A case-based reasoning design method for squeeze casting process parameters

Abstract

The present invention provides a case reasoning design method for extrusion casting process parameters, which belongs to the field of casting process parameter design. The method includes considering the collected process data as cases according to the characteristics of the extrusion casting process data, identifying problems and representing knowledge for the case library of the extrusion casting process data, determining the influencing factors according to the process parameters to be designed, constructing an uncertainty network structure between the process influencing factors and the process parameters, completing the simplification of the process influencing factors, optimizing the weights according to the Markov blanket structure between the process parameters and the process influencing factors to remove redundant attributes; based on the similarity between the cases, storing the predicted process parameter values ​​and their process influencing factors in the case library as new cases. Directly using existing data to complete the design of extrusion casting process parameters has good application value, high design efficiency, and can greatly save design time.

Description

Technical Field

[0001] The invention relates to the field of casting process parameter design, and in particular to a case-based reasoning design method for extrusion casting process parameters. Background Art

[0002] Squeeze casting is a near-net-shape manufacturing technology that combines the advantages of forging and casting. It can produce parts with high surface finish and good mechanical properties. It has been successfully used in the preparation of high-performance product parts and composite materials and has very good development potential.

[0003] Squeeze casting process parameters are key factors in the quality of extrusion casting, mainly including pouring temperature, extrusion pressure, holding time and mold preheating temperature, which have an important influence on the microstructure and performance of extrusion castings. Many scholars at home and abroad have conducted a lot of research on the influence of different extrusion casting process parameters on castings, but there are few studies specifically on the design (or selection) of extrusion casting process parameters. By summarizing the relevant research on extrusion casting process parameters, the current methods of obtaining extrusion casting process parameters can be summarized into four types: (1) obtaining process parameters based entirely on physical experiments, (2) obtaining process parameters based on the optimization of physical experimental results, (3) obtaining process parameters based on numerical simulation experiments and (4) calculation based on empirical formulas. The first method is usually used for the preparation of extrusion casting materials. By designing physical experiments to study the influence of specific process parameters on casting properties, the optimal process parameters and the best material properties are determined. The optimal process parameters selected by method (1) are the best results at the experimental level, and there is no guarantee that there are better potential optimal parameters. For this reason, a small number of researchers have adopted a mathematical model of the relationship between process parameters and performance based on the results of physical experiments to optimize and determine the process parameters, forming the second method, but it is still rare. However, they are all optimization models of process parameters established for a specific component or a specific alloy. At the same time, physical experiments need to be carried out before optimization. Like the first method, it is costly and has a long cycle. With the maturity of casting simulation technology, some people gradually use virtual experiments to analyze the molding process of specific extrusion castings, observe the influence of process parameters, and then obtain optimized process parameters. Although the fourth method can directly calculate the process parameters, it is impossible to fully quantitatively describe the entire process because the mechanism of extrusion casting is not fully understood at present. Only a small number of empirical formulas are used to calculate parameters such as extrusion pressure, and the formulas are complex and the accuracy is not high. In summary, experiments and experience are still the main means of obtaining extrusion casting process parameters at present. Both physical experiments and virtual experiments (numerical simulations) are targeted at specific (material + geometry) castings, and are not universal. A universal method for designing (or selecting) extrusion casting process parameters has not yet been formed.

[0004] Process (parameter) design is a comprehensive design process that integrates experience, knowledge and mechanism. Successful design cases undoubtedly provide a wealth of reference experience and knowledge, while avoiding starting from scratch. Case-Based Reasoning (CBR) is an artificial intelligence reasoning paradigm that uses past experience to quickly solve new problems. Due to its good compatibility, case-based reasoning can be combined with other data mining methods (i.e., forming integrated case-based reasoning) to achieve the acquisition of mechanism knowledge and high-precision parameter prediction, greatly improving the efficiency of obtaining solutions to new problems. Case-based reasoning is now widely used in parameter design in the fields of welding, machining, electronic design, etc., but there are few reports on its application in the field of extrusion casting. To this end, the present invention establishes an extrusion casting process parameter design method based on case-based reasoning, which provides a new method for extrusion casting process parameter design, saves design time, and strengthens the use of existing data. Summary of the invention

[0005] The purpose of the present invention is to provide a case-based reasoning design method for squeeze casting process parameters to solve the existing technical problems.

