A large component multi-source measurement error coupling compensation and calibration method

Abstract

This invention provides a method for multi-source measurement error coupling compensation and calibration of large components, belonging to the field of industrial measurement and intelligent sensing technology. It includes: collecting multi-source data, constructing a measurement state description vector, introducing random disturbances, obtaining a low-dimensional representation through a state coding network, and mapping it to generate an optimized control factor. Error compensation actions and state-action modeling are defined, and a state-driven joint optimization objective is constructed. An adaptive search domain is established based on the control factor and confidence constraints, mapping the data source confidence to the dynamic range of the compensation amount, generating a hierarchical candidate compensation action set, and obtaining the optimal compensation amount through two-stage incremental objective evaluation and coordinate-by-coordinate progressive optimization. The optimal compensation amount is converted into compensated data, and consistency is determined through the joint optimization objective until a threshold is met, completing multi-source fusion calibration and geometric reconstruction. This method can reduce the impact of low-quality data and improve compensation accuracy and stability.

Description

Technical Field

[0001] This invention relates to a method for multi-source measurement error coupling compensation and calibration of large components, belonging to the field of industrial measurement and intelligent sensing technology. Background Technology

[0002] With the rapid development of large-scale equipment manufacturing and precision industrial assembly, higher demands are being placed on the high-precision 3D measurement and assembly inspection of large components. In practical engineering applications, large components are typically characterized by large size, complex structure, and fine local geometric features. A single measuring device cannot simultaneously meet the needs of global high-precision control and local fine-scale morphological acquisition. Therefore, multi-source fusion measurement technology, based on the collaborative work of multiple measuring devices such as 3D coordinate sequence laser trackers, structured light 3D scanners, industrial robots, and mobile measurement platforms, has become an important technical means to achieve high-precision measurement of large components.

[0003] Currently, multi-source measurement fusion and error calibration methods mainly include methods based on geometric calibration models, methods based on unified coordinate system registration, and error compensation methods based on optimization solutions. Some methods establish extrinsic parameter calibration relationships between sensors, mapping data from different measuring devices to a unified global coordinate system to achieve multi-source data fusion. Other methods use point cloud registration and feature matching techniques to align local scan data for component surface reconstruction. Furthermore, some optimization methods construct error functions and employ iterative optimization strategies to compensate for system errors in the measurement system, thereby improving measurement accuracy.

[0004] However, existing technologies still have the following shortcomings: First, multi-source measurement data generally suffer from heterogeneous data types, inconsistent sampling frequencies, and inconsistent spatial coordinate systems, making it difficult to achieve high-quality collaborative fusion between different data sources and affecting overall measurement consistency. Second, most existing error modeling methods rely on fixed error models or pre-set empirical parameters, which are difficult to adapt to the dynamic changes in system states under complex measurement environments, resulting in a lack of adaptive capability in error compensation strategies. Third, traditional optimization methods usually adopt a unified global optimization strategy, failing to fully consider the reliability differences of different measurement data sources under different operating conditions, which may cause low-quality data to interfere with the final results during the optimization process, thereby reducing the overall calibration accuracy and stability. Fourth, existing methods often only focus on the results of single error corrections and lack a global constraint mechanism for the consistency of compensation results, making it difficult to guarantee the overall consistency and interpretability of multi-source fusion results in spatial geometry.

[0005] Therefore, there is an urgent need for a method that can address complex large-scale component measurement scenarios, introduce a data-driven measurement state assessment mechanism based on a unified multi-source data representation, and combine reliability adaptive constraints and state-driven optimization strategies to achieve adaptive coupling compensation and high consistency calibration of multi-source errors, so as to improve the accuracy, stability and engineering applicability of large-scale component measurement in complex industrial scenarios. Summary of the Invention

[0006] The purpose of this invention is to provide a method for multi-source measurement error coupling compensation and calibration of large components, so as to solve the problems of difficulty in fusion of multi-source heterogeneous data, lack of adaptability of error compensation, susceptibility to interference from low-quality data and poor consistency of fusion results in the prior art.

[0007] To achieve the above objectives, the present invention employs the following technical solution: The raw observation data from the three-dimensional coordinate sequence laser tracker, structured light three-dimensional scanner, industrial robot and mobile measurement platform are collected, and time synchronization and coordinate unification processing are performed to obtain a multi-source fusion measurement data set; A measurement state description vector is constructed based on a multi-source fusion measurement data set. After introducing random perturbations into the measurement state description vector, a state encoding network with shared parameters with the measurement state description vector input obtains a low-dimensional state representation. Decoupling mapping is then performed based on the perturbation similarity distribution to generate an optimized control factor. Define and model the multi-source error compensation action and state-action, construct the state-driven joint optimization objective, construct the adaptive search domain based on the optimization control factor, generate a hierarchical candidate compensation action set in the adaptive search domain, and obtain the optimal error compensation amount through a two-stage incremental objective evaluation and coordinate-by-coordinate progressive optimization strategy. The optimal error compensation amount is converted into the optimal data amount after error compensation, and consistency judgment is performed based on the state-driven joint optimization objective. When the judgment result is less than the preset threshold, multi-source fusion calibration and geometric reconstruction are completed, and a structured measurement result set is output.

[0008] Preferably, the original observation data includes three-dimensional spatial coordinates, local high-density point cloud data, pose matrix and joint angle sequence, and pose change information; The multi-source fusion measurement data set is constructed as follows: (1) The three-dimensional spatial coordinates of the three-dimensional coordinate sequence laser tracker are processed by interpolating the time continuity of the spherical coordinate observation sequence and removing outliers to obtain a stable three-dimensional coordinate sequence of spatial control points; The pose matrix and joint angle sequence of the industrial robot system are used to obtain the pose sequence of the end effector through the forward kinematics solution method. The pose change information of the mobile measurement platform is processed by trajectory smoothing to obtain the platform spatial pose sequence; (2) Using the three-dimensional coordinate sequence of spatial control points as the global component coordinate system, the local high-density point cloud data is mapped to the global component coordinate system by using the initial external parameter calibration relationship between the three-dimensional coordinate sequence laser tracker and the structured light three-dimensional scanner, so as to obtain point cloud data with unified spatial expression; a k-nearest neighbor local point set is constructed for each sampling point of the point cloud data, the local normal vector of each point is estimated by the principal component analysis method, and the curvature feature is calculated based on the feature value distribution of the local neighborhood, so as to obtain the point cloud normal vector set and the curvature feature set; Based on the end effector pose sequence, the external parameter transformation relationship between the industrial robot base coordinate system and the 3D coordinate sequence laser tracker coordinate system is introduced. Homogeneous coordinate transformation is used to map the robot motion trajectory to the global component coordinate system, so as to obtain robot trajectory data with unified spatial expression. Based on the platform spatial pose sequence, the spatial transformation relationship between the coordinate system of the mobile measurement platform and the coordinate system of the three-dimensional coordinate sequence laser tracker is introduced, and the homogeneous coordinate transformation is used to convert the multi-station pose data into a unified spatial expression. (3) The three-dimensional coordinate sequence of spatial control points, the set of point cloud normal vectors, the set of point cloud curvature features, robot trajectory data and multi-station pose data are spliced ​​together into a multi-source fusion measurement data set.

