3D size measurement method suitable for curved surface structure

Abstract

The invention relates to a 3D size measurement method suitable for a curved surface structure. Capturing a target curved surface by using a camera and a laser scanner; using structured light to generate point cloud data; performing point cloud registration among the plurality of views by means of a SIFT feature matching algorithm to generate a unified point cloud model; using a deep learning algorithm to train the feature recognition model, and extracting key geometric features of the curved surface; identifying curved surface features with different edges, angles and curvatures by marking and classifying, and converting the features into mathematical representation by using a Bezier curve parameterized modeling method; establishing a comprehensive error model, evaluating a plurality of error sources, and fitting an error propagation function through experimental data; adjusting dimension measurement by using a least square method; a real-time data acquisition and feedback system is introduced, environment change is monitored by using a temperature and humidity sensor, and scanning parameters including scanning speed and light source brightness are adjusted through a PID controller.

Description

Technical Field

[0001] The present invention relates to the field of 3D dimension measurement of curved surface structures, and in particular to a 3D dimension measurement method suitable for curved surface structures. Background Art

[0002] The 3D dimension measurement methods currently applied to curved surface structures still have many shortcomings and disadvantages in practical applications, which are mainly reflected in measurement accuracy, adaptability, efficiency and stability. First, the traditional 3D dimension measurement methods mainly rely on fixed geometric rules or preset feature extraction parameters, and are usually unable to adapt to the complex and changeable geometric characteristics of the surface. For example, the high curvature and sharp edge areas commonly seen in curved surface structures often cannot be effectively identified and fitted in traditional methods. This is because traditional methods usually rely on manually preset thresholds and rules to extract surface features, such as edges, angles or curvature changes, and are difficult to deal with highly irregular and multi-scale surface features, resulting in severe limitations in the measurement accuracy on complex curved surface structures. Secondly, in terms of error correction, traditional measurement methods often ignore the diversity and complexity of error sources. Usually, the existing technology will correct a single error source, such as compensating for equipment errors through simple scanner calibration, or compensating for environmental errors through environmental monitoring. However, in practical applications, the error sources of 3D dimension measurement are often diverse, such as equipment errors, changes in material properties, and environmental influences, which work together to affect the measurement results. Traditional methods lack a systematic error model to comprehensively consider the synergistic effects of different error sources, especially the deformation of material properties under changes in external conditions such as temperature and humidity, which often cannot be effectively compensated, resulting in reduced stability and measurement accuracy in diverse measurement environments. In addition, existing error correction methods are usually static and cannot be dynamically adjusted according to different measurement conditions. The flexibility and adaptability of error compensation are obviously insufficient.

[0003] Another common problem is the lack of data processing efficiency of traditional methods. In the 3D dimension measurement of curved surface structures, a large amount of point cloud data needs to be processed, especially in high curvature areas with complex details, and the amount of data grows exponentially. The processing process of existing methods is usually inefficient, especially when processing large amounts of data or complex surfaces, which takes a lot of time and is easily limited by computing resources. In order to reduce computational complexity, traditional methods often use low-order geometric fitting methods, such as low-order Bezier curves or polynomial fitting to simplify the model. However, this simplification usually sacrifices fitting accuracy, resulting in an increase in the error between the measurement result and the actual size. Even if high-order curves are used for fitting, traditional methods lack a dynamic adjustment mechanism for controlling point density, and cannot flexibly allocate computing resources according to the geometric characteristics of different areas of the surface. Therefore, in practical applications, traditional methods face the contradiction between high precision and high efficiency, especially in application scenarios that require real-time measurement and real-time feedback. It is difficult to meet the requirements. Furthermore, the adaptability of existing 3D dimension measurement methods in complex surface structures is also obviously insufficient. Traditional methods rely on a single feature extraction strategy or a fixed error compensation model, which cannot effectively adapt to surface features of different types and sizes, especially in complex structures with multiple curvatures and angles that change significantly. For example, for the surface of certain special materials, due to the material's own characteristics (such as elastic modulus, thermal expansion coefficient, etc.), its surface may deform at a specific temperature or humidity, resulting in inaccurate measurement results. Existing methods lack the ability to dynamically adjust and flexibly respond to different measurement environments, especially when the environment changes drastically. The measurement accuracy is easily affected and the stability is poor, resulting in the measurement results being unrepeatable and unreliable. Summary of the invention

[0004] The purpose of the present invention is to provide a 3D dimension measurement method suitable for curved surface structures, thereby solving some of the drawbacks and shortcomings pointed out in the background technology.

[0005] The present invention solves the above-mentioned technical problems by adopting the following technical solutions: A 3D dimension measurement method applicable to curved surface structures, comprising: S1, multi-view curved surface reconstruction:

[0006] S1.1, use camera and laser scanner to capture the target surface; use structured light to generate point cloud data;

[0007] S1.2, using SIFT feature matching algorithm, point cloud registration is performed between multiple views to generate a unified point cloud model;

[0008] S2. Intelligent feature recognition and parametric modeling:

[0009] S2.1. Use deep learning algorithms to train feature recognition models and extract key geometric features of surfaces.

[0010] S2.2. Identify surface features including edges, angles, and curvatures by labeling and classification, and convert the features into mathematical representations using the parametric modeling method of Bezier curves;

[0011] S3. Dimension measurement based on error model:

[0012] S3.1. Establish a comprehensive error model to evaluate multiple error sources, including the geometric error of the scanner itself, deformation caused by material properties, and the influence of environmental factors; fit the error propagation function through experimental data;

[0013] S3.2, use the least squares method to adjust the size measurement;

[0014] S4, dynamic adjustment and feedback control:

[0015] S4.1. Introduce a real-time data acquisition and feedback system, use temperature and humidity sensors to monitor environmental changes, and adjust scanning parameters including scanning speed and light source brightness through a PID controller.

