Ink multi-element titanium dioxide compounding system based on big data analysis

Abstract

The present application belongs to the technical field of industrial data collection, and particularly relates to information retrieval, database indexing and big data analysis technology, and especially to an ink multi-element titanium dioxide compounding system based on big data analysis. The system receives titanium dioxide raw material inspection data, ink system conditions, compounding ratio data and ink performance results, maps them into a raw material particle size layer, a compounding combination layer, an ink system layer and a performance result layer, generates a compounding fingerprint containing a single raw material fingerprint, a component interaction fingerprint and a system context fingerprint, and retrieves historical compounding records through rule indexing, inverted indexing and vector indexing. The system further resolves conflict samples according to detection conditions, system context distance and performance deviation, and outputs candidate compounding records, key difference fields and compatibility confidence levels. The present application can improve the comparability and traceability of historical compounding data, and reduce the false matching caused by detection caliber and system differences.

Description

Technical Field

[0001] This invention pertains to industrial data acquisition, specifically involving information retrieval, database indexing, and big data analysis technologies, particularly a multi-component titanium dioxide compounding system for inks based on big data analysis. Background Technology

[0002] Titanium dioxide blending for inks involves adding titanium dioxide of different types, surface treatments, or particle size distributions to the same ink system in specific proportions to achieve a balance between hiding power, whiteness, tinting strength, gloss, fineness, viscosity stability, and sedimentation properties. In existing technologies, the production or experimental stages typically begin with a raw material inspection report recording indicators such as crystal form, whiteness, oil absorption, particle size distribution, dispersion grade, and pH. Then, a formulation test table records the type of binder, dispersant, pigment volume fraction range, blending ratio, grinding conditions, and ink formation test results. Conventional information systems often use relational tables or formulation management databases to store this data, using titanium dioxide type, batch number, applicable ink type, and several key performance indicators as search fields. When the system receives a new blending request, it typically filters by target ink type, target hiding power range, target viscosity range, or historical model name. Technical personnel then review similar historical formulations and select several proportion points for retesting. The implementation logic of this type of solution mainly relies on record archiving, field querying, and manual comparison. The relationships between data are primarily reflected in model correspondence or single indicator correspondence. For indicators with the same name generated by different testing methods, the system usually only performs unit conversion or provides remarks, rarely processing how the testing conditions themselves participate in subsequent retrieval calculations.

[0003] In more conventional technical solutions, some systems add big data analysis capabilities to historical formula databases. This involves cleaning, deduplicating, converting units, and annotating outliers on multiple batches of raw material testing data and ink performance results. Candidate blending ratios are then recommended using statistical regression, similarity calculations, or clustering methods. These solutions typically treat each blending record as a sample, using a feature vector composed of titanium dioxide type, several raw material indicators, blending ratio, and target performance. Similar samples are retrieved based on distance, correlation coefficient, or weighted scoring. Other solutions set performance tags in the database, such as high opacity, high gloss, low viscosity, or easy dispersibility, and use tag combinations to narrow down the candidate pool. While these solutions can perform basic data retrieval and ratio recommendations within an empirical range, their feature construction is often centered on individual raw material indicators and the final ink result. Differences in indicators between blending components, indicator ratios, primary and secondary relationships, surface treatment consistency, and the context of the ink system are usually not treated as independent data objects. Differences in detection methods, missing fields, and variations in detection conditions are often simplified into unified fields, and the system defaults to direct comparison of samples from different sources. The resulting search results reflect similarity at the field level more than at the level of complex relationships.

[0004] Data on multi-component titanium dioxide blends for inks exhibits component interactions, system dependencies, and variations in testing conditions. Conventional database searches or ordinary similarity analyses cannot uniformly express raw material particle size information, blending relationships, ink system conditions, and ink-forming performance results into calculable and traceable blending characteristics. This leads to the system's inability to reliably determine whether samples are truly comparable when retrieving historical blending records. Specifically, two titanium dioxide blends may be similar in individual indicators such as whiteness, oil absorption, or particle size, but due to differences in surface treatment or the main-component relationship in the blending ratio, they may exhibit different dispersibility and rheological results in the same ink system. Other blends may have significant differences in individual raw material indicators, but produce similar ink-forming performance due to component complementarity. Existing systems, when searching by model, label, or single vector distance, are prone to mixing historical records with different detection calibers, different system contexts, or conflicting performance into the candidate results. The essence of the problem is not the lack of historical data, but the lack of indexes that can simultaneously constrain raw materials, combinations, systems, and detection conditions. As a result, it is difficult to provide candidate records with comparable boundaries for new compounding tasks, making subsequent proportional verification based on unstable samples. Summary of the Invention

[0005] The purpose of this invention is to provide an ink multi-component titanium dioxide compounding system based on big data analysis, which can solve the problems mentioned in the background art.

[0006] To achieve the above objectives, the technical solution adopted by the present invention is as follows: The ink multi-component titanium dioxide compounding system based on big data analysis includes a data access module, a hierarchical modeling module, a compounding fingerprint construction module, a composite index module, and a candidate output module. The data access module is used to receive titanium dioxide raw material inspection data, ink system conditions, compounding ratio data, and ink forming performance results. The layered modeling module is used to map the received data into a raw material particle size layer, a compound combination layer, an ink system layer, and a performance result layer. The composite fingerprint construction module is used to generate composite fingerprints that include monomer raw material fingerprints, component interaction fingerprints, and system context fingerprints. The composite index module is used to retrieve historical composite records based on rule indexes, inverted indexes, and vector indexes; The candidate output module is a system used to output candidate titanium dioxide compound records corresponding to the target ink performance constraints.

[0007] Preferably, the hierarchical modeling module is used to establish a field semantic table and a detection method coding table for the titanium dioxide raw material inspection data, mapping the name, unit, sampling batch and test conditions of the same indicator in different test reports to a unified field, and attaching data source tags to missing fields, conversion fields and cross-method fields respectively; The raw material particle size layer records at least the crystal form, surface treatment type, particle size distribution, oil absorption, whiteness, tinting strength, pH value, and dispersion grade. The ink system layer records at least the binder type, dispersant type, pigment volume fraction range, and target application performance. The performance result layer records at least the hiding power, fineness, gloss, viscosity stability, and sedimentation trend.

[0008] Preferably, the compound fingerprint construction module is used to extract monomer index sequences from the raw material particle size layer, extract the proportional relationship, index difference, index ratio and surface treatment consistency mark between different titanium dioxide from the compound combination layer, extract system context constraints from the ink system layer, and encode the monomer index sequences, inter-component interaction features and system context constraints into compound fingerprints according to the preset field order; The matching fingerprint corresponding to the same historical matching record is simultaneously associated with the ink performance result, detection condition marker and matching record source identifier, and a gap marker is generated for historical matching records that lack any component interaction field.

