Industrial part production nondestructive testing system based on 3D point cloud matching

Abstract

The invention discloses an industrial part production nondestructive testing system based on 3D point cloud matching. The system comprises a sensor data acquisition module; a data information preprocessing module; a data cutting processing module; a data feature acquisition module; a part basic model; a feature matching processing module; an error analysis calculation module and a quality analysis report module. According to the method, the three-dimensional model is scanned, the three-dimensional point cloud data of the three-dimensional model is obtained, the collected three-dimensional point cloud data is processed, the three-dimensional point cloud data is filtered through voxel grid filtering, the accuracy of the three-dimensional model is improved through point cloud registration, establishment of the three-dimensional model of the part is facilitated, and the accuracy of the part is improved. And cutting the three-dimensional model, reducing the calculation pressure, improving the matching between the detected three-dimensional point cloud data and the basic model data of the part, and judging whether the part has defects or not according to an error threshold value between the matching, thereby completing the nondestructive detection of the industrial part production.

Description

Technical Field

[0001] The present invention relates to the technical field of parts production detection, and in particular to a nondestructive detection system for industrial parts production based on 3D point cloud matching. Background Art

[0002] In the fierce competition of the global economy, various parts processing companies are exploring different methods to ensure the quality of their products. Among them, shape and size detection of produced parts is an important part of parts product quality assurance. At present, in the parts manufacturing industry with intensive parts processing demand and high quality requirements, such as aircraft parts manufacturing and processing industry, parts detection all relies on three-coordinate measuring instruments. However, its operation is complicated. Not only the detection speed cannot meet the requirements of online real-time detection, but also the detection accuracy is related to the proficiency of the operator. It can be seen that with the further improvement of the detection requirements of the industrial sector, the previous detection methods (three-coordinate measuring machine, theodolite / total station industrial measurement system, etc.) are difficult to meet the requirements, which is the direct reason for the research and development of visual measurement methods. The research and development of visual measurement methods and systems can make up for the shortcomings of current detection methods and further solve the practical problems of industrial detection departments. Although visual measurement technology has reached the practical stage and many products have been put into use, visual measurement technology is still in the development stage. Therefore, it is necessary and urgent to independently study the corresponding visual measurement methods and develop practical systems with high performance-price ratio and convenient operation.

[0003] A 3D point cloud is a data structure used to represent objects or scenes in three-dimensional space. In its most basic form, it is a collection of multiple three-dimensional coordinate points (X, Y, Z). These points are obtained by discrete sampling of the surface of the actual object or scene, so the point cloud can be regarded as a discrete representation of the scene surface in a given coordinate system.

[0004] Compared with traditional conventional measurement technology, the measurement technology based on 3D point cloud (3D Point Cloud) has many advantages such as fast speed, strong penetration, non-contact, full digital, high precision and high density.

[0005] In parts inspection, the applications of 3D point cloud matching technology include:

[0006] Defect detection: Identify defects or deviations on the part surface by comparing the actual point cloud with the point cloud of the design model.

[0007] Quality control: Real-time monitoring of parts quality during the production process to reduce defective rates.

[0008] Helps to improve production efficiency and product quality.

[0009] In defect detection, 3D point cloud matching technology can be applied in the following ways:

[0010] Surface defect identification: Compare point cloud data with a reference model to detect surface defects such as cracks, pits or wear.

[0011] Geometric deviation detection: Detect the deviation between the geometric dimensions of the part and the design model, and identify errors in manufacturing.

[0012] However, the existing 3D point cloud matching industrial parts production non-destructive testing system has problems such as large workload, large amount of data, inability to achieve effective and fast calculation and processing, large three-dimensional model, complex direct processing, inconvenience for fast calculation and analysis, and inability to accurately determine defects due to the inability to match features.

[0013] Currently, no effective solution has been proposed for the problems in the related technologies. Summary of the invention

[0014] In view of the problems in the related art, the present invention proposes a non-destructive testing system for industrial parts production based on 3D point cloud matching to overcome the above-mentioned technical problems existing in the existing related art.

