High-precision positioning device and positioning method for mechanical bridge box component machining

Abstract

The application relates to the technical field of image processing, in particular to a high-precision positioning device and positioning method for machining of mechanical bridge box parts, which comprises the following steps: high-precision first point cloud is collected from a first visual angle by using a first structured light measurement method based on phase shift-Grey code coding, and a structure continuous region is extracted therefrom; fast second point cloud is collected from a second visual angle by using a second structured light measurement method based on a pure phase shift method. The first point cloud matched with the second point cloud in the structure continuous region is taken as a reliable reference, and the second phase shift corresponding to the second point cloud is corrected in combination with a matching error, so that final point cloud is obtained. The final point cloud and the first point cloud are combined to form reference point cloud of the surface. When the next surface is machined, the positioning error correction of the reference point cloud is carried out. The application realizes dynamic reference reconstruction based on a process live, effectively cuts off an error transmission chain in multi-surface machining, and guarantees the precision of machining positioning.

Description

Technical Field

[0001] This invention relates to the field of image processing technology, and more specifically to a high-precision positioning device and positioning method for machining mechanical bridge box parts. Background Technology

[0002] As a key transmission component in engineering machinery, the machining accuracy of the mechanical bridge box (especially the coaxiality of the multi-bearing hole system) directly determines the performance and reliability of the entire machine. These components are typically hexahedral box-shaped structures, requiring multi-process, multi-faceted sequential machining (e.g., machining the bearing hole end face of face A first, then the corresponding hole on opposite face B, and subsequently machining other faces). In traditional high-precision machining paradigms, whether based on precision fixtures, zero-point positioning systems, or in-machine contact probes (such as Renishaw probes), the core relies on the precise construction and long-term stability of a preset workpiece coordinate system, using this coordinate system for positioning all processes.

[0003] However, in actual machining, factors such as the initial shape and position errors of the workpiece blank, the repeated positioning errors of the fixture, the micro-deformation caused by clamping force, and the stress release and thermal deformation during the cutting process are difficult to completely avoid. During the multiple flipping and reclamping processes of "A surface → B surface → C surface", these errors will be transmitted and accumulated step by step, causing the positioning datum of the subsequent machined surfaces to actually deviate from the theoretical position, ultimately damaging the relative positional accuracy between key features. For example, it can cause the coaxiality of the bearing holes at both ends to exceed the tolerance range, becoming a major bottleneck in improving machining accuracy and consistency. Summary of the Invention

[0004] To address the aforementioned problems, this invention provides a high-precision positioning device and method for machining mechanical bridge box components.

[0005] The high-precision positioning device and positioning method for machining mechanical bridge box parts of the present invention adopts the following technical solution: One embodiment of the present invention provides a high-precision positioning method for machining mechanical bridge box components, the method comprising the following steps: After each surface of a component is processed, a first point cloud of the surface is acquired from a first viewpoint using a first structured light measurement method, and several structurally continuous regions in the first point cloud are obtained. A second point cloud of the surface is acquired from a second viewpoint using a second structured light measurement method. In the first structured light measurement method, a phase-shift Gray code encoding algorithm is used to obtain the first phase shift of each first pixel, and in the second structured light measurement method, a phase shift algorithm is used to obtain the second phase shift of each second pixel. The first point cloud is obtained by the first phase shift, and the second point cloud is obtained by the second phase shift. For the first point cloud that matches the second point cloud and is located within the structural continuous region, and the matching error during matching, the second phase shift of all second pixels is corrected using the matched first point cloud and the matching error. The final point cloud obtained after the correction of the second phase shift is merged with the first point cloud to obtain the reference point cloud of the surface. When processing another surface of the component, the positioning error is corrected based on the reference point cloud.

[0006] Preferably, the specific steps for obtaining several structurally continuous regions in the first point cloud are as follows: Cluster all the first point clouds to obtain all categories. The categories containing the most first point clouds are denoted as reference categories, and the first point clouds in each reference category are regarded as structurally continuous regions.

[0007] Preferably, the specific steps for obtaining the first point cloud that matches the second point cloud and is located within a structurally continuous region are as follows: A point cloud registration algorithm is used to spatially transform the first point cloud so that the first point cloud and the second point cloud are aligned. The first point cloud in all structurally continuous regions is denoted as a continuous point cloud after spatial transformation. The second point cloud is matched with the continuous point cloud to minimize the matching distance of all matching pairs. The matching distance refers to the Euclidean distance between the second point cloud and the continuous point cloud in each matching pair. The first point cloud corresponding to the continuous point cloud in any matching pair is regarded as the first point cloud that is matched with the second point cloud and is located in the structurally continuous region.

[0008] Preferably, the method for calculating the matching error includes: Calculate the first mean of the matching distances of all matching pairs, and use the ratio of the matching distance of any matching pair to the first mean as the matching error of the first and second point clouds of any matching pair.

[0009] Preferably, the specific steps for correcting the second phase shift of all second pixels using the matched first point cloud and the matching error are as follows: For any second point cloud, if there exists a matching first point cloud within a structurally continuous region, then the second point cloud is denoted as the target point cloud, and the second pixel corresponding to the target point cloud is denoted as the reference pixel. For the first point cloud that matches the target point cloud, the point cloud coordinates after the first point cloud is transformed and aligned to the target point cloud are used as the new point cloud coordinates of the target point cloud. Based on the new point cloud coordinates, the final second phase shift of the reference pixel is solved. The correction weight of the reference pixel is determined by the matching error when matching the target point cloud with the matching first point cloud. The correction weight is negatively correlated with the matching error. The final second phase shift of the second pixel other than all reference pixels is determined based on the final second phase shift of all reference pixels and the correction weight. The final second phase shift of all second pixels is the result of correcting the second phase shift of all second pixels.

[0010] Preferably, the specific steps for determining the final second phase shift of the second pixel point other than all reference pixels based on the final second phase shift of all reference pixels and the corrected weights are as follows: For any reference pixel, the order of the reference pixel is determined based on the final second phase shift of the reference pixel. Starting from the reference pixel, traverse the second pixel in the same row that is outside the reference pixel, and assign a level to the traversed second pixel according to the level of the reference pixel; the level assigned to all second pixels in the same row that is outside the reference pixel is marked as the reconstruction sub-level of the reference pixel; For any second pixel other than all reference pixels, mark all reference pixels in the same row as the second pixel as the target pixels of the second pixel; for all reconstruction sub-levels obtained by the second pixel based on all target pixels, use the correction weights of all target pixels to fuse all reconstruction sub-levels as the final level of the second pixel, and determine the final second phase shift of the second pixel based on the final level of the second pixel.

