Multi-source feature extraction, storage and intelligent retrieval system for adhesive formulations

Abstract

This invention discloses a multi-source feature extraction, storage, and intelligent retrieval system for adhesive formulations, including a multi-source formulation data receiving module, a raw material morphology identification module, an effective feed conversion module, an effective proportion vector generation module, a formulation feature storage module, and a similarity retrieval and ranking module. The system receives the original feed data and raw material attribute data of the adhesive formulation, identifies emulsions, resin solutions, dispersions, solvent-based additives, or diluent functional components, and converts the original feed amount into the effective feed amount based on solid content, effective component ratio, or active component ratio, generating an effective formulation proportion vector. During retrieval, the system calculates the similarity between the target formulation and historical formulations based on their effective formulation proportion vectors, and corrects the ranking results when the original feed proportions are similar but the effective formulation proportions deviate, thereby reducing misjudgments of formulation similarity caused by inconsistencies between the nominal feed amount and the effective component ratio.

Description

Technical Field

[0001] This application relates to the field of adhesive formulation data processing technology, and in particular to a multi-source feature extraction, storage and intelligent retrieval system for adhesive formulations. Background Technology

[0002] In the process of adhesive formulation development, researchers typically need to search, reuse, and compare historical formulations to determine the degree of similarity between the target formulation and existing formulations. Existing formulation management systems usually establish formulation databases using fields such as raw material name, raw material category, initial feed amount, mass fraction, mass percentage, and performance test results, and output similar formulations based on the degree of overlap in raw material names, the similarity of feed ratios, or the similarity of performance indicators. However, adhesive raw materials do not always exist in the form of pure active components. Common raw materials may manifest as emulsions, resin liquids, dispersions, solvent-based additives, diluted functional components, or rheology modifiers containing carriers. For such raw materials, the initial feed amount in the formulation record usually only represents the total amount of material added to the system and does not directly represent the amount of active ingredients actually involved in film formation, curing, tackification, plasticization, crosslinking, or rheology regulation.

[0003] For example, two water-based adhesive formulations both state an initial feed amount of 100 parts of water-based acrylic emulsion. However, the first emulsion has a solid content of 50%, while the second has a solid content of 35%. Therefore, the corresponding effective polymer feed amounts in these formulations are 50 parts and 35 parts, respectively. If the search system still calculates similarity based on the initial feed amount of 100 parts, it will misclassify two formulations with significantly different effective main resin contents as highly similar. Similarly, for additive dispersions, resin solutions, or diluted functional components, there may be significant discrepancies between the initial feed amount and the effective component feed amount. This results in the system outputting similar formulations that are only close at the nominal feed amount level, but differ considerably at the actual effective component level, making it difficult to accurately reflect the formulation structure.

[0004] As adhesive R&D processes expand from simple historical formula searches to intelligent formula generation, raw material substitution analysis, batch adaptation correction, and process parameter recommendations, historical formula search results are no longer just for manual review by R&D personnel. Instead, they may serve as input for subsequent intelligent design platforms as candidate base formulas. If the initial search stage still relies on the original feed amount as the primary criterion for similarity judgment, historical formulas with similar nominal feed ratios but significantly different effective component ratios may be output as candidate base formulas. This could negatively impact subsequent determination of target effective component ratios, analysis of additive effective dosage, determination of raw material compensation, and recommendation of process parameters. Therefore, it is necessary to establish formula ratio characteristics based on the effective component feed amount during the historical formula search stage. This ensures that the candidate base formulas output by the search system are comparable not only in terms of original feed amounts but also in terms of effective component structure, thereby reducing misjudgments of formula similarity and providing a more reliable data foundation for subsequent AI-driven formula generation, batch adaptation correction, and process parameter recommendations. Summary of the Invention

[0005] This application provides a multi-source feature extraction, storage, and intelligent retrieval system for adhesive formulations. It can solve the technical problem that existing adhesive formulation retrieval systems directly use the original feed amount to calculate the formulation similarity, which leads to the failure to reflect the differences in the effective components of emulsions, resin liquids, dispersions, solvent-based additives, or diluent functional components. Consequently, historical formulations with similar nominal feed amounts but different proportions of effective components are misjudged as similar formulations.

[0006] Firstly, this application provides a multi-source feature extraction, storage, and intelligent retrieval system for adhesive formulations. The system includes a multi-source formulation data receiving module, a raw material morphology identification module, an effective feed conversion module, an effective proportion vector generation module, and a similarity retrieval and ranking module. The multi-source formulation data receiving module receives original feed data and raw material attribute data of the adhesive formulation. The original feed data includes the raw material name and original feed amount, and the raw material attribute data includes the raw material morphology, solid content, and effective ingredient ratio or active component ratio. The raw material morphology identification module identifies raw materials whose original feed amount differs from the effective ingredient feed amount based on the raw material attribute data. The effective feed conversion module determines the effective ingredient feed amount of the raw material to be converted based on its original feed amount and corresponding solid content, effective ingredient ratio, or active component ratio. The effective proportion vector generation module generates an effective formulation proportion vector based on the effective ingredient feed amount of each raw material. The similarity retrieval and ranking module, upon receiving a target formulation retrieval request, calculates the formulation similarity based on the effective formulation proportion vectors of the target formulation and historical formulations, and outputs the historical formulation retrieval and ranking results based on the formulation similarity.