[0015] To this end, the specific technical solution adopted by the present invention is as follows:

[0016] A nondestructive testing system for industrial parts production based on 3D point cloud matching, including a sensor data acquisition module, a data information preprocessing module, a data cutting processing module, a data feature acquisition module, a part basic model, a feature matching processing module, an error analysis calculation module and a quality analysis report module:

[0017] The sensor data acquisition module is used to collect data on industrial parts through sensors, that is, to scan the industrial parts multiple times through scanning equipment to obtain three-dimensional point cloud data of the industrial parts;

[0018] The data information preprocessing module is used to preprocess the three-dimensional point cloud data collected by the sensor, wherein the data information preprocessing module includes denoising, data alignment, point cloud clipping and point cloud registration of the data information to obtain a three-dimensional model of the industrial part;

[0019] The data cutting processing module is used to cut the collected three-dimensional model and cut the three-dimensional model of the industrial part into several small model parts;

[0020] The data feature acquisition module is used to perform feature acquisition processing on the several small model parts after cutting, and obtain a feature set of each small model part;

[0021] The part basic model is used to input the manufacturing model parameters of the industrial parts and obtain the basic data information and basic feature information of the industrial parts;

[0022] The feature matching processing module is used to match the basic feature information of the industrial parts with the information in the collected feature set;

[0023] The error analysis and calculation module is used to calculate the matching error of the matched basic feature information and the information in the collected feature set, and detect the surface defects and geometric errors of industrial parts by calculating the distance or geometric error between the three-dimensional point cloud data and the basic model;

[0024] The quality analysis report module is used to generate a report on the error analysis results to obtain a quality inspection report for industrial parts.

[0025] As a preferred embodiment, the sensor used in the sensor data acquisition module includes a laser scanner, a structured light scanner or an optical measurement system. The laser scanner, the structured light scanner or the optical measurement system scans the produced industrial parts to obtain detailed three-dimensional point cloud data.

[0026] As a preferred embodiment, the point cloud registration includes initial registration and fine registration;

[0027] The initial registration is calculated using the SIFT algorithm, and the SIFT algorithm locates the extreme points through Gaussian blur calculation, and performs preliminary alignment on the three-dimensional point cloud data scanned multiple times;

[0028] The calculation formula is as follows:

[0029] L(x,y,z)=G(x,y,z)*I(n,m);

[0030] Among them, L(x, y, z) is the positioning extreme point in the calculated three-dimensional point cloud data, G(x, y, z) is the Gaussian function, and I(n, m) is the light intensity during scanning.

[0031] As a preferred embodiment, the fine registration uses the ICP algorithm to perform fine registration on the three-dimensional point cloud data, specifically:

[0032] Assume the midpoint of the first scanned 3D point cloud data is a i , then the midpoint of the 3D point cloud data of the second scan is b i , find a in the 3D point cloud data of the first scan in the 3D point cloud data of the second scan i The closest point b i ;

[0033] Calculate the distance between two scanned points of the 3D point cloud data. The formula is:

[0034] d(a i ,b i )=||a i -b i ||.

[0035] As a preferred embodiment, the denoising in the data information preprocessing module adopts voxel grid filtering;

[0036] Set the voxel size V to 2, and the entire point cloud space will be divided into multiple voxel units of the same size;

[0037] Randomly select a data point from the 3D point cloud data, set the data point to (x, y, z), take the integer down in the voxel according to the data point, and average aggregate the obtained data points;

[0038] The mean aggregation is expressed as:

[0039]

[0040] Among them, (x', y', z') is the representative point position of the data point in the voxel, (x n ,y n , z n ) is the coordinate of the nth data point within the voxel, and N is the total number of data points within the voxel;

[0041] The filtered three-dimensional point cloud data is generated according to the calculated representative points of the data points in the voxel.