[0011] Preferably, the specific steps of fusing all reconstructed secondary components are as follows: For any second pixel other than all reference pixels and all target pixels of that second pixel, calculate the absolute value of the difference between the x-coordinate of each target pixel and the pixel, and denote it as the expansion distance of each target pixel. Based on the expansion distance of each target pixel and the correction weight of each target pixel, determine the fusion weight of each target pixel. The fusion weight is positively correlated with the correction weight and negatively correlated with the expansion distance. For the second pixel, all the reconstructed sub-levels obtained based on all the target pixels are weighted and summed using the fusion weights of all the target pixels. The weighted sum is used as the final level of the second pixel.

[0012] Preferably, when machining another surface of the component, the positioning error correction based on the reference point cloud includes the following specific steps: The CAD model of the component is converted into an initial point cloud. A point cloud registration algorithm is used to spatially transform the reference point cloud to align it with the initial point cloud. For the aligned reference point cloud, the measured geometric vectors of the structural features in the reference point cloud are extracted, and the theoretical geometric vectors of the structural features in the initial point cloud are extracted. The transformation parameters when transforming the theoretical geometric vectors into the measured geometric vectors are calculated and denoted as the machining error. When machining another surface of the component, the machining error is used to correct the workpiece coordinate system preset during the machining process. The measured or theoretical geometric vector of the structural feature refers to the vector of the central axis of the cylinder in the reference point cloud or initial point cloud.

[0013] Preferably, the specific steps of assigning a level to the traversed second pixel based on the level of the reference pixel are as follows: Starting from the reference pixel, traverse to the right or left of the second pixel in the same row that is outside the reference pixel. Calculate the absolute value of the difference between the wrapping phase of the currently traversed second pixel and the previously traversed second pixel. If the absolute value of the difference is greater than π, the order of the currently traversed second pixel is equal to the order of the previously traversed second pixel plus or minus 1. If the absolute value of the difference is less than or equal to π, the order of the currently traversed second pixel is equal to the order of the previously traversed second pixel. Wherein, if the currently traversed second pixel is an adjacent pixel of the reference pixel, the previously traversed second pixel is the reference pixel.

[0014] Another embodiment of the present invention provides a high-precision positioning device for machining mechanical bridge components. The device includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor runs the computer program, it implements all the steps of the high-precision positioning method for machining mechanical bridge components described above.

[0015] The beneficial effects of the technical solution of the present invention are: This invention breaks the error propagation chain in multi-process machining, achieving precise positioning for subsequent processes. Traditional multi-faceted machining of mechanical bridge boxes relies on a preset, static workpiece coordinate system. Any clamping, deformation, or machining errors accumulate step-by-step during multi-process flow, ultimately compromising critical precision such as the coaxiality of bearing holes. This invention fundamentally changes this paradigm by acquiring the point cloud of each surface after machining and using this reference point cloud to correct positioning errors when machining the next surface. It resets the positioning reference for each process from a workpiece coordinate system that may already contain accumulated errors to a reference point cloud reconstructed based on the actual machining results of the previous process. This is equivalent to setting up an error correction link at each process node, effectively blocking the backward propagation of errors and providing direct support for ensuring the overall positional accuracy of the final part.

[0016] Furthermore, this invention provides a highly accurate and efficient online 3D measurement method for the aforementioned error correction process by integrating a first structured light measurement method (phase-shift Gray code encoding) and a second structured light measurement method (pure phase-shift method) into a multi-view measurement process. Specifically: the phase-shift Gray code encoding method (used to acquire the first point cloud) has high accuracy and can serve as reliable reference data, but its measurement speed is relatively slow because it requires projecting multiple Gray code patterns. The pure phase-shift method (acquiring the second point cloud from different perspectives) has a fast measurement speed and is suitable for online detection, but the second phase shift obtained through spatial phase unfolding is prone to cumulative errors, resulting in insufficient accuracy when used alone. This invention does not simply stitch together point clouds from different perspectives, but rather uses the structurally continuous and reliable portion of the high-precision first point cloud as a reference, performs directional correction on the second phase shift corresponding to the rapidly acquired second point cloud, and then merges the corrected, accurate final point cloud with the first point cloud. This method cleverly leverages the advantages of the two technologies mentioned above while avoiding their respective disadvantages, ultimately achieving the acquisition of a complete reference point cloud that can be used for high-precision positioning correction within a limited time. This provides an important data foundation for real-time positioning error correction after each process.

[0017] In summary, this invention balances the core contradiction between measurement accuracy and measurement efficiency in the reference point cloud reconstruction results after each process through dual structured light measurement fusion and data correction process. This allows the invention to be based on relative positioning of the processed features rather than absolute positioning of a preset coordinate system (i.e., not relying on the original workpiece coordinate system). This clears the positioning accuracy error before each processing step and locks it at the best result of the current process. During processing, a closed-loop processing flow of processing, measurement, and positioning error correction is formed, achieving the fundamental goal of efficiently and accurately cutting off the processing error transmission chain. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0019] Figure 1 This is a flowchart illustrating the steps of a high-precision positioning method for machining mechanical bridge box components, provided in an embodiment of the present invention. Detailed Implementation

[0020] To further illustrate the technical means and effects adopted by the present invention to achieve its intended purpose, the following, in conjunction with the accompanying drawings and preferred embodiments, details the specific implementation, structure, features, and effects of the high-precision positioning device and positioning method for machining mechanical bridge box components proposed according to the present invention. In the following description, different "one embodiment" or "another embodiment" do not necessarily refer to the same embodiment. Furthermore, specific features, structures, or characteristics in one or more embodiments can be combined in any suitable form.

[0021] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0022] The following description, in conjunction with the accompanying drawings, details the specific scheme of the high-precision positioning device and positioning method for machining mechanical bridge box parts provided by the present invention.

[0023] Please see Figure 1 The diagram illustrates a flowchart of a high-precision positioning method for machining mechanical bridge box components according to an embodiment of the present invention. The method includes the following steps: Step S101: After each surface of a component is processed, a first point cloud of the surface is acquired from a first viewpoint using a first structured light measurement method, and several structurally continuous regions in the first point cloud are obtained. The first structured light measurement method uses a phase-shift Gray code encoding algorithm to obtain the first phase shift of each first pixel; the first point cloud is obtained from the first phase shift.