[0016] Furthermore, the intelligent feature recognition and parameterized modeling include:

[0017] A multi-scale analysis strategy is introduced to enable the neural network to extract local feature information at different scales. The multi-scale strategy gradually extracts features at different levels by introducing convolutional layers, so that the edge, angle and curvature information in the surface can be identified at a detailed scale. For the local curvature area of ​​the surface, the second-order derivative is calculated and adaptively adjusted to identify the local surface features. The local curvature is calculated by the following formula:

[0018]

[0019] Among them, C local (x,y) represents the local curvature of the surface at position (x,y); Represents the second-order derivative of the surface height function Z(x,y) in the x direction, which is used to calculate the curvature change of the surface; represents the second-order derivative of the surface height function Z(x,y) in the y direction; w i The weight of a local feature indicates the influence of the feature in a specific area and is usually adjusted dynamically according to the geometric characteristics of the surface. i The area of ​​region i, used to weigh spatial factors in the local curvature calculation.

[0020] Furthermore, the intelligent feature recognition and parameterized modeling include:

[0021] The Bezier curve is selected for local surface fitting; the density of control points is dynamically adjusted according to different requirements of surface features; high-order Bezier curves are used to fit surface changes; and low-order Bezier curves are used to simplify modeling in flat or small curvature areas; the parametric modeling method of Bezier curves is used to interpolate and fit the control points in each local area through the following formula:

[0022]

[0023] Among them, S Bezier (u,v) represents the value of the Bezier surface at parameters u and v, that is, the geometric shape of the local surface;

[0024] P ij Represents the control points of the Bezier surface, whose positions and numbers are adaptively adjusted according to the surface characteristics; in high curvature areas, the density of control points will be greater, while in flat areas, the control points are sparse; represents the Bezier basis function, the i-th basis function on parameter u, with order m; represents the Bezier basis function, the jth basis function on parameter v, with order n; m and n represent the order of the surface in the u and v directions, respectively, and are adjusted according to changes in surface characteristics.

[0025] Furthermore, the intelligent feature recognition and parameterized modeling include:

[0026] By analyzing the changes in angles and curvatures in different areas of the surface, the number and density of control points are dynamically adjusted; the mechanism of dynamic control point adjustment can be expressed by the following formula:

[0027]

[0028] Among them, Density controlpoints Indicates the density of control points, that is, the spacing or density between control points; It represents the second-order derivative of the surface height function Z(x,y) in the x direction, reflecting the degree of change of the surface in the x direction; It represents the second-order derivative of the surface height function Z(x,y) in the y direction, reflecting the degree of change of the surface in the y direction; n represents the number of sample points in the surface calculation area, which is the total number of local feature points in the area.

[0029] Furthermore, the dimension measurement method based on the error model includes:

[0030] A comprehensive error model is established to integrate the joint effects of scanner, material and environment on the measurement results; the model is expressed in the following form:

[0031] E total =f(E scanner ,E material ,E environment )

[0032] Among them, E total Represents the final comprehensive error, which indicates the deviation between the measurement result and the true value; E scanner It is the geometric error caused by the scanner, which comes from the hardware problem of the scanner and improper calibration. material Represents the error caused by material properties, including thermal expansion and elastic modulus changes of materials at different temperatures; E environment is the error caused by environmental factors, including the impact of changes in environmental conditions such as temperature and humidity on the measurement; f(·) is the error propagation function, which is used to describe the interaction between different error sources and their contribution to the final measurement error.

[0033] Furthermore, the dimension measurement method based on the error model includes:

[0034] The experimental data provide error performance under different conditions including different materials and different environments, so that the error propagation function can be dynamically adjusted according to the actual measurement conditions; through regression analysis, the form of the error propagation function is fitted, which is expressed as:

[0035] ΔD=α 1 ·E scanner +α 2 ·E material +α 3 ·E environment +∈

[0036] Among them, ΔD represents the deviation of dimensional measurement, that is, the difference between the corrected measurement result and the true value; α 1 ,α 2 ,α 3 are regression coefficients, representing the influence of scanner, material and environmental factors on the measurement deviation. The regression coefficients are obtained by fitting the experimental data and quantify the contribution of each error source to the final error. scanner ,E material ,E environment They represent the errors caused by scanner, material and environmental factors respectively; ∈ is the fitting error, which represents the deviation between the error propagation function and the actual experimental data.

[0037] Furthermore, the dimension measurement method based on the error model includes:

[0038] The error propagation function is obtained and the regression coefficients are fitted according to the experimental data, and then the data is used to make the size correction; the parameters are optimized by minimizing the sum of squares of the errors between the model prediction value and the actual measurement value; the size measurement value is adjusted by the least squares method, which is achieved through the following objective function:

[0039]

[0040] in, Indicates the corrected dimensional measurement result; the dimensional value obtained by the least squares optimization method minimizes the error; D i is the i-th data point in the original experimental measurement data, representing the original measurement result without error correction; It is the predicted value calculated by the error propagation function, which is dynamically adjusted according to the influence of error sources such as scanner, material and environment; n is the number of experimental data points, that is, the number of samples used for training and optimization.