[0007] Through the aforementioned system, adhesive formulation retrieval no longer relies solely on the original feed amount or nominal mass fraction. Instead, it first identifies raw materials whose original feed amount may differ from the effective ingredient feed amount, then converts these differences into effective ingredient feed amounts, and generates an effective formulation ratio vector based on these effective ingredient feed amounts. This allows effective components in emulsions, resin solutions, dispersions, solvent-based additives, or diluent functional components to participate in formulation structure characterization, reducing misjudgments of formulation similarity caused by differences in solid content, effective ingredient ratios, or active ingredient ratios despite identical original feed amounts, and improving the reliability of historical formulation retrieval ranking results.

[0008] In one possible design, the original material input data received by the multi-source formulation data receiving module comes from at least one of the following sources: experimental formulation records, production batch records, raw material supplier information, performance test reports, or historical formulation databases. The multi-source formulation data receiving module is also used to merge data from different sources corresponding to the same adhesive formulation under the same formulation identifier. Through this design, the original material input amounts, raw material attribute data, performance test results, or formulation identifiers recorded separately from different sources can be uniformly associated, avoiding duplicate filing or incomplete feature records for the same formulation due to different data sources.

[0009] In one possible design, the raw material form identification module is used to determine the raw material form based on the raw material form field when it exists in the raw material attribute data; when the raw material form field is missing, the raw material form identification module is used to determine the raw material form based on the raw material name field, solid content field, effective ingredient ratio field, active ingredient ratio field, solvent ratio field, carrier medium field, or historical similar raw material attribute data; through this design, even if the raw material form field is missing, the system can still identify whether it belongs to the raw material that needs to be converted by combining the raw material name and attribute fields, thereby improving the adaptability of historical formula data processing.

[0010] In one possible design, the raw material to be converted includes at least one of emulsion, resin liquid, dispersion, solvent-based additive, and diluent functional component; wherein the emulsion includes film-forming emulsion or aqueous dispersion emulsion, the resin liquid includes resin solution or liquid resin, the dispersion includes additive dispersion, filler dispersion, or functional slurry, and the diluent functional component includes a carrier-containing tackifier, plasticizer, crosslinking agent, or rheology modifier; this design can cover raw material forms in adhesive formulations where the original feed amount differs from the effective ingredient feed amount.

[0011] In one possible design, when the same raw material has at least two of the fields of solid content, effective ingredient ratio, and active component ratio, the effective feed conversion module is used to select the effective ratio field for conversion according to the raw material type; wherein, solid content is preferred for film-forming emulsions, effective ingredient ratio is preferred for resin liquids, dispersions, solvent-based additives or diluent functional components, and active component ratio is preferred for reactive additives; through this design, arbitrary selection of conversion basis can be avoided when multiple effective ratio fields exist at the same time, thereby improving the consistency and accuracy of effective ingredient feed conversion.

[0012] In one possible design, the effective feed conversion module is used to combine the original feed amount of the raw material to be converted with the corresponding solid content, effective ingredient ratio or active component ratio of the raw material to be converted to obtain the effective ingredient feed amount of the raw material to be converted; through this design, the original feed amount in the formula record can be converted into an effective ingredient feed amount that better reflects the actual formula structure.

[0013] In one possible design, for raw materials not identified as raw materials to be converted, the effective proportion vector generation module is used to use the original amount of the raw material as the effective ingredient amount, or to convert the original amount of the raw material into the effective ingredient amount according to a preset effective proportion, and to incorporate the raw materials not identified as raw materials to be converted into the effective formula proportion vector; through this design, both raw materials to be converted and non-raw materials to be converted can enter the same effective formula proportion vector, ensuring the integrity of the formula structure characteristics.

[0014] In one possible design, the system further includes a converted confidence level generation module. This module generates a converted confidence level based on the data source of the effective proportion field, batch matching degree, raw material morphology identification results, or historical attribute data of similar raw materials. The similarity retrieval and ranking module adjusts the weight of the corresponding raw material in the formula similarity calculation based on the converted confidence level. Through this design, even when some raw material attribute data is incomplete or the reliability of the effective proportion field source differs, the system can still retain the raw material in the similarity calculation and control its impact on the retrieval and ranking results through the converted confidence level.

[0015] In one possible design, the effective proportion vector generation module is used to sum the effective component input amounts of each raw material, and to determine the ratio of the effective component input amount of each raw material to the summation value as the vector element of that raw material in the effective formula proportion vector; through this design, adhesive formulas with different total input amounts can be converted to a uniform proportion scale, so that the similarity between the target formula and the historical formula can be calculated according to the effective component proportion.

[0016] In one possible design, the similarity retrieval and ranking module is used to calculate the effective formula ratio similarity based on the effective formula ratio vectors of the target formula and historical formulas, and to calculate the original ingredient ratio similarity based on the original ingredient ratio vectors of the target formula and historical formulas. When the original ingredient ratio similarity is higher than a first threshold and the effective formula ratio similarity is lower than a second threshold, the similarity retrieval and ranking module determines a weighting coefficient based on the difference between the original ingredient ratio similarity and the effective formula ratio similarity, and corrects the ranking score of the historical formula based on the weighting coefficient. Through this design, historical formulas with seemingly similar nominal ingredient amounts but significantly deviating effective ingredient ratios can be identified, avoiding such historical formulas from being preferentially recommended due to their similar original ingredient ratios.

