Knowledge graph-based conductive adhesive dispensing parameter assisted decision method

Abstract

The application discloses a conductive glue dispensing parameter auxiliary decision-making method based on a knowledge graph and belongs to the technical field of microwave assembly. The conductive glue dispensing process knowledge graph can provide processing parameters for new conductive glue, provides technical support for the application of the new conductive glue, and solves the problem that microwave assembly processing units cannot provide effective basis and processing parameters for processability. The verification frequency is saved through a fluid mechanics model. After multiple autonomous learning and iteration, the training error can be less than 8%, the verification efficiency of new materials is improved, and the verification cost and verification time are saved. Through big data learning, processing parameter suggestions can be provided according to the characteristics of each device, the problem that traditional process parameters do not consider the processing difficulty of different devices on the same object is solved, and the processing qualification rate is improved by more than 20%.

Description

Technical Field

[0001] This invention relates to the field of microwave component assembly technology, and more specifically to a knowledge graph-based method for assisting decision-making on conductive adhesive dispensing parameters. Background Technology

[0002] Conductive adhesive is a crucial raw material for bare-chip interconnection in microwave components. It not only secures microwave components but also serves as the electrical interconnect medium for internal interconnection and telecommunications functions. While nearly a hundred types of conductive adhesives exist internationally, the variety and grades of domestically produced conductive adhesives are increasing with the progress of independent and controllable research and development of domestic materials. This has brought a significant workload to the selection of conductive adhesives and the setting of process parameters in microwave component manufacturing.

[0003] Conductive adhesive is transferred from the packaging container to the bonding interface through a dispensing process, completing the fixation and electrical connection between the upper and lower layers. The principle of dispensing, the equipment used, the control method, and the specific process steps all directly affect the adhesive bonding quality and reliability. With the trend of miniaturization of bonding surfaces, there are requirements for controlling the volume of extracted adhesive to be smaller, for more precise positioning of the dispensing device, for further improvement in dispensing time, and for more precise repeatability and consistency of dispensing. When faced with new conductive adhesives with different solid contents, viscosities, and open times, it is difficult to quickly design and verify a scheme to obtain the optimal dispensing process parameters.

[0004] A knowledge graph is a knowledge base that represents entities, concepts, and their relationships in the objective world in the form of graphs. It is a network-like knowledge structure composed of "entity-relationship-entity" triples, and concept retrieval can be achieved through reasoning. The lack of a knowledge graph for conductive adhesives has limited the promotion and use of some new conductive adhesives, and has also made the dispensing process of conductive adhesives unstable and unsustainable, frequently resulting in problems such as "satellite drips" that contaminate the surface of parts, clog nozzles, adhesive stringing, and tailing.

[0005] Currently, research on applying knowledge graphs to conductive adhesive dispensing is still lacking. Traditional DOE (Document of Knowledge) is mainly used for process design and optimization analysis. Although these methods can obtain information on process influencing factors and their importance, they lack the requirement for extensive process experiments and cannot establish a correlation between the intrinsic properties of conductive adhesives and process parameters. Therefore, they have the disadvantages of not being able to perform correlation analysis on process parameters when applying new conductive adhesives and not being able to systematically summarize and utilize historical cases.

[0006] Fluid dynamics simulation can quickly simulate the morphology of adhesive dots and the stress and strain during the dispensing process based on fluid characteristics and processing parameters. Simulation can rapidly verify the feasibility of process parameters, saving significant manpower and costs. However, the presentation of simulation results depends on the skill level of the simulation technicians, hindering their utilization and reference. Therefore, this paper proposes a knowledge graph-based auxiliary decision-making method for conductive adhesive dispensing parameters. Summary of the Invention

[0007] The technical problem this invention aims to solve is: how to address the lack of theoretical basis for the dispensing process settings of existing novel conductive adhesives, the lack of optimization direction for process parameters, the failure to consider the differences in processing difficulty of the same object by different equipment, and the problems of low yield, low production efficiency, and increased processing costs due to improper parameter selection. This invention provides a knowledge graph-based auxiliary decision-making method for conductive adhesive dispensing parameters. This method provides dispensing processes for novel conductive adhesives through knowledge graphs and completes the evaluation in the knowledge graph using fluid dynamics simulation results, providing scientific and effective processing methods and parameters for the use of novel conductive adhesives.

