Structural entropy-based tin-based material formulation prediction method and application thereof

Abstract

The application relates to the technical field of formula matching, and particularly relates to a tin-based material formula prediction method based on structural entropy and application thereof. Feature extraction and entity alignment are performed on raw material, process and performance data, different types of entities are mapped into nodes, a complete entity combination in a historical formula is represented by a directed formula hyperedge, and a tin-based material formula hypergraph is constructed; structural contribution of different correlations is evaluated by comparing structural entropy differences before and after correlation change, and the correlation strength between entity nodes and formula hyperedges is adjusted according to the structural entropy contribution, so that the correlation which plays an important role in formula formation and performance change is enhanced, and weak correlation and noise correlation are inhibited, thereby the candidate formula search space is compressed, and the formula prediction efficiency is improved. The application aims to solve the problem of how to narrow the candidate formula search space to realize reliable tin-based material formula prediction.

Description

Technical Field

[0001] This application relates to the field of formulation technology, and in particular to a method for predicting the formulation of tin-based materials based on structural entropy and its application. Background Technology

[0002] The formulation of tin-based materials is typically determined by a combination of raw materials and processes. Different raw materials can have synergistic or restrictive effects, and the process can further influence material properties by altering the crystal phase, interface, and microstructure. As the types, content ranges, and process conditions of raw materials continue to increase, the number of available formulations is growing rapidly. However, relying on manual experience to adjust the raw materials and their proportions one by one makes it difficult to quickly obtain new formulations that meet comprehensive performance requirements within a large candidate pool.

[0003] In related technical solutions, to enhance the extraction capability of complex features and component relationships, known methods for predicting tin-based material formulations based on deep learning and probabilistic graphical models utilize deep neural networks, Bayesian networks, or graph attention networks to extract high-order features, component associations, and component-performance dependencies from material data, and thereby predict candidate components, processes, or performance. For example, Chinese Patent Publication No. CN118247262A discloses a solder paste printing quality prediction system and method based on an XGBoost-LSTM hybrid model, which uses a long short-term memory network to extract temporal features of solder paste printing data and combines XGBoost to predict printing quality; Chinese Patent Publication No. CN202511479055.X discloses a method for inferring the composition and performance of tin-based materials based on Bayesian networks.

[0004] While the methods described above can extract high-order features from material data and learn the dependencies between components, they mostly use co-occurrence frequency, attention weight, or probability dependency strength to measure the importance of associations. However, they lack association evaluation methods that can quantitatively distinguish between key associations that have a stable effect on formulation composition and performance formation and noisy associations formed by data sparsity, experimental errors, and accidental co-occurrence. This makes it difficult for the methods to effectively identify noisy associations and constrain the candidate space of new formulations when generating new formulations in reverse based on target performance, resulting in an excessively large candidate formulation search space.

[0005] In view of this, this application proposes a new method for predicting tin-based material formulations, which aims to uniformly characterize the complex relationship between raw materials, processes and performance in historical formulation data, and on this basis, narrow the candidate formulation search space to achieve reliable prediction of tin-based material formulations. Summary of the Invention

[0006] The main objective of this application is to provide a method for predicting tin-based material formulations based on structural entropy, aiming to solve the problem of how to narrow the candidate formulation search space in order to achieve reliable tin-based material formulation prediction.

[0007] To achieve the above objectives, this application provides a method for predicting tin-based material formulations based on structural entropy, the method comprising:

[0008] S10: Extract features and align entities from the raw material, process and performance data in the collected historical formula data, and map them to their respective raw material nodes, process nodes and performance nodes to form a set of entity nodes.

[0009] S20, construct a directed formula hyperedge from the historical formula data, and construct a tin-based material hypergraph based on the formula hyperedge and the set of entity nodes;

[0010] S30, determine the structural entropy difference between the tin-based material hypergraph and the tin-based material hypergraph after correlation change, wherein the tin-based material hypergraph after correlation change is obtained by reducing or removing the correlation value between the entity nodes and the formulation hyperedges in the tin-based material hypergraph;

[0011] S40, based on the structural entropy difference, determine the correlation strength between the entity nodes and the formulation superedges in the tin-based material hypergraph, enhance the formulation superedge characterization in the tin-based material hypergraph based on the correlation strength, and determine the entity enhancement characterization based on the formulation superedge characterization;

[0012] S50, determine a target performance representation that meets the preset performance conditions based on the entity augmentation representation, determine a candidate node set based on the association score between the target performance representation and the entity augmentation representation, connect the candidate node set with the target performance node set associated with the preset performance conditions to form a candidate recipe hyperedge, and calculate the connection score between the candidate recipe hyperedge and the preset performance conditions.

[0013] S60, predict the tin-based material formulation based on the target formulation superedge with a connection score greater than a preset score threshold.

[0014] Optionally, in step S10, feature extraction and entity alignment are performed on the raw material, process, and performance data in the collected historical formula data, including:

[0015] The continuous attributes of candidate entities in the historical formula data are normalized, and the semantic features, category features and numerical features of the candidate entities are extracted and fused with the normalized continuous attributes as the initial features of the candidate entities.

[0016] Entity alignment scores are calculated based on the consistency of the initial features of the candidate entities in terms of semantics, attributes, type, and units.

[0017] Candidate entities with an entity alignment score not lower than a preset entity alignment threshold and of the same type are aligned to the same standard entity node. The standard entity node includes raw material entity nodes, process entity nodes and performance entity nodes.

[0018] Using the credibility of each data record in the historical formula data as a weight, the features of candidate entities in the same standard entity node are weighted and fused to obtain the fused features of the standard entity node.

[0019] The various fused features are arranged in node order to form a node feature matrix, thereby completing entity alignment.

[0020] Optionally, step S20 specifically includes:

[0021] Raw material nodes in historical formula data process nodes as input node , performance nodes As an output node Each historical recipe is constructed as a corresponding directed recipe superedge. :

[0022]

[0023] In the formula, Indicates the formula type. For the weight of the superedge;

[0024] Based on each of the aforementioned formulations, the super-edge The set of superedges of the formula and entity node set Constructing a supermap of tin-based materials :

[0025]

[0026] In the formula, These represent the input correlation matrix, which describes the degree of involvement of raw materials and processes in the formulation, and the output correlation matrix, which describes the performance level obtained by a formulation.

[0027] in:

[0028]

[0029]

[0030] In the formula, , , This represents the normalized content of raw material node j in the i-th formulation. This represents the normalized process parameter value. This represents the normalized performance measurement.

[0031] Optionally, step S30 specifically includes:

[0032] The preset weights of the recipe hyperedges are assigned to the entity nodes contained in the recipe hyperedges to obtain the degree of each entity node, and the structural proportion is calculated based on the degree of each entity node.

[0033] Construct an encoding tree based on the participation degree of the superedges of each entity node, and calculate the multi-level structural entropy based on the hierarchical partitioning of the entity nodes using the encoding tree. :

[0034]

[0035] In the formula, Indicates the number of levels in the coding tree. Represents the root node of the coding tree. This represents a non-root node in the coding tree. express The sum of the structural participation degrees of the corresponding entity nodes, This represents the sum of the recipe superedge weights between this entity node and other entity nodes. express The parent node, This represents the sum of the structural participation degrees of all entity nodes;

[0036] Select the target coding tree with the minimum multilayer structure entropy from the candidate coding trees, and use the multilayer structure entropy corresponding to the target coding tree as the structure entropy of the tin-based material hypergraph.

[0037] The association value between the entity node and the recipe hyperedge in the tin-based material hypergraph is adjusted to ρ times the original value to obtain the tin-based material hypergraph after association change, and the structural entropy of the tin-based material hypergraph after association change is calculated, where ρ∈[0,1], and ρ=0 indicates that the association is removed;

[0038] The difference between the structural entropy of the tin-based material hypergraph and the structural entropy of the tin-based material hypergraph after correlation change is taken as the structural entropy difference.