[0017] In one possible design, the similarity retrieval and ranking module is used to simultaneously output information on the effective formula ratio differences between the target formula and historical formulas, as well as the corresponding raw material conversion basis, when outputting the adhesive formula retrieval and ranking results. The effective formula ratio difference information includes differences in the original feed amount of at least one raw material, differences in the feed amount of the effective component, and differences in the effective formula ratio. Through this design, the retrieval results not only include the ranking conclusion but also the source of the effective component differences that led to the ranking differences, facilitating researchers' judgment on whether historical formulas have practical reference value.

[0018] In one possible design, the historical formula retrieval and ranking results are used as candidate basic formula inputs for the adhesive intelligent formula generation platform. The candidate basic formula inputs include the effective ingredient dosage, effective formula ratio vector, and corresponding raw material conversion basis of the historical formula. Through this design, the historical formulas output by the system can not only be manually viewed by R&D personnel, but also provide a more realistic basis for the effective component ratios for subsequent adhesive intelligent formula generation, raw material substitution analysis, batch adaptation correction, or preparation process parameter recommendations, reducing subsequent intelligent design deviations caused by distortion of the front-end candidate basic formulas.

[0019] In summary, the multi-source feature extraction, storage, and intelligent retrieval system for adhesive formulations provided in this application receives the original feed data and raw material attribute data of the adhesive formulation, identifies raw materials whose original feed amounts differ from the effective ingredient feed amounts, converts these raw materials into effective ingredient feed amounts, generates an effective formulation ratio vector based on the effective ingredient feed amounts of each raw material, and performs similarity search and ranking. Because this system can incorporate the effective components from emulsions, resin solutions, dispersions, solvent-based additives, or diluent functional components into the formulation ratio features, it can reduce misjudgments of formulation similarity caused by inconsistencies between nominal feed amounts and effective ingredient feed amounts, thereby improving the data reliability for historical adhesive formulation reuse, candidate basic formulation recommendation, and subsequent intelligent formulation design. Attached Figure Description

[0020] Figure 1 This is a schematic diagram of the structure of a multi-source feature extraction, storage and intelligent retrieval system for adhesive formulations provided in an embodiment of this application; Figure 2 This is a schematic diagram of a process for receiving multi-source formula data and merging formula identifiers provided in an embodiment of this application; Figure 3 This is a schematic diagram of a process for raw material morphology identification and effective feeding conversion provided in an embodiment of this application; Figure 4 This is a schematic diagram illustrating an effective formula ratio vector generation and associated storage process provided in an embodiment of this application; Figure 5 This is a schematic diagram of a similarity retrieval ranking and difference weighting process provided in an embodiment of this application; Figure 6 This is a schematic diagram of an interface for outputting search result difference information provided in an embodiment of this application. Detailed Implementation

[0021] The following section, with reference to the accompanying drawings, details the multi-source feature extraction, storage, and intelligent retrieval system for adhesive formulations provided in this application. First, the key terms used in the embodiments of this application are introduced.

[0022] The initial feed amount refers to the total amount of a particular raw material added to the system as recorded in the adhesive formulation record. This initial feed amount can be expressed in parts by mass, mass percentage, kilograms, grams, or other units that can represent the quantity of material added. It is understood that for pure solid resins, powder fillers, or pure active additives, the initial feed amount can be essentially equivalent to the amount of the active ingredient added; for emulsions, resin solutions, dispersions, solvent-based additives, or diluent functional components, the initial feed amount usually also includes non-active components such as water, solvents, carriers, or diluents.

[0023] The effective ingredient dosage refers to the amount of a raw material that can actually participate in the film formation, curing, tackification, plasticization, crosslinking, rheology adjustment, or other formulation functions of the adhesive. For example, for an aqueous acrylic emulsion with a solid content of 50% and an initial dosage of 100 parts, the effective ingredient dosage can be 50 parts.

[0024] An effective formulation ratio vector refers to a proportional characteristic generated based on the amount of effective components added to each raw material in an adhesive formulation. Each vector element in this effective formulation ratio vector can be used to represent the proportion of the effective component added to the total effective component added to the formulation.

[0025] The converted confidence level refers to the degree of confidence that a system can assess when determining the amount of effective component in a raw material, based on the solid content, effective component ratio, or active component ratio used. This converted confidence level can be determined based on the completeness of the raw material attribute fields, the reliability of the data source, the matching of supplier batch information, and the consistency of historical attribute data for similar raw materials.

[0026] like Figure 1 As shown, in some embodiments, a multi-source feature extraction, storage, and intelligent retrieval system for adhesive formulations may include a multi-source formulation data receiving module, a raw material morphology recognition module, an effective material input conversion module, a conversion confidence generation module, an effective proportion vector generation module, a formulation feature storage module, and a similarity retrieval and ranking module.

[0027] The multi-source formulation data receiving module can receive the original feed data and raw material attribute data of the adhesive formulation. The original feed data may include the raw material name and the original feed amount. The raw material attribute data may include the raw material form, solid content, and the proportion of active ingredients or active components. For example, if the original feed amount of a certain water-based acrylic emulsion is 100 parts and the solid content is 50%, then this solid content can be received as the raw material attribute data for that raw material.

[0028] The raw material morphology identification module can identify emulsions, resin solutions, dispersions, solvent-based additives, or diluent functional components based on raw material attribute data, and determine them as raw materials to be converted. The effective feed conversion module can determine the effective component feed amount of the corresponding raw material based on the original feed amount and the solid content, effective ingredient ratio, or active ingredient ratio of the corresponding raw material.