[0042] As a preferred embodiment, the data alignment matches the three-dimensional point cloud data scanned multiple times, so that points of the same three-dimensional point cloud data scanned multiple times can match each other;

[0043] Let the point of the first scanned 3D point cloud data be a i =(x i ,y i ,z i ), then the point of the 3D point cloud data scanned for the second time is b i =(x j ,y j ,z j );

[0044] By calculating the rotation matrix R and the displacement vector t corresponding to the three-dimensional point cloud data, the points of the three-dimensional point cloud data of the first scan after transformation are matched as much as possible with the points of the three-dimensional point cloud data of the second scan;

[0045] The calculation formula is as follows:

[0046]

[0047] in, is the midpoint between two points, and the midpoint is used to rigidly transform the two points to minimize the distance between them.

[0048] As a preferred embodiment, the data cutting processing module adopts bounding box segmentation:

[0049] By calculating the minimum bounding box of the model, the bounding box is divided into multiple parts along the x, y or z axis;

[0050] Formula: Bounding box coordinates:

[0051] Min x ≤x≤Max x ,

[0052] Min y ≤y≤Max y ,,

[0053] Min z ≤z≤Max z ,

[0054] Among them, Min x 、Min y 、Min z are the minimum values ​​of the boundaries on the x-axis, y-axis, and z-axis, Max x , Max y , Max z They are the maximum values ​​of the boundaries on the x-axis, y-axis, and z-axis respectively. By determining the size of the bounding box, the three-dimensional point cloud can be cut.

[0055] As a preferred embodiment, the data feature acquisition module is used to acquire features in the cut 3D model, and calculate the distribution and discreteness of the 3D point cloud data of the 3D model through statistical features, including:

[0056] The mean calculation formula is as follows:

[0057]

[0058] The formula for calculating variance is as follows:

[0059]

[0060] The formula for calculating the standard deviation is as follows:

[0061]

[0062] The features in the 3D model are collected and processed through the distribution and discreteness of the 3D point cloud data.

[0063] As a preferred embodiment, the part basic model is used to input the production data information of the part, and the feature matching processing module matches the features in the three-dimensional model with the basic feature information inputted by the part basic model, and the matching calculation formula is as follows:

[0064]

[0065] Among them, (x i ,y i ,z i ) is the collected 3D point cloud data, (x o ,y o ,z o ) is the data of the basic model of the part, and the features are matched by calculating the distance between two points.

[0066] As a preferred embodiment, the calculation formula of the error analysis calculation module is as follows:

[0067] C=|E i-1 -E i |,

[0068] Among them, E i is the error value of the i-th iteration, and C is the change in error;

[0069] The error is calculated in each iteration to determine whether the detection error gradually decreases below a preset threshold. Usually, the algorithm stops iterating when the error change is less than a certain threshold. When the change in C tends to zero and the error tends to the threshold, it is determined that there is a production error in the industrial parts.

[0070] The beneficial effects of the present invention are:

[0071] 1. The present invention scans and processes industrial parts through sensors to obtain three-dimensional point cloud data of parts, and in order to improve the accurate description of the three-dimensional model of parts, the industrial parts are scanned and collected at multiple angles and multiple times, and the data information obtained by scanning is analyzed and processed through a data information preprocessing module, so as to improve the accuracy of the data information, and can realize point cloud alignment of data information to realize the generation of three-dimensional models, and remove unnecessary point cloud parts through point cloud clipping to reduce the calculation burden;

[0072] 2. When the present invention is used, the accuracy of the three-dimensional model is improved by point cloud registration. The point cloud registration includes initial registration and fine registration. The point cloud data obtained from different scanning angles can be processed, and the accuracy of the three-dimensional model generation can be improved;

[0073] 3. When the present invention is used, the three-dimensional model is cut and processed by the data cutting processing module to reduce the size of the three-dimensional model, so that each cut part is processed by a separate feature matching processing module, that is, the three-dimensional model is cut and processed by the bounding box, so that the three-dimensional model is easy to form a specific range size, easy to match with the basic model of the part, realize matching detection, and complete the defect identification of the part;