[0039] Optionally, step S40 specifically includes:

[0040] After normalizing the structural entropy difference, the correlation strength is calculated. :

[0041]

[0042] In the formula, , , and These represent the obtained input correlation value and the output correlation value, respectively. Indicates the correlation adjustment factor. This represents the normalized structural entropy difference.

[0043] The correlation strength After normalization, the entity features in the formulation hyperedge are weighted and converged to obtain the representation of the i-th formulation hyperedge. :

[0044]

[0045] In the formula, d represents the feature of the entity node; This is the normalized correlation strength value; For entity nodes;

[0046] Based on the super-edge characterization of the formula Determine the augmented representation of entities :

[0047]

[0048] In the formula, Indicates the presence of entity nodes The set of superedges of the recipe, To prevent positive numbers with a denominator of zero.

[0049] Optionally, step S50 specifically includes:

[0050] The types, target measurements, and allowable deviations of each performance element in the preset performance conditions are normalized, fused with the entity augmentation representation of the corresponding target performance node, and then weighted and aggregated according to the importance of each performance element to obtain the target performance representation. :

[0051]

[0052] In the formula, Represents entity nodes The importance of , where d represents the dimension of representation;

[0053] The similarity between the target performance representation and the entity augmentation representation is used as the association score. , will correlate scores Above the candidate raw material screening threshold The raw material node was determined as a candidate raw material node. Those with a correlation score higher than the candidate process screening threshold will be considered. The process node was determined as a candidate process node. :

[0054]

[0055]

[0056] The candidate raw material node With the candidate process node Form a set of candidate input nodes The candidate input node set and the target performance node set are connected to form the candidate recipe hyperedge. ;

[0057] The connection score between the candidate recipe superedge and the preset performance condition is calculated using a scoring function. .

[0058] Optionally, step S60 specifically includes:

[0059] Determine the entity nodes in the target recipe hyperedge The feasible region corresponding to the conditions of satisfying preset constraints, wherein the preset constraints include raw material content constraints and process parameter constraints;

[0060] Determine the raw material ratios and process parameters of the target formulation superedge within the feasible region;

[0061] Based on the raw material ratio and the process parameters, a tin-based material formulation that meets the preset performance conditions is predicted.

[0062] In addition, to achieve the above objectives, this application also provides an application of the tin-based material formulation prediction method based on structural entropy as described in any of the preceding claims in tin-based material formulation prediction or formulation design.

[0063] In addition, to achieve the above objectives, this application also provides a computer system comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the computer program, when executed by the processor, implements the steps of the method for predicting tin-based material formulations based on structural entropy as described in any of the preceding claims.

[0064] In addition, to achieve the above objectives, this application also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of the method for predicting tin-based material formulations based on structural entropy as described in any of the preceding claims.

[0065] This application has at least the following beneficial effects:

[0066] 1. Feature extraction and entity alignment are performed on raw material, process and performance data. Different types of entities are mapped to nodes, and a directed formulation hyperedge is used to represent a complete combination of entities in a historical formulation. This constructs a tin-based material formulation hypergraph to improve the unified representation and reuse capability of historical formulation structural information.

[0067] 2. By comparing the structural entropy differences before and after the association changes, the structural contribution of different associations is evaluated. Based on the structural entropy contribution, the association strength between entity nodes and formulation hyperedges is adjusted to enhance associations that play an important role in formulation formation and performance changes, suppress weak associations and noisy associations, thereby compressing the candidate formulation search space and improving the formulation prediction efficiency. Attached Figure Description

[0068] Figure 1 This is a flowchart illustrating the method for predicting tin-based material formulations based on structural entropy, as described in an embodiment of this application.

[0069] Figure 2 This is a schematic diagram of the construction of a tin-based material hypergraph in an embodiment of this application;

[0070] Figure 3 This is a schematic diagram illustrating the key association identification based on structural entropy changes in an embodiment of this application;

[0071] Figure 4 This is a schematic diagram illustrating the structural entropy-guided association reweighting and representation enhancement involved in an embodiment of this application;

[0072] Figure 5 This is a schematic diagram of the hardware operating environment of the computer system involved in the embodiments of this application.

[0073] The realization of the purpose, functional features and advantages of this application will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0074] To better understand the above technical solutions, exemplary embodiments of this disclosure will be described in more detail below with reference to the accompanying drawings. While exemplary embodiments of this disclosure are shown in the drawings, it should be understood that this disclosure can be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of this disclosure to those skilled in the art.

[0075] First Embodiment

[0076] Reference Figure 1 This embodiment provides a method for predicting tin-based material formulations based on structural entropy, the method comprising the following steps:

[0077] S10: Extract features and align entities from the raw material, process and performance data in the collected historical formula data, and map them to their respective raw material nodes, process nodes and performance nodes to form a set of entity nodes.

[0078] In this step, historical formula data includes, but is not limited to, the names and contents of raw materials, process parameters and performance test data corresponding to existing tin-based material formulas, such as raw material names and contents, melting temperature, holding time, cooling method, welding parameters and other process data, as well as performance measurements such as melting point, wettability, hardness, tensile strength and reliability.

[0079] For example, to ensure that the superedge of the subsequently constructed formulation can fully correspond to the raw materials, processes, and performance information in the historical tin-based material formulation, the i-th historical formulation is represented as:

[0080]

[0081] In the formula, m represents the number of historical recipes. This represents the set of raw materials and their quantities for the i-th formula. Represents a set of processes. This represents a set of performance measurements.

[0082] Entity alignment refers to the process of grouping records from different data sources that differ in naming, attributes, type, or unit for the same real-world entity (e.g., the same raw material, the same process parameter, or the same performance indicator) into a unified entity. In this embodiment, entity alignment is used to consolidate raw material, process, and performance information that was originally scattered across various records into a standard set of entities that can be directly used for subsequent hypergraph construction.

[0083] After completing entity alignment, feature extraction is performed on each entity so that each entity node carries numerical, semantic, and categorical information that reflects its own attributes and participation in the formulation.

[0084] It should be noted that the three types of entity nodes—raw materials, process, and performance—are mutually exclusive; that is, there is no instance in the hypergraph where an entity node belongs to two types of node sets simultaneously.

[0085] S20, construct a directed formula hyperedge from the historical formula data, and construct a tin-based material hypergraph based on the formula hyperedge and the set of entity nodes;

[0086] In this step, to preserve the overall correspondence between various raw materials, process conditions, and performance results in a tin-based material formulation, step S20 constructs each historical formulation as a directed formulation hyperedge: using the raw material nodes and process nodes in the formulation as the input node set, and the performance nodes in the formulation as the output node set. In this embodiment, the direction of the formulation hyperedge is obtained by considering the actual physical process of raw material composition and process control affecting performance formation. Finally, a tin-based material hypergraph is constructed on the formulation hyperedge.

[0087] S30, determine the structural entropy difference between the tin-based material hypergraph and the tin-based material hypergraph after correlation change, wherein the tin-based material hypergraph after correlation change is obtained by reducing or removing the correlation value between the entity nodes and the formulation hyperedges in the tin-based material hypergraph;

[0088] In this embodiment, structural entropy is a metric that measures the overall information organization of the tin-based material hypergraph. It is used to reflect the distribution and combination relationship of entity nodes in the formulation hyperedge. The difference in structural entropy before and after the association change is used to evaluate the role of the corresponding association in maintaining the structural relationship between raw material combination, process conditions and performance results.

[0089] In this step, unlike conventional association importance evaluation methods such as attention weight, this embodiment weakens or removes the association value between a certain entity node and a certain recipe superedge, compares the difference between the structural entropy before and after the operation, and determines the difference as the structural entropy contribution of the association. In this way, associations with higher structural entropy contributions are identified as key associations.

[0090] It should be noted that in this embodiment, structural entropy is used to measure the overall information organization of a hypergraph. The change in structural entropy after weakening or removing a certain association can measure the role that association plays in maintaining the overall structure. When an association supported by the true formula rules is removed, the overall structural organization will change, while when a weak association formed by data sparsity, experimental error or accidental co-occurrence is removed, the overall structure is almost unaffected. This distinguishes between key associations that have made a high contribution to the historical formula structure and weak associations formed by data sparsity, experimental error or accidental co-occurrence.

