Method for predicting butt joint quality of large aviation parts based on force-position cooperation

By constructing a coordinated docking quality prediction model, using force level data from multiple sets of docking process, the problem of real-time docking quality prediction in large aviation components is solved, and accurate prediction of docking quality and early warning of potential problems is achieved. , improves assembly efficiency and safety.

CN118963293APending Publication Date: 2024-11-15DALIAN UNIV OF TECH
View PDF 13 Cites 0 Cited by

Patent Information

Application Number
CN202411128183.5
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-08-16
Publication Date
2024-11-15

AI Technical Summary

Technical Problem

The prior art cannot achieve real-time and accurate docking quality prediction during the assembly process of large aviation components, resulting in difficulty in early warning of potential quality problems and affecting assembly efficiency and safety.

Method used

The docking quality prediction method of large aviation components with force level coordination is adopted. By collecting force sensor data from different attitudes and docking processes of components, a force level coordinated docking quality prediction model is constructed, and a force level data from multiple sets of force level adjustment process is used for model training. Achieve accurate prediction of component docking quality.

Benefits of technology

Real-time and accurate prediction of component docking quality during the docking process of adjusting posture docking is realized, reducing the risk of potential quality problems during the assembly process, and improving the docking quality and assembly efficiency of large aviation components.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118963293A_ABST
    Figure CN118963293A_ABST
Patent Text Reader

Abstract

The invention belongs to the technical field of intelligent assembly of large aviation parts, and discloses a force-position collaborative large aviation part butt joint quality prediction method. According to the method, force sensor data and geometric quantity data of different postures and different docking processes of parts are collected, a force-position collaborative docking quality prediction model is constructed, a proper algorithm and corresponding parameters are selected, force-position data obtained after data processing serve as input of the model, and whether docking quality is qualified or not serves as output. And training the force-position collaborative docking quality prediction model. In the actual docking process, the attitude adjusting system industrial personal computer issues an instruction to the programmable logic controller to control the positioner to drive the component to perform attitude adjusting docking motion, the data acquisition industrial personal computer is used for acquiring and measuring force position data in real time, and whether a horizontal measuring point for supporting the wing component exceeds a tolerance range is judged based on a model; therefore, accurate prediction of the butt joint quality of the parts is realized. The method has certain engineering value.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The invention belongs to the technical field of intelligent assembly of large aviation components, and relates to a method for predicting the docking quality of large aviation components with force-position coordination. Background Art

[0002] In the field of aviation manufacturing, the assembly of large aviation components is a key link in the entire production process, accounting for more than 50% of the workload of the entire production process. As the most complex and time-consuming link in the assembly, the component posture adjustment and docking process plays a vital role in ensuring the assembly quality. In recent years, with the continuous progress of the aviation industry, the automated assembly technology based on digital posture adjustment and docking and measurement systems has been widely used in the posture adjustment and docking process of large aviation components, significantly improving the manufacturing efficiency of large aviation components. At present, in the process of component docking, only the theoretical measurement point coordinates and the actual measurement point coordinates are used to solve the movement of the locator drive shaft, and then the locator is driven to perform posture adjustment and docking. After the posture adjustment and docking process is completed, the docking quality is reflected by judging whether the geometric measurement data of the laser tracker meets the tolerance requirements. When the target posture cannot be achieved after multiple posture adjustments and docking, the locator is driven to move in a certain direction only by relying on manual experience, and repeated trials and adjustments are performed. This assembly process only relies on geometric measurement data, and it is difficult to achieve real-time and accurate docking quality prediction, and thus it is impossible to warn potential quality problems in the assembly process. Therefore, how to design a reasonable prediction method to achieve rapid prediction of the docking quality of components in the actual docking process has become the main problem and research direction at present.

[0003] The patent "A system and method for aircraft component assembly based on robot force-position hybrid control" by Zheng Yuhong et al., patent number 202011176106.9, discloses a new assembly technology. By integrating flexible grippers, articulated robots, AGV carts, product fixtures, six-dimensional force sensors and controllers, an aircraft component assembly system is constructed. During the process of grasping, transporting the workpiece to the docking and returning to the workstation, the force-position hybrid control algorithm is used to monitor and adjust the contact force between the aircraft components and the robot in real time to ensure smooth assembly.

