A method for designing temperature parameters of squeeze casting process

Abstract

The present invention provides a method for designing process temperature parameters of an extrusion casting process, which belongs to the technical field of extrusion casting. The process parameter design method first performs an initial design of the pouring temperature and the mold preheating temperature from the material composition level based on the KNN algorithm with reference to the temperature process parameters of similar material castings. Then, the mold preheating temperature with a large design error is corrected by combining the casting parameters and a generalized regression neural network (GRNN), and an error correction model is established. The design method is based on existing research data, fully references and utilizes existing related research results, does not require tedious experimental research, and has a high design accuracy of the process parameters. Compared with the existing process parameter design method, a large amount of time cost and experimental cost are saved. The method is universal and is not only applicable to the design of extrusion casting process parameters, but also to parameter design in other similar fields.

Description

Technical Field

[0001] The invention relates to the technical field of extrusion casting, and in particular to a method for designing temperature parameters of an extrusion casting process. Background Art

[0002] The process parameters of squeeze casting directly affect the mechanical properties and microstructure of the casting. Reasonable process parameters are the premise and key to produce squeeze castings with the above advantages. Among them, temperature parameters (pouring temperature and mold preheating temperature) are one of the most important parameters. Only at a reasonable temperature can the casting solidify smoothly and produce a complete and high-quality casting. Inappropriate temperature will cause insufficient fluidity of the molten metal, resulting in insufficient pouring or premature solidification, uneven distribution of the solidification area, shrinkage holes, shrinkage and other defects, affecting the mechanical properties of the squeeze casting.

[0003] The existing squeeze casting process parameter design methods mainly include three methods: obtaining process parameters completely based on experimental research, obtaining process parameters based on experimental data modeling optimization, and theoretical calculation. The method of obtaining process parameters completely based on experimental research and the method of obtaining process parameters based on experimental data modeling optimization require experiments, rely on experimental data, and can only be tested on a single material and casting each time, while the material composition and casting geometry are diverse, so their adaptability and flexibility are poor (even for numerical simulation, replacement of castings requires remodeling, mold design, iteration, and time-consuming); and the cost is high, and it takes a long time to obtain process parameters. For the third method, there are currently only some process parameter design formulas, which are mostly for a certain type of casting with uniform characteristics. There are great limitations, ignoring the coupling effect between process parameters, and rarely taking into account the interaction between process parameters, resulting in large errors between theory and practice. It can be seen that the existing process parameter design methods have their own shortcomings and are difficult to meet the needs of rapid development and rating application of squeeze casting process parameters.

[0004] With the advent of the big data era, data has become a means of production. There is an urgent need to strengthen the application of existing extrusion casting process data and to build a data-driven process parameter design method with reference to existing process parameter data in order to improve the design efficiency of extrusion casting process parameters and reduce costs. Summary of the invention

[0005] The purpose of the present invention is to provide a method for designing temperature parameters of an extrusion casting process, so as to solve the technical problems of low efficiency and high cost of the existing extrusion casting process.

[0006] The temperature process parameter design method firstly performs an initial design of the pouring temperature and mold preheating temperature from the material composition level based on the KNN algorithm and the temperature process parameters of similar material castings. Then, the mold preheating temperature with large design error is corrected by combining the casting parameters and the generalized regression neural network (GRNN), and an error correction model is established. This design method is based on existing research data and fully refers to and utilizes the existing relevant research results. It does not require tedious experimental research and has high design accuracy of process parameters. Compared with the existing process parameter design methods, it saves a lot of time and experimental costs. This method is universal and is not only suitable for the design of extrusion casting process parameters, but also for parameter design in other similar fields.

[0007] According to the extrusion casting process parameter design method of the present invention, the influencing factors of the extrusion casting process parameters and their weights are comprehensively considered, and the process parameter design of the pouring temperature and the mold preheating temperature is realized.

[0008] This method combines the K-Nearest Neighbor (KNN) algorithm and the generalized regression neural network (GRNN). The former preliminarily designs the initial values ​​of the temperature parameters of extrusion casting, and the latter corrects the process parameters with large errors in the initial design values, and finally solves the required process parameters. The proposed temperature process parameter design method does not require tedious experimental research, reduces the cost of obtaining process parameters, and achieves the purpose of quickly designing process parameters. It is not only applicable to the process parameter design in the field of extrusion casting, but also to the process parameter design in other similar fields, and has universal applicability. In addition, this method helps to strengthen the application of existing data, which is in line with the trend of the times that emphasizes data utilization in the context of the big data era, and is conducive to digital and intelligent manufacturing in the manufacturing field.

