Hinge beam stress-strain rapid calculation method based on global agent model

Abstract

The invention provides a hinge beam stress-strain rapid calculation method based on a global agent model. The method comprises the following steps: firstly, modeling the hinge beam of the cubic press through three-dimensional modeling software, then sampling working condition data by adopting a Latin hypercube sampling method according to the working condition of the hinge beam, importing a finite element to carry out stress-strain static structure simulation analysis, obtaining stress and strain simulation results of three-dimensional coordinate points, and storing the stress and strain simulation results in a database of a global agent model; secondly, on the basis of a random forest algorithm, the stress and strain prediction of the hinge beam is constructed; then, constructing stress and strain proxy models about different working conditions at each three-dimensional coordinate point to form a global proxy model; and finally, stress and strain distribution of the hinge beam in different areas under different working conditions can be directly obtained through a global agent model, and three-dimensional visual display is carried out.

Description

Technical Field

[0001] The invention relates to the technical field of stress and strain analysis, global proxy model and visualization of a six-sided top press, and in particular to a fast calculation method for stress and strain of a hinge beam. Background Art

[0002] In the field of diamond production, the six-sided top press is a key equipment for synthesizing artificial diamonds. During operation, it undergoes a high-temperature and high-pressure synthesis process with multiple stages of pressurization, pressure maintenance, and pressure relief, including ultra-high pressure and high temperature. However, under the action of long-term alternating loads, the lugs and bottom of the hinge beam of the six-sided top press are prone to cracks and fractures, making it difficult to adapt to the requirements of larger diamond presses. In order to extend the life of the equipment and meet its higher-intensity operation requirements, an in-depth analysis of the stress and strain distribution of the hinge beam is required. Traditional multi-condition analysis methods, especially in the finite element analysis (FEA) process, have large computational complexity and are time-consuming, making it difficult to achieve real-time monitoring of the operation process of the six-sided top press and process adjustment requirements.

[0003] The introduction of the proxy model can improve the efficiency of stress and strain analysis of the hinge beam of the six-sided top press, and provide the possibility for real-time monitoring and process adjustment of the operation process of the six-sided top press. By constructing a global proxy model of the stress and strain of the hinge beam of the six-sided top press, the stress and strain distribution of the hinge beam under different working conditions can be quickly predicted at a low computational cost. Compared with traditional methods, the global proxy model improves the calculation efficiency of the stress and strain of the hinge beam, and intuitively displays the stress state of the hinge beam through visualization, thus providing the possibility for real-time monitoring and process adjustment of the operation process of the six-sided top press.

[0004] Digital twin technology provides an innovative solution for the real-time monitoring and fault prediction of the hinge beam of the six-sided top press. By using digital twin technology, sensors are arranged at key positions of the hinge beam to collect stress, temperature, strain and other data in real time. Combined with three-dimensional visualization technology, the real-time monitoring data is integrated with the virtual model of the hinge beam to form a dynamically updated visualization interface. Using the real-time data and historical operating data fed back by the sensors, the digital twin can realize real-time simulation and prediction of the stress and strain state of the hinge beam. However, there is a coupling effect of thermal-mechanical-fluid multi-physical fields during the operation of the six-sided top press, and the simulation of the stress and strain state requires a lot of time, which makes it difficult to realize the real-time simulation of the stress and strain of the hinge beam. The present invention proposes a fast calculation method for the stress and strain of the hinge beam based on the global proxy model, which provides an important idea for the real-time simulation of the stress and strain of the hinge beam.

[0005] The present invention performs feature extraction on the stress and strain data of the hinge beam of the six-sided press, and constructs a global proxy model of stress and strain for different working conditions. It is expected to break through the computing speed bottleneck of the thermal-mechanical-fluid multi-physical field coupling simulation during the operation of the six-sided press, cooperate with the digital twin system to improve the simulation and prediction efficiency of the digital twin, and realize real-time monitoring of the operation process of the six-sided press and process adjustment and optimization.