[0024] When using CNC machine tools to process original workpieces (such as hexahedral castings or welded blanks) into high-precision mechanical bridge box parts, it is necessary to perform multiple flips and multiple processes on multiple surfaces of the original workpiece. For example, after machining (finishing or roughing) the bearing hole end face on surface A of the original workpiece, the corresponding hole is machined on surface B opposite A, and then other holes are machined on surface C.

[0025] Traditional high-precision machining (such as solutions based on zero-point positioning and contact probes) heavily relies on the precise alignment and long-term stability of a preset workpiece coordinate system. In the multi-process, multi-face machining of the original workpiece, any minute clamping repeatability error, fixture wear, or workpiece micro-deformation will be transmitted and amplified with each process iteration, resulting in large positioning errors in subsequent processes and ultimately damaging the machining accuracy of the parts (e.g., damaging the coaxiality of the bearing holes at both ends).

[0026] Based on this, in this embodiment, after each surface of a component is processed, the first point cloud of the processed surface is acquired from the first perspective using the first structured light measurement method. The positioning error of the processing process is corrected based on the first point cloud to ensure the accuracy of the processing positioning. This embodiment breaks the chain of cumulative errors. It does not assume that the clamping and positioning of the previous process is perfect. Instead, it directly acquires the actual geometric state of the processed surface (described by the first point cloud) and uses it as the unique and objective positioning reference for the next process.

[0027] The first structured light measurement method refers to the process of acquiring an image of the processed surface using a 3D structured light camera, denoted as the first image, and reconstructing a three-dimensional point cloud (i.e., the first point cloud) based on the first image. The projector in the 3D structured light camera projects a sinusoidal fringe pattern onto the surface of the component, and the CCD camera in the 3D structured light camera acquires the first image of the component surface; the CCD camera acquires grayscale images of 1920×1080 pixels. Because the surface undulations (i.e., depth information) of the component will modulate and deform the phase of the sinusoidal fringe pattern, the stripes in the first image undergo phase shift, referred to as phase shift. The process of reconstructing the three-dimensional point cloud (i.e., the first point cloud) includes: obtaining the first phase shift of each first pixel in the first image using a phase shift-Gray code encoding algorithm; and then obtaining the three-dimensional point cloud coordinates corresponding to each first pixel based on the first phase shift of each first pixel. The point cloud coordinates corresponding to all first pixels in the first image constitute the first point cloud.

[0028] It should be noted that this embodiment uses a semantic segmentation model (such as the DeepLabV3 model) to segment the semantic regions of the components in the first image. All pixels mentioned in this embodiment refer to pixels within the semantic regions of the components, and background pixels outside the semantic regions of the components are not considered.

[0029] The 3D structured light camera and phase-shift Gray code encoding algorithm are well-known technologies, and their specific processes will not be described in detail in this embodiment.

[0030] The method for obtaining point cloud coordinates is also a well-known technique, and the specific calculation formula involved is as follows: The coordinates of any pixel (e.g., any first pixel) are (u, v), and its corresponding point cloud coordinates are (x, y, z): Where z represents the depth value of a pixel. ; , This represents the principal point coordinates of the CCD camera.

[0031] This represents the first phase shift of a pixel, equivalent to parallax in binocular ranging technology.

[0032] in ; The x-coordinate of the projected sinusoidal fringes. ; Represents the absolute phase of a pixel. ;in The wrap phase of a pixel is represented, and the value of the wrap phase ranges from [-π, π]. k represents the order of the sine fringe corresponding to the first pixel. The methods for obtaining the wrap phase, absolute phase, and order will be described later.

[0033] It should be noted that when the calculated first phase shift for any pixel is less than or equal to 0, the point cloud coordinates corresponding to that pixel will no longer be obtained (i.e., the point cloud is missing). In subsequent steps, when calculating any phase shift (such as the first similarity, the second phase shift, etc.), as long as there is a case where the phase shift is less than or equal to 0, the point cloud coordinates will no longer be calculated, and it will be regarded as a missing point cloud.

[0034] T represents the period of the sinusoidal stripe (in pixels); B represents the baseline distance between the optical center of the CCD camera and the optical center of the projector in the 3D structured light camera. , This represents the horizontal and vertical focal lengths of the CCD camera. In this embodiment, T is 48 pixels. , They are 1500 pixels and 1800 pixels respectively, and B is 180mm.

[0035] The above formulas are all well-known technologies, and the specific reasoning process will not be described in detail in this embodiment.

[0036] The continuous structural region refers to the area where the structure of the surface changes continuously after processing.

[0037] Step S102: The second point cloud of the surface is acquired from the second viewpoint using the second structured light measurement method. In the second structured light measurement method, the second phase shift of each second pixel is obtained using a phase shift algorithm. The second point cloud is obtained by the second phase shift.

[0038] In this embodiment, it is considered that the surface of the processed parts is a relatively smooth surface with relatively strong specular reflection capability. The sinusoidal stripe pattern projected by the projector mentioned above may be overexposed or underexposed in some areas. The phase information of the sinusoidal stripe pattern is completely lost, which makes it impossible to accurately obtain the first phase shift of some first pixels. This leads to holes or severe noise in the first point cloud. At this time, the first point cloud cannot be used to correct the positioning error of the processing process.

[0039] In this embodiment, the second point cloud of the surface is acquired from the second perspective using the second structured light measurement method. The second structured light measurement method refers to the process of acquiring an image of the processed surface (denoted as the second image) using a 3D structured light camera and reconstructing the processed surface into a three-dimensional point cloud (i.e., the second point cloud) based on the second image.

[0040] Since the second and first viewpoints refer to different viewing angles of the 3D structured light camera, the overexposed or underexposed areas of the sinusoidal fringe pattern in the first and second structured light measurement methods may be different. At this time, although both the second and first point clouds have holes or severe noise, the areas where the holes or severe noise appear are different. In this embodiment, the second and first point clouds will be merged in the future, and the two will complement each other to obtain an accurate point cloud without holes or severe noise, which helps to accurately correct the positioning error in the processing.

[0041] This embodiment further considers that, when acquiring the first point cloud, a phase-shift Gray code encoding algorithm is used. The advantage of this well-known algorithm is that it can accurately acquire the first phase shift of the first pixel, thereby ensuring the accuracy of the first point cloud (except for areas with holes or severe noise). The disadvantage of this well-known algorithm is that it mainly consists of two processes: The first process is to use a phase-shift method (e.g., a four-step phase-shift method) to acquire the wrapping phase of each first pixel, where the value range of the wrapping phase is [-π, π]. However, what is needed to acquire the first phase shift or the first point cloud is the absolute phase of each first pixel (i.e., the first phase shift cannot be directly obtained from the wrapping phase). The absolute phase is equal to the wrapping phase plus 2π times k (see the formula in step S101), where k represents the order of the stripe (i.e., the sinusoidal stripe projected by the projector) corresponding to each first pixel. In the second process, a large amount of Gray code (e.g., 10-bit Gray code) is projected onto the surface of the component using the projector, and k is obtained using the Gray code encoding method, thereby finally obtaining the first phase shift of each first pixel (see the formula in step S101).