[0009] Preferably, the construction of the measurement state description vector based on the multi-source fusion measurement data set specifically includes: The farthest point sampling method is used to iteratively select representative points with the largest spatial coverage in the point cloud space of local high-density point cloud data; A local neighborhood is constructed with each representative point as the center. The set of point cloud normal vectors and curvature feature sets in the neighborhood are aggregated to obtain compressed point cloud features and measurement state description vectors, including: three-dimensional coordinate sequence of spatial control points, compressed point cloud features, robot trajectory data and multi-station pose data.

[0010] Preferably, the state encoding network that shares parameters with the input of the measurement state description vector after introducing random perturbations into the measurement state description vector obtains a low-dimensional state representation is specifically implemented as follows: The measurement state description vector is used to generate a perturbation state vector through a random perturbation strategy. The random perturbation strategy includes adding small random noise to the three-dimensional coordinate sequence of spatial control points, randomly masking the compressed point cloud features, applying small pose perturbations to the robot trajectory data, and adding random position offsets to the pose data of multiple stations. The measured state description vector and the perturbation state vector are respectively input into a parameter-shared state coding network to obtain the original low-dimensional state representation and the perturbation low-dimensional state representation. The parameter-sharing state coding network is constructed using a multi-layer nonlinear mapping network with randomly initialized network parameters. Consistency constraints are established by minimizing the Euclidean distance between the original low-dimensional state representation and the perturbed low-dimensional state representation, and the state coding network is optimized with the goal of minimizing the consistency constraints.

[0011] Preferably, the multi-layer nonlinear mapping network structure is as follows: The three-dimensional coordinate sequence of spatial control points, compressed point cloud features, robot trajectory data, and multi-station pose data in the measurement state description vector or perturbation low-dimensional state representation are respectively passed through a lightweight feature mapping layer to obtain the corresponding data source feature representation; the lightweight feature mapping layer consists of two fully connected layers; The high-order features of the four data sources are extracted by a self-attention mechanism. The fusion weights of each data source are learned by a lightweight gating network. The high-order features are then fused by weighting based on the weights to obtain the original low-dimensional state representation or the perturbed low-dimensional state representation.

[0012] Preferably, the specific method for decoupling mapping based on perturbation similarity distribution is as follows: Random noise following a Gaussian distribution is applied to the original low-dimensional state representation to generate a set of perturbed state samples; For each data source feature representation, the similarity distribution between it and each sample in the perturbation state sample set is calculated to obtain the similarity order; The mean and standard deviation of each similarity sequence are calculated to obtain the importance response distribution of each data source; A random sampling mechanism is constructed based on the importance response distribution to randomly sample from the response distribution of each data source and obtain the corresponding data source informationity sample. The information content samples from the four types of data sources were normalized to obtain the optimized control factor.

[0013] Preferably, the multi-source error compensation action includes: compensation for measurement deviation of the three-dimensional coordinate sequence laser tracker, compensation for geometric error of structured light point cloud, compensation for robot trajectory pose error, and compensation for drift of mobile platform station. The state space consists of compensated state variables: compensated spatial control point 3D coordinate sequence, compressed point cloud features, robot trajectory data, and multi-station pose data; The state-driven joint optimization objectives include: spatial consistency objective, local geometry preservation objective, and compensation constraint objective; The spatial consistency objective is as follows: the compensated three-dimensional coordinate sequence of spatial control points is used as a common control point. Based on the compensated robot trajectory data and the compensated multi-station pose data, it is uniformly mapped to the global component coordinate system. The Euclidean distance error between it and the common control point is calculated, and minimizing this Euclidean distance is the optimization objective. The local geometry preservation objective is as follows: for the compensated compressed point cloud features, calculate the changes in the mean local normal vector and mean curvature before and after compensation, and minimize these changes as the optimization objective; The objective of the compensation amount constraint is as follows: calculate the change range of the compensation amount in the compensation action vector, and take minimizing the norm of the change range as the optimization objective.

[0014] Preferably, the specific method for obtaining the optimal error compensation amount is as follows: (1) The optimization control factors are mapped to the adaptive search boundary of the compensation action vector, and the search ranges of the four data sources are jointly constructed to form a reliable search domain; search range as follows: , in, For the first Search boundaries of each data source, For the first The preset search space for each data source. , No. Optimization control factors for each data source; (2) For each compensation amount, a candidate update set is generated based on a three-level scale partition within its search interval: The first layer is a set of conservative correction values: To constrain the range, randomly generate From a sample, we obtain the sample set. Used for minor error correction; The second layer is a set of moderately corrected amounts: To constrain the range, randomly generate From a sample, we obtain the sample set. Used for routine error correction; The third layer is a set of strong correction quantities: with To constrain the range, randomly generate From a sample, we obtain the sample set. Used for significant bias correction; The conservative correction set, the medium correction set, and the strong correction set in the candidate update set generated by each compensation amount are combined by Cartesian combination to form three global candidate compensation action sets. (3) The candidate samples in the global candidate compensation action set are superimposed on the state variables in the state space to obtain the corrected state corresponding to the sample, and the joint optimization objective function value is calculated. (4) Introduce a two-stage evaluation mechanism: The first stage is the coarse screening stage, in which the joint optimization objective function value of each candidate objective is calculated in the global candidate compensation action set. And select the top N best candidate actions as a candidate subset; The second stage is the local incremental evaluation stage. Within the neighborhood of the candidate subset, each candidate solution is updated with fine-grained perturbation, i.e., a small increment is generated in its neighborhood, and this small increment is superimposed on the candidate solution to obtain a new candidate solution; and the joint optimization objective function value is recalculated. Adopting the greedy acceptance criterion, only when... Update the candidate solutions in a timely manner; otherwise, retain the original candidate solutions. In the process of gradually updating the candidate subset, a coordinate-wise progressive update strategy is adopted for optimization: only one dimension of the candidate solution is updated each time, while other dimensions remain unchanged. The progressive convergence optimization of the compensation vector is achieved by iteratively updating the dimension one by one. (5) When the maximum number of iterations is reached, the optimization process is terminated, and the compensation action vector that minimizes the joint objective function is selected as the optimal error compensation amount.