[0008] The present invention solves the above-mentioned technical problems through the following technical solution, and the present invention includes the following steps:

[0009] S1: Knowledge Graph Embedding

[0010] Import the product manuals for conductive adhesives into a knowledge graph database;

[0011] S2: Recommended Dispensing Process Parameters

[0012] Natural language processing technology is used to extract conceptual entities from product manuals, and the text corpus is expressed through sequence labeling. This corpus serves as the input vector for a trained process parameter recommendation model, which selects dispensing process parameters from a knowledge graph database.

[0013] S3: Simulation Verification

[0014] Using a fluid dynamics model, the dispensing process with the dispensing parameters provided in step S2 is simulated to quickly verify its feasibility; based on the simulation results, a decision is made, and if the requirements are not met, the process is returned to step S2 until the requirements are met; based on the simulation results, the final conclusion is output.

[0015] S4: Manual Judgment

[0016] Actual processing is carried out based on the final conclusions, and adjustments are made based on experience to obtain the final dispensing process parameters;

[0017] S5: Iterative Optimization

[0018] The actual effect of the final dispensing process parameters is used as the judgment result and embedded into the knowledge graph to complete the optimization and upgrading of the knowledge graph relationship.

[0019] Furthermore, in step S1, when constructing the knowledge graph, entities and relationships need to be constructed. When constructing entities, important elements in the entity are determined based on axioms, theorems, formulas, and common knowledge from academia or industry. This is achieved by defining core concepts, hierarchical structures, and concept attributes, then defining attribute types and attribute relationships, and finally creating the entity. When constructing relationships, the ComplEx model is selected as the knowledge graph embedding model, and relationships are constructed through knowledge extraction and instance filling.

[0020] Furthermore, in step S1, based on the entities and relationships in the knowledge graph, the recommended dispensing process parameters for different conductive adhesives at different interfaces are transformed into entity searches for related questions in the knowledge graph; the entities include adhesive type, adhesive coating, adhesive surface type, adhesive surface coating, conductive adhesive properties, dispensing equipment type, dispensing process parameters, and usage environment.

[0021] Furthermore, in step S1, a BiLSTM network based on a feedback mechanism is selected to extract the conductive adhesive properties. The output vector of the BiLSTM is represented as follows:

[0022]

[0023] Among them, h i ∈R 2D Let be the output vector of the BiLSTM network at time i, and D be the number of hidden layer nodes in the BiLSTM network.

[0024] Furthermore, in step S1, the conductive adhesive bonding-related entities and the dispensing process parameter entities are defined as complex vectors, with the head entity h... i ∈C K Tail entity t∈C K And relation embedding r∈C K Existing in complex space, the ComplEx model relation matrix M is a diagonal matrix, and the model scoring function is:

[0025]

[0026] in, It is the conjugate of t, diag(r) denotes the relational embedding part of the diagonal matrix M, <.> denotes the dot product operation, Re(.) denotes the real part of the complex number, Im(.) denotes the imaginary part of the complex number, r = iIm(r) indicates an antisymmetric relation, and r = x∈R indicates a symmetric relation.

[0027] Furthermore, in step S1, the existing conductive adhesive dispensing processing parameters are imported into a knowledge graph database, and a combination of rule-based reasoning, deep learning, and reinforcement learning is used to train and determine the process parameter recommendation model using a generative model, an identification model, and an optimizer.

[0028] The process parameter recommendation model is formed by combining a generation model, an identification model, and an optimizer. The generation model uses a recurrent neural network model, the identification model uses a convolutional neural network model, and the optimizer uses the Radam optimizer.

[0029] Furthermore, in step S2, the conductive adhesive type, bonding type, and selected processing equipment are input. The process parameter recommendation model is used to determine the relationship based on the knowledge graph. Reasoning is performed according to reasoning and conflict resolution strategies. The corresponding dispensing process parameters are obtained from the parameters obtained by the ComplEx model. The output dispensing process parameters include dispensing time, dispensing pressure, and adhesive dot distribution.