[0091] S40, based on the structural entropy difference, determine the correlation strength between the entity nodes and the formulation superedges in the tin-based material hypergraph, enhance the formulation superedges in the tin-based material hypergraph based on the correlation strength to obtain the formulation superedge characterization, and determine the entity enhancement characterization based on the formulation superedge characterization;

[0092] In this step, after obtaining the structural entropy difference, the association strength between entity nodes and recipe hyperedges is determined based on the structural entropy difference of each association. This strengthens associations with higher structural entropy contributions and suppresses associations with lower structural entropy contributions. Subsequently, the recipe hyperedges are weighted and aggregated based on the obtained association strength to obtain the recipe hyperedge representation. Furthermore, the representations of each recipe hyperedge in which the entity participates are back aggregated to the entity node according to the normalized association strength to obtain the entity enhancement representation.

[0093] It should be noted that the formulation hyperedge representation is used to compress the overall compositional information into a unified vector of the formulation, while the entity enhancement representation is used to incorporate the compositional contextual information into the representation of the entity itself. For example, for trace amounts of added raw materials and special processes, their frequency of occurrence is often low, but they may play an important role in target properties such as melting point, wettability, and mechanical strength. These features are enhanced in this embodiment due to their strong correlation strength.

[0094] S50, determine a target performance representation that meets the preset performance conditions based on the entity augmentation representation, determine a candidate node set based on the association score between the target performance representation and the entity augmentation representation, connect the candidate node set with the target performance node set associated with the preset performance conditions to form a candidate recipe hyperedge, and calculate the connection score between the candidate recipe hyperedge and the preset performance conditions.

[0095] S60, predict the tin-based material formulation based on the target formulation superedge with a connection score greater than a preset score threshold.

[0096] Steps S50 and S60 constitute the formulation prediction stage. The preset performance conditions consist of the type, target measured value, and allowable deviation of each target performance, and after normalization, are fused with the entity enhancement representation of the corresponding performance node to form the target performance representation. For example, a combination of multiple target requirements such as "liquidothermal temperature 140℃±3℃, wetting angle 28°±3°" is used. The correlation score represents the degree of matching between each raw material node, each process node, and the target performance representation, and is a predicted value. Candidate raw material nodes and candidate process nodes with high matching degrees are selected based on the correlation score; then, the set of candidate input nodes is connected to the set of target performance nodes to form candidate formulation hyperedges; finally, the connection score between each candidate formulation hyperedge and the target performance condition is calculated based on pre-trained parameters.

[0097] Target formulation superedges whose connection scores reach a preset score threshold are identified as structurally feasible target formulation superedges, and tin-based material formulations that can be practically prepared are predicted.

[0098] In the technical solution provided in this embodiment, feature extraction and entity alignment are performed on raw material, process, and performance data. Different types of entities are mapped as nodes, and a directed formulation hyperedge is used to represent a complete combination of entities in a historical formulation, thus constructing a tin-based material formulation hypergraph. The structural contribution of different associations is evaluated by comparing the structural entropy differences before and after the association changes. The association strength between entity nodes and formulation hyperedges is adjusted according to the structural entropy contribution to enhance the associations that play an important role in formulation formation and performance changes, suppress weak associations and noisy associations, thereby compressing the candidate formulation search space and improving the formulation prediction efficiency.

[0099] Second Embodiment

[0100] Based on any embodiment, in this embodiment, step S10 involves feature extraction and entity alignment of the raw material, process, and performance data from the collected historical formula data, including:

[0101] S11, normalize the continuous attributes of candidate entities in the historical formula data, extract the semantic features, category features and numerical features of candidate entities and fuse the normalized continuous attributes as the initial features of candidate entities.

[0102] In this embodiment, continuous attributes refer to attributes whose values ​​can change continuously within a certain numerical range and can be directly used for numerical calculations. These include raw material content, melting temperature, holding time, cooling rate, and measured values ​​such as melting point, wetting angle, and tensile strength. For example, an Ag content of 0.80%, a melting temperature of 205°C, and a holding time of 30 min are all continuous attributes.

[0103] Specifically, the raw material content, process parameters, and performance measurements from historical formulas are used for entity alignment and formula association strength calculation. To eliminate differences in dimensions and numerical ranges between different continuous attributes and to avoid the model focusing excessively on attributes with large value ranges, continuous attributes are normalized. The k-th attribute of the j-th entity is normalized as follows:

[0104]

[0105] In the formula, , This indicates the number of candidate entities before entity alignment. , Indicates the total number of continuous numerical attributes. This represents the original value of the k-th attribute of the j-th entity. and These represent the minimum and maximum values ​​of the k-th attribute in the historical recipe data, respectively. This indicates a positive number used to prevent the denominator from being zero.

[0106] To enable comparison of raw material, process, and performance records from different data sources based on attribute, category information, and name description, this invention extracts the semantic features, name description features, and category features of the j-th candidate entity based on the numerical normalization results, and merges them into the initial features of the candidate entity:

[0107]

[0108] In the formula, Numerical characteristics representing entities, The semantic features that represent entities || represents the category features of an entity, and || represents vector concatenation. Indicates entity type, and These represent the feature mapping matrix and bias vector for the corresponding entity type, respectively.

[0109] S12, Calculate the entity alignment score based on the consistency of the initial features of the candidate entities in terms of semantics, attributes, type, and unit:

[0110] In historical formulations from different sources, the same raw material may appear as a chemical formula, Chinese name, English name, or abbreviation, and the same performance indicator may also use different units. To avoid the same raw material forming multiple nodes in the hypergraph, which would lead to repeated splitting of the formulation structure and interfere with subsequent key association identification, this invention comprehensively considers semantics, attributes, type, and unit consistency, and aligns candidate entities by calculating entity alignment scores. and entity The entity alignment score is calculated as follows:

[0111]

[0112] In the formula, and , and Represents a collection of entity attributes. For indicator functions, This indicates that two entities have the same or mutually convertible units of measurement, and satisfy the following conditions: .

[0113] S13, Align candidate entities with entity alignment scores not lower than the preset entity alignment threshold and of the same type into the same standard entity node;

[0114] When two entity types are the same and satisfy When aligning them to the same standard entity, This represents the entity alignment threshold.

[0115] After entity alignment, the resulting Each standard entity node is denoted as .

[0116] S14, using the credibility of each data record in the historical formula data as weight, the features of candidate entities in the same standard entity node are weighted and fused to obtain the fused features of the standard entity node;

[0117] To preserve the contribution of original records from different sources to the features of standard entities based on data credibility, standard entity nodes are... The fusion feature is denoted as:

[0118]

[0119] In the formula, , Indicates alignment to standard entities The original set of entities, This indicates the reliability of the corresponding data record.

[0120] S15: The various fused features are arranged in node order to form a node feature matrix to complete entity alignment.

[0121] To replace the raw material, process, and performance records in historical formulations with corresponding standard entities, and to provide node and feature inputs for the subsequent construction of the tin-based material hypergraph, the fused features of the standard entity nodes with unified names, consistent units, and identical feature dimensions obtained after weighted fusion are arranged in node order to form a node feature matrix:

[0122]

[0123] In the formula, This indicates the number of standard entities after entity alignment. This represents the dimension of entity features.

[0124] Third Embodiment

[0125] Based on any embodiment, this embodiment provides a detailed description of the construction of a tin-based material hypergraph. Specifically, the tin-based material formulation is formed by the combined action of multiple raw materials and multiple processes to achieve specific properties. However, edges in a conventional graph can only represent the relationship between two entities. To avoid the loss of raw material and process combination information after splitting the complete formulation into multiple pairwise relationships, raw materials, processes, and properties are represented as different types of entity nodes, and the set of tin-based material entity nodes is denoted as:

[0126]

[0127] In the formula, Represents the set of raw material nodes. Represents the set of process nodes. Let each set represent a set of performance nodes, and these three sets are mutually exclusive.

