaging plane P1, i.e. corresponding i∈[1,num], the circular arc interval between adjacent imaging docking points Point(i) and Point(i+1) is equal to 360 / num, and each camera only images at the imaging docking point Point(i) corresponding to the inclination angle α1. The horizontally adjustable mobile camera 2 satisfies the movement in the imaging plane P2, and the corresponding horizontally adjustable mobile camera 2 has an inclination angle α2. There are num imaging docking points Point(i) on the corresponding imaging plane P2, i.e. corresponding i∈[1,num], the circular arc interval between adjacent imaging docking points Point(i) and Point(i+1) is equal to 360 / num, and each camera only images at the imaging docking point Point(i) corresponding to the inclination angle α2. The horizontally adjustable mobile camera 3 satisfies the movement in the imaging plane P3, and the corresponding horizontally adjustable mobile camera 3 has an inclination angle α3. There are num imaging docking points Point(i) on the corresponding imaging plane P3, i.e. corresponding i∈[1,num], the circular arc interval between adjacent imaging docking points Point(i) and Point(i+1) is equal to 360 / num, and each camera only images at the imaging docking point Point(i) corresponding to the inclination angle α3; The corresponding to-be-ground structure placement strategy is: after the structure is placed, the corresponding grinding area of the to-be-ground structure faces upward, if the structure needs to be ground on both sides, one side is selected to face upward first, and the other side is ground after the grinding is completed; the geometric center of the structure after being placed is fixed at the same position as the fixed-position camera to ensure the integrity of the structure area.

2. The method of claim 1, wherein, Corresponding mathematical model In order to prevent the influence of the BX and GT position and size on the regression scale, wherein BX is a model prediction area and GT is an actual result area, the BX coordinates need to be normalized, that is: BX→ GT→ wherein, is a value for a center of a prediction box x ; is a value for a center of a prediction box y ; is a value for a center of a standard box x ; is a value for a center of a standard box y ; is a value for a center of a potential target region box x ; is a value for a center of a prediction box y ; is a value for a width of a prediction box is a value for a height of a prediction box is a value for a center width of a prediction box is a value for a height of a potential target region is a value for a width of a standard box is a value for a height of a standard box , , , and , , , respectively represent the normalized center coordinates and the width and height values of the BX and the standard box. i is the index of the center of the prediction box; represents the prediction probability of the prediction box center as the target; represents the value corresponding to the background and foreground, the background is 0, and the foreground is 1; and respectively represent the coordinates of the GT enclosure and the coordinates of the prediction box, is a coordinate vector composed of ; represents the log loss corresponding to the identification and positioning of the workbench background and the area to be polished, that is: ; represents the regression deviation loss of the prediction box coordinates, and the corresponding mathematical model is represented as , wherein R represents the loss evaluation function, and let ; in the formula , the coordinate position deviation loss exists only when the foreground ( =1) exists; The output results of the classification model and the regression model are represented as } and } respectively, while represents the normalization processing, a balance weight is introduced and the normalization processing is implemented, cls the normalization is performed on a small batch value, while reg the normalization is performed on a value corresponding to the number of predicted frame center positions, wherein the balance weight corresponds to a positive integer; A mapping table Table(x, y) of the area position and the physical dimension is constructed, recorded in the original analysis image O The actual physical distance data Dis(point(x, y), C(x, y)) of each point point(x, y) in Obj to the geometric center / placement center C(x, y) of the structure to be polished.

3. The method of claim 2, wherein, The to-be-ground structure placement pose is corrected jointly with the design model based on the pose information of the placed structure, so that the pose of the structure in the design model is consistent with the pose of the to-be-ground structure. The specific joint correction strategy steps are: in the recognition result image Obj, four orientations are set, which are vertical upward, downward, horizontal left and right, and the reference point of the orientation is the center of Obj, that is, (rows / 2, cols / 2) is the reference, wherein row represents the number of rows of Obj, and cols represents the number of columns of Obj; in the design model software, the features corresponding to the vertical upward, downward, horizontal left and right of the recognition result image are adjusted respectively, and the features presented on the screen remain consistent with the features in Obj.

4. The method of claim 1, wherein, The tool mark transition region identification and positioning is specifically: based on the region mark G(m) of polishing to determine the coordinates of the web region to be polished. Since the obtained data D is obtained through a series of transformations to obtain G(m), the data D and G(m) have the same image coordinate system, that is, G(m) determines the web region to be polished and has a one-to-one correspondence with the data D, and the data D and the original analysis image O are obtained based on the same position camera, so that the coordinates of the web region to be polished in the original analysis image O are obtained based on G(m) mapping, and the coordinates of the web region to be polished in the digital model are completed at the same time to obtain an accurate polishing tool path.