[0030] Furthermore, in step S3, based on the Navier-Stokes equations, the dispensing process parameters and conductive adhesive properties from step S2 are used to perform numerical simulation calculations using ANSYS FLUENT software. A continuous surface tension (CSF) model and a wall adhesion model are added to the ANSYS FLUENT software. A VOF model is selected to track the interface between two or more immiscible fluids on a fixed Euler grid. During the simulation, the sum of all phase volume fractions in each computational cell is 1, and all other variables and physical parameters are shared within the cell. The value of the phase volume fraction determines whether the grid represents a mixture of one or more phases.

[0031] Furthermore, in step S3, the final conclusions include the viscosity of the adhesive, contact angle, surface tension, and the quality and stability of the adhesive spot.

[0032] Furthermore, in step S5, the actual performance after processing according to the dispensing process parameters given in step S3 is input into the knowledge graph and compared with the evaluation value. The weights and thresholds of the trained process parameter recommendation model are saved for subsequent performance prediction to reduce training errors.

[0033] Compared with existing technologies, this invention has the following advantages: This knowledge graph-based conductive adhesive dispensing parameter auxiliary decision-making method can provide processing parameters for novel conductive adhesives through a knowledge graph of conductive adhesive dispensing processes, providing technical support for the application of novel conductive adhesives and solving the problem that microwave component processing units cannot provide effective evidence and processing parameters regarding processability; the use of fluid dynamics models reduces the number of verification attempts, and after multiple self-learning and iterations, the training error can be less than 8%, improving the verification efficiency of new materials and saving verification costs and time; through big data learning, it can provide targeted processing parameter suggestions based on the characteristics of each device, solving the problem that traditional process parameters do not consider the processing difficulty of different devices for the same object, thus increasing the processing pass rate by more than 20%. Attached Figure Description

[0034] Figure 1 This is a flowchart illustrating the knowledge graph-based conductive adhesive dispensing parameter auxiliary decision-making method of the present invention.

[0035] Figure 2 This is a schematic diagram of the recurrent neural network structure in Embodiment 1 of the present invention;

[0036] Figure 3 It is the knowledge graph vector graph in Embodiment 1 of the present invention;

[0037] Figure 4 This is a fluid dynamics simulation diagram based on H2OE in Embodiment 2 of the present invention;

[0038] Figure 5 This is a dispensing morphology diagram based on H2OE in Embodiment 2 of the present invention;

[0039] Figure 6 This is an example diagram of the J-427 product manual obtained by crawling in Embodiment 3 of the present invention;

[0040] Figure 7 This is a fluid dynamics simulation diagram based on J-427 in Embodiment 3 of the present invention;

[0041] Figure 8 This is the J-427 dispensing morphology diagram of the dispensing process parameters obtained from the knowledge graph in Embodiment 3 of the present invention;

[0042] Figure 9 This is a J-427 dispensing morphology diagram showing the dispensing process parameters after manual adjustment in Embodiment 3 of the present invention. Detailed Implementation

[0043] The embodiments of the present invention are described in detail below. These embodiments are implemented based on the technical solution of the present invention, and provide detailed implementation methods and specific operation processes. However, the scope of protection of the present invention is not limited to the following embodiments.

[0044] Example 1

[0045] like Figure 1 As shown, this embodiment provides a technical solution: a knowledge graph-based auxiliary decision-making method for conductive adhesive dispensing parameters, comprising the following steps:

[0046] (1) Knowledge graph embedding: Import the product manual of the new conductive adhesive into the knowledge graph database.

[0047] (2) Recommendation of dispensing process parameters: Natural language processing technology is used to extract conceptual entities from product manuals and express the text corpus through sequence labeling (such as online product manuals, which are descriptive language) as the input vector of the trained process parameter recommendation model, and process parameters are selected from the knowledge graph database.

[0048] (3) Simulation Verification: Using a fluid dynamics model, the dispensing process with provided dispensing parameters is simulated to quickly verify its feasibility. Decisions are made based on the simulation results; for those that do not meet the requirements, the process is reverted to the recommended steps for the dispensing parameters until the requirements are met. The final conclusion is output based on the simulation results.