[0128] To achieve the reverse determination of raw material composition and process conditions based on target performance, this invention uses raw materials and processes as inputs for performance formation, and melting point, wettability, and mechanical strength as outputs, to represent the actual process by which raw material composition and process treatment jointly affect material properties. Specifically, the raw material and process nodes included in the i-th historical formula are used as input nodes, and performance nodes are used as output nodes. The set of input nodes is denoted as... ( The set of output nodes is denoted as :

[0129]

[0130] To preserve the overall correspondence between various raw materials, process conditions, and performance results in the i-th historical formula, this historical formula is represented as a directed formula hyperedge:

[0131]

[0132] In the formula, Indicates the formula type. The superedge weights are determined based on data integrity, experimental repeatability, and the reliability of the data source.

[0133] Even if the same raw materials and processes are used in different historical formulations, the content of raw materials, process parameters, and final performance measurements may vary. To preserve the differences in the content, parameters, and corresponding performance measurements of the same raw material or process in different formulations, and to enable subsequent formulation predictions to distinguish the specific composition and performance results of different formulations, this invention further constructs an input correlation matrix and an output correlation matrix. The input correlation matrix describes the degree of participation of raw materials and processes in the formulation, while the output correlation matrix describes the actual performance level achieved by a formulation.

[0134]

[0135]

[0136] In the formula, , , This represents the normalized content of raw material node j in the i-th formulation. This represents the normalized process parameter value. This represents the normalized performance measurement, thereby preserving the complete entity combination of each historical formula in the correlation matrix, as well as the interaction between raw materials and processes that influence performance.

[0137] Based on the aforementioned node set, formulation hyperedges, and input-output relationships, a tin-based material hypergraph is constructed. :

[0138]

[0139] In the formula, Represents a set of entity nodes. Denotes the set of superedges of the recipe. and Let represent the input and output incidence matrices of the hyperedge, respectively.

[0140] The results were:

[0141] S21, the raw material nodes in the historical formula data process nodes as input node , performance nodes As an output node Each historical recipe is constructed as a corresponding directed recipe superedge. :

[0142]

[0143] In the formula, Indicates the formula type. For the weight of the superedge;

[0144] S22, based on each of the aforementioned formulation superedges The set of superedges of the formula and entity node set Constructing a supermap of tin-based materials :

[0145]

[0146] In the formula, These represent the input correlation matrix, which describes the degree of involvement of raw materials and processes in the formulation, and the output correlation matrix, which describes the performance level obtained by a formulation.

[0147] in:

[0148]

[0149]

[0150] In the formula, , , This represents the normalized content of raw material node j in the i-th formulation. This represents the normalized process parameter value. This represents the normalized performance measurement.

[0151] Fourth embodiment

[0152] Based on any embodiment, this embodiment provides a detailed description of the structural entropy perturbation contribution measurement and key correlation identification. Specifically, step S30 includes:

[0153] S31, the preset weight of the recipe superedge is assigned to the entity nodes contained in the recipe superedge to obtain the degree of each entity node, and the structure proportion is calculated based on the degree of each entity node.

[0154] Specifically, to distinguish between stable associations that contribute to the composition and performance of historical formulations and weak associations formed by data sparsity, experimental errors, and accidental co-occurrence, this invention calculates structural entropy based on the structural participation of each entity in the formulation hyperedge and evaluates the association contribution by the difference in structural entropy before and after the association change. The weight of each formulation hyperedge is assigned to the entity nodes it contains, and the entity nodes... The degree is recorded as :

[0155]

[0156] In the formula, , Indicates the presence of entity nodes Recipe superedge set and They represent the first The number of input and output nodes in a recipe superedge.

[0157] To convert the degrees of entities with different raw materials, processes, and performance into a unified and comparable scale, entity nodes are... The structural proportion is denoted as:

[0158]

[0159] In the formula, ,and .

[0160] S32, Construct an encoding tree based on the participation degree of the superedges of each entity node, and calculate the multi-level structural entropy based on the hierarchical partitioning of the entity nodes using the encoding tree. :

[0161]

[0162] In the formula, Indicates the number of levels in the coding tree. Represents the root node of the coding tree. This represents a non-root node in the coding tree. express The sum of the structural participation degrees of the corresponding entity nodes, This represents the sum of the recipe superedge weights between this entity node and other entity nodes. express The parent node, This represents the sum of the structural participation degrees of all entity nodes;

[0163] Specifically, to measure the impact of weakening or removing a certain raw material, process, or performance correlation on the formulation regularity of tin-based materials, the one-dimensional structural entropy of the original tin-based material hypergraph is first calculated based on the structural proportion of each entity node, serving as a reference value before the correlation change:

[0164]

[0165] In the formula, ,when When the value of the corresponding term in the formula is zero, it represents an entity node. It did not participate in any valid historical formula, and its corresponding structural entropy is zero.

[0166] To distinguish local combination patterns within different formulation groups, and to identify the combination relationships between specific raw materials and processes in tin-based material formulation prediction while preserving the overall patterns between similar formulation groups and performance results, this embodiment constructs a coding tree based on the situation where entities jointly participate in formulation hyperedges. And the corresponding multi-layer structure entropy is denoted as .

[0167] S33, Select the target coding tree with the minimum multilayer structure entropy from the candidate coding trees, and use the multilayer structure entropy corresponding to the target coding tree as the structure entropy of the tin-based material hypergraph;

[0168] Specifically, to form corresponding formulation groups for entities with similar raw material combinations, process conditions, and performance results, the coding tree with the minimum multi-layer structure entropy is selected from the candidate coding trees and denoted as:

[0169]

[0170] In the formula, Indicates that the number of layers does not exceed The set of candidate coding trees.

[0171] S34, adjust the association value between the entity node and the recipe hyperedge in the tin-based material hypergraph to ρ times the original value to obtain the tin-based material hypergraph after association change, and calculate the structural entropy of the tin-based material hypergraph after association change, where ρ∈[0,1], and ρ=0 indicates that the association is removed;

[0172] S35, the difference between the structural entropy of the tin-based material hypergraph and the structural entropy of the tin-based material hypergraph after the correlation change is taken as the structural entropy difference.

[0173] Finally, to quantify the role of different raw materials, processes, and performance correlations in maintaining the correspondence between raw material combinations, process conditions, and performance results, and to provide a basis for subsequent candidate raw material screening and process condition determination, entity nodes were... With the formula super edge The structural entropy contribution between them is denoted as:

[0174]

[0175] In the formula, .

[0176] Furthermore, after obtaining the structural entropy difference, step S40 specifically includes:

[0177] S41, after normalizing the structural entropy difference, calculate the correlation strength. :

[0178]

[0179] In the formula, , , and These represent the obtained input correlation value and the output correlation value, respectively. Indicates the correlation adjustment factor. This represents the normalized structural entropy difference.

[0180] Specifically, to eliminate the influence of different numerical ranges of structural entropy changes and to uniformly compare raw material correlation, process correlation, and performance correlation, the contribution of normalized structural entropy is denoted as... :

[0181]

[0182] In the formula, It contributes the largest positive structural entropy among all associations. To prevent positive numbers with a denominator of zero, .

[0183] when At that time, the formula will be extended beyond the edge. Identified as a key association This indicates the threshold for determining key associations.

[0184] Considering that in the prediction of tin-based material formulations, some trace raw materials or special processes, although occurring infrequently, may play a significant role in target properties such as melting point, wettability, and mechanical strength, this study aims to preserve such high-contribution correlations and reduce the interference of weak correlations on new formulation predictions during candidate raw material screening and process condition determination. Based on the structural entropy contribution, the correlation strength between raw materials, processes, performance entities, and formulation superedges is adjusted. The adjusted correlation strength is denoted as... .

[0185] It is easy to see from the formula that... At that time, the correlation strength is enhanced; At that time, the correlation strength is suppressed; At that time, the original association strength remains unchanged.