[0042] In summary, the phase-shift Gray code encoding algorithm requires not only projecting a sine stripe pattern but also projecting Gray code multiple times, thus incurring a significant time cost.

[0043] This step utilizes a phase-shifting algorithm to obtain the second phase shift of each second pixel, where the second point cloud is obtained from the second phase shift (similar to the method used to obtain the first point cloud). This phase-shifting algorithm mainly consists of two processes: the first process is the same as described above. In the second process, instead of obtaining the order through Gray code projection, the order of each second pixel is obtained through spatial phase expansion, ultimately yielding the second phase shift of each second pixel.

[0044] In summary, compared to the phase-shift Gray code algorithm, the phase-shift algorithm, although obtaining a second phase shift with a larger error (e.g., accumulated error due to shadows and noise), has a lower time cost and faster computation speed. Based on this, and considering subsequent processes, it can be seen that when facing numerous processing steps, after each step (i.e., after processing each surface), point cloud data can be quickly and accurately obtained to correct positioning errors during processing. This avoids the problem of excessively long processing time caused by breaking the chain of accumulated errors, thus improving processing efficiency (i.e., accurate and fast).

[0045] The spatial phase unrolling method is a conventional method. The process is as follows: traverse each second pixel in the same row from left to right. If the absolute value of the difference between the wrapped phase of the traversed second pixel and the second pixel to its right is greater than π, it indicates that these two adjacent second pixels are not in the same period. In this case, the order (k value) of the traversed second pixel is equal to the order (k value) of the second pixel to its right plus 1. If the absolute value of the difference is less than or equal to π, it indicates that these two adjacent second pixels are in the same period. In this case, the order (k value) of the traversed second pixel is equal to the order (k value) of the second pixel to its right. Specifically, if there is no second pixel to its right, the order (k value) of the traversed second pixel is equal to 0.

[0046] Therefore, it can be seen that the second image, under the interference of shadows and noise, will cause errors in the wrapping phase of the second pixel. These errors may accumulate during the acquisition process, which will lead to the cumulative error between the absolute phase or the second phase shift of the second pixel and the second point cloud.

[0047] Step S103: Obtain the first point cloud that matches the second point cloud and is located within the structurally continuous region, and calculate the matching error during matching.

[0048] For a first point cloud that matches the second point cloud and is located within a structurally continuous region, on the one hand, the first point cloud in this part matches the second point cloud; the matching means that the first point cloud and the second point cloud in this part represent the same area of ​​the component surface and describe the same structure of the component surface; on the other hand, the first point cloud in this part is located within a structurally continuous region, that is, the first point cloud in this part and the second point cloud that matches it not only describe the same structure of the component surface, but the structure is a smooth and continuous component area without abrupt changes.

[0049] This embodiment only performs matching on the first point cloud within the continuous structural region. The purpose is that the component structure within the continuous structural region is continuous without abrupt changes. This can suppress the influence of the point cloud in the second point cloud that has abrupt position changes due to large errors, and avoid confusion and incorrect matching between the first point cloud of the non-smooth continuous structure of the component (the point cloud in this part has abrupt position changes) and the second point cloud that has abrupt position changes due to errors.

[0050] The smaller the matching error, the smaller the error between the first point cloud and the second point cloud that it matches, and the more accurately and reliably they can describe the same smooth and continuous structure on the surface of the component.

[0051] Step S104: Correct the second phase shift of all second pixels using the matched first point cloud and the matching error. Merge the final point cloud obtained by the corrected second phase shift with the first point cloud to obtain the reference point cloud of the surface.

[0052] As described in step S102, the process of obtaining the second phase shift based on the phase shift algorithm is computationally fast, but the accuracy of the second phase shift (and the second point cloud) is low and the error is large. However, the process of obtaining the first phase shift based on the phase shift-Gray code encoding algorithm is slower, but the accuracy of the first phase shift (and the first point cloud) is higher. For the matching first and second point clouds, the second phase shift of the second pixel corresponding to the second point cloud is corrected by the first point cloud, so that the error of the corrected second phase shift is reduced and its accuracy is improved. Moreover, the correction process is based on the matching error, avoiding the problem of incorrect correction of the second phase shift when the matching error is caused by the error of the second point cloud itself. This avoids the situation where the error of the second point cloud is fed back into the corrected second phase shift. It is equivalent to the high accuracy of the first point cloud providing a reliable correction basis for the inaccurate second point cloud and second phase shift, thereby ensuring the accuracy of the corrected second phase shift and the accuracy of the final point cloud.

[0053] In summary, step S102 employs a phase-shift algorithm (without continuing to use the phase-shift-Gray code encoding algorithm) to improve speed, which results in low accuracy of the second phase shift. This step, however, improves the accuracy of the corrected second phase shift while retaining the high speed characteristic. This helps to accurately correct the positioning error of the processing process based on the reference point cloud, and provides a data foundation for accurate positioning in the subsequent processing process.

[0054] Step S105: When machining another surface of the part, perform positioning error correction based on the reference point cloud.

[0055] Since the reference point cloud is obtained again after each process, it does not rely on the absolute positioning of the preset coordinate system (that is, it does not rely on the original workpiece coordinate system), but is based on the relative positioning of the processed features. This clears the positioning accuracy error before each processing step and locks it at the best result of the current process. In subsequent processing, a closed-loop processing flow of processing, measurement, and positioning error correction is formed, which ensures the accuracy and efficiency of the entire processing process.

[0056] As a preferred example, the method for obtaining structurally continuous regions is as follows: Clustering is performed on all point cloud coordinates of the first point cloud to obtain all categories. Each category contains multiple point cloud coordinates. The categories containing the most point cloud coordinates are denoted as reference categories. The first point clouds in the reference categories have a continuous and concentrated distribution, and they are most likely to describe the regions where the surface structure of the part changes smoothly and continuously. The set of point cloud coordinates of the first point clouds in each reference category is denoted as a structurally continuous region, or in other words, the first point clouds in each reference category constitute a structurally continuous region.