[0004] This method will only stop workpiece assembly and alarm when the contact force and / or robot posture are abnormal. It fails to achieve real-time prediction of the force-position relationship during the assembly of aircraft parts, cannot provide early warning of assembly force deviations, and is difficult to respond to potential assembly quality problems in real time, thus failing to effectively reduce the risk of damage to aircraft parts due to assembly. Summary of the invention

[0005] The technical problem to be solved by the present invention is to take into account the characteristics of nonlinear correlation between geometric quantities and mechanical quantities, overcome the defect that traditional technology cannot predict the docking quality of components, and invent a method for predicting the docking quality of large-scale aviation components with force-position coordination. The method first builds a docking quality prediction hardware system, collects force-position data during the docking process and performs data processing, selects appropriate algorithms and parameters to establish a docking quality prediction model, uses multiple sets of force-position data during the attitude adjustment docking process to train the model, and finally uses the model to predict the docking quality of components. This method avoids the problem that the attitude adjustment internal force cannot be determined and the physical model of the docking force and position is difficult to construct. It effectively uses a data-driven approach to solve the problem that the various error factors in the physical model mechanism analysis make the analysis difficult, and realizes the accurate prediction of the docking quality of components during the attitude adjustment docking process. In addition, this method can meet the real-time docking quality prediction needs of all attitude adjustment docking systems.

[0006] The technical solution of the present invention:

[0007] A method for predicting the docking quality of large aviation components with force-position coordination is proposed. By collecting force sensor data and geometric data of components in different postures and different docking processes, a force-position coordination docking quality prediction model is constructed. The appropriate algorithm and corresponding parameters are selected, and the force-position data after data processing is used as the input of the model. Whether the docking quality is qualified is used as the output to train the force-position coordination docking quality prediction model. In the actual docking process, the attitude adjustment system industrial computer sends instructions to the programmable logic controller to control the positioner to drive the components to perform attitude adjustment and docking movements. The data acquisition industrial computer is used to collect and measure force-position data in real time. Based on the model, it is judged whether the horizontal measurement point of the supporting wing component exceeds the tolerance range, thereby realizing accurate prediction of the component docking quality. The specific steps are as follows:

[0008] Step 1: Build the docking quality prediction hardware system and collect the docking process force and position data

[0009] Before the experiment begins, each force sensor 3 is first installed and fixed on the supporting part of the positioner 2 to ensure that the force sensor 3 accurately measures the force on the supporting wing component during the movement of the positioner 2; then the force sensor 3 is connected to the signal amplifier 5, and the mechanical signal converted by the signal amplifier 5 is connected to the programmable logic controller 6; at the same time, the position signal of the positioner servo motor is connected to the programmable logic controller 6, and the programmable logic controller 6 is connected to the attitude adjustment system industrial computer 7 to establish communication between the attitude adjustment system industrial computer 7 and the data acquisition industrial computer 8; finally, the force position data is collected and processed in real time by the data acquisition industrial computer 8; at the same time, the laser tracker 1 is deployed to ensure that the horizontal measurement point 4 on the supporting wing component can be measured, and the global coordinate system is established by measuring the ground ERS point;

[0010] At the beginning of the experiment, the programmable logic controller 6, the attitude adjustment system industrial computer 7 and the data acquisition industrial computer 8 are started; the geometric position data of the horizontal measurement point 4 on the supporting wing component is measured by the laser tracker 1, and it is converted into the current six-dimensional attitude data of the wing component through SVD decomposition; then the transition path from the current attitude to the target attitude of the supporting wing component in the theoretical digital model is calculated by the interpolation method, and the corresponding attitude conversion path data of the supporting wing component is generated; this attitude conversion path data describes how to gradually adjust the attitude of the supporting wing component through the positioner 2, and convert the current six-dimensional attitude data of the supporting wing component into the target attitude in the theoretical digital model; the attitude adjustment system industrial computer 7 sends instructions to the programmable logic controller 6 to control the positioner 2 to drive the supporting wing component to perform attitude adjustment and docking movement; during the movement, the data acquisition industrial computer 8 collects the three-dimensional force data of the force sensor 3 and the position data of the positioner 2 in real time; repeat the above attitude adjustment and docking process m times, obtain a total of m groups of measurement data, and complete the collection of force and position data of multiple groups of attitude adjustment and docking processes;

