Intelligent optimization method for integrated casting forming defect parameters of large complex component

Abstract

The invention discloses an intelligent optimization method for integrated casting forming defect parameters of a large complex component, and belongs to the technical field of casting machining. The method comprises the following steps: simulating a die-casting process through finite element software to obtain prediction data, carrying out nondestructive detection on a physical integrated casting to obtain detection data, and fusing the prediction data and the detection data to divide into a training set and a test set; the method comprises the following steps: establishing a relation model between forming defects and a forming process, obtaining a process parameter interval covered by the relation model, normalizing process parameters in the process parameter interval, importing the normalized process parameters into the model, predicting the internal defect probability of the integrated casting, and optimizing the forming process parameters by using a particle swarm optimization model. Through precise prediction of internal defects, precise adjustment of technological parameters and automatic control, the quality and efficiency of integrated casting forming can be improved, resource waste can be greatly reduced, the cost is reduced, and the flexibility and the intelligent level of the production process are enhanced.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of casting processing, and particularly relates to a large complex component integrated casting forming defect parameter intelligent optimization method. BACKGROUND

[0002] The vigorous development of the aerospace, national defense and military industry and new energy automobile industry has put forward very high requirements on the performance and quality of super-large integrated castings. With the increase of the size and the improvement of the complexity of castings, integrated casting is facing many technical challenges, one of the most prominent problems being the casting defect problem.

[0003] Casting products usually have defects such as shrinkage, shrinkage holes, deformation and the like. Especially in large complex integrated castings, these defects not only affect the mechanical properties of the castings, but also cause poor forming quality of the castings, reduce the yield and increase the production cost. At present, the quality improvement and defect control of integrated castings are generally controlled by optimizing process parameters by artificial experience. However, due to the complexity of the structure of super-large integrated castings, the single manufacturing cost is high, and the relationship between the forming process parameters is difficult to control by traditional experience, which makes the process optimization of super-large integrated castings very difficult, resulting in frequent occurrence of various defects. Related researchers have developed professional integrated casting finite element simulation software, such as ProCAST, Flow3d, magma, etc., which greatly reduces the product development cost. However, the grid number of large integrated components is generally in the order of hundreds of millions, and the traditional simulation process is still relatively time-consuming for defect prediction, and the calculation cost is still high. Moreover, it is difficult to comprehensively evaluate the overall defect position of integrated casting by using a single numerical simulation software. Therefore, it is urgent to develop a fast and low-cost casting defect prediction and control method in the field of integrated casting. SUMMARY

[0004] The application aims to provide a large complex component integrated casting forming defect parameter intelligent optimization method, which solves the above problems existing in the prior art.

[0005] Technical scheme: A large complex component integrated casting forming defect parameter intelligent optimization method comprises the following steps:

[0006] S1, by constructing a large integrated structure three-dimensional model, at least three kinds of finite element software such as ProCAST, Flow3d and magma are used to simulate the casting process, the position, type and range of integrated casting forming defects are comprehensively predicted to obtain at least three kinds of prediction data, at the same time, the initial verification test of integrated casting process is carried out, the actual position, type and range of integrated casting forming defects are obtained through CT detection of nondestructive testing of physical integrated casting, the prediction data and the test data are fused to obtain a data set, and the data set is divided into a training set and a test set;

[0007] S2, a design matrix is preset, the training set is imported into the design matrix, a relationship model between forming defects and forming process is built, and the test set is used to evaluate the accuracy of the relationship model, and the process parameter interval covered by the relationship model is obtained;

[0008] S3, the process parameters in the process parameter interval are normalized to obtain normalized process parameters, the normalized process parameters are imported into the prediction model, the internal defect probability of the integrated casting is predicted by using the prediction model, the model most consistent with the actual detection defect is taken as the benchmark, the forming defect is taken as the target, and the forming process parameters are optimized, a function based on the internal defect probability of the integrated casting is predefined, and then the corresponding defect probability of the process parameter combination is calculated, the defect probability parameter combination corresponding to the normalized process parameters is obtained, and the particle swarm iteration is used to make the defect probability parameter combination gradually close to the low defect probability process parameter combination, the normalized process parameters are restored to actual physical quantities through inverse normalization, and the optimized process parameter combination is obtained.

