Virtual pre-assembly method of steel structure components based on 3D laser scanning

Abstract

The present invention relates to a virtual pre-assembly method for steel structure components based on three-dimensional laser scanning. An adaptive hierarchical scanning strategy is designed, and the parameters of the scanning equipment are dynamically adjusted to obtain data from each area of ​​the component; abnormal points are detected and eliminated by fusing the scanning data at different levels; reference points are extracted and coordinates are calibrated, and the data at each level are spliced ​​to form a three-dimensional point cloud model of the entire component; the point cloud is classified using a multidimensional data analysis algorithm; a closed-loop data verification system is established; a multiple feature point matching algorithm is designed; docking and position calibration are performed through algorithm iteration; the spatial deviation between components during the assembly process is quantitatively analyzed; after the deviation is detected, a fine-tuning plan is generated and the component position is adjusted; a virtual pre-assembly path is generated according to the shape, connection method and installation order of the component, and the assembly order and spatial position layout are controlled; the assembly status of each component is monitored in real time, and the path is optimized and adjusted through a feedback mechanism.

Description

Technical Field

[0001] The invention relates to the field of virtual pre-assembly of steel structure components, and in particular to a virtual pre-assembly method of steel structure components based on three-dimensional laser scanning. Background Art

[0002] At present, the virtual pre-assembly method of steel structure components based on 3D laser scanning plays an important role in industrial applications, but it still has many shortcomings and disadvantages, which are specifically reflected in key links such as data collection, feature recognition, and assembly path optimization, which affect the accuracy and efficiency of pre-assembly.

[0003] First, in terms of data acquisition, current methods usually rely on fixed scanning parameters, such as resolution, angle, and scanning frequency, but lack dynamic adaptive adjustment capabilities, and cannot flexibly adjust the parameters of the scanning equipment according to the shape, complexity, and position of the component. This fixed mode easily leads to poor data quality of detailed features (such as welds, holes, etc.), especially when facing components with different materials, surface smoothness, or high structural complexity, the noise problem of reflection or light spot is prominent, resulting in a large number of invalid points and noise points in the point cloud data, which increases the difficulty and calculation amount of subsequent data processing. The data fusion and denoising methods in the prior art are difficult to accurately identify and remove noise points, especially on metal components. Due to the serious surface reflection phenomenon, the noise points are confused with the actual feature points, affecting the accuracy of the assembly model. In addition, the existing methods also have deficiencies in multi-level splicing and coordinate calibration. When processing the data of multiple assembled components, due to inaccurate reference point identification, errors accumulate during the splicing process, which ultimately affects the integrity of the virtual model.

[0004] Secondly, in terms of feature recognition and parametric modeling of steel structure components, existing methods rely on traditional edge detection, feature template matching and other algorithms, but lack multi-dimensional data analysis of complex features, making it difficult to adapt to the feature recognition of components with complex geometric structures. In particular, when facing steel structures with many curved surfaces and complex shapes, traditional edge detection methods have limited recognition capabilities for curvature and cannot accurately distinguish feature areas from non-feature areas on complex shapes; feature template matching is also only applicable to templates of specific standard shapes, and the automatic recognition effect for non-standard holes or welds is poor, resulting in low accuracy in parametric modeling. Traditional feature recognition methods are usually unable to handle nonlinear and non-uniformly distributed features in the data, especially in scenes with more noise, and are prone to recognition errors. In addition, during the recognition process, traditional methods are usually unable to dynamically adjust the size and shape of the feature template. When the size and angle of the component change, the adaptability of feature recognition is poor, which affects the accuracy of the model and the assembly effect.

[0005] In terms of assembly path optimization and adjustment, existing methods usually use static path planning and lack the ability to optimize dynamic paths based on real-time feedback mechanisms. The assembly process of steel structure components is complex and large, and the assembly order and position layout between components are crucial to the stability of the final structure. However, existing technologies usually ignore the installation order and possible spatial limitations that may occur during the assembly process when generating paths. In particular, when the components are large or complex in shape, path planning fails to dynamically adjust according to the connection method of the components, resulting in a large amount of manual adjustment and re-planning during the assembly process, reducing assembly efficiency. Summary of the invention

[0006] The purpose of the present invention is to provide a virtual pre-assembly method for steel structure components based on three-dimensional laser scanning, thereby solving some of the drawbacks and shortcomings pointed out in the background technology.

[0007] The present invention solves the above-mentioned technical problems by adopting the following technical solutions: A virtual pre-assembly method of steel structure components based on three-dimensional laser scanning, comprising: S1, adaptive hierarchical acquisition of three-dimensional laser data of steel structure components:

[0008] S1.1. Design an adaptive layered scanning strategy and dynamically adjust the parameters of the scanning equipment, including resolution, angle, and scanning frequency, to obtain data from each area of ​​the component;

[0009] S1.2, by fusing the scan data at different levels, detect and remove abnormal points, including the noise point cloud caused by surface reflection;

[0010] S1.3. After completing the layered data collection, extract the reference points and perform coordinate calibration, and splice the data at each level to form a three-dimensional point cloud model of the entire component;

[0011] S2. Intelligent feature recognition and parametric modeling of steel structure components:

[0012] S2.1. Use multi-dimensional data analysis algorithms to classify point clouds and identify key features of components, including connection holes, weld lines, and overlap areas, through edge detection and curvature analysis methods;

