Method for measuring binding pre-tightening force of multilayer binding container based on artificial intelligence algorithm

Abstract

A method for measuring binding pre-tightening force of a multi-layer binding container based on an artificial intelligence algorithm comprises the following steps: firstly, measuring part of point position stress in a delivery test process of the multi-layer binding high-pressure container, then establishing a corresponding finite element model, and calculating the binding pre-tightening force of the multi-layer binding high-pressure container by taking a simulation result close to an actual measurement result of a corresponding part as a target; the method comprises the following steps: fitting a basic function form of pre-tightening force distribution between laminates, enveloping an actual measurement result by using a simulation result after the basic function form is obtained, determining a function coefficient value range, sampling by using a Latin super-cubic method, establishing a coefficient-stress database, and obtaining a pre-tightening force distribution model based on an agent model between a training coefficient and stress. And finally, obtaining an accurate distribution function of the binding pre-tightening force of the multi-layer container by taking an actual measurement result as input and the function coefficient as output. According to the invention, accurate measurement of the binding pre-tightening force is realized, so that accurate boundary and load conditions are applied in finite analysis, and the purpose of reproducing stress distribution when the multilayer binding container is loaded is achieved.

Description

Technical Field

[0001] The present invention belongs to the field of pressure vessel design and analysis, and in particular to the technical field of application of emerging NEW IT technology in pressure vessel design and analysis. Specifically, it is a method for measuring the wrapping preload force of a multi-layer wrapping container based on an artificial intelligence algorithm. Background Art

[0002] Multi-layer wrapped containers have become the main choice for high-pressure hydrogen storage due to their low processing and manufacturing costs, failure mode of leakage but not explosion, and relatively dispersed defects. However, multi-layer wrapped containers still face problems such as unclear calculation and transfer theory of wrapping prestress, unclear factors affecting the ultimate load-bearing capacity and safety performance of the container, and unclear weld fatigue damage characteristics and supervision and inspection points. If the stress distribution of multi-layer wrapped containers under load can be accurately reproduced, its potential weak points can be analyzed, thereby improving the production process, achieving iterative upgrades of products, and formulating targeted supervision and inspection strategies to ensure the safety of its service process. However, due to the difficulty in precise control of the robotic arm during the wrapping process and the randomness of the production and manufacturing process, the wrapping preload between the layers cannot be accurately described, resulting in the inability to calculate the stress results in theoretical analysis, resulting in a large deviation between the finite element analysis results and the actual stress distribution. Summary of the Invention

[0003] The present invention provides a method for measuring the wrapping preload of a multi-layered wrapping container based on an artificial intelligence algorithm, which applies accurate boundary and load conditions in a finite analysis to achieve the purpose of reproducing the stress distribution of the multi-layered wrapping container when it is loaded.

[0004] The technical solution adopted in the present invention is: A method for measuring the preload force of a multi-layered wrapping container based on an artificial intelligence algorithm includes three stages. The first stage aims to match the trends of finite element simulation results and measured results, and fits the basic functional relationship obeyed by the preload force between the layers of the multi-layered wrapping container. The second stage is database construction and proxy model training. First, based on the principle that the stress simulation results of each measuring point envelop the measured results, the range of the function coefficients determined in the first stage is determined, and sampling is performed using the Latin hypercube sampling method to construct the preload force distribution function relationship of the corresponding number of groups. Then, different preload forces are applied between the layers in the finite element model, the stress of each measuring point is calculated, and the relationship is established. Preload function-stress database; finally, the constructed data is used to train the proxy model between the preload function and stress, and the iteration is repeatedly adjusted until the root mean square error and determination coefficient of the proxy model output meet the requirements; the third stage is the wrapping preload measurement, and the measured stress results are input into the proxy model trained in the second stage to obtain the preload distribution function relationship, which is input into the finite element model as a boundary condition to calculate the simulated stress of each measuring point. If the error between the simulation result and the measured result is less than the specified value, it is considered that the accurate preload distribution is obtained, otherwise the preload function distribution form, sampling results and proxy model are adjusted, and the above process is repeated until the error meets the requirements.