[0041] The present invention is applicable to the 3D dimension measurement method of the curved surface structure and has significant beneficial effects. First, it effectively extracts and processes the complex geometric features of the surface through intelligent feature recognition and parametric modeling, especially under the support of multi-scale analysis and dynamic adjustment of control point density, so that different curvature regions can achieve adaptive fitting, thereby improving the accuracy of surface modeling. Secondly, the present invention introduces a comprehensive error model, systematically considers the influence of scanners, material properties and environmental factors on measurement accuracy, fits the error propagation function through experimental data and dynamically adjusts the regression coefficient, so that the model can adapt to different measurement conditions. Thirdly, through the optimization of the least squares method, the deviation of the error source on the dimension measurement is further reduced, and the accuracy and stability of the measurement results are improved. The method has a high degree of adaptability under a variety of environmental and material conditions, and is particularly suitable for the precise measurement of complex curved surface structures in industrial environments. It can meet the needs of high-precision and high-efficiency 3D dimension measurement, significantly improves the reliability and applicability of the measurement system, and its automated and intelligent processing method also greatly reduces human intervention, making it more advantageous in actual operation. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] Figure 1 The present invention is a flow chart of a 3D dimension measurement method applicable to curved surface structures.

[0043] Figure 2 This is a flow chart of the intelligent feature recognition and parametric modeling of the present invention.

[0044] Figure 3 This is a flow chart of the dimension measurement method based on the error model of the present invention. DETAILED DESCRIPTION

[0045] The specific implementation modes of the present invention will be described in detail below in conjunction with the accompanying drawings.

[0046] Combined with Figure 1 In the 3D dimension measurement method applicable to curved surface structures of the present invention, the first step is multi-view surface reconstruction. This process mainly involves how to obtain information of the target surface from different angles, and integrate the data from different perspectives together through algorithms to form a unified three-dimensional point cloud model. S1.1 involves using a camera and a laser scanner to capture the target surface, and generating point cloud data with the help of structured light. Structured light is a technology commonly used in three-dimensional scanning. It projects a known pattern of light onto the surface of an object, and then uses a camera to capture the reflected light to deduce the three-dimensional coordinates of the surface. Combined with the use of cameras and laser scanners, every detail on the surface can be efficiently captured and corresponding point cloud data can be generated. Point cloud data refers to a model composed of a set of discrete points in three-dimensional space, each point has three-dimensional coordinates (X, Y, Z). These points are obtained through the measurement process of a laser scanner or structured light, which can accurately reflect the geometric shape of the surface of the object, and the resolution can be adjusted as needed to ensure sufficient accuracy. Next, S1.2 realizes point cloud registration with the help of SIFT (Scale-Invariant Feature Transform) algorithm. SIFT is a feature matching algorithm widely used in computer vision, especially good at extracting and matching feature points under different viewing angles. In the process of multi-view shooting, the target surface will have different image data due to different viewing angles. Through the SIFT algorithm, the same feature points can be found under different viewing angles, thereby achieving matching between point clouds. The SIFT algorithm has scale invariance, rotation invariance and partial illumination invariance, which enables it to effectively extract stable feature points when processing image data acquired at different angles and different lighting conditions, thereby achieving accurate registration of point clouds between multiple views. In this way, the point cloud data captured under different viewing angles are spliced ​​together to finally generate a unified and complete point cloud model. This point cloud model contains all the geometric information of the target surface and is the basis for further 3D reconstruction and dimensional measurement. In this process, the accuracy of point cloud registration directly determines the accuracy of the final surface model. Therefore, an efficient and accurate algorithm is needed to handle the matching problem and ensure that the registration error between point clouds is minimized, thereby improving the accuracy and reliability of the entire 3D reconstruction process.

[0047] The second step is intelligent feature recognition and parametric modeling. The core purpose of this process is to identify and extract the key geometric features in the target surface through deep learning algorithms, and convert these features into mathematical representations for further dimensional measurement and model analysis. S2.1 First, the feature recognition model is trained using a deep learning algorithm to extract the key geometric features of the surface from the acquired three-dimensional point cloud data. These features include the edges, angles, curvatures, etc. of the surface, which are important signs of the surface shape and determine the geometric properties and structural characteristics of the surface. To achieve this goal, deep learning models, especially convolutional neural networks (CNNs), are widely used in feature extraction tasks of images and three-dimensional data. Through a large amount of training data, deep learning models can learn how to identify these key geometric features from surface point cloud data without artificial definition or manual extraction. This process effectively solves the problem of relying on manual experience for feature recognition in traditional methods, and improves the automation and accuracy of feature extraction. S2.2 further identifies different surface features including edges, angles, curvatures, etc. through labeling and classification, and converts these identified features into mathematical models. Specifically, the edge area usually has a large curvature change, the angle change represents the local shape change of the surface, and the curvature reflects the overall curvature of the surface. By classifying these features, it can help to better understand the morphology of the surface and the geometric characteristics of its various parts. Subsequently, the parametric modeling method of Bezier curves is used to convert these geometric features into mathematical representations. Bezier curves are a common parametric curve that is widely used in surface modeling. They flexibly fit the curve through control points and have good accuracy and controllability. In this method, according to the features identified on the surface, such as edges and curvature, suitable Bezier curves are selected for local fitting. By adjusting the control points and order of the Bezier curve, the local features of the surface can be accurately described, such as increasing the density of control points in the edge area to adapt to the curvature change, and reducing the control points in the flat area to simplify the model. Finally, these local Bezier curves are combined to form a complete parametric surface model, which can not only accurately describe the shape of the surface, but also has good flexibility and operability due to the parametric characteristics of the Bezier curve, which is convenient for subsequent dimensional measurement and surface analysis.

