A structural component butt joint pose prediction method, device, equipment and storage medium

Abstract

The application discloses a kind of structural component butt joint pose prediction method, device, equipment and storage medium. By acquiring training data set in virtual butt joint process, the corresponding relationship of first position difference and pose parameter can be efficiently constructed, without a large number of entity butt joint test, reduce data acquisition cost. By training neural network model, the nonlinear mapping relationship of first position difference and pose can be fitted, to provide model basis for fast pose prediction, collect second position difference in actual butt joint and input model to obtain initial adjustment pose, can quickly output high-precision initial pose, greatly shorten the optimization path of quasi-newton iteration;With initial pose as quasi-newton iteration starting point, avoid iteration divergence, fall into local optimum and other problems, improve the iteration convergence speed and solution accuracy, finally realize the structural component butt joint pose solution that considers solution efficiency and butt joint accuracy, adapt to the actual demand of aviation equipment automation, intelligent assembly.

Description

Technical Field

[0001] This invention relates to the field of mechanical manufacturing, and in particular to a method, apparatus, equipment, and storage medium for predicting the docking posture of structural components. Background Technology

[0002] In fields such as machinery manufacturing and aerospace, the docking of structural components is a core process, and the accuracy of docking directly affects product quality and performance. With the development of industrial intelligence, traditional docking methods can no longer meet the demands for high precision and high efficiency. Developing accurate and efficient pose prediction methods has become crucial for driving the upgrading of assembly processes.

[0003] Existing pose prediction methods are mainly divided into two categories: one is the traditional mathematical model method, which establishes a geometric model, combines position data collected by sensors, and uses iterative methods to solve for pose parameters; the other is the artificial intelligence method, which uses algorithms such as neural networks to train models by collecting data to achieve pose prediction.

[0004] Existing technologies are highly dependent on the initial pose, and when the initial deviation is large, they are prone to getting stuck in local optima, affecting docking accuracy and efficiency. Although pure artificial intelligence methods can output results quickly, their accuracy is difficult to meet the requirements of high precision, and model training requires a large amount of real data, resulting in high collection costs, weak generalization ability, and poor docking effect. Summary of the Invention

[0005] This invention provides a method, apparatus, device, and storage medium for predicting the pose of docking structural components. It constructs a neural network model to initially estimate and adjust the pose, and combines it with a quasi-Newton iterative method to fine-tune the pose, thus solving the technical problems of low efficiency, poor convergence, and insufficient accuracy in pose solving during aircraft structural component docking.

[0006] According to one aspect of the present invention, a method for predicting the docking pose of structural components is provided, the method comprising:

[0007] The training dataset is obtained based on the virtual docking process. The training dataset includes the first position difference between the docking point and the target point during the virtual docking process, as well as the pose parameters of the structural components.

[0008] Train a neural network model based on the training dataset;

[0009] During the actual docking process, the second position difference between the docking point and the target point is collected, and the second position difference is input into the neural network model to obtain the initial adjustment pose of the output.

[0010] The initial adjustment pose is used as the initial point for iteration of the quasi-Newton iteration method to obtain the final adjustment pose of the structural component docking.

[0011] Optionally, the training dataset is obtained based on the virtual docking process, including: constructing a virtual debugging model of the docking station; repeatedly conducting virtual docking experiments based on the virtual debugging model, collecting the position of the docking point and the target point during each virtual docking process, and collecting pose parameters after successful docking; calculating the first position difference between the position of the docking point and the target point, and constructing a training dataset based on the correspondence between the first position difference and the pose parameters.

[0012] Optionally, a virtual debugging model for the docking station is constructed, including: importing three-dimensional models of aircraft structural components and tooling; assigning rigid body properties to the three-dimensional model and configuring assembly constraints, drive shafts and communication signals to form a virtual debugging model, wherein the assembly constraints include fixed constraints, hinge constraints, ball joint constraints, linear sliding constraints and lead screw constraints.

[0013] Optionally, the neural network model is trained based on the training dataset, including: obtaining a feedforward neural network, taking the first position difference in the training dataset as input and the corresponding pose parameters as output, and iteratively training the feedforward neural network; when the mean square error between the actual output and the expected output of the feedforward neural network reaches the minimum value, the training is completed, and the trained neural network model is obtained.

[0014] Optionally, the initial adjusted pose is used as the initial iteration point of the quasi-Newton iteration method to perform iterative solutions to obtain the final adjusted pose of the structural component docking. This includes: setting an iteration stopping criterion, using the initial adjusted pose as the initial iteration point of the quasi-Newton iteration method, and calculating the gradient of the current iteration point; determining whether the gradient satisfies the iteration stopping criterion. If so, the current iteration point is used as the final adjusted pose of the structural component docking; otherwise, the pose of the current iteration point is updated and the gradient is recalculated according to the search direction and iteration step size until the gradient satisfies the iteration stopping criterion, and the current iteration point is used as the final adjusted pose of the structural component docking.

[0015] Optionally, based on the search direction and iteration step size, the pose of the current iteration point is updated and the gradient is recalculated, including: calculating the search direction based on the gradient of the current iteration point; determining the objective function based on the second position difference and selecting an iteration step size that makes the objective function decrease; updating the pose of the current iteration point based on the search direction and iteration step size and recalculating the gradient.

[0016] Optionally, the objective function is determined based on the second position difference, including: establishing a local coordinate system on the plane to be docked and establishing a global coordinate system on the fixed plane; and constructing an objective function corresponding to the second position difference based on the local coordinate system and the global coordinate system.

[0017] According to another aspect of the present invention, a structural component docking pose prediction device is provided, the device comprising:

[0018] The training dataset acquisition module is used to acquire a training dataset based on the virtual docking process. The training dataset includes the first position difference between the docking point and the target point during the virtual docking process, as well as the pose parameters of the structural components.

[0019] The neural network model training module is used to train neural network models based on the training dataset.

[0020] The initial adjustment pose determination module is used to collect the second position difference between the docking point and the target point during the actual docking process, input the second position difference into the neural network model, and obtain the output initial adjustment pose.

[0021] The final adjustment pose determination module is used to iteratively solve the problem using the initial adjustment pose as the initial point of the quasi-Newton iteration method to obtain the final adjustment pose of the structural component docking.

[0022] According to another aspect of the present invention, an electronic device is provided, the electronic device comprising:

[0023] At least one processor;

[0024] and a memory communicatively connected to the at least one processor;

[0025] The memory stores a computer program that can be executed by the at least one processor, which is then executed by the at least one processor to enable the at least one processor to perform a structural component docking pose prediction method according to any embodiment of the present invention.

