Engine pipeline measuring method based on laser scanning

Abstract

The invention provides an engine pipeline measurement method based on laser scanning, which belongs to a non-contact measurement technology, and comprises the following steps: before measurement, selecting a three-dimensional laser scanner, configuring and calibrating the three-dimensional laser scanner, carrying out a scanning test, adjusting a scanning environment, and then obtaining pipeline point cloud data; the outlier error of the pipeline point cloud data is denoised by using a KD-tree (K-dimension tree), and the outlier error of the pipeline point cloud data is denoised by using the KD-tree (K-dimension tree); de-noising system errors and random errors of the pipeline point cloud data by using a Gaussian filter; segmenting the point cloud data of the pipeline, and identifying and isolating all parts (linear segments, elbows and joints) of the pipeline; performing reverse modeling on the pipeline point cloud data to obtain a digital model of the pipeline; and comparing and analyzing the original point cloud of the pipeline and the pipeline model, and evaluating the error of pipeline measurement. According to the invention, the feature extraction and processing analysis of the pipeline point cloud are realized, the time cost of engine pipeline assembly is reduced, and the wear of the pipeline is reduced.

Description

Technical Field

[0001] The invention belongs to the technical field of non-contact measurement, and in particular relates to an engine pipeline measurement method based on laser scanning. Background Art

[0002] In the current process of assembling the external pipelines of engines, there is a technical defect of lack of pipeline measurement before assembly, which makes it impossible to effectively predict and evaluate the performance after assembly. This technical limitation leads to a lack of systematic guidance in the assembly process, forcing technicians to perform multiple trial assembly during the assembly process. This practice not only increases the time cost of assembly, but also causes unnecessary wear on the pipeline joints. In order to solve this problem, it is necessary to introduce a more efficient non-contact measurement technology that will not scratch the parts. Before assembly, the pipeline of the engine is scanned to obtain the point cloud data of the pipeline, and the data is used for reverse modeling to construct a measured model of the engine pipeline. By comparing and analyzing this model with the theoretical design model, the assembly feature deviation of the engine pipeline can be obtained. More importantly, with the assistance of subsequent in-depth research, the assembly feature deviation data of the engine pipeline can be used to predict the sealing performance of the pipeline after assembly. In order to achieve this goal, it is particularly important to introduce a suitable non-contact measurement technology.

[0003] Pipes are parts with large manufacturing errors, and the ball-cone structure of pipe joints has a certain tolerance, so the measurement accuracy required for pipes is relatively low. At the same time, there are many pipes with different shapes, so measurement efficiency is particularly important in the selection of measurement methods. 3D laser scanning, as a method with slightly lower accuracy but efficient point cloud acquisition, is suitable for the above scenarios. Summary of the invention

[0004] In view of the above situation, the present invention provides an engine pipeline measurement method based on laser scanning, which standardizes the measurement environment and measurement process, as well as the reverse modeling process of pipeline point cloud data, to achieve rapid extraction of pipeline assembly features from pipeline point cloud data.

[0005] In order to achieve one or more of the above purposes, the technical solution adopted by the present invention is:

[0006] Step S1: Before measuring, select a 3D laser scanner, configure and calibrate it, perform a scanning test, adjust the scanning environment, and then obtain pipeline point cloud data;

[0007] Step S2: Use KD-tree (K-dimensional tree) to denoise the outlier errors of the pipeline point cloud data;

[0008] Step S3: using a Gaussian filter to remove the systematic error and random error of the pipeline point cloud data;

[0009] Step S4: segmenting the pipeline point cloud data to identify and isolate the various parts of the pipeline (straight segments, elbows, joints);

[0010] Step S5: reverse modeling is performed on the pipeline point cloud data to obtain a digital model of the pipeline;

[0011] Step S6: Compare and analyze the original point cloud of the pipeline and the pipeline model to evaluate the error of pipeline measurement.

[0012] In one embodiment, the step S1 comprises the following steps:

[0013] Step S101: before measuring, select a 3D laser scanner, including a tracking laser scanner and a handheld laser scanner, and configure an optical dynamic tracker;

[0014] Step S102: Install an optical tracker at the measurement position, ensure that the tracker is firmly placed on a stable surface, and its line of sight can cover the area that the 3D laser scanner can reach during the entire measurement process, and connect the 3D laser scanner to the computer, configure and calibrate it, and set the scanning speed, resolution and scanning mode;

[0015] Step S103: After the calibration is completed, a scanning test is performed to verify whether the configuration and calibration of the device are correct, and the scanning environment is adjusted;

[0016] Step S104: different scanning strategies are adopted according to the pipeline structure, and pipeline point cloud data is acquired using a three-dimensional laser scanner.

[0017] In one embodiment, step S2 comprises the following steps:

[0018] Step S201: constructing a KD-tree (K-dimensional tree) using pipeline point cloud data;

[0019] Step S202: setting the search parameters of KD-tree to search for neighboring points of the point cloud;

[0020] Step S203: Compare the local density of each point cloud with the threshold value to identify outliers. Points with local density far below the average level will be identified as outliers. ;

[0021] Step S204: removing the calibrated outlier points from the pipeline point cloud data;

[0022] Step S205: After removing outliers, post-process the pipeline point cloud data to improve the quality of the pipeline point cloud data.