[0049] (4) Manual judgment: Actual processing is carried out based on the dispensing process parameters in the final conclusion, and adjustments are made based on experience to obtain the final dispensing process parameters.

[0050] (5) Iterative optimization: The actual effect of the final dispensing process parameters obtained in step (4) is used as the judgment result and embedded into the knowledge graph to complete the optimization and upgrading of the knowledge graph relationship.

[0051] In step (1), the knowledge graph is constructed primarily using a top-down approach, mainly building entities and relationships.

[0052] More specifically, entities mainly include three types: bonding environment, intrinsic properties of conductive adhesive, and dispensing process parameters. Based on axioms, theorems, formulas, and common knowledge from academia or industry, the important elements in the entity are determined. The entity is created by defining core concepts, hierarchical structure, and conceptual attributes, then defining attribute types and attribute relationships.

[0053] More specifically, entities include, but are not limited to, adhesive type, adhesive coating, adhesive surface type, adhesive surface coating, conductive adhesive properties, dispensing equipment type, dispensing process parameters, and usage environment. Adhesive type, adhesive coating, adhesive surface type, and adhesive surface coating are entities of the adhesive environment type; conductive adhesive properties are entities of the intrinsic characteristics of conductive adhesive; and dispensing process parameters and usage environment are entities of the dispensing process parameter type.

[0054] More specifically, the explicit or implicit relationships, such as the constraints and synergies among various properties of conductive adhesives, the mutual influence between conductive adhesive properties and process parameters, and the constraints and synergies between conductive adhesives and the environment, are expressed using vector diagrams.

[0055] More specifically, based on the entities and relationships in the knowledge graph, the recommended process parameters for different conductive adhesives at different interfaces are transformed into entity searches for related questions in the knowledge graph.

[0056] More specifically, embedding models that utilize deep interactive information about entities and relationships, such as neural network models and bilinear models, are used.

[0057] More specifically, the ComplEx model is selected as the knowledge graph embedding model, and relationships are constructed through knowledge extraction and instance filling.

[0058] In step (1), since the conductive adhesive product manual is a professional technical document, the performance of the conductive adhesive is usually listed directly in tabular form. Therefore, natural language processing technology is used to extract the concept entities from the semi-structured text.

[0059] More specifically, information such as conductive adhesive product manuals and equipment used (e.g., whether dispensing is automatic or manual, obtained through web crawlers) is obtained by manually constructing and using web crawlers to extract raw data.

[0060] More specifically, a BiLSTM network based on a feedback mechanism was selected to extract conductive adhesive performance information.

[0061]

[0062] Among them, h i ∈R 2D Let D be the output vector of the network at time i (where D is the number of hidden nodes in the BiLSTM layer).

[0063] In step (1), the conductive adhesive bonding-related entities (adhesive surfaces and their characteristics) and the dispensing process parameter entities are defined as complex vectors, with the head entity h... i ∈C K Tail entity t∈C K And relation embedding r∈C K It exists in complex space. The ComplEx model relation matrix M is a diagonal matrix, and the model scoring function is:

[0064]

[0065] in, Let be the conjugate of t, diag(r) denote the relational embedding of the diagonal matrix M, <.> denote the dot product operation, Re(.) denote the real part of the complex number, and Im(.) denote the imaginary part of the complex number. When r = iIm(r) (the real part is 0), it represents an antisymmetric relation, and when r = x∈R, it represents a symmetric relation.

[0066] In step (1), a knowledge graph database was established using the Neo4j graph structure database. The set of feature paths for conductive adhesive entities was calculated using Cypher statements in Java to manipulate the Neo4j database.

[0067] In step (1), the existing conductive adhesive dispensing processing parameters are imported into the knowledge base. Rule-based reasoning, combined with deep learning and reinforcement learning, is used to train and determine the process parameter recommendation model using a generative model, an identification model, and an optimizer. Thresholds are set for determining relationships.

[0068] More specifically, the process parameter recommendation model is formed by combining a generative model, a qualification model, and an optimizer.

[0069] More specifically, the generative model uses a recurrent neural network (RNN) model, the identification model uses a convolutional neural network (CNN) model, and the optimizer uses the Radam optimizer.

