A Real-time Sensing Method for the Global Strain State of Aircraft Tooling Positioners during the Assembly Process

Abstract

The present invention belongs to the technical field of intelligent manufacturing, and proposes a method for real-time perception of the global strain state of aircraft tooling locators during the assembly process. Perform finite element simulation on the locator, analyze the strain-sensitive areas under different loading conditions, and select strain measurement points in combination with the on-site measurement conditions; discretize the locator according to the finite element mesh, establish a non-linear mapping relationship between the strain measurement points and the strain values of the remaining nodes, and use the strain big data set to train the global strain real-time perception model; perform actual strain measurement at the strain measurement points, input the measured strain into the trained real-time perception model, and finally realize the real-time perception of the global strain of the aircraft tooling locator during the assembly process. This method can effectively use the big data set formed by simulation for the training of machine learning models, and organically combine the finite element method and machine learning, avoiding the need to establish a complex strain prediction mathematical model due to the complex structure of the locator, and solving the problem of insufficient real-time performance of global strain perception by the finite element method.

Description

Technical Field

[0001] The present invention relates to the field of intelligent manufacturing technology, and in particular to a real-time sensing method for the global strain state of an aircraft tooling locator during an assembly process. Background Art

[0002] As a key structure for the assembly and positioning of aircraft parts, the force and position state of the tooling directly affects the assembly accuracy of aircraft parts. In the assembly process, due to the inevitable manufacturing errors of components, forced assembly is very likely to occur, resulting in local stress concentration, deformation of the tooling locator, and loss of assembly accuracy. Therefore, in order to ensure that the aircraft assembly process meets various technical indicators, it is necessary to establish an effective tooling locator strain online monitoring system, analyze the strain field distribution state of the tooling locator, and then monitor the abnormal stress state of the locator in real time to guide the adjustment of the tooling and the implementation of aircraft assembly work. However, due to the complex aircraft assembly process and the diverse loading conditions of the locator, it is difficult to directly determine the fixed strain sensitive area to set the strain measurement points in a targeted manner. In addition, the structure of aircraft tooling components is complex and the number is large, resulting in a small and almost closed measurement space at the assembly site, making it difficult to set strain measurement points in all strain sensitive areas on the locator during the assembly process. The above series of problems make it difficult to effectively carry out strain monitoring of aircraft tooling locators during the assembly process. Therefore, it has become an urgent problem to realize the real-time perception of the global strain state of the locator during the assembly process to accurately evaluate the global stress-strain state of the locator under any loading state.

[0003] At present, common global strain state perception methods include finite element method and modal superposition method. In the paper "Predictions and measurements of residual stress in repair welds in plates" published in Volume 83, Issue 11-12 of the International Journal of Pressure Vessels and Piping in 2006, Brown et al. used ABAQUS simulation analysis software to establish a finite element model of the plate-type mailbox repair weld to predict the strain distribution of the weld, and verified the accuracy of the prediction by comparing it with the actual measurement results of the strain. However, the finite element method takes a long time to calculate, and it is difficult to achieve real-time perception of global strain for complex structural parts such as aircraft tooling locators. At the same time, since accurate boundary conditions cannot be obtained at the assembly site, the accuracy of global perception is difficult to guarantee.

[0004] In 2018, Yu et al. proposed a method for reconstructing the global strain field of a thin-walled disk during the machining process using the modal superposition method in the 23rd volume, issue 3 of the journal "IEEE / ASME Transactions on Mechatronics". This method converts the global strain information of the disk into the modal coordinate system, and uses displacement sensors to monitor the displacements of a limited number of points in real time. Then, the participation factors of each order of mode are calculated and the strain mode shapes of each order are weighted and superposed to achieve the purpose of reconstructing the global strain field. This method has high computational efficiency and good accuracy. However, different from the thin-walled disk, the structure of the fixture locator is complex, and the mathematical expressions of its strain mode shapes of each order will be very complex and cannot be accurately obtained. Therefore, this method cannot effectively realize the global strain perception of the locator.

[0005] In summary, the current global strain perception methods cannot achieve accurate and efficient perception effects for the fixture locator with complex structures, and it is difficult to be directly applied to the real-time perception of the strain state of the fixture during the aircraft assembly process. Summary of the Invention

[0006] The technical problem to be solved by the present invention is to overcome the defects of the prior art and invent a method for real-time perception of the global strain state of an aircraft fixture locator during the assembly process, which can use the measurement results of a small number of limited points to perceive the global strain of the aircraft fixture locator in real time.