[0048] The third step is the dimensional measurement based on the error model, which aims to evaluate and adjust the various error sources that occur during the measurement process by establishing a comprehensive error model to ensure that the final dimensional data is more accurate. S3.1 The first step is to establish a comprehensive error model and evaluate multiple error sources. Specifically, the error sources can be divided into several categories, including the geometric error of the scanner itself, the deformation caused by material properties, and the influence of environmental factors. The geometric error of the scanner usually includes systematic errors and random errors, which are caused by the accuracy of the equipment, inaccurate calibration of the sensor, or any mechanical errors that occur during the scanning process. In addition, the deformation caused by material properties includes surface deformation due to the elasticity, thermal expansion or other physical properties of the material, which varies under different environmental conditions, such as temperature, humidity, etc. The influence of environmental factors, such as temperature changes, lighting conditions or air humidity, can also lead to inconsistencies in the scan data, thereby affecting the accuracy of the measurement. In order to comprehensively evaluate these error sources, it is necessary to use experimental data and fit the error propagation function through the analysis of multiple measurement results. The error propagation function is a mathematical model used to describe the impact of each error source on the final measurement result, and can quantify the contribution of each error source to the overall measurement accuracy. By analyzing the contribution of different error sources, we can accurately understand the specific impact of each error on the measurement results and provide a basis for subsequent error correction. S3.2 Use the least squares method to adjust the dimensional measurement. The least squares method is a commonly used mathematical optimization method, which is mainly used to find the parameters that best match the actual situation in data with large errors. In the process of dimensional measurement, the measured data is compared with the expected true value through the established error model, and the error between the measured result and the true value is calculated. The basic principle of the least squares method is to obtain the optimal solution by minimizing the sum of squares of the errors, that is, to weight all the measured data and adjust the model parameters to minimize the error. In this way, the deviation caused by various error sources can be eliminated, thereby improving the measurement accuracy. For example, when measuring the size of a curved surface, the least squares method can be used to adjust the scanned data to minimize the deviation caused by the geometric error of the scanner or the deformation of the material, ensuring that the final measured size is more accurate. With the support of the comprehensive error model, the least squares method can effectively reduce the propagation of errors, so that the measurement results are more consistent with the geometric characteristics of the actual object, and improve the accuracy and reliability of the curved surface size measurement.

[0049] The fourth step is dynamic adjustment and feedback control, which aims to ensure that all conditions in the measurement process are always maintained in the optimal state by real-time monitoring of environmental changes and dynamically adjusting scanning parameters to improve the accuracy and stability of the measurement. S4.1 first introduces a real-time data acquisition and feedback system to continuously monitor changes in the measurement environment, especially factors such as temperature and humidity, which have a significant impact on the scanning results. For example, fluctuations in temperature and humidity will affect the accuracy and stability of the laser scanner, especially the propagation characteristics of the laser beam, surface reflectivity, and physical properties of the material during the scanning process will change with changes in environmental conditions. Therefore, these factors need to be monitored in real time. By arranging temperature and humidity sensors in the measurement environment, real-time data of environmental parameters can be accurately obtained and passed to the system for processing. Based on the real-time temperature and humidity data, the system identifies environmental changes that affect the measurement accuracy and makes corresponding adjustments. S4.1 involves adjusting scanning parameters, including scanning speed, light source brightness, etc., through a PID controller (proportional-integral-differential controller). The PID controller is a classic feedback control method that is widely used in real-time regulation of dynamic systems. By adjusting the feedback of the measurement error, the control variables are adjusted to make the system stable and reach the optimal working state. In this application, the PID controller adjusts the scanning parameters such as scanning speed and light source brightness according to the real-time data of environmental changes to ensure that the scanning process is always in the best state. Specifically, the PID controller calculates the scanning parameters that need to be adjusted according to the current environmental changes (such as temperature and humidity), and adjusts these parameters through three parts: proportional, integral and differential. The proportional part adjusts the parameters according to the size of the current error, the integral part accumulates past errors to eliminate long-term deviations, and the differential part predicts future error trends and responds in advance. Through this precise control, the PID controller can achieve fast real-time adjustments, thereby maintaining the consistency of scanning accuracy under changing environmental conditions. Finally, through the combined effects of real-time data acquisition, environmental monitoring and PID control, the scanning parameters can be dynamically adjusted during the measurement process, effectively offsetting the impact of the external environment on the measurement accuracy, and ensuring that the results of each scan can be kept within the accuracy requirements.

[0050] Embodiment 1:

[0051] Combined with Figure 2In this embodiment, in the 3D dimension measurement method of the present invention applicable to curved surface structures, intelligent feature recognition and parametric modeling are one of the key technologies, which can accurately extract and model the geometric features of the surface. In particular, a multi-scale analysis strategy is introduced to enable the neural network to extract local feature information at different scales to improve the recognition ability of surface features. In practical applications, a complex metal component surface is set, and there are different curvature areas on the surface, especially at some curved edges and corners. The geometric features of these areas are crucial to the final dimension measurement.

[0052] First, the point cloud data of the target surface is collected by using a multi-view laser scanning device. These data contain detailed geometric information of different areas, and due to errors in the scanning process, there are some irregularities and noise in the surface data. Therefore, it is crucial to use intelligent feature recognition and parametric modeling methods. In order to improve the accuracy of surface feature extraction, a multi-scale analysis strategy is adopted to introduce the gradual extraction of features at different levels in different convolutional layers of the neural network, so that information such as edges, angles and curvatures can be fully recognized at a detailed scale. For example, in high-level convolutional layers, the large-scale features recognized by the neural network include the overall contour and large curvature areas of the surface, while in low-level convolutional layers, they are detailed to local small changes and detailed features. This enables accurate geometric information to be extracted on complex surfaces, especially in some areas where the curvature changes dramatically.