[0074] 4. When the present invention is used, the three-dimensional point cloud data is filtered by voxel grid filtering to remove clutter in the three-dimensional point cloud data, and the calculated voxel representative points are used to generate the filtered point cloud. Each voxel unit will replace all the points in the voxel with one point, thereby significantly reducing the number of points in the point cloud, and the voxel grid filtering is used to remove or filter out unnecessary points, thereby obtaining a more accurate or more demand-compliant point cloud data set, and the point at the center of the voxel is used to represent all the points in the voxel, thereby removing unnecessary points, reducing the amount of point cloud data, and making the processing more efficient;

[0075] 5. When the present invention is used, the data information of the production part detection and the basic data of the part are matched through the feature matching processing module. The distribution and discreteness between the three-dimensional point cloud data information are used to facilitate the collection and processing of the three-dimensional point cloud data information features. The error is calculated in each iteration to determine whether the detection error is gradually reduced to below a preset threshold. Usually, the algorithm stops iterating when the error change is less than a certain threshold. When the change in C tends to zero and the error tends to the threshold, it is determined that the part has a production error.

[0076] The present invention has the advantages of obtaining 3D point cloud data of the 3D model by scanning the 3D model and processing the collected 3D point cloud data, thereby facilitating the establishment of a 3D model of a part, and reducing the pressure of calculation by cutting the 3D model, improving the matching of the detected 3D point cloud data with the basic model data of the part, and determining whether the part has defects through the error threshold between the matches, thereby completing non-destructive testing of the production of industrial parts. BRIEF DESCRIPTION OF THE DRAWINGS

[0077] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0078] Figure 1It is a structural schematic diagram of a nondestructive testing system for industrial parts production based on 3D point cloud matching according to an embodiment of the present invention. DETAILED DESCRIPTION

[0079] To further illustrate each embodiment, the present invention provides drawings, which are part of the disclosure of the present invention and are mainly used to illustrate the embodiments and can be used in conjunction with the relevant descriptions in the specification to explain the operating principles of the embodiments. With reference to these contents, ordinary technicians in the field should be able to understand other possible implementations and advantages of the present invention. The components in the figures are not drawn to scale, and similar component symbols are generally used to represent similar components.

[0080] According to an embodiment of the present invention, a nondestructive testing system for industrial parts production based on 3D point cloud matching is provided.

[0081] The present invention is further described with reference to the accompanying drawings and specific embodiments. Figure 1 As shown, a nondestructive testing system for industrial parts production based on 3D point cloud matching according to an embodiment of the present invention includes a sensor data acquisition module, a data information preprocessing module, a data cutting processing module, a data feature acquisition module, a part basic model, a feature matching processing module, an error analysis calculation module and a quality analysis report module:

[0082] The sensor data acquisition module is used to collect data on industrial parts through sensors, that is, to scan the industrial parts multiple times through scanning equipment to obtain three-dimensional point cloud data of the industrial parts;

[0083] The data information preprocessing module is used to preprocess the three-dimensional point cloud data collected by the sensor, wherein the data information preprocessing module includes denoising, data alignment, point cloud clipping and point cloud registration of the data information to obtain a three-dimensional model of the industrial part;

[0084] The data cutting processing module is used to cut the collected three-dimensional model and cut the three-dimensional model of the industrial part into several small model parts;

[0085] The data feature acquisition module is used to perform feature acquisition processing on the several small model parts after cutting, and obtain a feature set of each small model part;

[0086] The part basic model is used to input the manufacturing model parameters of the industrial parts and obtain the basic data information and basic feature information of the industrial parts;

[0087] The feature matching processing module is used to match the basic feature information of the industrial parts with the information in the collected feature set;

[0088] The error analysis and calculation module is used to calculate the matching error of the matched basic feature information and the information in the collected feature set, and detect the surface defects and geometric errors of industrial parts by calculating the distance or geometric error between the three-dimensional point cloud data and the basic model;

[0089] The quality analysis report module is used to generate a report on the error analysis results to obtain a quality inspection report for industrial parts.