[0013] S2.2. Establish a closed-loop data verification system to control the high matching between the parametric model and the actual component through dynamic comparison between the scanned data and the model data;

[0014] S3. Virtual pre-assembly positioning based on intelligent matching algorithm:

[0015] S3.1. Design a multiple feature point matching algorithm to align key points in the model including bolt holes and lap joints; perform docking and position calibration through algorithm iteration;

[0016] S3.2, quantitatively analyze the spatial deviation between components during assembly; after detecting the deviation, generate a fine-tuning plan and adjust the position of the components;

[0017] S4. Dynamic optimization and adjustment of virtual pre-assembly path:

[0018] S4.1. Generate a virtual pre-assembly path based on the shape, connection method and installation sequence of the components to control the assembly sequence and spatial position layout; when pre-assembling large components, give priority to assembling the main beam and supporting structure;

[0019] S4.2, monitor the assembly status of each component in real time, and optimize and adjust the path through the feedback mechanism;

[0020] S4.3. Add a feasibility assessment module to simulate construction site conditions, including space limitations or equipment failures, and generate emergency adjustment plans in advance;

[0021] S5. Virtual assembly accuracy prediction and dynamic error compensation:

[0022] S5.1. Based on the virtual assembly model and historical assembly data, an error prediction model is established using a machine learning algorithm to predict assembly deviations.

[0023] S5.2. Design a real-time compensation algorithm to dynamically adjust the component positions or connection point positions in the virtual model to control the high degree of fit between the virtual assembly and the actual assembly state.

[0024] Furthermore, the intelligent feature recognition and parametric modeling method of the steel structure component:

[0025] A deep learning network is used to process the spatial position, density and color multi-dimensional attributes of point clouds through self-supervised learning, and adaptive labels are generated through point cloud data to set intelligent classification functions:

[0026]

[0027] Among them, f(x, y, z) represents the classification function, which is used to classify each point according to the spatial coordinate position of the point cloud, where x, y, and z represent the three-dimensional coordinates of the point cloud respectively; the parameter α i is a weighting factor related to the classification result, which is used to control the influence weight of different feature points, and β i is the attenuation coefficient, which is used to adjust the influence range of different features in space; exponential function part Control the distribution of point cloud density in spatial position.

[0028] Furthermore, the intelligent feature recognition and parametric modeling method of the steel structure component:

[0029] Through multi-scale curvature analysis, the curvature changes at different scales are detected in layers to identify the weld and hole characteristics on the structural parts. In each layer, the sudden changes on the component surface are marked by identifying the local curvature differences. The feature detection formula is defined as follows:

[0030]

[0031] Among them, g(x,y,z) represents the curvature analysis function used for edge detection, which is used to determine whether a point is an edge where the surface changes suddenly; the first part of the formula The second-order derivative in the y direction is calculated, and the partial derivative in the x direction is used to measure the change in curvature of the surface. The second part It is the second-order derivative in the x direction, and the partial derivative in the y direction.

[0032] Furthermore, the intelligent feature recognition and parametric modeling method of the steel structure component:

[0033] The dynamic template matching and region growing algorithm are introduced to adaptively adjust the template size and shape to adapt to components of different sizes and angles. In the process of region growing, the boundary of the overlapping area is refined by an adaptive expansion method, and the edge of the overlapping area is fine-tuned. The formula of the region growing and thinning algorithm is:

[0034]

[0035] Among them, h(x,y) represents the regional characteristic function of adaptive expansion, which is used to calculate the edge expansion result of the target area; φ(x,y) in the function represents the edge expansion coefficient of the area, which is obtained by integrating ∫ A φ(x,y)dx dy is used to accumulate the expansion results, A is the range of the expansion area; the parameter γ is the growth rate coefficient, which controls the speed of boundary expansion, and δ j represents the boundary adjustment factor, which is used to adapt to boundary changes when refining the region; finally, θ j It is the angle that controls the direction of the boundary and is used to make angle corrections based on the geometry of the area.

[0036] Furthermore, the virtual assembly accuracy prediction and dynamic error compensation method:

[0037] By analyzing historical assembly data, the key factors affecting the error are identified, including the shape of the component, assembly sequence, material properties, ambient temperature and mechanical accuracy; the factors are used as feature input models for deep learning to create a feature mapping space; the feature processing function F(x 1 ,x 2 ,…,x n ), where each feature x i Indicates variables that affect the error, including component shape or mechanical accuracy:

[0038]

[0039] Among them, x 1 ,x 2 ,…,x n Represents various characteristic variables in the assembly process, including component size, number of connection points or environmental parameters; γ i is a weight parameter used to adjust the importance of each feature in the overall error prediction; η i is the attenuation coefficient, which controls the influence range of each feature on the error; C is a constant term, which is used to smooth the eigenvalue range.

[0040] Furthermore, the virtual assembly accuracy prediction and dynamic error compensation method:

[0041] The assembly error is modeled using the regression model in machine learning. The relationship between the feature variables and the assembly error is identified and learned by learning the patterns in the historical data. The error prediction value P(y|F(x)) is described by the integral formula to capture the nonlinear relationship between the feature variables and the error:

[0042]

[0043] Among them, y represents the predicted value of the assembly error; F(x) is the feature mapping result obtained by the feature processing function, which is used to integrate the influence of each feature on the error; λ is the linear adjustment parameter, which is used to adjust the direct influence of the feature mapping on the error; θ is the nonlinear adjustment coefficient, which controls the amplitude of the sine term; k is the frequency coefficient, which adjusts the periodicity of the sine term.