[0053] For the local curvature area of ​​the surface, its second-order derivative is further calculated and adaptively adjusted. This process helps to capture local changes more accurately, especially in places where the surface changes rapidly. The local curvature calculation formula is as follows:

[0054]

[0055] In this formula, C local (x,y) represents the local curvature of the surface at position (x,y), and They represent the second-order derivatives of the surface height function Z(x,y) in the x-direction and y-direction respectively, which are used to calculate the curvature change of the surface. The second-order derivative of the surface reflects the curvature of the surface and can accurately capture the drastic degree of surface change. i is the weight of the local feature, which indicates the influence of each feature in a specific area, and is usually adjusted dynamically according to the geometric characteristics of the surface. For example, in areas of high curvature, the weight value will be larger, while in flat areas it will be smaller. i is the area of ​​region i, which is used to weigh the spatial factors in the local curvature calculation. This area can provide more precise spatial adjustments when the surface changes greatly, thereby further optimizing the calculation of local curvature.

[0056] The measurement is set to be performed on the curved area of ​​a metal plate. First, the point cloud data of the area is obtained by scanning, and a 3D model containing multiple data points is obtained. When processing this data, the neural network extracts the characteristic information of the curved area through multi-scale analysis, especially at the edge of the area, where the curvature changes greatly. Then the local curvature is calculated, and the height function at a certain position (x, y) is set to Z(x, y) = 3x 2 +4y 2 , its second-order derivative is:

[0057]

[0058] Set the weight w of this position 1 =0.8, area A 1 =2, the local curvature is calculated as:

[0059] C local (x,y)=0.8×(6+8)×2=22.4

[0060] This method can accurately identify and adjust the local features of the surface, especially for areas with large changes in curvature. In this way, during the entire 3D measurement process, the fitting parameters of the scanning model can be adjusted according to the changes in local curvature to ensure more accurate measurement results.

[0061] It is necessary not only to accurately capture the surface features, but also to efficiently fit and model the surfaces. In this process, the parametric modeling method of Bezier curves plays a vital role, especially for those surface areas with more complex changes. Suppose that a metal part with multiple surface changes is being measured, some areas have high curvature, while others are relatively flat. In this case, the use of Bezier curves can help better adapt to the geometric characteristics of different areas.

[0062] First, when modeling the surface, use Bezier curves to fit the local area. Bezier curves themselves have good interpolation performance and can accurately describe the geometric shape of the surface. First, each local area is segmented, and then Bezier curves of different orders are selected for fitting according to the curvature characteristics of each area. For high curvature areas, the order of Bezier curves is usually higher in order to more carefully describe the sharp changes in the surface, while for flat areas, the order of Bezier curves is lower, which can simplify calculations and improve efficiency.

[0063] Set to model a region with high curvature, and use a fourth-order Bezier curve to fit the region. The formula of the Bezier curve is:

[0064]

[0065] Among them, S Bezier (u, v) represents the value of the Bezier surface at parameters u and v, that is, the geometric shape of the local surface; P ij are the coordinates of the control points, indicating the specific shape of the surface; and They are Bezier basis functions, the i-th and j-th basis functions in the u and v directions, with orders m and n respectively; m and n represent the orders of the surface in the u and v directions, which are usually adjusted according to changes in surface characteristics.

[0066] When the surface is in a high curvature area, the density of control points will be increased. For example, if the number of control points in a certain area is 16, these control points will be distributed more densely to better fit the surface. The settings of m=4 and n=4 mean that the fourth-order Bezier basis function is used for fitting in the u and v directions respectively. By comparison, the Bezier fitting result set in this area is very consistent with the actual surface, which can greatly reduce the error caused by low-order curve fitting.

[0067] On the other hand, for flat areas, low-order Bezier curves (such as second-order curves) are used for fitting. In such areas, the surface changes little, and low-order curves can meet the needs while reducing the computational complexity. For example, if a second-order Bezier curve is used to fit a flat area, the number of control points will be reduced to 4, and its parameters are set as follows:

[0068]

[0069] In this way, the density of control points can be effectively adjusted dynamically to adapt to different surface features and ensure the accuracy of the fitting results.

[0070] First, the scanned point cloud data is used to extract local features of different scales through the multi-scale analysis strategy in the neural network, and then the appropriate Bezier curve order and control point density are selected according to the changes in the surface area. In areas with large curvature, fine fitting is performed by increasing the density of control points and the order of Bezier curves, while in flat areas, the modeling is simplified by reducing the order and control point density, thereby balancing accuracy and computational efficiency.

[0071] The surface of a metal gear is set to be measured. The edge of the gear has a large curvature, while the tooth surface area is relatively flat. In the edge area, the fourth-order Bezier curve is selected, and the density of control points is increased through the algorithm to make it more accurate in subtle surface changes; in the tooth surface area, the second-order Bezier curve fitting is selected to reduce the amount of calculation while maintaining a relatively accurate modeling result.

[0072] The dynamic control point adjustment mechanism of the Bezier curve is crucial for different surface areas, especially when the geometric features of the surface vary greatly. In the previous steps, we have introduced how to select the appropriate Bezier curve order based on the surface curvature. The key to this stage is to dynamically adjust the number and density of control points according to the angle and curvature changes of the surface area, so as to ensure that the fitting effect of each local area is more accurate and efficient.

[0073] Suppose that a metal mold is being measured in 3D dimensions, which contains multiple curved surface areas of different shapes: the outer contour area of ​​the mold has a large curvature and slight deformation; while the inner surface is relatively flat. In order to deal with this complex surface morphology, a dynamic adjustment mechanism is needed to adjust the density of control points according to local curvature and angle changes. In this process, the density of control points is calculated by analyzing the second-order derivatives of the surface at different positions, and the number and spacing of control points are determined based on the density value, so as to achieve accurate local fitting.