[0005] The basic functional relationship used in the first stage is: For a certain layer of the multi-layer container cylinder, it is assumed that the wrapping preload it bears is symmetrically distributed along the circumference of the cylinder and uniformly attenuates, with the maximum and minimum values at the longitudinal weld and the 180° angle position of the longitudinal weld respectively. For different layers, as the inner diameter increases, the preload between adjacent layers is also considered to gradually attenuate, and the attenuation coefficient can be obtained according to the design theory; therefore, a cylindrical coordinate system is established with the container axial direction as the Z axis, the radial direction as the r axis, and the circumferential direction as the θ axis. The basic function form of the interlayer wrapping preload P', unit: MPa, is set as: P'= f (θ)* μ n , 0°≤θ≤180°, μ is the interlayer preload attenuation coefficient, n is the number of layers; f The specific form of the (θ) function was fitted with the goal of matching the trends of the finite element simulation results with the measured results.

[0006] described f(θ) Function fitting method: First, apply a uniformly distributed preload between each layer, with the goal of enveloping the test results of all measuring points, determine the maximum, minimum and mean values of the preload, and then try to fit different forms of functions. After obtaining a function form that meets the requirements, apply the preload between the layers in the form of a function, and calculate the stress results of the corresponding measuring points by the finite element method. If the theory and simulation are close, it is determined as the basic function form for the subsequent establishment of the database, otherwise try a new function form.

[0007] The interlayer wrapping preload function is in the form of:

[0008] Where P' is the pressure generated by the wrapping pre-tightening force, a 1 、a 2 、a 3 、a 4 are function coefficients, μ represents the preload attenuation coefficient between laminates, n is the current number of layers, then the parameter vector to be corrected for the function is A=[ a 1 a 2 a 3 a 4 μ ], set A=[ a 1 a 2 a 3 a 4 μ ] The value ranges are: a 1 ∈(0, 1×10 5 ), a 2 ∈[-5.5×10 5 , 5×10 5 ]、 a 3 ∈[0, 0.005], a 4 ∈[0.2, 1], μ ∈[0.8, 1.0].

[0009] The proxy model used in the second stage is: Gaussian radial basis function-back propagation neural network GRBF-BPNN, whose activation function is a mixed Sigmoid function embedded in a Gaussian function, in the following form:

[0010] Where x, , σ represent the input sample and the mean and variance of the sample respectively.

[0011] This method, using artificial intelligence algorithms and measured local stresses in loaded multilayer containers, enables precise measurement of the wrapping preload. This allows accurate boundary and loading conditions to be applied in a finite element analysis, reproducing the stress distribution of loaded multilayer containers. This method provides technical support for studying stress distribution patterns and analyzing potential failure modes in multilayer containers. It also establishes a mapping relationship between processing parameters and stress distribution, facilitating improvements in production processes and product upgrades. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] Figure 1 It is a structural block diagram of the present invention; Figure 2 Schematic diagram of the arrangement of measuring points for the water pressure test of a multi-layer wrapped container according to the present invention; Figure 3 for Figure 2 Middle BB view; Figure 4 for Figure 2 AA view in the; Figure 5 A schematic diagram of a partial layer plate of a container in the present invention; Figure 6 for Figure 5 Schematic diagram of the pre-tightening force of the container's laminate wrapping in various directions; (a) n Laminate q Distribution of preload force of the cylinder segment A'-A' wrapping; (b) n Laminate q Distribution of preload force of cylinder segment AA wrapping; (c) n -1 layer board q Distribution of preload force of barrel section B'-B' wrapping; (d) n -1 layer board q Distribution of pre-tightening force of barrel section BB wrapping; Figure 7 A finite element model of the cylinder section for analysis of the present invention; Figure 8 is the local coordinate system of the present invention; Figure 9 This is a sample function curve of the multi-layer wrapping pre-tightening force of the present invention; Figure 10The result of applying pre-tightening force to a certain layer of the present invention; Figure 11 The comparison results of the measured and simulated stresses of the present invention at 21.10 MPa are shown; Figure 12 The comparison results of the measured and simulated stresses at 25.50 MPa are shown in the figure. Figure 13 This is the comparison result of the measured and simulated stress at 26.00 MPa in the present invention. DETAILED DESCRIPTION

[0013] The present invention will be described in further detail below with reference to the accompanying drawings.