[0186] S42, the correlation strength After normalization, the entity features in the formulation hyperedge are weighted and converged to obtain the representation of the i-th formulation hyperedge. :

[0187]

[0188] In the formula, d represents the feature of the entity node; This is the normalized correlation strength value; For entity nodes;

[0189] Specifically, after adjusting the association strength, the relative contributions of each entity within the same recipe's hyperedge will change. To ensure the comparability of the internal composition of each historical recipe, the entity nodes... The normalized correlation strength in the i-th formulation hyperedge is denoted as:

[0190]

[0191] In the formula, , and Let these represent the set of input nodes and the set of output nodes for the i-th superedge. This prevents positive numbers with a denominator of zero. .

[0192] To fully characterize the overall synergistic relationship between the target performance of tin-based materials and the combined effects of various raw materials and process conditions, and to enable subsequent candidate formulation predictions to utilize verified raw material content, process conditions, and performance results from historical formulations, a weighted aggregation of entity features in the formulation hyperedges is performed based on the normalized enhanced correlation strength to obtain the enhanced characterization of the i-th historical formulation. .

[0193] S43, Super-edge characterization according to the stated formula Determine the augmented representation of entities :

[0194]

[0195] In the formula, Indicates the presence of entity nodes The set of superedges of the recipe, To prevent positive numbers with a denominator of zero.

[0196] Specifically, after obtaining the historical formulation enhancement characterization, in order to ensure that the same raw material, process, or performance entity can reflect its role in different historical formulation combinations and carry information on the raw material combinations, process conditions, and performance results that co-occur with it, the entity nodes are classified according to the normalized enhancement correlation strength. The historical recipe representations involved are back-converged to the entity, resulting in entity nodes. Entity augmentation representation .

[0197] Fifth embodiment

[0198] Based on any embodiment, this embodiment provides a detailed description of the formulation prediction of multi-constrained tin-based materials based on hypergraph link analysis. Specifically, step S50 includes:

[0199] S51, the type, target measurement, and allowable deviation of each performance in the preset performance conditions are normalized, fused with the entity augmentation representation of the corresponding target performance node, and weighted and aggregated according to the importance of each performance to obtain the target performance representation. :

[0200]

[0201] In the formula, Represents entity nodes The importance of , where d represents the dimension of representation;

[0202] Specifically, for the predicted tin-based material formulation, one or more target properties such as melting point, wettability, mechanical strength, corrosion resistance, and service reliability can be set according to application requirements. To ensure clear performance constraints for candidate raw materials and process screening, the type, target measurement, and allowable deviation of each target property are combined to form target performance conditions, and the corresponding set of target performance nodes is denoted as... To standardize the representation of multiple target performance requirements, the measured values ​​and allowable deviations of each target are first normalized and then fused with the enhanced representations of the corresponding performance nodes. Finally, the representations are weighted and aggregated according to the importance of each target performance to obtain the target performance characterization. .

[0203] S52, the similarity between the target performance representation and the entity enhancement representation is used as the association score. , will correlate scores Above the candidate raw material screening threshold The raw material node was determined as a candidate raw material node. Those with a correlation score higher than the candidate process screening threshold will be considered. The process node was determined as a candidate process node. :

[0204]

[0205]

[0206] In the formula, Indicates the screening threshold for candidate raw materials. This indicates the threshold for screening candidate processes.

[0207] Specifically, in order to screen entities that can support the target performance from all raw material nodes and process nodes, and to reduce raw material and process combinations that lack historical formulation correlation, based on the entity enhancement characterization obtained in step 2, the raw material or process entity nodes are calculated. and The correlation score between them is used to screen raw materials or process nodes:

[0208]

[0209] In the formula, , This represents the function for calculating similarity. .

[0210] The raw materials in tin-based material formulations include both the main tin-based material components and trace additives that significantly contribute to the target performance. The processes also involve special techniques such as segmented heat treatment, specific heat treatments, and welding parameter control. While these trace raw materials or special processes may only appear in a few historical high-performance formulations, they have a significant impact on target properties such as melting point, wettability, and mechanical strength. To avoid them being overlooked in the candidate screening process due to their low frequency of occurrence, this invention filters candidate raw material nodes and candidate process nodes separately based on hypergraph connectivity scores, and denotes the candidate raw material set as follows: The candidate process set is denoted as .

[0211] S53, the candidate raw material node With the candidate process node Form a set of candidate input nodes The candidate input node set and the target performance node set are connected to form the candidate recipe hyperedge. ;

[0212] Specifically, according to the input and output node structure of the recipe hyperedge, the candidate input node set is... With the target performance node set Connect to form the a-th candidate recipe superedge:

[0213]

[0214] S54, calculate the connection score between the candidate recipe hyperedge and the preset performance condition using a scoring function. .

[0215] Specifically, to eliminate candidate solutions that are only related to the target performance by a single entity but lack historical formulation structure support for the overall raw material-process combination, candidate formulation hyperedge characterization is formed based on the enhanced characterization of each entity in the candidate formulation hyperedge. And calculate the connectivity score between the candidate formulation hyperedge and the target performance:

[0216]

[0217] In the formula, , Represents the scoring function. With candidate formulation super-edge screening threshold Compare and satisfy The candidate formulation superedge is used for subsequent raw material ratio and process parameter determination.

[0218] In some alternative implementations, the scoring function may be a cosine similarity function, inner product function, Euclidean distance transformation function, or a multilayer perceptron-based scoring function, used to calculate a connectivity score based on the degree of matching between the candidate recipe hyperedge representation and the target performance representation.

[0219] Furthermore, after determining the connection score Subsequently, the actual performance of tin-based materials depends not only on the use of a particular raw material, but also on the specific content of each raw material and the synergistic relationship between melting temperature, holding time, cooling method, and heat treatment conditions. Therefore, after determining the types of raw materials and process types that may collectively contribute to the target performance, it is necessary to further determine the raw material ratios and process parameters based on the candidate formulation's super-edge, transforming the structurally compatible candidate combinations for the target performance into practically achievable tin-based material formulations. Specifically, this includes:

[0220] S61, determine the entity nodes in the target recipe hyperedge. The feasible region corresponding to the conditions of satisfying preset constraints, wherein the preset constraints include raw material content constraints and process parameter constraints;

[0221] S62, determine the raw material ratio and process parameters of the target formulation superedge within the feasible domain;

[0222] S63, Based on the raw material ratio and the process parameters, predict the tin-based material formulation that meets the preset performance conditions.

[0223] Specifically, entity nodes The content in the a-th candidate formulation is denoted as And set raw material content constraints based on historical formulation range, material system and preparation requirements:

[0224]

[0225] In the formula, and They represent The lower and upper limits of the permissible content.

[0226] Given the specific types and contents of raw materials, and considering that the same combination of raw materials may result in different microstructures and properties under different process conditions, to ensure that the process parameters match the candidate raw material ratios, the value of the r-th process parameter in the a-th candidate formulation is denoted as... Process parameter constraints are set based on historical preparation records, raw material melting characteristics, and equipment operability.

[0227]

[0228] In the formula, , For the number of process parameters, and These are the lower and upper limits of the r-th process parameter, respectively.

[0229] During the determination of raw material ratios and process parameters, candidate formulations must simultaneously meet requirements regarding target performance, raw material content, raw material compatibility, process implementation, material costs, restricted element usage, and environmental protection. To ensure that these constraints are effective together during the same formulation determination process, they are grouped into the feasible domain of the a-th candidate formulation. Since the same candidate formulation structure may correspond to multiple sets of raw material ratios and process parameters, in order to determine the formulation parameters that best meet the target performance from the feasible region, the performance prediction results are calculated based on the superedge characterization of the candidate formulation, raw material content, and process parameters, and the raw material ratios and process parameters of the a-th candidate formulation are determined accordingly:

[0230]

[0231]

[0232] In the formula, Let be the vector of raw material content for the a-th candidate formulation. For process parameter vectors, For the corresponding performance prediction results, To predict performance based on the raw material-process-performance enhancement correlation determined in historical formulations, This represents the overall degree to which the candidate formulation meets the target performance conditions. and These are the determined raw material ratios and process parameters, respectively.