[0026] According to another aspect of the present invention, a computer-readable storage medium is provided, the computer-readable storage medium storing computer instructions for causing a processor to execute and implement a structural component docking pose prediction method according to any embodiment of the present invention.

[0027] The technical solution of this invention, by acquiring a training dataset during virtual docking, can efficiently construct the correspondence between the first position difference and pose parameters, eliminating the need for extensive physical docking experiments and reducing data acquisition costs. By training a neural network model, the nonlinear mapping relationship between the first position difference and pose can be fitted, providing a model foundation for rapid pose prediction. In actual docking, the second position difference is acquired and input into the model to obtain the initial adjustment pose, enabling rapid output of a high-precision initial pose and significantly shortening the optimization path of the quasi-Newton iteration. Using the initial pose as the starting point of the quasi-Newton iteration avoids problems such as iteration divergence and getting trapped in local optima, improving the iteration convergence speed and solution accuracy. Ultimately, it achieves a solution for the docking pose of structural components that balances solution efficiency and docking accuracy, adapting to the actual needs of automated and intelligent assembly of aerospace equipment.

[0028] It should be understood that the description in this section is not intended to identify key or essential features of the embodiments of the present invention, nor is it intended to limit the scope of the invention. Other features of the invention will become readily apparent from the following description. Attached Figure Description

[0029] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0030] Figure 1 This is a flowchart of a structural component docking pose prediction method provided in Embodiment 1 of the present invention;

[0031] Figure 2 This is a schematic diagram of a feedforward neural network according to Embodiment 1 of the present invention;

[0032] Figure 3 This is a flowchart of another structural component docking pose prediction method provided in Embodiment 2 of the present invention;

[0033] Figure 4 This is a schematic diagram of the coordinate system matching with the target point for the docking of structural components according to Embodiment 2 of the present invention;

[0034] Figure 5 This is a schematic diagram of a structural component docking pose prediction device according to Embodiment 3 of the present invention;

[0035] Figure 6 This is a schematic diagram of the structure of an electronic device that implements a structural component docking pose prediction method according to an embodiment of the present invention. Detailed Implementation

[0036] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. 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 should fall within the scope of protection of the present invention.

[0037] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of the invention described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0038] Example 1

[0039] Figure 1 This is a flowchart illustrating a structural component docking pose prediction method according to Embodiment 1 of the present invention. This embodiment is applicable to aircraft installation scenarios. The method can be executed by a structural component docking pose prediction device, which can be implemented in hardware and / or software. This device can be configured in electronic devices such as computers or aircraft assembly controllers. Figure 1 As shown, the method includes:

[0040] S110. Obtain a training dataset based on the virtual docking process, wherein the training dataset includes the first position difference between the docking point and the target point during the virtual docking process, as well as the pose parameters of the structural components.

[0041] Structural component docking refers to the process of precisely assembling and connecting large structural components such as the fuselage, wings, and tail of an aircraft into a complete aircraft body according to design requirements. Virtual docking refers to the simulated docking operation performed in a structural component docking simulation environment built within commercial virtual debugging software. The training dataset is a dataset collected based on the virtual docking process and used to train neural network models. The first position difference refers to the spatial position vector difference between the docking point and the target point. For example, by collecting the initial spatial positions of six docking points and six target points in the virtual docking using a laser tracker, the first position difference is obtained by subtracting the spatial position vectors formed by the docking points and target points. Pose parameters refer to the attitude and position parameters required for successful docking of the aircraft's structural components. For example, pose parameters include six parameters: the position coordinates of the structural component on the x, y, and z axes, and the rotation angles of the structural component around each of the x, y, and z axes.

[0042] Optionally, the training dataset is obtained based on the virtual docking process, including: constructing a virtual debugging model of the docking station; repeatedly conducting virtual docking experiments based on the virtual debugging model, collecting the position of the docking point and the target point during each virtual docking process, and collecting pose parameters after successful docking; calculating the first position difference between the position of the docking point and the target point, and constructing a training dataset based on the correspondence between the first position difference and the pose parameters.

[0043] Among them, the virtual debugging model refers to the digital simulation model built in commercial virtual debugging software to simulate the actual docking scenario of aircraft structural components, and is the basic carrier for carrying out virtual docking experiments.

[0044] Specifically, electronic devices can repeatedly conduct virtual docking experiments based on a pre-built virtual debugging model, with the experiments repeated 1000 times. During each virtual docking process, a laser tracker collects the initial spatial positions of six docking points and six target points. After each successful virtual docking experiment, the adjusted pose parameters of the structural components to be docked are simultaneously recorded. Then, the first position difference between the docking point positions and the target point positions is calculated. This is done by converting the initial spatial positions of the six docking points and six target points collected in each virtual docking experiment into spatial position vectors, and then subtracting the two sets of spatial position vectors to obtain the spatial position vector difference. Finally, the spatial position vector difference obtained from each experiment is used as a set of input data, and the six pose parameters recorded after successful docking in this experiment are used as the corresponding output data, forming a set of data corresponding to the first position difference and pose parameters. The 1000 experiments yielded 1000 sets of corresponding data. These 1000 sets of data were integrated to construct a training dataset for subsequent neural network model training. Of these, 800 sets of data can be used as the actual training set for model training, and the remaining 200 sets are used as a test set to evaluate the performance of the subsequently trained neural network model.

[0045] Optionally, a virtual debugging model for the docking station is constructed, including: importing three-dimensional models of aircraft structural components and tooling; assigning rigid body properties to the three-dimensional model and configuring assembly constraints, drive shafts and communication signals to form a virtual debugging model, wherein the assembly constraints include fixed constraints, hinge constraints, ball joint constraints, linear sliding constraints and lead screw constraints.

[0046] Specifically, the 3D models of core structural components such as the aircraft fuselage and tooling locators are imported into the modeling software. Due to the complex geometry and enormous size of these components, the original 3D models contain a large amount of data. Direct use would reduce the efficiency of the simulation software and affect the smoothness and accuracy of the virtual docking experiment. Therefore, the imported models need to be lightweighted. Lightweighting involves simplifying redundant structures and data while retaining key geometric features and docking references required for core simulations. Then, the lightweight 3D models are assigned rigid body properties. This operation is based on the actual physical characteristics of aircraft structural component docking. The aircraft fuselage, tooling locators, and other structural components are rigid bodies during docking and assembly, and will not undergo plastic deformation. Assigning rigid body properties to the model allows the virtual model to follow the motion and force laws of rigid bodies during simulation, restoring the motion state of the structural components during actual docking, avoiding simulation data distortion caused by model deformation, and ensuring that the subsequently acquired position and pose data are consistent with the actual working conditions.