[0014] Reference Figure 1 A method for measuring the preload force of a multi-layered wrapping container based on an artificial intelligence algorithm is proposed. The method includes three stages. The first stage aims to match the trend of finite element simulation results with the measured results, and fits the basic functional relationship obeyed by the preload force between the layers of the multi-layered wrapping container. The second stage is database construction and proxy model training. First, based on the principle that the stress simulation results of each measuring point envelop the measured results, the range of the function coefficients determined in the first stage is determined, and sampling is performed using the Latin hypercube sampling method to construct the preload force distribution function relationship of the corresponding number of groups. Then, different preload forces are applied between the layers in the finite element model, the stress of each measuring point is calculated, and the preload force is established. Function-stress database; finally, the constructed data is used to train the proxy model between the preload function and stress, and the iteration is repeatedly adjusted until the root mean square error and determination coefficient of the proxy model output meet the requirements; the third stage is to measure the actual distribution function of the wrapping preload, and the measured stress results are input into the proxy model trained in the second stage to obtain the preload distribution function relationship, which is input into the finite element model as a boundary condition to calculate the simulated stress of each measuring point. If the error between the simulation result and the measured result is less than the specified value, it is considered that the accurate preload distribution is obtained, otherwise the preload function distribution form, sampling results and proxy model are adjusted, and the above process is repeated until the error meets the requirements.

[0015] The basic functional relationship used in the first stage is: the distribution of circumferential and longitudinal welds of multi-layer container cylinders is generally more complicated, the axial horizontal distance between adjacent circumferential welds is generally greater than 200mm, and adjacent longitudinal welds are staggered at a certain angle, so the wrapping preload force borne by each section of the plate after wrapping is different. For a certain layer of plate, it can be assumed that the wrapping preload force it bears is symmetrically distributed along the circumference of the cylinder and uniformly attenuates, with maximum and minimum values at the longitudinal weld and the 180° angle position of the longitudinal weld respectively. For different layers, as the inner diameter increases, the preload force between adjacent layers is also considered to gradually attenuate, and the attenuation coefficient can be obtained according to the design theory; therefore, a cylindrical coordinate system is established with the container axial direction as the Z axis, the radial direction as the r axis, and the circumferential direction as the θ axis, then the basic functional form of the interlayer wrapping preload force P' (MPa) can be set as: P'= f (θ)* μ n , 0°≤θ≤180°, μ is the interlayer preload attenuation coefficient, n is the number of layers; f The specific form of the (θ) function was fitted with the goal of matching the trends of the finite element simulation results with the measured results.

[0016] described f (θ) Function fitting method: First, apply a uniformly distributed preload between each layer, with the goal of enveloping the test results of all measuring points, determine the maximum, minimum and mean values of the preload, and then try to fit different forms of functions. After obtaining a function form that meets the requirements, apply the preload between the layers in the form of a function, and calculate the stress results of the corresponding measuring points by the finite element method. If the theory and simulation are close, it is determined as the basic function form for the subsequent establishment of the database, otherwise try a new function form.

[0017] The interlayer wrapping preload function is:

[0018] Where P' is the pressure generated by the wrapping pre-tightening force, a 1 、a 2 、a 3 、a 4 are the function coefficients, μ represents the preload attenuation coefficient between laminates, n is the current number of layers, then the parameter vector to be corrected for the function is A=[ a 1 a 2 a 3 a 4 μ ], set A=[ a 1 a 2 a 3 a 4 μ ] The value ranges are: a 1 ∈(0, 1×10 -5 ), a 2 ∈[-5.5×10 -5 , 5×10 -5 ]、 a 3 ∈[0, 0.005], a 4 ∈ [0.2, 1], μ ∈[0.8, 1.0].