[0006] The basic idea of ​​the present invention is: first, through modeling and simulation using 3D modeling software, a sample database is constructed by combining Latin hypercube sampling with finite element analysis results under different working conditions. Subsequently, the stress and strain data of each grid node are trained to construct stress and strain proxy models for each node under different working conditions to form a global proxy model. Finally, the stress and strain distribution of different regions of the hinged beam under different working conditions can be directly obtained through the global proxy model for three-dimensional visualization. Summary of the invention

[0007] In view of the shortcomings of the prior art, the present invention provides a method for quickly calculating the stress and strain of a hinge beam based on a global proxy model, comprising the following steps:

[0008] S1: Extract working condition samples;

[0009] S2: using improved Latin hypercube sampling;

[0010] S3: Establish a three-dimensional model of the six-sided top press and perform finite element analysis;

[0011] S4: Establish database;

[0012] S5: Train the agent model using the random forest algorithm;

[0013] S6: Establish a global proxy model;

[0014] S7: Obtain the three-dimensional coordinate data (x, y, z) and the corresponding scalar number (magnitude of stress and strain) from the database to prepare for scalar visualization;

[0015] S8: Combine stress value data with spatial position point coordinates to display stress distribution;

[0016] S9: perform color mapping;

[0017] S10: define color space;

[0018] S11: Perform scalar prediction visualization.

[0019] Compared with the prior art, this patent application has the following beneficial effects:

[0020] This paper proposes a fast calculation method for stress and strain of hinged beams based on a global proxy model. First, the hinged beam of the six-sided press is modeled by three-dimensional modeling software. Then, according to the working condition of the hinged beam, the Latin hypercube sampling method is used to sample the working condition data, and the finite element is imported to perform stress and strain static structural simulation analysis to obtain the stress and strain simulation results of the three-dimensional coordinate points, which are stored in the database of the global proxy model. Secondly, based on the random forest algorithm, the prediction of the stress and strain of the hinged beam is constructed. After that, at each three-dimensional coordinate point, the stress and strain proxy model for different working conditions is constructed to form a global proxy model. Finally, the stress and strain distribution of different regions of the hinged beam under different working conditions can be directly obtained through the global proxy model for three-dimensional visualization.

[0021] 1) Significantly improved computing efficiency

[0022] When calculating the stress and strain of hinged beams, the existing technology often consumes a lot of time and computing resources to process complex working condition data and model analysis. However, the method based on the global proxy model proposed in this patent reasonably collects working condition data through the Latin hypercube sampling method, and combines the random forest algorithm to build a prediction and global proxy model. This method avoids the tedious repetitive calculation process in the traditional method, can quickly obtain the stress and strain results of the hinged beam under different working conditions, significantly shorten the calculation time, and improve the calculation efficiency.

[0023] 2) High accuracy and reliability

[0024] In the calculation process of this patent, the finite element method is first used to perform static structural simulation analysis to obtain accurate stress and strain simulation results of three-dimensional coordinate points, and a global proxy model is constructed with each node coordinate. Each three-dimensional coordinate point under different working conditions has a targeted model construction, making the calculation results of stress and strain more accurate, providing a reliable basis for the design, evaluation and optimization of hinged beams.

[0025] 3) Good 3D visualization effect

[0026] Current technologies may not be intuitive enough when displaying stress and strain distribution, which is not conducive to rapid understanding and analysis by engineers and researchers. The global proxy model of this patent can realize the three-dimensional visualization of stress and strain distribution in different areas of the hinge beam under different working conditions. This visualization method enables staff to intuitively observe the distribution characteristics of stress and strain, and promptly identify potential problem areas, providing strong support for further improving design and optimizing processes. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] Figure 1 This is the overall process of this method;

[0028] Figure 2 It is a three-dimensional structure of hinged beam;

[0029] Figure 3 A pressurized process for producing diamonds;

[0030] Figure 4 Random forest algorithm structure framework;

[0031] Figure 5 Random forest algorithm process;

[0032] Figure 6 Hinge beam visualization predictions. DETAILED DESCRIPTION

[0033] The technical solutions in the embodiments of the present invention will be described clearly and completely below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments.

[0034] A fast calculation method for stress and strain of hinge beam based on global proxy model includes the following steps:

[0035] S1: Extract working condition samples.

[0036] Since the pressure and temperature rise time in the synthesis process of the diamond six-sided top press is relatively long, in order to obtain the corresponding stress and strain, if n pressure points and m temperature points are jointly simulated, there are n×m situations, which will inevitably cause too many simulation times and too long simulation time. Therefore, the Latin hypercube sampling method is adopted to sample at different temperature and pressure stages of pressurization, pressure holding, secondary pressurization and ultra-high pressure of the hinge beam of the six-sided top press, and a limited number of sample points are selected to replace the entire simulation process, so as to achieve sample optimization with a small amount of calculation.

[0037] S2: Using improved Latin hypercube sampling:

[0038] (1) Determine the sampling dimension d and the number of samples N, and perform Latin hypercube sampling.