[0047] It should be noted that assembly constraints are used to simulate the actual connections and motion relationships between aircraft structural components and tooling, as well as between various structural components. Based on the mechanical fit requirements of actual docking, five types of constraints are configured: fixed constraints, hinge constraints, ball joint constraints, linear sliding constraints, and lead screw constraints. Different constraint types correspond to different connection and motion forms of structural components. For example, fixed constraints are used to fix structural components without relative motion into a whole; hinge and ball joint constraints are used to realize the rotational motion of structural components; and linear sliding and lead screw constraints are used to realize the linear movement of structural components. By configuring assembly constraints, the connection relationships and degrees of freedom of motion in the virtual model can be completely matched with the actual docking station, ensuring that the motion trajectory and posture adjustment of structural components during virtual docking meet the actual assembly requirements.

[0048] Furthermore, after completing the assembly constraint configuration, the model's drive axes will also be configured. The drive axes are the core power source for adjusting the virtual model's pose, corresponding to the power mechanisms that drive the movement of structural components in the actual docking station. Based on the actual power transmission and motion control logic, the lead screw spindle and linear guide sliding pair are both configured as drive axes. The lead screw spindle and linear guide sliding pair are key structural components for adjusting the linear displacement of structural components during aircraft docking. By setting them as drive axes, motion control can be achieved through software, simulating the position adjustment actions of structural components during actual docking, providing a simulation basis for pose adjustment in subsequent virtual docking experiments. Finally, communication signals are configured for the model to enable industrial communication between the virtual debugging model and the motion controller, allowing for precise control of the virtual model's motion state. Simultaneously, monitoring of the assembly position of sections at the workstation can also be achieved. Specifically, an industrial communication interface protocol for the Mechatronics Concept Designer (MCD) model will be developed first to build a communication bridge between the software model and the controller, followed by the sequential execution of signal adapter encapsulation and external signal configuration. The signal adapter encapsulation is for standardizing the transmission format of communication signals. External signal configuration involves matching the signal input and output requirements of the controller. Ultimately, the signals in the signal adapter and the input signals selected in the motion controller are mapped and matched to each axis, so that the control signals issued by the controller can accurately correspond to each power axis and moving structural component of the model. This enables precise control of the virtual model docking process and allows the model's assembly position information to be fed back to the controller in real time, completing the entire process configuration of communication signals.

[0049] S120. Train the neural network model based on the training dataset.

[0050] Optionally, the neural network model is trained based on the training dataset, including: obtaining a feedforward neural network, taking the first position difference in the training dataset as input and the corresponding pose parameters as output, and iteratively training the feedforward neural network; when the mean square error between the actual output and the expected output of the feedforward neural network reaches the minimum value, the training is completed, and the trained neural network model is obtained.

[0051] Specifically, training the neural network model based on the training dataset utilizes supervised machine learning. This allows the feedforward neural network to learn the mapping relationship between the first position difference and the pose parameters. By iteratively optimizing the network's weights and thresholds, the model's predicted output is made as close as possible to the actual expected output. First, the electronic device acquires the feedforward neural network. In this embodiment, a 6×64×64×6 feedforward neural network is selected. The input layer of this network has 6 nodes, corresponding to the first position difference between the 6 docking points and the target point. The output layer also has 6 nodes, corresponding to the pose parameters adjusted for the 6 aircraft structural components. Two hidden layers are set, each with 64 nodes. The number of nodes in the hidden layers can be determined based on the accuracy requirements of the aircraft docking pose prediction and the dimensionality of the data features. For example, Figure 2 This invention provides a schematic diagram of a feedforward neural network structure according to Embodiment 1. Figure 2 The system employs a 4-layer fully connected structure, consisting of an input layer, hidden layer 1, hidden layer 2, and an output layer. Figure 2 The layers are fully connected. to It can be used as an input feature for the network. This represents the first position difference between the first docking point and the corresponding target point. This indicates the first position difference between the second docking point and the corresponding target point. This represents the first position difference between the 6th docking point and the corresponding target point. Figure 2 middle, , and These are the pose parameters of the structural components output by the output layer. Actually, there are six pose parameters in total. , and ( Figure 2 (Not shown in the image) indicates the translational adjustment amount of the aircraft's structural components to be docked. ( Figure 2 (not shown in the image) ( Figure 2 (not shown in the image) and These represent the amount of rotational adjustment of the structural components of the aircraft to be docked.

[0052] Then, the electronic device divides the 1000 sets of corresponding data on the first position difference and pose parameters, constructed through the virtual docking experiment, into a training set and a test set in an 8:2 ratio. 800 sets are used as the training set for iterative training of the network, and 200 sets are used as the test set to evaluate network performance during the process. The training process follows the error backpropagation principle of feedforward neural networks. First, the first position difference is input into the network. The data is passed through the input layer to two hidden layers. After a nonlinear transformation by the activation function of the hidden layers, it is passed to the output layer to obtain the actual output of the model, i.e., the predicted pose parameters. Finally, the electronic device calculates the mean squared error between the actual output and the expected output. The mean squared error is used to measure the deviation between the predicted and actual values ​​in a regression task. By calculating the squared average of the difference, the current prediction error of the model can be reflected. Based on the calculated mean squared error, the error is propagated back from the output layer to the hidden layer and then to the input layer, following the error backpropagation path. The weights and thresholds of each level of the network are iteratively updated according to the error magnitude. Weights and thresholds are parameters of the feedforward neural network, determining the propagation and transformation of data within the network. Each iteration adjusts these parameters based on the error, bringing the model's actual output closer to the desired output. The next set of training data is then input into the network with updated parameters, and the forward propagation, error calculation, backpropagation, and parameter update steps are repeated until all training data has been traversed, completing one iteration. This full iteration process is then repeated multiple times. During continuous iterative training, the network's weights and thresholds are continuously optimized, and the mean squared error between the model's actual and desired outputs gradually decreases. Simultaneously, 200 sets of test data are used periodically to evaluate the network's generalization ability, preventing overfitting. When, during iteration, the mean squared error between the actual and desired outputs of the feedforward neural network on both the training and test sets no longer decreases and reaches its global minimum, it indicates that the network parameters have accurately learned the mapping relationship between the first position difference and the pose parameters. The model's prediction accuracy and generalization ability have reached their optimal levels. At this point, iterative training of the feedforward neural network is stopped, resulting in the neural network model.

[0053] S130. During the actual docking process, the second position difference between the docking point and the target point is collected, and the second position difference is input into the neural network model to obtain the initial adjustment pose of the output.