[0009] Preferably, the composite index module is used to establish a three-level retrieval path for the compound fingerprint, wherein the rule index is used to filter historical compound records according to ink system compatibility conditions and target application performance, the inverted index is used to recall candidate records according to titanium dioxide model, surface treatment type, key indicator range and performance label, and the vector index is used to recall compound records according to the joint similarity of monomer raw material fingerprint, component interaction fingerprint and system context fingerprint; The search results of the three-level search path are merged by matching record identifiers, and the hit fields corresponding to each search path are retained.

[0010] Preferably, the candidate output module is used to perform conflict sample resolution on the historical compound records recalled by the composite index module. The conflict sample resolution includes classifying historical compound records with similar titanium dioxide raw material indicators but inconsistent ink performance results into a first conflict group, classifying historical compound records with similar ink performance results but different raw material indicators exceeding the historical distribution of the same type into a second conflict group, and generating sorting information, key difference fields and compatibility confidence levels of candidate titanium dioxide compound records based on detection condition markers, system context distance and performance deviation fields, and writing the conflict group type into the retrieval log of candidate titanium dioxide compound records.

[0011] Preferably, the hierarchical modeling module is further used to establish a detection condition transformation graph, wherein the nodes of the detection condition transformation graph correspond to the detection methods in the detection method coding table, and the edges correspond to the comparable fields generated by the same batch of titanium dioxide under different detection methods; When new raw material inspection data is received, the fields that can be directly compared, the fields that need to be converted, and the fields that are not comparable are determined according to the test condition conversion diagram. The fields that need to be converted are written into the converted index set, and the fields that are not comparable are written into the isolation index set. The composite fingerprint construction module references the transformed indicator set when generating composite fingerprints, and only generates record labels that do not involve similarity for the isolated indicator set.

[0012] Preferably, the compound fingerprint construction module is further used to establish a directed interaction code for the interaction features between the components. The directed interaction code starts with the main contributing titanium dioxide in the target ink system and ends with the auxiliary compensating titanium dioxide, respectively recording the relative change direction of the masking related index, the dispersion related index and the rheological related index. When the compound ratio changes, the directed interactive code is regenerated according to the order of components before and after the ratio change, and the codes corresponding to different component orders are written into the compound index module as different compound fingerprints. For the same two-component combination, the corresponding codes are retained when the main and auxiliary orders are reversed.

[0013] Preferably, the composite index module is further used to establish a composite record relationship graph, wherein the nodes of the composite record relationship graph are historical composite records, the edges are the raw material similarity relationship, performance similarity relationship or system context similarity relationship between composite fingerprints, and the corresponding set of similar fields and the source of detection conditions are recorded for each edge; When performing conflict sample resolution, the candidate output module locates the intermediary record that connects the first conflict group and the second conflict group along the re-matching record relationship graph, and adds the detection condition mark, re-matching ratio difference and performance deviation fields of the intermediary record to the candidate sorting information.

[0014] Preferably, it also includes an index incremental update module, which is used to determine whether the new compounding experimental data contains existing titanium dioxide models, existing ink system conditions and newly added component interaction characteristics after the new compounding experimental data is written. When new component interaction features are included, a temporary compound fingerprint is generated for the new compound experimental data and written into the verification index area; When the new composite experimental data and the records in the index area to be verified form a stable correspondence in the ink forming performance results, the temporary composite fingerprint is converted into a formal composite fingerprint, and the detection condition transformation graph or directed interactive coding is updated simultaneously.

[0015] Preferably, it also includes a retrieval request correction module, which is used to receive the target ink performance constraints and the list of available titanium dioxide, determine the fields in the target ink performance constraints that lack reference bases according to the compounding record relationship diagram, formal compounding fingerprint and candidate sorting information, convert the fields that lack reference bases into fields to be confirmed, and generate a retrieval request containing fields to be confirmed, comparable fields and prohibited extrapolation fields before rule index filtering; The composite index module performs similarity retrieval only on comparable fields and uses fields that are prohibited from extrapolation as filtering conditions.

[0016] Compared with the prior art, the beneficial effects of the present invention are as follows: 1. This invention divides titanium dioxide raw material inspection data, compounding ratio data, ink system conditions, and ink-forming performance results into a raw material particle size layer, a compounding combination layer, an ink system layer, and a performance result layer. It generates compound fingerprints containing individual raw material fingerprints, component interaction fingerprints, and system context fingerprints. Historical compounding records can participate in retrieval and comparison using a unified data structure. The rule index first filters by ink system compatibility conditions and target application performance. The inverted index then recalls records by model, surface treatment type, key indicator range, and performance tag. The vector index further calculates joint similarity relationships based on the compound fingerprints, thus ensuring that the process of obtaining candidate records is simultaneously constrained by raw materials, combinations, and system conditions. Conflict sample resolution groups records with similar raw material indicators but inconsistent ink-forming results, and records with similar ink-forming results but large differences in raw material indicators. These groups are then combined with detection condition labels, system context distance, and performance deviation fields to form ranking information, key difference fields, and compatibility confidence levels. Hit fields from each retrieval path are retained along with the candidate records, facilitating the differentiation of the sources of rule hits, tag hits, and fingerprint similarity hits. Therefore, the search results can exclude records that lack compatibility with the target ink system and retain the source of difference for samples with contradictory performance, reducing the probability of incomparable historical records entering the candidate results.

[0017] 2. The detection method coding table, field semantic table, and detection condition transformation diagram ensure that raw material inspection data from different sources undergo field normalization, transformation relationship marking, and incomparable field isolation before entering the compound fingerprint, reducing the interference of detection caliber differences on similarity calculation. Directed interactive coding incorporates the relative change direction between the main contributing titanium dioxide and the auxiliary compensating titanium dioxide into the component interactive fingerprint, enabling the same two-component combination to form different data expressions when the compounding order and ratio change. The compound record relationship diagram uses historical compound records as nodes to record the similarity of raw materials, performance, and system context, and can introduce detection conditions and performance deviation information from intermediate records during candidate ranking. The incremental update module distinguishes between temporary and formal compound fingerprints after new experimental data is written, keeping the newly added component interactive features in a pending verification state before forming a stable correspondence, avoiding direct changes to the formal index from a single experimental data. The retrieval request correction module converts fields lacking a reference basis into fields to be confirmed and uses prohibited extrapolation fields as filtering conditions, ensuring that the compound data retrieval process remains consistent with the range of verified data, and preventing unverified fields from directly participating in similarity calculation. The above data processing establishes a consistent referencing relationship between the matching fingerprints, index records, and retrieval requests. Attached Figure Description