[0015] Preferably, the multi-source fusion calibration and geometric reconstruction specifically include: The global geometric reference information of the component is reconstructed based on the three-dimensional coordinate sequence of the compensated optimal spatial control points, and the global spatial reference measurement results of the large component are obtained. Local geometric reconstruction is performed based on the compensated optimal compressed point cloud features. The local continuous geometric morphology of the component is restored by local interpolation and surface fitting based on k-nearest neighbors, and the expression of local continuous geometric morphology is obtained. Based on the compensated optimal robot trajectory data and multi-station pose data, the motion trajectory is recovered from the spatial sampling path during the measurement process.

[0016] A device for multi-source measurement error coupling compensation and calibration of large components includes a processor and a memory storing program instructions. The processor is configured to execute the method for multi-source measurement error coupling compensation and calibration of large components when running the program instructions.

[0017] The advantages of this invention are as follows: It achieves automatic modeling of measurement states through self-supervised learning, obtaining stable and reliable state representations without manual annotation, and adapting to dynamic changes in different measurement scenarios. By introducing a dynamic search mechanism driven by data source credibility, the error compensation process can automatically adjust the optimization range, effectively reducing the impact of low-quality data on the results and improving compensation accuracy and stability. Through consistency constraints and fusion reconstruction mechanisms, it achieves unified calibration and structured output of multi-source measurement results, providing a reliable data foundation for high-precision inspection and assembly calibration of large components. Attached Figure Description

[0018] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used together with the embodiments of the invention to explain the invention and do not constitute a limitation thereof.

[0019] Figure 1 A schematic diagram of the method flow of this invention.

[0020] Figure 2 Flowchart of a method for measurement state evaluation and optimization control factor generation based on self-supervised learning.

[0021] Figure 3 Flowchart of a state-driven multi-source error compensation optimization method.

[0022] Figure 4 Trend chart of optimized control factor under different measurement noise conditions.

[0023] Figure 5 Comparison of error compensation convergence curves under different optimization strategies.

[0024] Figure 6 A comparison of error consistency before and after multi-source fusion calibration. Detailed Implementation

[0025] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0026] Example 1 like Figure 1As shown, a method for multi-source measurement error coupling compensation and calibration of large components is proposed. This method first addresses the heterogeneity of data types and inconsistencies in spatial coordinate systems among a 3D coordinate sequence laser tracker, a structured light 3D scanner, an industrial robot, and a mobile measurement platform. It then performs time synchronization and coordinate unification processing on the raw observation data of the multi-source measurement system to obtain a multi-source fused measurement data set. ;based on Constructing measurement state description vectors And by introducing random perturbations to construct ,Will and A state-encoding network with shared input parameters is used for comparative learning to obtain a low-dimensional state representation. Furthermore, decoupling mapping is performed based on the perturbation similarity distribution to generate an optimized control factor. This is used to characterize the credibility and influence weight of four types of data sources in the current measurement state; based on this, A state-driven multi-source error compensation optimization method is constructed, transforming the multi-source measurement error correction problem into the optimal solution problem of a four-dimensional continuous compensation action vector, and a state-driven joint optimization objective is constructed. At the same time The dynamic search range of the compensation variable is mapped to form an adaptive search domain with credibility constraints. Within this constraint space, a hierarchical set of candidate compensation actions is generated, and the optimal error compensation amount is obtained through a two-stage incremental target evaluation and coordinate-by-coordinate progressive optimization strategy. Finally based on Obtain the optimal data volume after error compensation and utilize State-driven joint optimization objective Perform consistency determination when When the value is less than a preset threshold, multi-source fusion calibration and geometric reconstruction are completed, and a structured measurement result set is output. This includes spatial reference information of the components, geometric information of local continuous surfaces, and motion trajectory information of the measurement system.

[0027] Specifically, including: S1: Collect raw observation data from a 3D coordinate sequence laser tracker, a structured light 3D scanner, an industrial robot, and a mobile measurement platform; perform time synchronization and coordinate unification processing to obtain a multi-source fusion measurement data set. S2: Construct a measurement state description vector based on a multi-source fusion measurement data set. After introducing random disturbances into the measurement state description vector, obtain a low-dimensional state representation by a state coding network that shares parameters with the measurement state description vector input. Then, perform decoupling mapping based on the disturbance similarity distribution to generate an optimized control factor. S3: Define and model the multi-source error compensation action and state-action, construct the state-driven joint optimization objective, construct the adaptive search domain based on the optimization control factor, generate a hierarchical candidate compensation action set in the adaptive search domain, and obtain the optimal error compensation amount through a two-stage incremental objective evaluation and coordinate-by-coordinate progressive optimization strategy. S4: Convert the optimal error compensation amount into the optimal data amount after error compensation, and perform consistency judgment based on the state-driven joint optimization objective. When the judgment result is less than the preset threshold, complete the multi-source fusion calibration and geometric reconstruction, and output the structured measurement result set.

[0028] As a refinement of the above embodiments, step S1 addresses the issue of heterogeneous data types and inconsistent spatial coordinate systems among the 3D coordinate sequence laser tracker, structured light 3D scanner, industrial robot, and mobile measurement platform during the multi-source measurement of large components. First, the key measurement devices in the multi-source measurement system are screened, selecting only four types of core data sources that can directly affect the geometric measurement accuracy of large components. These include measurement data from the 3D coordinate sequence laser tracker, point cloud data from the structured light 3D scanner, pose data from the industrial robot, and state data of the mobile platform and the environment. This avoids redundant data from interfering with subsequent error analysis.

[0029] Specifically, it includes the following: S1-1 Multi-source measurement data type selection: In a multi-source measurement and coordinate fusion system for large-scale components, to achieve unified cross-scale measurement of large-scale spatial structures and local minute key geometric features, a multi-sensor collaborative measurement system based on a mobile measurement platform is constructed. This system drives measurement sensors to perform multi-view observations from different spatial positions through a mobile platform and robot system, thereby achieving unified acquisition of global geometric constraints and local fine structures. In this system, four types of core data sources with complementary measurement capabilities that work together on the same measurement object are selected, including 3D coordinate sequence laser tracker measurement data, structured light 3D scanner data, industrial robot motion data, and mobile measurement platform status data. Among them: (1) The three-dimensional coordinate sequence laser tracker measurement data is used to observe the spherical coordinates of the spatial control points of large components, and obtain the three-dimensional coordinates of the control points through spatial calculation, which is used to construct a global spatial reference coordinate system and a high-precision geometric constraint system.

[0030] (2) Structured light 3D scanner data is used to obtain high-density 3D point cloud and corresponding surface geometric structure information of local areas of large components, and to characterize the fine morphology of key local features of components.

[0031] (3) Industrial robot system data, used to record the pose changes and joint angle sequences generated by the structured light scanner moving with the robot end effector during the scanning process, thereby describing the spatial trajectory and posture evolution relationship during the measurement process.