[0054] In this context, the actual docking process refers to the docking operation of structural components carried out on-site at the actual assembly site of aircraft structural parts. This differs from the virtual simulation docking scenario. It requires the acquisition of the actual spatial positions of the docking points and target points using a laser tracker, representing the practical application scenario in this implementation. The docking points are six selected feature points on the aircraft structural components to be docked. These are the measurement objects of the laser tracker, and their spatial positions reflect the current actual position of the docking components. Their initial spatial positions need to be acquired in both virtual and actual docking. The target points are the aircraft docking reference structural components, i.e., the six selected feature points on the fixed structural components, serving as the docking position reference. The second position difference refers to the spatial position vector difference between the six docking points and the six target points calculated using the laser tracker during the actual docking process. The initial pose adjustment refers to inputting the second position difference into a trained neural network model, and then using the aircraft structural component pose adjustment parameters directly output by the model as the initial iteration point for the quasi-Newton iterative method, replacing the random initial point of the traditional quasi-Newton method and improving iteration efficiency and accuracy. The quasi-Newton iteration method is a numerical iterative method for solving unconstrained optimization problems. In this embodiment, it is used to iteratively optimize the initial adjustment pose output by the neural network to find the optimal pose for aircraft docking. The core is to gradually search for the pose solution that minimizes the objective function by approximately updating the inverse of the Hessian matrix. The iteration process follows a preset stopping criterion.

[0055] S140. Using the initial adjustment pose as the initial point of the quasi-Newton iteration method, iteratively solve the problem to obtain the final adjustment pose of the structural component docking.

[0056] In this context, the initial iteration point refers to the starting position of the iteration in the quasi-Newton iteration method. In this embodiment, it is the initial adjustment pose output by the neural network. Compared to the initial point such as the identity matrix in the traditional quasi-Newton method, this initial point is closer to the optimal solution, reducing the number of iterations and improving the solution speed. Iterative solution refers to the process of repeatedly calculating according to the fixed steps of the quasi-Newton iteration method. Starting from the initial adjustment pose, the identity matrix is ​​selected as an approximation of the inverse of the initial Hessian matrix. The gradient of the current point is calculated, and it is checked whether the stopping criterion is met. If not, the search direction is calculated, an appropriate step size is selected, a new pose point is obtained, and its gradient is calculated. Then, the inverse of the Hessian matrix is ​​updated using the quasi-Newton update formula. The above steps are repeated until the stopping criterion is met. The final adjustment pose refers to the adjustment pose parameters of the aircraft structural components output by the quasi-Newton iteration method after the iteration stopping criterion is met.

[0057] The technical solution of this invention, by acquiring a training dataset during virtual docking, can efficiently construct the correspondence between the first position difference and pose parameters, eliminating the need for extensive physical docking experiments and reducing data acquisition costs. By training a neural network model, the nonlinear mapping relationship between the first position difference and pose can be fitted, providing a model foundation for rapid pose prediction. In actual docking, the second position difference is acquired and input into the model to obtain the initial adjustment pose, enabling rapid output of a high-precision initial pose and significantly shortening the optimization path of the quasi-Newton iteration. Using the initial pose as the starting point of the quasi-Newton iteration avoids problems such as iteration divergence and getting trapped in local optima, improving the iteration convergence speed and solution accuracy. Ultimately, it achieves a solution for the docking pose of structural components that balances solution efficiency and docking accuracy, adapting to the actual needs of automated and intelligent assembly of aerospace equipment.

[0058] Example 2

[0059] Figure 3 This is a flowchart of a structural component docking pose prediction method provided in Embodiment 2 of the present invention. This embodiment adds a specific process to Embodiment 1, using the initial adjusted pose as the initial iteration point of the quasi-Newton iterative method to obtain the final adjusted pose of the structural component docking. The specific content of steps S210-S230 is largely the same as steps S110-S130 in Embodiment 1, therefore, it will not be repeated in this embodiment. Figure 3 As shown, the method includes:

[0060] S210. Obtain a training dataset based on the virtual docking process, wherein the training dataset includes the first position difference between the docking point and the target point during the virtual docking process, as well as the pose parameters of the structural components.

[0061] Optionally, the training dataset is obtained based on the virtual docking process, including: constructing a virtual debugging model of the docking station; repeatedly conducting virtual docking experiments based on the virtual debugging model, collecting the position of the docking point and the target point during each virtual docking process, and collecting pose parameters after successful docking; calculating the first position difference between the position of the docking point and the target point, and constructing a training dataset based on the correspondence between the first position difference and the pose parameters.

[0062] Optionally, a virtual debugging model for the docking station is constructed, including: importing three-dimensional models of aircraft structural components and tooling; assigning rigid body properties to the three-dimensional model and configuring assembly constraints, drive shafts and communication signals to form a virtual debugging model, wherein the assembly constraints include fixed constraints, hinge constraints, ball joint constraints, linear sliding constraints and lead screw constraints.

[0063] S220. Train the neural network model based on the training dataset.

[0064] Optionally, the neural network model is trained based on the training dataset, including: obtaining a feedforward neural network, taking the first position difference in the training dataset as input and the corresponding pose parameters as output, and iteratively training the feedforward neural network; when the mean square error between the actual output and the expected output of the feedforward neural network reaches the minimum value, the training is completed, and the trained neural network model is obtained.

[0065] S230. During the actual docking process, the second position difference between the docking point and the target point is collected, and the second position difference is input into the neural network model to obtain the initial adjustment pose of the output.

[0066] S240. Set the iteration stopping criterion, using the initial adjusted pose as the initial iteration point of the quasi-Newton iteration method, and calculate the gradient of the current iteration point.

[0067] Specifically, using the initial adjusted pose as the initial point for iteration in the quasi-Newton iteration method involves further optimizing the initial adjusted pose output by the feedforward neural network using the quasi-Newton iteration method. By continuously iterating and approximating the optimal pose that minimizes the spatial positional deviation between the docking point and the target point, the accuracy of structural component docking is improved. This operation is carried out using the numerical calculation principle of unconstrained optimization. First, an iteration stopping criterion is set. This criterion is the termination condition set for the quasi-Newton iteration method and is the basis for determining whether the optimal pose has been obtained. In this embodiment, the criterion is set to stop the iteration when either the iteration results meet a precision threshold or the iteration calculation exceeds the preset maximum number of iterations N, thus avoiding invalid repeated calculations. Then, the electronic device directly uses the initial adjusted pose output by the feedforward neural network as the initial point for iteration. This initial point is a pose close to the optimal solution obtained after model learning, which can reduce the number of iterations.