[0044] Furthermore, the virtual assembly accuracy prediction and dynamic error compensation method:

[0045] Combining the virtual assembly model and real-time data feedback, the assembly parameters, including component position or assembly order, are adjusted according to the predicted deviation during the assembly process. The error compensation function T(z′) is defined, and the assembly parameters are dynamically adjusted in a timely manner according to the real-time error value z′:

[0046]

[0047] Among them, z′ is the error deviation currently detected, which is used to provide real-time feedback on the deviation in the current assembly process; σ is the proportional coefficient, which controls the overall strength of the compensation; φ(t) is the adjustment function, which performs smooth adjustments within the error range through integration; δ is the compensation amplitude coefficient, which is used to control the adjustment amplitude; ω is the compensation frequency parameter, which is used to adjust the frequency of the cosine term.

[0048] The virtual pre-assembly method of steel structure components based on three-dimensional laser scanning of the present invention has significant beneficial effects, which are specifically manifested as follows: First, the method obtains accurate point cloud data of components through three-dimensional laser scanning technology, combines adaptive layered scanning and multi-dimensional data analysis algorithms, so that the geometric features and details of steel structure components can be completely captured and clearly presented, providing a data basis for high-precision virtual assembly. Secondly, the intelligent feature recognition and parametric modeling technology introduces advanced algorithms such as edge detection, curvature analysis, and dynamic template matching, which can automatically identify key features (such as welds, holes, overlap areas, etc.), and establish parametric models through deep learning to ensure that the virtual model is highly consistent with the actual components. In the virtual assembly positioning, the alignment and iterative calibration of multiple feature points based on the intelligent matching algorithm significantly improves the accuracy and efficiency of virtual assembly, ensuring that the key connection parts can be accurately docked. In addition, the method solves the dynamic adjustment problem in the assembly path planning of large and complex steel structures through real-time path optimization, feedback mechanism and emergency plan, so that virtual assembly can simulate the actual situation of the construction site, and effectively reduces the assembly interruption and rework caused by on-site space limitations or emergencies. In terms of assembly accuracy prediction and error compensation, this method introduces an error prediction model and a real-time compensation algorithm. By analyzing historical assembly data and combining it with real-time data feedback, it can dynamically predict and adjust assembly deviations to ensure a high degree of fit between virtual assembly and actual assembly status. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] Figure 1 This is a flow chart of the virtual pre-assembly method of steel structure components based on three-dimensional laser scanning of the present invention.

[0050] Figure 2 The present invention is a flow chart of the intelligent feature recognition and parametric modeling method of steel structure components.

[0051] Figure 3 This is a flow chart of the virtual assembly accuracy prediction and dynamic error compensation method of the present invention. DETAILED DESCRIPTION

[0052] The specific implementation modes of the present invention will be described in detail below in conjunction with the accompanying drawings.

[0053] Combined with Figure 1Flowchart steps, the present invention is based on the virtual pre-assembly method of steel structure components based on three-dimensional laser scanning, the first step is the adaptive layered acquisition of three-dimensional laser data of steel structure components. S1.1 Design an adaptive layered scanning strategy, the core of which is to dynamically adjust the parameters of the scanning equipment to ensure that all areas of the steel structure components can be fully covered. During the scanning process, the resolution, scanning angle and scanning frequency are adjusted in real time according to the complexity of the component and the required fineness to achieve the best scanning effect. For example, in some areas with complex geometric shapes, the scanning resolution will be increased to capture more details; while in relatively flat areas, the resolution can be reduced accordingly to save scanning time and data processing resources. Through this adaptive adjustment, more complete and clear component surface data can be obtained. S1.2 In the data acquisition process, data fusion at different levels is a key step. Since the surface of the steel structure will cause the generation of noise point clouds due to the characteristics of smoothness and reflection, in order to ensure the accuracy of the data, it is necessary to fuse the scanning data of different layers. By comparing and calculating the multi-layer data, the system can detect the position of abnormal point clouds and remove them, such as removing redundant noise points in areas with severe reflection. After this processing, the remaining point cloud data can more realistically reflect the surface shape of the steel structure, eliminating deviations caused by reflection or other external factors. S1.3 After completing the acquisition of all levels of scanning data, extract the reference points and perform coordinate calibration. The extraction of reference points is to establish an alignment relationship between different scanning layers. Usually, specific positions or geometric features with clear shapes on steel structure components are selected as reference points. Through these reference points, the coordinates of the scanning data at each level can be calibrated to eliminate the errors between different scanning layers. After the calibration is completed, the splicing algorithm is used to synthesize the data at all levels to generate a complete three-dimensional point cloud model. This model accurately reflects the spatial form of steel structure components and provides basic data for subsequent virtual pre-assembly.