[0039] A variable is divided into n evenly spaced segments in the interval (a, b), where the i-th segment is (a i ,b i ), then

[0040]

[0041] For convenience, note:

[0042]

[0043] (2) Randomly select the representative number of partitions in the equally probable partitions of each variable;

[0044] Define a number η and 0≤η<1 as the coefficient of randomly selected representative numbers, then in the range a i ≤x i The random number chosen is:

[0045] (b i -a i )η i +a i =(1-η i ) i +η i b i (5)

[0046] The number of random representatives of each interval of multiple variables (C nm ) is calculated as follows:

[0047]

[0048] At this point, the selection of random representative "numbers" for each partition of the multivariate is completed, and the matrix (each column represents the representative "number" of each partition of each parameter) is output.

[0049] (3) Calculate the distance d between all random representative “numbers” ij , and for d ij Sort and get the distance list {d1, d2…d s} and the index list {j1,j2…j s}. Where d i Represents different distance values, j i Represents the distance d i The number of points;

[0050] The basic principle of the maximum and minimum distance criterion is to maximize the minimum distance between the random representative "numbers" so that the distance between sample points in space can be as large as possible. The commonly used formula for calculating the distance between sample points is:

[0051]

[0052] Where m is the dimension of the sample point, Manhattan distance is used when t=1, and Euclidean distance is used when t=2.

[0053] (4) Calculation, P is a positive integer;

[0054] (5) If the Latin hypercube sampling φ p If the value is the smallest, then the sample set is selected as the optimal sampling, otherwise return to (1).

[0055] S3: Build a 3D model of the six-sided press and perform finite element analysis

[0056] ​S3.1: Use 3D modeling software to model the six-sided top press, export it into x_t format, and import it into ANSYS software.

[0057] S3.2: Setting material properties

[0058] Set the hinge beam material to 42CrMo, mass density, elastic modulus, Poisson's ratio, shear modulus, and thermal expansion coefficient.

[0059] S3.3: Create analysis unit types

[0060] Tetrahedral Solid187 elements are used to adapt to the complex geometry of the hinge beam to improve simulation accuracy.

[0061] S3.4: Gridding

[0062] In view of the complex shape of some areas of the hinge beam, the mesh division adopts a combination of overall coarse mesh and local fine mesh. The overall hinge beam adopts a 10 mm mesh; the bottom stress-bearing part is chamfered; the mesh of the piston stroke part is refined to 5 mm; and for the parts that are easy to deform and fracture, such as all ear holes and both sides of the ear, they are also refined to 5 mm mesh.

[0063] S3.5: Setting fixed constraints

[0064] Limit the degrees of freedom at the hinge beam axis hole. Add constraints to the double ear holes: limit its rotation in the y direction and movement in the x direction, and add constraints to the three ear holes to limit its rotation in the x direction and movement in the y direction.

[0065] S3.6: Apply boundary conditions

[0066] Two boundary conditions are set: temperature T and pressure P. n temperature boundary conditions Ti (i = 1...n) and m pressure boundary conditions F are applied respectively. j (j=1……m), where ε ij Corresponding to the stress distribution under the action of temperature Ti and pressure Fj, where σ ij is the strain under the corresponding working condition).

[0067] S3.7: Perform sampling point simulation

[0068] Input temperature and pressure. According to the key points of Latin hypercube sampling, sample points of temperature and pressure are sampled. The working conditions of local key points replace the overall working conditions of the hinge beam of the six-sided top press. First, simulate and analyze the temperature T in Ansys, and then import it into the static structure to continue simulating the pressure P.

[0069] S3.8 Temperature field analysis

[0070] Based on the heat transfer, energy conservation law and temperature field analysis equations, the transient heat transfer control equation of the T (x, y, z, t) structure is established:

[0071]

[0072] Where ρ is the material density (kg / m 3 );C T is the specific heat capacity of the material J / (kg·K); J x , J y , J z are the thermal conductivity coefficients along the X, Y, and Z directions (W / (m·K)); R(x,y,z,t) is the heat source intensity inside the object (W / kg); and T is the temperature (K).