[0029] The confidence level generation module can generate a confidence level based on the data source of the effective proportion field, batch matching degree, or historical data of similar raw materials. The effective proportion vector generation module can generate an effective formula proportion vector based on the effective component input amount and the confidence level of each raw material. The formula feature storage module can store the original input data, raw material attribute data, effective component input amount, confidence level, and effective formula proportion vector in association.

[0030] The similarity retrieval and ranking module, upon receiving a target formula retrieval request, calculates the similarity of the original ingredient ratios and the similarity of the effective formula ratios between the target formula and historical formulas. When the original ingredient ratio similarity is higher than a first threshold and the effective formula ratio similarity is lower than a second threshold, the corresponding historical formulas are ranked with reduced weight. Therefore, the system can convert the original ingredient quantities into effective ingredient quantities before formula retrieval, and then use the effective formula ratio vector for similarity retrieval, avoiding misjudgments of formula structure similarity due to differences in the solid content of different raw materials or the proportion of effective ingredients.

[0031] like Figure 2As shown, in some embodiments, the raw material input data received by the multi-source formulation data receiving module can come from at least one of the following: experimental formulation records, production batch records, raw material supplier information, performance test reports, or historical formulation databases. The experimental formulation records can record the raw material name and mass fraction of a certain adhesive sample; the production batch records can record the actual input amount of the adhesive sample; the raw material supplier information can record the solid content of a certain emulsion or resin liquid; the performance test reports can record the viscosity, peel strength, or shear strength of the sample; and the historical formulation database can record the historical version number, formulation identifier, and usage scenario of the formulation.

[0032] The multi-source formulation data receiving module can merge data from different sources corresponding to the same adhesive formulation under the same formulation identifier. For example, when both the experimental formulation record and the production batch record correspond to formulation number A-, the system can associate the original feed amount and raw material attribute data from the two sources with the formulation record corresponding to formulation number A-. If the formulation identifiers from different sources are inconsistent, the system can merge them based on project number, sample number, experimental date, batch number, or similarity of raw material combinations. Thus, the system can avoid duplicate filing of the same adhesive formulation due to different sources and can associate the solid content field in the supplier information with the original feed amount field in the experimental formulation record, providing a data basis for subsequent calculation of the effective ingredient feed amount.

[0033] like Figure 3 As shown, in some embodiments, the raw material form recognition module can identify the raw materials to be converted in the adhesive formulation based on raw material attribute data, raw material name field, supplier information field, raw material effective proportion field, and historical similar raw material attribute data. It is understood that the raw materials to be converted refer to raw materials whose original feed amount may differ from the effective component feed amount.

[0034] For example, the raw material form recognition module can determine the raw material form according to the rules shown in Table 1: Table 1. Rules for Identifying Raw Material Morphology In the actual identification process, if the raw material attribute data already contains a clear raw material form field, the raw material form identification module can prioritize using that raw material form field for identification; if the raw material form field is missing, identification can be performed based on keywords in the raw material name field; if the raw material name field cannot be identified, identification can be performed based on solid content, effective ingredient ratio, active ingredient ratio, solvent ratio, carrier medium field, or historical data of similar raw materials.

[0035] For pure active additives, when the proportion of active component is greater than or equal to a preset purity threshold, the original feed amount can be used as the effective component feed amount; when the proportion of active component is less than the preset purity threshold, it can be calculated based on the proportion of active component. For example, the preset purity threshold can be 0.95%.

[0036] For example, if a raw material is named water-based acrylic emulsion A and the supplier's information states that its solid content is 50%, the raw material form recognition module can identify the raw material as an emulsion and determine it as a raw material to be converted. As another example, if a raw material is named thermally conductive filler B, its form field is powder, and there is no carrier medium field, the raw material form recognition module may not determine it as a raw material to be converted.

[0037] In some embodiments, the effective feed conversion module can multiply the original feed amount of the raw material to be converted by the corresponding solid content, effective ingredient ratio, or active component ratio of the raw material to be converted, to obtain the effective ingredient feed amount of the raw material to be converted. For example, if the original feed amount of a certain water-based acrylic emulsion is 100 parts and the solid content is 50%, then the effective feed conversion module can multiply 100 parts by 50% to obtain the effective ingredient feed amount of the water-based acrylic emulsion as 50 parts. As another example, if the original feed amount of a certain auxiliary agent dispersion is 2 parts and the effective ingredient ratio is 20%, then the effective feed conversion module can determine that its effective ingredient feed amount is 0.4 parts.

[0038] In some embodiments, when at least two of the fields of solid content, effective ingredient ratio and active component ratio exist simultaneously in the raw material attribute data of the same raw material, the effective feed conversion module can determine the selection priority of the effective ratio field according to the raw material form and raw material function type.

[0039] For example, the effective proportion field can be selected according to the priority shown in Table 2: Table 2 Priority Table for Selecting Effective Ratio Fields It is understandable that solid content is mainly used to indicate the proportion of non-volatile solid components in emulsions, dispersions, or slurries; effective ingredient ratio is mainly used to indicate the proportion of components with target functional effects in resin liquids, solvent-based additives, or diluent functional components; and active ingredient ratio is mainly used to indicate the proportion of components in reactive additives that can participate in crosslinking, curing, or other chemical reactions.