[0006] Squeeze casting process parameters are the key to extrusion casting. Currently, they are mainly determined through experiments and experience, which is inefficient. Combining the trend of data-driven manufacturing, the present invention proposes a method for designing extrusion casting process parameters based on case reasoning. The collected process data are regarded as cases, and then a Markov blanket is constructed to simplify non-critical influencing factors of extrusion casting process parameters, remove redundant influencing factors and establish the relationship between case attributes. On this basis, information entropy is used to determine each other's weights, and then the process parameters of new material castings are obtained based on the principle of case reasoning. Through experiments and applications, the effectiveness and accuracy of this method have been proved, which provides a new method for the design of extrusion casting process parameters, can greatly save design time, and also enhance the utilization of existing data.

[0007] In order to achieve the above object, the technical solution adopted by the present invention is as follows:

[0008] A case-based reasoning design method for extrusion casting process parameters, the method comprising the following steps:

[0009] Step 1: According to the characteristics of squeeze casting process data, problem identification and knowledge representation are performed on the case library of squeeze casting process data, and its influencing factors are determined according to the process parameters to be designed;

[0010] Step 2: Based on the characteristics of the extrusion casting process data, the continuous data is discretized, the Markov blanket of the process parameters is searched, and the uncertainty network structure between the process influencing factors and the process parameters, namely the Markov blanket structure, is constructed to simplify the process influencing factors and remove redundant attributes;

[0011] Step 3: Calculate the weights of process influencing factors based on information entropy, and optimize the weights according to the Markov blanket structure between process parameters and process influencing factors;

[0012] Step 4: Based on the similarity between cases, the process parameter values ​​of the most similar case are used as recommended values;

[0013] Step 5: Store the predicted process parameter values ​​and their process influencing factors in the case library as a new case.

[0014] Furthermore, the specific process in step 1 is:

[0015] The process parameters of squeeze casting are affected by the material composition and geometry of the casting, as well as the squeeze casting method, namely direct squeeze casting and indirect squeeze casting. When direct squeeze casting is used and the shape is simple and the wall thickness is thick, a lower pouring temperature is used. The extrusion pressure will affect the solidification time of the squeeze casting, and then affect the holding time. The material composition, geometry and squeeze casting method of the casting together constitute the case scenario of squeeze casting process parameter design, and together with the process parameters, constitute the case characteristics. The collected squeeze casting process data set contains process parameter variables, material composition and casting performance parameter data, which are directly converted into basic The extrusion casting process data collected by the specification related to the process parameter case characteristics is used as a case set. The extrusion casting process data case is set to be described by three parts: case number, process parameters and process influencing factors. The process parameters include temperature parameters, pressure parameters, time parameters and speed parameters. The temperature parameters include pouring temperature and mold preheating temperature. The process influencing factors include material composition, casting method, shape complexity, wall thickness and equivalent thickness. The process parameters are continuous variables. The material setting is described by the material composition value. The casting method uses 0 and 1 to represent direct and indirect extrusion casting respectively. The casting shape is described according to the structural shape complexity level of the casting.

[0016] Let case = (S i , X k , Y j ) represents the squeeze casting process data case set, where S represents the case number S i ={S1, S2, ..., S m}(i=1,2,...,m), m represents the number of cases; X represents the process influencing factor X k = {X1, X2, ..., X t}, (k = 1, 2, ..., t), t represents the number of process influencing factors; Y represents the process parameter set Y j ={Y1, Y2, ..., Y n}(j=1, 2, ..., n), where n represents the number of process parameters.

[0017] Further, the specific process of Step 2 is as follows:

[0018] In the squeeze casting process data case = (S i , X k , Y j ), the relationships between all or some of the variables are unknown. By obtaining its Markov blanket, redundant factors can be eliminated to obtain the process parameter Y j 's λ j (λ j <n) direct influencing factors X η = {X1, X2,..., X λj}, η = (1, 2,..., λ j ), and their mutual relationships. Finally, the logic of the influencing relationship is determined to provide more accurate reasoning; the IPCMB (Iterative parents - children based search of Markov blanket) algorithm is used to obtain the mutual influence structure between process parameters and construct the Markov blanket of the squeeze casting process parameters:

[0019] (1) Input the discrete data X k of the process influencing factors and the process parameter Y j to be predicted;