[0090] As a preferred embodiment, the sensor used in the sensor data acquisition module includes a laser scanner, a structured light scanner or an optical measurement system. The laser scanner, the structured light scanner or the optical measurement system scans the produced parts to obtain detailed point cloud data.

[0091] It should be noted that the industrial parts are non-destructively scanned by laser scanners, structured light scanners or optical measurement systems to obtain the three-dimensional point cloud data information of the parts.

[0092] As a preferred embodiment, the steps of the sensor scanning the production parts are as follows:

[0093] Check the equipment: ensure that the laser scanner, the structured light scanner or the optical measurement system is in good condition, including calibration, cleaning and normal operation; check the power supply and software version of the scanner;

[0094] Prepare parts: clean the surface of the parts to remove dirt or grease that may affect the scan. If the surface of the parts is smooth and highly reflective, spray scanning powder to improve the scan quality.

[0095] Set up the scanning environment: Choose an appropriate scanning environment to avoid the effects of strong light, shadows, and temperature changes on the scanning results; ensure that the scanning area is stable and vibration-free;

[0096] Multiple scans: Aim the scanner at the part and perform a comprehensive scan; depending on the type of scanner and the complexity of the part, perform multi-angle and multi-position scans to obtain complete 3D data. Ensure that the scanner and the part remain stable during the scanning process to avoid movement of the part or scanner. The collected data should be dense enough to ensure high resolution.

[0097] It should be noted that in order to achieve accurate scanning and processing of parts, both the scanning equipment and the parts need to be processed to improve the accuracy of scanning parts, and multiple scanning processes should be performed on the parts at different angles to facilitate the acquisition of complete three-dimensional data. The collected data should be dense enough to ensure high resolution, and then the three-dimensional establishment of the parts is completed through matching between the data.

[0098] As a preferred embodiment, the point cloud registration includes initial registration and fine registration;

[0099] The initial registration is calculated using the SIFT algorithm, and the SIFT algorithm locates the extreme points through Gaussian blur calculation, and performs preliminary alignment on the three-dimensional point cloud data scanned multiple times;

[0100] The calculation formula is as follows:

[0101] L(x,y,z)=G(x,y,z)*I(n,m);

[0102] Among them, L(x, y, z) is the positioning extreme point in the calculated three-dimensional point cloud data, G(x, y, z) is the Gaussian function, and I(n, m) is the light intensity during scanning.

[0103] The point cloud data acquired at different scanning angles are initially aligned by initial registration to complete the preliminary alignment of the three-dimensional point clouds in the first scan and the second scan.

[0104] As a preferred embodiment, the fine registration uses the ICP algorithm to perform fine registration on the three-dimensional point cloud data, specifically:

[0105] Assume the midpoint of the first scanned 3D point cloud data is a i , then the midpoint of the 3D point cloud data of the second scan is b i , find a in the 3D point cloud data of the first scan in the 3D point cloud data of the second scan i The closest point b i ;

[0106] Calculate the distance between two scanned points of the 3D point cloud data. The formula is:

[0107] d(a i ,b i )=||a i -b i ||;

[0108] It should be noted that the ICP algorithm is used to improve the accuracy of three-dimensional point cloud registration, which is convenient for improving the familiarity between the first scan and the second scan.

[0109] As a preferred embodiment, the denoising in the data information preprocessing module adopts voxel grid filtering;

[0110] Set the voxel size V to 2, and the entire point cloud space will be divided into multiple voxel units of the same size;

[0111] Randomly select a data point from the 3D point cloud data, set the data point to (x, y, z), take the integer down in the voxel according to the data point, and average aggregate the obtained data points;

[0112] The mean aggregation is expressed as:

[0113]

[0114] Among them, (x', y', z') is the representative point position of the data point in the voxel, (x n ,y n , z n ) is the coordinate of the nth data point within the voxel, and N is the total number of data points within the voxel;

[0115] The filtered three-dimensional point cloud data is generated according to the calculated representative points of the data points in the voxel.