[0070] More specifically, the generative model is as follows:

[0071]

[0072] Among them, R T It is the reward when the entire sequence is fully generated. It is a signal generated by the identification model D. P(·) represents the action function value when the entire sequence is generated. It is the expected cumulative reward obtained by the current generation model G taking the current action Z from the current state S until the end.

[0073] More specifically, the identification model ultimately determines the probability that the generated sequence is true, and then feeds it back to the generation model as a reward. The generation model then continues to generate sequences based on the latest reward.

[0074] More specifically, through deep learning iterations, the criteria for knowledge graph judgment are solidified into a judgment system.

[0075] In step (2), the conductive adhesive type, bonding type, and selected processing equipment are input. The process parameter recommendation model is used to determine the relationship based on the knowledge graph. Reasoning is performed according to the reasoning and conflict resolution strategies. The corresponding dispensing process parameters are obtained from the parameters obtained by the ComplEx model.

[0076] In step (2), the output dispensing process parameters include, but are not limited to: dispensing time, dispensing pressure, glue dot distribution, conductive adhesive type, bonding type and processing equipment.

[0077] More specifically, conductive adhesive types include, but are not limited to, grade, capacity, open time, solvent content (or absence), viscosity, and solids content.

[0078] More specifically, bonding types include, but are not limited to, bonding (non-bonding) chips, substrates (non-substrate), and passing (non-passing) microwave signals.

[0079] More specifically, processing equipment includes manual dispensing machines and automatic dispensing machines; manual dispensing is not within the scope of the knowledge graph.

[0080] More specifically, automatic dispensing machines should be labeled as either contact or non-contact.

[0081] In step (3), based on the Navier-Stokes equations, the dispensing process parameters and conductive adhesive properties from step (2) are used to perform numerical simulation calculations using ANSYS FLUENT software.

[0082] More specifically, the VOF (Volume of Fluid) model is used to trace the interface between two or more immiscible fluids on a fixed Euler grid.

[0083] More specifically, during simulation, the sum of all phase volume fractions in each computational cell grid is 1, and all other variables and physical parameters are shared within the grid cell. The value of the phase volume fraction determines whether the grid represents a mixture of one or more phases.

[0084] More specifically, add a continuous surface tension (CSF) model to FLUENT:

[0085]

[0086] Where σ is the surface tension coefficient; k is the surface curvature; ρ1 and ρ2 are the densities of air and liquid, respectively; α q This represents the volume ratio of the function.

[0087] More specifically, to realistically reflect the flow of adhesive in capillaries and the actual situation after adhesive adheres to the substrate surface, a wall adhesion model was added to FLUENT.

[0088] More specifically, in order to quantitatively analyze the changes in field quantities during fluid flow, it is necessary to establish and solve the governing equations that describe the process. These mainly include the mass conservation equation (continuity equation), the momentum conservation equation, and the energy conservation equation. The mathematical description of these equations is called the governing equations.

[0089] Among them, the mass conservation equation is:

[0090]

[0091] In the formula, ρ is density and t is time.

[0092] Considering that the polymer is an incompressible fluid, the density ρ is constant, and the equation can be rewritten as:

[0093]

[0094] Momentum conservation equation:

[0095] In a three-dimensional coordinate system, this law must be satisfied in the x, y, and z directions.

[0096]

[0097]

[0098]

[0099] In the formula, p is the pressure on the fluid; τ xx τ xy and τ xz It is the viscous component generated by the viscous effect; F is the volume force on the fluid.

[0100] Energy conservation equation:

[0101]

[0102] In the formula, c p Specific heat capacity, T is temperature, k is thermal conductivity, and S is... T This is the viscous dissipation component.

[0103] Component mass conservation equation:

[0104]

[0105] In the formula, c s The volume concentration of component s, ρc s For mass concentration, D s S is the diffusion coefficient. s The change in mass of component s per unit time.

[0106] In step (3), the simulation yields conclusions on the viscosity of the adhesive, contact angle, surface tension, and the quality and stability of the adhesive spot.

