Multi-objective optimization method for processing quality prediction

Abstract

The invention discloses a multi-objective optimization method for processing quality prediction. The method comprises the following steps of: establishing a high-dimensional curved surface agent model of a mapping relationship between process parameters and product quality, and performing multi-time stratified sampling and fitting on a high-dimensional curved surface through stage processing and interval merging to realize refinement of a part where an optimal parameter of a high-dimensional model is located; and according to the model, calculating a comprehensive entropy weight of a product quality target parameter, and adopting an NSGA3 algorithm to realize multi-target optimization of a process parameter. According to the invention, through layered sampling of the high-dimensional curved surface, the precision of solving the optimal process parameters by the proxy model is improved; and the comprehensive entropy weight of the product quality target parameter is used as an optimization basis, so that the reliability and efficiency of the NSGA3 algorithm are improved. The method can solve the problems that the technological parameter coupling relation is complex and difficult to accurately and efficiently solve, and is beneficial to improving the processing quality of products.

Description

Technical Field

[0001] The invention relates to a multi-objective optimization method, to the technical field of industrial automation, and in particular to a multi-objective optimization method for products for processing quality prediction. Background Art

[0002] The world is focusing on precision scenes in industrial production processes, and the production processes in these scenes often involve multiple complex parameter settings. Taking injection molding as an example, plastic products are widely used in industries such as automobiles, aviation, medical and electronics due to their excellent chemical stability, excellent electrical insulation and thermal insulation, high shock absorption and high mechanical strength.

[0003] Precision injection molding is one of the main methods for mass production of precision plastic products due to its high precision, high degree of automation and good adaptability to various complex geometric shapes. However, there are many process parameters in injection molding, and the coupling relationship is complex. It is very difficult to accurately quantify the impact of each parameter on the molding quality of plastic products. How to quickly and accurately obtain the optimal process parameters according to multiple quality requirements has always been a hot issue in many technical research. Summary of the invention

[0004] In order to solve the problems existing in the background technology, the present invention provides a multi-objective optimization method for production products for predicting processing quality.

[0005] The technical solution adopted by the present invention is:

[0006] The multi-objective optimization method for machining quality prediction of the present invention comprises:

[0007] Step 1: Establish a preliminary high-dimensional surface proxy model of the production product based on Latin hypercube sampling.

[0008] Step 2: The preliminary high-dimensional surface proxy model of the production product is graded and interval merged, and then multiple stratified sampling and fitting are performed to obtain a Gaussian regression high-dimensional model with refined key parts; the key parts are the areas with the optimal quality target parameters.

[0009] Step 3: According to the Gaussian regression high-dimensional model, the comprehensive entropy weights of the quality target parameters of the production products are obtained. According to the comprehensive entropy weights, the multi-objective optimization NSGA3 algorithm (Non-dominated Sorting Genetic Algorithm III) is used to optimize the various process parameters of the production products to obtain the relatively optimal process parameters, and then the products are processed to achieve multi-objective optimization of the production products.

[0010] The method of the present invention determines a proxy model that fits more accurately near the optimal process parameters through stratified sampling and multiple Gaussian regression, and uses the multi-objective optimization method of NSGA3 combined with comprehensive entropy weights to finally obtain a globally relatively optimal process parameter combination to achieve multi-objective optimization of production products.

[0011] The step 1 is specifically as follows:

[0012] Step 1.1: Set the m initial process parameters and n quality target parameters to be optimized for the product, and determine the value range A0 of each initial process parameter, A0 = {C1 0 ,C2 0 ,…,C i 0 ,…,C m 0}, C1 0 、C2 0 ,…,C i 0 ,…,C m 0 are the value ranges of the 1st, 2nd, …, i, …, mth initial process parameters respectively;

[0013] Step 1.2: Perform Latin hypercube sampling within the value range A0 of each initial process parameter, so as to obtain an initial sample group B0 of each process parameter after k0 groups of sampling.

[0014] Step 1.3: Concatenate the initial sample group B0 of each process parameter after sampling and the n quality target parameters to obtain k0 groups of recombinant sample groups B0 of each process parameter after sampling based on the quality target parameters. * For each process parameter, the initial sample group B0 is sampled, and n quality target parameters are spliced ​​after m sampled process parameters to obtain the recombined sample group B0. * .