[0068] S250. Determine whether the gradient meets the iteration stopping criterion. If yes, execute S260; otherwise, execute S270.

[0069] It should be noted that the gradient refers to the gradient of the objective function at the current iteration point. The gradient reflects the rate of change of the objective function at that point and the steepest descent direction, and is the basis for subsequently determining the pose optimization direction and calculating the search direction.

[0070] Specifically, it determines whether the gradient of the current iteration point meets the iteration stopping criterion, that is, whether the pose corresponding to the current iteration point has reached the preset docking accuracy requirement, or whether the number of iterations has reached the upper limit.

[0071] S260. Use the current iteration point as the final adjustment pose for the docking of structural components.

[0072] The current iteration point is the aircraft structural component's pose adjustment point at a certain moment during the iterative solution process. The updated current iteration point is obtained after each iteration step is completed, until the point meets the iteration stopping criterion.

[0073] Specifically, if the gradient satisfies the iteration stopping criterion, it means that the pose corresponding to the current iteration point can achieve high-precision docking of structural components. In this case, the pose corresponding to the current iteration point is directly used as the final adjusted pose for docking of structural components.

[0074] S270. Based on the search direction and iteration step size, update the pose of the current iteration point and recalculate the gradient until the gradient meets the iteration stopping criterion. Use the current iteration point as the final adjusted pose for the docking of structural components.

[0075] The gradient refers to the gradient of the objective function at the current iteration point, reflecting the rate of change and steepest descent direction of the objective function at that point. It serves as the basis for calculating the search direction and determining the pose optimization direction. The search direction is the pose optimization direction calculated based on the gradient of the current iteration point and an approximation of the inverse of the Hessian matrix. It is crucial in determining the updated current iteration point in the quasi-Newton iteration method, ensuring that the objective function decreases rapidly and that the iteration process approaches the optimal pose. The iteration step size is the distance scale from the current iteration point to the updated current iteration point along the determined search direction. Updating the pose of the current iteration point refers to the process of calculating and adjusting the pose parameters of the current iteration point based on the calculated search direction and iteration step size to obtain a new adjusted pose point for the aircraft structural component. Each update brings the current iteration point closer to the optimal docking pose. The final adjusted pose refers to the final adjusted pose parameters of the aircraft structural component output after the iterative calculation of the quasi-Newton iteration method meets the preset iteration stopping criteria.

[0076] Specifically, when the gradient does not meet the iteration stopping criterion, the electronic device continues iterative optimization. First, it calculates the search direction by combining the gradient of the current iteration point and the approximate value of the inverse of the Hessian matrix, following the computational logic of the quasi-Newton iteration method. The search direction is the direction that allows the objective function to decrease rapidly, ensuring that subsequent iterations can approach the optimal pose. After determining the search direction and iteration step size, the pose parameters of the current iteration point are updated based on both, resulting in the updated current iteration point, which is the new structural component adjustment pose. Then, based on the updated current iteration point, the gradient of the objective function at this new point is recalculated, and it is checked whether the gradient of the updated current iteration point meets the preset iteration stopping criterion. If it still does not meet the criterion, the above operations of calculating the search direction, selecting the step size, updating the pose of the current iteration point, and recalculating the gradient are repeated, continuously iterating and optimizing until the gradient meets the iteration stopping criterion after a certain iteration. At this point, all iteration calculations are stopped, and the pose corresponding to the current iteration point that meets the criterion is output as the final result. This output pose is the final adjustment pose of the structural component docking obtained after optimization by the quasi-Newton iteration method.

[0077] Optionally, based on the search direction and iteration step size, the pose of the current iteration point is updated and the gradient is recalculated, including: calculating the search direction based on the gradient of the current iteration point; determining the objective function based on the second position difference and selecting an iteration step size that makes the objective function decrease; updating the pose of the current iteration point based on the search direction and iteration step size and recalculating the gradient.

[0078] Optionally, the objective function is determined based on the second position difference, including: establishing a local coordinate system on the plane to be docked and establishing a global coordinate system on the fixed plane; and constructing an objective function corresponding to the second position difference based on the local coordinate system and the global coordinate system.

[0079] Specifically, Figure 4 This invention provides a schematic diagram of coordinate system and target point matching for structural component docking in Embodiment 2. This diagram is used to establish a spatial position reference between the structural component to be docked and a fixed plane, providing a coordinate framework for pose calculation. The electronic device can establish a local coordinate system on the docking plane of the structural component to be docked. This coordinate system serves as the reference for describing the coordinates of the docking point, and can characterize the spatial position of the docking point within the dimensions of the structural component to be docked. Given the docking point... The coordinates in the local coordinate system are respectively Simultaneously, the electronic equipment will establish a global coordinate system on the fixed plane that mates with the structural component to be docked. This coordinate system serves as the benchmark for describing the target point's position coordinates and can be used as a reference system to measure the overall docking position of the structural components to be docked. Given the target point... The coordinates in the global coordinate system are respectively Then, the electronic equipment adjusts its pose based on the spatial position of the docking point and the target point, as well as the aircraft structural components, to establish a functional association. The expression for the functional association is shown in the following formula (1):

[0080] (1)

[0081] in, Indicates the first The first position difference corresponding to each target point Indicates the target index. Indicates the first The coordinate vectors of each target point in the global coordinate system. Indicates the first The coordinate vectors of each target point in the local coordinate system This represents the translation vector, that is, the translation transformation relationship between the global coordinate system and the local coordinate system. Indicates the first The coordinate vectors of each target point in the global coordinate system. This represents the aircraft's adjusted pose. The function expression clearly demonstrates the intrinsic relationship between the pose change and the spatial positions of the docking point and the target point. Based on the spatial position vector difference between the docking point and the target point, and combined with the above functional relationship, a theoretical model for solving the pose using the quasi-Newton iteration method is constructed, as shown in the following formula (2):

[0082] (2)

[0083] in, Indicates the first Constraint error function for each target point Indicates the first The first position difference corresponding to each target point Indicates the target index. Indicates the first The pose parameters that need to be adjusted for each target point, among which... , and This indicates the amount of translational adjustment required for the structural components of the aircraft to be docked. , and These represent the rotational adjustment amounts of the aircraft's structural components to be docked. Indicates the first The first point to be connected with the first The first positional difference between the target points.