[0009] In order to achieve the above object, the technical solution adopted by the present invention is as follows:

[0010] A method for designing temperature parameters of an extrusion casting process, the method comprising the following steps:

[0011] Step 1: Representation of squeeze casting process data. For an existing squeeze casting process parameter data set DB, it is represented as DB = {S i |i=1,2,…,m}, m is the number of data samples;

[0012] Step 2: Initial design of process parameters based on KNN algorithm. Take the existing data samples in DB as reference objects, use KNN algorithm to find K neighbor samples with similar material composition to the designed casting, and determine the influence weight according to the distance between the K neighbor samples and the material composition of the designed casting. Then, the weighted regression value of the corresponding temperature process parameter of the K neighbor samples is used as the initial design value of the temperature process parameter of the designed casting. Suppose the pouring temperature and mold preheating temperature of the i-th sample are y i1 and i2 , let the target casting be O, and the corresponding material composition be {o1,o2,…,o θ}, the pouring temperature and mold preheating temperature are y o1 and o2 .

[0013] Step 3: Temperature parameter correction based on GRNN algorithm. The pouring temperature predicted by KNN algorithm has a small error and can be directly applied. However, the mold preheating temperature prediction value has a large error, and the average design error is 12-14%. Because the selection of mold preheating temperature is related to the solidification temperature of the alloy and the size and shape of the casting, only the mold preheating temperature is corrected below. First, we try to analyze the influencing factors other than material composition based on the data in the database DB, and obtain the correction model structure of the mold preheating temperature based on GRNN. It is a mentor network. The error correction model can be obtained through sample training. The error correction amount of the model fitting is summed with the initial value of the mold preheating temperature to complete the design of the process parameters.

[0014] Furthermore, the specific process of step 1 is as follows: each sample data contains attribute variables of squeeze casting process parameters, material composition data and shape feature data, and the sample attributes are divided into the influencing factors X of the process parameters according to the properties. k , k = 1, 2, ..., t, t represents the number of process influencing factors, process parameter Y j , j = 1, 2, ..., n, n represents the number of process parameters, data sample S i The corresponding attribute data are x ik and ij ,The process parameters include extrusion pressure, pouring temperature and mold preheating temperature.

[0015] Furthermore, the specific process of step 2 is:

[0016] Step 2.1: Preprocess the data and use data normalization method to eliminate the influence of different dimensions between different related attributes:

[0017]

[0018] In the formula, max(k) and min(k) represent the maximum and minimum values ​​of the kth column data in the process parameter database respectively;

[0019] Step 2.2: Calculate the distance between samples. The distance between samples in the feature space is measured by calculating the Euclidean distance, and K samples that are close to the target sample are selected according to the size of the distance. The sample distance calculation is as follows:

[0020]

[0021] Step 2.3: According to the distance d i The size of , in order from small to large, select K samples with the smallest distance from the material composition of O, that is, the nearest neighbor clustering sample set CS, CS = {S i′ |i′=1,2…,K}, and determine the weight ω of each cluster sample according to the distance relationship i′ , represents the degree of similarity to O, and the sample weight calculation is shown as follows:

[0022]

[0023] ω i′ is the weight of the i′th cluster sample, d i' is the calculated distance between the i′th cluster sample and the target case;

[0024] Step 2.4: According to the sample weights obtained in step 2.3, the required process parameters are weighted and summed to obtain the process parameters required for the target case. The target case process parameters are calculated as shown in the following formula:

[0025]

[0026]

[0027] In order to improve the accuracy, the optimal K value is obtained based on the KNN algorithm to construct the optimal clustering scale. After optimization, when K=8, the relative error and mean square error of the initial design values ​​of the pouring temperature and mold preheating temperature are close to the lowest level. At this time, the prediction effect of the KNN model is the most accurate and stable.