[0015] Step 1.4: For each set of reorganized sample groups B0 based on the initial process parameters of the quality target parameters * Perform Gaussian regression to obtain a preliminary high-dimensional surface proxy model O0.

[0016] For recombinant sample group B0 * When performing Gaussian regression, the distance and semi-variance between each two quality target parameters are first calculated, then the optimal coefficient is solved, and finally the optimal coefficient is used to restructure the sample group B0 * The semi-variance of * The target estimated values ​​of other unknown points will eventually reorganize the sample group B0 *Combined with the target estimated values ​​of other unknown points, the high-dimensional surface of the proxy model is fitted together to obtain the Bayesian high-dimensional surface proxy model.

[0017] The step 2 is specifically as follows:

[0018] Step 2.1: Set the total number of stratifications N and the number of grading levels L for each stratification, L = {l1,l2,…,l i ,…,l N}, l1, l2, …, l i ,…,l N These are the average grading levels of the 1st, 2nd, …, i, …, and Nth stratification processes respectively.

[0019] Step 2.2: Layer each quality target parameter in the preliminary high-dimensional surface proxy model O0. In the i-th layer, each quality target parameter is divided into l i The levels are averaged and graded to obtain the initial sample groups of each optimized process parameter when the quality target parameters of each level are optimal. Then, the initial sample groups of each optimized process parameter are interval-merged to obtain the final sample groups of each optimized process parameter.

[0020] Step 2.3: Perform Latin hypercube sampling on the final sample group of each optimized process parameter in step 2.2 to obtain k i Initial sample group B of each optimized process parameter i Then, the initial sample group B of each group of optimized process parameters is i After concatenating with n quality target parameters, k i Group B: Recombination sample group of each optimized process parameter based on quality target parameter i * .

[0021] Step 2.4: Reorganize the sample group B with each optimized process parameter i * And the reorganized sample group B of each optimized process parameter obtained after the last stratification process i-1 * After performing Gaussian surface fitting together, the high-dimensional Gaussian fitting surface O of the current i-th layer processing is obtained i .

[0022] Step 2.5: Determine whether the N-th layering process is completed. If not, repeat steps 2.2 to 2.4 until the N-th layering process is completed, and obtain the final high-dimensional Gaussian fitting surface O N As a Gaussian regression high-dimensional model for key part refinement, this surface is the hyperplane of the mapping relationship between multiple process parameters and multiple target quality parameters.

[0023] In the step 2.2, when the initial sample groups of the optimized process parameters when the quality target parameters at each level are optimal are mutually exclusive, the initial sample groups of the optimized process parameters when the quality target parameters at each level are optimal are taken as the union to obtain the final sample groups of the optimized process parameters; when the initial sample groups of the optimized process parameters when the quality target parameters at each level are optimal intersect or contain each other, the initial sample groups of the optimized process parameters when the quality target parameters at each level are optimal are taken as the intersection to obtain the final sample groups of the optimized process parameters.

[0024] The step 3 is as follows:

[0025] Step 3.1: Fit the high-dimensional Gaussian to the surface O N After normalization, the normalized high-dimensional Gaussian fitting surface O is obtained. N , calculate and obtain the normalized high-dimensional Gaussian fitting surface O N The entropy value of each quality target parameter in .

[0026] High-dimensional Gaussian fitting surface O N The details are as follows:

[0027]

[0028] Among them, x ij For the high-dimensional Gaussian fitting surface O N The target quality parameter is the i-th row and j-th column in .

[0029] Fit the high-dimensional Gaussian surface O N After normalizing the target quality parameters in the

[0030] Normalized entropy value W of the jth target quality value j as follows:

[0031]

[0032] Among them, g j is the entropy value W of the jth target quality value j The coefficient of variation, g j =1-e j ;e j is the weight of the j-th target quality value, as follows:

[0033]

[0034] Among them, P ij is the weight of the target quality parameter in the i-th row and j-th column.

[0035] Step 3.2: Obtain the comprehensive entropy weight according to the entropy value of each target quality parameter.

[0036] Step 3.3: Take each target quality parameter and the product of each target quality parameter and the comprehensive entropy weight as the fitness function of the multi-objective optimization NSGA3 algorithm, take each optimized process parameter when each target quality parameter is optimal as the ideal point of the multi-objective optimization NSGA3 algorithm, take each target quality parameter and the minimum comprehensive entropy weight as the optimization target of the multi-objective optimization NSGA3 algorithm to perform multi-objective optimization, and obtain the final Pareto frontier as the relatively optimal process parameter combination.