[0011] Step 2: Establish a prediction model for the docking quality of supporting wing components

[0012] The data acquisition industrial computer 8 stores multiple sets of three-dimensional force data and position data collected during the attitude adjustment and docking process. By building a docking quality prediction model, the three-dimensional force data collected by the force sensor 3 and the position data of the locator 2 are used as the input of the model, and whether the horizontal measurement point of the supporting wing component exceeds the tolerance range is used as the quality judgment standard. If it exceeds the tolerance range, it is considered that the assembly quality is unqualified, and the docking quality judgment result is used as the model output; finally, a nonlinear mapping relationship between input x and output y is established through a data-driven method:

[0013] f:x→y (1)

[0014] The construction process of the docking quality prediction model is as follows:

[0015] Firstly, the grid search method is used to optimize the parameters of the docking quality prediction model, including duality selection, penalty coefficient, residual convergence condition and loss function. By searching different parameter combinations, the optimal parameter settings are determined to improve the performance of the docking quality prediction model in quality prediction in different docking processes. The optimized objective function is as follows:

[0016]

[0017] Among them, w is the optimal hyperplane slope, b is the optimal hyperplane intercept, C is the penalty coefficient, ξ i is a slack variable, k is the number of training samples, x i is the feature vector of the i-th training sample, y iis the label of the i-th training sample; in order to better fit the nonlinear mapping relationship between input data and output data, the following radial basis function (RBF) kernel function is introduced:

[0018] K(x i ,x j ) = exp(-γ·||x i -x j || 2 ) (3)

[0019] Among them, K(x i , x j ) is the kernel function, and γ is the kernel parameter. By mapping the original input data to a high-dimensional space, the complex nonlinear relationship between the input and output data can be captured more effectively, significantly improving the fitting accuracy and prediction performance of the model.

[0020] At the same time, by introducing the Lagrange multiplier, the original problem is transformed into its corresponding Lagrange function, and the corresponding dual function is constructed:

[0021]

[0022] Among them, α i , α j is the Lagrange multiplier vector. By solving the dual function, the optimal solution w of the original problem is obtained. * and b * as follows:

[0023]

[0024] Among them, α * is the optimal solution of α in the dual problem; grid search is used to optimize the hyperparameters C and γ to ensure the training effect of the docking quality prediction model. The final adjusted model and the corresponding prediction values ​​are as follows:

[0025]

[0026]

[0027] in, The value of determines the predicted category of sample x, where represents the positive class, Represents the negative class; Step 3: Train the docking quality prediction model and perform docking quality prediction based on measured data

[0028] First, mean filling is used to process the variables with null values ​​in the force data, and the maximum and minimum values ​​of each dimension are normalized to complete the feature preprocessing process:

[0029]

[0030] The processed force and position data are input into the optimized prediction model for training, which can predict whether the posture adjustment and docking quality is qualified and meet the requirements of quality prediction in complex docking processes. The evaluation criteria for prediction accuracy are as follows:

[0031]

[0032] Among them, TP refers to the number of positive samples predicted correctly by the model, TN refers to the number of negative samples predicted correctly by the model, FP refers to the number of samples that the model mistakenly predicts as positive from negative, and FN refers to the number of samples that the model mistakenly predicts as negative from positive. The accuracy of the model is evaluated by calculating the proportion of the model predictions that are consistent with the actual observations.