[0032] (4) Mobile measurement platform status data, used to record the position and pose changes of the measurement system in different spatial areas of large components, and combined with basic environmental parameters (including temperature changes and system stability parameters), to describe the spatial transformation relationship between multiple stations and the overall operating status of the system.

[0033] S1-2 Multi-source Measurement Data Synchronous Acquisition: Synchronous data acquisition is performed on a 3D coordinate sequence laser tracker, a structured light 3D scanner, an industrial robot system, and a mobile measurement platform. Specifically, the 3D coordinate sequence laser tracker outputs spherical coordinate observation data of the target control point, including the distance from the 3D coordinate sequence laser tracker to the target point. Horizontal angle With pitch angle The corresponding three-dimensional spatial coordinates are obtained through geometric calculation; the structured light 3D scanner outputs local high-density point cloud data. The pose matrix of the end effector during the scanning process of the industrial robot system outputs. and joint angle sequence It is used to characterize the spatial motion trajectory of the scanning path; the mobile measurement platform outputs information on the pose changes of the platform between different measurement stations. It is used to describe the spatial movement state of the measurement system and the environmental disturbance situation.

[0034] Furthermore, prior to measurement, a collaborative calibration target with known spatial coordinates was deployed. A 3D coordinate sequence laser tracker and a structured light 3D scanner simultaneously observed the same calibration target. Based on corresponding feature points, a spatial rigid body transformation model was constructed, and the initial extrinsic parameter calibration relationship between the structured light 3D scanner and the global coordinate system was obtained using the least squares optimization method. Meanwhile, to ensure the consistency of multi-source measurement data in both time and space, a unified timestamp mechanism is used to synchronize and align data from various sensors.

[0035] S1-3 Multi-source Measurement Data Preprocessing: Based on the acquired multi-source synchronous measurement data, the data from different sensors are preprocessed to improve the computability of subsequent multi-source fusion and error modeling.

[0036] Specifically, firstly regarding three-dimensional spatial coordinates By performing time continuity interpolation and outlier removal on the spherical coordinate observation sequence, a stable three-dimensional coordinate sequence of spatial control points is obtained. .

[0037] Pose matrix for industrial robot systems and joint angle sequence The pose sequence of the end effector is obtained through the forward kinematics method. It is used to uniformly describe spatial motion information during the measurement process.

[0038] Pose change information for mobile measurement platform Trajectory smoothing is performed to obtain the platform spatial pose sequence. It is used to describe the spatial change process of multiple monitoring stations.

[0039] S1-4 Spatial unification and global coordinate fusion of multi-source measurement data: Three-dimensional coordinate sequence of spatial control points measured by a three-dimensional coordinate sequence laser tracker The coordinate system is designated as the global component coordinate system (GCS) and serves as a unified spatial reference for multi-source measurement data.

[0040] Based on this, the obtained initial external parameter calibration relationship is used Local high-density point cloud data A coordinate transformation is performed, mapping the data from the scanner coordinate system to the global coordinate system (GCS) through a homogeneous coordinate transformation, to obtain point cloud data with a unified spatial representation. ,Right now For point cloud data with unified spatial representation For each sampling point, a set of k nearest neighbor local points is constructed. The local normal vector of each point is estimated using principal component analysis (PCA), and the curvature features are calculated based on the eigenvalue distribution of the local neighborhood, thus obtaining the point cloud normal vector set NI and the curvature feature set. .

[0041] Meanwhile, the pose sequence of the end effector in the industrial robot system The external parameter transformation relationship between the robot base coordinate system and the three-dimensional coordinate sequence laser tracker coordinate system is introduced. Similarly, homogeneous coordinate transformation is used to map the robot's motion trajectory to the GCS coordinate system, resulting in robot trajectory data with a unified spatial representation. .

[0042] Furthermore, pose sequences based on mobile measurement platforms The spatial transformation relationship between the platform coordinate system and the three-dimensional coordinate sequence laser tracker coordinate system is introduced. Using homogeneous coordinate transformation to Multi-station pose data converted into a unified spatial representation .

[0043] The final result is a spatially consistent multi-source fusion measurement data set. This is used for subsequent cross-sensor error modeling and adaptive compensation analysis.

[0044] As a refinement of the above embodiments, in step S2, due to measurement noise between multiple source sensors, fluctuations in system operating status, and the non-fixed optimal error compensation strategies for different measurement tasks, the method of relying on manual experience to set optimization parameters is difficult to accurately reflect the true state of the current measurement system. Therefore, this invention employs a physically constrained self-supervised learning method to automatically learn the intrinsic correlation between multi-source measurement data, establish a measurement state representation without manual error labeling, and further generate optimization control factors with clear physical meaning. The flowchart of the measurement state evaluation and optimization control factor generation method based on self-supervised learning is as follows: Figure 2 As shown, the specific steps include: S2-1 Measurement State Description Vector Construction: Due to the point cloud normal vector set and curvature feature set The data volume is large and the local redundancy is strong. Directly inputting it into the subsequent model will significantly increase the computational complexity. This invention adopts a structure-preserving compression method based on farthest point sampling and local geometric feature aggregation for lightweight representation.

[0045] Specifically, firstly, the farthest point sampling method (FPS) is used to iteratively select representative points with the largest spatial coverage in the point cloud space; then, a local neighborhood is constructed with each representative point as the center, and the normal vectors of the point cloud within the neighborhood are set. and curvature feature set Aggregation is performed, that is, the mean of the local normal vectors is calculated. mean curvature and curvature variance Forming a geometric description of the corresponding region Thus, compressed point cloud features are obtained. Therefore, the final measurement state description vector is formed. .

[0046] S2-2 Self-supervised state representation learning: To improve the robustness of state representation to random perturbations, this invention constructs a self-supervised state representation learning model based on perturbation consistency, which learns the low-dimensional state representation of the measurement system without manual annotation.

[0047] Specifically, the vector is described by the measurement state. As the original input, the corresponding perturbation state vector is generated through a random perturbation strategy. The random perturbation strategy includes the three-dimensional coordinate sequence of spatial control points. Add small amounts of random noise to compress point cloud features Perform random feature masking and analyze robot trajectory data. Applying small pose perturbations and analyzing pose data from multiple stations Add a random position offset.

[0048] Subsequently, the measurement state description vector With the perturbation state vector By inputting each parameter into a state-encoding network with shared parameters, the corresponding original low-dimensional state representations are obtained. and perturbation low-dimensional state representation The state encoding network is constructed using a multi-layer nonlinear mapping network. Its network parameters are randomly initialized and iteratively optimized end-to-end through backpropagation during self-supervised learning. Furthermore, to ensure the encoding network learns a state representation invariant to measurement perturbations, the original low-dimensional state representation is minimized. Representation of perturbation in low-dimensional states Consistency constraints are established using the Euclidean distance between the input vectors, and the state encoding network is optimized with the goal of minimizing these constraints, ultimately yielding the trained state encoding network. This network measures the state description vector for any input. The corresponding original low-dimensional state representation can be obtained from all of them. .