[0040] If there are multiple valid proportion fields for the same raw material, and the difference between different fields exceeds the preset difference threshold, the valid feed conversion module can select the valid proportion field for conversion according to the priority in Table 2, and use the unselected field as the verification field; if the difference between the verification field and the selected field exceeds the preset difference threshold, the system can reduce the conversion confidence of the raw material.

[0041] For example, if an aqueous emulsion has a solid content of 50% and an effective ingredient ratio of 48%, then since the raw material is a film-forming emulsion, the effective feed conversion module can preferentially use the solid content of 50% as the effective ratio field. As another example, if a reactive crosslinking agent has an active component ratio of 70% and an effective ingredient ratio of 80%, then since the raw material is a reactive additive, the effective feed conversion module can preferentially use the active component ratio of 70% as the effective ratio field.

[0042] In some embodiments, for raw materials not identified as raw materials to be converted, the effective proportion vector generation module can use the original feed amount of the raw material as the effective component feed amount, or convert the original feed amount of the raw material into the effective component feed amount according to a preset effective proportion; if the raw material attributes record purity, the preset effective proportion can be the purity; if the raw material attributes record the effective substance content, the preset effective proportion can be the effective substance content; if the raw material attributes do not have the above fields and the raw material form is powder filler or pure resin, the preset effective proportion can be one.

[0043] like Figure 4 As shown, in some embodiments, the effective proportion vector generation module can sum the effective component amounts of each raw material and determine the ratio of the effective component amount of each raw material to the summation value as the vector element of that raw material in the effective formulation proportion vector. For example, an adhesive formulation includes a main emulsion, a tackifying resin liquid, and a powder filler. The effective component amount of the main emulsion is fifty parts, the effective component amount of the tackifying resin liquid is twenty parts, and the effective component amount of the powder filler is thirty parts. The effective proportion vector generation module can determine that the summation value of the effective component amounts of the three is one hundred parts, and determine the vector elements corresponding to the main emulsion, the tackifying resin liquid, and the powder filler as 0.5, 0.2, and 0.3, respectively.

[0044] In some embodiments, the conversion confidence level generation module can generate a conversion confidence level based on the data source of the effective proportion field, batch matching degree, raw material morphology identification results, and historical similar raw material attribute data. The conversion confidence level is used to represent the reliability of the conversion result of the effective ingredient input amount. The conversion confidence level can take a value between zero and one; the closer the conversion confidence level is to one, the more reliable the conversion basis is; the closer the conversion confidence level is to zero, the less reliable the conversion basis is.

[0045] For example, the confidence level generation module can generate basic confidence levels according to Table 3: Table 3. Reconstructed Confidence Level Classification Table In some embodiments, similar raw materials can be determined based on at least one of the following: raw material functional category, chemical category, supplier category, raw material form, and application system. When multiple conditions are met, data with the same raw material form, functional category, and chemical category are preferred; when the number of historical similar samples is lower than a preset number, a lower basic confidence level can be used.

[0046] In some embodiments, the confidence level generation module can also adjust the base confidence level based on the dispersion of similar historical data. If the effective proportion in similar historical data fluctuates little, a higher confidence level can be maintained; if the effective proportion in similar historical data fluctuates greatly, the confidence level can be reduced.

[0047] For example, the reduced confidence level can be calculated as follows: Ci = C0i × (1 - min(σi / μi,λ)); Where Ci represents the converted confidence level of the i-th raw material; C0i represents the basic confidence level corresponding to the i-th raw material; μi represents the average effective proportion of similar raw materials in history; σi represents the standard deviation of the effective proportion of similar raw materials in history; and λ represents the maximum correction coefficient, which can take the value of 0.5.

[0048] If the effective proportion field of a certain raw material comes from the same supplier and the same batch of data, then its basic confidence level is one. If the field comes from the historical average of similar raw materials, and the historical average effective proportion of similar raw materials is 0.5 and the standard deviation is 0.05, then: Ci = 0.60 × (1 - min(0.05 / 0.50, 0.50)) = 0.54; The similarity search and ranking module can adjust the weight of the corresponding raw material in the formula similarity calculation based on the converted confidence level. For example, if the base weight of the i-th raw material is Wi and the converted confidence level is Ci, then the adjusted weight Wi' can be determined as follows: Wi' = Wi × Ci; After the weights of each raw material are adjusted, the similarity search and ranking module can normalize the adjusted weights so that the sum of all weights involved in the calculation is one. Thus, when the effective proportion field of a certain raw material is unreliable, the raw material can still participate in the search calculation, but its impact on the formula similarity will be reduced.

[0049] The formulation feature storage module can associate and store the original feed amount, active ingredient feed amount, solid content, active ingredient ratio, active component ratio, conversion confidence level, and data source identifier for the same raw material. For cases where the solid content of the same raw material varies in different batches, the formulation feature storage module can store the corresponding raw material attribute data and conversion results separately according to the batch identifier.

[0050] In some embodiments, the similarity retrieval and ranking module can calculate the effective formula ratio similarity based on the effective formula ratio vector corresponding to the target formula and the effective formula ratio vector corresponding to the historical formula. The effective formula ratio vector corresponding to the target formula can be represented as P=(p1,p2,...,pn), and the effective formula ratio vector corresponding to the historical formula can be represented as Q=(q1,q2,...,qn). Here, pi represents the effective formula ratio of the i-th raw material in the target formula, qi represents the effective formula ratio of the i-th raw material in the historical formula, and n represents the number of raw materials involved in the comparison; if a certain raw material exists only in the target formula or the historical formula, the corresponding effective formula ratio in the other formula can be recorded as zero.