[0033] Beneficial effects of the present invention: Compared with the traditional method for detecting the docking quality of supporting wing components, this method effectively solves the problem of the influence of various error factors in theoretical analysis through a data-driven approach. There is no need to calculate the geometric position of the horizontal attitude adjustment point of the component corresponding to different force values. It is only necessary to obtain a large amount of attitude adjustment and docking process data, and by establishing a suitable docking quality prediction model and conducting model training, the real-time prediction of the component docking quality can be achieved. This method is suitable for the prediction of the docking quality of components in any attitude adjustment and docking system, avoiding complex problems such as the determination of the attitude adjustment internal force and the construction of the docking force position physical model, further enhancing the real-time warning function for unqualified attitude adjustment and docking quality, and improving the docking quality of large aviation components. This method has certain engineering value. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] Figure 1 Shown is the layout diagram of the docking hardware system.

[0035] Figure 2 Shown is a flowchart of the force-position coordinated docking quality prediction process in the present invention.

[0036] In the figure, 1-laser tracker, 2-positioner, 3-force sensor, 4-horizontal measurement point, 5-signal amplifier, 6-programmable logic controller, 7-posture adjustment system industrial computer, 8-data acquisition industrial computer. DETAILED DESCRIPTION

[0037] The specific implementation of the present invention is further described below in conjunction with the accompanying drawings and technical solutions.

[0038] Figure 1This is the installation diagram of the docking hardware system. The programmable logic controller used in this embodiment is AC802, which can drive 64+64 axes. The industrial computer of the posture adjustment system uses IPC-610L. The industrial computer for data acquisition uses MIC-770V3H-BTO, and the sampling frequency is 0.1s. The force sensor is MCS10, with a maximum accuracy of 0.1, an axial range of 5-200kN, and a radial range of 0.05 to 3.5kNm. The signal amplifier is ClipX, with an accuracy level of 0.01, through which the electrical signal is converted into a mechanical quantity for display.

[0039] Figure 2 This is a flowchart of the docking quality prediction method. The entire prediction method is divided into the following three parts: building a docking hardware system and collecting force and position data during the docking process, establishing a component docking quality prediction model, training the docking quality prediction model, and predicting docking quality based on measured data. The specific steps of the method are as follows:

[0040] (1) Build the docking hardware system and collect force and position data during the docking process

[0041] In this embodiment, the installation diagram of the docking hardware system is shown in the attached Figure 1 As shown. Before the experiment begins, each force sensor 3 is first installed and fixed on the supporting part of the positioner 2 to ensure that the force sensor 3 can accurately measure the force exerted on the component during the movement of the positioner 2. Then the force sensor 3 is connected to the signal amplifier 5, and the mechanical signal converted by the signal amplifier 5 is connected to the programmable logic controller 6. At the same time, the position signal of the positioner servo motor is connected to the programmable logic controller 6, and the programmable logic controller 6 is connected to the attitude adjustment system industrial computer 7 to establish communication between the attitude adjustment system industrial computer 7 and the data acquisition industrial computer 8. Finally, the force and position data are collected and processed in real time by the data acquisition industrial computer 8. The laser tracker 1 is arranged at a suitable position to ensure that the horizontal measurement point 4 on the component can be measured, and the global coordinate system is established by measuring the ground ERS point.

[0042] At the beginning of the experiment, start the programmable logic controller 6, the attitude adjustment system industrial computer 7 and the data acquisition industrial computer 8. Use the laser tracker 1 to measure the geometric position data of the horizontal measurement point 4 on the component, and convert it into the six-dimensional posture data of the component through SVD decomposition. Then match these data with the target posture to generate the corresponding posture conversion path data. The attitude adjustment system industrial computer 7 sends instructions to the programmable logic controller 6 to control the positioner 2 to drive the component to perform attitude adjustment and docking movement. During the movement, the data acquisition industrial computer 8 collects the three-dimensional force data of the force sensor 3 and the position data of the positioner 2 in real time. Repeat the above attitude adjustment and docking process 20 times to obtain a total of 20 sets of measurement data, and complete the collection of multiple sets of attitude adjustment and docking process force and position data.