[0023] The beneficial effect of the above further scheme is: for outlier errors caused by factors such as the chaotic measurement environment, the reflective properties of the surface of the pipeline material, the obstruction of the light path, or the point cloud data of other objects such as desktops or fixtures that are inevitably collected during the scanning process, a KD-tree-based clustering algorithm is used to eliminate such outliers. KD-tree can efficiently query the neighboring points of data points. The KD-tree algorithm is used to search for the neighboring points of a point, calculate the distance between the points, and set a reasonable threshold to remove the neighboring points whose distance is greater than the threshold. Iterating this process can achieve the purpose of removing most of the outliers, thereby improving the quality of the pipeline point cloud data.

[0024] In one embodiment, step S3 comprises the following steps:

[0025] Step S301: setting the Gaussian kernel size and standard deviation of the Gaussian filter. The size of the Gaussian kernel determines the number of neighboring points considered in the filtering process, and the standard deviation of the Gaussian filter determines the shape of the weight distribution.

[0026] Step S302: Calculate Gaussian weights. For each point in the Gaussian kernel, calculate the Gaussian weight according to its distance from the center point. The farther the point is from the center point, the smaller its weight is.

[0027] Step S303: weighting the neighboring points of each point cloud using Gaussian weights, and weighted averaging the coordinates of the neighboring points using Gaussian weights, thereby updating the position of each point;

[0028] Step S304: repeatedly applying a Gaussian filter to the pipeline point cloud data to remove higher-order noise;

[0029] Step S305: After Gaussian filtering, the pipeline point cloud data is post-processed to recalculate normals and other joint features.

[0030] The beneficial effect of the above further scheme is that Gaussian filtering uses Gaussian functions as weights to perform weighted averaging on points in the point cloud, which can effectively smooth the point cloud data and reduce random noise in the data while retaining important structural information.

[0031] In one embodiment, step S4 comprises the following steps:

[0032] Step S401: extracting the pipeline axis using the pipeline point cloud data, the direction of the pipeline can be identified. First, a boundary line of the pipeline point cloud data is extracted, and a cross section is generated using the points on the boundary line and the normal vector. The cross section intersects with the point cloud to form a circle, and the center of the circle is the point on the pipeline axis.

[0033] Step S402: For each point cloud on the pipeline axis, the curvature is calculated to identify the curved part and the connection area of ​​the pipeline. The high curvature area corresponds to the elbow or connection part of the pipeline, and the low curvature area represents the straight line segment;

[0034] Step S403: Calculate the surface normal direction of each point in the pipeline point cloud. By using principal component analysis in the local neighborhood of each point, the surface normal can be estimated. The change in the normal direction indicates the change in the geometric shape of the pipeline.

[0035] Step S404: clustering point clouds with similar features into continuous regions, starting from selecting a seed point, and then gradually adding neighboring point clouds with similar features (such as similar curvature, normal direction) into the same region;

[0036] Step S405: performing point cloud clustering according to the geometric features of the pipeline point cloud data, dividing the pipeline point cloud data into different geometric shape groups, each group representing a specific part of the pipeline;

[0037] Step S406: fitting the pipeline point cloud data to a geometric model through a fitting algorithm, and performing model segmentation according to the geometric model of the pipeline point cloud;

[0038] Step S407: Merge regions with similar geometric features to form a more complete and continuous pipeline section.

[0039] In one embodiment, step S5 comprises the following steps:

[0040] Step S501: creating a parametric model of the pipeline, formulating models for different parts of the pipeline, and these models need to have adjustable parameters, which enable the models to adapt to the size and shape of the actual pipeline;

[0041] Step S502: using the segmented pipeline point cloud data and the parameterized model to solve the specific parameters of each model so that it is as consistent as possible with the actual point cloud data;

[0042] Step S503: converting the point cloud data of the pipeline into a segmented model, and for each part identified by feature segmentation, using the corresponding parameterized model to replace the original point cloud data;

[0043] Step S504: All segmented models are combined and smoothed to complete the reverse modeling of the pipeline point cloud data.

[0044] In one embodiment, step S6 comprises the following steps:

[0045] Step S601: Use Geomagic Design X software to perform accuracy analysis on the reconstructed pipeline model. The software can provide detailed error measurement and visualization, which helps to identify and quantify errors. The software can be used to solve the distance from the initial point cloud data to the reconstructed model.

[0046] Step S602: Create a deviation cloud map between the reconstructed pipeline model and the original point cloud data. Different colors are assigned to the point cloud data according to the distance from the initial point cloud data to the reconstructed model. This chart can intuitively display the size and distribution of the error.

[0047] Step S603: Compare and analyze the fitting accuracy between the reconstructed pipeline model and the original point cloud data, and evaluate the error of pipeline measurement.

[0048] The beneficial effects of the present invention include at least one of the following:

[0049] 1) The present invention proposes a process specification for non-contact measurement technology in the process of acquiring three-dimensional point clouds of pipeline parts.