[0054] The second step is intelligent feature recognition and parametric modeling of steel structure components. S2.1 uses a multidimensional data analysis algorithm to classify point cloud data to identify key features in steel structure components. Through edge detection and curvature analysis methods, the system can extract key areas of components from point cloud data, such as connection holes, weld lines, and overlap areas. These areas are usually important locations in the assembly process of steel structures, so they need to be identified and processed in the virtual assembly process. The edge detection algorithm can accurately identify the boundary areas in the point cloud data, while the curvature analysis method can identify locations with large changes in surface morphology. By combining these two methods, the system can accurately locate and mark these key features to provide data support for parametric modeling. S2.2 After completing feature recognition, the system will then establish a closed-loop data verification system, which dynamically compares the scanned data and the parametric model data to ensure that the virtual model is highly matched with the actual component. In this closed-loop verification process, the system will continuously detect and update the accuracy of the parametric model, compare the scanned data with the model at multiple levels, and dynamically adjust the parameters of the model to ensure that the real features of the component are fully and accurately represented in the virtual model. If deviations are detected, the system will make corresponding corrections until the accuracy of the parametric model meets the requirements. In this way, the parametric model can fully and truly reflect the actual size and shape of the component, making the subsequent virtual pre-assembly more accurate and efficient, avoiding assembly errors caused by model deviations, and improving the accuracy and quality of the overall assembly.

[0055] The third step is virtual pre-assembly positioning based on the intelligent matching algorithm. In this process, S3.1 designed a multiple feature point matching algorithm to align key points in the model, such as bolt holes and overlapping surfaces. Bolt holes and overlapping surfaces are important locations for connecting components during the assembly process, so their precise alignment is the key to ensuring successful assembly. The core of the multiple feature point matching algorithm is to use an iterative algorithm to perform step-by-step docking and position calibration by analyzing the position and angle characteristics of these key points. In each iteration, the algorithm detects and adjusts the relative positions of the key points to ensure that the positioning of these key points in the model is completely consistent with that of the actual components. This iterative calibration process can eliminate subtle deviations and gradually achieve precise alignment of key points. As the iterations proceed, the matching accuracy gradually improves, ultimately forming a highly accurate virtual assembly positioning.

[0056] S3.2 After completing the key point matching, the system will further quantify the spatial deviations between components during the assembly process. Since components may have slight deviations such as displacement and rotation during the assembly process, quantitative analysis is a necessary step to ensure the accuracy of component position. In this analysis, the system compares the current spatial position of the component with its target position and accurately calculates the size and direction of the position deviation. The key to this process is to generate a fine-tuning plan after detecting the deviation to eliminate the impact of these deviations on the overall assembly accuracy. When a position deviation of a component is detected, the system will generate a fine-tuning plan, which includes the specific adjustment amount and direction. Through this plan, the system can make small adjustments to the position of the component until the position of the component accurately matches the designed position. This virtual pre-assembly positioning method based on the intelligent matching algorithm not only ensures the high-precision alignment of key connection points, but also can correct the slight deviations that occur during the assembly process in real time, ensuring that the accuracy of the virtual pre-assembly reaches the standard of actual construction, thereby effectively improving the quality and efficiency of the overall assembly.

[0057] The fourth step is the dynamic optimization and adjustment of the virtual pre-assembly path. In this stage, S4.1 generates a virtual pre-assembly path based on the shape, connection method and installation sequence of the components. The generation of the pre-assembly path takes into account the geometric shape of the components, the location of the connection points and the preset assembly sequence to ensure that the components can be arranged and installed in the best way during assembly. For example, when pre-assembling a large steel structure, in order to ensure the stability and accuracy of the structure, the system gives priority to assembling the main beam and the supporting structure, which can provide a stable support for other components in the early stage and reduce the cumulative error caused by subsequent assembly. S4.2 In order to ensure the feasibility of the path in actual operation, the system monitors the assembly status of each component in real time. This real-time monitoring continuously collects the actual assembly position and status data of the components through a feedback mechanism and compares it with the preset path. When deviations or unexpected situations are detected during the assembly process, the system can instantly optimize and adjust the assembly path based on the feedback results to ensure that the assembly process is carried out according to the precise designed route. The S4.3 system has added a feasibility assessment module to simulate the conditions in real construction scenarios in a virtual environment, including potential problems such as space limitations and equipment failures. The feasibility assessment module can detect these potential problems in advance in a virtual environment and generate emergency adjustment plans for each situation. For example, if it is detected that the construction space is too narrow or the equipment fails, the system will generate alternative assembly paths or provide new assembly sequence suggestions so that operators can flexibly adjust according to the actual situation and reduce the impact of unexpected factors on the assembly process.

[0058] The fifth step is virtual assembly accuracy prediction and dynamic error compensation. In this step, S5.1 predicts assembly deviations by building an error prediction model. Specifically, the system uses machine learning algorithms to train based on the virtual assembly model and a large amount of historical assembly data to generate an error prediction model. The model can identify factors that cause errors in the assembly process, such as small deviations in component size and accumulated errors at connection points. The machine learning model learns the error patterns in historical data and understands the impact of various factors on assembly accuracy under different conditions, so that potential assembly deviations can be predicted before the actual assembly begins. This error prediction capability enables the system to identify assembly problems in advance and provide guidance for subsequent error adjustments. S5.2 After completing the error prediction, the system further designs a real-time compensation algorithm to ensure a high degree of consistency between the virtual assembly model and the actual assembly state. The core of the real-time compensation algorithm is to dynamically adjust the component position or connection point position in the virtual model. In particular, after the deviation is found, the algorithm can adjust the relevant parameters of the virtual assembly model according to the size, direction and position of the deviation to minimize the impact of the error on the overall assembly accuracy. For example, if the actual position of a component is found to deviate from the expected position during real-time assembly, the compensation algorithm will calculate the appropriate adjustment amount and direction, and update the component position of the virtual model in real time, so that the virtual assembly state is consistent with the actual state.