[0107] In step (5), the actual performance after processing according to the parameters given in step (3) is input into the knowledge graph and compared with the evaluation value. The weights and thresholds of the trained process parameter recommendation model are saved for subsequent performance prediction to reduce training error.

[0108] Example 2

[0109] like Figure 1 As shown, for conductive adhesives already included in the knowledge graph, the specific operation flow of the knowledge graph-based conductive adhesive dispensing parameter auxiliary decision-making method is as follows:

[0110] (1) Knowledge graph embedding: Input conductive adhesive grade H20E.

[0111] (2) Recommended dispensing process parameters.

[0112] The knowledge graph retrieves the processed conductive adhesive parameters, as shown in Table 1 below. The user confirms whether the parameters need modification. If not, proceed to the next step; if filtering is required, a change request must be submitted to the administrator for review and authorization.

[0113] Table 1 Key Parameters of H20E

[0114]

[0115] After determining the conductive adhesive material, select the bonding material and bonding interface parameters, as shown in Table 2 below.

[0116] Table 2 Adhesion Parameters

[0117]

[0118] Based on existing models in the knowledge graph, and by comparing the correlation between the bonding objects and dispensing process parameters of 106 existing H20E samples, the recommended dispensing process parameters are shown in Table 3 below.

[0119] Table 3 Dispensing process parameters

[0120] Recommended solution glue dispensing time Dispensing pressure Glue dot distribution Automatic dispensing 0.05～0.08s 40-45 psi 4×4 Manual dispensing 1～1.5s 45-50 psi 9×9

[0121] (3) Simulation verification:

[0122] Based on the obtained dispensing process parameters, the knowledge graph automatically calls the embedded ANSYS FLUENT software to perform simulation, and the simulation results are shown in the attached figure. Figure 4 As shown.

[0123] By comparing the simulated graphics with the actual graphics using the knowledge graph, the system determines that the dispensing process parameters are usable and then provides these parameters back to the user.

[0124] (4) Manual judgment:

[0125] Based on the given dispensing process parameters, an automatic dispensing machine is used to dispense adhesive, resulting in adhesive dots of the shape shown in the attached figure. Figure 5 As shown, it meets the usage requirements.

[0126] (5) Iterative optimization:

[0127] Image of the morphology of the glue dots ( Figure 5 Input the knowledge graph to complete one process.

[0128] Example 3

[0129] like Figure 1 As shown, for conductive adhesives not included in the knowledge graph, the specific operation flow of the knowledge graph-based conductive adhesive dispensing parameter auxiliary decision-making method is as follows:

[0130] (1) Knowledge graph embedding: Input conductive adhesive grade: J-427 from the Petrochemical Research Institute of Heilongjiang Academy of Sciences.

[0131] The product manual was obtained from the internet via web scraping, as shown in the attached image. Figure 6 As shown.

[0132] (2) Recommended dispensing process parameters:

[0133] The performance of the J-427 was extracted using BiLSTM, as shown in Table 4 below.

[0134] Table 4 Key Parameters of J-427

[0135]

[0136] Among them, viscosity and chip shear strength are determined by converting units using natural language from a knowledge graph, and then using a unified scale.

[0137] After determining the conductive adhesive material, select the bonding material and bonding interface parameters, as shown in Table 5 below.

[0138] Table 5 Adhesion Parameters

[0139]

[0140]

[0141] The recommended dispensing process parameters are predicted by establishing a predictive model using an adversarial neural network. The nodes "J-427" and "adhesion parameters" are associated with the relationship "dispensing process parameters". The dispensing equipment, dispensing time, and dispensing pressure are used as single output layer nodes of the three neural network models, and a multi-input single-output mapping mode under a three-layer neural network is adopted.

[0142] The seven factors of open time, solvent content (including / excluding), viscosity, chip shear strength, volume resistivity, coefficient of thermal expansion, and glass transition temperature were used as input data for the generation model; the three factors of dispensing time, dispensing pressure, and glue dot distribution were used as output data for the evaluation model.

[0143] The Radam optimizer compares the expected output data with the actual output data of the model during subsequent training, setting relevant parameters required during training, such as: maximum number of training iterations (time), learning accuracy (eps), number of hidden layer nodes, initial learning rate (η), momentum factor (α), weight matrix, and threshold matrix.