[0233] Fifth embodiment

[0234] Based on any embodiment, in this embodiment, in order to enable the candidate formulation connection score to accurately distinguish between effective formulation structures and ineffective combinations, and to ensure that the performance prediction results are consistent with the actual experimental measurements, the hypergraph connection judgment and performance prediction process are jointly trained using historical tin-based material formulations.

[0235] During the training phase, complete formulation hyperedges that actually exist in the historical data are used as positive samples, while non-historical formulation hyperedges formed by replacing raw material nodes, process nodes, or performance nodes are used as negative samples. To enable formulation structure judgment to identify whether effective connections are formed between raw materials, processes, and performance, the formulation connection loss is calculated based on the true connection labels and predicted connection scores of the training samples.

[0236]

[0237] In the formula, The number of superedge samples used for training; The connection label for the h-th training sample is 1 for positive samples and 0 for negative samples. This represents the corresponding predicted connection score.

[0238] The fact that the formulation's super-edge can form an effective bond does not necessarily mean that its specific raw material content and process parameters will yield accurate performance results. To ensure that the predicted melting point, wettability, mechanical strength, and other properties are close to historical experimental values, performance prediction losses are calculated based on the importance of each property:

[0239]

[0240] In the formula, For the number of performance indicators, For historical formulas beyond the edge The corresponding k-th performance prediction value, This is the actual measured value for this performance. Let k be the importance of the k-th performance attribute.

[0241] Simply reducing the connection judgment error and performance prediction error may still result in excessive raw material content, unexecutable process parameters, or failure to meet cost and environmental requirements. To ensure the prediction process simultaneously complies with all formulation constraints in step 3.2, the degree of violation of the b-th constraint on the a-th candidate formulation is denoted as... And calculate the multi-constraint loss:

[0242]

[0243] In the formula, To constrain the quantity, For constrained indexes, This represents the importance of the corresponding constraint.

[0244] To ensure that candidate formulation structure judgment, performance prediction, and actual preparation constraints are adjusted collaboratively during the same training process, the above losses are weighted and combined to obtain the total loss for tin-based material formulation prediction:

[0245]

[0246] In the formula, The weights for the formulation connection loss, performance prediction loss, and multi-constraint loss are respectively, satisfying... .

[0247] During training, the parameters used in candidate recipe connection calculation, performance prediction, and recipe parameter determination are updated based on the total loss until the total loss meets the preset convergence condition. After training is complete, the target performance node set is input. Then, by sequentially executing steps 3.1 and 3.2, a candidate tin-based material formulation with clearly defined raw material composition, raw material content, process conditions, and performance prediction results is obtained:

[0248]

[0249] In the formula, This refers to the collection of raw materials used in tin-based material formulations. To determine the content of each raw material, This is a set of process types used. For the corresponding process parameters, These are the performance indicators of the formula.

[0250] To ultimately obtain a tin-based material formulation that meets the preset requirements, the candidate formulations are screened, retaining only those with a connection score reaching the specified threshold and simultaneously satisfying the requirements for target performance, raw material content, raw material compatibility, process implementation, cost, and environmental protection.

[0251] .

[0252] Furthermore, as an implementation scheme, this application also relates to the application of a tin-based material formulation prediction method based on structural entropy as described in any of the preceding claims in the manufacture of tin-based materials.

[0253] Specifically, in this embodiment, 500 tin-based material formulation data provided by a tin-based material manufacturing company are used. Each formulation data includes process data such as raw material name and content, melting temperature, holding time, cooling rate, and protective atmosphere, as well as performance measurements such as liquidus temperature, wetting angle, tensile strength, corrosion resistance, and service reliability. 400 of these historical formulation data are used as a training set for correlation strength adjustment, formulation characterization enhancement, candidate formulation connection judgment, and performance prediction parameter training. The remaining 100 historical formulation data are used as a test set to verify the candidate formulation screening and performance prediction results. Some historical tin-based material formulation data are shown in Table 1.

[0254] surface Tin-based material formulations and performance data

[0255]

[0256] (1) Feature extraction and solid alignment of tin-based materials

[0257] First, missing values, outliers, duplicate records, and inconsistent units of measurement in 500 historical formula data entries were processed. Raw material content was uniformly converted to mass percentage, melting temperature to degrees Celsius, holding time to minutes, and tensile strength to megapascals. In different data sources, the same raw material might be recorded as "Sn," "tin," or "metallic tin," and the same performance indicator might be recorded as "liquidite temperature," "melting temperature," or "melting point." Alignment scores between candidate entities were calculated, and the entity alignment threshold was set to 0.90. For example, the alignment score between "Sn" and "tin" is 0.98, and the alignment score between "liquidite temperature" and "melting point" is 0.93, therefore they are aligned to the same standard entity. The alignment score between "rapid cooling" and "furnace cooling" is 0.31, therefore they are retained as two different process entities.

[0258] After entity alignment, the 61 candidate entities in the original data were unified into 24 standard entities, including 10 raw material entities, 9 process entities, and 5 performance entities.

[0259] Then, the continuous attributes are normalized. Taking the Ag content in the first historical recipe as an example, the minimum and maximum values ​​of Ag content in the training set are 0.20% and 1.00%, respectively. The Ag content in this recipe is 0.80%, so the normalization result is:

[0260]

[0261] The melting temperature range for the training set is 180–240℃. The melting temperature for the first formula is 205℃, and the corresponding normalized result is:

[0262]

[0263] We further extract the numerical features, name semantic features, and entity category features of each entity, and map the different features to a 64-dimensional unified feature space.

[0264] In this embodiment, the numerical feature dimension is 16, the semantic feature dimension is 32, and the category feature dimension is 16. After fusion, a node feature matrix is ​​obtained. .

[0265] (2) Construction of tin-based material hypergraph

[0266] Raw materials such as Sn, Bi, Ag, Cu, and Ni are represented as raw material nodes; processes such as melting temperature, holding time, cooling rate, and protective atmosphere are represented as process nodes; and liquidus temperature, wetting angle, tensile strength, corrosion resistance, and service reliability are represented as performance nodes.

[0267] For the first historical recipe, its set of input nodes is:

[0268]

[0269] The output node set is:

[0270]

[0271] Represent the first historical recipe as a directed recipe superedge:

[0272]

[0273] Based on the data integrity, experimental repeatability, and data source reliability of the first formulation, the weight of the superedge of this formulation is set to... Each of the 500 historical recipes is converted into a recipe hyperedge in the same way, with the total number of historical recipes being m=500. Input the association matrix. Record raw material content and process parameters, and output the correlation matrix. Record the corresponding performance measurements. For example, the normalized correlation values ​​for Ag content, melting temperature, and tensile strength in formula 1 are as follows:

[0274]

[0275] This yields a tin-based material hypergraph containing 24 entity nodes, 500 recipe hyperedges, input-output association matrices, and node semantic hierarchy, as shown below. Figure 2 As shown.

[0276] (3) Identification of key correlations in tin-based materials based on structural entropy changes

[0277] Reference Figure 3 and Figure 4 The structural participation of each entity node is obtained by assigning the weights of the hyperedges to the entities they contain. After calculating all 24 entity nodes, the total volume of the tin-based material hypergraph is 612.4. The structural participation of some entity nodes is shown in Table 2.

[0278] Table 2. Structural Participation of Some Entity Nodes

[0279]

[0280] Calculate the structural proportion of each entity node and the one-dimensional structural entropy to obtain... The maximum number of levels in the coding tree is set to K=3, and candidate coding trees are constructed based on the situation where entities jointly participate in the recipe hyperedge. After merging and adjusting the coding tree nodes, the coding tree with the minimum entropy of the multi-level structure is obtained. Its multi-layer structure entropy is The coding tree aggregates Sn-Bi low-temperature solder formulations, Sn-Ag-Cu high-reliability formulations, and formulations containing trace amounts of Ni, Sb, and other raw materials into formulation groups with different raw material combinations, process conditions, and performance results.