[0037] In step 4.3, firstly, the entropy weight Δ is obtained according to the entropy value of each target quality parameter, as follows:

[0038]

[0039] Among them, W j is the normalized high-dimensional Gaussian fitting surface O N The target quality parameter in the i-th row and j-th column The entropy value of .

[0040] Then the entropy value of each target quality parameter is multiplied by the target quality parameter to obtain the comprehensive entropy weight.

[0041] The electronic device of the present invention comprises: a memory and a processor coupled to each other, wherein the memory stores program data, and the processor calls the program data to execute the method as described above.

[0042] The computer-readable storage medium of the present invention stores program data thereon, and when the program data is executed by a processor, the method described above is implemented.

[0043] After determining the preliminary process parameter interval of the product to be produced, the method of the present invention obtains the initial sample points through Latin hypercube sampling, obtains the target quality value of the corresponding sample through experiments or simulations; uses a Gaussian process to perform high-dimensional surface fitting on the sample points containing the target quality value, grades the target quality, and obtains the process parameter combination corresponding to the optimal level; merges multiple process parameter intervals according to an interval merging criterion, and performs Latin hypercube sampling in the merged interval, that is, secondary stratified sampling; performs experiments on new samples to obtain the target quality value and re-fits the high-dimensional surface; calculates the entropy weight of each target quality value based on all corresponding sample points of each target quality value, and obtains the comprehensive entropy weight of each target quality value; performs multi-objective optimization on the injection molding process parameters based on the NSGA3 method in combination with the comprehensive entropy weight, obtains the relatively optimal process parameter combination, and finally processes the product to achieve a multi-objective optimization method for the product.

[0044] The beneficial effects of the present invention are:

[0045] The present invention reduces the cost of fitting a high-progress proxy model of a key part by layered sampling of a high-dimensional surface, reduces the experimental cost under the same accuracy, and improves the accuracy of solving the optimal process parameters by the proxy model.

[0046] The present invention adopts the comprehensive entropy weight of the product quality target parameter as the optimization basis of the new NSGA3, improves the reliability of the NSGA3 optimization process, improves the efficiency of the NSGA3 optimization process, and finally performs better than the response surface method.

[0047] The method of the present invention can solve the problem that the coupling relationship of process parameters is complex and difficult to quantify accurately and efficiently, thereby improving the quality of products. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] Figure 1 Schematic diagram of a three-dimensional model of a thin-walled propeller blade in an embodiment of the present invention;

[0049] Figure 2 This is a preliminary high-dimensional surface proxy model fitting diagram in an embodiment of the present invention, where: Figure 2 (a) is a preliminary relationship fitting diagram of the melt temperature and mold surface temperature of the thin-walled propeller blade and the maximum warpage value in the embodiment, Figure 2 (b) is a preliminary relationship fitting diagram between the injection time and injection pressure of the thin-walled propeller blade in the embodiment and the maximum shrinkage mark, Figure 2 (c) is a preliminary relationship fitting diagram of the injection pressure and cooling time of the thin-walled propeller blades in the embodiment and the maximum volume shrinkage rate;

[0050] Figure 3 This is a Gaussian regression high-dimensional model fitting diagram after completing a stratified sampling in an embodiment of the present invention, wherein: Figure 3 (a) is a fitting diagram of the relationship between the melt temperature and the mold surface temperature of the thin-walled propeller blade in the embodiment and the maximum warpage value after completing a layered sampling, Figure 3 (b) is a fitting diagram of the relationship between the injection time and injection pressure of the thin-walled propeller blade in the embodiment and the maximum shrinkage mark after completing the first layered sampling. Figure 3 (c) is a fitting diagram of the relationship between the injection pressure and cooling time of the thin-walled propeller blade in the embodiment and the maximum volume shrinkage rate after completing the first layered sampling;

[0051] Figure 4 This is a comparison diagram of random fitting sample points and simulation points in an embodiment of the present invention, where: Figure 4 (a) is a comparison diagram of random fitting sample points and simulation points of the maximum warpage value of the thin-walled propeller blade in an embodiment of the present invention, Figure 4(b) is a comparison diagram of random fitting sample points and simulation points of the maximum volume shrinkage rate of the thin-walled propeller blade in an embodiment of the present invention, Figure 4 (c) is a comparison diagram of random fitting sample points and simulation points of the maximum shrinkage mark of the thin-walled propeller blade in an embodiment of the present invention;