[0049] Specifically, the network structure of the state-coded network is as follows: First, for each... , , as well as Each feature is passed through an independent lightweight feature mapping layer, each consisting of two fully connected layers, and yields the corresponding data source feature representation. , , , The high-order features of the four data sources are extracted using a self-attention mechanism, and then the fusion weights for each data source are learned through a lightweight gating network. The higher-order features are then weighted and fused according to their respective weights to obtain the original low-dimensional state representation. (or perturbation of low-dimensional state representation) ).

[0050] S2-3 State Factor Decoupling Mapping: Due to the original low-dimensional state representation It is difficult to directly reflect the actual operating status of each measuring device and the measuring process in the current measurement system. Therefore, this invention further constructs a state factor decoupling mapping module based on physical constraints to represent the original low-dimensional state. It is converted into an optimization control factor with a clear physical meaning.

[0051] Specifically, this invention constructs an information measurement method based on perturbation similarity distribution, firstly based on the original low-dimensional state representation output by S2-2. Through the Apply random noise that follows a Gaussian distribution to generate multiple perturbation state samples. ,in To preset the number of perturbation samples to generate, and preset... It is 50.

[0052] Subsequently, feature representations were performed for each data source. , , , Calculate the similarity distribution between each sample and the perturbation state set; specifically, for ,calculate The cosine similarity between the sample and the perturbed state yields the corresponding similarity sequence. This is used to characterize the degree to which the data source matches the current state under different perturbation conditions.

[0053] Furthermore, for similar sequences Calculate the mean and standard deviation The importance response distribution of the data source was obtained. ;in, This indicates the average degree of consistency between the data source and the current measurement state, and The closer the value is to 1, the stronger the data source's ability to interpret the current state. ≥0 indicates the degree of fluctuation in the response of the data source under disturbance conditions. A smaller value indicates higher stability of the data source. Finally, based on this method, the importance response distributions of three other data sources were obtained. , , .

[0054] Subsequently, this invention is based on the importance response distribution , , , A random sampling mechanism is constructed to randomly sample from the response distribution of each data source to obtain the corresponding data source informationity samples. , , , Subsequently, information quality samples from the four types of data sources were analyzed. , , , Normalization is performed to obtain the optimized control factor. ,therefore, , , , The reliability of the data sources is respectively assessed using a 3D coordinate sequence laser tracker, a structured light scanner, an industrial robot motion trajectory, and a mobile measurement platform.

[0055] As a refinement of the above embodiments, in step S3, based on the optimized control factor output in S2, a state-driven multi-source error compensation optimization method is constructed. This transforms the error correction problem of the multi-source measurement system into an optimal solution problem for four-dimensional continuous control variables, thereby achieving adaptive generation of the error compensation strategy. The flowchart of the state-driven multi-source error compensation optimization method is as follows: Figure 3 As shown, the specific steps include: S3-1 Multi-source error compensation action definition and state-action modeling: This invention models the multi-source measurement error compensation problem as a continuous action optimization problem, and defines the compensation action vector. ,in: For measuring the deviation compensation amount of the three-dimensional coordinate sequence laser tracker, This is the amount of geometric error compensation for structured light point clouds. This is the amount of compensation for robot trajectory and pose error. This is the drift compensation amount for the mobile platform monitoring station.

[0056] At the same time, using the compensated motion vector Perform data compensation and define the state space. The corrected data volume, i.e. ,in , , , .

[0057] S3-2 State-Driven Joint Optimization Objective Construction: A joint optimization objective is constructed, comprising a spatial consistency objective, a local geometry preservation objective, and a compensation constraint objective. This includes: Spatial consistency objective: to adjust the compensated spatial control points Defined as a common control point, based on the compensated robot trajectory and the compensated mobile platform pose The three types of data are mapped to the global component coordinate system (GCS) using the mapping method in S1-4. Then, the Euclidean distance error between the data and the common control point is calculated, and the optimization objective is to minimize this Euclidean distance so that the data sources maintain spatial consistency in the unified coordinate system.

[0058] Local geometry preservation objective: targeting compensated point cloud features The changes in the mean local normal vector and mean curvature before and after compensation are calculated respectively, and the optimization objective is to minimize these changes, so as to ensure that the original geometric structure of large components is not damaged during the error compensation process.

[0059] Compensation quantity constraint objective: Calculate the compensation action vector The system considers the variation range of four types of compensation variables and aims to minimize the norm of the variation range, so that the system can meet the error correction requirements while avoiding excessive compensation.

[0060] Finally, the three optimization objectives are summed to obtain the state-driven joint optimization objective applicable to the current measurement state. .

[0061] S3-3 Optimization of continuous compensation actions based on credibility constraints: To address the issues of a fixed search space for compensation variables and susceptibility to interference from low-reliability data sources during multi-source measurement error compensation, a method based on optimizing control factors is proposed. The reliability-constrained continuous compensation action optimization method maps the data source reliability to the search range of compensation actions and combines it with a progressive coordinate optimization strategy to achieve adaptive high-precision solution of compensation actions. Specifically, it includes: Optimization control factor based on S2 output (These respectively characterize the reliability of the three-dimensional coordinate sequence laser tracker, structured light point cloud, robot trajectory, and mobile platform data.) First, Mapped to compensation action vector An adaptive search boundary is used to construct a credibility-constrained search domain. Specifically, the search range... Specifically: , in, For the first Search boundaries of each data source, For the first The preset search space for each data source. , No. Optimization control factors for each data source.

[0062] Therefore, when A larger value indicates higher reliability of the data source, resulting in a smaller range of variation for the corresponding compensation variable, thus maintaining stability. Conversely, a smaller value indicates lower reliability. A smaller range indicates greater uncertainty in the data source, allowing for a larger range of compensation adjustments. The dynamic search ranges of the four data sources together constitute the trusted search domain. .

[0063] Hierarchical candidate compensation action generation: in the trusted search domain To avoid excessive computational complexity caused by directly performing a global search in the continuous space, this invention constructs a hierarchical discrete candidate compensation action generation mechanism. Specifically, for each dimension of compensation variable... Within its search interval, a candidate update set is generated based on a three-level scale partition: The first layer is a set of conservative correction values: To constrain the range, randomly generate From a sample, we obtain the sample set. It is used for minor error correction.

[0064] The second layer is a set of moderately corrected amounts: To constrain the range, randomly generate From a sample, we obtain the sample set. It is used for routine error correction.

[0065] The third layer is a set of strong correction quantities: with To constrain the range, randomly generate From a sample, we obtain the sample set. It is used for significant deviation correction.