[0019] The proxy model used in the second stage is the Gaussian Radial Basis Function-Back Propagation Neural Network (GRBF-BPNN), whose activation function is a hybrid Sigmoid function embedded in a Gaussian function. This function can retain the global nonlinearity of the Sigmoid function while introducing local nonlinearity through the Gaussian function to enhance the model's ability to fit strong nonlinear relationships and its robustness. The hybrid Sigmoid function embedded in the Gaussian function is expressed as follows:

[0020] In the formula x、 , σ Represent the input sample, the mean value and the variance of the sample respectively.

[0021] To facilitate a better understanding of the present invention by those skilled in the art, the following further describes the present invention with reference to the accompanying drawings and a technical solution for measuring the preload force of a cylinder section of a multi-layer high-pressure reactor. Measurement solutions employing different proxy models also fall within the scope of protection of the present invention. The term "pressure vessel" herein refers to a sealed container for containing fluid media as defined in GB / T 26929.

[0022] Step 1: Water pressure gradient test and strain measurement of multi-layer wrapped container First, a water pressure gradient test is carried out on a multi-layer high-pressure reactor, and the strain at the corresponding measuring point is measured in real time using a dynamic strain gauge. The equivalent stress at each measuring point is calculated based on the three-dimensional strain results. The schematic diagram of the measuring point arrangement is shown in the figure. Figure 2 、 Figure 3 、 Figure 4 shown.

[0023] The third section of the multi-layer wrapped container (i.e., the section where measuring points 4 to 9 are located) was taken as the research object to measure its wrapping preload. Based on the measured strain, the equivalent stress results of measuring points 4 to 9 under internal pressures of 21.10 MPa, 25.50 MPa, and 26.00 MPa were calculated, as shown in Table 1.

[0024] Table 1 Equivalent stress at measuring points 4 to 9 (MPa)

[0025] Step 2: Establish the finite element model of the corresponding cylinder section of the multi-layer wrapped container The local section of the cylinder where measuring points 4 to 9 are located is as follows Figure 5 As shown in Figure 2. As far as the wrapping preload is concerned, the force on the nth layer of laminate is determined by the n +1 layer of laminate is determined, so n When the layer-by-layer plate is meshed, the plate area is cut by n The prestress distribution of adjacent cylinder segments across different thicknesses is affected by the longitudinal welds between the layers, resulting in a 98° offset. However, the prestress distribution of adjacent cylinder segments within the same layer is offset by 180°. Furthermore, the circumferential weld distance between the layers is 200mm, resulting in different prestress distribution trends within the same cylinder segment. Figure 6 (b) and (c) show the n Layered board q The distribution of preload force along the A'-A' and AA sections of the cylinder segment, Figure 6 (d) and (e) show the n Preload distribution along sections B'-B' and BB for the -1 layer q cylinder segment. The preload reaches its maximum value at the longitudinal weld. Assuming it decreases uniformly along the circumference of the cylinder, the preload reaches its minimum value at the 180° angle to the longitudinal weld.

[0026] Based on its stress characteristics, the cylinder section where measuring points 4 to 9 are located is divided and the finite element model is established as follows: Figure 7 As shown. Two cylindrical local coordinate systems are established in each longitudinal weld, as shown Figure 8 As shown in the figure, the y-axis corresponds to the θ-axis, 0°≤θ≤180°, ensuring that the preload force is symmetrically distributed based on the weld.

[0027] Step 3: Measure the preload function with the goal of simulating the stress envelope test results exist Figure 7 A uniformly distributed constant preload is applied between the laminates of the finite element model shown in the figure. The corresponding pressure curve of the water pressure gradient test is set on the inner cylinder wall. With the goal of enveloping the test results in Table 1, the maximum, minimum, and average values of the preload under the stress simulation results are first determined, which are 1.2 MPa, 0.2 MPa, and 0.7 MPa, respectively.

[0028] Based on this, different function forms were fitted, and linear functions, polynomial functions, exponential functions, etc. were tried to explore the adaptability of different functions in describing the relationship between preload and angle corresponding load. The load variation curve fitted by the selected function was strived to reflect the true trend presented by the experimental data to the greatest extent. After multiple simulations and calculations, the basic function form finally selected was a cubic polynomial. The preload function form of the cylinder segment wrapping is:

[0029] Where P' is the pressure generated by the wrapping pre-tightening force, a 1 、a 2 、a 3 、a 4 are function coefficients, μ represents the preload attenuation coefficient between laminates, n is the current number of layers, then the parameter vector to be corrected for the function is A=[ a 1 a 2 a 3 a 4 μ ].