[0059] Embodiment 1:

[0060] Combined with Figure 2 In this embodiment, in a large steel structure project, the engineering team adopted a virtual pre-assembly method based on three-dimensional laser scanning to ensure the high-precision assembly of complex steel structure components. Due to the complexity of the project and the diversity of components, we are faced with the challenge of how to accurately identify and model the key features of the components. For this reason, intelligent feature recognition and parametric modeling methods are used to accurately classify and identify key features in point cloud data. Specifically, through a deep learning network, a self-supervised learning method is used to process the point cloud data generated by laser scanning, and adaptive labels are generated based on multi-dimensional attributes such as spatial position, density and color of the point cloud. In order to achieve accurate classification, an intelligent classification function is set:

[0061]

[0062] In this formula, f(x, y, z) represents the function used for classification, the purpose of which is to classify each point cloud point to identify the key feature points of the component. Here, x, y, and z represent the three-dimensional coordinates of the point cloud, respectively, and the parameter α i is a weighting factor related to the classification result, which controls the weight of different feature points in the classification, and β iis the attenuation coefficient, which adjusts the influence range of the feature in space. To ensure the effectiveness of the classification function, the parameter α i and β i A specific value range is set, where α i The value of is between [0.1, 1.0] to ensure that different feature points have appropriate influence weights to reflect their importance in the component structure; and β i The value range of is set to [0.05, 0.5], which is used to control the distribution of point cloud density in space so as to effectively identify important component features. Through this intelligent classification function, the point cloud data is divided into multiple different feature areas, such as weld areas, connection holes, etc.

[0063] In the project, complex steel structure components were scanned and 3D point cloud data was generated, which contained millions of points. In order to facilitate analysis and feature extraction, the scan area was divided into multiple sub-areas. For example, in the weld area, α was set according to the formula i = 0.8 and β i = 0.3, ensuring that the point cloud density in this area is higher, thus highlighting the characteristics of the weld during classification. For flat component surface areas, set α i = 0.3 and β i =0.1 to reduce the weight of this area and give priority to key parts during classification. By substituting these actual data calculation results, the classification function will generate adaptive labels for each point. The system will clearly mark key features such as welds, holes and overlap areas, and complete parametric modeling, laying the foundation for subsequent virtual pre-assembly.

[0064] For example, when scanning a main beam, it is found that the point cloud data density of the weld area is high.

[0065]

[0066] Set α separately 1 =0.8, β 1 = 0.3 for weld features, α 2 =0.4, β 2 = 0.2 is used for the plane area features, resulting in a clear classification of the weld and other areas. The exponential function in the formula This ensures that the distribution of point clouds in space gradually decreases with distance, so that areas with dense point clouds (such as welds) are marked with higher accuracy in the virtual assembly model. Substituting the above parameters into the calculation results, not only can the specific location of each weld and hole be clearly marked in the virtual model, but also high-quality basic data support can be provided for subsequent assembly path optimization and precision control.

[0067] In this steel structure project, the engineering team further used a multi-scale curvature analysis method to detect curvature changes at different scales in layers, so as to accurately identify key features such as welds and hole positions. Due to the complexity of steel structures, the feature points in the point cloud data contain different degrees of surface changes. By setting multi-scale detection layers, it can adapt to the distribution of different curvature features. In each layer, the feature detection formula is applied:

[0068]

[0069] To detect whether there is a sudden change in curvature on the surface of the component. The main function of this formula is to identify edge features in the point cloud, especially for complex steel structure surface feature recognition.

[0070] g(x,y,z) represents the curvature analysis function, which is used to determine whether a point is located at a surface mutation point. The first part of the formula By calculating the second-order derivative in the y direction and taking the partial derivative in the x direction, the changing trend of the surface can be captured, so that local protrusions or depressions such as welds can be detected. The second-order derivative in the x direction is used to perform partial derivative in the y direction to identify the boundary between relatively smooth areas and edge areas. Through the comprehensive calculation of these two parts, the system can effectively identify areas with different curvatures.

[0071] When applying this method for actual inspection, a certain supporting component was scanned and point cloud data containing millions of points were obtained. In the weld area, the key parameters for curvature recognition are set to highlight the curvature change of the weld. The position point cloud of x=0.5 and y=0.7 is set for inspection, and the local curvature value of the weld position can be obtained by substituting it into the formula. The partial derivative value range of the inspection is further set, where the coefficient range of the derivative term is between [0.01,1.0] to adapt to the curvature changes in different areas and ensure that the curvature mutation points in the weld area are accurately marked. Due to the large curvature changes of features such as welds, the values ​​of g(x,y,z) in these areas are usually significantly higher than those in smooth areas. In this way, key positions such as welds and holes can be clearly marked in the virtual model, providing more detailed reference data for component parametric modeling and subsequent assembly path optimization.

[0072] In this steel structure project, the engineering team further optimized the virtual pre-assembly process and introduced dynamic template matching and region growing algorithms to accurately identify and model the overlap area boundaries of components. Since the shape and size of the overlap area boundaries vary greatly in different components, an adaptive method is used to allow the template size and shape to be dynamically adjusted according to the actual situation of the component. In this adaptive adjustment process, the region growing and refinement algorithms are used to expand and fine-tune the boundaries of the overlap area. Specifically, the region growing algorithm sets the expansion rate, adjusts the refinement of the boundaries, and uses angle adjustment to control the direction to ensure that the edges of the overlap area can accurately match the actual structure.