[0007] The technical solution of the present invention:

[0008] A method for real-time perception of the global strain state of an aircraft fixture locator during the assembly process, which uses the measurement results of limited points to perceive the global strain of the aircraft fixture locator in real time; conducts finite element simulation on the aircraft fixture locator, analyzes the strain-sensitive areas of the aircraft fixture locator under different loading conditions, and combines the on-site measurement conditions to select the positions in the strain-sensitive areas where there is sufficient space to paste strain sensors as strain measurement points; discretizes the aircraft fixture locator according to the finite element mesh division, and selects an algorithm and parameters to establish a non-linear mapping relationship between the selected strain measurement points and the strain values of the remaining nodes, and uses the large dataset of strain states generated by the finite element simulation to train the global strain real-time perception model; pastes strain sensors at the selected strain measurement points of the aircraft fixture locator, conducts actual strain measurement, and inputs the measured strain into the trained global strain real-time perception model to finally realize the real-time perception of the global strain of the aircraft fixture locator during the assembly process; specifically includes the following steps:

[0009] Step 1: Simulation analysis of the aircraft fixture locator and construction of the large dataset of strain states

[0010] Analyze the loaded state of the aircraft tooling locator during the assembly process, conduct a large number of finite element simulation analyses under different loading conditions, a total of k groups, and determine the strain-sensitive area of the aircraft tooling locator according to the simulation results; according to the on-site measurement conditions, select the positions on the aircraft tooling locator where there is sufficient space to paste strain sensors as strain measurement points, use the simulated strain values of the nodes at the strain measurement points as inputs, and the strain values of all other nodes, i.e., the non-measurable points, as outputs to construct a large strain state dataset:

[0011] E={(ε ai ,ε bi )|i=1,2,…,k} (1)

[0012] where ε ai is the strain simulation result vector at the strain measurement point in the i-th group of simulations, and ε bi is the strain simulation result vector at the non-measurable points in the i-th group of simulations;

[0013] Step 2: Establishment of the global strain real-time perception model

[0014] According to the large strain state dataset established in Step 1, write a software program to obtain the non-linear mapping relationship between the strain values at the strain measurement points and the strain values at the non-measurable points, and establish a global strain real-time perception model. Using ε ai in the large strain state dataset as the input and the corresponding ε bi as the output, establish the non-linear mapping relationship between the two parts, that is:

[0015] ε bi =Π(ε ai ) (2)

[0016] To solve the above non-linear mapping relationship, introduce an optimization problem and establish a support vector machine model:

[0017]

[0018] where ω is the slope of the optimal hyperplane, b is the intercept of the optimal hyperplane, C is the penalty coefficient, and ξ i is the slack variable;

[0019] To better fit the non-linear mapping relationship between the strain at the strain measurement points and the strain at the non-measurable points, introduce the following RBF kernel function:

[0020] K(ε ai ,ε aj )=exp(-Υ·||ε ai -ε aj || 2 ) (4)

[0021] Υ is the gamma parameter in the RBF kernel function, and at the same time, the original optimization problem is transformed into the following dual problem:

[0022]

[0023] The optimal solution ω of the original optimization problem is obtained * and b* are as follows:

[0024]

[0025] where α * is the optimal solution of α in the dual problem. Grid search is used to optimize the penalty coefficient C and the gamma parameter Υ to ensure the training effect of the model. Finally, the real-time perception model of the global strain state is constructed as:

[0026]

[0027] where is the strain prediction result of the non-measurable point; by inputting the measured strain values of the strain measurement points, this model can realize the real-time perception of the global strain state of the locator.

[0028] Step 3: Real-time perception of the global strain based on the measured values of finite discrete strain measurement points

[0029] A distributed optical fiber strain monitoring system for the tooling locator is established to realize the real-time measurement of the strain signals of finite discrete strain measurement points on the aircraft tooling locator; specifically: the fiber Bragg grating strain sensor 1 is arranged at the strain measurement point position of the aircraft tooling locator 2, and the fiber Bragg grating strain sensor 1 is connected to the corresponding channel of the demodulator 3 to collect the change of the wavelength signal; the demodulator 3 is connected to the computer 4, which is used to read the wavelength change detected by each fiber Bragg grating strain sensor 1 and the corresponding strain measurement value in the corresponding software system; thus, a distributed optical fiber strain monitoring system for the tooling locator is established to realize the real-time measurement of the strain signals of finite discrete points on the locator. After the above sensor arrangement and wiring are completed, during the normal aircraft component assembly work, the fiber Bragg grating strain sensor 1 collects the strain values of the finite discrete strain measurement points during the assembly process and inputs them into the trained real-time perception model of the global strain state, and the model will output the global strain state of the aircraft tooling locator to realize the real-time perception of the global strain state of the aircraft tooling locator during the assembly process.