[0018] Figure 1 This is a flowchart of the overall data processing of the system. Figure 2 Flowchart for raw field standardization and detection condition processing; Figure 3 Flowchart for generating composite fingerprints; Figure 4 Flowchart for multi-level index retrieval and conflict sample handling; Figure 5 Flowchart for incremental index update and retrieval request correction. Detailed Implementation

[0019] refer to Figure 1 In one embodiment, a multi-element titanium dioxide compounding system for inks based on big data analysis is used for unified data processing of ink formulation development, compounding experiment records, and ink formation test records. The system includes a data access module, a hierarchical modeling module, a compounding fingerprint construction module, a composite index module, and a candidate output module. The data access module receives titanium dioxide raw material inspection data, ink system conditions, compounding ratio data, and ink formation performance results. Titanium dioxide raw material inspection data includes fields such as crystal form, surface treatment type, particle size distribution, oil absorption, whiteness, tinting strength, pH, and dispersion grade; ink system conditions include binder type, dispersant type, pigment volume fraction range, and target application performance; compounding ratio data represents the proportion of multiple titanium dioxide components in the same ink system; and ink formation performance results include hiding power, fineness, gloss, viscosity stability, and sedimentation trend. The data access module performs preliminary parsing of field names, units, and batch identifiers in the source files and assigns a unique compounding record identifier to each record, enabling subsequent modules to use the same compounding record as a data association benchmark. In cases where the same compound record contains multiple titanium dioxides, the system records each titanium dioxide as a component and writes the component and compound ratio together into the compound combination layer.

[0020] In this embodiment, the hierarchical modeling module maps the received data into a raw material particle size layer, a compound combination layer, an ink system layer, and a performance result layer. The raw material particle size layer is used to express the inspection attributes of a single titanium dioxide component; the compound combination layer is used to express the proportional relationship and index differences between different titanium dioxide components; the ink system layer is used to express the ink system boundary where the compound record is located; and the performance result layer is used to express the corresponding ink formation detection output. For the compound ratio, the system uses a proportionality conservation constraint to represent the ratio relationship of each titanium dioxide component within the same compound record, calculated as follows: Where r represents the compounding record, k represents the k-th titanium dioxide component in the compounding record, and m r p represents the amount of titanium dioxide component in the compound record r. r,k This represents the mass percentage of the k-th component in the total amount of multi-component titanium dioxide. The physical meaning of this formula is: the sum of the proportions of each titanium dioxide component within the same ink system equals the overall proportion of the multi-component titanium dioxide addition. For example, if a compound contains three types of titanium dioxide with mass percentages of 0.5, 0.3, and 0.2 respectively, then m... r=3), p r,1 =0.5, p r,2 =0.3, p r,3 =0.2, substituting it into the equation, we get 0.5 + 0.3 + 0.2 = 1, which satisfies the proportionality constraint. If the entered data is 50, 30, and 20, the system will convert it to 0.5, 0.3, and 0.2 after unit recognition before writing it into the composite layer.

[0021] Furthermore, the compound fingerprint construction module generates compound fingerprints containing monomer raw material fingerprints, component interaction fingerprints, and system context fingerprints based on the hierarchical modeling results. The monomer raw material fingerprint corresponds to the normalized index sequence of a single titanium dioxide component; the component interaction fingerprint corresponds to the proportional differences, index differences, index ratios, and surface treatment consistency among multiple titanium dioxide components; and the system context fingerprint corresponds to the ink system boundary and target application performance. The composite index module retrieves historical compound records based on rule indexes, inverted indexes, and vector indexes. The rule index is used to exclude records with incompatible ink system conditions; the inverted index is used for candidate recall by model, surface treatment type, key index range, and performance tag; and the vector index is used to recall compound records based on the joint similarity of the compound fingerprints. The candidate output module merges, sorts, and processes conflicting samples from the recalled records and outputs candidate titanium dioxide compound records corresponding to the target ink performance constraints. This embodiment, through the above data flow method, transforms historical compound records from scattered fields into data objects with hierarchical structure and fingerprint expression, providing a definite data foundation for subsequent indexing and conflict discrimination.

[0022] refer to Figure 2 In one embodiment, the hierarchical modeling module is configured with a field semantic table and a detection method coding table. The field semantic table is used to map synonymous fields from different data sources to a unified field. For example, oil absorption value, oil absorption amount, and oil absorption quantity are mapped to a unified oil absorption amount field, and whiteness value, Hunter whiteness, is mapped to a unified whiteness field. The detection method coding table is used to record the detection method, sampling batch, test conditions, and source file identifier corresponding to each inspection field. For data obtained under different detection methods for the same field, the system does not directly merge them into a single sourceless value, but retains the detection method code after unifying the field. For missing fields, the system writes a missing source marker; for fields obtained through unit conversion or method conversion, the system writes a conversion field marker; for cross-method fields that cannot establish a comparable relationship, the system writes a cross-method isolation marker. Fields in the raw material particle size layer, ink system layer, and performance result layer can be stored as shown in Table 1. The table only shows an example of an implementable data structure, and the actual number of fields can be expanded as the content of the compounded records increases.

[0023] Table 1. Example of data layer field mapping Raw material particle size layer Oil absorption Oil absorption value, oil absorption capacity Unified field and reserved method encoding Monomer raw material fingerprint Raw material particle size layer Particle size distribution Average particle size, particle size range Unified fields, preserved distribution sources Monomer raw material fingerprint Composite layer proportional relationship Proportioning, weight Converted to mass percentage Component Interaction Fingerprint Ink system layer Connector Material Category Resin type, binder type System context fields System context fingerprint Performance Result Layer viscosity stability Viscosity change, storage viscosity Performance result field Candidate sorting field In this embodiment, the system normalizes the continuous test field to allow fields with different dimensions to participate in subsequent similarity calculations. The normalization calculation is as follows: Where, x r,k,f L represents the original test value of the k-th titanium dioxide component in the compound record r on field f. f U represents the lower bound of field f in comparable historical samples. f This represents the upper bound of field f in comparable historical samples. This represents the normalized field value. The physical meaning of this formula is to map raw material indicators with different value ranges to a unified scale, facilitating the formation of composite fingerprints. For example, if the lower bound of the oil absorption field in a certain class of comparable samples is 10 and the upper bound is 30, and the oil absorption of a certain titanium dioxide component is 18, then the corresponding normalized value is... When the upper and lower bounds come from historical samples encoded by the same detection method, the normalized value participates in the similarity calculation; when a field is labeled as a cross-method isolated field, the field is only saved as a record label and is not included in the vector index calculation.