[0050] 2) The present invention summarizes a set of processing methods suitable for engine pipeline data point cloud, realizing feature extraction and processing analysis of pipeline point cloud.

[0051] 3) The present invention measures the engine pipeline by a three-dimensional laser scanner, which can not only reduce the time cost of assembling the engine pipeline, but also reduce the wear on the pipeline. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] Figure 1 is a flow chart of the method of the present invention;

[0053] Figure 2 Schematic diagram of scanning measurement process;

[0054] Figure 3 This is a diagram of the pipeline 3D laser scanning process;

[0055] Figure 4 Comparison chart before and after point cloud denoising;

[0056] Figure 5 This is a schematic diagram of the point cloud data segmentation results;

[0057] Figure 6 This is a schematic diagram of the reverse modeling results;

[0058] Figure 7 This is a three-dimensional model structure diagram of the pipeline. DETAILED DESCRIPTION

[0059] The specific implementation modes of the present invention are described below so that those skilled in the art can understand the present invention. However, it should be clear that the present invention is not limited to the scope of the specific implementation modes. For those of ordinary skill in the art, as long as various changes are within the spirit and scope of the present invention as defined and determined by the attached claims, these changes are obvious, and all inventions and creations utilizing the concept of the present invention are protected.

[0060] The present invention proposes an engine pipeline measurement method based on laser scanning, which can realize the process specification of non-contact measurement technology in the process of acquiring three-dimensional point clouds of pipeline parts, and summarizes a set of processing methods suitable for engine pipeline point cloud data, realizing the feature extraction and processing analysis of pipeline point clouds, which can not only reduce the time cost of engine pipeline assembly, but also reduce the wear on the pipeline.

[0061] The engine pipeline measurement method based on laser scanning proposed in the present invention comprises the following steps:

[0062] Step S1: Before measuring, select a 3D laser scanner, configure and calibrate it, perform a scanning test, adjust the scanning environment, and then obtain the pipeline point cloud data. Preferably, the specific implementation method of this step is as follows:

[0063] Step S101: before measuring, select a 3D laser scanner, including a tracking laser scanner and a handheld laser scanner, and configure an optical dynamic tracker;

[0064] Step S102: Install an optical tracker at the measurement position, ensure that the tracker is firmly placed on a stable surface, and its line of sight can cover the area that the 3D laser scanner can reach during the entire measurement process, and connect the 3D laser scanner to the computer, configure and calibrate it, and set the scanning speed, resolution and scanning mode;

[0065] Step S103: After the calibration is completed, a scanning test is performed to verify whether the configuration and calibration of the device are correct, and the scanning environment is adjusted;

[0066] Step S104: different scanning strategies are adopted according to the pipeline structure, and pipeline point cloud data is acquired using a three-dimensional laser scanner.

[0067] In this embodiment, the reverse modeling result schematic diagram is as follows: Figure 6 As shown, adjust the scanner's alignment and focus, and complete the registration through the automatic registration module in the equipped software. The external structure of the pipeline is shown in Figure 7As shown in the figure, the external structure of the pipeline can be mainly divided into four types: ball joint, cone joint, conduit and outer nut. These four parts have their own structural characteristics and different focuses during the measurement process, requiring different scanning strategies.

[0068] The spherical and conical joints are the most critical feature measurement areas, which will eventually reflect the radial deviation and angular deviation of the pipe joints. Relying solely on the overall contour scan is not enough to capture all the details, so it is necessary to cooperate with the appropriate scanning posture and lighting conditions to improve the scanning accuracy of the inner surface of the joint. Because after scanning the external contours of the ball head and the cone surface, the pipe joint boundary will be extracted based on the point cloud and the end face shape, centroid and normal vector will be fitted. The accuracy of the point cloud will directly affect the accuracy of the feature extraction analysis, so the accuracy of the scanned point cloud is required to be high.

[0069] The point cloud data collection work of the catheter part is mainly focused on the contour of the pipeline. The center line will be solved based on the contour to have an overall control of the direction of the pipeline. The reconstruction of the pipeline contour is to obtain the length and direction of the pipeline on the one hand, which is conducive to the evaluation of the tolerance and deformation margin of the pipeline; on the other hand, it is to ensure the continuity of the reverse modeling of the pipeline structure and does not affect the final evaluation of the pipeline assembly error. Therefore, the measurement accuracy requirement is relatively low. As long as enough points are obtained through multiple scans, the pipeline model can be reconstructed through denoising, segmentation, fitting and other methods. Therefore, the scanning of this part can take relatively less time.

[0070] Finally, the outer nut part is a non-critical feature that can be discarded in pipeline modeling and analysis. However, since the outer nut is an accessory that is placed on the pipeline and has a non-fixed position, the relative position of the nut will change when the pipeline posture is changed, which will make the registration of the pipeline point cloud data difficult. In addition, the nut will block the scanning and acquisition of part of the point cloud data. Therefore, in the process of 3D point cloud acquisition, the outer nut needs to be moved to the straight section of the catheter and fixed with non-reflective tape. This operation strategy not only improves the registration efficiency of scanning pipeline point clouds at different angles, but also ensures the effective capture of key features in the pipeline, and improves the efficiency and accuracy of the point cloud data acquisition process.