[0020] (2) Use the G 2 conditional independence test to remove the process influencing factors that have no information transfer with the process parameter Y j to be predicted;

[0021] (3) According to the definition of the Markov blanket, search for the Markov blanket set of the process parameter Y j to be predicted; according to the V structure in the Markov blanket set of Y j , obtain the pointing relationship in the Markov blanket set of the process parameter Y j to be predicted. Suppose X1 is the parent - child node in the Markov blanket set of Y1, Y1 is the parent - child node of the Markov blanket of X2, X1 is not the parent - child node of X2, and X1 and X2 are independent under the union of all conditional sets of Y1 with X1 and X2. Then there exists a pointing structure of X1 → Y1 ← X2 in the Markov blanket of Y1;

[0022] (4) Obtain the Markov blanket of the process parameter Y j to be predicted and output the structure diagram.

[0023] Further, the process of using the G 2 conditional independence test to remove the process influencing factors that have no information transfer with the process parameter Y j to be predicted is as follows:

[0024] There are discrete data and continuous data in squeeze casting process data. Firstly, Canopy K-means algorithm is used to discretize the continuous data in squeeze casting.

[0025] For case = (S i , X k , Y j ), the assumption is that they are all discretized, let v k (i) Represents a discrete variable X k =x ik The number of times, v j (i) Represents Y j =y ij The number of times, if X k With Y j If the random variables X k Y j The expected value of is:

[0026]

[0027] Let {case-Y j -X k} is their condition set, denoted as C u , and C u ={c iu}, if X k With Y j In C u If the random variable X is independent under k Y j C u The expected value of is:

[0028]

[0029] Where: v u (i) C u =c iu The number of times, v ku (i) For case X k =x ik At the same time C u =c iu The number of times, v ju (i) For case Y j =y ij At the same time C u =c iu The number of times.

[0030] Constructing the chi-square statistic for two discrete variables

[0031]

[0032] The chi-square statistic has f degrees of freedom:

[0033] f=(r k -1)(r j -1) (4)

[0034] Where: r k For X k The possible value categories, r j Y j The possible value categories, such as X1 represents the casting method, X1 may be 0 or 1, representing direct extrusion casting and indirect extrusion casting respectively, then r k =2;

[0035] Through the chi-square statistic and the degree of freedom f, the p-value corresponding to the chi-square value is looked up in the table. If the p-value is less than the significance level, the two discrete variables are independent of each other.

[0036] Construct conditional independence G 2 Statistics to test X k With Y j In C u Under conditional independence, if G 2 If the value of the statistic is greater than the chi-square value under a given degree of freedom f, then the p-value is less than the significance level, indicating that X k With Y j Conditional independence, conditional independence test G 2 The statistics are:

[0037]

[0038] Substituting into formula (2), formula (5) becomes:

[0039]

[0040] According to formula (4), the degrees of freedom under the conditional independence test is:

[0041]

[0042] Where: r u C u The possible value category, |C u | for C u The number of variables in C u Dimension.

[0043] Furthermore, the specific process of discretizing continuous data in squeeze casting using the Canopy K-means algorithm is as follows:

[0044] (1) Use K-fold cross validation algorithm to cross-validate and determine the hyperparameters T1 and T2 of the Canopy algorithm;

[0045] (2) Use the Canopy algorithm to perform multiple clustering iterations, output a set {K} of the number of cluster centers, and multiple cluster center sets, and select the number of clusters K with the silhouette coefficient closest to 1 optimum and cluster centers;

[0046] (3) Use the K-means algorithm to transform K optimum The cluster center points are used as the initial center points for clustering, and the discrete data of each input variable are clustered to complete the data transformation.

[0047] Furthermore, the specific process of step 3 is:

[0048] The weight of the process influencing factors of the process parameters to be predicted will directly affect the accuracy of case reasoning. Since the case is a data set, the information entropy is used to weight the influencing factors of the specific process parameters based on the data and the Markov blanket of the process parameters to be predicted. The method is as follows:

[0049] For continuous value data, mining the process parameter Y to be predicted j Factors affecting process X k The relationship between the existing process parameters and their simplified influencing factors is combined into a new matrix V*:

[0050]

[0051] The V* is standardized by the extreme difference method. Normalize For process data with discrete values, let

[0052]

[0053] The information entropy of the kth influencing factor in the case is defined as:

[0054]

[0055] Prediction of process parameters Y based on Markov blanket j After simplifying the process influencing factors, we get Y j Lambda j Process influencing factors X η ={X1,X2,…X λj}, the weight needs to be based on Xη With λ j-1 If there is information transmission and adjustment between the process influencing factors, then the process influencing factor X η Process parameter Y j The influence weight is:

[0056]

[0057] Where: α is the dependence or independence relationship between process influencing factors, that is There are mainly three situations:

[0058] (a) Process influencing factors X η are independent of each other, α = 0;

[0059] (b) Process influencing factors X η The relationship between the influencing factors of information transmission is: η →{X1, X2, ..., X λj-1}, then α=-1;

[0060] (c) Process influencing factors X η The relationship between the influencing factors of information transmission is: η ←{X1, X2, ..., X λj-1}, then α=1.

[0061] Furthermore, the specific process of step 4 is to use the nearest neighbor strategy to calculate the similarity of the cases, the target case S l With source case S i The similarity calculation formula is:

[0062]

[0063] Then, the results of the source case with the highest similarity score to the target case are assigned to the target case to achieve the design of the squeeze casting process parameters of the new casting.

[0064] The present invention has the following beneficial effects due to the adoption of the above technical solution:

[0065] The present invention regards the existing extrusion casting process parameter data of the same material as a case, establishes an extrusion casting process parameter design method based on integrated case reasoning, directly uses the existing data to complete the design of extrusion casting process parameters, and provides an implementation method or strategy for extrusion casting to adapt to the trend of intelligent manufacturing and material digital manufacturing. It has good application value, high design efficiency, and can greatly save design time. The method strengthens the application of existing research results (process data) of extrusion casting. It can further promote the application of extrusion casting and reduce its application cost. By combining the use of Markov blanket and information entropy, the distribution of case reasoning feature attribute weights can be improved, the accuracy and rationality of weight distribution, and the robustness of reasoning results can be improved, ensuring that the method has high accuracy and design quality. BRIEF DESCRIPTION OF THE DRAWINGS

[0066] Figure 1 is a flow chart of the method of the present invention;

[0067] Figure 2 is a logical schematic diagram of the Markov blanket of the present invention;

[0068] Figure 3 It is a weight distribution diagram based on the Markov blanket structure and information entropy of the present invention;

[0069] Figure 4 It is a Markov blanket information entropy structure diagram of the pouring temperature of the present invention. DETAILED DESCRIPTION

[0070] 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.

[0071] like Figure 1-4 As shown, a case-based reasoning design method for extrusion casting process parameters comprises the following steps:

[0072] Step 1: Case representation. Based on the characteristics of squeeze casting process data, problem identification and knowledge representation are performed on the case library of squeeze casting process data, and the influencing factors are determined according to the process parameters to be designed.

[0073] According to the principle of squeeze casting and existing research, the process parameters of squeeze casting are mainly affected by the material composition and geometry of the casting, as well as the squeeze casting method (direct squeeze casting and indirect squeeze casting) (when direct squeeze casting is used and the shape is simple and the wall thickness is thick, a lower pouring temperature can be used). Some process parameters also affect each other (such as extrusion pressure affects the solidification time of the squeeze casting, and then affects the holding time), and the material composition is the main factor. Therefore, the material composition, geometry and squeeze casting method of the casting together constitute the case scenario of squeeze casting process parameter design, and together with the process parameters, constitute the case characteristics. The process data set of squeeze casting that is easy to collect generally contains process parameter variables (such as temperature), material composition and casting performance parameter data (such as strength), because they are all data. In order to perform case reasoning process parameter design based on squeeze casting process data or its extended set and promote data-driven squeeze casting, the squeeze casting process data collected based on the specifications related to the process parameter case characteristics is directly used as the case set, and the squeeze casting process data case is set to be described by three parts: case number, process parameters and process influencing factors. Process parameters include temperature parameters (such as pouring temperature, mold preheating temperature), pressure parameters (such as extrusion pressure), time parameters (such as holding time) and speed parameters (such as extrusion speed), etc.; process influencing factors include material composition, casting method, shape complexity, wall thickness, equivalent thickness, etc. Process parameters are a series of values, mainly continuous variables, so other case scenarios are also expressed by values: material conventions are described by material composition values, casting methods are represented by 0 and 1 for direct and indirect extrusion castings, respectively, and casting shapes are described according to the complexity level of the structural shape of the castings. For example, based on the existing axisymmetric rotating bodies (such as round bars, etc.), cup-shaped parts and other complex parts (such as wheels with side holes and reinforcing ribs, etc.), they are divided into three levels: simple, relatively complex and complex, represented by 0, 1, and 2 respectively (the following focuses on establishing the process parameter design method affected by material composition, and does not elaborate on geometric shapes).