[0009] Preferably, the prediction model comprises at least one of SVM vector machine, SVR support vector regression, RF random forest, LIN linear regression and XGBoost extreme gradient boosting.

[0010] Preferably, in the process of nondestructive testing of physical integrated casting by CT detection, the feedback test data of the sensor is continuously received, the physical defect of the physical integrated casting is compared with the model defect of the integrated casting predicted by the model, when there is a difference between the physical defect of the physical integrated casting and the integrated casting predicted defect, the parameters of the prediction model or the particle swarm optimization model are automatically adjusted and called to adapt, and the defect probability parameter combination corresponding to the normalized process parameters is optimized.

[0011] Preferably, the prediction model training process is as follows:

[0012] The training set is imported into each prediction model, the data in the training set is used to train the prediction model, a defect prediction model for predicting defect occurrence probability is obtained, the probabilities and degrees of each defect type under different process parameters are obtained through the defect prediction model, a hyperplane is determined through the defect prediction model, the linear data of the process parameters in the training set are classified according to whether there is a defect through the hyperplane, a defect sample set and a non-defect sample set are obtained, for the nonlinear data of the process parameters in the training set, the nonlinear data is mapped to a high-dimensional space through a kernel function, and linear separability is completed.

[0013] Preferably, the classification process of linear data is as follows:

[0014] First, the hyperplane is calculated through formula (1), and the formula is as follows:

[0015] w T x+b=0 (1);

[0016] wherein w represents the normal vector of the hyperplane, x represents the forming process parameter, b represents the bias term, and T represents the ratio of the regression coefficient to the standard error; after the calculation of the hyperplane is completed, the prediction model determines the hyperplane by maximizing the interval, and after the constraint condition is given, the linear data of the process parameters in the training set is classified into defect process parameters and non-defect process parameters according to whether there is a defect, wherein the interval is located as:

[0017]

[0018] Constraint condition:

[0019] The optimization problem is finally converted into:

[0020]

[0021] Preferably, the classification process of nonlinear data is as follows:

[0022] The SVM regression model maps the nonlinear data to a high-dimensional space through a kernel function k(x i ,x j ), and processes the nonlinear data into linearly separable data, and the calculation formula is as follows:

[0023] k(x i ,x j )=Φ(x i ) T Φ(x j ) (5);

[0024] In the formula, k(x i ,x j ) represents a kernel function, x i represents i data points in an input space, and xj representing j data points in the input space;

[0025] After the nonlinear data is converted into linearly separable data, the classification decision function of the prediction model is used to classify the linearly separable data, so that the nonlinear data can be classified into defective process parameters and non-defective process parameters according to whether there is a defect.

[0026] Preferably, at least the finite element software simulation casting process of ProCAST, Flow3d and magma is imported, and the process of predicting the position and range of the forming defect is as follows:

[0027] A three-dimensional model of the integrated casting is established, which is at least imported into ProCAST, Flow3d, magma finite element software and finite difference simulation software, the casting forming simulation under the corresponding process parameters is completed, and the shrinkage, porosity, gas entrapment or forming defect of the casting under the corresponding forming process parameters is quantitatively predicted by extracting the temperature field, solidification field, shrinkage and porosity data under different simulation process parameters.

[0028] Preferably, the training set and the test set are normalized by using normalization, and the process is as follows:

[0029]

[0030] In the formula: x normalized representing the data subjected to normalization processing, x represents the variable in the data set, x max represents the maximum value of the variable x in the data set, x min represents the minimum value of the variable x in the data set.

[0031] Preferably, the process of optimizing the forming process parameters by the particle swarm optimization model is as follows:

[0032] Randomly generate N groups of process parameter combinations in the training set, and each group of process parameter combinations represents a group of particles i. The iteration update speed and position are calculated by substituting the particle i into the particle swarm optimization formula, wherein the position and speed of the particle i are respectively shown in formula (8) and formula (9):

[0033] x i =[x i1 ,x i2 ,···,x iD ] (8);

[0034] v i =[v i1 ,v i2 ,···,v iD ] (9);

[0035] In the formula: x ia position representing a combination of process parameters, v i representing a velocity corresponding to the direction and magnitude of movement of the control particle in the next step;

[0036] The low defect probability is calculated by a target function, and the calculation formula of the target function is as follows:

[0037] f(x i )=defect_probability(x i ) (10)。

[0038] Preferably, after the calculation of the defect probability is completed, the position of each group of particles is updated according to the position of each group of particles in the global and the position of each group of particles in the global The velocity and position of the particle group are updated, and the particle swarm optimization update velocity formula is as follows:

[0039]

[0040] wherein: v i (t) is the velocity of particle i at time step t, x i (t) is the position of particle i at time step t, w is the inertia weight, c1 and c2 are learning factors, and r1 and r2 are random numbers in the range [0, 1];

[0041] The particle swarm optimization update position formula is as follows:

[0042] x i (t+1)=x i (t)+v i (t+1) (12)

[0043] wherein: v i (t+1) represents the particle swarm optimization update velocity.

[0044] Beneficial effects: The present application relates to a large complex component integrated casting forming defect parameter intelligent optimization method, (1), a plurality of prediction models of SVM vector machine, SVR support vector regression, RF random forest, LIN linear regression and XGBoost extreme gradient boosting are combined with particle swarm optimization model through a plurality of numerical simulation methods to optimize the integrated casting forming process parameters and predict internal defects, and the process parameters in the casting defect optimization forming process are predicted in multiple ways. Compared with the ordinary single casting simulation prediction method, the precision is improved from 80% to 98%; compared with the defect rate of the single simulation process optimization algorithm before production, the defect rate is reduced by 60%-80%, which solves the problems of low defect prediction accuracy, high prediction difficulty, long prediction cycle and high scrap rate caused by single simulation method.

[0045] (2) The results of various process simulations such as finite element and finite difference are combined with the defect detection results of actual experiments using different casting temperatures, pouring speeds and process parameters, and used as input for deep learning simulation of defect prediction to perform defect prediction; not only is feature extraction of three-dimensional simulation structure introduced, but also feature values ​​of various simulation structures are integrated with process parameters, which greatly improves the generalization performance of defect prediction of castings.

[0046] (3) By using three finite element software, ProCAST, Flow3d and Magma, to simulate the casting process, the location, type and range of defects in the integrated casting are predicted to obtain three prediction data. At the same time, an initial verification test of the integrated casting process is carried out. The three simulation prediction data are integrated with the test detection data. The process parameters within the process parameter range are normalized. The probability of internal defects in the integrated casting is predicted simultaneously using five prediction models, so that the defect prediction is more in line with the engineering application standards. Attached Figure Description

[0047] Figure 1 This is a system block diagram of the present invention;

[0048] Figure 2 This is a three-dimensional model of the integrated front engine bay of the present invention;

[0049] Figure 3 This is the test matrix table for the present invention;

[0050] Figure 4 This is the model accuracy evaluation table for the present invention;

[0051] Figure 5 This is a comparison between the prediction results and test results of the particle swarm optimization model of the present invention;

[0052] Figure 6 This is a schematic diagram of the Procast defect prediction process of the present invention;

[0053] Figure 7 This is a schematic diagram of the FLow3d defect prediction process of the present invention;

[0054] Figure 8 This is a data reference table for the present invention. Detailed Implementation

[0055] like Figures 1 to 8 As shown, the present invention provides a technical solution: an intelligent optimization method for defect parameters in the integrated casting forming of large and complex components, comprising the following steps:

[0056] Step one, by constructing a large integrated structure three-dimensional model, at least three kinds of finite element software such as ProCAST, Flow3d, magma are used to simulate the casting process, the position, type and range of integrated casting forming defects are comprehensively predicted to obtain at least three kinds of prediction data, at the same time, the initial verification test of integrated casting process is carried out, the actual position, type and range of integrated casting forming defects are obtained by nondestructive testing of the physical integrated casting through CT detection, the prediction data and the test data are fused to obtain a data set, and the data set is divided into training set and test set;

[0057] Step two, a design matrix is preset, the training set is imported into the design matrix, a relationship model between forming defects and forming process is built, the accuracy of the relationship model is evaluated by using the test set, the process parameter interval covered by the relationship model is obtained, the process parameters in the process parameter interval are normalized to obtain normalized process parameters, wherein the process parameters at least include mold temperature, casting pressure and pouring speed, wherein the training set and the test set are normalized by using normalization, and the processing process is as follows:

[0058]

[0059] In the formula: x normalized represents the data subjected to normalization processing, x represents the variable in the data set, x max represents the maximum value of the variable x in the data set, x minThe minimum value of a variable x in the data set is represented, the process parameters in the process parameter interval are normalized to obtain normalized process parameters, and the normalized process parameters are imported into a prediction model, wherein the prediction model at least includes one of an SVM vector machine, an SVR support vector regression, an RF random forest, a LIN linear regression, and an XGBoost extreme gradient boosting, the prediction model is used to simultaneously predict the probability of internal defects of an integrated casting, a model most consistent with actual detection defects is taken as a benchmark, a forming defect is taken as a target, a forming process parameter is optimized, a function based on the predicted probability of internal defects of the integrated casting is predefined, and then the corresponding defect probability under the process parameter combination is calculated to obtain a defect probability parameter combination corresponding to the normalized process parameter, and the particle swarm iteration is used to gradually approach the low defect probability process parameter combination. The normalized process parameter is restored to an actual physical quantity through inverse normalization to obtain an optimized process parameter combination. In the process of non-destructive testing of the physical integrated casting through CT detection, the feedback test data of the sensor is continuously received, the physical defects of the integrated casting are compared with the model predicted defects of the integrated casting, and when there is a difference between the physical defects of the integrated casting and the predicted defects of the integrated casting, the parameters of the adaptive prediction model or the particle swarm optimization model are automatically adjusted and called to optimize the defect probability parameter combination corresponding to the normalized process parameter, and the particle swarm iteration is used to gradually approach the low defect probability process parameter combination. The normalized process parameter is restored to an actual physical quantity through inverse normalization to obtain an optimized process parameter combination. In the embodiment, the adaptive optimization of the model is realized, that is, if the mold temperature in the process parameter predicted by the SVM vector machine model is too high to cause defects, the temperature of the molten metal is reduced by adjusting the parameters of the SVM vector machine model or the particle swarm optimization model. If the result of the SVM vector machine model prediction shows that the fast pouring speed may cause porosity or uneven cooling, the pouring speed is appropriately slowed down by adjusting the parameters of the SVM vector machine model or the particle swarm optimization model. If the result of the SVM vector machine model prediction shows that low pressure may cause incomplete filling of the mold, the pressure is increased to ensure that the mold is filled with the mold by adjusting the parameters of the SVM vector machine model or the particle swarm optimization model.

[0060] In further embodiments, the prediction model training process is as follows:

[0061] The training set is imported into each prediction model, the data in the training set is used to train the prediction model, the prediction model for predicting the probability of defect occurrence is obtained, the probability and degree of each defect type under different process parameters are obtained through the prediction model, the hyperplane is determined through the prediction model, the linear data of the process parameters in the training set is classified according to whether there is a defect through the hyperplane, the defect sample set and the non-defect sample set are obtained, and the non-linear data of the process parameters in the training set is classified as follows:

[0062] First, the hyperplane is calculated by formula (1), whose formula is as follows:

[0063] w T x+b=0 (1);

[0064] Wherein, w represents the normal vector of the hyperplane, x represents the forming process parameter, b represents the bias term, T represents the ratio of regression coefficient and standard error; after the calculation of the hyperplane is completed, the SVM vector machine model determines the hyperplane by maximizing the interval, and after the given constraint condition, the linear data of the process parameters in the training set is divided into the defective process parameters and the non-defective process parameters according to whether there is a defect, wherein the interval is positioned as:

[0065]

[0066] The optimization problem is finally converted into:

[0067] The SVM regression model maps the nonlinear data to the high-dimensional space through the kernel function k(x i ,x j ), and completes the linear separability, and the classification process for the nonlinear data is as follows after the classification is completed through the prediction model:

[0068] The SVM regression model maps the nonlinear data to the high-dimensional space through the kernel function k(x i ,x j ), and processes the nonlinear data into linear separable data, and the calculation formula is as follows:

[0069] k(x i ,x j )=Φ(x i ) T Φ(x j ) (5);

[0070] Wherein, k(x i ,x j ) represents the kernel function, x i represents the i data points in the input space, and x j represents the j data points in the input space.