[0073] In order to achieve fine expansion of the boundary, the regional characteristic function is used:

[0074]

[0075] In this formula, h(x,y) represents the adaptively extended regional characteristic function, which is used to calculate the extended result of the overlapped region edge. A φ(x,y)dx dy is responsible for accumulating edge expansion results in a given area A, where φ(x,y) is the edge expansion coefficient of the area. By adjusting its value, the specific range and density of edge expansion can be controlled. In actual operation, the value of φ(x,y) is set in a range between [0.1,0.5] to adjust the expansion coefficient to adapt to the edge characteristics of different components.

[0076] In addition, to ensure the accuracy of the expansion process, the parameter γ in the formula is used as a growth rate coefficient to control the speed of boundary expansion. The value range of γ is set between [0.05, 0.2] to ensure that the boundary expansion speed is moderate, which can effectively cover the edge area without causing excessive expansion and deviating from the actual boundary. Another key parameter δ in the formula is j is the boundary adjustment factor, which is used to adapt to small changes in the boundary. Specifically, setting δ j The value range of is [0.02, 0.1], so as to adjust the sensitivity of the region refinement, so that when the boundary changes slightly, the system can respond quickly and adjust the boundary shape. Finally, θ in the formula j The parameter is used to control the direction angle of boundary expansion. Set θ j The value range of is between [0,π / 4] to ensure that the direction of boundary extension is consistent with the actual geometric shape of the overlap area, thereby achieving accurate boundary matching.

[0077] An extended test was conducted on the overlap area of ​​a large steel structure. The initial boundary shape of the area was set to be complex with multiple angle changes. By applying the formula:

[0078]

[0079] Substitute the actual parameters and set φ(x,y)=0.3, γ=0.1, δ j = 0.05, and θ j =π / 8, and the adaptive expansion of the boundary was successfully completed. During the region growth process, the boundary of the overlap area was accurately adjusted to completely match the actual component, ensuring a high degree of consistency between the virtual model and the edge of the real steel structure.

[0080] Embodiment 2:

[0081] See attached Figure 3 In this embodiment, during the virtual pre-assembly process of the steel structure project, the engineering team further improved the assembly accuracy prediction and dynamic error compensation capabilities to ensure that high-precision requirements can be met in actual construction. To this end, an error analysis method based on historical assembly data is used to identify the key factors that affect errors in assembly, including component shape, assembly sequence, material properties, ambient temperature, and mechanical accuracy. By analyzing the potential impact of these factors on errors, a feature mapping space is created, and these factors are used as feature input models for deep learning training, thereby constructing an error prediction model. Feature processing function:

[0082]

[0083] It is used to describe the contribution of each feature in error prediction. Each feature x in the formula i Represents specific variables in the assembly process, such as component size, number of connection points, or current ambient temperature.

[0084] In this formula, x 1 ,x 2 ,…,x n Represents the various characteristic variables in the assembly process. Specifically, factors such as component size, thermal expansion and contraction coefficient of materials, number of connection points, and accuracy of mechanical equipment are included in the model. In order to enable the model to reflect the importance of each feature in error prediction, the weight parameter γ is set i For example, for the key feature of mechanical accuracy, setting γ i = 0.9 to highlight the significant impact of this factor on assembly accuracy. The weight parameter range is set between [0.1, 1.0] to dynamically adjust the importance of each feature. In addition, η in the formula i is the attenuation coefficient, which is used to control the range of the influence of the feature on the error. It is found that different features have different influences on the error, so η iThe value of is set between [0.05, 0.3] to ensure that important features have a greater impact on the error within a certain range, while the impact of other minor features is gradually attenuated. Finally, the constant term C is used to smooth the value range of each feature to avoid model instability due to large differences in feature values. The value of C is set to 1.0 to ensure the stability of the model in actual predictions.

[0085] Suppose a large support structure needs to be assembled. Historical data shows that when the ambient temperature changes greatly, the assembly error increases significantly due to the thermal expansion and contraction effect of the material. 3 Input the model and set γ 3 =0.7 and η 3 =0.2, ensuring that the influence of ambient temperature is moderately expressed in the model. Substituting these parameters into the formula for calculation, we can get the predicted value of the feature processing function F(x 1 ,x 2 ,x 3 ,…), and the results are used to further analyze the trend of assembly errors. For example, the current temperature change causes F(x 1 ,x 2 ,x 3 ,…) increases to 1.2, the system determines that there will be an error of about 1mm, so it makes adjustment suggestions in advance.

[0086] Through this error prediction and dynamic compensation method, potential errors can be predicted before actual assembly, and the component position or connection point position can be pre-adjusted according to the model output value.

[0087] During the virtual pre-assembly process of the steel structure project, the engineering team continued to optimize the assembly accuracy prediction and dynamic error compensation methods to ensure accuracy requirements. To this end, the assembly error was modeled using a regression model in machine learning, and an error prediction model was constructed by learning the relationship between each characteristic variable and the assembly error in historical data. The model uses an integral formula to describe the error prediction value P(y|F(x)) to capture the nonlinear relationship between the characteristic variable and the error, thereby more accurately predicting the assembly error. Specifically, the error prediction value formula is:

[0088] P(y|F(x))=∫ 0 1 (λ·F(x)+θsin(k·x))dx

[0089] In this formula, y represents the predicted value of the assembly error, and F(x) is the feature mapping result obtained by the feature processing function, which is used to comprehensively analyze the impact of each feature on the error.