[0024] Preferably, the hierarchical modeling module also establishes a detection condition transformation graph. The nodes of the detection condition transformation graph correspond to the detection methods in the detection method coding table, and the edges correspond to comparable fields generated by different detection methods for the same batch of titanium dioxide. After receiving new raw material inspection data, the system classifies the fields into directly comparable fields, fields requiring transformation, and incomparable fields according to the detection condition transformation graph. Directly comparable fields are written into a unified field and then normalized; fields requiring transformation are calculated based on the transformation edges to obtain the transformed indicators; incomparable fields are written into an isolated indicator set. The calculation formula for the transformed indicators is as follows: Where, x r,k,f,a This represents the original value of the k-th component in the compound record r obtained in field f using detection method a. This represents the field value after conversion to the detection method b caliber, a f,a,b This represents the ratio coefficient by which field f is converted from detection method a to detection method b, where b is the ratio coefficient. f,a,bThis represents the corresponding offset term. The physical meaning of this formula is: using the linear transformation relationship formed by comparable samples in the same batch, the same field under different detection methods is brought into the same comparison caliber. For example, if the original value of a field under method a is 22, the conversion ratio to method b is 0.9, and the offset term is 1.5, then the converted field value is 0.9 × 22 + 1.5 = 21.3. If there is no transformation edge from method a to method b in the detection condition transformation graph, the system will write the corresponding field into the isolation index set and only generate record labels that do not involve similarity when generating the matching fingerprint. This embodiment explicitly saves source differences, detection caliber differences, and field missing differences through a field semantic table, a detection method encoding table, and a detection condition transformation graph, avoiding the mixing of data from different sources into similarity calculations when the comparability relationship is not confirmed.

[0025] refer to Figure 3 In one embodiment, the compound fingerprint construction module generates codes for monomer raw material fingerprints, component interaction fingerprints, and system context fingerprints, respectively. The monomer raw material fingerprint extracts normalized crystal form codes, surface treatment type codes, particle size distribution indices, oil absorption, whiteness, tinting strength, pH, and dispersion grade from the raw material particle size layer. For categorical fields, the system generates stable categorical codes based on the field semantic table; for continuous fields, the system uses normalized values; for missing fields, the system writes a missing marker and excludes the corresponding position from distance calculation. The monomer raw material fingerprint is represented as follows: Where, q r,k This represents the monomeric raw material fingerprint of the k-th titanium dioxide component in the compound record r, f1 to f2. n Indicates the continuous raw material field involved in the coding, κ r,k This represents the categorical field combination code for the k-th component. The physical meaning of this formula is: to write the comparable raw material indicators and category information of a single titanium dioxide component into a data vector with a fixed order. For example, if a component has a normalized oil absorption value of 0.4, a normalized whiteness value of 0.7, and a normalized tinting strength value of 0.6, and the categorical field combination code is code 8 formed by the titanium dioxide crystal form and surface treatment type, then a single raw material fingerprint containing elements such as 0.4, 0.7, 0.6, and 8 can be formed. Categorical fields are not directly added to continuous fields, but participate in similarity calculations according to category consistency rules.

[0026] Specifically, the component interaction fingerprint extracts the proportional relationships, index differences, index ratios, and surface treatment consistency markers between different titanium dioxide components from the compound layer. For any two components, the system determines the default component order according to the compound ratio from high to low. When there are markers for both main contributing titanium dioxide and auxiliary compensating titanium dioxide, a directed interaction code is generated starting with the main contributing titanium dioxide and ending with the auxiliary compensating titanium dioxide. The calculation formula for the component interaction field is as follows: Among them, hr,k,l p represents the interactive fingerprint segment in the compound record r from the k-th component to the l-th component. r,k -p r,l Indicates the difference in proportion between the two components. This represents the normalized difference on field f. This represents the smoothing ratio value on field f, where η represents the amount of positive smoothing used to avoid a denominator of 0, and e r,k,l This indicates a surface treatment consistency marker. The physical meaning of this formula is: to express the relative proportion and index direction between two titanium dioxide components within the same compound record. For example, if the proportions of the two components are 0.6 and 0.4 respectively, a normalized field value is 0.8 and 0.5 respectively, the smoothing amount is 0.01, and the surface treatment consistency marker is 1, then the proportion difference is 0.2, the field difference is 0.3, and the smoothing ratio is... This result is written into a component interaction fingerprint to distinguish the differences in complex combinations of the same component set under different primary and secondary relationships. See Table 2 for an example of complex fingerprint composition.

[0027] Table 2 Examples of fingerprint matching Monomer raw material fingerprint Raw material particle size layer Normalized indicators and category codes Missing fields are written to missing tags Component Interaction Fingerprint Composite layer Proportional difference, indicator difference, indicator ratio, consistency marker Re-encoding when the primary and secondary order changes System context fingerprint Ink system layer Binder type, dispersant type, pigment volume fraction range, target application performance Incompatible systems do not enter vector alignment Result associated fields Performance Result Layer Covering power, fineness, gloss, viscosity stability, sedimentation tendency Participating in candidate ranking and conflict grouping Furthermore, the compounding fingerprint corresponding to the same historical compounding record is associated with the ink-forming performance result, detection condition marker, and compounding record source identifier. If a historical compounding record lacks any component interaction field, the system does not replace the field with a zero value, but generates a void marker. The void marker includes the void field name, the missing component relationship, and the source file identifier. During vector index calculation, the void field does not participate in distance calculation; during inverted index retrieval, the void field can be used as a record source integrity label. In this embodiment, the compounding fingerprint is not simply a concatenation of raw material indicators into a vector, but rather a unified association of monomer information, component interrelationships, and system context after separate modeling, enabling the system to express two types of data relationships: similar raw materials but different compounding performance, and raw materials with large differences but similar ink-forming performance. This embodiment, through directed interactive coding and void markers, enables compounding records to have traceable data representation in terms of component order, proportion relationship, and field integrity.

[0028] refer to Figure 4In one embodiment, the composite index module establishes a three-level search path for compounding fingerprints. The rule index uses ink system compatibility conditions and target application performance as filtering conditions, mainly including whether the binder category belongs to the same comparable family, whether the dispersant category has the same dispersing effect, whether there is overlap in pigment volume fraction ranges, and whether the target application performance belongs to the same constraint category. Historical compounding records excluded by the rule index are not included in subsequent vector comparisons. The inverted index is used to recall candidate records according to titanium dioxide type, surface treatment type, key indicator range, and performance label. For example, the system establishes inverted tables for normalized oil absorption range, normalized whiteness range, dispersion grade, and surface treatment type. When the target compounding task includes a list of available titanium dioxide and target performance constraints, the inverted index returns the identifier of the compounding record that matches the corresponding field. The vector index is used to recall compounding records according to the joint distance of monomer raw material fingerprints, component interaction fingerprints, and system context fingerprints. The joint distance is calculated as follows: Among them, D r,v This represents the joint distance between the historical matchmaking record r and the target matchmaking request v. Indicates the fingerprint distance of the monomer raw material. Indicates the distance between component interaction fingerprints. Represents the system context fingerprint distance. λ represents the detection condition distance. q , λ h , λ s and λ m These represent the weights of the corresponding distances. The physical meaning of this formula is: simultaneously constraining the similarity of raw materials, compound relationships, system conditions, and detection calibers during the same retrieval process. For example, if a historical record has a fingerprint distance of 0.2 from the target raw material fingerprint, a component interaction fingerprint distance of 0.3, a system context fingerprint distance of 0.1, and a detection condition distance of 0.4, with weights of 0.35, 0.35, 0.2, and 0.1 respectively, then the joint distance is 0.35×0.2+0.35×0.3+0.2×0.1+0.1×0.4=0.235. The system forms a vector index recall order based on the joint distance from smallest to largest, and retains the component fields of each distance.