[0116] It should be noted that all the 3D point cloud information of parts is processed through voxel grid filtering to reduce the 3D point cloud information, that is, each voxel unit will replace all the points in the voxel with one point, thereby significantly reducing the number of points in the point cloud, and then filtering out abnormal clutter data information and improving the accuracy of data information.

[0117] As a preferred embodiment, the data alignment matches the three-dimensional point cloud data scanned multiple times, so that points of the same three-dimensional point cloud data scanned multiple times can match each other;

[0118] Let the point of the first scanned 3D point cloud data be a i =(x i ,y i ,z i ), then the point of the 3D point cloud data scanned for the second time is b i =(x j ,y j ,z j );

[0119] By calculating the rotation matrix R and the displacement vector t corresponding to the three-dimensional point cloud data, the points of the three-dimensional point cloud data of the first scan after transformation are matched as much as possible with the points of the three-dimensional point cloud data of the second scan;

[0120] The calculation formula is as follows:

[0121]

[0122] in, is the midpoint between two points, and the midpoint is used to rigidly transform the two points to minimize the distance between them.

[0123] It should be noted that in order to achieve matching of three-dimensional point cloud data collected by multiple scans, improve the accuracy of three-dimensional model establishment, and match between two points, the matching between the two points is determined by rigid transformation, and in order to improve the matching between the two points, the center point between the two points is used as the origin for rigid transformation adjustment.

[0124] As a preferred embodiment, the point cloud cropping removes or filters out unnecessary points through voxel grid filtering to obtain a more accurate or more demand-compliant point cloud data set, by dividing the point cloud data into voxels of a fixed size, and the voxels are cubic grids with a side length of 2, and using the point at the center of the voxel to represent all the points in the voxel, thereby removing unnecessary points, reducing the amount of point cloud data, and making processing more efficient.

[0125] It should be noted that in order to reduce the collected 3D point cloud, the point at the center of the voxel is used to represent all the points in the voxel, thereby removing unnecessary points, reducing the amount of point cloud data, making the processing more efficient, facilitating calculation processing, improving calculation efficiency and reducing calculation pressure.

[0126] As a preferred embodiment, the data cutting processing module adopts bounding box segmentation:

[0127] By calculating the minimum bounding box of the model, the bounding box is divided into multiple parts along the x, y or z axis;

[0128] Formula: Bounding box coordinates:

[0129] Min x ≤x≤Max x ,

[0130] Min y ≤y≤Max y ,,

[0131] Min z ≤z≤Max z ,

[0132] Among them, Min x 、Min y 、Min z are the minimum values ​​of the boundaries on the x-axis, y-axis, and z-axis, Max x , Max y , Max z They are the maximum values ​​of the boundaries on the x-axis, y-axis, and z-axis respectively. By determining the size of the bounding box, the three-dimensional point cloud can be cut.

[0133] It should be noted that the setting of the minimum bounding box makes it easier to cut and segment the three-dimensional model, and to perform calculations on each feature after segmentation, thereby improving calculation efficiency and reducing calculation pressure.

[0134] As a preferred embodiment, the data feature acquisition module is used to acquire features in the cut three-dimensional point cloud, because the data information acquired by scanning is used to describe the distribution and discreteness of the data through statistical features.

[0135] The mean calculation formula is as follows:

[0136]

[0137] The formula for calculating variance is as follows:

[0138]

[0139] The formula for calculating the standard deviation is as follows:

[0140]

[0141] The distribution and discreteness of three-dimensional point cloud data information make it easier to collect and process the information features of three-dimensional point cloud data.

[0142] It should be noted that in order to realize feature extraction of data information, the distribution and discreteness of data are described by statistical features, which facilitates feature collection of data and feature matching of data.

[0143] As a preferred embodiment, the part basic model is used to input the production data information of the part, and the features of the 3D cloud data information are matched with the feature information inputted by the part basic model by searching the feature matching processing module, and the matching calculation formula is as follows:

[0144]

[0145] Among them, (x i ,y i ,z i ) is the collected 3D point cloud data, (x o ,y o ,z o ) is the data of the basic model of the part, and the features are matched by calculating the distance between two points.