[0052] Figure 5 This is a multi-objective optimization result diagram of a thin-walled propeller blade in an embodiment of the present invention, where: Figure 5 (a) is an optimization result diagram of the maximum warpage value of the thin-walled propeller blade in an embodiment of the present invention, Figure 5 (b) is an optimization result diagram of the maximum volume shrinkage of the thin-walled propeller blade in the embodiment of the present invention, Figure 5 (c) is an optimization result diagram of the maximum shrinkage mark of the thin-walled propeller blade in an embodiment of the present invention;

[0053] Figure 6 The multi-objective optimization result diagram of the thin-walled propeller blade of the comparative response surface method of the present invention is shown in FIG. Figure 6 (a) is a graph showing the maximum warpage value optimization result of a thin-walled propeller blade using the response surface method of the present invention. Figure 6 (b) is a graph showing the maximum volume shrinkage optimization result of a thin-walled propeller blade using the response surface method of the present invention. Figure 6 (c) is a graph showing the maximum shrinkage mark optimization result of a thin-walled propeller blade using the comparative response surface method of the present invention;

[0054] Figure 7 The figure is an overall flow chart of the method of the present invention. DETAILED DESCRIPTION

[0055] The present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0056] like Figure 7 As shown, the multi-objective optimization method for machining quality prediction of the present invention is characterized by comprising:

[0057] Step 1: Establish a preliminary high-dimensional surface proxy model of the production product based on Latin hypercube sampling. The preliminary high-dimensional surface proxy model reflects the mapping relationship between process parameters and product quality, as follows:

[0058] Step 1.1: Set the m initial process parameters and n quality target parameters to be optimized for the product, and determine the value range A0 of each initial process parameter, A0 = {C1 0 ,C2 0 ,…,C i 0 ,…,C m 0}, C1 0 、C20 ,…,C i 0 ,…,C m 0 are the value ranges of the 1st, 2nd, …, i, …, m initial process parameters respectively; in specific implementation, when processing injection molded products, the surface temperature, melt temperature, cooling time, injection pressure range and total cooling time of the injection molded products are used as process parameters, and the maximum warpage value, maximum volume shrinkage rate and maximum shrinkage mark size are used as quality target parameters, and the quality target parameters are the defect values ​​of the injection molded products.

[0059] Step 1.2: Perform Latin hypercube sampling within the value range A0 of each initial process parameter, so as to obtain an initial sample group B0 of each process parameter after k0 groups of sampling.

[0060] Step 1.3: Concatenate the initial sample group B0 of each process parameter after sampling and the n quality target parameters to obtain k0 groups of recombinant sample groups B0 of each process parameter after sampling based on the quality target parameters. * For each process parameter, the initial sample group B0 is sampled, and n quality target parameters are spliced ​​after m sampled process parameters to obtain the recombined sample group B0. * .

[0061] Step 1.4: For each set of reorganized sample groups B0 based on the initial process parameters of the quality target parameters * Perform Gaussian regression to obtain a preliminary high-dimensional surface proxy model O0.

[0062] For recombinant sample group B0 * When performing Gaussian regression, the distance and semi-variance between each two quality target parameters are first calculated, then the optimal coefficient is solved, and finally the optimal coefficient is used to restructure the sample group B0 * The semi-variance of * The target estimated values ​​of other unknown points will eventually reorganize the sample group B0 * Combined with the target estimated values ​​of other unknown points, the high-dimensional surface of the proxy model is fitted together to obtain the Bayesian high-dimensional surface proxy model.

[0063] Step 2: The preliminary high-dimensional surface proxy model of the product is graded and interval merged, and then multiple stratified sampling and fitting are performed to obtain a Gaussian regression high-dimensional model with key parts refined; the key parts are the areas with the best quality target parameters. Step 2 is as follows:

[0064] Step 2.1: Set the total number of stratifications N and the number of grading levels L for each stratification, L = {l1,l2,…,l i,…,l N}, l1, l2, …, l i ,…,l N These are the average grading levels of the 1st, 2nd, …, i, …, and Nth stratification processes respectively.