[0029] In this embodiment, the search results of the three-level search paths are merged using the compound record identifier. When the same compound record is hit by multiple search paths, the system retains the hit fields corresponding to each search path, such as the binder category hit by the rule index, the surface treatment type hit by the inverted index, and the component interaction field hit by the vector index. Records hit only by the inverted index but not filtered by the rule index are not written into the candidate output set; records that pass the rule index but have a large vector distance are retained as low-priority candidates or reference records. When outputting candidate titanium dioxide compound records, the candidate output module outputs the compound record identifier, compound ratio, corresponding titanium dioxide components, hit fields, detection condition markers, and ink-forming performance results together. The output is not a simple data display, but a structured candidate result based on the search path, field hits, and distance composition. This embodiment uses a three-level search path to enable compatibility filtering, field recall, and fingerprint similarity comparison to operate under the same data structure, while retaining the search source, facilitating subsequent conflict sample resolution.

[0030] In one embodiment, the candidate output module performs conflict sample resolution on the historical compound records recalled by the composite index module. Conflict sample resolution is based on the similarity of raw material indicators, differences in ink-forming performance, and differences in detection conditions. Historical compound records with similar titanium dioxide raw material indicators but inconsistent ink-forming performance results are classified into a first conflict group, while historical compound records with similar ink-forming performance results but differences in raw material indicators exceeding the historical distribution of the same type are classified into a second conflict group. The conflict intensity is calculated as follows: Among them, G r,v This represents the conflict strength of the historical matchmaking record r relative to the target matchmaking request v. This indicates differences in raw material indicators. This indicates a deviation in ink-forming performance. Indicates differences in detection conditions, μ a μ y and μ mThis indicates the weight of the corresponding difference item. The physical meaning of this formula is: to express the degree of contradiction between candidate records in terms of raw materials, results, and detection methods using the same numerical value. For example, if a record has a raw material index difference of 0.1, an ink-forming performance deviation of 0.6, and a detection condition difference of 0.2, with weights of 0.3, 0.5, and 0.2 respectively, then the conflict intensity is 0.3×0.1 + 0.5×0.6 + 0.2×0.2 = 0.37. If the record has a small raw material index difference but a large ink-forming performance deviation, it can enter the first conflict group; if the ink-forming performance deviation is small but the raw material index difference is large, it can enter the second conflict group. Specifically, the formation process of the first conflict group is as follows: within the same ink system context, the system searches for records where the monomer raw material fingerprint distance is lower than the median interval of the comparable sample distribution and the component cross-fingerprint distance is close, and then compares performance fields such as hiding power, fineness, gloss, viscosity stability, and sedimentation trend. If performance deviations are concentrated in the viscosity stability or sedimentation trend fields, the system further reads the detection condition markers and surface treatment consistency markers to determine whether the source of the difference is related to the detection aperture or the direction of component interaction. The formation process of the second conflict group is as follows: the system reads the raw material particle size layer and compound combination layer in records with similar ink-forming performance results to check whether there are cases where different raw material indicators form similar performance results through component complementarity. The candidate output module generates sorting information, key difference fields, and compatibility confidence levels based on the detection condition markers, system context distance, and performance deviation fields, and writes the conflict group type into the retrieval log of the candidate titanium dioxide compound records. For an example of conflict sample resolution records, refer to Table 3.

[0031] Table 3 Examples of Conflict Sample Resolution Records First Conflict Group Similar raw material properties, but inconsistent ink-forming performance Detection condition markers, performance deviation fields Reduce sorting or grouping display Second Conflict Group Similar ink-forming properties, but significant differences in raw material properties Complementary indicators, component interaction fields Retained as an alternative candidate Non-conflict candidates The raw materials, composition, system, and performance are all within a comparable range. Hit fields, joint distance Enter the regular candidate sorting Isolation Candidate There are incomparable detection fields. Set of isolation indicators and source identification Not involved in vector distance calculation In this embodiment, the ranking information is not based solely on a single similarity score, but rather on a combination of rule-based filtering results, inverted index hit fields, vector distance, conflict intensity, detection condition sources, and key difference fields. Key difference fields describe the main differences between candidate records and the target request, such as different surface treatment types, different primary / secondary component orders, a performance field being in a boundary region, or different detection method calibers. The compatibility confidence level is generated based on whether the candidate record is indexed by rules, whether it contains incomparable fields, whether it enters a conflict group, and its joint distance distribution position. This embodiment resolves conflicting samples, enabling the candidate output module to retain the source of differences in contradictory samples and preventing historical records with similar fields but unreliable matching relationships from being directly output as equivalent candidates.

[0032] In one embodiment, the composite index module also establishes a composite record relationship graph. The nodes of the composite record relationship graph are historical composite records, and the edges represent the raw material similarity, performance similarity, or system context similarity relationships between composite fingerprints. Each edge records the set of similar fields and the source of detection conditions. Raw material similarity relationships are formed by the distance between individual raw material fingerprints, performance similarity relationships are formed by the distance between ink-forming performance results, and system context similarity relationships are formed by the comparability of binder type, dispersant type, pigment volume fraction range, and target application performance. The edge weights are calculated as follows: Among them, W r,v The graph represents the edge weights of the relationship between historical matchmaking records r and matchmaking records or target requests v. Indicating the similarity of raw materials, Indicates the similarity of ink-forming properties. Indicates the similarity of system context. Indicating differences in detection conditions, θ a θ y θ s and θ m This represents the weight of the corresponding terms. The physical meaning of this formula is: the greater the difference in detection conditions, the lower the comparable edge weight between the two records; the higher the similarity in raw materials, performance, and system, the higher the comparable edge weight. For example, if two records have a raw material similarity of 0.8, an ink-forming performance similarity of 0.7, a system context similarity of 0.9, and a detection condition difference of 0.2, with four weights of 0.3, 0.3, 0.3, and 0.5 respectively, then the numerator is 0.3 × 0.8 + 0.3 × 0.7 + 0.3 × 0.9 = 0.72, the denominator is 1 + 0.5 × 0.2 = 1.1, and the edge weight is approximately... .