[0074] Let case = (S i , X k , Y j ) represents the squeeze casting process data case set, where S represents the case number S i ={S1, S2, ..., S m}(i=1,2,...,m), m represents the number of cases; X represents the process influencing factor X k = {X1, X2, ..., X t}, (k = 1, 2, ..., t), t represents the number of process influencing factors; Y represents the process parameter set Y j ={Y1, Y2, ..., Y n}(j = 1, 2,..., n), where n represents the number of process parameters. An example of squeeze casting process data is shown in Table 1.

[0075] Table 1 Example of Squeeze Casting Process Data

[0076]

[0077] In Table 1, x and y can be continuous variables such as the pouring temperature of 720 °C or discrete variables such as 0 - direct squeeze casting, 1 - indirect squeeze casting, etc.

[0078] Table 2 Example (Instance) of Squeeze Casting Process Data

[0079]

[0080] Step 2: Feature reduction based on Markov blanket. Based on the characteristics of squeeze casting process data, discretize the continuous data, search for the Markov blanket of process parameters, and construct an uncertainty network structure between process influencing factors and process parameters, i.e., the Markov blanket structure, to complete the reduction of process influencing factors (removing redundant attributes).

[0081] For specific squeeze casting process parameters, initially, in the squeeze casting process data case = (S i , X k , Y j ), the relationships between various variables are all or partially unknown. By obtaining its Markov blanket, redundant factors can be eliminated, and λ j (λ j < n) direct influencing factors X j = {X1, X2,..., X η}, η = (1, 2,..., λ λj ) and their mutual relationships can be obtained, and finally, the logic of the influencing relationship can be determined to provide more accurate reasoning. By performing a G j conditional independence test between variables to determine whether there is an influencing relationship and independence between variables. 2

[0082] There are discrete data (such as casting method, shape complexity, etc.) and continuous data (such as pouring temperature, extrusion pressure, etc.) in the squeeze casting process data. Therefore, first, use the Canopy K - means algorithm to discretize the continuous data in squeeze casting, and the process is as follows:

[0083] (1) Use the K - fold cross - validation algorithm to cross - validate and determine the hyperparameters T1 and T2 of the Canopy algorithm.

[0084] ​(2) Use the Canopy algorithm to perform multiple clustering iterations and output a set {K} of the number of cluster centers and multiple cluster center sets. Select the number of clusters K with the silhouette coefficient closest to 1. optimum and cluster centers.

[0085] (3) Use the K-means algorithm to transform K optimum The cluster center points are used as the initial center points for clustering, and the discrete data of each input variable are clustered to complete the data transformation.

[0086] The discrete results are shown in Table 3.

[0087] Table 3 Discretization results of squeeze casting process data case (partial)

[0088]

[0089] Squeeze Casting Process Data 2 The specific method of conditional independence testing is as follows.

[0090] For case = (S i , X k , Y j ), the assumption is that they are all discretized, let v k (i) Represents a discrete variable X k =x ik The number of times, v j (i) Represents Y j =y ij The number of times, if X k With Y j If the random variables X k Y j The expected value of is:

[0091]

[0092] Let {case-Y j -X k} is their condition set, denoted as C u , and C u ={c iu}, if X k With Y j In C u If the random variable X is independent under k Y j C u The expected value of is:

[0093]

[0094] Where: vu (i) C u =c iu The number of times, v ku (i) For case X k =x ik At the same time C u =c iu The number of times, v ju (i) For case Y j =y ij At the same time C u =c iu The number of times.

[0095] Constructing the chi-square statistic for two discrete variables

[0096]

[0097] The chi-square statistic has f degrees of freedom:

[0098] f=(r k -1)(r j -1) (4)

[0099] Where: r k For X k The possible value categories, r j Y j The possible value category, such as X1 represents the casting method, X1 may be 0 or 1, representing direct casting and indirect casting respectively, then r k =2;

[0100] Through the chi-square statistic and the degree of freedom f, the p-value corresponding to the chi-square value is looked up in the table. If the p-value is less than the significance level, the two discrete variables are independent of each other.