[0074] Specifically, the calculation formula for the local curvature of the surface is:

[0075]

[0076] In this formula, Density controlpoints Indicates the density of control points, reflecting the spacing or density of control points in the surface area; and are the second-order derivatives of the surface in the x and y directions, respectively, indicating the degree of change of the surface in these two directions; and n represents the number of sample points in the calculation area, that is, the total number of local feature points.

[0077] The measurement is set at a certain edge area of ​​the mold with high curvature. It is found that the second-order derivative of this area changes by 0.15 in the x direction and 0.2 in the y direction. By substituting these values ​​into the above formula, the control point density of this area can be calculated:

[0078]

[0079] This means that in this area, relatively dense control points are required to accurately fit the surface shape, thereby ensuring that the modeling of the high curvature part does not produce large errors.

[0080] Next, the density of control points is applied to the actual Bezier curve fitting. Set the fourth-order Bezier curve to be used in this high curvature area, and according to the calculation results of the control point density, set the number of control points to 16. By increasing the density of control points, the slight curvature changes in this area can be better captured. The formula of the Bezier curve is as follows:

[0081]

[0082] Among them, P ij is the position of the control point, and is the Bezier basis function, m and n are the orders of the Bezier curve in the u and v directions respectively. A fourth-order curve is selected for fitting, so the values ​​of m and n are 4 respectively.

[0083] For relatively flat areas inside the mold, the second-order derivative value will be smaller, for example, the second-order derivative in the x direction is 0.05, and the second-order derivative in the y direction is 0.05. According to the formula, the density of the control point is:

[0084]

[0085] In this flat area, the density of control points is high, so a lower-order Bezier curve (such as a second-order curve) can be used for fitting to reduce the amount of calculation. In the fitting process, a second-order Bezier curve was selected, the number of control points was reduced to 4, and the geometric shape of the area was obtained by calculation.

[0086] By dynamically adjusting the density of control points, Bezier curves of different orders are used in different curvature areas to accurately fit the surface. In high curvature areas, dense control points and high-order Bezier curves ensure measurement accuracy; while in flat areas, lower-order Bezier curves and sparse control points reduce computational complexity while still maintaining a relatively accurate modeling effect.

[0087] Ultimately, this method not only improves the overall measurement accuracy, but also effectively reduces the computational burden and completes the 3D dimension measurement of the metal mold.

[0088] Embodiment 2:

[0089] See also Figure 3In this embodiment, the 3D dimension measurement process of a metal mold is used to illustrate how to use a comprehensive error model to evaluate and adjust the combined effects of the scanner, material properties, and environmental factors on the final measurement results. In applications, the complexity of the mold and the changing working environment result in a variety of error sources, including the geometric error of the scanner itself, thermal expansion and elastic changes caused by material properties, and the effects of environmental changes (such as temperature and humidity). Therefore, by establishing a comprehensive error model, these error sources can be effectively combined to optimize the measurement results.

[0090] First, we start with the error E caused by the scanner. The scanner used is a high-precision laser scanner.

[0091] scanner

[0092] After calibration, there are still small geometric errors, which usually include the accuracy error of the equipment and the calibration error of the equipment. For example, the calibration error of the scanner leads to the deviation between the measurement points, and its error is set to ±0.05mm. This error can be given by the resolution of the equipment and the calibration error model. Usually, the hardware error of the scanner under standard test conditions ranges from 0.01mm to 0.1mm.

[0093] Next is the error E caused by material properties material The measurement is set to be an aluminum alloy mold, whose elastic modulus and thermal expansion coefficient are greatly affected by temperature changes. In actual measurement, the ambient temperature fluctuates, and the thermal expansion coefficient of aluminum alloy is approximately α Al =23×10 -6 ℃ -1 , which means that for every 1°C increase, the length of the aluminum alloy will expand by about 0.023 mm. When the temperature of the mold is set to increase from 20°C to 25°C, the thermal expansion error of the material can be calculated by the formula:

[0094] ΔL=L 0 α Al ΔT

[0095] Among them, L 0 =500mm is the initial length of the mold, and ΔT=5℃ is the temperature change. After substituting the data, the thermal expansion error is:

[0096] ΔL=500mm×23×10 -6 ℃ -1 ×5℃=0.0575mm

[0097] This means that under this temperature change, the size of the aluminum alloy mold will increase by about 0.0575 mm.

[0098] Finally, consider the error E caused by environmental factors environment. Changes in temperature and humidity in the environment have a significant impact on the measurement results. The ambient humidity of the measurement is set to change greatly, from 50% to 80%, which causes a slight deformation of the metal material or amplification of the scanning error. For example, humidity changes cause changes in the moisture adsorption of the material, resulting in a slight deviation in the position of the measuring point. The error caused by this humidity change is set to ±0.03mm, which is obtained through modeling of changes in environmental conditions and actual experimental data.

[0099] Now, the total error is estimated by combining the above error sources into a comprehensive error model. According to the given model:

[0100] E total =f(E scanner ,E material ,E environment )

[0101] The error propagation function f is set to be a linear weighted sum, where the weight of each error source depends on their contribution to the final measurement. The weight of the scanner error is set to 0.6, the weight of the material error is set to 0.3, and the weight of the environmental error is set to 0.1. In this way, the total error can be expressed by the following formula:

[0102] E total =0.6·E scanner +0.3·E material +0.1·E environment

[0103] After substituting the values, we get:

[0104] E total =0.6·0.05mm+0.3·0.0575mm+0.1·0.03mm

[0105] =0.03mm+0.01725mm+0.003mm=0.05025mm

[0106] Therefore, the final comprehensive error is about 0.05025mm. This means that after comprehensively considering the scanner error, the thermal expansion error of the material, and the influence of the change in environmental humidity, the error of the measurement result is about 0.05025mm. In this way, the errors caused by different factors can be more accurately estimated and corrected, thereby improving the overall measurement accuracy.