[0033] Furthermore, when performing conflict sample resolution, the candidate output module locates intermediate records that simultaneously connect the first and second conflict groups along the compound record relationship graph. Intermediate records refer to historical compound records that have a raw material similarity relationship with records in the first conflict group and a performance similarity relationship with records in the second conflict group, or a system context similarity relationship with both conflict groups. The system reads the detection condition markers, compounding ratio differences, and performance deviation fields of the intermediate records and adds these fields to the candidate ranking information. For example, an intermediate record may be similar to records in the first conflict group in the titanium dioxide type and oil absorption fields, but similar to records in the second conflict group in the gloss and viscosity stability results. The system retains this as an intermediate sample to explain the candidate relationship and records its connection relationship in the retrieval log. If the detection condition source of the intermediate record is consistent with the target request, the edge weight corresponding to the intermediate record participates in the candidate ranking; if the intermediate record has cross-method isolation fields, only the connection description is retained, and it does not participate in distance correction. This embodiment, through the compound record relationship graph, enables the indexing and tracing of multiple similarity relationships between historical compound records, and the candidate ranking is not limited to a single distance calculation between a single record and the target request.

[0034] In one embodiment, the compound fingerprint construction module extends the directed interactive coding. The system starts with the primary contributing titanium dioxide in the target ink system and ends with the secondary compensating titanium dioxide, recording the relative changes in masking-related, dispersion-related, and rheological-related indices, respectively. The primary contributing titanium dioxide can be determined based on the target performance field. For example, when the target performance is primarily constrained by masking force, the component with the dominant masking-related index is designated as the primary contributing titanium dioxide; when the target performance is primarily constrained by viscosity stability or dispersion fineness, the component with the dominant dispersion-related or rheological-related index is designated as the primary contributing titanium dioxide. The secondary compensating titanium dioxide is the component used to supplement another performance boundary. When generating the directed interactive code, the system does not treat the relationship between the two components as an unordered combination, but rather retains the direction from the primary contributing component to the secondary compensating component. When the direction changes, the code position and symbol are regenerated. For the same two-component combination, if the compounding ratio or target performance constraint changes, resulting in a reversed primary-secondary order, the system retains the corresponding codes and writes them as different compound fingerprints into the compound index module.

[0035] Specifically, in the occlusion-related indicators, the system records the relative directions of whiteness, tinting strength, and particle size distribution; in the dispersion-related indicators, the system records the relative directions of dispersion grade, surface treatment consistency, and fineness correlation fields; and in the rheology-related indicators, the system records the relative directions of oil absorption, viscosity stability, and sedimentation trend correlation fields. Each direction field is represented by a positive, negative, or identical label. If the normalized oil absorption value of the main contributing titanium dioxide is higher than that of the auxiliary compensating titanium dioxide, the rheology-related direction field is positive; if it is lower than that of the auxiliary compensating titanium dioxide, it is negative; and if they are in the same comparable range, they are identically labeled. The system writes the direction field, along with the ratio difference, indicator difference, and indicator ratio, into the component interaction fingerprint. For compound records of more than two types of titanium dioxide, the system generates paired directed interaction codes and arranges them from high to low according to the main contribution intensity. Components that do not form paired relationships are marked with empty slots. Through this embodiment, the same set of components has different computable expressions under different target performance constraints or different compounding ratios, reducing retrieval confusion caused by disordered combination coding.

[0036] refer to Figure 5 In one embodiment, the system further includes an index incremental update module. After new compounding experimental data is written, the index incremental update module determines whether the new compounding experimental data contains existing titanium dioxide models, existing ink system conditions, and newly added component interaction features. If the titanium dioxide model and ink system conditions of the new data already exist, but the component interaction features do not appear in the formal compounding fingerprint, the system generates a temporary compounding fingerprint and writes it to the verification index area. The verification index area is stored separately from the formal index. The temporary compounding fingerprint can participate in retrieval log recording and candidate source tracking, but does not directly change the recall boundary of the formal vector index. The system compares the ink-forming performance results of the new compounding experimental data with the prediction interval of similar historical records, and determines whether a stable correspondence is formed after the same newly added component interaction feature appears multiple times. The discriminant for a stable correspondence is as follows: Where z represents the interaction feature of the newly added component, R z Let N represent the set of compound records to be validated, containing the interaction feature z of the newly added component. z Represents set R z The number of records in |ε r | represents the residual inking performance of record r, δ z Z represents the residual tolerance determined by the bias of similar historical repeated experiments, I represents the indicator function, which takes the value 1 when the condition in parentheses is true, and 0 otherwise. zThis indicates the degree of stable correspondence of the newly added component interaction feature. The physical meaning of this formula is: to statistically determine whether the newly added component interaction feature forms a stable correspondence with the ink performance results across multiple records to be verified. For example, if a newly added component interaction feature corresponds to 5 records to be verified, and the absolute value of the performance residual in 4 of these records does not exceed the residual tolerance, then N... z =5), the sum of the 5 indicator function values ​​is 4. =0.8). The system can determine whether to convert a temporary matched fingerprint into a formal matched fingerprint based on the coverage of similar data.

[0037] In this embodiment, when a temporary composite fingerprint is converted into a formal composite fingerprint, the index incremental update module synchronously updates the detection condition transformation graph or the directed interactive coding. If the new data contains a new comparable relationship of detection methods, a transformation edge is added to the detection condition transformation graph, and the composite record identifier used to establish the transformation relationship is recorded. If the new data indicates that the same component set forms a new comparable direction under different primary and secondary orders, the directed interactive coding table is updated, and the corresponding code is written to the composite index module. If the new data only contains the results of a single experiment, or the performance residual does not have a stable correspondence, the system retains the temporary composite fingerprint and does not write it into the formal composite fingerprint set. The temporary composite fingerprint can appear as a field to be verified in the retrieval request, but it is not used as the main basis for formal candidate ranking. This embodiment separates the index area to be verified and the formal index area, so that the new experimental data does not change the historical index boundary before forming a stable data relationship, maintaining the traceability of data accumulation and index updates in the process of big data analysis.