[0071] After the nonlinear data is converted into linear separable data, the linear separable data is classified by using the classification decision function of the SVM, and the nonlinear data is divided into the defective process parameters and the non-defective process parameters according to whether there is a defect.

[0072] In further embodiments, at least the ProCAST, Flow3d, magma finite element software is used to simulate the casting process, and the process of predicting the forming defect position and range is as follows:

[0073] A three-dimensional model of the integrated casting is established and imported into at least ProCAST, Flow3d, Magma finite element software, and finite difference simulation software to complete the casting forming simulation under the corresponding process parameters. By extracting data such as temperature field, solidification field, shrinkage cavity, and porosity under different simulated process parameters, the shrinkage cavity, air entrapment, or forming defects of the casting under the corresponding forming process parameters are predicted and quantified. The process of simulating the die casting process using ProCAST, Flow3d, and Magma finite element software to predict the location and range of forming defects is as follows:

[0074] A three-dimensional model of the integrated casting is established and imported into various finite element and finite difference simulation software to complete the casting forming simulation under the corresponding process parameters. By extracting data such as temperature field, solidification field, shrinkage cavity, and porosity under different simulated process parameters, the shrinkage cavity, air entrapment, or deformation forming defects of the casting under the corresponding forming process parameters are predicted and quantified. Forming tests are conducted on the three-dimensional model of the integrated casting to obtain the predicted data of the three-dimensional morphology and internal defects of the integrated casting.

[0075] In a further embodiment, the particle swarm optimization model optimizes the molding process parameters as follows:

[0076] N sets of process parameter combinations are randomly generated from the training set. Each set of process parameter combinations represents a set of particles i. The iterative update velocity and position are calculated by substituting particles i into the particle swarm optimization formula. The position and velocity of particles i are shown in formulas (8) and (9), respectively.

[0077] x i =[x i1 ,x i2 ,···,x iD (8);

[0078] v i =[v i1 ,v i2 ,···,v iD (9);

[0079] In the formula: x i The position representing a combination of process parameters, v i This indicates the velocity corresponding to the direction and magnitude of the particle's movement in the next step;

[0080] The probability of low defects is then calculated using an objective function, the formula for which is as follows:

[0081] f(x i =defect_probability(xi ) (10), after the calculation of the probability of defects is completed, the position of each group of particles is determined according to the position of each group of particles with the position in the global The velocity and position of the particle swarm are updated, and the particle swarm optimization updates the velocity formula as follows:

[0082]

[0083] In the formula: v i (t) is the velocity of particle i at time step t, x i (t) is the position of particle i at time step t, w is the inertia weight, c1 and c2 are learning factors, and r1 and r2 are random numbers in the range [0, 1];

[0084] The particle swarm optimization updates the position formula as follows:

[0085] x i (t+1) = x i (t) + v i (t+1) (12)

[0086] In the formula: v i (t+1) represents the particle swarm optimization update velocity.

[0087] Through the above technical scheme, the process of the integrated die casting forming of the new energy automobile front bin is as follows:

[0088] Taking the integrated die casting automobile front bin as an example, the forming material of the integrated front bin is mainly Al9Si0.5Mn, and the forming material herein can be changed arbitrarily according to actual use requirements, and the finite element model of the integrated front bin is as shown in Figure 3 The integrated front bin is imported into ProCAST, Flow3d and magma software, and the filling speed, vacuum degree, casting pressure and other forming process parameters are taken as variables to carry out filling and solidification simulation under different forming process parameters, and the integrated front bin casting defect distribution results under different forming process parameters are obtained and the quantitative data are exported, that is, the prediction data, wherein the casting defect results can be any one or more of the defect types such as shrinkage, oxide slag distribution and gas entrapment, in addition, the defect data of the integrated front bin casting can also be obtained by CT detecting the integrated front bin casting physical object, that is, the detection data, the prediction data and the detection data are fused to obtain a data set, and the data set is divided into a training set and a test set.

[0089] In the data set, 75% of the data is used as the training set, and 25% of the data is used as the test set. Taking the shrinkage and porosity defect as an example, the quantitative data can be directly exported by the software. Based on this, the training set and test set are improved, a complete test matrix is constructed, and the process parameter interval is obtained. The specific results are shown in Figure 3 The process parameter interval is normalized using normalization to obtain normalized process parameters. The particle swarm optimization model is trained using the normalized process parameters to optimize the forming process parameters. The accuracy of the particle swarm optimization model is verified using the optimized forming process parameters, i.e., the fitting degree (R 2 ) and the root mean square error (RMSE) of the test sample set are calculated. Figure 3 The test set prediction value of the PSO model is given, and the results are shown in Figure 4 .