[0030] In order to evaluate the strain perception effect of this method, some parts to be assembled on the aircraft tooling 5 are removed to expand the measurement space, m fiber Bragg grating strain sensors are arranged in the original non-measurable area, and the simulated assembly is continued and the measured strain is compared with the strain value output by the perception model. With the absolute error E of each strain measurement point AEvaluate the accuracy of the global strain real-time perception model as shown in Equation (8):

[0031]

[0032] Where ε si is the measured strain value, is the strain value output by the global strain real-time perception model.

[0033] This method combines the finite element method and machine learning. By conducting a large number of finite element simulations on the locator under various different load conditions, a strain big data set is formed. Appropriate machine learning algorithms and training parameters are selected to establish a prediction model, so as to realize the real-time perception of the global strain field through the strain values of the measuring points. This method can effectively use the big data set formed by the simulation for the training of the machine learning model, and organically combine the finite element method and machine learning, avoiding the need to establish a complex strain prediction mathematical model due to the complex structure of the locator, and solving the problem of insufficient real-time performance of the single finite element method for global strain perception.

[0034] The beneficial effects of the present invention are as follows: Compared with other existing global strain state prediction methods, this method effectively combines machine learning and finite element analysis, avoiding the problems brought by the uncertain factors in the establishment of complex mathematical models and mechanism analysis processes, solving the problem that it is difficult to ensure the real-time performance of global strain perception in finite element analysis, and can real-time perceive the global strain state of the aircraft tooling locator during the assembly process. Only by conducting a large number of finite element simulation analyses on the locator, establishing a suitable support vector machine model, conducting effective machine learning, obtaining a global perception model, and then inputting the strain values measured by the finite discrete point strain sensors during the assembly process, the real-time perception of the global strain state of the locator can be realized. This method is applicable to the real-time perception of the global strain field of the aircraft tooling locator under any assembly working conditions, and at the same time, the operation is relatively simple and convenient, and it is easy to promote. Description of the Drawings

[0035] Figure 1 It is a schematic diagram of the loads on the aircraft tooling during the assembly process.

[0036] Figure 2 It is a schematic diagram of the loads on the aircraft tooling locator during the assembly process.

[0037] Figure 3 It is a schematic diagram of the distributed optical fiber strain monitoring system for the aircraft tooling locator.

[0038] Figure 4 It is a flow chart of the method for real-time perception of the global strain state of the aircraft tooling locator during the assembly process.

[0039] Figure 5 It is a schematic diagram of the simulation loading method for the aircraft tooling locator.

[0040] Figure 6 It is a schematic diagram for the selection of strain measurement points of the strain sensor.

[0041] Figure 7 It is a curve graph of the predicted result of the application model.

[0042] In the figure: 1 - Fiber Bragg grating strain sensor, 2 - Aircraft tooling locator, 3 - Demodulator, 4 - Computer, 5 - Aircraft tooling. Specific implementation manner

[0043] The specific implementation manner of the present invention will be described in detail below in combination with the technical solution and the drawings.

[0044] In this embodiment, the used demodulator 3 is the si255 - 16 - ST / 160 - NO fiber Bragg grating demodulator of MOI Company, with a strain detection range of - 15000 - 15000 με, a demodulation accuracy of 1 pm and a resolution of 0.5 με. The used fiber Bragg grating strain sensor 1 is the JMFSS - 01 fiber Bragg grating strain sensor of Shenzhen Jiance Company, with a full range of ±2000 με and a resolution of 0.5 με. The global strain state perception experiment of the locator was carried out on the horizontal tail elevator tooling of a certain type of aircraft.

[0045] Step 1: Simulation analysis of the aircraft tooling locator and construction of the strain state big data set

[0046] In this embodiment: First, considering the influence of the gravity of different components and the drilling force during drilling and riveting on the tooling and the locator during the assembly process as shown in Figure 1 and Figure 2 , a large number of finite element simulations were carried out on the tooling locator. As shown in Figure 5 , 3 point positions, 3 line positions and 3 surface positions were set on the locator, and point loading, line loading, surface loading and mixed loading were carried out respectively, and finally a big data set composed of 293895 groups of simulation results was formed. The specific loading forms are shown in Table 1.