[0144]

[0145] Where k = 0, 1, 2, ..., q, v ki Here, f1 is the weight between the input layer and the hidden layer, f1() is the activation function of the hidden layer, and b k This represents the threshold of neurons in the hidden layer.

[0146] Establish the glue dispensing time matrix T = {t i ,i=1,…,K}, where K represents the types of glue dispensing time already existing in the knowledge graph.

[0147] Establish the dispensing pressure matrix P = {p i ,i=1,…,K}, where K represents the types of dispensing pressures already existing in the knowledge graph.

[0148] Establish the glue dot distribution matrix S = {s} i ,i=1,…,K}, where K represents the existing types of glue dot distributions in the knowledge graph.

[0149] The three matrices are linked by the relation "togerther" to form an adjacency matrix.

[0150] Generate vectors Each element:

[0151]

[0152] Based on existing models in the knowledge graph, it was inferred that the J-427 product is similar to Henkel's 84-1A and 84-1Lmit1 products. By comparing the correlation between the bonding objects and dispensing process parameters of 1593 existing 84-1A and 84-1Lmit1 products, the optimized dispensing process parameters were obtained as shown in Table 6 below.

[0153] Table 6 Dispensing Process Parameters

[0154] Recommended solution glue dispensing time Dispensing pressure Glue dot distribution Automatic dispensing 0.05～0.1s 40-45 psi 4×4 Manual dispensing 2～3s 50-60 psi 6×6

[0155] (3) Simulation verification:

[0156] Based on the obtained dispensing process parameters, the knowledge graph automatically calls the embedded ANSYS FLUENT software to perform simulation, and the simulation results are shown in the attached figure. Figure 7 As shown.

[0157] By comparing the simulated graphics with the actual graphics using a knowledge graph, the system determines that the dispensing process parameters are usable and provides the process parameters back to the user.

[0158] (4) Manual judgment:

[0159] Based on the given dispensing process parameters, an automatic dispensing machine is used to dispense adhesive, resulting in adhesive dots of the shape shown in the attached figure. Figure 8 As shown, the glue dot shape is slightly pointed. Based on experience, the adjusted dispensing process parameters are shown in Table 7, and the glue dot shape is as shown in the attached figure. Figure 9 The results show that the usage requirements are met.

[0160] Table 7 Dispensing Process Parameters

[0161] Recommended solution glue dispensing time Dispensing pressure Glue dot distribution Automatic dispensing 0.05～0.08s 50-52 psi 4×4 Manual dispensing 1.8～2.2s 55-68psi 6×6

[0162] (5) Iterative optimization:

[0163] Image of the morphology of the glue dots ( Figure 9 Input the knowledge graph to complete one process.

[0164] In summary, the knowledge graph-based conductive adhesive dispensing parameter auxiliary decision-making method in the above embodiments can provide processing parameters for novel conductive adhesives through a knowledge graph of conductive adhesive dispensing processes, providing technical support for the application of novel conductive adhesives and solving the problem that microwave component processing units cannot provide effective evidence and processing parameters for processability; the fluid dynamics model reduces the number of verification times, and after multiple self-learning and iterations, the training error can be less than 8%, improving the verification efficiency of new materials and saving verification costs and time; through big data learning, it can provide targeted processing parameter suggestions based on the characteristics of each device, solving the problem that traditional process parameters do not consider the processing difficulty of different devices for the same object, thus increasing the processing pass rate by more than 20%.

[0165] Although embodiments of the present invention have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of the present invention.