[0043] (2) Establish a docking quality prediction model for the docking process

[0044] After the posture adjustment and docking process of this embodiment is completed, the data acquisition industrial computer 8 is operated to process the data to obtain data that can be used to establish a machine learning model. The docking quality prediction model is established through Python software programming, and the mapping relationship between the actual measurement value x of the sensor and the predicted expected output value y is established using formula (1):

[0045] f:x→y (10)

[0046] The parameter C value is selected as 150, and the model is adjusted by establishing a weighted support vector machine (SVM) to deal with the imbalance problem of the data set and improve the generalization ability of the model. Through formula (2), the problem is converted into the following optimization problem:

[0047]

[0048] By deriving and simplifying formula (4), the original problem is transformed into its dual form, laying the foundation for further optimization solution:

[0049]

[0050] Through the mapping of radial basis function (RBF kernel), the complex nonlinear characteristics in the data are effectively captured and the dual problem is solved. Based on formula (5) and formula (6), the model is adjusted to accurately predict the docking quality in high-dimensional space, and the final optimization model is as follows:

[0051]

[0052] (3) Training the docking quality prediction model and performing docking quality prediction based on measured data

[0053] In this embodiment, after the docking quality prediction model is established, the model is trained by running the program. A total of 20 groups of data are obtained as training sets, and 4 groups of data are obtained as data test sets. Multiple groups of force position data are preprocessed by formula (8), and the 4 groups of force position data in the processed test set are imported into the input end, and the trained docking quality prediction model is used for prediction to obtain an accurate prediction value of the docking quality. The accuracy of the predicted output is calculated by formula (9). Under laboratory conditions, the accuracy of the training data is 0.952, and the accuracy of the predicted data is 0.940. The results show that in the task of predicting the docking quality of large aviation components, the prediction accuracy of the model is not less than 90%, and the prediction model can achieve high-precision prediction effects.

[0054] The above describes the specific embodiments of the present invention. The present invention proposes a force-position coordinated docking quality prediction method for the docking process of large-scale aviation components. Compared with the traditional method, this method overcomes the defect that the docking quality of components can only be judged as unqualified when the force value exceeds the tolerance during the docking process of large-scale aviation components. There is no need to calculate the theoretical force values ​​corresponding to different locator positions, which avoids the difficulties in determining the internal force of the posture adjustment and constructing the physical model of the docking force position in practical applications. By collecting multiple groups of posture adjustment docking process data and establishing and training the force position prediction model, this method can effectively use the data-driven approach to solve the analysis difficulties caused by various error factors in the mechanism analysis, accurately obtain the predicted force value and perform real-time docking quality warning for the posture adjustment docking system. This method is a docking quality prediction method with practical engineering application value, which significantly improves the automation level of the docking process of large-scale aviation components, improves the safety and reliability of the posture adjustment docking process, and provides a strong guarantee for the improvement of the assembly quality of large-scale aviation components.