[0073] The temperature load remains unchanged during the steady-state heat transfer process, that is:

[0074]

[0075] The temperature field analysis uses the same interpolation method as the structural analysis to apply the temperature load to the unit node and determine T β (x, y, z) is the temperature field function of the unit:

[0076]

[0077] Combining the unit temperature field function with the steady-state heat transfer boundary conditions and calculating the variational extreme value, the unit heat transfer equation can be obtained as follows (where is the unit heat transfer matrix, is the element node temperature array, is the equivalent temperature load array of the unit node):

[0078]

[0079] S3.9: Perform structural stress analysis

[0080] The distribution of temperature field in the process of thermal-structural coupling analysis affects the size and distribution of thermal stress.Thermal-structural coupling analysis adopts the indirect one-way coupling method, and based on the thermal analysis physical equation, the virtual work principle is used to perform virtual work variation of the physical equation to obtain the finite element analysis equation of structural thermal analysis.

[0081] The physical equation for thermal stress analysis is as follows:

[0082] Due to the existence of the internal temperature difference ΔT(x,y,z) of the structure, the α T ·ΔT(x,y,z) expansion (where α T is the thermal expansion coefficient), and the expansion amount is introduced into the structural physical equation to obtain the physical equation for structural thermal stress analysis:

[0083]

[0084] Thermal stress analysis and traditional elastic problem analysis use the same virtual work principle of δU-δW=0 (which can be expressed as formula (13), and its matrix form is formula (14), Formula (15) is obtained by combining the thermal analysis physical equation of the structure with the thermal analysis virtual work equation of formula (9):

[0085]

[0086]

[0087] Substituting the physical equation (12) into equation (13), we get:

[0088]

[0089] The finite element unit node displacement array is shown in formula (17), and the unit node displacement relationship is used to express the basic mechanical parameters in the unit as shown in formulas (18) and (19):

[0090] q e =[u1v1w1...u n v n w n ] T (17)

[0091] u=Nq e (18)

[0092] ε=Bq e (19)

[0093] σ=Sq e -D·α T ΔT[1 1 1 0 0 0] T (20)

[0094] Where: N is the shape function matrix, B is the geometry matrix, D is the elastic coefficient matrix, and S is the stress matrix. Where (20) is the structural thermal stress finite element analysis formula considering temperature distribution.

[0095] S4: Building a database

[0096] S4.1: Get the sample data, export the sample data in the Finite Element Solution - Equivalent Stress / Equivalent Strain - Export Text File module, and save the file type as .xls

[0097] S4.2 extracts the simulation result information: the number of mesh nodes (ID), the values ​​of the node coordinates X, Y, and Z, and the stress and strain corresponding to the node.

[0098] S4.3: Store the simulated data of the sampling points as a database

[0099] Under different working conditions, the consistency of the grid and grid nodes is maintained, and multiple tables containing simulation data information are merged into one data table. The first column is the node ID, and the second, third, and fourth columns are the three-dimensional coordinates X, Y, and Z values ​​of the grid nodes. When training the proxy model, the proxy model is established with each three-dimensional coordinate value. The subsequent columns are the applied pressure, and the corresponding stress σ and strain ε under different pressures. The processed data is the database for proxy model training.

[0100] S5: Training the surrogate model using the random forest algorithm

[0101] (1) Random forest is composed of multiple decision trees {h(x,θ t ),t=1,2,...,T}, where θ t is a random variable that obeys an independent distribution, x is the independent variable, and T is the number of decision subtrees.

[0102] (2) The classification model prediction results are:

[0103]

[0104] In the above formula, Represents the regression prediction result, h(x,θ t ) represents the relationship between x and θ t Output.

[0105] (3) The prediction results of the regression model are:

[0106]

[0107] In the above formula, Represents the regression prediction result, h(x,θ t ) represents the relationship between x and θ t Output.

[0108] (4) Establishing a CART decision tree

[0109] Metric properties of CART classification tree

[0110] Suppose the data set is D with n different categories C i ,C i,D is C in data set D i The set of class tuples, D| is the number of D tuples, |C i,D |It is C i,D The number of tuples. The CART decision tree uses the Gini index Gini (D) calculation formula as follows:

[0111]

[0112] Among them, pi is the frequency of occurrence of Ci type tuples.

[0113] The Gini index performs a binary division on each attribute. For a discrete attribute A, the binary division of A divides D into D1 and D2. Under the given division conditions, the Gini index of D is

[0114]

[0115] The impurity caused by the binary partition of attribute A is reduced to:

[0116] ΔGini A (D) = Gini(D) - Gini A (D) (25)

[0117] By observing the possible binary divisions of each attribute, we select the subset with the smallest Gini index among all cases and use it as the split subset. A The larger (D) is, the smaller Gini(D) is, and the better the splitting effect on A is.

[0118] (5) Analyze the metric properties of the CART regression tree

[0119] For the data set: D = {(x1,y1),(x2,y2),...,(x n ,y n )}, use feature j and split point s to divide all input regions into two sub-regions.