[0028] Furthermore, in step 3, the mold preheating temperature correction process based on the GRNN algorithm is:

[0029] Step 3.1: Identification of influencing factors of mold preheating temperature correction. Classify the relevant attributes of part complexity of existing samples in the database. The relevant attributes are attributes other than material composition (casting shape complexity, minimum wall thickness grade, etc.). Then, according to the classification, count the number of existing samples in the database belonging to the category, and calculate the arithmetic mean of the mold preheating temperature of the same type as the mold preheating temperature of this type of casting. On this basis, the least squares method is used to fit the relationship between the corresponding relevant attributes and the mold preheating temperature. The actual correlation is judged by the goodness of fit results. The most relevant factors with a goodness of fit exceeding 80% are taken as the most relevant factors and included in the input of GRNN. The least squares method fitting is implemented based on the PyCharm platform. The goodness of fit is calculated as follows:

[0030]

[0031] In the formula, is the mold preheating temperature value fitted by the least square method;

[0032] Step 3.2: Based on the mold preheating temperature related factors selected in step 3.1, a GRNN error correction model is established.

[0033] Furthermore, the GRNN error correction model includes an input layer, a pattern layer, a summation layer, and an output layer;

[0034] The input layer is the influencing factor variable. Since the correction is made based on the initial mold preheating temperature, the pouring temperature, the initial value of the mold preheating temperature, the minimum wall thickness level and the part complexity are taken as input vectors, expressed as D, and the corresponding output is the mold preheating temperature error value;

[0035] The pattern layer is used to calculate the distance information between the test and training sample input vectors. The number of neurons in the pattern layer is the same as the number of training samples of the model, that is, each pattern layer neuron corresponds to a training sample, and is fully connected with the input layer neurons, indicating that all input variables have an impact on the mold preheating temperature error. The neurons in the layer are not connected, that is, the mold preheating temperatures of different castings are independent, and the transfer function is:

[0036]

[0037] Where D λ is the input vector of the training sample corresponding to the λth neuron of the pattern layer, Do is the input vector of the target casting, P λ is the calculated value of the λth neuron in the pattern layer, q is the number of training samples, and σ is the hyperparameter smoothing factor;

[0038] The summation layer uses two types of neurons for summation, namely type A neurons and type B neurons;

[0039] There is only one type A neuron, and the calculation value is the output arithmetic sum of all pattern layer neurons:

[0040]

[0041] The number of B-type neurons is the same as the dimension of the output vector. It is necessary to perform weighted summation on all neurons in the pattern layer. The connection weight of the Nth B-type neuron in the summation layer and the λth neuron in the pattern layer is the Nth element l in the output vector of the learning sample corresponding to the neuron in the pattern layer. λN , N is a positive integer, and its calculation formula is:

[0042]

[0043] Where: l λN is the Nth output value corresponding to the learning sample corresponding to the λth pattern layer neuron.

[0044] The output layer is calculated according to the obtained output, that is, the correction value of the mold preheating temperature:

[0045]

[0046] Substituting into the following formula, we can get the final mold preheating temperature design value y′ o2 ;

[0047] y′ o2 =y o2 -Δy o2

[0048] In the formula, y′ o2 This is the corrected mold preheating temperature value.

[0049] The present invention has the following beneficial effects due to the adoption of the above technical solution:

[0050] The present invention does not require experimental research, escapes the tedious experimental research process, and enhances the utilization of existing data. It is not only suitable for the design of extrusion casting process parameters, but also has a reference role in the design of process parameters in other similar fields. The model combines the influence of material composition and casting factors on process parameters, and can design more accurate process parameters. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] Figure 1 It is the overall flow chart of the present invention.

[0052] Figure 2 It is a structural diagram of the GRNN model of the present invention. DETAILED DESCRIPTION

[0053] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below with reference to the accompanying drawings and preferred embodiments. However, it should be noted that many details listed in the specification are only for the purpose of enabling the reader to have a thorough understanding of one or more aspects of the present invention, and these aspects of the present invention can be implemented even without these specific details.

[0054] like Figure 1-2 As shown, a method for designing process parameters of extrusion casting comprises the following steps:

[0055] Step 1: Representation of squeeze casting process data. For an existing squeeze casting process parameter data set DB, it is represented as DB = {S i |i=1,2,…,m}, m is the number of data samples. Each sample data contains attribute variables of squeeze casting process parameters, material composition data and shape feature data. According to the properties, the sample attributes are divided into the influencing factors of process parameters X k , k = 1, 2, ..., t, t represents the number of process influencing factors, process parameter Y j , j = 1, 2, ..., n, n represents the number of process parameters, data sample S i The corresponding attribute data are x ik and ij ,The process parameters include extrusion pressure, pouring temperature and mold preheating temperature.