[0038] In one embodiment, the system further includes a retrieval request correction module. This module receives the target ink performance constraints and a list of available titanium dioxide, and determines the fields in the target ink performance constraints that lack a reference basis based on the compounding record relationship graph, formal compounding fingerprints, and candidate ranking information. Fields lacking a reference basis refer to fields that do not have sufficient comparable historical records to support them in the formal compounding fingerprint, relationship graph edges, or candidate ranking records. For example, the target request proposes a specific performance field, but in historical data, this field only exists in a set of isolation indicators or only in a temporary compounding fingerprint. The system converts fields lacking a reference basis into fields to be confirmed, and generates a retrieval request containing fields to be confirmed, comparable fields, and prohibited extrapolation fields before rule index filtering. Comparable fields participate in similarity retrieval, fields to be confirmed are written to the retrieval log and a source hint is output, and prohibited extrapolation fields are used as filtering conditions to avoid vector extrapolation in areas not covered by historical data. See Table 4 for an example of retrieval request field status.

[0039] Table 4 Examples of Search Request Field Status Comparable fields The formal matching fingerprints contain fields with the same caliber. Participating in rule indexes, inverted indexes, and vector indexes Record hit fields Fields to be confirmed Only temporary fingerprint matching or insufficient sample quantity exists. Not included in similarity calculation Write the list of fields to be confirmed Fields that cannot be extrapolated Lack of comparability with the historical system context As a filtering condition Write a list of fields to prevent extrapolation Isolation fields The detection condition transformation graph has no transformation edges. Not involved in vector distance calculation Write to the isolation metric set Specifically, after receiving the target ink performance constraints, the retrieval request correction module matches the performance fields with the performance result layer fields in the formal compound fingerprint, matches the available titanium dioxide list with the model, surface treatment type, and detection condition markers in the raw material particle size layer, and then uses the compound record relationship graph to check whether comparable fields are connected to a sufficient number of historical compound records. For existing comparable fields, the system generates rule index conditions and vector index fields; for fields to be confirmed, the system does not replace them with default values, nor does it directly merge fields with similar names; for fields prohibited from extrapolation, the system restricts candidate records to have the same or comparable system context during the rule index filtering stage. The composite index module only performs similarity retrieval on comparable fields and uses fields prohibited from extrapolation as filtering conditions. When outputting candidate records, the candidate output module retains the classification results of fields to be confirmed, comparable fields, and fields prohibited from extrapolation. This embodiment, through retrieval request correction, ensures that the target request is consistent with the range of verified historical data, avoiding fields lacking a reference basis from entering the compound fingerprint distance calculation.

[0040] In one embodiment, the data access module, hierarchical modeling module, compound fingerprint construction module, composite index module, candidate output module, index incremental update module, and retrieval request correction module associate data using the same compound record identifier. When parsing raw material inspection data and ink performance results, the data access module generates association keys according to batch, component, and ink system. The hierarchical modeling module breaks down the association keys into raw material component keys, compound combination keys, system context keys, and performance result keys. The compound fingerprint construction module reads fields at each layer centered on the compound combination key to generate compound fingerprints. The composite index module writes the compound fingerprints into the regular index, inverted index, and vector index, and binds the index entries to the compound record identifier. When recalling candidate records, the candidate output module traces back the original fields, detection conditions, and conflict groups based on the compound record identifier. When writing new experimental data, the index incremental update module determines whether the new data supplements existing records or generates new records based on the compound record identifier. When generating retrieval requests, the retrieval request correction module uses the same identifier system to check whether fields are supported by formal compound fingerprints.

[0041] Specifically, the system's data processing can be stored in parallel at the database level using structured tables and vector index tables. The structured tables store field semantic tables, detection method encoding tables, raw material granularity layers, compound combination layers, ink system layers, performance result layers, conflict sample grouping tables, and retrieval logs. The vector index tables store individual raw material fingerprints, component interaction fingerprints, and system context fingerprints. The relationship graph can be stored using record node tables and record edge tables. The node tables record compound record identifiers, source files, and official status, while the edge tables record edge types, similar field sets, detection condition sources, and edge weights. Status fields are established for the verification index area and the official index area, including temporary, official, isolated, and invalid states. After a retrieval request enters the system, the rule index reads the system context fields from the structured tables, the inverted index reads the label and interval fields, the vector index reads the fingerprint fields, and the candidate output module reads the conflict sample grouping table and the relationship graph edge table to form candidate records. This embodiment enables the access, modeling, fingerprint generation, index retrieval, conflict handling, incremental update, and request correction of complex data to run under the same data link through unified identification and partitioned storage. The results of each process can be traced back to the original fields and the source of the detection conditions.

[0042] In a preferred embodiment, the system performs a preservative processing of abnormal and boundary data. For compounding records where ink production performance fails to meet standards, the system does not delete the corresponding data but instead writes it into the performance result layer and marks the non-compliant fields. Non-compliant records can participate in conflict sample resolution and relationship graph construction, but are sorted and controlled according to target constraints and conflict intensity during candidate output. For records with obvious missing data in raw material inspection data, the system writes the missing fields into empty space markers and generates partial compounding fingerprints based on the remaining fields; partial compounding fingerprints can participate in inverted index recall, but only fields with comparable relationships are calculated in the vector index. For boundary data caused by changes in detection conditions, the system determines whether it can be converted based on the detection condition transformation graph; fields that cannot be converted are written into the isolation index set. This processing method ensures that failed samples, boundary samples, and incomplete samples retain their data sources without unconditionally merging them.

[0043] Furthermore, the system categorizes candidate titanium dioxide blending records into regular candidates, alternative candidates, conflict candidates, and candidates to be verified during candidate output. Regular candidates originate from formal blending fingerprints constrained by rule indexes, inverted indexes, and vector indexes; alternative candidates originate from the second conflict group, representing historical records with similar ink-forming performance but significant differences in raw material combinations; conflict candidates originate from the first conflict group or records exhibiting multiple differences; and candidates to be verified originate from the candidate verification index area. The candidate output module outputs blending ratios, titanium dioxide components, system context, ink-forming performance results, key difference fields, and retrieval logs for each type of candidate record. The retrieval logs record the path each candidate takes to enter the output set, including rule index hit conditions, inverted index hit tags, vector index distance composition, conflict group type, and intermediate records in the relationship graph. Through this embodiment, the system does not mix and sort all historical records by a single score but retains the candidate source and data status, enabling technicians to conduct blending verification based on comparability boundaries.

[0044] In one embodiment, the system supports extended processing for combinations of three or more components in multi-component titanium dioxide blends. For a combination of three titanium dioxides, the blend fingerprint construction module generates a monomer raw material fingerprint for each component and generates multiple directed interaction fingerprints according to the main contribution strength. The main contribution strength can be determined by the direction of the raw material index corresponding to the target performance constraint. For example, the hiding power constraint corresponds to whiteness, tinting strength, and particle size distribution; the dispersion fineness constraint corresponds to dispersion grade and surface treatment type; and the viscosity stability constraint corresponds to oil absorption and related rheological fields. The system generates directed interaction codes from the main contributing component to each auxiliary compensation component within the same blend record and generates compensation interaction codes between the auxiliary compensation components. If an auxiliary compensation component is only used to improve dispersion performance and does not directly participate in masking the main contribution, the system retains this difference in the component interaction fingerprint and associates it with the fineness field in the performance result layer.