[0281] Reduce the association value between entity nodes and recipe hyperedges to their original value. This embodiment sets the value to times the original value. That is, remove the corresponding association and preserve the coding tree. Without changing the correlation, calculate the difference in multilayer structure entropy before and after the correlation change, and then normalize it. The calculation results of some correlations in the first historical formula are shown in Table 3.

[0282] Table 3. Structural entropy contributions of some related historical recipes in item 1

[0283]

[0284] Set the key association determination threshold to Therefore, the correlation between entities such as Bi, Ag, Cu, Ni, rapid cooling, and wetting angle in the first formulation and the formulation superedges is identified as the key correlation.

[0285] Set the correlation adjustment factor to Adjust the association strength between entities and recipe hyperedges. Taking the Ag association in recipe 1 as an example, its original input association value is 0.75, and its normalized structural entropy contribution is 0.81. The adjusted association strength is:

[0286]

[0287] For a weak association with a structural entropy contribution of 0.24, assuming its original association value is 0.45, the adjusted association strength is:

[0288]

[0289] The above treatment enhances the correlations of Ag, Ni, and other structural elements that occur less frequently but contribute significantly to the target performance, while suppressing weak correlations formed by accidental co-occurrence. The correlation strength within the first formulation superedge is normalized, and the partially normalized correlation strengths are shown in Table 4.

[0290] Table 4. Normalized Correlation Strength of Historical Recipes in Item 1

[0291]

[0292] We perform weighted aggregation on the entity features in the first recipe hyperedge to obtain the recipe hyperedge representation:

[0293]

[0294] The historical recipe hyperedge representations involving entity nodes are back-converged to their corresponding entities, resulting in enhanced representations for each entity node. This yields 500 historical recipe hyperedge representations and 24 enhanced entity node representations, providing input for candidate recipe generation and performance prediction.

[0295] (4) Formulation prediction of multi-constrained tin-based materials based on hypergraph link analysis

[0296] 400 existing historical formulation hyperedges were used as positive samples. For each positive sample, one raw material node or one process node was replaced, resulting in 800 negative samples, for a total of 1200 training formulation hyperedge samples. The connection score of each training sample was calculated. Taking the first historical formulation hyperedge as an example, its connection score between the raw material and process combination and the performance result was 0.91. After replacing the Ag raw material node with a Zn raw material node to form a negative sample, the calculated connection score was 0.24.

[0297] The training batch size was set to 64, the initial learning rate to 0.001, and the maximum number of training epochs to 300. When the total loss decreased by less than [a certain value] over 20 consecutive epochs... Training is stopped when the time is right. The weights in the total loss are set to... The calculation results of some losses during the training process are shown in Table 5.

[0298] Table 5. Loss Variation During Joint Training

[0299]

[0300] After training, save the connection score calculation parameters, performance prediction parameters, and relevant parameters used in determining the formulation parameters.

[0301] Taking the prediction of a Sn-Bi-based lead-free solder formulation for low-temperature electronic packaging as an example, the target formulation needs to simultaneously meet the requirements of low liquidus temperature, small wetting angle, high tensile strength, low corrosion rate, and high service reliability. The target performance conditions are shown in Table 6.

[0302] Table 6 Target performance conditions of the formulation to be predicted

[0303]

[0304] The performance types, target measurements, and permissible deviations in Table 6 are used to form the target performance conditions, and the corresponding set of target performance nodes is denoted as... The target measured values ​​and allowable deviations are normalized and fused with the corresponding performance entity enhancement characterizations. Then, they are weighted and aggregated according to importance to obtain the target performance characterization:

[0305]

[0306] The connection scores between each raw material node and process node and the target performance characterization were calculated, and some results are shown in Table 7.

[0307] Table 7. Connection Score of Raw Material and Process Nodes

[0308]

[0309] By setting both the candidate raw material screening threshold and the candidate process screening threshold to 0.70, we obtain:

[0310]

[0311]

[0312] Thirty-six candidate input node combinations were formed from the candidate raw material set and the candidate process set, and the connection score between the candidate input node combination and the target performance was calculated. Some of the calculation results are shown in Table 8.

[0313] Table 8. Connection scores of some candidate formulation structures

[0314]

[0315] Set the candidate formulation structure screening threshold to The first four candidate input node combinations in Table 8 are retained, and each is combined with the target performance node set. Connect to form candidate formulation superedges to .

[0316] Based on the retained candidate formulation superedges, constraints are set for raw material content, raw material compatibility, process parameters, cost, and environmental protection. Some of the constraints in this embodiment are shown in Table 9.

[0317] Table 9. Constraints on Candidate Formulations

[0318]

[0319] The above constraints constitute the feasible region for each candidate formulation. Within the feasible region, the raw material content and process parameters are adjusted, and the performance prediction results are calculated based on the superedge characterization of the candidate formulation, the raw material content vector, and the process parameter vector. After parameter iteration, the candidate formulations shown in Table 10 are obtained.

[0320] Table 10. Formulation and Performance Prediction Results of Candidate Tin-Based Materials

[0321]

[0322] In Table 10, the formula The target performance conditions were not met simultaneously, therefore the formula was deleted, but the formula was retained. , and This is the final output formula. Among them, the formula that best meets the overall target performance is... Therefore, the preferred tin-based material formulation output in this embodiment is as follows: Sn content 57.42%, Bi content 41.35%, Ag content 0.72%, Cu content 0.43%, Ni content 0.08%, melting temperature 205℃, holding time 28 min, and cooling at a rate of 24℃ / min under nitrogen protection. The predicted liquidus temperature of this formulation is 140.8℃, wetting angle is 27.6°, tensile strength is 76.3 MPa, and corrosion rate is 0.062 mg / (cm²·h).

[0323] As one implementation scheme, Figure 5 This is a schematic diagram of the hardware operating environment of the computer system involved in the embodiments of this application.

[0324] like Figure 5As shown, the computer system may include: a processor 1001, such as a CPU; a memory 1005; a user interface 1003; a network interface 1004; and a communication bus 1002. The communication bus 1002 is used to enable communication between these components. The user interface 1003 may include a display screen or an input unit such as a keyboard; optionally, the user interface 1003 may also include a standard wired interface or a wireless interface. The network interface 1004 may optionally include a standard wired interface or a wireless interface (such as a Wi-Fi interface). The memory 1005 may be high-speed RAM or non-volatile memory, such as a disk drive. Optionally, the memory 1005 may also be a storage device independent of the aforementioned processor 1001.

[0325] Those skilled in the art will understand that Figure 5 The computer system architecture shown does not constitute a limitation on the computer system and may include more or fewer components than shown, or combine certain components, or have different component arrangements.

[0326] like Figure 5 As shown, the memory 1005, as a storage medium, may include an operating system, a network communication module, a user interface module, and computer programs. The operating system is a program that manages and controls the hardware and software resources of the computer system, as well as the operation of the computer programs and other software or programs.

[0327] exist Figure 5 In the computer system shown, the user interface 1003 is mainly used to connect to the terminal and communicate with the terminal; the network interface 1004 is mainly used to communicate with the backend server; and the processor 1001 can be used to call the computer program stored in the memory 1005.

[0328] In this embodiment, the computer system includes: a memory 1005, a processor 1001, and a computer program stored in the memory and executable on the processor, wherein:

[0329] When the processor 1001 calls the computer program stored in the memory 1005, it implements each step of the method for predicting tin-based material formulations based on structural entropy as described in the above embodiment.

[0330] Furthermore, those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program includes program instructions and can be stored in a storage medium, which is a computer-readable storage medium. The program instructions are executed by at least one processor in a computer system to implement the process steps of the embodiments of the above methods.

[0331] Therefore, this application also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the various steps of the tin-based material formulation prediction method based on structural entropy as described in the above embodiments.