[0056] Table 1 is the squeeze casting process data DB

[0057]

[0058] Table 2 is a sample of squeeze casting process data

[0059]

[0060] Step 2: Initial design of process parameters based on KNN algorithm. Take the existing data samples in DB as reference objects, use KNN algorithm to find K neighbor samples with similar material composition to the designed casting, and determine the influence weight according to the distance between the K neighbor samples and the material composition of the designed casting. Then, the weighted regression value of the corresponding temperature process parameter of the K neighbor samples is used as the initial design value of the temperature process parameter of the designed casting. Suppose the pouring temperature and mold preheating temperature of the i-th sample are y i1 and i2 , let the target casting be O, and the corresponding material composition be {o1,o2,…,o θ}, the pouring temperature and mold preheating temperature are y o1 and o2 .

[0061] Step 2.1: Preprocess the data and use data normalization method to eliminate the influence of different dimensions between different related attributes:

[0062]

[0063] Where max(k) and min(k) represent the maximum and minimum values ​​of the kth column of data in the process parameter database, respectively.

[0064] Step 2.2: Calculate the distance between samples. The distance between samples in the feature space is measured by calculating the Euclidean distance, and K samples that are close to the target sample are selected according to the size of the distance. The sample distance calculation is as follows:

[0065]

[0066] Step 2.3: According to the distance d i The size of , in order from small to large, select K samples with the smallest distance from the material composition of O, that is, the nearest neighbor clustering sample set CS, CS = {S i′ |i′=1,2…,K}, and determine the weight ω of each cluster sample according to the distance relationship i′ , represents the degree of similarity to O, and the sample weight calculation is shown as follows:

[0067]

[0068] ω i′ is the weight of the i′th cluster sample, d i' is the calculated distance between the i′th cluster sample and the target case.

[0069] Step 2.4: According to the sample weights obtained in step 2.3, the required process parameters are weighted and summed to obtain the process parameters required for the target case. The target case process parameters are calculated as shown in the following formula:

[0070]

[0071]

[0072] Based on the KNN algorithm, the optimal K value is obtained to construct the optimal clustering scale. After optimization, when K = 8, the relative error and mean square error of the initial design values ​​of the pouring temperature and mold preheating temperature are close to the lowest level. At this time, the prediction effect of the KNN model is the most accurate and stable. The initial design values ​​of some pouring temperatures and mold preheating temperatures obtained based on K = 8 are shown in Table 3.

[0073] Table 3 Initial design values ​​of some temperature parameters

[0074]

[0075] Step 3: Temperature parameter correction based on GRNN algorithm. The pouring temperature predicted by KNN algorithm has a small error and can be directly applied. However, the mold preheating temperature prediction value has a large error, and the average design error is 12-14%. Because the selection of mold preheating temperature is related to the solidification temperature of the alloy and the size and shape of the casting, only the mold preheating temperature is corrected below. First, we try to analyze the influencing factors other than material composition based on the data in the database DB, and obtain the correction model structure of the mold preheating temperature based on GRNN. It is a mentor network. The error correction model can be obtained through sample training. The error correction amount of the model fitting is summed with the initial value of the mold preheating temperature to complete the design of the process parameters.

[0076] The mold preheating temperature correction process based on the GRNN algorithm is:

[0077] Step 3.1: Identification of influencing factors of mold preheating temperature correction. Classify the relevant attributes of part complexity of existing samples in the database. The relevant attributes are attributes other than material composition. Then, according to the classification, count the number of existing samples in the database belonging to the category, and calculate the arithmetic mean of the mold preheating temperature of the same type as the mold preheating temperature of this type of casting. On this basis, the least squares method is used to fit the relationship between the corresponding relevant attributes and the mold preheating temperature. The actual correlation is judged by the goodness of fit results. The most relevant factors with a goodness of fit exceeding 80% are taken as the input of GRNN. The least squares method fitting is implemented based on the PyCharm platform. The goodness of fit is calculated as follows:

[0078]

[0079] In the formula, It is the mold preheating temperature value fitted by the least squares method.

[0080] Table 4 shows the classification of squeeze casting shape complexity levels and their statistical relationships

[0081]

[0082] Table 5 shows the minimum thickness grades of extrusion castings and their statistical relationships

[0083]

[0084] At the same time, the cross-effect of pouring temperature on mold preheating temperature is considered, and the corresponding goodness of fit is obtained. As shown in Table 6, compared with part complexity and minimum casting thickness, the influence of pouring temperature on mold preheating temperature is more significant.

[0085] Table 6 shows the correlation of mold preheating temperature related properties

[0086]

[0087] Step 3.2: Based on the mold preheating temperature related factors selected in step 3.1, a GRNN error correction model is established.