5. The method of claim 1, wherein, The regional refinement data scanning specifically comprises: G(m) determining the web region coordinates needing polishing to complete the position information Loc(m) needed to be polished in the digital model, setting the refined scanning moving track and path based on the position information Loc(m), obtaining the accurate data PData(m) of each scanning region in a cyclic scanning manner according to the generated path, processing the m regions, analyzing the data PData(m), removing the regions not needing polishing from the m regions, retaining the regions needing polishing, completing the secondary judgment of determining the web candidate regions needing polishing based on G(m), and finally determining the web regions needing polishing as DM a (i,j) represents, a represents the number of final regions needing polishing, and (i,j) represents the coordinates of the corresponding region in G(m).

6. The method of claim 1, wherein, The final web area polishing boundary data acquisition is specifically: based on the final determined web area to be polished, DM a (i,j) is used to complete the mapping of the polishing area in the digital model, and the corresponding boundary data Cdata in the calculation digital model.

7. The method of claim 1, wherein, The first web grinding area planning is specifically: the distances dis from the geometric centers of the a grinding areas to the initial grinding head are calculated respectively, wherein the grinding area with the smallest distance dis to the grinding head is the first grinding area, the diameter of the surrounding area corresponding to the grinding is calculated based on the grinding head radius DR, and the surrounding area calculation method has wherein r is the actual radius of the polishing head, r r represents the corresponding polishing head torus radius, w represents the side rake angle of the polishing head, represents the front rake angle of the polishing head; the polishing path optimization function of the designed polishing path is represented as wherein M i represents the polishing path length of the Nv polishing areas, S j represents the corresponding closed area boundary polishing path length, F z represents the auxiliary path length walked by the polishing head; Nk is the number of polishing areas, i represents the polishing path length index, j represents the closed area boundary polishing path length index, and z represents the auxiliary path length index walked by the polishing head. The generation of the polishing head machining track is completed based on the path optimization function, and the generation of the polishing head point data is realized based on the boundary data Cdata.

8. The method of claim 1, wherein, The spatial non-interference distance measurement method is: based on the obtained first grinding area, the boundary data point Bdata(i,j) is calculated, the radius data GR of the rigid rotation area in the radius direction of the grinding head during grinding is calculated, the point Mpoint(i,j) to which the center of the grinding head moves is calculated, and the Euclidean distance DMB between the point Mpoint(i,j) and the point Bdata(i,j) is directly calculated; in the plane, if the Euclidean distance DMB between the point Mpoint(i,j) and the point Bdata(i,j) satisfies DMB≤GR, the flag is marked as 0, and if DMB>GR, the flag is marked as 1; According to the optimal spatial non-interference distance measurement method, the workpiece interference condition is judged to correct the grinding path. If there is spatial grinding interference flag=0, the path Mpoint(i,j) corresponding to the interference area is directly modified, and if there is no interference flag=1, the Mpoint(i,j) does not need to be modified.

9. The method of claim 1, wherein, The remaining to-be-ground webs are determined based on the loop planning method, and the first grinding area is taken as the starting point to plan the a-1 grinding areas, and the corresponding planning evaluation function is Wherein, c1 represents the motion path weight coefficient, and the corresponding meaning is the importance evaluation of the polishing head motion path index, F(T k ) The first term represents the total length of the polishing head motion path, c2 represents the area polishing difficulty coefficient, D i H i represents the center clustering of the adjacent two polishing areas, and H k represents the polishing difficulty coefficient; global planning is carried out, which is converted into F(T k ) Extreme value problem, based on the global optimal solution to obtain the second polishing area, and in turn, h represents the index of the center clustering measurement of the adjacent two polishing areas, and a represents the number of global polishing areas.

10. The method of claim 1, wherein, The grinding program is generated to grind specifically: the numerical control program for controlling the movement of the grinding head is generated.

11. The method of claim 1, wherein, The grinding process is specifically: based on the generated grinding program, the grinding head is driven to complete the grinding of the a grinding areas, the first grinding of each area is completed, and the grinding area re-inspection is used for judgment.

12. The method of claim 1, wherein: The polishing area re-inspection specifically refers to: evaluating the already polished area, if the parameters corresponding to the features of the area such as the tool joint mark and the step difference to be polished do not meet the requirements, further polishing is needed, and the further polishing corresponds to the area selected from the first polishing area which is not polished well, until all the a areas meet the requirements.

Images