[0065] Step 2.2: Layer each quality target parameter in the preliminary high-dimensional surface proxy model O0. In the i-th layer, each quality target parameter is divided into l i The levels are averaged and graded to obtain the initial sample groups of each optimized process parameter when the quality target parameters of each level are optimal, and then the initial sample groups of each optimized process parameter are interval-merged to obtain the final sample groups of each optimized process parameter. When the initial sample groups of each optimized process parameter when the quality target parameters of each level are optimal are mutually exclusive, the initial sample groups of each optimized process parameter when the quality target parameters of each level are optimal are taken as the union to obtain the final sample groups of each optimized process parameter; when the initial sample groups of each optimized process parameter when the quality target parameters of each level are optimal intersect or contain each other, the initial sample groups of each optimized process parameter when the quality target parameters of each level are optimal are taken as the intersection to obtain the final sample groups of each optimized process parameter.

[0066] Step 2.3: Perform Latin hypercube sampling on the final sample group of each optimized process parameter in step 2.2 to obtain k i Initial sample group B of each optimized process parameter i Then, the initial sample group B of each group of optimized process parameters is i After concatenating with n quality target parameters, k i Group B: Recombination sample group of each optimized process parameter based on quality target parameter i * .

[0067] Step 2.4: Reorganize the sample group B with each optimized process parameter i * And the reorganized sample group B of each optimized process parameter obtained after the last stratification process i-1 * After performing Gaussian surface fitting together, the high-dimensional Gaussian fitting surface O of the current i-th layer processing is obtained i .

[0068] Step 2.5: Determine whether the N-th layering process is completed. If not, repeat steps 2.2 to 2.4 until the N-th layering process is completed, and obtain the final high-dimensional Gaussian fitting surface O NAs a Gaussian regression high-dimensional model for key part refinement, this surface is the hyperplane of the mapping relationship between multiple process parameters and multiple target quality parameters.

[0069] Step 3: Obtain the comprehensive entropy weights of each quality target parameter of the product according to the Gaussian regression high-dimensional model, and use the multi-objective optimization NSGA3 algorithm to optimize the process parameters of the product according to the comprehensive entropy weights to obtain the relatively optimal process parameters, and then process the product to achieve multi-objective optimization of the product. Step 3 is as follows:

[0070] Step 3.1: Fit the high-dimensional Gaussian to the surface O N After normalization, the normalized high-dimensional Gaussian fitting surface O is obtained. N , calculate and obtain the normalized high-dimensional Gaussian fitting surface O N The entropy value of each quality target parameter in .

[0071] High-dimensional Gaussian fitting surface O N The details are as follows:

[0072]

[0073] Among them, x ij For the high-dimensional Gaussian fitting surface O N The target quality parameter is the i-th row and j-th column in .

[0074] Fit the high-dimensional Gaussian surface O N After normalizing the target quality parameters in the

[0075] Normalized entropy value W of the jth target quality value j as follows:

[0076]

[0077] Among them, g j is the entropy value W of the jth target quality value j The coefficient of variation, g j =1-e j ;e j is the weight of the j-th target quality value, as follows:

[0078]

[0079] Among them, P ij is the weight of the target quality parameter in the i-th row and j-th column.

[0080] Step 3.2: Obtain the comprehensive entropy weight according to the entropy value of each target quality parameter. First, obtain the entropy weight Δ according to the entropy value of each target quality parameter, as follows:

[0081]

[0082] Among them, W j is the normalized high-dimensional Gaussian fitting surface O N The target quality parameter in the i-th row and j-th column The entropy value of .

[0083] Then the entropy value of each target quality parameter is multiplied by the target quality parameter to obtain the comprehensive entropy weight.

[0084] Step 3.3: Take each target quality parameter and the product of each target quality parameter and the comprehensive entropy weight as the fitness function of the multi-objective optimization NSGA3 algorithm, take each optimized process parameter when each target quality parameter is optimal as the ideal point of the multi-objective optimization NSGA3 algorithm, take each target quality parameter and the minimum comprehensive entropy weight as the optimization target of the multi-objective optimization NSGA3 algorithm to perform multi-objective optimization, and obtain the final Pareto frontier as the relatively optimal process parameter combination.