[0146] It should be noted that by calculating and analyzing the distance between feature points, it is convenient to complete the matching of two points and to realize feature matching.

[0147] As a preferred embodiment, the calculation formula of the error analysis calculation module is as follows:

[0148] C=|E i-1 -E i |,

[0149] Among them, E i is the error value of the i-th iteration, and C is the change in error;

[0150] The error is calculated in each iteration to determine whether the detection error is gradually reduced to below the preset threshold. Usually, the algorithm stops iterating when the error change is less than a certain threshold. When the change in C tends to zero and the error tends to the threshold, it is determined that the part has a production error.

[0151] It should be noted that through the error analysis calculation module, that is, through the iterative error calculation processing, the control error gradually reaches below the threshold, so as to determine that the production of the parts is qualified and there is no error. If the error amount is in the iterative process, when the change of C tends to zero, when the error tends to the threshold, it is determined that the part has a production error.

[0152] In summary, the present invention has the advantages of obtaining three-dimensional point cloud data of the three-dimensional model by scanning the three-dimensional model, and processing the collected three-dimensional point cloud data, so as to improve the establishment of the three-dimensional model of the part, and reduce the calculation pressure by cutting the three-dimensional model, improve the matching of the detected three-dimensional point cloud data with the basic model data of the part, and determine whether the part is defective through the error threshold between the matches, thereby completing the non-destructive testing of the production of industrial parts.

[0153] 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 spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A non-destructive testing system for industrial parts production based on 3D point cloud matching, characterized in that: It includes sensor data acquisition module, data information preprocessing module, data cutting processing module, data feature acquisition module, part basic model, feature matching processing module, error analysis calculation module and quality analysis report module: The sensor data acquisition module is used to collect data on industrial parts through sensors, that is, to scan the industrial parts multiple times through scanning equipment to obtain three-dimensional point cloud data of the industrial parts; The data information preprocessing module is used to preprocess the three-dimensional point cloud data collected by the sensor, wherein the data information preprocessing module includes denoising, data alignment, point cloud clipping and point cloud registration of the data information to obtain a three-dimensional model of the industrial part; The data cutting processing module is used to cut the collected three-dimensional model and cut the three-dimensional model of the industrial part into several small model parts; The data feature acquisition module is used to perform feature acquisition processing on the several small model parts after cutting, and obtain a feature set of each small model part; The part basic model is used to input the manufacturing model parameters of the industrial parts and obtain the basic data information and basic feature information of the industrial parts; The feature matching processing module is used to match the basic feature information of the industrial parts with the information in the collected feature set; The error analysis and calculation module is used to calculate the matching error of the matched basic feature information and the information in the collected feature set, and detect the surface defects and geometric errors of industrial parts by calculating the distance or geometric error between the three-dimensional point cloud data and the basic model; The quality analysis report module is used to generate a report on the error analysis results to obtain a quality inspection report for industrial parts.

2. The industrial parts production nondestructive testing system based on 3D point cloud matching according to claim 1 is characterized in that: The sensors used in the sensor data acquisition module include laser scanners, structured light scanners or optical measurement systems. The laser scanners, structured light scanners or optical measurement systems scan the produced industrial parts to obtain detailed three-dimensional point cloud data.

3. The industrial parts production nondestructive testing system based on 3D point cloud matching according to claim 1 is characterized in that: The point cloud registration includes initial registration and fine registration; The initial registration is calculated using the SIFT algorithm, and the SIFT algorithm locates the extreme points through Gaussian blur calculation, and performs preliminary alignment on the three-dimensional point cloud data scanned multiple times; The calculation formula is as follows: L(x,y,z)=G(x,y,z)*I(n,m); Among them, L(x, y, z) is the extreme point of positioning in the calculated three-dimensional point cloud data, G(x, y, z) is the Gaussian function, and I(n, m) is the light intensity during scanning.