[0057] As an example, clustering is performed using the mean-shift clustering algorithm, with the clustering bandwidth set to 2.8 mm, the distance threshold set to 0.8 mm, and the maximum number of iterations set to 300.

[0058] As an example, the 20 categories containing the most point cloud coordinates are used as reference categories. Specifically, if there are multiple categories containing the most point cloud coordinates (more than 20), then only all of these categories should be used as reference categories. Similarly, if there are multiple categories with the most and second-most point cloud coordinates (more than 20), then only all of these categories should be used as reference categories, and so on, to obtain all reference categories.

[0059] As a preferred example, obtaining the first point cloud that matches the second point cloud and is located within a structurally continuous region includes the following methods: A point cloud registration algorithm is used to perform spatial transformations (including rotation and translation) on the first point cloud to align it with the second point cloud. The first point cloud within all structurally continuous regions is denoted as a continuous point cloud after the spatial transformation. The second point cloud is then matched with the continuous point cloud to obtain all matching pairs. Each matching pair contains one continuous point cloud and one second point cloud. The Euclidean distance between the two point clouds in each matching pair is denoted as the matching distance (or matching cost) for each pair. The matching distance (matching cost) of all matching pairs is minimized.

[0060] The first point cloud corresponding to any consecutive point cloud in any matching pair is the first point cloud that matches the second point cloud and is located within the structurally continuous region.

[0061] As an example, matching the second point cloud with the continuous point cloud to obtain all matching pairs includes the following methods: The KM matching algorithm is used to match all second point clouds with all continuous point clouds to obtain all matching pairs with the minimum matching cost.

[0062] As an example, the point cloud registration algorithm used is the Point-to-Plane ICP algorithm, which can perform spatial transformations (including rotation and translation transformations) on the first point cloud to align the first point cloud and the second point cloud. This algorithm is well known, and its specific process will not be described in detail in this embodiment. Other point cloud registration algorithms may also be used in other embodiments, and this embodiment does not impose any specific limitations.

[0063] As a preferred example, the matching error is calculated as follows: Calculate the first mean of the matching distances for all matching pairs. Use the ratio of the matching distance of any matching pair to the first mean as the matching error between any pair of matched first and second point clouds. Specifically, when the first mean equals 0, set the matching error of any pair of matched first and second point clouds to 0. The purpose of calculating the ratio to the first mean is to remove dimensions.

[0064] As a preferred example, the second phase shift of all second pixels is corrected using the matched first point cloud and the matching error, including the following method: For any second point cloud, if there exists a matching first point cloud within a structurally continuous region, then the second point cloud is denoted as the target point cloud, and the second pixel corresponding to the target point cloud is denoted as the reference pixel.

[0065] For the first point cloud that matches the target point cloud, the point cloud coordinates of the first point cloud after transformation and alignment with the target point cloud are used as the new point cloud coordinates of the target point cloud. Based on the new point cloud coordinates, the final second phase shift of the reference pixel corresponding to the target point cloud is solved; the specific formula is... ,in This represents the depth value in the new point cloud coordinates (i.e., the third dimension of the new point cloud coordinates). This represents the final second phase shift of the reference pixel.

[0066] For all target point clouds in the second point cloud, and for all reference pixels corresponding to each target point cloud, the final second phase shift of all reference pixels is solved by inverse solving using the method described above.

[0067] The correction weight of the reference pixel is determined by the matching error between the target point cloud and the matching first point cloud, and the correction weight is negatively correlated with the matching error.

[0068] The final second phase shift of the second pixel outside of all reference pixels is determined based on the final second phase shift of all reference pixels and the corrected weights.

[0069] The final second phase shift of all second pixels is the result of correcting the second phase shift of the second pixel.

[0070] In this process, since the target point cloud matches the first point cloud and the point cloud coordinates of the first point cloud are more accurate, the final second phase shift of the reference pixel is solved by reverse calculation from the first point cloud with more accurate point cloud coordinates. This makes the final second phase shift of some second pixels (i.e., all reference pixels) more accurate. The correction weight is a quantification and evaluation of the accuracy of the final second phase shift. The larger the correction weight (i.e., the smaller the matching error), the smaller the distance between the target point cloud and the matching first point cloud (the smaller the matching error). The more accurately and reliably the target point cloud and the matching first point cloud can describe the same smooth and continuous structure of the component surface, the more the point cloud coordinates of the matched first point cloud can be replaced with the new point cloud coordinates of the target point cloud, thus ensuring the accuracy of the final second phase shift of the reference pixel. Therefore, it is more necessary to use the final second phase shift of the reference pixel as a reference anchor point to accurately and reliably determine the final second phase shift of other second pixels.

[0071] In summary, this process ensures the accuracy of the correction result of the second phase shift (i.e., the final second phase shift), which helps to accurately correct the positioning error of the machining process based on the reference point cloud, and provides a data foundation for accurate positioning in the subsequent machining process.

[0072] As an example, the correction weights for the reference pixels are determined by the matching error between the target point cloud and the first matching point cloud, using the following formula: Let w = exp(-d), where w represents the correction weight, d represents the matching error, and exp() represents an exponential function with the natural constant as the base.

[0073] As a preferred example, the final second phase shift of the second pixel other than all reference pixels is determined based on the final second phase shift of all reference pixels and the corrected weights, including the following method: For any reference pixel, the level of the reference pixel is determined based on the wrapping phase and the final second phase shift (determined by the formula in step S101).

[0074] Starting from the reference pixel, traverse to the right (or left) of the second pixel in the same row, excluding the reference pixel. Calculate the absolute value of the difference between the wrap-around phase of the currently traversed second pixel and the previously traversed second pixel. If the absolute value of this difference is greater than π, the order of the currently traversed second pixel is equal to the order of the previously traversed second pixel plus (or minus) 1; if the absolute value of the difference is less than or equal to π, the order of the currently traversed second pixel is equal to the order of the previously traversed second pixel. Where the currently traversed second pixel is an adjacent pixel of the reference pixel, the previously traversed second pixel is the reference pixel.

[0075] It should be noted that in the above process, when traversing to the right, the order is incremented by 1, and when traversing to the left, the order is decremented by 1.

[0076] At this point, for any given reference pixel, and for all other second pixels in the same row besides the reference pixel, all other second pixels are reassigned a level, and the level reassigned to all other second pixels in the same row is marked as the reconstructed secondary level of the reference pixel.