[0066] For each compensation variable, a corresponding sample set is generated, and within each set ( , , The candidate sets of the four compensation variables are combined using a Cartesian combination to ultimately form three global candidate compensation action sets. , , And the candidate samples in these three sets are defined as This is used for subsequent optimization and filtering.

[0067] State-driven incremental target evaluation strategy: for candidate samples defined as This is then superimposed on the state variables defined in S3-1 to obtain the corrected state corresponding to the sample. And calculate the joint optimization objective function value. .

[0068] A two-stage evaluation mechanism is introduced: the first stage is the coarse screening stage, which involves screening the global candidate compensation actions. , , Quickly calculate the function value for each candidate target. and select the first The optimal candidate actions are selected as a subset of candidates. The second stage is the local incremental evaluation stage, which involves evaluating the candidate subset. Within the neighborhood, a fine-grained perturbation update is performed on each candidate solution, that is, a tiny increment is generated within its neighborhood. and will The new candidate solution is obtained by superimposing it onto the candidate solution. And recalculate the joint optimization objective function value. And it adopts the greedy acceptance criterion, that is, only when The candidate solution is updated in time, otherwise the original solution is kept, thus forming a local optimum enhanced search process and taking the candidate solution with smaller function value as the new candidate optimum.

[0069] Progressive strategy for optimizing the coordinates of the credibility domain: This involves progressively updating the candidate subset. During the process, a coordinate-by-coordinate progressive update strategy is adopted for optimization, that is, only candidate solutions are updated each time. One dimension is selected while the other dimensions remain unchanged. The compensation vector is progressively converged and optimized by updating iteratively in one dimension, thereby obtaining a stable optimal compensation solution.

[0070] Optimal Compensation Vector Output and Termination Determination: After completing multiple rounds of candidate solution updates, the optimization process terminates when the maximum number of iterations is reached. The compensation action vector that minimizes the joint objective function is selected as the optimal error compensation amount, denoted as... It is used for final error correction of multi-source measurement data.

[0071] As a refinement of the above embodiment, in step S4, due to the optimal error compensation amount output by S3 This only represents the error correction amount corresponding to each data source, and has not yet formed the final measurement result of large components under a unified spatial coordinate system. Therefore, this invention further constructs an error recovery and calibration method based on the optimal compensation amount, which synchronously applies the compensation result to various types of measurement data, and finally obtains high-precision fused measurement results of large components. Specifically, it includes the following steps: S4-1 Multi-Source Measurement Data Error Recovery: Optimal error compensation amount The superposition method in S3-1 is applied to the corresponding data sources respectively to obtain the optimal data volume after error compensation. ,in , , , The error compensation amounts correspond to , , , .

[0072] S4-2 Consistency Judgment of Compensation Results: Based on the optimal data volume after error compensation Recalculate the joint optimization objective function constructed in S3-2 and use it as the consistency evaluation index of the compensation result. When the value of the joint optimization objective function is less than the preset threshold (which is set to 5% of the average value of the joint optimization objective function obtained in the first 3 rounds of calculation), it is considered that the current compensation result meets the requirements of multi-source measurement fusion of large components, and proceed to step S4-3. Otherwise, return to step S3-3 to continue to perform the compensation action optimization, re-search for the optimal error compensation amount, until the compensation termination condition is met.

[0073] S4-3 Multi-source fusion calibration and final measurement result recovery: The optimal amount of data to pass the consistency test Finally, based on the optimal data volume after error compensation... The following measurement results were obtained respectively: Compensated data based on a 3D coordinate sequence laser tracker By utilizing its absolute measurement characteristics in the global component coordinate system (GCS), the global geometric datum information of the component can be directly reconstructed, thus obtaining the global spatial datum measurement results of the large component. That is, the absolute spatial coordinates of the component; right Local geometric reconstruction is performed, and the local continuous geometric morphology of the component is restored by using a k-nearest neighbor-based local interpolation and surface fitting method, thus obtaining a representation of the local continuous geometric morphology. That is, the geometric information of the local position of the component; based on and By utilizing the motion chain constraints in the unified global coordinate system (GCS), the motion trajectory of the spatial sampling path during the measurement process is recovered, thus obtaining the motion trajectory. This reflects the overall measurement trajectory.

[0074] The final output is a set of structured measurement results. ,in, It reflects the global spatial reference information (absolute coordinates) of the component. It reflects the local continuous surface geometry (morphology) of the component. It reflects the motion trajectory information of the measurement system.

[0075] Experimental Analysis To verify the effectiveness of the proposed method for multi-source measurement error coupling compensation and calibration of large components, a simulation experimental environment for multi-source measurement of large components was constructed. The experiment simulated a multi-source collaborative measurement process under complex industrial measurement scenarios, collecting data from a 3D coordinate sequence laser tracker, a structured light 3D scanner, an industrial robot, and a mobile measurement platform, and constructing a multi-source fusion measurement dataset (DG). Based on this, the measurement state evaluation method based on self-supervised learning, the error compensation optimization method based on credibility constraints, and the multi-source fusion calibration and reconstruction method were experimentally verified.

[0076] (1) Measurement status assessment and stability analysis of optimization control factors Figure 4 This study demonstrates the changing trends of the optimized control factors for various data sources under different measurement noise levels. Experiments were conducted with varying intensities of measurement perturbations to simulate instability factors in real-world industrial environments, including sensor noise enhancement, robot trajectory disturbances, and platform pose drift. It was observed that the 3D coordinate sequence laser tracker maintained a consistently high confidence weight under low-noise conditions, while its weight decreased slightly under increased noise conditions but still maintained a stable and dominant role. The structured light scanner exhibited higher weight fluctuations in locally geometrically complex regions. Industrial robot trajectory data was highly sensitive to motion stability, with a significant decrease in weight under high perturbation conditions. The mobile measurement platform data played a supporting role in the overall structural constraints, and its weight showed a smooth trend with changes in environmental perturbation.

[0077] Experimental results show that the optimized control factor constructed by the perturbation similarity distribution in this invention can effectively reflect the reliability changes of different data sources under different measurement states, and has good adaptability and stability.

[0078] (2) Analysis of the effect of credibility constraint error compensation optimization Figure 5 The figure shows the comparison results of the error compensation convergence process and the final error level under different optimization strategies. The pink area around the curve of the method of this invention is the standard deviation confidence band of the convergence curve obtained by multiple independent repeated experiments, which is used to characterize the fluctuation range and convergence stability of the optimization process under random sampling conditions.