[0047] Table 1 Loading configuration method for positioning simulation

[0048]

[0049] After obtaining the simulation results, according to the simulation results and considering the on - site measurement conditions (whether there is enough space to arrange sensors), appropriate strain measurement points were selected, as shown in Figure 6 , and based on this, a big data set of the strain distribution state of the locator was constructed:

[0050] E = {(ε ai , ε bi )|i = 1, 2, …, 293895} (9)

[0051] where ε ai is the strain simulation result vector at the measurement point in the i-th group of simulations, and ε bi is the strain simulation result vector at the non-measurement point in the i-th group of simulations.

[0052] Step 2: Establishment of the global strain real-time perception model

[0053] Relying on the large dataset of the strain state of the tooling locator established in Step 1, write a software program to solve the non-linear mapping relationship between the strain values at the measurement points and the strain values at the non-measurement points, and establish a global strain real-time perception model. Taking ε ai in the large dataset as the input and the corresponding ε bi as the output, establish the non-linear mapping relationship between the two parts, that is:

[0054] ε bi = Π(ε ai ) (10)

[0055] To solve the above non-linear mapping relationship, introduce the following optimization problem, that is, establish a support vector machine model:

[0056]

[0057] where ω is the slope of the optimal hyperplane, b is the intercept of the optimal hyperplane, C is the penalty coefficient, and ξ i is the slack variable. To better fit the non-linear mapping relationship between the strains at the measurement points and the non-measurement points, introduce the following RBF kernel function:

[0058] K(ε ai , ε aj ) = exp(-Υ·||ε ai - ε aj || 2 ) (12)

[0059] where Υ is the gamma parameter in the kernel function. At the same time, convert the original optimization problem into the following dual problem:

[0060]

[0061] Furthermore, the optimal solutions ω * and b* of the original problem can be solved as follows:

[0062]

[0063]

[0064] where α *is the optimal solution of α in the dual problem. Grid search is used to optimize the penalty coefficient C and the gamma parameter Υ. After grid search, C = 99.85 and Υ = 1 are selected to train the support vector machine model. Finally, the real-time perception model of the global strain state of the locator is constructed as follows:

[0065]

[0066] Among them, is the strain prediction result of the non-measurable point, and the measured strain result of the input data measurement point. This model can realize the real-time perception of the global strain state of the locator. In this embodiment, 50 simulation result groups are selected from 293,895 groups of simulation data as the test set, and the remaining 293,845 groups of simulation results are used as the training set. The program is run to train the support vector machine model, and finally the real-time perception model of the global strain state of the aircraft tooling locator is obtained.

[0067] Step 3: Real-time global strain perception based on the measured strain of finite discrete points

[0068] According to the selected measuring point positions in Step 1, the fiber Bragg grating strain sensor 1 is arranged at the corresponding positions of the aircraft tooling locator 2. At the same time, the fiber Bragg grating strain sensor 1 is connected to the corresponding channels of the demodulator 3 to collect the change of the wavelength signal. The demodulator 3 is connected to the computer 4, which is used to read the wavelength change detected by each fiber Bragg grating strain sensor 1 and the corresponding strain measurement value in the corresponding software system. Thus, a distributed fiber optic strain monitoring system for the tooling locator is established, as Figure 3 shown, to realize the real-time measurement of the strain signals of finite discrete points on the locator. After the above sensor arrangement and wiring are completed, the aircraft component assembly work can be carried out normally. During this period, the fiber Bragg grating sensor collects the strain values of the finite discrete measurement points during the assembly process and inputs them into the trained real-time perception model of the global strain state. The model will output the global strain state of the locator. Thus, the real-time perception of the global strain state of the aircraft tooling locator during the assembly process is realized, and the average single global perception time does not exceed 0.48 s, meeting the real-time requirement.

[0069] In order to evaluate the effect of strain perception in this embodiment, some of the parts to be assembled on the tooling are removed to expand the measurement space. 8 fiber Bragg grating strain sensors are arranged in the original non-measurable area, and the simulation assembly is continued. The measured strain is compared with the strain value output by the perception model. The absolute error E of each measuring point A is used to evaluate the accuracy of the perception model, as shown in formula (8):

[0070]

[0071] Among them, ε si is the measured strain value, is the strain value output by the sensing model. The final results are as Figure 7 shown. The three curves in the figure represent the measured strain, the predicted strain output by the sensing model, and the absolute error of the model.