[0030] Step 4: Sample the function coefficients according to the Latin hypercube method and establish a database between the preload function and the stress at the measuring point Based on the maximum, minimum and mean values of the wrapping preload determined in step 3, the value range of each parameter in formula (1) can be estimated. This range can be dynamically adjusted. The size of the range will control the number of samples of the envelope function parameters. Initially set A=[ a 1 a 2 a 3 a 4 μ ] The value ranges are: a 1 ∈(0, 1×10 -5 ), a 2 ∈[-5.5×10 -5 , 5×10 -5 ]、 a 3 ∈[0, 0.005], a 4∈ [0.2, 1.2], μ ∈[0.8, 1.0].

[0031] When sampling, we need to set the sample screening criteria, such as the monotonicity, concavity, rate of change, severity of fluctuation and range of function value. Based on this, we use the Latin hypercube method to sample and generate 200 groups of functions that meet the conditions, such as Figure 9 shown.

[0032] exist Figure 7 The preload is applied between the layers of the finite element model shown in the figure according to the function form of formula (1), and the result of applying a certain layer is as follows: Figure 10 shown.

[0033] Will Figure 9 The functions shown are applied as load conditions in sequence, the simulation results of measuring points 4 to 9 are calculated, and the “wrapping preload function-measuring point stress” database is constructed.

[0034] Step 5: Proxy model building and function coefficient inversion Based on the database constructed in step 4, the GRBF-BPNN model is trained as a proxy model between the “wrapping preload function-measurement point stress”, and the mean absolute error (MAE) and root mean square error (RMSE) are close to zero, and the coefficient of determination (R 2 ) approaches 1 as the evaluation standard for the accuracy of the trained model. After obtaining the proxy model that meets the accuracy requirements, the measured data of measuring points 4 to 9 under 21.1MPa are used as input, and the preload function coefficient of the cylinder segment is measured as follows: A=[ a 1 a 2 a 3 a 4 μ ] =[5.87828028e-07, -5.12347894e-05, 2.83786654e-03, 1.06075147, 0.84] Step 6: Verification of preload function inversion results Apply the function obtained in step 5 as a boundary condition to Figure 7 Calculate the equivalent stress results of measuring points 4-9 under internal pressures of 21.10MPa, 25.50MPa and 26.00MPa. The comparison results of the measured stress at each measuring point under different internal pressures and the stress before and after considering the preload are as follows: Figure 11-13 As shown in the figure, σ1, σ2 and σ3 represent the measured stress results, the simulation results without considering the bandage preload and the simulation results with the bandage preload respectively; ε1 means that the error between the simulation result and the measured result of the wrapping preload force is not considered. ε 2 represents the error between the simulation result and the measured result after considering the wrapping preload obtained in step 5.

[0035] Depend on Figure 11-13 It can be seen that the measured stress results of measuring points 4-9 at each pressure level are significantly different, that is, the stress of the cylinder of the multi-layer container after loading shows a certain degree of randomness. When the wrapping preload is not considered, the simulation results of each measuring point are basically the same, and the error with the measured results is about 20%-50%. When the preload distribution function obtained in step 5 is used as the boundary condition, the simulation results show randomness. Under 21.10MPa ( Figure 11 The error between the simulated and measured results at each measuring point was less than 10%. At 25.50 MPa and 26.00 MPa, the error between the simulated and measured results for measuring points 5-8 was reduced to 15%, while the error for measuring points 4 and 9 was as low as 5%. In summary, after inverting the wrapping preload force distribution function of a multi-layer container using the proposed method, the simulated stress distribution results exhibited a randomized nature consistent with actual conditions. The error in the simulated stress at each measuring point was reduced from 20%-50% to 5%-15%. Considering the inevitable errors in the measurement process, the proposed method for measuring the wrapping preload force distribution of a multi-layer container is considered effective.