[0332] The computer-readable storage medium can be any computer-readable storage medium capable of storing program code, such as a USB flash drive, portable hard drive, read-only memory (ROM), magnetic disk, or optical disk.

[0333] It should be noted that, since the storage medium provided in the embodiments of this application is the storage medium used to implement the methods of the embodiments of this application, those skilled in the art can understand the specific structure and variations of the storage medium based on the methods described in the embodiments of this application, and therefore will not be repeated here. All storage media used in the methods of the embodiments of this application fall within the scope of protection of this application.

[0334] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0335] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0336] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0337] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0338] Although preferred embodiments of this application have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments as well as all changes and modifications falling within the scope of this application.

[0339] Obviously, those skilled in the art can make various modifications and variations to this application without departing from the spirit and scope of this application. Therefore, if such modifications and variations fall within the scope of the claims of this application and their equivalents, this application also intends to include such modifications and variations.

Claims

1. A method for predicting the formulation of tin-based materials based on structural entropy, characterized in that, The method includes the following steps: S10: Extract features and align entities from the raw material, process and performance data in the collected historical formula data, and map them to their respective raw material nodes, process nodes and performance nodes to form a set of entity nodes. S20, construct a directed formula hyperedge from the historical formula data, and construct a tin-based material hypergraph based on the formula hyperedge and the set of entity nodes; S30, determine the structural entropy difference between the tin-based material hypergraph and the tin-based material hypergraph after correlation change, wherein the tin-based material hypergraph after correlation change is obtained by reducing or removing the correlation value between the entity nodes and the formulation hyperedges in the tin-based material hypergraph; S40, based on the structural entropy difference, determine the correlation strength between the entity nodes and the formulation superedges in the tin-based material hypergraph, enhance the formulation superedge characterization in the tin-based material hypergraph based on the correlation strength, and determine the entity enhancement characterization based on the formulation superedge characterization; S50, determine a target performance representation that meets the preset performance conditions based on the entity augmentation representation, determine a candidate node set based on the association score between the target performance representation and the entity augmentation representation, connect the candidate node set with the target performance node set associated with the preset performance conditions to form a candidate recipe hyperedge, and calculate the connection score between the candidate recipe hyperedge and the preset performance conditions. S60, predict the tin-based material formulation based on the target formulation superedge with a connection score greater than a preset score threshold.

2. The method for predicting tin-based material formulations based on structural entropy as described in claim 1, characterized in that, In step S10, feature extraction and entity alignment are performed on the raw material, process, and performance data in the collected historical formula data, including: The continuous attributes of candidate entities in the historical formula data are normalized, and the semantic features, category features and numerical features of the candidate entities are extracted and fused with the normalized continuous attributes as the initial features of the candidate entities. Entity alignment scores are calculated based on the consistency of the initial features of the candidate entities in terms of semantics, attributes, type, and units. Candidate entities with an entity alignment score not lower than a preset entity alignment threshold and of the same type are aligned to the same standard entity node. The standard entity node includes raw material entity nodes, process entity nodes and performance entity nodes. Using the credibility of each data record in the historical formula data as a weight, the features of candidate entities in the same standard entity node are weighted and fused to obtain the fused features of the standard entity node. The various fused features are arranged in node order to form a node feature matrix, thereby completing entity alignment.

3. The method for predicting tin-based material formulations based on structural entropy as described in claim 1, characterized in that, Step S20 specifically includes: Raw material nodes in historical formula data process nodes as input node , performance nodes As an output node Each historical recipe is constructed as a corresponding directed recipe superedge. : ； In the formula, Indicates the formula type. For the weight of the superedge; Based on each of the aforementioned formulations, the super-edge The set of superedges of the formula and entity node set Constructing a supermap of tin-based materials : ； In the formula, These represent the input correlation matrix, which describes the degree of involvement of raw materials and processes in the formulation, and the output correlation matrix, which describes the performance level obtained by a formulation. in: ； ； In the formula, , , This represents the normalized content of raw material node j in the i-th formulation. This represents the normalized process parameter value. This represents the normalized performance measurement.

4. The method for predicting tin-based material formulations based on structural entropy as described in claim 1, characterized in that, Step S30 specifically includes: The preset weights of the recipe hyperedges are assigned to the entity nodes contained in the recipe hyperedges to obtain the degree of each entity node, and the structural proportion is calculated based on the degree of each entity node. Construct an encoding tree based on the participation degree of the superedges of each entity node, and calculate the multi-level structural entropy based on the hierarchical partitioning of the entity nodes using the encoding tree. : ； In the formula, Indicates the number of levels in the coding tree. Represents the root node of the coding tree. This represents a non-root node in the coding tree. express The sum of the structural participation degrees of the corresponding entity nodes, This represents the sum of the recipe superedge weights between this entity node and other entity nodes. express The parent node, This represents the sum of the structural participation degrees of all entity nodes; Select the target coding tree with the minimum multilayer structure entropy from the candidate coding trees, and use the multilayer structure entropy corresponding to the target coding tree as the structure entropy of the tin-based material hypergraph. The association value between the entity node and the recipe hyperedge in the tin-based material hypergraph is adjusted to ρ times the original value to obtain the tin-based material hypergraph after association change, and the structural entropy of the tin-based material hypergraph after association change is calculated, where ρ∈[0,1], and ρ=0 indicates that the association is removed; The difference between the structural entropy of the tin-based material hypergraph and the structural entropy of the tin-based material hypergraph after correlation change is taken as the structural entropy difference.

5. The method for predicting tin-based material formulations based on structural entropy as described in claim 1, characterized in that, Step S40 specifically includes: After normalizing the structural entropy difference, the correlation strength is calculated. : ； In the formula, , , and These represent the obtained input correlation value and the output correlation value, respectively. Indicates the correlation adjustment factor. This represents the normalized structural entropy difference. The correlation strength After normalization, the entity features in the formulation hyperedge are weighted and converged to obtain the representation of the i-th formulation hyperedge. : ； In the formula, d represents the feature of the entity node; This is the normalized correlation strength value; For entity nodes; Based on the super-edge characterization of the formula Determine the augmented representation of entities : ； In the formula, Indicates the presence of entity nodes The set of superedges of the recipe, To prevent positive numbers with a denominator of zero.

6. The method for predicting tin-based material formulations based on structural entropy as described in claim 1, characterized in that, Step S50 specifically includes: The types, target measurements, and allowable deviations of each performance metric in the preset performance conditions are normalized and then fused with the entity-enhanced representation of the corresponding target performance entity node. Finally, the target performance representation is obtained by weighting and aggregating the data according to the importance of each performance metric. : ； In the formula, Represents entity nodes The importance of , where d represents the dimension of representation; The similarity between the target performance representation and the entity augmentation representation is used as the association score. , will correlate scores Above the candidate raw material screening threshold The raw material node was determined as a candidate raw material node. Those with a correlation score higher than the candidate process screening threshold will be considered. The process node was determined as a candidate process node. : ； ； The candidate raw material node With the candidate process node Form a set of candidate input nodes The candidate input node set and the target performance node set are connected to form the candidate recipe hyperedge. ; The connection score between the candidate recipe superedge and the preset performance condition is calculated using a scoring function. .

7. The method for predicting tin-based material formulations based on structural entropy as described in claim 1, characterized in that, Step S60 specifically includes: Determine the entity nodes in the target recipe hyperedge The feasible region corresponding to the conditions of satisfying preset constraints, wherein the preset constraints include raw material content constraints and process parameter constraints; Determine the raw material ratios and process parameters of the target formulation superedge within the feasible region; Based on the raw material ratio and the process parameters, a tin-based material formulation that meets the preset performance conditions is predicted.

8. The application of the tin-based material formulation prediction method based on structural entropy as described in any one of claims 1 to 7 in tin-based material formulation prediction or formulation design.

9. A computer system, characterized in that, The computer system includes: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the computer program, when executed by the processor, implements the steps of the method for predicting tin-based material formulations based on structural entropy as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, implements the steps of the method for predicting tin-based material formulations based on structural entropy as described in any one of claims 1 to 7.

Images