[0079] The comparative methods include fixed search range optimization methods, global optimization methods without confidence constraints, and the confidence-constrained hierarchical candidate optimization method proposed in this invention. Fixed search range methods, due to their failure to consider differences in data source reliability, have slow convergence speeds and are prone to getting trapped in local optima. While methods without confidence constraints offer a large search space, low-quality data sources can interfere with the optimization direction, leading to unstable convergence. In contrast, the method of this invention maps the optimization control factor to a dynamic search domain and combines a hierarchical candidate compensation strategy with a two-stage incremental evaluation mechanism, enabling the optimization process to rapidly decrease in the early stages and achieve stable convergence in the later stages, resulting in the lowest final error level. Experimental results show that the method of this invention outperforms the comparative methods in both convergence speed and final error control capability, verifying the effectiveness of the confidence-constrained optimization strategy.

[0080] (3) Consistency analysis of multi-source fusion calibration and reconstruction results Figure 6 This paper presents a consistency analysis of the results after multi-source fusion calibration and geometric reconstruction based on Sbest, including a comparison of spatial reference error, local geometric deviation, and trajectory reconstruction deviation. When only initial extrinsic parameter calibration is performed, significant spatial deviations exist between the data sources. After error compensation optimization based on this invention, the spatial reference error is significantly reduced, the local geometric topography maintains good continuity, and the measured trajectory is highly consistent with the actual motion path. Particularly in complex curved surface regions, the deviation between the structured light point cloud and the laser tracking control point is significantly reduced, indicating a significant improvement in the global spatial consistency of the multi-source fusion results. Experimental results demonstrate that this invention, through a consistency constraint judgment mechanism and a multi-source fusion reconstruction method, can effectively improve the spatial consistency and geometric integrity of the measurement results for large components.

[0081] Example 2 This disclosure also provides a device for multi-source measurement error coupling compensation and calibration of large components, including a processor and a memory. Optionally, the device may further include a communication interface and a bus. The processor, communication interface, and memory can communicate with each other via the bus. The communication interface can be used for information transmission. The processor can call logical instructions in the memory to execute the multi-source measurement error coupling compensation and calibration method for large components described in the above embodiments.

[0082] Furthermore, the logical instructions in the aforementioned memory can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium.

[0083] Memory, as a computer-readable storage medium, can be used to store software programs and computer-executable programs, such as the program instructions / modules corresponding to the methods in the embodiments of this disclosure. The processor executes functional applications and data processing by running the program instructions / modules stored in the memory, thereby realizing the multi-source measurement error coupling compensation and calibration method for large components in the above embodiments.

[0084] The memory may include a program storage area and a data storage area. The program storage area may store the operating system and applications required for at least one function; the data storage area may store data created based on the use of the terminal device. Furthermore, the memory may include high-speed random access memory and may also include non-volatile memory.

[0085] Finally, it should be noted that the above descriptions are merely preferred embodiments of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for multi-source measurement error coupling compensation and calibration of large components, characterized in that, Includes the following steps: The raw observation data from the three-dimensional coordinate sequence laser tracker, structured light three-dimensional scanner, industrial robot and mobile measurement platform are collected, and time synchronization and coordinate unification processing are performed to obtain a multi-source fusion measurement data set; A measurement state description vector is constructed based on a multi-source fusion measurement data set. After introducing random perturbations into the measurement state description vector, a state encoding network with shared parameters with the measurement state description vector input obtains a low-dimensional state representation. Decoupling mapping is then performed based on the perturbation similarity distribution to generate an optimized control factor. Define and model the multi-source error compensation action and state-action, construct the state-driven joint optimization objective, construct the adaptive search domain based on the optimization control factor, generate a hierarchical candidate compensation action set in the adaptive search domain, and obtain the optimal error compensation amount through a two-stage incremental objective evaluation and coordinate-by-coordinate progressive optimization strategy. The optimal error compensation amount is converted into the optimal data amount after error compensation, and consistency judgment is performed based on the state-driven joint optimization objective. When the judgment result is less than the preset threshold, multi-source fusion calibration and geometric reconstruction are completed, and a structured measurement result set is output.

2. The method for multi-source measurement error coupling compensation and calibration of large components according to claim 1, characterized in that, The original observation data includes three-dimensional spatial coordinates, local high-density point cloud data, pose matrix and joint angle sequence, and pose change information; The multi-source fusion measurement data set is constructed as follows: (1) The three-dimensional spatial coordinates of the three-dimensional coordinate sequence laser tracker are processed by interpolating the time continuity of the spherical coordinate observation sequence and removing outliers to obtain a stable three-dimensional coordinate sequence of spatial control points; The pose matrix and joint angle sequence of the industrial robot system are used to obtain the pose sequence of the end effector through the forward kinematics solution method. The pose change information of the mobile measurement platform is processed by trajectory smoothing to obtain the platform spatial pose sequence; (2) Using the three-dimensional coordinate sequence of spatial control points as the global component coordinate system, the local high-density point cloud data is mapped to the global component coordinate system by using the initial external parameter calibration relationship between the three-dimensional coordinate sequence laser tracker and the structured light three-dimensional scanner, so as to obtain point cloud data with unified spatial expression; a k-nearest neighbor local point set is constructed for each sampling point of the point cloud data, the local normal vector of each point is estimated by the principal component analysis method, and the curvature feature is calculated based on the feature value distribution of the local neighborhood, so as to obtain the point cloud normal vector set and the curvature feature set; Based on the end effector pose sequence, the external parameter transformation relationship between the industrial robot base coordinate system and the 3D coordinate sequence laser tracker coordinate system is introduced. Homogeneous coordinate transformation is used to map the robot motion trajectory to the global component coordinate system, so as to obtain robot trajectory data with unified spatial expression. Based on the platform spatial pose sequence, the spatial transformation relationship between the coordinate system of the mobile measurement platform and the coordinate system of the three-dimensional coordinate sequence laser tracker is introduced, and the homogeneous coordinate transformation is used to convert the multi-station pose data into a unified spatial expression. (3) The three-dimensional coordinate sequence of spatial control points, the set of point cloud normal vectors, the set of point cloud curvature features, robot trajectory data and multi-station pose data are spliced ​​together into a multi-source fusion measurement data set.

3. The method for multi-source measurement error coupling compensation and calibration of large components according to claim 2, characterized in that, The construction of the measurement state description vector based on the multi-source fusion measurement data set specifically includes: The farthest point sampling method is used to iteratively select representative points with the largest spatial coverage in the point cloud space of local high-density point cloud data; A local neighborhood is constructed with each representative point as the center. The set of point cloud normal vectors and curvature feature sets in the neighborhood are aggregated to obtain compressed point cloud features and measurement state description vectors, including: three-dimensional coordinate sequence of spatial control points, compressed point cloud features, robot trajectory data and multi-station pose data.