[0077] For all reference pixels, the level is reassigned to all other second pixels in the same row using the method described above. It should be noted that when a row contains one or more reference pixels, the same second pixel in the same row will receive one or more reconstruction levels; these reconstruction levels are reassigned based on one or more reference pixels in the same row.

[0078] Based on this, for any second pixel other than all reference pixels, all reference pixels in the same row as the second pixel are marked as target pixels of the second pixel. For all reconstructed sub-levels obtained for the second pixel based on all target pixels, all reconstructed sub-levels are fused using the correction weights of all target pixels to form the final level of the second pixel. The final second phase shift of the second pixel is determined based on the final level of the second pixel and the wrapping phase.

[0079] It should be noted that the determination of the order of the reference pixel by the wrapping phase and the final second phase shift, and the determination of the final second phase shift of the second pixel based on the final order and the wrapping phase, are both well-known techniques. The specific formulas are recorded in step S101. It is only necessary to substitute the variables and perform the calculations in the formulas in step S101. This example will not elaborate on the details.

[0080] In this example, by adjusting the weights for secondary reconstruction fusion, on the one hand, the larger the adjusted weight of the target pixel, the more attention is paid to the secondary reconstruction corresponding to the target pixel, reducing the interference of the second point cloud, which cannot match well with the first point cloud, on the final level. For example, it reduces the interference in the following situation: the error of the second phase shift obtained in step S102 causes the second point cloud to not match well with the first point cloud (that is, when the matching error is large), and the error of the second phase shift is excessively introduced into the final level, thus causing significant interference to the final level. On the other hand, the level obtained in step S102 has serious cumulative error. This example assigns multiple secondary reconstructions to the same second pixel. These secondary reconstructions are also obtained by spatial unfolding from different starting points (that is, the target pixel). When multiple secondary reconstructions are fused, they can offset a certain degree of cumulative error. Although the final level obtained cannot truly describe the level of the stripe where the second pixel is located (that is, it may not be an integer), it can significantly reduce the magnitude of the cumulative error.

[0081] In summary, by adjusting the weights for secondary fusion reconstruction, the accuracy of the final second phase shift of the second pixel beyond all reference pixels is ensured.

[0082] Specifically, when some rows have no reference pixels, the second phase shift of the second pixel in these rows cannot be corrected; therefore, the final point cloud corresponding to the second pixel in these rows is not acquired subsequently. The missing final point cloud can be compensated for by acquiring point clouds from other perspectives later.

[0083] As a preferred example, all reconstructed secondary levels are fused using the corrected weights of all target pixels to form the final level for the second pixel. The methods include: For any second pixel other than all reference pixels and all target pixels of that second pixel, calculate the absolute value of the difference between the x-coordinate of each target pixel and the first pixel, and denote it as the expansion distance of each target pixel. Based on the expansion distance of each target pixel and the correction weight of each target pixel, determine the fusion weight of each target pixel. The fusion weight is positively correlated with the correction weight and negatively correlated with the expansion distance.

[0084] For the second pixel, based on all the reconstructed secondary levels obtained from all the target pixels, the weighted sum of all the reconstructed secondary levels is calculated using the fusion weights of all the reference pixels, and the weighted sum is used as the final level of the second pixel.

[0085] Specifically, when the second pixel has only one reconstruction sub-level (for example, when the second pixel has only one reference pixel or target pixel in the same row), its final level is equal to its reconstruction sub-level.

[0086] The greater the extension distance, the farther the second pixel is from the target pixel. In this case, the cumulative error of the reconstruction level assigned to the second pixel based on the target pixel may be greater. At this time, the smaller the fusion weight, the less attention the reconstruction level with a large cumulative error will receive.

[0087] As a preferred example, the fusion weight of each target pixel is determined based on the extended distance and the corrected weight of each target pixel, including the following methods: The normalized extended distance is obtained by comparing the extended distance of each target pixel with the number of second pixels in each row (i.e., the width of the second image). This process removes the dimensions by calculating the ratio.

[0088] The difference between the corrected weight of each target pixel and the normalized extended distance is recorded as the first difference of each target pixel. The first differences of all target pixels of the second pixel are normalized using the softmax formula to obtain the fusion weight of each target pixel.

[0089] As a preferred example, the final point cloud obtained by the corrected second phase shift is merged with the first point cloud to obtain the reference point cloud of the surface, including the following method: The final point cloud is obtained based on the corrected second phase shift of all second pixels (the calculation formula has been recorded in step S101). The final point cloud is spatially transformed using a point cloud registration algorithm so that the final point cloud and the first point cloud are aligned and combined to obtain a reference point cloud.

[0090] After obtaining the reference point cloud, abnormal point clouds are removed from the reference point cloud to avoid interference from abnormal point clouds caused by noise; for example, the LOF algorithm is used to remove abnormal points. This algorithm is a well-known technology and will not be described in detail in this embodiment; the min_samples in the LOF algorithm is set to 30 and the LOF threshold is set to 1.3.

[0091] As a preferred example, after obtaining the final point cloud, the viewing angle of the 3D structured light camera is changed again, and the third point cloud corresponding to the third pixel and the fourth point cloud corresponding to the fourth pixel are acquired from the third and fourth viewing angles respectively using the second structured light measurement method. Note that the second structured light measurement method has relatively high acquisition efficiency and low time cost because it does not require multiple projections of Gray code. Following the methods in steps S102 and S104, the third phase shift of all third pixels and the fourth phase shift of all fourth pixels are corrected, and the final point clouds under the third and fourth viewing angles are obtained respectively. The final point clouds obtained from the second, third, and fourth viewing angles are merged with the first point cloud to obtain the reference point cloud.

[0092] Since the reference point cloud is obtained from multiple perspectives, it can relatively completely describe the structural details of the processed surface. In other embodiments, point clouds can also be acquired from more perspectives using a second structured light measurement method and merged with the first point cloud, thereby further improving the accuracy and completeness of the reference point cloud.

[0093] As an example, the method for changing the viewpoint of a 3D structured light camera is as follows: A 3D structured light camera is mounted on a circular camera track, with the camera facing the machined surface of the part at each position on the track. The circular camera track is randomly divided into four equal parts, and the viewing angles of the 3D structured light camera at the four division points are respectively designated as the first, second, third, and fourth viewing angles.

[0094] Specifically, for all the reference categories obtained above, the normal vector of each first point cloud within each reference category is calculated, and the mean vector of the normal vectors of all first point clouds within each reference category is obtained. The variance of the cosine similarity between the normal vectors of all first point clouds within each reference category and the mean vector is denoted as the distribution feature of each reference category. The normal vector of each first point cloud is a well-known technique. In this embodiment, the normal vector of the plane formed by the three closest first point clouds to each first point cloud is obtained, and the normal vector of the plane formed by the three first point clouds is used as the normal vector of each first point cloud.