Claims

1. A knowledge graph-based conductive adhesive dispensing parameter assisted decision-making method, characterized in that, The method comprises the following steps: S1: knowledge graph embedding Import the product manual of conductive adhesive into the knowledge graph database; S2: dispensing process parameter recommendation Extract the concept entity in the product manual by using natural language processing technology, express the text corpus by sequence labeling, use the trained process parameter recommendation model as the input vector, and select the dispensing process parameter from the knowledge graph database; S3: simulation verification Use the fluid mechanics model to simulate the dispensing process of the dispensing process parameter provided in step S2, quickly verify the feasibility, make a decision according to the simulation result, return to step S2 if it does not meet the requirements, and output the final conclusion according to the simulation result; S4: artificial judgment According to the final conclusion, the actual processing is carried out, and the final dispensing process parameter is obtained according to the experience; S5: iterative optimization The actual effect of the final dispensing process parameter is embedded into the knowledge graph as a judgment result to complete the knowledge graph relationship optimization and upgrade; In the step S1, when constructing the knowledge graph, the entity and the relationship need to be constructed; when constructing the entity, the important elements in the entity are determined according to the axioms, theorems, formulas, and public knowledge of the academic or industry, the core concept, hierarchical structure, and concept attribute are defined, the attribute type and attribute relationship are defined, and finally the entity is created; when constructing the relationship, the ComplEx model is selected as the knowledge graph embedding model, and the relationship is constructed by knowledge extraction and instance filling; In the step S1, according to the entities and relationships in the knowledge graph, the dispensing process parameters of different conductive adhesives in different interfaces are recommended and converted into entity search of related problems in the knowledge graph; the entity includes the adhesive type, the adhesive coating, the adhesive surface type, the adhesive surface coating, the conductive adhesive performance, the dispensing equipment type, the dispensing process parameter, and the use environment; In the step S1, the conductive adhesive bonding related entity and the dispensing process parameter entity are defined as a complex vector, the head entity h i ∈C K , the tail entity t∈C K and the relationship embedding r∈C K exist in a complex space, the ComplEx model relationship matrix M is a diagonal matrix, and the model score function is: wherein is the conjugate complex of t, diag(r) denotes the relational embedding part of the diagonal matrix M, <.> denotes the dot product operation, Re(.) denotes the real part of a complex number, Im(.) denotes the imaginary number of a complex number, r = i Im(r) denotes the anti-symmetric relation, and r = x e R denotes the symmetric relation; In the step S1, the BiLSTM network based on the feedback mechanism is selected to extract the conductive adhesive performance, and the output vector of BiLSTM is represented as follows: wherein h i ∈R 2D is the output vector of the BiLSTM network at time i, and D is the number of hidden layer nodes of the BiLSTM network. In the step S1, the existing dispensing process parameters of conductive adhesive are imported into the knowledge graph database, the rule reasoning, deep learning, and reinforcement learning are combined, and the generation model, identification model, and optimizer are used to train and judge the process parameter recommendation model; The process parameter recommendation model is formed by combining the generation model, identification model, and optimizer, the generation model uses the recurrent neural network model, the identification model uses the convolutional neural network model, and the optimizer uses the Radam optimizer; In the step S2, the conductive adhesive type, the adhesive type, and the selected processing equipment are input, the process parameter recommendation model is used to determine according to the relationship of the knowledge graph, the reasoning and conflict resolution strategy are used for reasoning, the corresponding dispensing process parameter is obtained from the parameters obtained by the ComplEx model, and the output dispensing process parameter includes the dispensing time, the dispensing pressure, and the glue point distribution; In the step S3, according to the Navier-Stokes equation, the point gluing process parameters and the related properties of the conductive glue in the step S2 are used for numerical simulation calculation by using the ANSYS FLUENT software, and the continuous surface tension CSF model and the wall adhesion model are added in the ANSYS FLUENT software; the VOF model is selected to track the interface of two or more mutually insoluble fluids on the fixed Euler grid; in each calculation unit grid during simulation, the sum of all phase volume fractions is 1, and the remaining variables and physical parameters are shared in the grid unit, and whether the grid represents one phase or a mixture of multiple phases is determined by the numerical value of the phase volume fraction; In the step S5, the performance after actual processing according to the point gluing process parameters given in the step S3 is input into the knowledge graph, compared with the evaluation value, the weight and threshold of the trained process parameter recommendation model are saved for subsequent performance prediction, and the training error is reduced.

2. The knowledge graph-based conductive adhesive dispensing parameter auxiliary decision-making method according to claim 1, characterized in that, In the step S3, the final conclusion includes the glue viscosity, the contact angle, the surface tension, and the quality and stability of the glue spot.

Images