4. The method for multi-source measurement error coupling compensation and calibration of large components according to claim 3, characterized in that, The state encoding network, which shares parameters with the input of the measurement state description vector after introducing random perturbations into the measurement state description vector, obtains a low-dimensional state representation. The specific method is as follows: The measurement state description vector is used to generate a perturbation state vector through a random perturbation strategy. The random perturbation strategy includes adding small random noise to the three-dimensional coordinate sequence of spatial control points, randomly masking the compressed point cloud features, applying small pose perturbations to the robot trajectory data, and adding random position offsets to the pose data of multiple stations. The measured state description vector and the perturbation state vector are respectively input into a parameter-shared state coding network to obtain the original low-dimensional state representation and the perturbation low-dimensional state representation. The parameter-sharing state coding network is constructed using a multi-layer nonlinear mapping network with randomly initialized network parameters. Consistency constraints are established by minimizing the Euclidean distance between the original low-dimensional state representation and the perturbed low-dimensional state representation, and the state coding network is optimized with the goal of minimizing the consistency constraints.

5. The method for multi-source measurement error coupling compensation and calibration of large components according to claim 4, characterized in that, The structure of the multi-layer nonlinear mapping network is as follows: The three-dimensional coordinate sequence of spatial control points, compressed point cloud features, robot trajectory data, and multi-station pose data in the measurement state description vector or perturbation low-dimensional state representation are respectively passed through a lightweight feature mapping layer to obtain the corresponding data source feature representation; the lightweight feature mapping layer consists of two fully connected layers; The high-order features of the four data sources are extracted by a self-attention mechanism. The fusion weights of each data source are learned by a lightweight gating network. The high-order features are then weighted and fused based on the weights to obtain the original low-dimensional state representation or the perturbed low-dimensional state representation.

6. The method for multi-source measurement error coupling compensation and calibration of large components according to claim 5, characterized in that, The specific method for decoupling mapping based on perturbation similarity distribution is as follows: Random noise following a Gaussian distribution is applied to the original low-dimensional state representation to generate a set of perturbed state samples; For each data source feature representation, the similarity distribution between it and each sample in the perturbation state sample set is calculated to obtain the similarity order; The mean and standard deviation of each similarity sequence are calculated to obtain the importance response distribution of each data source; A random sampling mechanism is constructed based on the importance response distribution to randomly sample from the response distribution of each data source and obtain the corresponding data source informationity sample. The information content samples from the four types of data sources were normalized to obtain the optimized control factor.

7. The method for multi-source measurement error coupling compensation and calibration of large components according to claim 6, characterized in that, The multi-source error compensation actions include: compensation for measurement deviation of the three-dimensional coordinate sequence laser tracker, compensation for geometric error of structured light point cloud, compensation for robot trajectory and pose error, and compensation for drift of mobile platform station. The state space consists of compensated state variables: compensated spatial control point 3D coordinate sequence, compressed point cloud features, robot trajectory data, and multi-station pose data; The state-driven joint optimization objectives include: spatial consistency objective, local geometry preservation objective, and compensation constraint objective; The spatial consistency objective is as follows: the compensated three-dimensional coordinate sequence of spatial control points is used as a common control point. Based on the compensated robot trajectory data and the compensated multi-station pose data, it is uniformly mapped to the global component coordinate system. The Euclidean distance error between it and the common control point is calculated, and minimizing this Euclidean distance is the optimization objective. The local geometry preservation objective is as follows: for the compensated compressed point cloud features, calculate the changes in the mean local normal vector and mean curvature before and after compensation, and minimize these changes as the optimization objective; The objective of the compensation amount constraint is as follows: calculate the change range of the compensation amount in the compensation action vector, and take minimizing the norm of the change range as the optimization objective.

8. The method for multi-source measurement error coupling compensation and calibration of large components according to claim 7, characterized in that, The specific method for obtaining the optimal error compensation amount is as follows: (1) The optimization control factors are mapped to the adaptive search boundary of the compensation action vector, and the search ranges of the four data sources are jointly constructed to form a reliable search domain; search range as follows: ， in, For the first Search boundaries of each data source, For the first The preset search space for each data source. , No. Optimization control factors for each data source; (2) For each compensation amount, a candidate update set is generated based on a three-level scale partition within its search interval: The first layer is a set of conservative correction values: To constrain the range, randomly generate From a sample, we obtain the sample set. Used for minor error correction; The second layer is a set of moderately corrected amounts: To constrain the range, randomly generate From a sample, we obtain the sample set. Used for routine error correction; The third layer is a set of strong correction quantities: with To constrain the range, randomly generate From a sample, we obtain the sample set. Used for significant bias correction; The conservative correction set, the medium correction set, and the strong correction set in the candidate update set generated by each compensation amount are combined by Cartesian combination to form three global candidate compensation action sets. (3) The candidate samples in the global candidate compensation action set are superimposed on the state variables in the state space to obtain the corrected state corresponding to the sample, and the joint optimization objective function value is calculated. (4) Introduce a two-stage evaluation mechanism: The first stage is the coarse screening stage, in which the joint optimization objective function value of each candidate objective is calculated in the global candidate compensation action set. And select the top N best candidate actions as a candidate subset; The second stage is the local incremental evaluation stage. Within the neighborhood of the candidate subset, each candidate solution is updated with fine-grained perturbation, i.e., a small increment is generated in its neighborhood, and this small increment is superimposed on the candidate solution to obtain a new candidate solution; and the joint optimization objective function value is recalculated. Adopting the greedy acceptance criterion, only when... Update the candidate solutions in a timely manner; otherwise, retain the original candidate solutions. In the process of gradually updating the candidate subset, a coordinate-wise progressive update strategy is adopted for optimization: only one dimension of the candidate solution is updated each time, while other dimensions remain unchanged. The progressive convergence optimization of the compensation vector is achieved by iteratively updating the dimension one by one. (5) When the maximum number of iterations is reached, the optimization process is terminated, and the compensation action vector that minimizes the joint objective function is selected as the optimal error compensation amount.

9. The method for multi-source measurement error coupling compensation and calibration of large components according to claim 8, characterized in that, The multi-source fusion calibration and geometric reconstruction specifically include: The global geometric reference information of the component is reconstructed based on the three-dimensional coordinate sequence of the compensated optimal spatial control points, and the global spatial reference measurement results of the large component are obtained. Local geometric reconstruction is performed based on the compensated optimal compressed point cloud features. The local continuous geometric morphology of the component is restored by local interpolation and surface fitting based on k-nearest neighbors, and the expression of local continuous geometric morphology is obtained. Based on the compensated optimal robot trajectory data and multi-station pose data, the motion trajectory is recovered from the spatial sampling path during the measurement process.

10. A multi-source measurement error coupling compensation and calibration device for large components, comprising a processor and a memory storing program instructions, characterized in that, The processor is configured to execute, when running the program instructions, the method for multi-source measurement error coupling compensation and calibration of large components as described in any one of claims 1-9.

Images