[0107] In order to further improve the accuracy of the model, experimental data will be introduced to dynamically adjust the error propagation function under different materials and environmental conditions, so as to achieve adaptability to different measurement conditions. This process uses regression analysis to fit the error propagation function so that the model can dynamically reflect the error performance under different conditions.

[0108] The mold dimensions of three different materials (such as aluminum alloy, stainless steel, and carbon fiber) were set and measured. These materials behave differently under environmental factors such as temperature and humidity. In order to quantify the impact of each factor on the final error, multiple experimental measurements were conducted on each material, and error data under different environmental conditions (such as temperature between 20°C and 30°C, humidity between 40% and 80%) were collected. These experimental data will provide the model with real error performance, which in turn supports the fitting of the error propagation function.

[0109] In model fitting, the error propagation function takes the form:

[0110] ΔD=α 1 ·E scanner +α 2 ·E material +α 3 ·E environment +∈

[0111] Among them, ΔD represents the measurement deviation, that is, the difference between the corrected measurement result and the true value; α 1 , α 2 and α 3 is the regression coefficient, obtained by fitting the experimental data, representing the influence of each error source on the measurement deviation. scanner 、E material and E environment denote the errors caused by scanner, material and environmental factors respectively, while ∈ denotes the fitting error, which reflects the deviation between the error propagation function and the actual experimental data.

[0112] After experimental fitting, the following regression coefficient value range is obtained: α 1 Between 0.5 and 0.7, α 2 Between 0.2 and 0.4, α 3 Between 0.1 and 0.2. The different ranges of these values ​​reflect the contribution of each error source to the final measurement error. Assume that in a certain measurement, the geometric error of the scanner is 0.06mm, the thermal expansion error of the material is 0.04mm, and the error caused by ambient humidity is 0.02mm.

[0113] According to the error propagation function form of experimental fitting, the data are substituted into the formula and the intermediate value is selected for calculation. 1 =0.6, α 2 =0.3, α 3 =0.15, then the total error deviation is:

[0114] ΔD=0.6·0.06mm+0.3·0.04mm+0.15·0.02mm=0.036mm+0.012mm+0.003mm

[0115] =0.051mm

[0116] In this way, the total measurement error deviation can be obtained to be 0.051 mm. This means that the error propagation function after fitting the experimental data under different conditions can dynamically adjust the error evaluation ability of the model under various environmental and material conditions.

[0117] In the application of 3D dimension measurement of metal molds mentioned above, the contribution of different error sources to the measurement deviation has been estimated by fitting the error propagation function and the regression coefficient, and the total error deviation is 0.051mm. Next, these data will be used for dimension correction, and the least squares method will be introduced to further optimize the results so that the measured value is closer to the true size.

[0118] In this step, the dimensional correction is based on minimizing the sum of squared errors between the model predictions and the actual measurements to optimize the parameters. The measured values ​​are adjusted by the least squares method, whose goal is to minimize the sum of squared errors of all experimental data points. The least squares method can help find the best measurement results, minimize the comprehensive error, and thus improve the measurement accuracy.

[0119] In this series of measurements, n experimental data points are obtained under different environmental conditions and material types, and the uncorrected measurement result of each data point is D i According to the model of the error propagation function, the predicted value It can be expressed as:

[0120]

[0121] Among them, α 1 , α 2 , α 3 is the regression coefficient obtained through experimental fitting, and ∈ is the fitting error. By comparing these predicted values ​​with the actual measured values, the least squares objective function can be established:

[0122]

[0123] In this formula, is the corrected dimensional measurement result, the goal is to minimize the sum of square errors of all experimental data points. i is the original data point of the experimental measurement, the measurement result without error correction, are the predicted values ​​calculated using the error propagation function. These values ​​are adjusted using the least squares method to minimize the error.

[0124] The collected measurement data is set to include the following samples:

[0125] Data Point 1: D 1=100.12mm

[0126] Data Point 2: D 2 =100.08mm

[0127] Data Point 3: D 3 =100.15mm

[0128] At each data point, the corresponding predicted value can be obtained through the previously fitted error propagation function. The predicted values ​​for these data points are calculated as follows:

[0129]

[0130] Next, calculate the least squares error sum of squares:

[0131] Error sum of squares = (100.12-100.10) 2 +(100.08-100.07) 2 +(100.15-100.12) 2

[0132] =(0.02) 2 +(0.01) 2 +(0.03) 2 =0.0004+0.0001+0.0009=0.0014

[0133] By minimizing the sum of squared errors, the optimal size correction value can be found, making the overall measurement result more accurate. This process can be continuously iterated and optimized through new experimental data, so that the accuracy of the model under different conditions can be further improved. The use of the least squares method not only improves the accuracy of the measurement data, but also further reduces the deviation caused by different error sources, providing a reliable basis for the final 3D size measurement. Therefore, by using the least squares method to correct the measurement value, the measurement result can be closer to the actual size.

[0134] The above shows and describes the basic principles, main features and advantages of the present invention. It should be understood by those skilled in the art that the present invention is not limited to the above embodiments, and the above embodiments and descriptions are only for explaining the principles of the present invention. Without departing from the spirit and scope of the present invention, the present invention may have various changes and improvements, which fall within the scope of the present invention to be protected. The scope of protection of the present invention is defined by the attached claims and their equivalents.

Claims

1. A 3D dimension measurement method suitable for curved surface structures, characterized in that The following steps are involved: S1. Multi-view surface reconstruction: S1.1, use camera and laser scanner to capture the target surface; use structured light to generate point cloud data; S1.2, using SIFT feature matching algorithm, point cloud registration is performed between multiple views to generate a unified point cloud model; S2. Intelligent feature recognition and parametric modeling: S2.