[0090] Each characteristic variable (such as component size, material thermal expansion and contraction coefficient, assembly environment temperature and humidity, mechanical accuracy, etc.) is input into the model to obtain the characteristic mapping result F(x). In order to effectively adjust the weight of each feature in the error prediction, the parameter λ in the formula is used as a linear adjustment parameter, and its value range is set between [0.5, 1.5] to adjust the direct impact of the characteristic mapping on the error. For example, if the impact of component size is more significant in error prediction, λ is set to 1.2 to highlight the linear contribution of size to the error. In addition, the nonlinear adjustment coefficient θ is used to control the amplitude of the sine term, and the value range is set to [0.1, 0.8], so that it can dynamically adjust the impact of nonlinear factors such as environmental factors. For example, in a high temperature environment, due to the complex impact of thermal expansion and contraction on assembly accuracy, θ = 0.5 is set to amplify the nonlinear contribution of ambient temperature, so that the model is more sensitive to temperature changes. The frequency coefficient k is used to adjust the periodicity of the sine term to capture periodic errors that occur under different assembly conditions. For some intermittent mechanical errors, k is set to 2 so that the sinusoidal term matches the frequency of the error change, effectively capturing the periodic impact of mechanical accuracy.

[0091] When applying this model for error prediction, the support structure assembly task that needs to be performed in a high temperature and high humidity environment is analyzed. By inputting parameters such as temperature (set to 30°C), humidity (set to 70%), component length and width, the characteristic mapping function F(x) obtains an initial value of 1.3. After substituting this value into the error prediction formula, the predicted error value is obtained:

[0092] P(y|F(x))=1.3·λ+0.5·sin(2·x)

[0093] After integrating x, the predicted error is 1.1 mm. Since this error affects the final assembly accuracy, the system recommends fine-tuning the position of the support structure before assembly to ensure that the error is within an acceptable range.

[0094] After completing the error prediction, the engineering team further used the virtual assembly accuracy prediction and dynamic error compensation method to ensure high accuracy and consistency during the assembly process. At this stage, the virtual assembly model and real-time data feedback were combined to adjust the assembly parameters through the detected real-time error value z, including adjusting the position of the component or optimizing the assembly sequence. For this purpose, an error compensation function was designed:

[0095]

[0096] Used to dynamically adjust assembly parameters according to the real-time error value z.

[0097] In this formula, z′ represents the error deviation currently detected, which is used to provide real-time feedback on deviations that occur during the assembly process. For example, during the assembly process, component position deviations may occur due to temperature changes or equipment accuracy issues. The value of q is obtained through real-time scanning and substituted into the compensation function for calculation. The parameter σ is a proportional coefficient that controls the overall strength of the compensation, with a value range of [0.5, 2.0]. For example, when the error value is large, σ is set to 1.5 to enhance the compensation strength and make the compensation response faster. Integral term It is used to make smooth adjustments within the error range, where φ(t) is the adjustment function, and the range of φ(t) is set in [0.1,0.4] to ensure that the adjustment process is smooth during the compensation process and avoid secondary offset of the component due to sudden and large adjustments.

[0098] In addition, the parameter δ is the compensation amplitude coefficient, which controls the size of the adjustment amplitude, and its value range is [0.05, 0.3]. In a certain operation, it was detected that a certain support beam had a deviation of 2mm during the assembly process. For this reason, δ = 0.2 was set to compensate for the deviation appropriately, so that the adjusted support beam position is closer to the design requirements. The parameter ω is the compensation frequency parameter, which is used to adjust the frequency of the cosine term. The value range is set to [1, 5] to adapt to error fluctuations of different frequencies. For example, when it is found that the assembly machine has periodic vibration during operation, ω = 3 is set to adjust the compensation frequency so that the response frequency of the compensation function is synchronized with the mechanical vibration period, thereby better eliminating the periodic error caused by vibration.

[0099] Assume that a 15-meter-long main beam is being assembled. Laser scanning detects that the main beam has a position deviation of 1.5 mm under high temperature conditions, that is, z = 1.5. Set σ = 1.2, δ = 0.2, and ω = 2, and substitute into the compensation function:

[0100]

[0101] The integral result is 0.9, and the calculated result of the cosine term is about -0.198, and finally T(z′) = 0.75 is obtained. This means that the system will adjust the position of the main beam by 0.75 mm to compensate for the currently detected deviation, so that the main beam returns to the ideal position.

[0102] Through this error compensation mechanism, the deviation detected in real time can be adjusted immediately during the assembly process to ensure assembly accuracy. This method effectively reduces the accumulation of errors caused by environmental or equipment factors and ensures high consistency between the virtual model and the actual assembly results. Using this method, the position of key components was successfully adjusted many times throughout the project, so that the final assembly accuracy was always controlled within 1mm, meeting the high-precision requirements of the project.

[0103] 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 virtual pre-assembly method for steel structure components based on 3D laser scanning, characterized in that The following steps are involved: S1. Adaptive layered acquisition of 3D laser data of steel structure components: S1.

1. Design an adaptive layered scanning strategy and dynamically adjust the parameters of the scanning equipment, including resolution, angle, and scanning frequency, to obtain data from each area of ​​the component; S1.2, by fusing the scan data at different levels, detect and remove abnormal points, including the noise point cloud caused by surface reflection; S1.