[0072] This method can be generally applied to the real-time sensing of the global strain state of the tooling locator under various assembly conditions, with high efficiency, strong universality, simple and convenient operation, and easy to promote.

Claims

1. A real-time perception method for the global strain state of an aircraft tooling locator during the assembly process, characterized in that, The real-time sensing method for the global strain state of the aircraft tooling locator in the assembly process uses the measurement results of finite points to sense the global strain of the aircraft tooling locator in real time; Perform finite element simulation on the aircraft tooling locator, analyze the strain-sensitive areas of the aircraft tooling locator under different loading conditions, and select the positions with sufficient space on the strain-sensitive areas to paste strain sensors as strain measurement points in combination with the on-site measurement conditions; Discretize the aircraft tooling locator according to the finite element mesh division, select the algorithm and parameters to establish the non-linear mapping relationship between the selected strain measurement points and the strain values of the remaining nodes, and use the large dataset of strain states generated by the finite element simulation to train the global strain real-time sensing model; Paste strain sensors at the selected strain measurement points of the aircraft tooling locator, conduct actual strain measurement, input the measured strain into the trained global strain real-time sensing model, and finally realize the real-time sensing of the global strain of the aircraft tooling locator during the assembly process; specifically including the following steps: Step 1: Simulation analysis of the aircraft tooling locator and construction of the large dataset of strain states Analyze the loading state of the aircraft tooling locator during the assembly process, conduct a large number of finite element simulation analyses under different loading conditions, a total of k groups, and determine the strain-sensitive areas of the aircraft tooling locator according to the simulation results; According to the on-site measurement conditions, select the positions on the aircraft tooling locator with sufficient space to paste strain sensors as strain measurement points, and construct a large dataset of strain states with the simulated strain values of the nodes at the strain measurement points as the input and the strain values of all the remaining nodes, that is, the non-measurement points, as the output; E = {(ε ai , ε bi ) | i = 1, 2, …, k} (1) where ε ai is the strain simulation result vector at the strain measurement point in the i-th group of simulations, and ε bi is the strain simulation result vector at the non-measurable point in the i-th group of simulations; Step 2: Establishment of the global strain real-time sensing model Based on the large strain state dataset established in step 1, obtain the non-linear mapping relationship between the strain values of the strain measurement points and the strain values of the non-measurable points, and establish a global real-time strain perception model. Using ε ai in the large strain state dataset as the input, and the corresponding ε bi as the output, establish the non-linear mapping relationship between the two parts, that is: ε bi = Π(ε ai ) (2) To solve the above non-linear mapping relationship, introduce an optimization problem and establish a support vector machine model; where ω is the slope of the optimal hyperplane, b is the intercept of the optimal hyperplane, C is the penalty coefficient, and ξ i is the slack variable; To better fit the non-linear mapping relationship between the strain measurement points and the strains of the non-measurement points, introduce the following RBF kernel function: K(ε ai , ε aj ) = exp(-Υ·||ε ai - ε aj || 2 ) (4) Υ is the gamma parameter in the RBF kernel function, and at the same time convert the original optimization problem into the following dual problem: The optimal solution ω of the original optimization problem is obtained as follows: * and b* are as follows: where α * is the optimal solution of α in the dual problem. Grid search is used to optimize the penalty coefficient C and the gamma parameter Υ to ensure the training effect of the model. Finally, the real-time perception model of the global strain state is constructed as: Among them, is the strain prediction result of the non-measurable point; Step 3: Real-time sensing of the global strain based on the actual measurement of finite discrete strain measurement points Establish a distributed optical fiber strain monitoring system for the tooling locator to realize the real-time measurement of the strain signals of the finite discrete strain measurement points on the aircraft tooling locator; specifically: arrange the fiber Bragg grating strain sensor (1) at the strain measurement point position of the aircraft tooling locator (2), connect the fiber Bragg grating strain sensor (1) to the corresponding channel of the demodulator (3), and collect the change of the wavelength signal; the demodulator (3) is connected to the computer (4) for reading the wavelength change detected by each fiber Bragg grating strain sensor (1) and the corresponding strain measurement value in the corresponding software system; during the normal aircraft component assembly work, the fiber Bragg grating strain sensor (1) collects the strain values of the finite discrete strain measurement points during the assembly process and inputs them into the trained global strain state real-time sensing model, and the model will output the global strain state of the aircraft tooling locator, realizing the real-time sensing of the global strain state of the aircraft tooling locator during the assembly process.

Images