[0101] Construct conditional independence G 2 Statistics to test X k With Y j In C u Under conditional independence, if G 2 If the value of the statistic is greater than the chi-square value under a given degree of freedom f, then the p-value is less than the significance level, indicating that X k With Y j Conditional independence, conditional independence test G 2 The statistics are:

[0102]

[0103] Substituting into formula (2), formula (5) becomes:

[0104]

[0105] According to formula (4), the degrees of freedom under the conditional independence test is:

[0106]

[0107] Where: r u C u The possible value category, |C u | for C u The number of variables in C u Dimension.

[0108] Based on the above principles, the IPCMB algorithm is used to obtain the interaction structure between various process parameters and construct a Markov blanket of squeeze casting process parameters. The whole process is as follows:

[0109] (1) Input discrete data X of each process influencing factor k , the process parameter Y to be predicted j ;

[0110] (2) Using G 2 Conditional independence test removes the process parameter Y to be predicted j Process influencing factors without information transmission;

[0111] (3) According to the definition of Markov blanket, search for the process parameter Y to be predicted j Markov blanket set; according to Y j The V structure in the Markov blanket set is used to obtain the process parameter Y to be predicted. j Pointing relationships in the Markov blanket set; an example is as follows: suppose X1 is the parent-child node in the Markov blanket set of Y1, Y1 is the parent-child node in the Markov blanket of X2, X1 is not the parent-child node of X2, and X1 and X2 are independent of each other under the union of all condition sets of Y1 and X1 and X2, then there exists a pointing structure of X1→Y1←X2 in the Markov blanket of Y1.

[0112] (4) Obtain the process parameter Y to be predicted j The Markov blanket outputs a structure diagram such as Figure 3 shown.

[0113] Step 3: Weight allocation: Calculate the weights of various process influencing factors based on information entropy, and optimize the weights according to the Markov blanket structure between process parameters and process influencing factors.

[0114] The weight of the process influencing factors of the process parameters to be predicted will directly affect the accuracy of case reasoning. Since the case is a data set, the information entropy is used to weight the influencing factors of the specific process parameters based on the data and the Markov blanket of the process parameters to be predicted. The method is as follows:

[0115] For continuous value data, mining the process parameter Y to be predicted j Factors affecting process X k The relationship between the existing process parameters and their simplified influencing factors is combined into a new matrix V*:

[0116]

[0117] The V* is standardized by the extreme difference method. Normalize For process data with discrete values, let

[0118]

[0119] The information entropy of the kth influencing factor in the case is defined as:

[0120]

[0121] Prediction of process parameters Y based on Markov blanket j After simplifying the process influencing factors, we get Y j Lambda j Process influencing factors X η ={X1,X2,…X λj}, the weight needs to be based on X η With other lambda j-1 If there is information transmission and adjustment between the process influencing factors, then the process influencing factor X η Process parameter Y j The influence weight is:

[0122]

[0123] Where: α is the dependence or independence relationship between process influencing factors, that is There are mainly three situations:

[0124] (a) Process influencing factors X η are independent of each other, α = 0;

[0125] (b) Process influencing factors X η The relationship between the influencing factors of information transmission is: η →{X1, X2, ..., X λj-1}, then α=-1;

[0126] (c) Process influencing factors X η The relationship between the influencing factors of information transmission is: η ←{X1, X2, ..., X λj-1}, then α=1.

[0127] The obtained Markov blanket information entropy structure is as follows Figure 4 shown.

[0128] Step 4: Case retrieval to obtain the designed process parameter values. Based on the similarity between cases, the process parameter values ​​of the most similar case are used as recommended values.

[0129] The nearest neighbor strategy is used to calculate the similarity of the cases. The target case S l With source case S i The similarity calculation formula is:

[0130]

[0131] Then, the results of the source case with the highest similarity score to the target case are assigned to the target case to achieve the design of the squeeze casting process parameters of the new casting.

[0132] Step 5: Case reuse: The predicted process parameter values ​​and their process influencing factors are stored in the case library as a new case.