[0120] R1(j,s)={x|x (j) ≤s} and R2(j,s)={x|x (j) >s} (26)

[0121] The minimum loss function is

[0122]

[0123] Where c1 and c2 represent the average of the output values ​​of regions R1 and R2 respectively. Only by continuously calculating the error of each split point by traversing different values ​​of each feature, selecting the point with the smallest split error to divide the input into two regions, and finally satisfying the termination condition, can the minimum loss function be obtained. The input space is finally divided into M regions {R1, R2, ..., R M}, the resulting regression tree is based on the least squares deviation:

[0124]

[0125] Among them C m For region Rm The average value of .

[0126] (6) Evaluation of model assessment and performance

[0127] Classification effect index

[0128] First consider the confusion matrix of the two-classification problem:

[0129]

[0130] Among them, TP represents the number of positive samples classified correctly; TN represents the number of negative samples classified correctly; FN represents the number of positive samples classified incorrectly; FP represents the number of negative samples classified incorrectly.

[0131] The indicators for measuring the classification effect of random forest are as follows:

[0132] ①Classification Accuracy

[0133]

[0134] Represents the overall classification accuracy of random forest. The higher its value, the better the classification effect.

[0135] ② Geometric mean G-mean

[0136]

[0137] As shown in the formula, the geometric mean G-mean refers to the mean of the classification accuracy of the random forest algorithm for positive and negative classes.

[0138] ③F-value

[0139]

[0140] Precision and recall are the precision and recall respectively. F-value not only takes into account the precision of the random forest algorithm, but also its recall. It is a comprehensive indicator to measure the classification performance of the random forest. The value range of β is (0,1], which can be determined according to the actual situation, and is usually taken as 1. The definitions of precision and recall are shown in the following formulas:

[0141]

[0142] (7) Evaluating model performance: Using R 2 The Coefficient of Determination is used to measure the performance of the model, which is called the coefficient of determination. The calculation formula for evaluating the fit of the model is:

[0143]

[0144] Among them, y i is the true value, is the predicted value, is the mean of the true value. SSR represents the regression sum of squares, SST represents the total deviation sum of squares, and SSE represents the residual sum of squares.

[0145] R 2 , the coefficient of determination, the value is between 0 and 1, the closer to 1, the stronger the explanatory power of the model, 1 means perfect fit, 0 means the model's prediction has nothing to do with the actual value, if R 2 =1, it means that the model fully explains the variation in the data. The trained model has been built based on the random forest regression algorithm, and both the model parameters and the tree structure are serialized and saved, so that the model can be reloaded at a later point in time without retraining.

[0146] S6: Building a global proxy model

[0147] The trained random forest network model is saved locally as the proxy model of the hinge beam. The number of grid nodes in the database is i=1.2.3...n. A proxy model is established for each node, and the number of proxy models is always equal to the number of nodes.

[0148] (1) Define grid nodes

[0149] Assume that the hinge beam is discretized into a grid in space, containing n nodes, and each node can be represented by N i , where i∈{1,2,3···,n}. Therefore, the set of grid nodes can be written as:

[0150] N∈{N1,N2,N3···,N n}(39)

[0151] Each N i Nodes represent points in a hinge beam at specific spatial locations.

[0152] (2) Define the proxy model

[0153] For each grid node N i , we need to build a corresponding proxy model M i The proxy model M i The core task is to accurately represent or predict the node N i Some characteristics or behaviors of, such as predicting stress, strain. Usually, the surrogate model can be regarded as some kind of function or mapping:

[0154] Mi=f i (N i),i=1,2,3,...n (40)

[0155] Here, f i is a node for N i The proxy model construction function.

[0156] (3) Dividing feature vectors

[0157] Assume that each node N i Corresponding to a certain eigenvector X i , which contains the physical or computational properties of the node. It can be expressed as:

[0158] Ni=X i =[x i1 ,x i2 ,x i3 ,...,x im ] (41)

[0159] Where m represents the number of features of the node, x ij is node N i The jth feature of .

[0160] (4) Building a global proxy model

[0161] Each node N i The proxy model M i It can be based on the feature vector x of the node i Generated by the above random forest machine learning method. The training and generation of the model can be written as:

[0162] M i =f(x i ,y i ) (42)

[0163] Among them, y i are the target variables, stress and strain, for that node, and f is the algorithm used to generate the model.