[0084] For example, the objective function is shown in the following formula (3):

[0085] (3)

[0086] in, This represents the 6-dimensional constraint error vector. This indicates that the aircraft is adjusting its attitude. arrive These represent the constraint error functions for the six sets of target points. arrive These represent the first position differences corresponding to the six groups of target points. arrive These represent the squares of the allowable error thresholds for the six target points. The objective function quantifies the positional deviation between the docking point and the target point. The quasi-Newton iteration process continuously optimizes and adjusts the pose parameters to reduce the value of the objective function until the required accuracy is achieved. Finally, parameter initialization is performed before iteration, and the identity matrix I is selected as an approximation of the inverse of the initial Hessian matrix. The identity matrix serves as an initial approximation, providing basic parameters for calculating the search direction in the first iteration. Subsequently, the initial adjusted pose output by the feedforward neural network can be used as the initial point for iteration to initiate the iterative solution process.

[0087] Specifically, when updating the pose of the current iteration point and recalculating the gradient, the search direction is first calculated based on the gradient of the current iteration point and the approximation of the inverse of the Hessian matrix. The iteration step size that makes the objective function decrease is then selected. The gradient of the current iteration point is the first-order partial derivative of the objective function with respect to the 6-dimensional pose adjustment, which is used to represent the rate of change and descent direction of the objective function under the current pose. The approximation of the inverse of the Hessian matrix is ​​used to correct the search direction, avoid the sawtooth phenomenon of the steepest descent method, and improve the iteration convergence efficiency.

[0088] For example, the identity matrix is ​​chosen as an approximation of the inverse of the initial Hessian matrix. Calculate the current iteration point gradient Then check if the iteration stopping criterion is met. If the iteration stopping criterion is met, stop the iteration and return to the current iteration point. Otherwise, continue with the following steps: Calculate the search direction. Choose an appropriate step size So that the objective function is The position can be lowered sufficiently. Then calculate the new point. And calculate the gradient at this point. The iterative process using the quasi-Newton method is shown in the following formula (4):

[0089] (4)

[0090] in, This represents an approximation of the inverse of the Hessian matrix in the (k+1)th iteration. This represents an approximation of the inverse of the Hessian matrix in the k-th iteration. This represents the pose increment vector. , Indicates the current iteration point. This indicates the current iteration point after the update. Represents the gradient increment vector. , This represents the gradient at the current iteration point. This represents the gradient at the updated current iteration point. Specifically, by using the pose increment and gradient increment of the current iteration, the approximation of the inverse of the Hessian matrix is ​​corrected, gradually bringing it closer to the true inverse of the Hessian matrix. This improves the accuracy of the search direction in subsequent iterations and ensures the superlinear convergence of the iterations.

[0091] Optionally, the approximation of the inverse of the Hessian matrix can be updated using a quasi-Newton method update formula; the updated approximation of the inverse of the Hessian matrix and the updated gradient of the current iteration point can be used as the basic parameters for the next iteration.

[0092] Specifically, the electronic device can use the approximation of the inverse of the updated Hessian matrix and the updated gradient of the current iteration point as the basic parameters for the next iteration. That is, after completing the k-th iteration, the updated... As an initial approximation of the inverse of the Hessian matrix in the (k+1)th iteration, As the initial gradient for the (k+1)th iteration, and simultaneously with As the initial pose for the next iteration, the above process of calculating the search direction, selecting the step size, updating the pose, calculating the gradient, and updating the Hessian matrix inversely is repeated until the iteration stopping criterion is met. At this point, the final pose output is the optimal docking pose of the aircraft structural component, completing the precise docking of the structural component to be docked with the fixed plane.

[0093] The technical solution of this invention shortens the optimization path, avoids iteration divergence, and improves convergence efficiency and solution accuracy by using the initial adjusted pose as the starting point of the quasi-Newton iteration. Setting an iteration stopping criterion allows for precise control of the iteration termination timing, ensuring the accuracy and efficiency of the final pose. Establishing a dual coordinate system and constructing an objective function provides a clear spatial reference and optimization objective for pose solving. Initializing the Hessian matrix inverse with an identity matrix simplifies the iteration preparation process and ensures initial iteration stability. Calculating the search direction and step size, updating the pose and gradient, and iteratively updating the Hessian matrix inverse continuously corrects the optimization direction, gradually approaching the optimal pose, and ultimately efficiently outputting the final adjusted pose that meets the accuracy requirements.

[0094] Example 3

[0095] Figure 5This is a schematic diagram of a structural component docking pose prediction device provided in Embodiment 3 of the present invention. Figure 5 As shown, the device includes: a training dataset acquisition module 310, used to acquire a training dataset based on the virtual docking process, wherein the training dataset includes a first position difference between the docking point and the target point during the virtual docking process, and the pose parameters of the structural components.

[0096] The neural network model training module 320 is used to train a neural network model based on the training dataset.

[0097] The initial adjustment pose determination module 330 is used to collect the second position difference between the docking point and the target point during the actual docking process, input the second position difference into the neural network model, and obtain the output initial adjustment pose.

[0098] The final adjustment pose determination module 340 is used to perform iterative solutions using the initial adjustment pose as the initial point of the quasi-Newton iterative method to obtain the final adjustment pose of the structural component docking.

[0099] Optionally, the training dataset acquisition module 310 specifically includes: a virtual debugging model construction unit, used to: construct a virtual debugging model of the docking station; a pose parameter acquisition unit, used to: repeatedly conduct virtual docking experiments based on the virtual debugging model, and collect the position of the docking point and the target point during each virtual docking process, and collect pose parameters after successful docking; and a training dataset construction unit, used to: calculate the first position difference between the position of the docking point and the target point, and construct a training dataset based on the correspondence between the first position difference and the pose parameters.

[0100] Optionally, the virtual debugging model construction unit is specifically used for: importing three-dimensional models of aircraft structural components and tooling; assigning rigid body properties to the three-dimensional model and configuring assembly constraints, drive shafts and communication signals to form a virtual debugging model, wherein the assembly constraints include fixed constraints, hinge constraints, ball joint constraints, linear sliding constraints and lead screw constraints.

[0101] Optionally, the neural network model training module 320 is specifically used to: obtain a feedforward neural network, take the first position difference in the training dataset as input and the corresponding pose parameters as output, and iteratively train the feedforward neural network; when the mean square error between the actual output and the expected output of the feedforward neural network reaches the minimum value, the training is completed and a trained neural network model is obtained.

[0102] Optionally, the final adjustment pose determination module 340 is specifically used for: setting an iteration stopping criterion, using the initial adjustment pose as the initial iteration point of the quasi-Newton iteration method, calculating the gradient of the current iteration point; determining whether the gradient satisfies the iteration stopping criterion, and if so, using the current iteration point as the final adjustment pose for the docking of the structural components; otherwise, updating the pose of the current iteration point and recalculating the gradient according to the search direction and iteration step size, until the gradient satisfies the iteration stopping criterion, and using the current iteration point as the final adjustment pose for the docking of the structural components.