[0051] The similarity of effective formulation ratios can be calculated using the following formula: Se = 1 - 1 / 2 × Σ(Wi' × |pi - qi|); Where Se represents the similarity of effective formulation ratios; Wi' represents the adjusted weight corresponding to the i-th raw material; pi represents the effective formulation ratio of the i-th raw material in the target formulation; and qi represents the effective formulation ratio of the i-th raw material in the historical formulation.

[0052] In other embodiments, the similarity retrieval and ranking module may also use cosine similarity to calculate the similarity of effective formula proportions. In this embodiment, the weighted absolute difference formula is preferred because it can directly reflect the difference in the effective formula proportions of each raw material and facilitates the adjustment of weights by combining the converted confidence level.

[0053] like Figure 5 As shown, in some embodiments, the similarity retrieval and ranking module can also calculate the original ingredient ratio similarity and the effective ingredient ratio similarity between the target formula and the historical formula, respectively. The original ingredient ratio similarity is used to represent the degree of closeness between the two formulas at the nominal ingredient ratio level, while the effective ingredient ratio similarity is used to represent the degree of closeness between the two formulas at the effective ingredient ratio level.

[0054] The similarity of the original feed ratios can be calculated using the following formula: So = 1 - 1 / 2 × Σ(Ui × |ai - bi|); Where So represents the similarity of the original ingredient ratios; ai represents the original ingredient ratio of the i-th ingredient in the target formula; bi represents the original ingredient ratio of the i-th ingredient in the historical formula; and Ui represents the weight of the i-th ingredient in the calculation of the similarity of the original ingredient ratios.

[0055] When the similarity search and ranking module determines that the similarity of the original feed ratio is higher than the first threshold and the similarity of the effective formula ratio is lower than the second threshold, it means that the target formula and the historical formula are similar in terms of nominal feed ratio, but there is a significant difference in terms of effective ingredient ratio. At this time, the similarity search and ranking module can sort the historical formula by lower weight.

[0056] For example, the first threshold can be set to 0.85, and the second threshold can be set to 0.75. The similarity retrieval ranking module can calculate the weighting coefficient according to the following formula: D=1-η(So-Se); Where D represents the weighting factor; η represents the weighting intensity factor, which can range from 0.2 to 0.8; So represents the similarity of the original feed ratio; and Se represents the similarity of the effective formula ratio. To avoid excessive weighting, the system can also set a minimum weighting factor Dmin; when D calculated by the above formula is less than Dmin, D can be set to Dmin. For example, Dmin can be 0.6.

[0057] The final ranking score for historical recipes can be calculated using the following formula: F = F0 × D; Where F represents the final ranking score of the historical formula; F0 represents the basic ranking score before the weighting of the difference between the original similarity and the effective similarity; D represents the weighting coefficient; the basic ranking score F0 can be the effective formula ratio similarity, or the weighted result of the effective formula ratio similarity, performance index similarity and application scenario matching degree; in the embodiment that only uses the effective formula ratio similarity, the basic ranking score F0 is equal to the effective formula ratio similarity; if the condition that the original feed ratio similarity is higher than the first threshold and the effective formula ratio similarity is lower than the second threshold is not met, the similarity retrieval ranking module can set D to one, that is, not perform difference weighting on the historical formula.

[0058] like Figure 6 As shown, in some embodiments, the similarity search and ranking module can output effective formula ratio difference information between the target formula and historical formulas when outputting the adhesive formula search and ranking results. This effective formula ratio difference information may include differences in the original feed amount of at least one raw material, differences in the feed amount of the active ingredient, and differences in the effective formula ratio.

[0059] The following section uses a water-based adhesive formulation retrieval scenario to illustrate the effective material input conversion, effective formulation ratio vector generation, effective formulation ratio similarity calculation, and difference weighting process provided in the embodiments of this application.

[0060] Assume the target formulation T includes three raw materials: aqueous acrylic emulsion A, tackifying resin liquid B, and powder filler C; wherein, the initial feed amount of aqueous acrylic emulsion A is 100 parts, and the solid content is 50%; the initial feed amount of tackifying resin liquid B is 40 parts, and the effective ingredient ratio is 50%; the initial feed amount of powder filler C is 30 parts, and the raw material form is powder filler, which does not need to be converted.

[0061] Historical formula H includes the same three raw materials: water-based acrylic emulsion A, tackifying resin liquid B, and powder filler C; among them, the original feed amount of water-based acrylic emulsion A is 100 parts, with a solid content of 35%; the original feed amount of tackifying resin liquid B is 40 parts, with an effective ingredient ratio of 50%; the original feed amount of powder filler C is 30 parts, and the raw material form is powder filler, which does not need to be converted.

[0062] The effective ingredient dosage of target formulation T is calculated as follows: the effective ingredient dosage of waterborne acrylic emulsion A is 100 × 0.50 = 50, the effective ingredient dosage of tackifying resin liquid B is 40 × 0.50 = 20, and the effective ingredient dosage of powder filler C is 30 × 1.00 = 30. The sum of the effective ingredient dosages of target formulation T is 50 + 20 + 30 = 100. Therefore, the effective formulation ratio vector of target formulation T is P = (0.50, 0.20, 0.30).