[0090] Based on this, the optimal process parameters for the Al9Si0.5Mn integrated front magazine die casting forming are: pouring temperature 670.8℃, mold temperature 162.0℃, vacuum degree 24.4mbar, and fast punch speed 3.2m / s. Under this process, the shrinkage and porosity of the integrated front magazine casting is reduced to 2.01%, which has a better optimization effect.

[0091] The SVM vector machine model is trained using the optimized forming process parameters, the probability and degree of each defect type under different process parameters are obtained, and the optimized forming process parameters are used for the die casting forming of the integrated front magazine. The defect type and occurrence probability of the casting under the current process parameters are predicted by inputting the SVM model, i.e., the forming data input into the SVM vector machine model can be obtained by online monitoring and real-time feedback to the SVM vector machine model. Different defect optimization objectives are set, and based on the output of the SVM vector machine model, if the prediction result shows that the defect probability is high, the adjustment signal is fed back to the control system. According to the defect type and occurrence probability of the SVM vector machine model, the process parameters are adjusted in real time by feeding back to the control system. The adjustment parameters can include mold temperature, casting pressure, pouring speed, etc. The adjustment method depends on the feedback result and the preset rules to ensure that the optimization objective defect is minimized.

[0092] Example 1: Real-time monitoring result: input parameter [690, 180, 500, 5], output defect probability: 0.51, adjust vacuum degree to 150, and reduce fast punch speed to 4;

[0093] Example 2: Real-time monitoring result: input parameter [730, 200, 150, 4], output defect probability: 0.19, process parameter does not need to be adjusted, and production continues.

[0094] The preferred embodiments of the present application are described in detail above, but the present application is not limited to the specific details of the above-described embodiments, and various equivalent transformations of the technical solutions of the present application can be made within the technical concept of the present application, and these equivalent transformations all belong to the protection scope of the present application.

Claims

1. A method for intelligent optimization of defect parameters in integrated casting forming of large complex components, characterized in that, Comprise the following steps: S1, by constructing a large integrated structure three-dimensional model, at least using ProCAST, Flow3d, magma three kinds of finite element software simulation casting process, comprehensive prediction of integrated casting forming defect position, type and scope of at least three kinds of prediction data, at the same time, the initial verification test of integrated casting process, through CT detection of physical integrated casting nondestructive testing, get the actual position, type and range of integrated casting forming defect test data, the prediction data and test data are fused to obtain data set, and the data set is divided into training set and test set; S2, preset design matrix, import training set into design matrix, build the relationship model between forming defect and forming process, and evaluate the accuracy of the relationship model by using the test set, and obtain the process parameter interval covered by the relationship model; S3, the process parameters in the process parameter interval are normalized to obtain the normalized process parameters, the normalized process parameters are imported into the prediction model, the internal defect probability of integrated casting is predicted by using the prediction model, the model most consistent with the actual detection defect is taken as the benchmark, the forming defect is taken as the target, and the forming process parameters are optimized, the function based on the internal defect probability of the predicted integrated casting is predefined, and then the corresponding defect probability of the process parameter combination is calculated, the defect probability parameter combination corresponding to the normalized process parameter is obtained, and the particle swarm iteration is used to make the defect probability parameter combination gradually close to the low defect probability process parameter combination, and the normalized process parameter is restored to the actual physical quantity through inverse normalization, to obtain the optimized process parameter combination.

2. The method according to claim 1, wherein The prediction model comprises at least one of SVM vector machine, SVR support vector regression, RF random forest, LIN linear regression and XGBoost extreme gradient boosting.

3. The method according to claim 1, wherein, In the nondestructive testing of physical integrated casting by CT detection, the feedback test data of the sensor is continuously received, the physical defect of the physical integrated casting is compared with the model defect of the integrated casting model predicted by the model, and when there is a difference between the physical defect of the physical integrated casting and the integrated casting predicted defect, the parameters of the prediction model or the particle swarm optimization model are automatically adjusted and called to optimize the defect probability parameter combination corresponding to the normalized process parameter.