[0085] The multi-objective optimization based on the multi-objective optimization NSGA3 algorithm is as follows:

[0086] First, generate the initial reference point, P t is the parent of the tth generation, whose size is N, and the children it generates are Q t , whose size is also N; combine the child and the parent to form R t =P t ∪R t , R t The size is 2N, and N individuals are selected from them, and R t Divide it into multiple non-dominated layers (F1, F2, ...) through non-dominated sorting, and then construct a new population S starting from F1 t , until its size is N or exceeds N for the first time, and the population exceeding N is eliminated.

[0087] Divide the population into non-dominated layers and place the population members from level 1 to level 1 into S t In; if |S t |=N, then there is no need to perform the following operations, and directly set P t+1 =S t ; if |S t |>N, then The remaining part K = N-|P t+1 |From F l Select in.

[0088] Then perform population adaptive normalization and define the population S t The ideal point is Minimum value of The ideal point of the present invention is to use the proxy model fitted by the Bayesian method. N The optimal point corresponding to each objective function obtained by direct single-objective optimization is the corresponding process parameter combination D*; the worst point is the process parameter combination D** corresponding to the maximum value of the single defect value in the proxy model fitted by the Bayesian method; then the intercept of the hyperplane and the coordinate axis is used to normalize the objective function: the individual and the reference point are connected, that is, the reference line of each individual closest to the reference point is found.

[0089] Then, the niche preservation operation is performed, one-to-many or one-to-one or no corresponding relationship; K reference points are selected for the next round of iteration. The principle is that the individuals corresponding to the reference points with few connections should be retained to maintain diversity, that is, J min is the reference point with the smallest connection, but there may be more than one, so we can select any one of them as the reference point j; let I j is the individual associated with reference point j, if Then reselect the reference point if Then look at the correlation coefficient ρ j :ρ j =0, from I j Select the individual with the smallest distance j to enter the next generation; ρ j ≠0, then from I j An individual is randomly selected to enter the next generation.

[0090] Then the genetic selection process is carried out to determine the fitness function and ideal point of NSGAIII, and then the multi-objective optimization of the optimization target is carried out to finally obtain the relatively optimal combination of process parameters.

[0091] The method of the present invention determines a proxy model that fits more accurately near the optimal process parameters through stratified sampling and multiple Gaussian regression, and uses the multi-objective optimization method of NSGA3 combined with comprehensive entropy weights to finally obtain a globally relatively optimal process parameter combination to achieve multi-objective optimization of production products.

[0092] The specific embodiments of the present invention are as follows:

[0093] Step 1: First determine the parameters for multi-objective optimization: Taking injection molding as an example, the mold surface temperature, melt temperature, cooling time, injection pressure range and total cooling time are used as process parameters, and the maximum warpage value, maximum volume shrinkage rate and maximum shrinkage mark size are used as quality target parameters.

[0094] Taking thin-walled propeller blades as an example, Moldflow synergy software was used for mold flow simulation to construct DualDomain meshes, with a total of 9120 mesh patches. Figure 1 As shown; using the pouring position analysis sequence, the appropriate injection gate matching distribution diagram is analyzed and obtained, and using thermoplastic injection molding, the glass fiber nylon material Vydyne 909Natural is selected as the injection material.

[0095] Then, Latin hypercube sampling was performed within the preliminary process parameter value range, and the preliminary sample points were set to 20. The results are shown in Table 1.

[0096] Table 1

[0097]

[0098]

[0099] Then, perform experiments or simulations on the sample point B0: Use Moldflow to simulate and obtain the data set B0 containing the quality objective function value corresponding to the sample point B0 * , the results are shown in Table 2;

[0100] Table 2

[0101]

[0102] Then for B0 * Perform Gaussian regression to obtain the preliminary high-dimensional surface proxy model O0. The final result is as follows Figure 2 (a) Figure 2 (b) and Figure 2 As shown in (c), the x-axis and y-axis of each row are melt temperature and melt surface temperature, injection time and injection pressure, and injection pressure and cooling time, respectively, and the ordinate of each column is maximum warpage value, maximum sink mark depth and maximum volume shrinkage rate, respectively.