[0095] The mean of the distribution characteristics of all reference categories is obtained and denoted as the comprehensive distribution characteristic. When the comprehensive distribution characteristic is large (e.g., greater than 0.52), it indicates that the continuous structural regions represented by all reference categories are not very smooth and continuous, and it may be impossible to acquire enough smooth and continuous structural regions from the first viewpoint. In this case, the four equidistant points of the circular camera track are rotated and offset by a certain angle (e.g., 45° clockwise) along the circular camera track, so that the positions of the four equidistant points are updated, thereby updating the viewpoint (that is, the viewpoints of the 3D structured light camera at the four equidistant points after the position update are respectively used as the first viewpoint, second viewpoint, third viewpoint, and fourth viewpoint), and this embodiment is repeated. When the above viewpoint is updated multiple times (e.g., 5 times), and there is still a case where the comprehensive distribution characteristic is large (e.g., greater than 0.52), then only reference categories with a distribution characteristic less than or equal to 0.52 are retained to implement all subsequent steps. In particular, when the number of retained reference categories is less than half, it is determined that it is impossible to acquire enough continuous structural regions on the surface of the component. At this time, this embodiment is not terminated, and the error of the processing process is not corrected. In other examples, a 3D structured light camera with a higher resolution can be used, or the lighting conditions can be manually adjusted, and then this embodiment can be reimplemented.

[0096] As a preferred example, when machining another surface of the part, positioning error correction is performed based on a reference point cloud. The methods include: The CAD model of the component is converted into an initial point cloud. The reference point cloud is spatially transformed using a point cloud registration algorithm. The reference point cloud and the initial point cloud are aligned. For the aligned reference point cloud, the measured geometric vectors of the structural features in the reference point cloud are extracted, and the theoretical geometric vectors of the structural features in the initial point cloud are extracted. The transformation parameters when transforming the theoretical geometric vectors into the measured geometric vectors are calculated and denoted as the machining error.

[0097] When machining another surface of a component, the workpiece coordinate system preset during the machining process is corrected using machining errors. The preset workpiece coordinate system refers to the coordinate system set on the workpiece by the machining operator based on the part drawing (CAD drawing). After correcting the workpiece coordinate system, the tool position for machining the other surface of the component is located through a tool setting operation. Both the preset workpiece coordinate system and the tool setting operation are existing technologies in machine tool processing and will not be described in detail in this embodiment.

[0098] As an example, the measured geometric vectors of structural features in the reference point cloud are extracted, the theoretical geometric vectors of structural features in the initial point cloud are extracted, and the transformation parameters when transforming the theoretical geometric vectors to the measured geometric vectors are calculated, denoted as the processing error. The methods include: A cylinder in the reference point cloud is fitted using RANSAC. The central axis vector (unit vector) of the cylinder is used as the measured geometric vector, and the center point of the circle formed by fitting the top of the cylinder is used as the measured reference point for the measured geometric vector. Similarly, the cylinder is extracted from the CAD model, and the central axis vector (unit vector) of the cylinder is used as the theoretical geometric vector, and the center point of the circle formed by fitting the top of the cylinder is used as the theoretical reference point for the theoretical geometric vector.

[0099] Solve for the translation matrix Tc when all theoretical geometric vectors are translated from the theoretical reference point to the measured reference point of all measured geometric vectors, and solve for the rotation matrix Rc when all theoretical geometric vectors are rotated to all measured geometric vectors. Use Tc and Rc as machining errors.

[0100] This method is well-known and will not be described in detail in this embodiment. For example, in some examples, the SVD method is used to solve for Tc and Rc. In other examples, the stochastic gradient descent method is used to optimize the translation and rotation matrices respectively, so that the center of all theoretical reference points and the center of the measured reference points have the minimum Euclidean distance, and the resultant vector of all theoretical geometric vectors and the resultant vector of all measured geometric vectors have the minimum Euclidean distance. The optimized translation and rotation matrices are then used as Tc and Rc.

[0101] In other examples, RANSAC can be used to fit a plane in the reference point cloud, and the plane's normal vector (unit vector) can be used as the measured geometric vector, while the plane's normal vector (unit vector) in the CAD model can be used as the theoretical geometric vector. The keypoints of the reference point cloud and the initial point cloud (obtained by the FPFH algorithm) are used as the measured reference point and the theoretical reference point, respectively.

[0102] As an example, methods for correcting the workpiece coordinate system using machining errors include: The origin of the workpiece coordinate system is corrected by translation, and the workpiece coordinate system is corrected by rotation. The correction values ​​(or transformation parameters) are Tc and Rc.

[0103] As an example, the methods for converting a component CAD model into an initial point cloud include: Export the CAD model as a triangular mesh. Divide the plane containing each triangular mesh into rectangular meshes of a preset size (e.g., 0.02 mm × 0.02 mm). The center points of all rectangular meshes within each triangular mesh are treated as a point cloud (point clouds outside the triangular meshes are deleted). The vertices of each triangular mesh are also treated as point clouds. The point clouds corresponding to all triangular meshes on the CAD model are recorded as the initial point cloud, with only one overlapping point cloud retained.

[0104] Another embodiment of the present invention provides a high-precision positioning device for machining mechanical bridge box components. The device includes a CNC machine tool, a 3D structured light camera, and a circular camera track. The CNC machine tool is used to machine the original workpiece into mechanical bridge box components. The 3D structured light camera is used to acquire or reconstruct point cloud data of the component surface during machining. The 3D structured light camera is slidably linked to the circular camera track via a linear motor (or track motor). By controlling the linear motor (or track motor), the 3D structured light camera slides on the circular camera track. When the 3D structured light camera slides on the circular camera track, its viewing angle is always directed towards and captures component information. The device also includes a ring-shaped LED light source for illuminating the components.

[0105] The device also includes a positioning error correction module, which includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor runs the computer program, it implements all the steps of the above embodiments.