Claims

1. A method for predicting the docking quality of large aviation components with force-position coordination, characterized in that: Here are the steps: Step 1: Build the docking quality prediction hardware system and collect the docking process force and position data Before the experiment begins, each force sensor (3) is first installed and fixed on the supporting part of the positioner (2) to ensure that during the movement of the positioner (2), the force sensor (3) accurately measures the force on the supporting wing component; Then, the force sensor (3) is connected to the signal amplifier (5), and the mechanical signal converted by the signal amplifier (5) is connected to the programmable logic controller (6); at the same time, the position signal of the positioner servo motor is connected to the programmable logic controller (6), and the programmable logic controller (6) is connected to the attitude adjustment system industrial computer (7), and the communication between the attitude adjustment system industrial computer (7) and the data acquisition industrial computer (8) is established; finally, the force position data is collected and processed in real time by the data acquisition industrial computer (8); at the same time, the laser tracker (1) is arranged to ensure that the horizontal measurement point (4) on the supporting wing component can be measured, and the global coordinate system is established by measuring the ground ERS point; At the beginning of the experiment, the programmable logic controller (6), the attitude adjustment system industrial computer (7) and the data acquisition industrial computer (8) are started; the geometric position data of the horizontal measurement point (4) on the supporting wing component is measured by a laser tracker (1), and converted into the current six-dimensional attitude data of the wing component through SVD decomposition; then the transition path from the current attitude to the target attitude of the supporting wing component in the theoretical digital model is calculated by interpolation method, and the corresponding supporting wing component attitude conversion path data is generated; this attitude conversion path data describes how to gradually adjust the attitude of the supporting wing component through the positioner (2), and convert the current six-dimensional attitude data of the supporting wing component into the target attitude in the theoretical digital model; The attitude adjustment system industrial control computer (7) sends instructions to the programmable logic controller (6) to control the positioner (2) to drive the supporting wing component to perform attitude adjustment docking movement; during the movement, the data acquisition industrial control computer (8) collects three-dimensional force data of the force sensor (3) and position data of the positioner (2) in real time; the above attitude adjustment docking process is repeated m times to obtain a total of m groups of measurement data, and the collection of multiple groups of force and position data of the attitude adjustment docking process is completed; Step 2: Establish a prediction model for the docking quality of supporting wing components The data acquisition industrial control computer (8) stores multiple sets of three-dimensional force data and position data collected during the attitude adjustment docking process. By constructing a docking quality prediction model, the three-dimensional force data collected by the force sensor (3) and the position data of the positioner (2) are used as the input of the model, and whether the horizontal measurement point of the supporting wing component exceeds the tolerance range is used as the quality judgment standard. If it exceeds the tolerance range, it is considered that the assembly quality is unqualified, and the docking quality judgment result is used as the model output; finally, a nonlinear mapping relationship between the input x and the output y is established through a data-driven method: f:x→y (1) The construction process of the docking quality prediction model is as follows: Firstly, the grid search method is used to optimize the parameters of the docking quality prediction model, including duality selection, penalty coefficient, residual convergence condition and loss function. By searching different parameter combinations, the optimal parameter settings are determined to improve the performance of the docking quality prediction model in quality prediction in different docking processes. The optimized objective function is as follows: Among them, w is the optimal hyperplane slope, b is the optimal hyperplane intercept, C is the penalty coefficient, ξ i is a slack variable, k is the number of training samples, x i is the feature vector of the i-th training sample, y i is the label of the i-th training sample; in order to better fit the nonlinear mapping relationship between input data and output data, the following radial basis function kernel function is introduced: K(x i ,x j )=exp(-γ·||x i -x j || 2 ) (3) Among them, K(x i , x j ) is the kernel function, and γ is the kernel parameter; by mapping the original input data into a high-dimensional space, the complex nonlinear relationship between the input and output data can be captured more effectively; At the same time, by introducing the Lagrange multiplier, the original problem is transformed into its corresponding Lagrange function, and the corresponding dual function is constructed: Among them, α i , α j is the Lagrange multiplier vector. By solving the dual function, the optimal solution w of the original problem is obtained. * and b * as follows: Among them, α * is the optimal solution of α in the dual problem; grid search is used to optimize the hyperparameters C and γ to ensure the training effect of the docking quality prediction model. The final adjusted model and the corresponding prediction values ​​are as follows: in, The value of determines the predicted category of sample x, where represents the positive class, Represents the negative class; Step 3: Train the docking quality prediction model and perform docking quality prediction based on measured data First, mean filling is used to process variables with null values ​​in the force data, and the maximum and minimum values ​​of each dimension are normalized to complete the feature preprocessing process: The processed force and position data are input into the optimized prediction model for training, which can predict whether the posture adjustment and docking quality is qualified and meet the requirements of quality prediction in complex docking processes. The evaluation criteria for prediction accuracy are as follows: Among them, TP refers to the number of positive samples predicted correctly by the model, TN refers to the number of negative samples predicted correctly by the model, FP refers to the number of samples that the model mistakenly predicts as positive from negative classes, and FN refers to the number of samples that the model mistakenly predicts as negative from positive classes.

Citation Information

Patent Citations

  • Method and system for simulating ligand molecule and target receptor reaction and calculating and forecasting thermodynamics and kinetics parameters of reaction

    CN102930152A

  • Driving fatigue detection system and identification method based on eye movement index data

    CN106934368A

  • Intelligent recognition system and method for butt-joint state of mechanical components

    CN108596210A

  • A stress-strain prediction method based on machine learning

    CN109766618A

  • Local strain globalization prediction method based on a prediction model

    CN109918831A