[0063] The effective ingredient dosage of historical formula H is calculated as follows: the effective ingredient dosage of waterborne acrylic emulsion A is 100 × 0.35 = 35, the effective ingredient dosage of tackifying resin liquid B is 40 × 0.50 = 20, and the effective ingredient dosage of powder filler C is 30 × 1.00 = 30. The sum of the effective ingredient dosages of historical formula H is 35 + 20 + 30 = 85. Therefore, the effective formula ratio vector of historical formula H is Q = (0.412, 0.235, 0.353).

[0064] Since the target formulation T and the historical formulation H have the same three raw materials, and the conversion of each raw material is based on supplier data, assuming that the weights are equal after adjustment, i.e., W1'=W2'=W3'=1 / 3, the similarity of the effective formulation ratio is Se=1-1 / 2×(1 / 3×|0.50-0.412|+1 / 3×|0.20-0.235|+1 / 3×|0.30-0.353|)=0.97065. This result shows that although the emulsion solid content in the historical formulation H is lower than that in the target formulation T, the overall effective formulation ratio in this three-component system still has a high degree of similarity.

[0065] If we further compare the original ingredient ratios, the original ingredient quantities of the target formula T and the historical formula H are both 100 parts, 40 parts, and 30 parts, respectively, so their original ingredient ratios are completely consistent. At this point, the similarity of the original ingredient ratios is So=1. If the first threshold is 0.85 and the second threshold is 0.75, then since Se=0.97065 is higher than the second threshold, the system does not trigger difference weighting, and the weighting coefficient D is one.

[0066] To illustrate the triggering process of differential weighting, we further assume that the solid content of the aqueous acrylic emulsion A in another historical formulation H2 is 20%, and the tackifying resin liquid B and powder filler C are the same as the target formulation. Then, the effective ingredient feed amounts of historical formulation H2 are 20 parts, 20 parts, and 30 parts, respectively. The sum of the effective ingredient feed amounts is 70 parts, and the effective formulation ratio vector is Q2=(0.286,0.286,0.428).

[0067] If this embodiment aims to more sensitively identify the effective ratio differences of the emulsion-type main resin, the adjusted weight of the main emulsion A can be increased; for example, setting W1'=0.70, W2'=0.15, W3'=0.15, then the similarity of the effective formulation ratio of historical formulation H2 is Se2=1-1 / 2×(0.70×|0.50-0.286|+0.15×|0.20-0.286|+0.15×|0.30-0.428|)=0.90905.

[0068] If the system sets the second threshold to 0.92, then historical formula H2 satisfies the condition that the similarity of the original feed ratio is higher than the first threshold and the similarity of the effective formula ratio is lower than the second threshold. Assuming the weighting intensity coefficient η is 0.5, then the weighting coefficient is D = 1 - 0.5 × (1 - 0.90905) = 0.954525. If the basic ranking score F0 of historical formula H2 is 0.9, then the final ranking score is F = 0.9 × 0.954525 = 0.8590725.

[0069] As demonstrated by the above examples, even if the target formula and historical formulas are identical in terms of original feed amounts, differences in emulsion solid content can lead to discrepancies in the effective formula ratio vector. By performing effective feed conversion, generating effective formula ratio vectors, calculating the similarity of effective formula ratios, and weighting differences, the system can identify historical formulas with the same nominal feed amount but different effective ingredient ratios, and adjust the search ranking results accordingly. This reduces misjudgments in similarity searches for aqueous emulsion, resin liquid, dispersion, or solvent-based adhesive formulas.

[0070] In summary, this application provides a multi-source feature extraction, storage, and intelligent retrieval system for adhesive formulations. The system receives original feed data and raw material attribute data of the adhesive formulation through a multi-source formulation data receiving module, identifies the raw materials to be converted through a raw material morphology identification module, converts the original feed amount into effective component feed amount through an effective feed conversion module, generates conversion confidence level through a conversion confidence level generation module, generates effective formulation ratio vector through an effective ratio vector generation module, and outputs adhesive formulation retrieval ranking results through a similarity retrieval ranking module based on the effective formulation ratio vector, conversion confidence level, and the difference between the original feed ratio and the effective formulation ratio. Because this system converts the effective components of emulsions, resins, dispersions, solvent-based additives, or diluent functional components before formulation retrieval, it can improve the authenticity of the adhesive formulation structural characteristics and reduce retrieval misjudgments caused by similar nominal feed amounts but different effective component ratios.

[0071] In the embodiments of this application, the terms "first," "second," and "third" are used for descriptive purposes only and should not be construed as indicating or implying relative importance. The term "at least one" refers to one or more, and "multiple" refers to two or more.

[0072] The above description is merely an optional implementation of this application, but the scope of protection of this application is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in this application, and these modifications or substitutions should all be covered within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A multi-source feature extraction, storage, and intelligent retrieval system for adhesive formulations, characterized in that, The system includes a multi-source formula data receiving module, a raw material morphology recognition module, an effective feeding conversion module, an effective proportion vector generation module, and a similarity retrieval and sorting module. The multi-source formulation data receiving module is used to receive the original feeding data and raw material attribute data of the adhesive formulation. The original feeding data includes the raw material name and the original feeding amount, and the raw material attribute data includes the raw material form, solid content, and proportion of effective ingredients or active components. The raw material morphology identification module is used to identify raw materials that need to be converted based on the raw material attribute data, where there is a difference between the original feed amount and the effective ingredient feed amount. The effective feed conversion module is used to determine the effective feed amount of the raw material to be converted based on the original feed amount of the raw material to be converted and the corresponding solid content, effective component ratio or active component ratio; The effective proportion vector generation module is used to generate an effective formula proportion vector based on the amount of effective components added to each raw material. The similarity retrieval and ranking module is used to calculate the formula similarity based on the effective formula ratio vector between the target formula and historical formulas after receiving a target formula retrieval request, and output the historical formula retrieval and ranking results based on the formula similarity.