[0103] Optionally, the final pose determination module 340 includes a gradient and pose update unit, used for: calculating the search direction based on the gradient of the current iteration point; determining the objective function based on the second position difference and selecting the iteration step size that makes the objective function decrease; updating the pose of the current iteration point based on the search direction and the iteration step size and recalculating the gradient.

[0104] Optionally, the final pose determination module 340 further includes: an objective function determination unit, used to: establish a local coordinate system on the plane to be docked, establish a global coordinate system on the fixed plane; and construct an objective function corresponding to the second position difference based on the local coordinate system and the global coordinate system.

[0105] The technical solution of this invention, by acquiring a training dataset during virtual docking, can efficiently construct the correspondence between the first position difference and pose parameters, eliminating the need for extensive physical docking experiments and reducing data acquisition costs. By training a neural network model, the nonlinear mapping relationship between the first position difference and pose can be fitted, providing a model foundation for rapid pose prediction. In actual docking, the second position difference is acquired and input into the model to obtain the initial adjustment pose, enabling rapid output of a high-precision initial pose and significantly shortening the optimization path of the quasi-Newton iteration. Using the initial pose as the starting point of the quasi-Newton iteration avoids problems such as iteration divergence and getting trapped in local optima, improving the iteration convergence speed and solution accuracy. Ultimately, it achieves a solution for the docking pose of structural components that balances solution efficiency and docking accuracy, adapting to the actual needs of automated and intelligent assembly of aerospace equipment.

[0106] The structural component docking pose prediction device provided in this embodiment of the invention can execute the structural component docking pose prediction method provided in any embodiment of the invention, and has the corresponding functional modules and beneficial effects of the method.

[0107] Example 4

[0108] Figure 6A schematic diagram of an electronic device 10 that can be used to implement embodiments of the present invention is shown. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device can also represent various forms of mobile devices, such as personal digital assistants, cellular phones, smartphones, wearable devices (e.g., helmets, glasses, watches, etc.), and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely illustrative and are not intended to limit the implementation of the invention described and / or claimed herein.

[0109] like Figure 6 As shown, the electronic device 10 includes at least one processor 11 and a memory, such as a read-only memory (ROM) or random access memory (RAM), communicatively connected to the at least one processor 11. The memory stores computer programs executable by the at least one processor. The processor 11 can perform various appropriate actions and processes based on the computer program stored in the ROM 12 or loaded into the RAM 13 from the storage unit 18. The RAM 13 can also store various programs and data required for the operation of the electronic device 10. The processor 11, ROM 12, and RAM 13 are interconnected via a bus 14. Input / output (I / O) interfaces are also connected to the bus 14.

[0110] Multiple components in electronic device 10 are connected to I / O interface 15, including: input unit 16, such as keyboard, mouse, etc.; output unit 17, such as various types of displays, speakers, etc.; storage unit 18, such as disk, optical disk, etc.; and communication unit 19, such as network card, modem, wireless transceiver, etc. Communication unit 19 allows electronic device 10 to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks.

[0111] Processor 11 can be a variety of general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of processor 11 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various special-purpose artificial intelligence (AI) computing chips, various processors running machine learning model algorithms, a digital signal processor (DSP), and any suitable processor, controller, microcontroller, etc. Processor 11 performs the various methods and processes described above, such as a structural component docking pose prediction method.

[0112] In some embodiments, a structural component docking pose prediction method may be implemented as a computer program tangibly contained in a computer-readable storage medium, such as storage unit 18. In some embodiments, part or all of the computer program may be loaded and / or installed on electronic device 10 via ROM 12 and / or communication unit 19. When the computer program is loaded into RAM 13 and executed by processor 11, one or more steps of the structural component docking pose prediction method described above may be performed. Alternatively, in other embodiments, processor 11 may be configured to perform a structural component docking pose prediction method by any other suitable means (e.g., by means of firmware).

[0113] Various embodiments of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field-programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), application-specific standard products (ASSPs), systems-on-a-chip (SoCs), payload-programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments may include implementations in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which may be a dedicated or general-purpose programmable processor, capable of receiving data and instructions from a storage system, at least one input device, and at least one output device, and transmitting data and instructions to the storage system, the at least one input device, and the at least one output device.

[0114] Computer programs used to implement the methods of the present invention may be written in any combination of one or more programming languages. These computer programs may be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing device, such that when executed by the processor, the computer programs cause the functions / operations specified in the flowcharts and / or block diagrams to be performed. The computer programs may be executed entirely on a machine, partially on a machine, or as a standalone software package, partially on a machine and partially on a remote machine, or entirely on a remote machine or server.

[0115] In the context of this invention, a computer-readable storage medium can be a tangible medium that may contain or store a computer program for use by or in conjunction with an instruction execution system, apparatus, or device. A computer-readable storage medium may include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination thereof. Alternatively, a computer-readable storage medium may be a machine-readable signal medium. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fibers, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof.

[0116] To provide interaction with a user, the systems and techniques described herein can be implemented on an electronic device having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user; and a keyboard and pointing device (e.g., a mouse or trackball) through which the user provides input to the electronic device. Other types of devices can also be used to provide interaction with the user; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including sound input, voice input, or tactile input).

[0117] The systems and technologies described herein can be implemented in computing systems that include backend components (e.g., as data servers), or middleware components (e.g., application servers), or frontend components (e.g., user computers with graphical user interfaces or web browsers through which users can interact with implementations of the systems and technologies described herein), or any combination of such backend, middleware, or frontend components. The components of the system can be interconnected via digital data communication of any form or medium (e.g., communication networks). Examples of communication networks include local area networks (LANs), wide area networks (WANs), blockchain networks, and the Internet.

[0118] A computing system can include clients and servers. Clients and servers are generally located far apart and typically interact through communication networks. The client-server relationship is created by computer programs running on the respective computers and having a client-server relationship with each other. The server can be a cloud server, also known as a cloud computing server or cloud host, which is a hosting product within the cloud computing service system. It addresses the shortcomings of traditional physical hosts and VPS services, such as high management difficulty and weak business scalability.

[0119] It should be understood that the various forms of processes shown above can be used, with steps reordered, added, or deleted. For example, the steps described in this invention can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution of this invention can be achieved, and this is not limited herein.

[0120] The specific embodiments described above do not constitute a limitation on the scope of protection of this invention. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this invention should be included within the scope of protection of this invention.