4. The industrial parts production nondestructive testing system based on 3D point cloud matching according to claim 3 is characterized in that: The fine registration uses the ICP algorithm to perform fine registration on the three-dimensional point cloud data, specifically: Assume the midpoint of the first scanned 3D point cloud data is a i , then the midpoint of the 3D point cloud data of the second scan is b i , find a in the 3D point cloud data of the first scan in the 3D point cloud data of the second scan i The closest point b i ; Calculate the distance between two scanned points of the 3D point cloud data. The formula is: d(a i ,b i )=||a i -b i ||。 5. The industrial parts production nondestructive testing system based on 3D point cloud matching according to claim 1 is characterized in that: The denoising in the data information preprocessing module adopts voxel grid filtering; Set the voxel size V to 2, and the entire point cloud space will be divided into multiple voxel units of the same size; Randomly select a data point from the 3D point cloud data, set the data point to (x, y, z), take the integer down in the voxel according to the data point, and average aggregate the obtained data points; The mean aggregation is expressed as: Among them, (x', y', z') is the representative point position of the data point in the voxel, (x n ,y n , z n ) is the coordinate of the nth data point within the voxel, and N is the total number of data points within the voxel; The filtered three-dimensional point cloud data is generated according to the calculated representative points of the data points in the voxel.

6. The industrial parts production nondestructive testing system based on 3D point cloud matching according to claim 1 is characterized in that: The data alignment matches the three-dimensional point cloud data scanned multiple times so that points of the same three-dimensional point cloud data scanned multiple times can match each other; Let the point of the first scanned 3D point cloud data be a i =(x i ,y i ,z i ), then the point of the 3D point cloud data scanned for the second time is b i =(x j ,y j ,z j ); By calculating the rotation matrix R and the displacement vector t corresponding to the three-dimensional point cloud data, the points of the three-dimensional point cloud data of the first scan after transformation are matched as much as possible with the points of the three-dimensional point cloud data of the second scan; The calculation formula is as follows: in, is the midpoint between two points, and the midpoint is used to rigidly transform the two points to minimize the distance between them.

7. The industrial parts production nondestructive testing system based on 3D point cloud matching according to claim 1 is characterized in that: The data cutting processing module adopts bounding box segmentation, including: By calculating the minimum bounding box of the three-dimensional model, the bounding box is divided into multiple parts along the x, y or z axis; Formula: Bounding box coordinates: Among them, Min x 、Min y 、Min z are the minimum values ​​of the boundaries on the x-axis, y-axis, and z-axis, Max x , Max y , Max z They are the maximum values ​​of the boundaries on the x-axis, y-axis, and z-axis respectively. By determining the size of the boundary bounding box, the three-dimensional model can be cut.

8. The industrial parts production nondestructive testing system based on 3D point cloud matching according to claim 6 is characterized in that: The data feature acquisition module is used to acquire features in the cut 3D model, and calculate the distribution and discreteness of the 3D point cloud data of the 3D model through statistical features, including: The mean calculation formula is as follows: The formula for calculating variance is as follows: The standard deviation is calculated as follows: The features in the 3D model are collected and processed through the distribution and discreteness of the 3D point cloud data.

9. The industrial parts production nondestructive testing system based on 3D point cloud matching according to claim 1, characterized in that: The part basic model is used to input the production data information of the part, and the feature matching processing module matches the features in the three-dimensional model with the basic feature information inputted by the part basic model, and the matching calculation formula is as follows: Among them, (x i ,y i ,z i ) is the collected 3D point cloud data, (x o ,y o ,z o ) is the data of the basic model of the part, and the features are matched by calculating the distance between two points.

10. The industrial parts production nondestructive testing system based on 3D point cloud matching according to claim 1, characterized in that: The calculation formula of the error analysis calculation module is as follows: C=|E i-1 -AND i |, Among them, E i is the error value of the i-th iteration, and C is the change in error; The error is calculated in each iteration to determine whether the detection error gradually decreases below a preset threshold. Usually, the algorithm stops iterating when the error change is less than a certain threshold. When the change in C tends to zero and the error tends to the threshold, it is determined that there is a production error in the industrial parts.

Images