3. After completing the layered data collection, extract the reference points and perform coordinate calibration, and splice the data at each level to form a three-dimensional point cloud model of the entire component; S2. Intelligent feature recognition and parametric modeling of steel structure components: S2.

1. Use multi-dimensional data analysis algorithms to classify point clouds and identify key features of components, including connection holes, weld lines, and overlap areas, through edge detection and curvature analysis methods; S2.

2. Establish a closed-loop data verification system to control the high matching between the parametric model and the actual component through dynamic comparison between the scan data and the model data; S3. Virtual pre-assembly positioning based on intelligent matching algorithm: S3.

1. Design a multiple feature point matching algorithm to align key points in the model including bolt holes and lap joints; Iterative docking and position calibration through algorithms; S3.2, quantitatively analyze the spatial deviation between components during assembly; after detecting the deviation, generate a fine-tuning plan and adjust the position of the components; S4. Dynamic optimization and adjustment of virtual pre-assembly path: S4.

1. Generate a virtual pre-assembly path based on the shape, connection method and installation sequence of the components to control the assembly sequence and spatial position layout; when pre-assembling large components, give priority to assembling the main beam and supporting structure; S4.2, monitor the assembly status of each component in real time, and optimize and adjust the path through the feedback mechanism; S4.

3. Add a feasibility assessment module to simulate construction site conditions, including space limitations or equipment failures, and generate emergency adjustment plans in advance; S5. Virtual assembly accuracy prediction and dynamic error compensation: S5.

1. Based on the virtual assembly model and historical assembly data, an error prediction model is established using a machine learning algorithm to predict assembly deviations. S5.

2. Design a real-time compensation algorithm to dynamically adjust the component positions or connection point positions in the virtual model to control the high degree of fit between the virtual assembly and the actual assembly state; The virtual assembly accuracy prediction and dynamic error compensation method: By analyzing historical assembly data, key factors affecting errors are identified, including component shape, assembly sequence, material properties, ambient temperature, and mechanical accuracy; these factors are used as feature inputs into the model for deep learning to create a feature mapping space; definition Feature processing function F(x1,x2,…,x n ), where each feature x i Indicates variables that affect the error, including component shape or mechanical accuracy: Among them, x1,x2,…,x n Represents various characteristic variables in the assembly process, including component size, number of connection points or environmental parameters; γ i is a weight parameter used to adjust the importance of each feature in the overall error prediction; η i is the attenuation coefficient, which controls the influence range of each feature on the error; C is a constant term, which is used to smooth the range of eigenvalues; The virtual assembly accuracy prediction and dynamic error compensation method: uses a regression model in machine learning to model assembly errors; identifies and learns the relationship between feature variables and assembly errors by learning patterns in historical data; The virtual assembly accuracy prediction and dynamic error compensation method: Combining the virtual assembly model and real-time data feedback, the assembly parameters, including component position or assembly order, are adjusted according to the predicted deviation during the assembly process. The error compensation function T(z′) is defined, and the assembly parameters are dynamically adjusted in a timely manner according to the real-time error value z′: Among them, z′ represents the currently detected error deviation, which is used to provide real-time feedback on the deviation in the current assembly process; σ is the proportional coefficient, which controls the overall strength of the compensation; φ(t) is the adjustment function, which performs smooth adjustments within the error range through integration; δ is the compensation amplitude coefficient, which is used to control the adjustment amplitude; ω is the compensation frequency parameter, which is used to adjust the frequency of the cosine term.

2. The method for virtual pre-assembly of steel structure components based on three-dimensional laser scanning according to claim 1, characterized in that Intelligent feature recognition and parametric modeling method of the steel structure component: A deep learning network is used to process the spatial position, density and color multi-dimensional attributes of point clouds through self-supervised learning, and adaptive labels are generated through point cloud data to set intelligent classification functions: Among them, f(z,y,z) represents the classification function, which is used to classify each point according to the spatial coordinate position of the point cloud, where x, y, and x represent the three-dimensional coordinates of the point cloud respectively; the parameter α i is a weighting factor related to the classification result, which is used to control the influence weight of different feature points, and β i is the attenuation coefficient, which is used to adjust the influence range of different features in space; exponential function part Controls the distribution of point cloud density in space.

3. The method for virtual pre-assembly of steel structure components based on three-dimensional laser scanning according to claim 2, characterized in that Intelligent feature recognition and parametric modeling method of the steel structure component: Through multi-scale curvature analysis, the curvature changes at different scales are detected in layers to identify the weld and hole characteristics on the structural parts. In each layer, the sudden changes on the component surface are marked by identifying the local curvature differences. The feature detection formula is defined as follows: Among them, g(x,y,z) represents the curvature analysis function used for edge detection, which is used to determine whether a point is an edge where the surface changes suddenly; the first part of the formula Calculate the second-order derivative in the y direction and make a partial derivative in the x direction to measure the change in curvature of the surface, and the second part It is the second-order derivative in the x direction, and the partial derivative in the y direction.

4. The method for virtual pre-assembly of steel structure components based on three-dimensional laser scanning according to claim 3, characterized in that Intelligent feature recognition and parametric modeling method of the steel structure component: Dynamic template matching and region growing algorithms are introduced to adaptively adjust the template size and shape to accommodate components of different sizes and angles. During the region growing process, the boundaries of the overlapping areas are refined and the edges of the overlapping areas are fine-tuned through an adaptive expansion method.

Images