[0103] Step 2: For the sake of convenience, the number of layers is 2, and the number of sampling points in each layer is 20; according to the obtained high-dimensional surface proxy model O0, its quality objective function is graded and calculated according to 5 to obtain the optimal corresponding process parameter combination, as shown in Table 3:

[0104] Table 3

[0105]

[0106] Latin hypercube sampling was used again to determine the parameters of the first batch, with the number of samples being 20, and Moldflow was used for simulation to obtain sample group B1, as shown in Table 4:

[0107] Table 4

[0108]

[0109] Then, using sample group B1, Moldflow is used for simulation to obtain data group B1 containing the quality objective function value corresponding to sample point B1. * , as shown in Table 5:

[0110] Table 5

[0111]

[0112]

[0113] Then the new high-dimensional Gaussian fitting surface O1 is obtained by joint fitting, such as Figure 3 (a) Figure 3 (b) and Figure 3 As shown in (c), it can be seen from the figure that compared with Figure 2 , Figure 3 The fitting at key locations becomes more refined. By randomly extracting sample points from the high-dimensional Gaussian fitting surface, and using the process parameters of the corresponding points for moldflow simulation, the sample points are compared with the fitting values ​​on the high-dimensional surface. The comparison results are shown in the figure. Figure 4 (a) Figure 4 (b) and Figure 4 As shown in (c), the average error between the fitted value and the actual value of the maximum warpage value is 1.80%, the average error between the maximum volume shrinkage rate and the actual value is 7.37%, and the average error between the fitted value and the actual value of the maximum sink mark estimation value is 2.60%. The fitting effect is good, which proves the reliability of the fitting surface.

[0114] Step 3: The calculated comprehensive entropy weights are shown in Table 6:

[0115] Table 6

[0116]

[0117] Then, the NSGA3 algorithm and the comprehensive entropy weight were used to calculate the melt temperature of 21.49°C, the melt surface temperature of 255.61°C, the injection time of 2.97s, the injection pressure of 101.13MPa, and the cooling time of 22.88s.

[0118] The results of moldflow simulation of thin-walled propeller blades are as follows Figure 5 (a) Figure 5 (b) and Figure 5 As shown in (c), the simulation results obtained using the standard commercial software design-expert are as follows Figure 6 (a) Figure 6 (b) and Figure 6 The simulation comparison results of moldflow and standard commercial software design-expert are shown in Table 7:

[0119] Table 7

[0120]

[0121] Comparing the results of the two methods, all defect values ​​obtained by the proposed method are smaller than those obtained by the response surface method: the maximum warpage value is reduced by 1.93%, the maximum sink mark estimation value is reduced by 3.64%, and the maximum volume shrinkage rate is reduced by 1.54%.

[0122] It can be seen that the method of the present invention improves the reliability of the NSGA3 optimization process, and the final performance effect is better than the response surface method. It can solve the problem that the coupling relationship of process parameters is complex and difficult to quantify accurately and efficiently, thereby improving the quality of products.

[0123] Although the present invention has been disclosed as above with preferred embodiments, it is not intended to limit the present invention; those with ordinary knowledge in the technical field to which the present invention belongs can make various changes and modifications without departing from the spirit and scope of the present invention; therefore, the protection scope of the present invention shall be determined by that defined in the claims.

Claims

1. A multi-objective optimization method for machining quality prediction, characterized in that: include: Step 1: Establish a preliminary high-dimensional surface proxy model of the production product based on Latin hypercube sampling; Step 2: The preliminary high-dimensional surface proxy model of the product is subjected to hierarchical processing and interval merging, and then multiple stratified sampling and fitting are performed to obtain a Gaussian regression high-dimensional model with key parts refined; Step 3: According to the Gaussian regression high-dimensional model, the comprehensive entropy weights of the quality target parameters of the production products are obtained. According to the comprehensive entropy weights, the multi-objective optimization NSGA3 algorithm is used to optimize the various process parameters of the production products to obtain the relatively optimal process parameters, and then the products are processed to achieve multi-objective optimization of the production products.

2. The multi-objective optimization method for machining quality prediction according to claim 1, characterized in that: The step 1 is specifically as follows: Step 1.1: Set the m initial process parameters and n quality target parameters to be optimized for the product, and determine the value range A0 of each initial process parameter, A0 = {C1 0 , C2 0 , ..., C i 0 , .., C m 0 }, C1 0 、C2 0 ,…,C i 0 ,…,C m 0 are the value ranges of the 1st, 2nd, …, i, …, mth initial process parameters respectively; Step 1.2: Perform Latin hypercube sampling within the value range A0 of each initial process parameter, so as to obtain the initial sample group B0 of each process parameter after k0 groups of sampling; Step 1.3: concatenate the initial sample group B0 of each set of sampled process parameters and the n quality target parameters to obtain k0 sets of recombined sample groups B0* of each set of sampled process parameters based on the quality target parameters; Step 1.4: Perform Gaussian regression on each reorganized sample group B0* of each initial process parameter based on the quality target parameter to obtain a preliminary high-dimensional surface proxy model O0.