4. The method according to claim 3, wherein, Wherein, The prediction model training process is as follows: The training set is imported into each prediction model, the data in the training set is used to train the prediction model, the defect prediction model of the defect prediction probability is obtained, the probability and degree of each defect type under different process parameters are obtained through the defect prediction model, the hyperplane is determined through the defect prediction model, the linear data of the process parameters in the training set is classified according to whether there is defect through the hyperplane, the defect sample set and the non-defect sample set are obtained, for the nonlinear data of the process parameters in the training set, the nonlinear data is mapped to high-dimensional space through kernel function, and linear separability is completed.

5. The method according to claim 4, wherein The classification process of linear data is as follows: Firstly, the hyperplane is calculated by formula (1), and its formula is as follows: w T x + b = 0 (1); Wherein, w represents the normal vector of the hyperplane, x represents the forming process parameters, b represents the bias term, T represents the ratio of regression coefficient and standard error; after the calculation of the hyperplane is completed, the prediction model determines the hyperplane by maximizing the interval, and after the constraint condition is given, the linear data of the process parameters in the training set is divided into defective process parameters and non-defective process parameters according to whether there is a defect, wherein the interval is located as: The optimization problem is finally converted, which converts the maximization problem into a minimization problem, that is, the conversion formula is as follows:

6. The method according to claim 4, wherein, The classification process of nonlinear data is as follows: The prediction model maps the nonlinear data to a high-dimensional space through a kernel function k(x i ,x j ), processes the nonlinear data into linearly separable data, and the calculation formula is as follows: k(x i ,x j ) = Φ(x i ) T Φ(x j ) (5) where k(x i ,x j ) denotes a kernel function, x i denotes i-th data point in the input space, and x j denotes j-th data point in the input space. After the nonlinear data is converted into linearly separable data, the classification decision function of the prediction model is used to classify the linearly separable data, that is, the nonlinear data is divided into defective process parameters and non-defective process parameters according to whether there is a defect.

7. The method according to claim 1, wherein At least the finite element software ProCAST, Flow3d and magma are used to simulate the casting process to predict the position and range of forming defects, and the process is as follows: A three-dimensional model of an integrated casting is established, which is imported into at least ProCAST, Flow3d, magma finite element software and finite difference simulation software, the casting forming simulation under the corresponding process parameters is completed, the shrinkage and porosity, gas entrapment or forming defects of the casting under the corresponding forming process parameters are quantified by extracting the temperature field, solidification field, shrinkage and porosity data under different simulation process parameters.

8. The method according to claim 3, wherein, The training set and the test set are normalized by using normalization, and the process is as follows: where: x normalized denotes the data after normalization, x denotes a variable in the data set, x max denotes the maximum value of the variable x in the data set, x min denotes the minimum value of the variable x in the data set.

9. The method according to claim 1, wherein, The process of the particle swarm optimization model for optimizing the forming process parameters is as follows: Randomly generate N groups of process parameter combinations in the training set, and each group of process parameter combinations represents a group of particles i, and the iteration update speed and position are calculated by substituting the particles i into the particle swarm optimization formula, wherein the position and speed of the particles i are represented by formula (8) and formula (9) respectively: x i = [x i1 , x i2 , ···, x iD ] (8); v i = [v i1 , v i2 , ···, v iD ] (9); wherein: x i a position representing a combination of process parameters, v i denotes a velocity corresponding to the direction and amplitude of the movement of the control particle in the next step; Then the low defect probability is calculated by the objective function, and the calculation formula of the objective function is as follows: f(x i ) = defect_probability(x i ) (10).

10. The method according to claim 9, wherein After the calculation of the probability of defects is completed, the position of each group of particles is determined and the position in the global The velocity and position of the particle swarm are updated, wherein the particle swarm optimization updates the velocity formula as follows: where: v i (t) is the velocity of particle i at time step t, x i (t) is the position of particle i at time step t, w is the inertia weight, c1 and c2 are learning factors, and r1 and r2 are random numbers in the range [0, 1]. The particle swarm optimization update position formula is as follows: x i (t+1) = x i (t) + v i (t+1) (12) where: v i (t+1) represents the particle swarm optimization update speed.

Images