[0088] The GRNN error correction model includes input layer, pattern layer, summation layer and output layer;

[0089] The input layer is the influencing factor variable. Since the correction is made based on the initial mold preheating temperature, the pouring temperature, the initial value of the mold preheating temperature, the minimum wall thickness grade and the part complexity are taken as the input vector, expressed as D, and the corresponding output is the mold preheating temperature error value;

[0090] The pattern layer is used to calculate the distance information between the test and training sample input vectors. The number of neurons in the pattern layer is the same as the number of training samples of the model, that is, each pattern layer neuron corresponds to a training sample, and is fully connected with the input layer neurons, indicating that all input variables have an impact on the mold preheating temperature error. The neurons in the layer are not connected, that is, the mold preheating temperatures of different castings are independent, and the transfer function is:

[0091]

[0092] Where D λ is the input vector of the training sample corresponding to the λth neuron of the pattern layer, Do is the input vector of the target casting, P λ is the calculated value of the λth neuron in the pattern layer, q is the number of training samples, and σ is the hyperparameter smoothing factor;

[0093] The summation layer uses two types of neurons for summation, namely type A neurons and type B neurons;

[0094] There is only one type A neuron, and the calculation value is the output arithmetic sum of all pattern layer neurons:

[0095]

[0096] The number of B-type neurons is the same as the dimension of the output vector. It is necessary to perform a weighted summation on all neurons in the pattern layer. For example, the connection weight between the first B-type neuron in the summation layer and the λth neuron in the pattern layer is the first element l in the output vector of the learning sample corresponding to the neuron in the pattern layer. λ1 (Mold preheating temperature error value), the calculation formula is:

[0097]

[0098] Where: l λ1 is the first output value corresponding to the learning sample corresponding to the λth pattern layer neuron.

[0099] The output layer is calculated according to the obtained output, that is, the correction value of the mold preheating temperature:

[0100]

[0101] Substituting into the following formula, we can get the final mold preheating temperature design value y′ o2 ;

[0102] y′ o2 =y o2 -Δy o2

[0103] In the formula, y′ o2 This is the corrected mold preheating temperature value.

[0104] Through experimental verification, the average error of the pouring temperature designed by this method is only 2.37%; the average error of the mold preheating temperature is only 6.38%. The error is very small and meets the use requirements of the squeeze casting process.

[0105] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principle of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.