Claims

1. A method for measuring the preload force of a multi-layered wrapping container based on an artificial intelligence algorithm, characterized by: The method for measuring the preload force of a multi-layered wrapping container includes three stages. The first stage aims to match the trend of the finite element simulation results with the measured results, and fits the basic functional relationship obeyed by the preload force between the layers of the multi-layered wrapping container. The second stage involves database construction and proxy model training. First, based on the principle that the stress simulation results at each measuring point envelop the measured results, the range of the function coefficients determined in the first stage is determined. Samples are then drawn using the Latin hypercube sampling method to construct a corresponding number of sets of preload distribution function equations. Different preloads are then applied between the plies in the finite element model, and the stress at each measuring point is calculated to establish a preload function-stress database. Finally, the constructed data is used to train a proxy model between the preload function and stress. Adjustments are made repeatedly until the root mean square error (RMS) and coefficient of determination (CDR) of the proxy model output meet the required values. The third stage involves wrapping preload measurement. The measured stress results are input into the proxy model trained in the second stage to obtain the preload distribution function equation. This equation is then used as a boundary condition in the finite element model to calculate the simulated stress at each measuring point. If the error between the simulated and measured results is less than a specified value, the preload distribution is considered accurate. Otherwise, the preload function distribution form, sampling results, and proxy model are adjusted, and the above process is repeated until the error meets the required value.

2. The method for measuring the preload force of a multi-layered wrapping container based on an artificial intelligence algorithm according to claim 1, characterized in that: The basic functional relationship used in the first stage is: For a certain layer of the multi-layer container cylinder, it is assumed that the wrapping preload it bears is symmetrically distributed along the circumference of the cylinder and uniformly attenuates, with the maximum and minimum values at the longitudinal weld and the 180° angle position of the longitudinal weld respectively. For different layers, as the inner diameter increases, the preload between adjacent layers is also considered to gradually attenuate, and the attenuation coefficient can be obtained according to the design theory; therefore, a cylindrical coordinate system is established with the container axial direction as the Z axis, the radial direction as the r axis, and the circumferential direction as the θ axis. The basic function form of the interlayer wrapping preload P', unit: MPa, is set as: P'= f (θ)* μ n , 0°≤θ≤180°, μ is the interlayer preload attenuation coefficient, n is the number of layers; f The specific form of the (θ) function was fitted with the goal of matching the trends of the finite element simulation results with the measured results.

3. The method for inverting the preload force of a multi-layered wrapping container based on an artificial intelligence algorithm according to claim 2 is characterized in that: described f (θ) Function fitting method: First, apply a uniformly distributed preload between each layer, with the goal of enveloping the test results of all measuring points, determine the maximum, minimum and mean values of the preload, and then try to fit different forms of functions. After obtaining a function form that meets the requirements, apply the preload between the layers in the form of a function, and calculate the stress results of the corresponding measuring points by the finite element method. If the theory and simulation are close, it is determined as the basic function form for the subsequent establishment of the database, otherwise try a new function form.

4. The method for inverting the preload force of a multi-layered wrapped container based on an artificial intelligence algorithm according to claim 2 or 3, characterized in that: The interlayer wrapping preload function is:

5. Where P' is the pressure generated by the wrapping pre-tightening force, a 1 、a 2 、a 3 、a 4 are function coefficients, μ represents the preload attenuation coefficient between laminates, n is the current number of layers, then the parameter vector to be corrected for the function is A=[ a 1 a 2 a 3 a 4 μ ], set A=[ a 1 a 2 a 3 a 4 μ ] The value ranges are: a 1 ∈(0, 1×10 5 ), a 2 ∈[-5.5×10 5 , 5×10 5 ]、 a 3 ∈[0,0.005], a 4 ∈ [0.2, 1], μ ∈[0.8, 1.0].

6. The method for measuring the pre-tightening force of a multi-layered wrapping container based on an artificial intelligence algorithm according to claim 1, characterized in that: The proxy model used in the second stage is: Gaussian radial basis function-back propagation neural network GRBF-BPNN, whose activation function is a mixed Sigmoid function embedded in a Gaussian function, in the following form:

7. In the formula x、 , σ Represent the input sample, the mean value and the variance of the sample respectively.

Images