[0071] Step S2: Use KD-tree (K-dimensional tree) to denoise the outlier errors of the pipeline point cloud data. Preferably, the specific implementation method of this step is as follows:

[0072] Step S201: constructing a KD-tree (K-dimensional tree) using pipeline point cloud data;

[0073] Step S202: setting the search parameters of KD-tree to search for neighboring points of the point cloud;

[0074] Step S203: Compare the local density of each point cloud with the threshold value to identify outliers. Points with local density far below the average level will be identified as outliers. ;

[0075] Step S204: removing the calibrated outlier points from the pipeline point cloud data;

[0076] Step S205: After removing outliers, post-process the pipeline point cloud data to improve the quality of the pipeline point cloud data.

[0077] Step S3: Use a Gaussian filter to remove the systematic error and random error of the pipeline point cloud data. Preferably, the specific implementation method of this step is as follows:

[0078] Step S301: setting the Gaussian kernel size and standard deviation of the Gaussian filter. The size of the Gaussian kernel determines the number of neighboring points considered in the filtering process, and the standard deviation of the Gaussian filter determines the shape of the weight distribution.

[0079] Step S302: Calculate Gaussian weights. For each point in the Gaussian kernel, calculate the Gaussian weight according to its distance from the center point. The farther the point is from the center point, the smaller its weight is.

[0080] Step S303: weighting the neighboring points of each point cloud using Gaussian weights, and weighted averaging the coordinates of the neighboring points using Gaussian weights, thereby updating the position of each point;

[0081] Step S304: repeatedly applying a Gaussian filter to the pipeline point cloud data to remove higher-order noise;

[0082] Step S305: After Gaussian filtering, the pipeline point cloud data is post-processed to recalculate normals and other joint features.

[0083] Step S4: segment the pipeline point cloud data, identify and isolate the various parts of the pipeline (straight segments, elbows, joints). Preferably, the specific implementation method of this step is as follows:

[0084] Step S401: extracting the pipeline axis using the pipeline point cloud data, the direction of the pipeline can be identified. First, a boundary line of the pipeline point cloud data is extracted, and a cross section is generated using the points on the boundary line and the normal vector. The cross section intersects with the point cloud to form a circle, and the center of the circle is the point on the pipeline axis.

[0085] Step S402: For each point cloud on the pipeline axis, the curvature is calculated to identify the curved part and the connection area of ​​the pipeline. The high curvature area corresponds to the elbow or connection part of the pipeline, and the low curvature area represents the straight line segment;

[0086] Step S403: Calculate the surface normal direction of each point in the pipeline point cloud. By using principal component analysis in the local neighborhood of each point, the surface normal can be estimated. The change in the normal direction indicates the change in the geometric shape of the pipeline.

[0087] Step S404: clustering point clouds with similar features into continuous regions, starting from selecting a seed point, and then gradually adding neighboring point clouds with similar features (such as similar curvature, normal direction) into the same region;

[0088] Step S405: performing point cloud clustering according to the geometric features of the pipeline point cloud data, dividing the pipeline point cloud data into different geometric shape groups, each group representing a specific part of the pipeline;

[0089] Step S406: fitting the pipeline point cloud data to a geometric model through a fitting algorithm, and performing model segmentation according to the geometric model of the pipeline point cloud;

[0090] Step S407: Merge regions with similar geometric features to form a more complete and continuous pipeline section.

[0091] Step S5: reverse modeling of the pipeline point cloud data to obtain a digital model of the pipeline. Preferably, the specific implementation method of this step is as follows:

[0092] Step S501: creating a parametric model of the pipeline, formulating models for different parts of the pipeline, and these models need to have adjustable parameters, which enable the models to adapt to the size and shape of the actual pipeline;

[0093] Step S502: using the segmented pipeline point cloud data and the parameterized model to solve the specific parameters of each model so that it is as consistent as possible with the actual point cloud data;

[0094] Step S503: converting the point cloud data of the pipeline into a segmented model, and for each part identified by feature segmentation, using the corresponding parameterized model to replace the original point cloud data;

[0095] Step S504: All segmented models are combined and smoothed to complete the reverse modeling of the pipeline point cloud data.

[0096] Step S6: Compare and analyze the original point cloud of the pipeline and the pipeline model to evaluate the error of pipeline measurement.

[0097] Preferably, the specific implementation method of this step is as follows:

[0098] Step S601: Use Geomagic Design X software to perform accuracy analysis on the reconstructed pipeline model. The software can provide detailed error measurement and visualization, which helps to identify and quantify errors. The software can be used to solve the distance from the initial point cloud data to the reconstructed model.

[0099] Step S602: Create a deviation cloud map between the reconstructed pipeline model and the original point cloud data. Different colors are assigned to the point cloud data according to the distance from the initial point cloud data to the reconstructed model. This chart can intuitively display the size and distribution of the error.