Claims

1. A method for designing temperature parameters of an extrusion casting process, characterized in that: The method comprises the following steps: Step 1: Representation of squeeze casting process data. For an existing squeeze casting process parameter data set DB, it is represented as DB = {S i |i=1,2,…,m}, m is the number of data samples; Step 2: Initial design of process parameters based on KNN algorithm. Take the existing data samples in DB as reference objects, use KNN algorithm to find K neighbor samples with similar material composition to the designed casting, and determine the influence weight according to the distance between the K neighbor samples and the material composition of the designed casting. Then, the weighted regression value of the corresponding temperature process parameter of the K neighbor samples is used as the initial design value of the temperature process parameter of the designed casting. Suppose the pouring temperature and mold preheating temperature of the i-th sample are y i1 and i2 , let the target casting be O, and the corresponding material composition be {o1,o2,…,o θ }, the pouring temperature and mold preheating temperature are y o1 and o2 ; Step 3: Correction of temperature parameters based on GRNN algorithm. The error of the pouring temperature predicted by KNN algorithm is small and can be directly applied. However, the error of the predicted value of mold preheating temperature is large, and the average design error is 12-14%. Because the selection of mold preheating temperature is related to the solidification temperature of the alloy and the size and shape of the casting, only the mold preheating temperature is corrected below. First, try to analyze the influencing factors other than material composition based on the data in the database DB, and obtain the correction model structure of mold preheating temperature based on GRNN. It is a mentor network. The error correction model can be obtained through sample training. The error correction amount of model fitting and the initial value of mold preheating temperature are summed to complete the design of process parameters. In step 3, the mold preheating temperature correction process based on the GRNN algorithm is: Step 3.1: Identification of influencing factors of mold preheating temperature correction. Classify the relevant attributes of part complexity of existing samples in the database. The relevant attributes are attributes other than material composition. Then, according to the classification, count the number of existing samples in the database belonging to the category, and calculate the arithmetic mean of the mold preheating temperature of the same type as the mold preheating temperature of this type of casting. On this basis, the least squares method is used to fit the relationship between the corresponding relevant attributes and the mold preheating temperature. The actual correlation is judged by the goodness of fit results. The most relevant factors with a goodness of fit exceeding 80% are taken as the input of GRNN. The least squares method fitting is implemented based on the PyCharm platform. The goodness of fit is calculated as follows: Where y is the mold preheating temperature value fitted by the least squares method; Step 3.2: Based on the mold preheating temperature related factors selected in step 3.1, a GRNN error correction model is established; The GRNN error correction model includes input layer, pattern layer, summation layer and output layer; The input layer is the influencing factor variable. Since the correction is made based on the initial mold preheating temperature, the pouring temperature, the initial value of the mold preheating temperature, the minimum wall thickness grade and the part complexity are taken as the input vector, expressed as D, and the corresponding output is the mold preheating temperature error value; The pattern layer is used to calculate the distance information between the test and training sample input vectors. The number of neurons in the pattern layer is the same as the number of training samples of the model, that is, each pattern layer neuron corresponds to a training sample, and is fully connected with the input layer neurons, indicating that all input variables have an impact on the mold preheating temperature error. The neurons in the layer are not connected, that is, the mold preheating temperatures of different castings are independent, and the transfer function is: Where D λ is the input vector of the training sample corresponding to the λth neuron in the pattern layer, D o is the input vector of the target casting, P λ is the calculated value of the λth neuron in the pattern layer, q is the number of training samples, and σ is the hyperparameter smoothing factor; The summation layer uses two types of neurons for summation, namely type A neurons and type B neurons; There is only one type A neuron, and the calculation value is the output arithmetic sum of all pattern layer neurons: The number of B-type neurons is the same as the dimension of the output vector. It is necessary to perform weighted summation on all neurons in the pattern layer. The connection weight of the Nth B-type neuron in the summation layer and the λth neuron in the pattern layer is the Nth element l in the output vector of the learning sample corresponding to the neuron in the pattern layer. λN , N is a positive integer, and its calculation formula is: Where: l λN is the Nth output value corresponding to the learning sample corresponding to the λth pattern layer neuron; The output layer is calculated according to the obtained output, that is, the correction value of the mold preheating temperature: Substituting into the following formula, we can get the final mold preheating temperature design value y′ o2 ; y' o2 =y o2 -Δy o2 In the formula, y′ o2 This is the corrected mold preheating temperature value.

2. The method for designing temperature parameters of a squeeze casting process according to claim 1, characterized in that: The specific process of step 1 is as follows: each sample data contains attribute variables of squeeze casting process parameters, material composition data and shape feature data. According to the properties, the sample attributes are divided into the influencing factors of process parameters X k , k = 1, 2, ..., t, t represents the number of process influencing factors, process parameter Y j , j = 1, 2, ..., n, n represents the number of process parameters, data sample S i The corresponding attribute data are x ik and ij ,The process parameters include extrusion pressure, pouring temperature and mold preheating temperature.

3. The method for designing temperature parameters of a squeeze casting process according to claim 1, characterized in that: The specific process of step 2 is: Step 2.1: Preprocess the data and use data normalization method to eliminate the influence of different dimensions between different related attributes: In the formula, max(k) and min(k) represent the maximum and minimum values ​​of the kth column data in the process parameter database respectively; Step 2.2: Calculate the distance between samples. The distance between samples in the feature space is measured by calculating the Euclidean distance, and K samples that are close to the target sample are selected according to the size of the distance. The sample distance calculation is as follows: Step 2.3: According to the distance d i The size of , in order from small to large, select K samples with the smallest distance from the material composition of O, that is, the nearest neighbor clustering sample set CS, CS = {S i′ |i′=1,2…,K}, and determine the weight ω of each cluster sample according to the distance relationship i′ , represents the degree of similarity to O, and the sample weight calculation is shown as follows: ω i′ is the weight of the i′th cluster sample, d i' is the calculated distance between the i′th cluster sample and the target case; Step 2.4: According to the sample weights obtained in step 2.3, the required process parameters are weighted and summed to obtain the process parameters required for the target case. The target case process parameters are calculated as shown in the following formula: The optimal K value is obtained based on the KNN algorithm to construct the optimal clustering scale. After optimization, when K is a certain value, the relative error and mean square error of the initial design values ​​of the pouring temperature and the mold preheating temperature are close to the lowest level. At this time, the prediction effect of the KNN model is the most accurate and stable.

Images