[0164] (5) Calculate the total number of proxy models

[0165] Assume that we construct a proxy model for each node independently, so there will be N proxy models in total. This relationship can be written as:

[0166] M={M1,M2,...,M n} (43)

[0167] Among them, the total number of proxy models |M|=N, that is, the total number of proxy models is always equal to the number of mesh nodes, ensuring the accurate analysis of the entire structure of the hinged beam.

[0168] (6) Setting the objective function

[0169] Suppose we want to predict or approximate a target function g through a proxy model, which depends on the node feature x i Then each proxy model M i The task is to approximate g(x i ):

[0170] Mi(x i )≈g(x i )(44)

[0171] (7) Model evaluation

[0172] The error of the surrogate model Mi can usually be measured by some loss function L. The mean square error (MSE) can be used to evaluate the performance of the model:

[0173]

[0174] Here, p is the number of sample points. This evaluation mechanism ensures the reliability and practicality of the model.

[0175] S7: Obtain the three-dimensional coordinate data (x, y, z) and the corresponding scalar number (magnitude of stress and strain) from the database to prepare for scalar visualization.

[0176]

[0177] in:

[0178] N is the number of hinge beam nodes, each row (x i ,y i ,z i ) represents the coordinates of a point in three-dimensional space.

[0179] The matrix P can be represented as a set of vectors:

[0180] P={p1,p2,···,p N} (47)

[0181] Each point can be expressed as:

[0182]

[0183] The output is a vector of N stress and strain values:

[0184]

[0185] S8: Combining stress value data with spatial position point coordinates to display stress distribution

[0186] Color mapping usually requires combining stress values ​​with point coordinates to accurately reflect stress distribution:

[0187] VisualizationData = {(p i ,σ i ,ε i )|i=1,2,…N} (50)

[0188] S9: Perform color mapping

[0189] Define a color mapping function to map data values ​​linearly or nonlinearly to the color space. i ,y i ,z i ), its stress value will be mapped to a certain color C(σ, ε), and this color gradient is applied to the data. The higher the data value, the closer the color is to the end of the gradient.

[0190] Assume that the stress value range is [σ min ,σ max ], the strain value range is [ε min ,ε max ], mapping each stress value σ, strain value ε, to a color value C in the color space i The formula is:

[0191]

[0192] Among them, C max and C min are the starting and ending colors of the color map, σ max ,ε max is the grid point p i The maximum value of a scalar, σ min ,ε min is the grid point p i The minimum value of a scalar.

[0193] S10: Defining the color space

[0194] A color is represented in RGB space as:

[0195] C RGB =(R,G,B) (52)

[0196] Where R, G, and B represent the red, green, and blue components, respectively, usually in the range [0,255]

[0197] The three-dimensional cube representation of the RGB color space is as follows:

[0198] 0≤R,G,B≤225 (53)

[0199] In three-dimensional space, the R, G, and B values ​​correspond to the red, green, and blue axes respectively. Any color can be considered as a linear combination of the three colors:

[0200] C RGB =R·(1,0,0)+G·(0,1,0)+B·(0,0,1) (54)

[0201] In data visualization, the scalar data S is mapped to the RGB space. Linear interpolation can be used to map stress values ​​to RGB colors:

[0202]

[0203] Among them, (R min ,G min ,B min ) and (R max ,G max ,B max ) are the color endpoints of the mapping range.

[0204] S11: Visualizing Scalar Predictions

[0205] Change the boundary conditions to predict stress and strain, temperature Ti and pressure Fi. The boundary conditions represent the magnitude of the force and temperature applied to the grid.

[0206] The process of model prediction can be expressed as:

[0207] σ nm ,ε nm =f(x i ,y i ,z i ,F m ,T n ) (58)

[0208] where F j , T i are the two boundary conditions pressure and temperature, σ nm is the stress value, ε nm is the strain value, f is the random forest model function, which is based on the coordinates and (F m ,T n ) for calculation.

Claims

1. A fast calculation method for stress and strain of hinge beam based on global proxy model, characterized in that: The steps include: S1: Extract working condition samples; S2: using improved Latin hypercube sampling; S3: Establish a three-dimensional model of the six-sided top press and perform finite element analysis; S4: Establish database; S5: Train the agent model using the random forest algorithm; S6: Establish a global proxy model; S7: Obtain three-dimensional coordinate data (x, y, z) and corresponding scalar numbers from the database to prepare for scalar visualization; S8: Combine stress value data with spatial position point coordinates to display stress distribution; S9: perform color mapping; S10: define color space; S11: Perform scalar prediction visualization.

Images