[0100] Step S603: Compare and analyze the fitting accuracy between the reconstructed pipeline model and the original point cloud data, and evaluate the error of pipeline measurement.

[0101] In one embodiment, Figure 2 Mount the tracker at the intended measurement location as shown in the figure. Make sure the tracker is firmly placed on a stable surface and its line of sight covers the area that the scanner can reach during the entire measurement process.

[0102] Connection and configuration: Connect the scanner to the computer and configure the scanner settings, including scanning speed, resolution, and scanning mode, through the corresponding software.

[0103] Perform equipment calibration to ensure the accuracy of the scanned data. Place the calibration plate on a stable and flat surface, check the calibration plate of the scanner to ensure there are no reflective objects near the calibration plate, adjust the scanner's alignment and focus, etc., and complete the registration through the automatic registration module in the equipped software.

[0104] Before the actual measurement, a test scan is performed to verify that the equipment is configured and calibrated correctly. Depending on the scanning environment during the test process, some adjustments may be required, such as ensuring appropriate lighting, reducing factors in the environment that may interfere with the scan, and designing the fixture required for the scanning measurement process to ensure that the pipeline measurement process is unobstructed.

[0105] Because pipelines have some common characteristics, different feature types have different operating points during the three-dimensional scanning and measurement of pipelines. Timely control of the operating techniques of the scanning equipment can control the details of the scanning, which can provide a good foundation for subsequent point cloud processing.

[0106] like Figure 3 As shown in the figure, during the laser 3D scanning measurement of the pipeline system, the external structure of the pipeline can be mainly divided into four types: ball joint, cone joint, conduit and outer nut. These four parts have their own structural characteristics and different focuses during the measurement process, requiring different scanning strategies.

[0107] First of all, the spherical and conical joint parts are the most critical feature measurement areas, which ultimately reflect the radial deviation and angular deviation of the pipe joint. Relying solely on the overall contour scanning is not enough to capture all the details, so it is necessary to cooperate with appropriate scanning posture and lighting conditions to improve the scanning accuracy of the inner surface of the joint. Because after scanning the external contours of the ball head and the cone surface, the pipe joint boundary will be extracted based on the point cloud and the end face shape, centroid and normal vector will be fitted. The accuracy of the point cloud will directly affect the accuracy of feature extraction and analysis, so the accuracy of the scanned point cloud is required to be high.

[0108] Secondly, it is the point cloud data collection of the catheter part. In most cases, the shape of the catheter is a swept body of a constant diameter circle along the center line. Although the bends will be slightly deformed due to the bending process of the pipeline, the shape characteristics are almost unchanged. When the pipeline branches, special treatment is required, but it can still be abstracted as the Boolean sum of two cylinders. Therefore, the point cloud data collection work of the catheter part is mainly focused on the contour of the pipeline. The center line will be solved based on the contour to have an overall control of the direction of the pipeline. On the one hand, the reconstruction of the pipeline contour is to obtain the length and direction of the pipeline, which is conducive to the evaluation of the tolerance capacity and deformation margin of the pipeline; on the other hand, it is to ensure the continuity of the reverse modeling of the pipeline structure, and does not affect the final evaluation of the pipeline assembly error. Therefore, the measurement accuracy requirement is relatively low. As long as enough points are obtained through multiple scans, the pipeline model can be reconstructed through various methods such as denoising, segmentation, and fitting. Therefore, the scanning of this part can take relatively less time.

[0109] Finally, the outer nut part is a non-critical feature that can be discarded in pipeline modeling and analysis. However, since the outer nut is an accessory that is mounted on the pipeline and has a non-fixed position, the relative position of the nut changes when the pipeline posture is changed, which will make the registration of the pipeline point cloud data difficult. In addition, the nut will block part of the point cloud data scanning and acquisition. Therefore, in the process of 3D point cloud acquisition, the outer nut needs to be moved to the straight section of the catheter and fixed with non-reflective tape. This operation strategy not only improves the registration efficiency of scanning pipeline point clouds at different angles, but also ensures the effective capture of key features in the pipeline, and improves the efficiency and accuracy of the point cloud data acquisition process.

[0110] like Figure 4 As shown in the figure, after 3D laser scanning, a relatively rough point cloud data of the pipeline is obtained. In the 3D laser scanning process, there are many sources of error such as pipeline reflection, light path occlusion, environmental influence, equipment error, calibration error, etc. There are a lot of noise points in the point cloud data, which seriously affects the subsequent reverse modeling process. Therefore, point cloud data denoising is a key step before reverse modeling.

[0111] The errors in pipeline point cloud data can generally be divided into two categories: one is outlier errors, which are usually caused by factors such as the chaotic measurement environment, the reflective properties of the pipeline material surface, or the obstruction of the optical path. These errors appear as outliers, which are usually scattered and disordered. In data processing, this type of error is generally solved by eliminating outliers. This involves carefully examining the data and determining which points should be considered outliers and excluded based on certain criteria (such as the distance between points, deviation from the model, etc.). The other is systematic errors and random errors: This type of error may be caused by optical path deviations caused by environmental factors (such as temperature and humidity), or introduced by resolution limitations of the measurement equipment and inaccurate calibration processes. This error is relatively evenly distributed throughout the data set. When dealing with this type of error, the point cloud is usually fitted into a vector model. This means that the actual measured data points are adjusted to fit a predefined mathematical model, which is usually established based on the theoretical model of the pipeline. Through this fitting, the impact of errors caused by the above factors on the measurement results can be reduced.