Claims

1. A method for predicting the docking pose of structural components, characterized in that, include: A training dataset is obtained based on the virtual docking process, wherein the training dataset includes the first position difference between the docking point and the target point during the virtual docking process, as well as the pose parameters of the structural components. Train a neural network model based on the training dataset; During the actual docking process, the second position difference between the docking point and the target point is collected, and the second position difference is input into the neural network model to obtain the initial adjustment pose of the output. Using the initial adjustment pose as the initial point for iteration of the quasi-Newton iteration method, the final adjustment pose of the structural component docking is obtained by iterative solution. The step of using the initial adjusted pose as the initial iteration point of the quasi-Newton iteration method to iteratively solve for the final adjusted pose of the structural component docking includes: Set the iteration stopping criterion, take the initial adjusted pose as the iteration starting point of the quasi-Newton iteration method, and calculate the gradient of the current iteration point; Determine whether the gradient satisfies the iteration stopping criterion. If so, take the current iteration point as the final adjustment pose for the docking of the structural components. Otherwise, based on the search direction and iteration step size, update the pose of the current iteration point and recalculate the gradient until the gradient satisfies the iteration stopping criterion, and use the current iteration point as the final adjusted pose for the docking of structural components. The iterative solution process of the quasi-Newton iteration method is as follows: ； in, This represents an approximation of the inverse of the Hessian matrix in the (k+1)th iteration. This represents an approximation of the inverse of the Hessian matrix in the k-th iteration. This represents the pose increment vector. , Indicates the current iteration point. This indicates the current iteration point after the update. Represents the gradient increment vector. , This represents the gradient at the current iteration point. This represents the gradient at the current iteration point after the update. The acquisition of the training dataset based on the virtual docking process includes: Construct a virtual debugging model for the docking station; Based on the virtual debugging model, virtual docking experiments are repeatedly carried out. During each virtual docking process, the position of the docking point and the position of the target point are collected, and the pose parameters of the structural components are collected after successful docking. Calculate the first position difference between the position of the docking point and the position of the target point, and construct a training dataset based on the correspondence between the first position difference and the pose parameters; The virtual debugging model for constructing the docking station includes: Import the 3D model of the aircraft structural components and tooling; The three-dimensional model is given rigid body properties and configured with assembly constraints, drive shafts and communication signals to form a virtual debugging model. The assembly constraints include fixed constraints, hinge constraints, ball joint constraints, linear sliding constraints and lead screw constraints. The fixed constraints are used to fix structural components without relative motion into a whole. The hinge constraints and ball joint constraints are used to realize the rotational motion of the structural components. The linear sliding constraints and lead screw constraints are used to realize the linear movement of the structural components.

2. The method according to claim 1, characterized in that, The step of training the neural network model based on the training dataset includes: A feedforward neural network is obtained by taking the first position difference in the training dataset as input and the corresponding pose parameters as output, and then iteratively training the feedforward neural network. When the mean square error between the actual output and the expected output of the feedforward neural network reaches its minimum value, the training is completed, and the trained neural network model is obtained.

3. The method according to claim 1, characterized in that, The step of updating the pose of the current iteration point and recalculating the gradient based on the search direction and iteration step size includes: Calculate the search direction based on the gradient of the current iteration point; The objective function is determined based on the second position difference, and the iteration step size that makes the objective function decrease is selected; The pose of the current iteration point is updated based on the search direction and iteration step size, and the gradient is recalculated.

4. The method according to claim 3, characterized in that, Determining the objective function based on the second position difference includes: Establish a local coordinate system on the plane to be docked, and establish a global coordinate system on the fixed plane; Construct an objective function corresponding to the second position difference based on the local coordinate system and the global coordinate system.

5. A structural component docking pose prediction device, characterized in that, include: The training dataset acquisition module is used to acquire a training dataset based on the virtual docking process, wherein the training dataset includes the first position difference between the docking point and the target point during the virtual docking process, as well as the pose parameters of the structural components. A neural network model training module is used to train a neural network model based on the training dataset. The initial adjustment pose determination module is used to collect the second position difference between the docking point and the target point during the actual docking process, input the second position difference into the neural network model, and obtain the output initial adjustment pose. The final adjustment pose determination module is used to perform iterative solutions using the initial adjustment pose as the initial point of the quasi-Newton iterative method to obtain the final adjustment pose of the structural component docking. Specifically, the final pose determination module is used for: Set the iteration stopping criterion, take the initial adjusted pose as the iteration starting point of the quasi-Newton iteration method, and calculate the gradient of the current iteration point; Determine whether the gradient satisfies the iteration stopping criterion. If so, take the current iteration point as the final adjustment pose for the docking of the structural components. Otherwise, based on the search direction and iteration step size, update the pose of the current iteration point and recalculate the gradient until the gradient satisfies the iteration stopping criterion, and use the current iteration point as the final adjusted pose for the docking of structural components. The iterative solution process of the quasi-Newton iteration method is as follows: ； in, This represents an approximation of the inverse of the Hessian matrix in the (k+1)th iteration. This represents an approximation of the inverse of the Hessian matrix in the k-th iteration. This represents the pose increment vector. , Indicates the current iteration point. This indicates the current iteration point after the update. Represents the gradient increment vector. , This represents the gradient at the current iteration point. This represents the gradient at the current iteration point after the update. The training dataset acquisition module 310 specifically includes: a virtual debugging model construction unit, used to: construct a virtual debugging model of the docking station; a pose parameter acquisition unit, used to: repeatedly conduct virtual docking experiments based on the virtual debugging model, and collect the position of the docking point and the position of the target point during each virtual docking process, and collect the pose parameters of the structural components after successful docking; and a training dataset construction unit, used to: calculate the first position difference between the position of the docking point and the position of the target point, and construct a training dataset according to the correspondence between the first position difference and the pose parameters. The virtual debugging model construction unit is specifically used for: Import the 3D model of the aircraft structural components and tooling; The three-dimensional model is given rigid body properties and configured with assembly constraints, drive shafts and communication signals to form a virtual debugging model. The assembly constraints include fixed constraints, hinge constraints, ball joint constraints, linear sliding constraints and lead screw constraints. The fixed constraints are used to fix structural components without relative motion into a whole. The hinge constraints and ball joint constraints are used to realize the rotational motion of the structural components. The linear sliding constraints and lead screw constraints are used to realize the linear movement of the structural components.

6. An electronic device, characterized in that, The electronic device includes: At least one processor; and a memory communicatively connected to the at least one processor; The memory stores a computer program that can be executed by the at least one processor, the computer program being executed by the at least one processor to enable the at least one processor to perform the method according to any one of claims 1-4.

7. A computer storage medium, characterized in that, The computer storage medium stores computer instructions that are used to cause a processor to execute the method of any one of claims 1-4.

Images