[0133] 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 case-based reasoning design method for squeeze casting process parameters, characterized in that: The method comprises the following steps: Step 1: According to the characteristics of squeeze casting process data, problem identification and knowledge representation are performed on the case library of squeeze casting process data, and its influencing factors are determined according to the process parameters to be designed; Step 2: Based on the characteristics of the extrusion casting process data, the continuous data is discretized, the Markov blanket of the process parameters is searched, and the uncertainty network structure between the process influencing factors and the process parameters, namely the Markov blanket structure, is constructed to simplify the process influencing factors and remove redundant attributes; Step 3: Calculate the weights of process influencing factors based on information entropy, and optimize the weights according to the Markov blanket structure between process parameters and process influencing factors; Step 4: Based on the similarity between cases, the process parameter values ​​of the most similar case are used as recommended values; Step 5: Store the predicted process parameter values ​​and their process influencing factors in the case library as a new case.

2. The case-based reasoning design method for squeeze casting process parameters according to claim 1, characterized in that: The specific process in step 1 is: The process parameters of squeeze casting are affected by the material composition and geometry of the casting, as well as the squeeze casting method, namely direct squeeze casting and indirect squeeze casting. When direct squeeze casting is used and the shape is simple and the wall thickness is thick, a lower pouring temperature is used. The extrusion pressure will affect the solidification time of the squeeze casting, and then affect the holding time. The material composition, geometry and squeeze casting method of the casting together constitute the case scenario of squeeze casting process parameter design, and together with the process parameters, constitute the case characteristics. The collected squeeze casting process data set contains process parameter variables, material composition and casting performance parameter data, which are directly converted into basic The extrusion casting process data collected by the specification related to the process parameter case characteristics is used as a case set. The extrusion casting process data case is set to be described by three parts: case number, process parameters and process influencing factors. The process parameters include temperature parameters, pressure parameters, time parameters and speed parameters. The temperature parameters include pouring temperature and mold preheating temperature. The process influencing factors include material composition, casting method, shape complexity, wall thickness and equivalent thickness. The process parameters are continuous variables. The material setting is described by the material composition value. The casting method uses 0 and 1 to represent direct and indirect extrusion casting respectively. The casting shape is described according to the structural shape complexity level of the casting. Let case = (S i , X k , Y j ) represents the squeeze casting process data case set, where S represents the case number S i ={S1, S2, ..., S m }(i=1,2,...,m), m represents the number of cases; X represents the process influencing factor X k = {X1, X2, ..., X t }, (k = 1, 2, ..., t), t represents the number of process influencing factors; Y represents the process parameter set Y j ={Y1, Y2, ..., Y n }(j=1, 2, ..., n), where n represents the number of process parameters.

3. The case-based reasoning design method for squeeze casting process parameters according to claim 1, characterized in that: The specific process of step 2 is: The relationship between various variables in the squeeze casting process data case=(S i , X k , Y j ) is all or partially unknown. By obtaining its Markov blanket, redundant factors can be eliminated to obtain the process parameter Y j 's λ j (λ j <n) direct influencing factors X η ={X1, X2,..., X λj}, η=(1, 2,..., λ j ) and their mutual relationships, and finally determine the logic of the influencing relationship to provide more accurate reasoning; use the IPCMB algorithm to obtain the mutual influence structure between process parameters and construct the Markov blanket of the squeeze casting process parameters: (1) Input discrete data X of process influencing factors k , the process parameter Y to be predicted j ; (2) Using G 2 Conditional independence test removes the process parameter Y to be predicted j Process influencing factors without information transmission; (3) According to the definition of Markov blanket, search for the process parameter Y to be predicted j Markov blanket set; according to Y j The V structure in the Markov blanket set is used to obtain the process parameter Y to be predicted. j Pointing relationship in Markov blanket set, suppose X1 is the parent-child node in the Markov blanket set of Y1, Y1 is the parent-child node in the Markov blanket of X2, X1 is not the parent-child node of X2, and X1 and X2 are independent of each other under the union of all condition sets of Y1 and X1 and X2, then there exists a pointing structure of X1→Y1←X2 in the Markov blanket of Y1; (4) Obtain the process parameter Y to be predicted j The Markov blanket of ,output structure diagram.