[0112] Data denoising methods are used differently for different sources of error. For outlier errors caused by factors such as the chaotic measurement environment, the reflective properties of the surface of the pipe material, the obstruction of the light path, or the point cloud data of other objects such as desktops or fixtures that are inevitably collected during the scanning process, a KD-tree-based clustering algorithm is used to remove such outliers. KD-tree (K-dimensional tree) is a data structure used to organize spatial data. It can efficiently query the neighboring points of data points. The KD-tree algorithm is used to search for the neighboring points of a point, calculate the distance between the points, and set a reasonable threshold to remove the neighboring points whose distance is greater than the threshold. Iterating this process can achieve the purpose of removing most of the outliers. The following are the general steps for removing outliers using a KD-tree-based clustering algorithm:

[0113] First, the point cloud data needs to be constructed into a KD-tree. This step can speed up the subsequent search for the nearest neighbors of each point.

[0114] Determines the parameters needed when searching for neighboring points, such as the search radius or the number of neighboring points. These parameters will be used to determine the local density of each point in the point cloud.

[0115] For each point in the point cloud, a KD-tree is used to quickly find its neighbor points within a specified radius, or to find the nearest N neighbor points. Then, the local density of each point is calculated based on the number or distribution of neighbors.

[0116] Outliers are identified by comparing the local density of each point with a preset threshold. Typically, points with local densities far below the average are identified as outliers. Data identified as outliers are removed from the point cloud.

[0117] After removing outliers, you can choose to perform post-processing steps such as smoothing or further data analysis to improve the quality of the point cloud data.

[0118] At the same time, the following points should be noted: Correct selection of search parameters is crucial to successfully remove outliers. Improper parameter selection may cause valid data to be mistakenly deleted or outliers to be effectively identified. Different point cloud data may require different processing strategies. Therefore, for different applications, it may be necessary to adjust the parameters of the algorithm.

[0119] For the resolution limitation of the measuring equipment and the systematic error and random error introduced by the equipment calibration process, common denoising methods include: Gaussian filtering, mean filtering, median filtering, bilateral filtering, non-local mean filtering, iterative closest point (ICP) algorithm and other methods, each of which has its own characteristics and applicable scenarios. Gaussian filtering uses Gaussian function as weight to perform weighted averaging on the points in the point cloud, which can effectively smooth the data while retaining the edges and important structural features. It is suitable for most random noises, especially when it is necessary to retain edge information while denoising. Mean filtering replaces the value of each point with the average value of its neighboring points. It is simple but may cause blurred edges and details. It is suitable for data with relatively uniform noise levels. Median filtering replaces the value of each point with the median of the neighboring points. It is effective in removing impulse noise and better retains edges than mean filtering. It is suitable for data containing impulse noise. Bilateral filtering takes into account the spatial proximity and intensity differences of points. It retains edges well while smoothing noise, but the computational cost is high. It is suitable for high-quality point cloud data processing that requires accurate edge retention. The non-local mean filter considers the average value of similar areas in the entire point cloud to remove noise, which is suitable for retaining detail information and is applicable to complex scenes that require a high degree of detail preservation. The iterative closest point (ICP) algorithm removes mismatched points during the alignment of the point cloud with the reference model, and is applicable to point cloud data that needs to be registered with a specific model.

[0120] From the above description, it can be seen that compared with other filtering methods, Gaussian filtering is a method based on Gaussian function (normal distribution curve), which can effectively smooth point cloud data, reduce random noise in the data, and retain important structural information. Gaussian filtering embodies the characteristics of balance between efficiency and effect, wide applicability, flexible parameter adjustment, relative simplicity, and easy implementation. Therefore, for the ordered point cloud data of this study, the Gaussian filtering algorithm is used to denoise the collected point cloud data. The following are the basic steps for applying Gaussian filtering to denoise point cloud data:

[0121] Kernel size: Determines the kernel (or window) size of the Gaussian filter. The kernel size determines the number of neighboring points considered during the filtering process.

[0122] Standard Deviation: The standard deviation of the Gaussian filter determines the shape of the weight distribution. A larger standard deviation means a smoother filtering effect, but may cause blurred edges.

[0123] For each point in the kernel, a Gaussian weight is calculated based on its distance from the center point. The farther away the point is, the smaller its weight is.

[0124] For each point in the point cloud, consider its neighboring points (determined by the kernel size) and update the position of each point by taking a weighted average of the coordinates of these neighboring points using Gaussian weights.

[0125] This step will reduce the impact of noise points since they will have less influence in the weighted average.

[0126] Higher-order noise may exist in the point cloud data, requiring repeated application of Gaussian filtering.