2. The multi-source feature extraction, storage, and intelligent retrieval system for adhesive formulations according to claim 1, characterized in that, The original feed data received by the multi-source formula data receiving module comes from at least one of the following: experimental formula records, production batch records, raw material supplier information, performance test reports, or historical formula database. The multi-source formulation data receiving module is also used to merge data from different sources corresponding to the same adhesive formulation under the same formulation identifier.

3. The multi-source feature extraction, storage, and intelligent retrieval system for adhesive formulations according to claim 1, characterized in that, The raw material form recognition module is used to determine the raw material form based on the raw material form field when the raw material attribute data contains the raw material form field. When the raw material form field is missing, the raw material form identification module is used to determine the raw material form based on the raw material name field, solid content field, effective ingredient ratio field, active component ratio field, solvent ratio field, carrier medium field, or historical similar raw material attribute data.

4. The multi-source feature extraction, storage, and intelligent retrieval system for adhesive formulations according to claim 1, characterized in that, The raw materials to be converted include at least one of emulsions, resin solutions, dispersions, solvent-based additives, and diluent functional components; The emulsion includes a film-forming emulsion or an aqueous dispersion emulsion; the resin liquid includes a resin solution or a liquid resin; the dispersion includes an additive dispersion, a filler dispersion, or a functional slurry; and the diluent functional component includes a carrier-containing thickener, plasticizer, crosslinking agent, or rheology modifier.

5. The multi-source feature extraction, storage, and intelligent retrieval system for adhesive formulations according to claim 1, characterized in that, When the same raw material has at least two of the fields of solid content, effective ingredient ratio and active component ratio, the effective feed conversion module is used to select the effective ratio field for conversion according to the raw material type; Among them, for film-forming emulsions, the solid content should be prioritized; for resin liquids, dispersions, solvent-based additives or diluent functional components, the proportion of effective ingredients should be prioritized; and for reactive additives, the proportion of active ingredients should be prioritized.

6. The multi-source feature extraction, storage, and intelligent retrieval system for adhesive formulations according to claim 5, characterized in that, The effective feed conversion module is used to combine the original feed amount of the raw material to be converted with the corresponding solid content, effective ingredient ratio or active ingredient ratio of the raw material to be converted to obtain the effective ingredient feed amount of the raw material to be converted.

7. The multi-source feature extraction, storage, and intelligent retrieval system for adhesive formulations according to claim 1, characterized in that, For raw materials not identified as raw materials to be converted, the effective proportion vector generation module is used to take the original amount of the raw material as the amount of effective ingredient, or to convert the original amount of the raw material into the amount of effective ingredient according to a preset effective proportion, and to incorporate the raw materials not identified as raw materials to be converted into the effective formula proportion vector.

8. The multi-source feature extraction, storage, and intelligent retrieval system for adhesive formulations according to claim 1, characterized in that, The system also includes a confidence level generation module; The conversion confidence generation module is used to generate conversion confidence based on the data source of the effective proportion field, the batch matching degree, the raw material morphology identification result, or historical similar raw material attribute data; The similarity retrieval and ranking module is used to adjust the weight of the corresponding raw materials in the formula similarity calculation based on the converted confidence level.

9. The multi-source feature extraction, storage, and intelligent retrieval system for adhesive formulations according to claim 1, characterized in that, The effective proportion vector generation module is used to sum the effective component input amounts of each raw material, and to determine the ratio of the effective component input amount of each raw material to the summation value as the vector element of that raw material in the effective formula proportion vector.

10. The multi-source feature extraction, storage, and intelligent retrieval system for adhesive formulations according to any one of claims 1 to 9, characterized in that, The similarity retrieval and sorting module is used to calculate the similarity of effective formula ratios based on the effective formula ratio vectors of the target formula and historical formulas, and to calculate the similarity of original feed ratios based on the original feed ratio vectors of the target formula and historical formulas. When the similarity of the original feed ratio is higher than the first threshold and the similarity of the effective formula ratio is lower than the second threshold, the similarity retrieval and ranking module determines the weighting coefficient based on the difference between the similarity of the original feed ratio and the similarity of the effective formula ratio, and corrects the ranking score of the historical formula based on the weighting coefficient.

11. The multi-source feature extraction, storage, and intelligent retrieval system for adhesive formulations according to any one of claims 1 to 9, characterized in that, The similarity search and sorting module is used to simultaneously output the effective formula ratio difference information between the target formula and historical formulas, as well as the conversion basis of the corresponding raw materials, when outputting the adhesive formula search and sorting results. The effective formula ratio difference information includes at least one difference in the original amount of raw material, the difference in the amount of effective ingredient, and the difference in the effective formula ratio; The historical formula retrieval and ranking results are used as candidate basic formula inputs for the adhesive intelligent formula generation platform. The candidate basic formula inputs include the effective ingredient dosage of the historical formula, the effective formula ratio vector, and the conversion basis of the corresponding raw materials.

Images