1. Use deep learning algorithms to train feature recognition models and extract key geometric features of surfaces. S2.

2. Identify surface features including edges, angles, and curvatures by labeling and classification, and convert the features into mathematical representations using the parametric modeling method of Bezier curves; S3. Dimension measurement based on error model: S3.

1. Establish a comprehensive error model to evaluate multiple error sources, including the geometric error of the scanner itself, deformation caused by material properties, and the influence of environmental factors; fit the error propagation function through experimental data; S3.2, use the least squares method to adjust the size measurement; S4, dynamic adjustment and feedback control: S4.

1. Introduce a real-time data acquisition and feedback system, use temperature and humidity sensors to monitor environmental changes, and adjust scanning parameters including scanning speed and light source brightness through a PID controller.

2. The 3D dimension measurement method applicable to curved surface structures according to claim 1, characterized in that The intelligent feature recognition and parameterized modeling include: A multi-scale analysis strategy is introduced to enable the neural network to extract local feature information at different scales. The multi-scale strategy gradually extracts features at different levels by introducing convolutional layers, so that the edge, angle and curvature information in the surface can be identified at a detailed scale. For the local curvature area of ​​the surface, the second-order derivative is calculated and adaptively adjusted to identify the local surface features. The local curvature is calculated by the following formula: Among them, C local (x,y) represents the local curvature of the surface at position (x,y); Represents the second-order derivative of the surface height function Z(x,y) in the x direction, which is used to calculate the curvature change of the surface; represents the second-order derivative of the surface height function Z(x,y) in the y direction; w i The weight of a local feature indicates the influence of the feature in a specific area and is usually adjusted dynamically according to the geometric characteristics of the surface. i The area of ​​region i, used to weigh spatial factors in the local curvature calculation.

3. The 3D dimension measurement method applicable to curved surface structures according to claim 2, characterized in that The intelligent feature recognition and parameterized modeling include: The Bezier curve is selected for local surface fitting; the density of control points is dynamically adjusted according to different requirements of surface features; high-order Bezier curves are used to fit surface changes; and low-order Bezier curves are used to simplify modeling in flat or small curvature areas; the parametric modeling method of Bezier curves is used to interpolate and fit the control points in each local area through the following formula: Among them, S Bezier (u,v) represents the value of the Bezier surface at parameters u and v, that is, the geometric shape of the local surface; P ij Represents the control points of the Bezier surface, whose positions and numbers are adaptively adjusted according to the surface characteristics; in high curvature areas, the density of control points will be greater, while in flat areas, the control points are sparse; represents the Bezier basis function, the i-th basis function on parameter u, with order m; represents the Bezier basis function, the jth basis function on parameter v, with order n; m and n represent the order of the surface in the u and v directions, respectively, and are adjusted according to changes in surface characteristics.

4. The 3D dimension measurement method applicable to curved surface structures according to claim 3, characterized in that The intelligent feature recognition and parameterized modeling include: By analyzing the changes in angles and curvatures in different areas of the surface, the number and density of control points are dynamically adjusted; the mechanism of dynamic control point adjustment can be expressed by the following formula: Among them, Density controlpoints Indicates the density of control points, that is, the spacing or density between control points; It represents the second-order derivative of the surface height function Z(x,y) in the x direction, reflecting the degree of change of the surface in the x direction; It represents the second-order derivative of the surface height function Z(x,y) in the y direction, reflecting the degree of change of the surface in the y direction; n represents the number of sample points in the surface calculation area, which is the total number of local feature points in the area.

5. The 3D dimension measurement method applicable to curved surface structure according to claim 1, characterized in that The dimension measurement method based on the error model includes: establishing a comprehensive error model to integrate the joint influence of the scanner, material and environment on the measurement result; the model is expressed in the following form: AND total =f(E scanner ,AND material ,AND environment ) Among them, E total Represents the final comprehensive error, which indicates the deviation between the measurement result and the true value; E scanner It is the geometric error caused by the scanner, which comes from the hardware problem of the scanner and improper calibration. material Represents the error caused by material properties, including thermal expansion and elastic modulus changes of materials at different temperatures; E environment is the error caused by environmental factors, including the impact of changes in environmental conditions such as temperature and humidity on the measurement; f(·) is the error propagation function, which is used to describe the interaction between different error sources and their contribution to the final measurement error.

6. The 3D dimension measurement method applicable to curved surface structures according to claim 5, characterized in that The dimension measurement method based on the error model comprises: The experimental data provide error performance under different conditions including different materials and different environments, so that the error propagation function can be dynamically adjusted according to the actual measurement conditions; through regression analysis, the form of the error propagation function is fitted, which is expressed as: ΔD=α1·E scanner +α2·E material +α3·E environment +∈ Among them, ΔD represents the deviation of dimensional measurement, that is, the difference between the corrected measurement result and the true value; α1, α2, α3 are regression coefficients, which represent the influence of scanner, material and environmental factors on the measurement deviation respectively; the regression coefficients are obtained by fitting the experimental data to quantify the contribution of each error source to the final error; E scanner ,E material ,E environment They represent the errors caused by scanner, material and environmental factors respectively; ∈ is the fitting error, which represents the deviation between the error propagation function and the actual experimental data.

7. The 3D dimension measurement method applicable to curved surface structures according to claim 6, characterized in that The dimension measurement method based on the error model includes: obtaining an error propagation function, fitting a regression coefficient according to experimental data, and then using the data to perform dimension correction; optimizing parameters by minimizing the sum of square errors between model prediction values ​​and actual measurement values; and adjusting the dimension measurement value by the least squares method.

Images