[0127] After Gaussian filtering, some post-processing is required, such as recalculating normals and other joint features.

[0128] At the same time, the following points should be noted: For Gaussian filtering, choosing the appropriate kernel size and standard deviation can achieve the expected data denoising effect. Too large a kernel size or standard deviation may over-smooth the data, thereby losing important details. In the process of point cloud data processing, it is important to preserve the edges of the data, and the filtering parameters need to be carefully selected to avoid the loss of edge information.

[0129] like Figure 5 As shown, the pipeline is composed of multiple segments. The goal of pipeline point cloud data segmentation is to identify and isolate the various parts of the pipeline. Since the geometric characteristics of the pipeline are relatively unique and regular, this segmentation usually focuses on identifying the straight segments, elbows, joints, etc. of the pipeline. This article mainly focuses on single pipelines without branches. For pipeline systems with multiple branches, axis extraction is more complicated and requires special processing strategies, but they are all based on single pipelines without branches. The following are the general steps for this segmentation:

[0130] Axis extraction is the basis of pipeline point cloud data processing. The purpose is to identify the direction of the pipeline. First, a boundary line of the point cloud data is extracted, and a cross section is generated based on the points on the boundary line and the normal vector. The cross section intersects with the point cloud to form a circle, and the center of the circle is the point on the pipeline axis. This step is crucial to understanding the overall layout and direction of the pipeline.

[0131] The purpose of curvature analysis is to quantify the local curvature of each point on the axis, which is critical for identifying bends and connections in the pipeline. The curvature of each point can be calculated by fitting a curve within the local neighborhood of each point on the axis. Areas of high curvature usually correspond to elbows or connections in the pipeline, while areas of low curvature represent straight segments.

[0132] The normal estimation step involves calculating the direction of the surface normal for each point in the point cloud, which is extremely important for distinguishing the characteristics of the pipe. The surface normal can be estimated by using techniques such as principal component analysis (PCA) within the local neighborhood of each point. Changes in the normal direction usually indicate changes in the geometry of the pipe, such as the transition from a pipe segment to a joint.

[0133] Region growing is a method of clustering points into larger regions based on their similarity. This step starts with selecting a seed point and then gradually adding neighboring points with similar characteristics (such as similar curvature, normal direction) to the same region. In this way, points with similar characteristics can be formed into continuous regions representing different parts of the pipeline.

[0134] Geometric clustering is the process of clustering points based on their geometric properties (e.g., distance, direction). Using methods such as Euclidean clustering, point cloud data is divided into different geometric shape groups, each representing a specific part of the pipeline, such as a straight pipe section, elbow, or pipe joint.

[0135] Model segmentation divides point cloud data into specific geometric models (such as cylinders, frustums, threads, and spheres). This is usually achieved through a fitting algorithm, for example, fitting point cloud data to a cylindrical model to identify straight pipe sections. By comparing the fit between point cloud data and the geometric model, data segmentation can be performed effectively.

[0136] The final step is to merge geometrically and characteristically similar regions to form a more complete and continuous pipe section. This involves analyzing the characteristics of each region, such as size, shape, location, and orientation, and merging similar regions based on these characteristics, which helps to simplify the model and improve the overall accuracy and usability of the data.

[0137] like Figure 6 As shown in the figure, the point cloud data is segmented according to the similar features, and on this basis, point cloud reverse modeling can be carried out. The specific steps are as follows:

[0138] The process of creating parametric models is the cornerstone of reverse modeling. For piping systems, this involves developing flexible models for different parts such as straight pipe segments, curved pipe segments, threaded segments, and joint segments. These models need to have adjustable parameters such as length, diameter, and bend angle. These parameters allow the model to adapt to the size and shape of the actual pipe. For example, a straight pipe model may only require length and diameter as parameters, while a curved pipe model may also require bend radius and angle.

[0139] Once feature segmentation is complete, the next step is to match the segmented point cloud data with the corresponding parameterized models. The goal of this step is to determine the specific parameters of each model so that it matches the actual point cloud data as closely as possible. Typically, this involves using optimization algorithms, such as the least squares method, to adjust the model parameters. This process may include iterative adjustments until the model achieves the best fit with the corresponding part of the point cloud.

[0140] Converting point cloud data to segmented models is a key step in converting digital data into a specific model. For each part identified by feature segmentation, the original point cloud data is replaced with the corresponding parameterized model. For example, a specific bend segment point cloud will be converted into a bend model with a specific radius and bend angle.

[0141] All segmented models need to be combined into a complete pipeline model. This includes stitching them together according to their relative positions and orientations in the point cloud data. To ensure the continuity and practicality of the model, the stitched model may require further smoothing to eliminate discontinuities or rough edges at the seams.

[0142] After completing the above steps, you will end up with a complete and accurate pipeline model that accurately reflects all the details of the original point cloud data. This model can be used for a variety of applications, including system analysis, maintenance planning, or further design improvements. Reverse modeling not only provides an accurate digital copy of the physical object, but also provides a basis for in-depth understanding and optimization of existing systems.