4. The case-based reasoning design method for squeeze casting process parameters according to claim 3 is characterized in that: Using G 2 Conditional independence test removes the process parameter Y to be predicted j The process of process influencing factors without information transfer is: There are discrete data and continuous data in squeeze casting process data. Firstly, Canopy K-means algorithm is used to discretize the continuous data in squeeze casting. For case = (S i , X k , Y j ), the assumption is that they are all discretized, let v k (i) Represents a discrete variable X k =x ik The number of times, v j (i) Represents Y j =y ij The number of times, if X k With Y j If the random variables X k Y j The expected value of is: Let {case-Y j -X k } is their condition set, denoted as C u , and C u ={c iu }, if X k With Y j In C u If the random variable X is independent under k Y j C u The expected value of is: Where: v u (i) C u =c iu The number of times, v ku (i) For case X k =x ik At the same time C u =c iu The number of times, v ju (i) For case Y j =y ij At the same time C u =c iu The number of times; Constructing the chi-square statistic for two discrete variables The chi-square statistic has f degrees of freedom: f＝(r k -1)(r j -1) (4) Where: r k For X k The possible value categories, r j Y j The possible value categories, such as X1 represents the casting method, X1 may be 0 or 1, representing direct extrusion casting and indirect extrusion casting respectively, then r k =2; Through the chi-square statistic and the degree of freedom f, the p-value corresponding to the chi-square value is looked up in the table. If the p-value is less than the significance level, the two discrete variables are independent of each other. Construct conditional independence G 2 Statistics to test X k With Y j In C u Under conditional independence, if G 2 If the value of the statistic is greater than the chi-square value under a given degree of freedom f, then the p-value is less than the significance level, indicating that X k With Y j Conditional independence, conditional independence test G 2 The statistics are: Substituting into formula (2), formula (5) becomes: According to formula (4), the degrees of freedom under the conditional independence test is: Where: r u C u The possible value category, |C u | for C u The number of variables in C u Dimension.

5. The case-based reasoning design method for squeeze casting process parameters according to claim 4, characterized in that: The specific process of discretizing continuous data in squeeze casting using the Canopy K-means algorithm is as follows: (1) Use K-fold cross validation algorithm to cross-validate and determine the hyperparameters T1 and T2 of the Canopy algorithm; (2) Use the Canopy algorithm to perform multiple clustering iterations, output a set of cluster center numbers {K}, and multiple cluster center sets, and select the cluster number K with the silhouette coefficient closest to 1 optimum and cluster centers; (3) Use the K-means algorithm to transform K optimum The cluster center points are used as the initial center points for clustering, and the discrete data of each input variable are clustered to complete the data transformation.

6. The case-based reasoning design method for squeeze casting process parameters according to claim 1, characterized in that: The specific process of step 3 is: The weight of the process influencing factors of the process parameters to be predicted will directly affect the accuracy of case reasoning. Since the case is a data set, the information entropy is used to weight the influencing factors of the specific process parameters based on the data and the Markov blanket of the process parameters to be predicted. The method is as follows: For continuous value data, mining the process parameter Y to be predicted j Factors affecting process X k The relationship between the existing process parameters and their simplified influencing factors is combined into a new matrix V*: The V* is standardized by the extreme difference method. Normalize For process data with discrete values, let The information entropy of the kth influencing factor in the case is defined as: Prediction of process parameters Y based on Markov blanket j After simplifying the process influencing factors, we get Y j Lambda j Process influencing factors The weights need to be based on X η With λ j-1 If there is information transmission and adjustment between the process influencing factors, then the process influencing factor X η Process parameter Y j The influence weight is: Where: α is the dependence or independence relationship between process influencing factors, that is There are mainly three situations: (a) Process influencing factors X η are independent of each other, α = 0; (b) Process influencing factors X η The relationship between the influencing factors of information transmission is: η →{X1, X2, ..., X λj-1 }, then α=-1; (c) Process influencing factors X η The relationship between the influencing factors of information transmission is: η ←{X1, X2, ..., X λj-1 }, then α=1.

7. The case-based reasoning design method for squeeze casting process parameters according to claim 1, characterized in that: The specific process of step 4 is to use the nearest neighbor strategy to calculate the similarity of the cases. l With source case S i The similarity calculation formula is: Then, the results of the source case with the highest similarity score to the target case are assigned to the target case to achieve the design of the squeeze casting process parameters of the new casting.

Images