[0106] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A high-precision positioning method for machining mechanical bridge box parts, characterized in that, The method includes the following steps: After each surface of a component is processed, a first point cloud of the surface is acquired from a first viewpoint using a first structured light measurement method, and several structurally continuous regions in the first point cloud are obtained. A second point cloud of the surface is acquired from a second viewpoint using a second structured light measurement method. In the first structured light measurement method, a phase-shift Gray code encoding algorithm is used to obtain the first phase shift of each first pixel, and in the second structured light measurement method, a phase shift algorithm is used to obtain the second phase shift of each second pixel. The first point cloud is obtained by the first phase shift, and the second point cloud is obtained by the second phase shift. For the first point cloud that matches the second point cloud and is located within the structural continuous region, and the matching error during matching, the second phase shift of all second pixels is corrected using the matched first point cloud and the matching error. The final point cloud obtained after the correction of the second phase shift is merged with the first point cloud to obtain the reference point cloud of the surface. When processing another surface of the component, the positioning error is corrected based on the reference point cloud.

2. The high-precision positioning method for machining mechanical bridge box components according to claim 1, characterized in that, The specific steps for obtaining several structurally continuous regions in the first point cloud are as follows: Cluster all the first point clouds to obtain all categories. According to the order of the number of first point clouds contained, the categories with the highest number of first point clouds contained are recorded as reference categories. The first point clouds in each reference category are regarded as structurally continuous regions.

3. The high-precision positioning method for machining mechanical bridge box components according to claim 1, characterized in that, The specific steps for obtaining the first point cloud, which matches the second point cloud and is located within a structurally continuous region, are as follows: A point cloud registration algorithm is used to spatially transform the first point cloud so that the first point cloud and the second point cloud are aligned. The first point cloud in all structurally continuous regions is denoted as a continuous point cloud after spatial transformation. The second point cloud is matched with the continuous point cloud to minimize the matching distance of all matching pairs. The matching distance refers to the Euclidean distance between the second point cloud and the continuous point cloud in each matching pair. The first point cloud corresponding to the continuous point cloud in any matching pair is regarded as the first point cloud that is matched with the second point cloud and is located in the structurally continuous region.

4. The high-precision positioning method for machining mechanical bridge box components according to claim 3, characterized in that, The method for calculating the matching error includes: Calculate the first mean of the matching distances of all matching pairs, and use the ratio of the matching distance of any matching pair to the first mean as the matching error of the first and second point clouds of any matching pair.

5. The high-precision positioning method for machining mechanical bridge box components according to claim 1, characterized in that, The specific steps for correcting the second phase shift of all second pixels using the matched first point cloud and the matching error are as follows: For any second point cloud, if there exists a matching first point cloud within a structurally continuous region, then the second point cloud is denoted as the target point cloud, and the second pixel corresponding to the target point cloud is denoted as the reference pixel. For the first point cloud that matches the target point cloud, the point cloud coordinates after the first point cloud is transformed and aligned to the target point cloud are used as the new point cloud coordinates of the target point cloud. Based on the new point cloud coordinates, the final second phase shift of the reference pixel is solved. The correction weight of the reference pixel is determined by the matching error when matching the target point cloud with the matching first point cloud. The correction weight is negatively correlated with the matching error. The final second phase shift of the second pixel other than all reference pixels is determined based on the final second phase shift of all reference pixels and the correction weight. The final second phase shift of all second pixels is the result of correcting the second phase shift of all second pixels.

6. The high-precision positioning method for machining mechanical bridge box components according to claim 5, characterized in that, The specific steps for determining the final second phase shift of the second pixel point other than all reference pixels based on the final second phase shift of all reference pixels and the corrected weights are as follows: For any reference pixel, the order of the reference pixel is determined based on the final second phase shift of the reference pixel. Starting from the reference pixel, traverse the second pixel in the same row that is outside the reference pixel, and assign a level to the traversed second pixel according to the level of the reference pixel. All second pixels in the same row, except for the reference pixel, are assigned a level that is the reconstructed secondary of the reference pixel. For any second pixel other than all reference pixels, mark all reference pixels in the same row as the second pixel as the target pixels of the second pixel; for all reconstruction sub-levels obtained by the second pixel based on all target pixels, use the correction weights of all target pixels to fuse all reconstruction sub-levels as the final level of the second pixel, and determine the final second phase shift of the second pixel based on the final level of the second pixel.

7. The high-precision positioning method for machining mechanical bridge box components according to claim 6, characterized in that, The specific steps involved in fusing all reconstructed secondary components are as follows: For any second pixel other than all reference pixels and all target pixels of that second pixel, calculate the absolute value of the difference between the x-coordinate of each target pixel and the second pixel, and denote it as the expansion distance of each target pixel. Based on the expansion distance of each target pixel and the correction weight of each target pixel, determine the fusion weight of each target pixel. The fusion weight is positively correlated with the correction weight and negatively correlated with the expansion distance. For the second pixel, all the reconstructed sub-levels obtained based on all the target pixels are weighted and summed using the fusion weights of all the target pixels. The weighted sum is used as the final level of the second pixel.

8. The high-precision positioning method for machining mechanical bridge box components according to claim 1, characterized in that, When machining another surface of the component, the positioning error correction based on the reference point cloud includes the following specific steps: The CAD model of the component is converted into an initial point cloud. A point cloud registration algorithm is used to spatially transform the reference point cloud to align it with the initial point cloud. For the aligned reference point cloud, the measured geometric vectors of the structural features in the reference point cloud are extracted, and the theoretical geometric vectors of the structural features in the initial point cloud are extracted. The transformation parameters when transforming the theoretical geometric vectors into the measured geometric vectors are calculated and denoted as the machining error. When machining another surface of the component, the machining error is used to correct the workpiece coordinate system preset during the machining process. The measured or theoretical geometric vector of the structural feature refers to the vector of the central axis of the cylinder in the reference point cloud or initial point cloud.

9. The high-precision positioning method for machining mechanical bridge box components according to claim 6, characterized in that, The specific steps involved in assigning a level to the traversed second pixel based on the level of the reference pixel are as follows: Starting from the reference pixel, traverse to the right or left to the second pixel in the same row that is outside the reference pixel. Calculate the absolute value of the difference between the wrapping phase of the currently traversed second pixel and the previously traversed second pixel. When the absolute value of the difference is greater than π, the order of the currently traversed second pixel is equal to the order of the previously traversed second pixel plus or minus 1. When the absolute value of the difference is less than or equal to π, the level of the second pixel point currently traversed is equal to the level of the second pixel point traversed in the previous traversal; wherein, when the second pixel point currently traversed is an adjacent pixel point of the reference pixel point, the second pixel point traversed in the previous traversal is the reference pixel point.

10. A high-precision positioning device for machining mechanical bridge box components, the device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor runs the computer program, it implements all the steps of the high-precision positioning method for machining mechanical bridge box parts as described in any one of claims 1 to 9.

Images