[0143] During the entire reverse modeling process, record each step that may introduce errors, from acquiring point clouds, preprocessing data to the final model fitting.

[0144] Accuracy analysis was performed using Geomagic Design X software, which provides detailed error measurements and visualizations to help identify and quantify errors. The software can be used to solve the distance from the initial point cloud data to the reconstructed model.

[0145] Create and analyze the deviation cloud graph between the reconstructed pipeline model and the original point cloud data. Different colors are assigned to the point cloud data according to the distance from the initial point cloud data to the reconstructed model. This graph can intuitively display the size and distribution of the error.

[0146] Compare and analyze the fitting accuracy between the reconstructed model and the point cloud data, paying special attention to the distribution of errors and the size of errors at key features.

[0147] The errors in the nut and threaded portions of the jacket are larger, but since these are non-critical features, they can be ignored in the analysis.

[0148] The maximum error at the key features of the pipe joint part is 0.2 mm, which is within the allowable error range of this study.

[0149] Overall, the error between the catheter part and its original point cloud data is within an acceptable range, and the error of important parts meets the accuracy requirements.

Claims

1. An engine pipeline measurement method based on laser scanning, characterized in that: The following steps are involved: Step S1: Before measuring, select a 3D laser scanner, configure and calibrate it, perform a scanning test, adjust the scanning environment, and then obtain pipeline point cloud data; Step S2: Use KD-tree (K-dimensional tree) to denoise the outlier errors of the pipeline point cloud data; Step S3: using a Gaussian filter to remove the systematic error and random error of the pipeline point cloud data; Step S4: segmenting the pipeline point cloud data to identify and isolate the various parts of the pipeline (straight segments, elbows, joints); Step S5: reverse modeling is performed on the pipeline point cloud data to obtain a digital model of the pipeline; Step S6: Compare and analyze the original point cloud of the pipeline and the pipeline model to evaluate the error of pipeline measurement.

2. The method for measuring an engine pipeline based on laser scanning according to claim 1, characterized in that: The step S1 comprises the following steps: Step S101: before measuring, select a 3D laser scanner, including a tracking laser scanner and a handheld laser scanner, and configure an optical dynamic tracker; Step S102: installing an optical tracker at the measurement position and connecting the 3D laser scanner to a computer for configuration and calibration; Step S103: After the calibration is completed, a scanning test is performed and the scanning environment is adjusted; Step S104: According to the pipeline structure, a three-dimensional laser scanner is used to obtain pipeline point cloud data.

3. The method for measuring an engine pipeline based on laser scanning according to claim 2, characterized in that: The step S2 comprises the following steps: Step S201: constructing a KD-tree using pipeline point cloud data; Step S202: setting the search parameters of KD-tree to search for neighboring points of the point cloud; Step S203: comparing the local density of each point cloud with a threshold value to identify outliers; Step S204: removing the calibrated outlier points from the pipeline point cloud data; Step S205: After removing outliers, post-process the pipeline point cloud data to improve the quality of the pipeline point cloud data.

4. The method for measuring an engine pipeline based on laser scanning according to claim 3, characterized in that: The step S3 comprises the following steps: Step S301: setting the Gaussian kernel size and standard deviation of the Gaussian filter; Step S302: Calculate Gaussian weights; Step S303: weighting the neighboring points of each point cloud using Gaussian weights; Step S304: repeatedly applying a Gaussian filter to the pipeline point cloud data to remove higher-order noise; Step S305: After Gaussian filtering, post-process the pipeline point cloud data.

5. The method for measuring engine pipelines based on laser scanning according to claim 4, characterized in that: The step S4 comprises the following steps: Step S401: extracting pipeline axis using pipeline point cloud data; Step S402: For each point cloud on the pipeline axis, calculate the curvature; Step S403: Calculate the surface normal direction of each point in the pipeline point cloud; Step S404: Aggregate point clouds with similar features into continuous areas; Step S405: performing point cloud clustering according to the geometric features of the pipeline point cloud data; Step S406: fitting the pipeline point cloud data to a geometric model through a fitting algorithm, and performing model segmentation according to the geometric model of the pipeline point cloud; Step S407: Merge regions with similar geometric features.

6. The method for measuring engine pipelines based on laser scanning according to claim 5, characterized in that: The step S5 comprises the following steps: Step S501: creating a parameterized model of the pipeline; Step S502: using the segmented pipeline point cloud data and the parameterized model to solve the specific parameters of each model; Step S503: converting the point cloud data of the pipeline into a segmented model; Step S504: All segmented models are combined and smoothed to complete the reverse modeling of the pipeline point cloud data.

7. The method for measuring engine pipelines based on laser scanning according to claim 6, characterized in that: The step S6 comprises the following steps: Step S601: using Geomagic Design X software to perform accuracy analysis on the reconstructed pipeline model; Step S602: creating a deviation cloud map between the reconstructed pipeline model and the original point cloud data; Step S603: Compare and analyze the fitting accuracy between the reconstructed pipeline model and the original point cloud data, and evaluate the error of pipeline measurement.

Images