3. The multi-objective optimization method for machining quality prediction according to claim 1, characterized in that: The step 2 is specifically as follows: Step 2.1: Set the total number of stratifications N and the number of grading levels L for each stratification in the stratification process, L = {l1, l2, .., l i , .., l N }, l1, l2, …, l i ,…,l N are the average grading levels of the 1st, 2nd, …, i, …, Nth stratification treatments respectively; Step 2.2: Layer each quality target parameter in the preliminary high-dimensional surface proxy model O0. In the i-th layer, each quality target parameter is divided into l j The levels are averaged and graded to obtain the initial sample groups of each optimized process parameter when the quality target parameters of each level are optimal. Then, the initial sample groups of each optimized process parameter are interval-merged to obtain the final sample groups of each optimized process parameter. Step 2.3: Perform Latin hypercube sampling on the final sample group of each optimized process parameter in step 2.2 to obtain k i Initial sample group B of each optimized process parameter i Then, the initial sample group B of each group of optimized process parameters is i After concatenating with n quality target parameters, k i Group B: Recombination sample group of each optimized process parameter based on quality target parameter j *; Step 2.4: Reorganize the sample group B with each optimized process parameter i * and the reorganized sample group B with each optimized process parameter obtained after the last stratification process i-1 * After performing Gaussian surface fitting together, the high-dimensional Gaussian fitting surface O of the current i-th layer processing is obtained i ; Step 2.5: Determine whether the N-th layering process is completed. If not, repeat steps 2.2 to 2.4 until the N-th layering process is completed, and obtain the final high-dimensional Gaussian fitting surface O N Gaussian regression high-dimensional model as key part refinement.

4. The multi-objective optimization method for machining quality prediction according to claim 3 is characterized in that: In the step 2.2, when the initial sample groups of the optimized process parameters when the quality target parameters at each level are optimal are mutually exclusive, the initial sample groups of the optimized process parameters when the quality target parameters at each level are optimal are combined to obtain the final sample groups of the optimized process parameters; When the initial sample groups of the optimized process parameters when the quality target parameters at each level are optimal intersect or contain each other, the final sample groups of the optimized process parameters are obtained by taking the intersection of the initial sample groups of the optimized process parameters when the quality target parameters at each level are optimal.

5. The multi-objective optimization method for machining quality prediction according to claim 1, characterized in that: The step 3 is as follows: Step 3.1: Fit the high-dimensional Gaussian to the surface O N After normalization, the normalized high-dimensional Gaussian fitting surface O is obtained. N , calculate and obtain the normalized high-dimensional Gaussian fitting surface O N The entropy value of each quality target parameter in; Step 3.2: Obtain comprehensive entropy weight according to the entropy value of each target quality parameter; Step 3.3: Take each target quality parameter and the product of each target quality parameter and the comprehensive entropy weight as the fitness function of the multi-objective optimization NSGA3 algorithm, take each optimized process parameter when each target quality parameter is optimal as the ideal point of the multi-objective optimization NSGA3 algorithm, take each target quality parameter and the minimum comprehensive entropy weight as the optimization target of the multi-objective optimization NSGA3 algorithm to perform multi-objective optimization, and obtain the final Pareto frontier as the relatively optimal process parameter combination.

6. The multi-objective optimization method for machining quality prediction according to claim 5, characterized in that: In step 3.2, firstly, the entropy weight Δ is obtained according to the entropy value of each target quality parameter, as follows: Among them, W j is the normalized high-dimensional Gaussian fitting surface O N The target quality parameter in the i-th row and j-th column The entropy value of Then the entropy value of each target quality parameter is multiplied by the target quality parameter to obtain the comprehensive entropy weight.

7. An electronic device, characterized in that: include: A memory and a processor coupled to each other, wherein the memory stores program data, and the processor calls the program data to execute the method according to any one of claims 1 to 6.

8. A computer-readable storage medium having program data stored thereon, characterized in that: When the program data is executed by a processor, the method according to any one of claims 1 to 6 is implemented.

Images