[0045] Specifically, the compounding layer stores a multi-component ratio matrix and paired interaction fields. Each element of the ratio matrix represents the ratio difference or proportion of one component relative to another, and the paired interaction fields store the corresponding raw material index differences. For a 3-component combination, the system generates 3 individual raw material fingerprints and at least 3 paired interaction fingerprints; for a 4-component combination, the system generates 4 individual raw material fingerprints and at least 6 paired interaction fingerprints. To avoid inconsistent vector lengths due to differences in the number of components, the system establishes fingerprint templates with the maximum number of comparable components. Missing component relationships are marked as empty slots, and historical records exceeding the target requested number of components are only retained as reference records if the rule index allows. The candidate output module displays the key difference fields corresponding to each paired interaction fingerprint in the multi-component candidate list. Through this embodiment, the system can process binary, ternary, and more component titanium dioxide compounding records and maintain the comparable structure of compounding fingerprints.

[0046] In a preferred embodiment, the system expresses the target ink performance constraints hierarchically. Target ink performance constraints may include mandatory fields, adjustable fields, and reference fields. Mandatory fields are filtered through a rule index, such as binder type, target application performance, and necessary system compatibility conditions; adjustable fields are entered into a vector index and candidate ranking, such as hiding power, gloss, fineness, and viscosity stability; reference fields are entered into the candidate output log, such as sedimentation trend or historical verification count. After receiving the target constraints, the retrieval request correction module first checks whether the fields exist in the formal compounding fingerprint and performance result layers, and then generates comparable fields, fields to be confirmed, and fields prohibited from extrapolation based on the field status. If the target request includes a list of available titanium dioxide, the system only performs candidate retrieval if a corresponding or comparable model exists at the raw material particle size layer; if a model in the list only has raw material inspection data but no compounding results, the system writes it as a component to be verified into the retrieval log and does not participate in the formal candidate ranking.

[0047] Furthermore, when the candidate output module generates candidate records based on the target ink performance constraints and the available titanium dioxide list, it does not change the original historical blending ratio. Instead, it outputs the historical blending ratio, candidate similarity relationships, and fields to be verified separately. If the blending ratio in a candidate record satisfies the ratio conservation constraint and passes the rule index filtering, the system considers it a verifiable candidate. If a candidate record is only similar in performance results but has significant differences in raw material indicators, it is considered a substitute candidate. If a candidate record contains incomparable fields under detection conditions, it is considered an isolated candidate or a candidate to be confirmed. This processing method ensures that the candidate output maintains the historical data source and does not generate ratio inferences without a source. Through this embodiment, when processing complex target requests, the system can distinguish between officially comparable data, data to be verified, and data prohibited from extrapolation, thus limiting the retrieval process to the range of verified data.

[0048] In one embodiment, the system's working principle can be summarized as follows: data standardization drives the generation of composite fingerprints; composite fingerprints drive composite index retrieval; conflict samples and relationship graphs correct candidate outputs; and incremental updates maintain the index state. Data standardization is not a simple unit conversion, but rather simultaneously preserves field semantics, detection methods, source identifiers, and missing status. Composite fingerprints are not single raw material vectors, but are composed of individual raw material fingerprints, component interaction fingerprints, and system context fingerprints. Composite indexes are not single-path retrievals, but a combination of rule indexes, inverted indexes, and vector indexes. Candidate outputs are not arranged by score, but rather candidate results are formed by combining conflict groups, mediating records, detection conditions, and field states. The system saves a retrieval log after each retrieval, which includes the target request, field state, index hit path, candidate record identifier, conflict group type, mediating record, and output state. When new experimental data arrives, the incremental index update module reads the retrieval log and experimental results to determine whether to update the temporary composite fingerprint, detection condition transformation graph, and directed interaction encoding.

[0049] In this embodiment, data transfer between modules has clear boundaries. The data access module is only responsible for source parsing and record identifier generation, and does not directly judge candidate similarity; the hierarchical modeling module is only responsible for data layer mapping and field status marking, and does not directly output the compounding scheme; the compounding fingerprint construction module is only responsible for generating computable data representation, and does not directly modify the original test data; the composite index module is only responsible for retrieving and hitting field records, and does not directly delete conflicting samples; the candidate output module is only responsible for candidate grouping, sorting, and log output, and does not directly write temporary data into the formal index; the index incremental update module only updates the formal fingerprint when the newly added data satisfies a stable correspondence; and the retrieval request correction module only adjusts the retrieval request field status according to the formal data range. Through the above boundaries, the system can perform structured, indexed, and traceable processing of ink multi-element titanium dioxide compounding data at the software level, and make each candidate record traceable back to the raw material field, compounding combination relationship, ink system conditions, and ink performance results.

Claims

1. A multi-element titanium dioxide compounding system for inks based on big data analysis, characterized in that, It includes a data access module, a hierarchical modeling module, a composite fingerprint construction module, a composite index module, and a candidate output module; The data access module is used to receive titanium dioxide raw material inspection data, ink system conditions, compounding ratio data, and ink forming performance results. The layered modeling module is used to map the received data into a raw material particle size layer, a compound combination layer, an ink system layer, and a performance result layer. The composite fingerprint construction module is used to generate composite fingerprints that include monomer raw material fingerprints, component interaction fingerprints, and system context fingerprints; The composite index module is used to retrieve historical composite records based on rule indexes, inverted indexes, and vector indexes; The candidate output module is a system used to output candidate titanium dioxide compound records corresponding to the target ink performance constraints.

2. The ink multi-component titanium dioxide compounding system based on big data analysis according to claim 1, characterized in that, The hierarchical modeling module is used to establish a field semantic table and a test method coding table for the titanium dioxide raw material inspection data. It maps the name, unit, sampling batch and test conditions of the same indicator in different test reports to a unified field, and adds data source tags to missing fields, conversion fields and cross-method fields respectively. The raw material particle size layer records at least the crystal form, surface treatment type, particle size distribution, oil absorption, whiteness, tinting strength, pH value, and dispersion grade. The ink system layer records at least the binder type, dispersant type, pigment volume fraction range, and target application performance. The performance result layer records at least the hiding power, fineness, gloss, viscosity stability, and sedimentation trend.

3. The ink multi-component titanium dioxide compounding system based on big data analysis according to claim 1, characterized in that, The compound fingerprint construction module is used to extract monomer index sequences from the raw material particle size layer, extract the proportional relationship, index difference, index ratio and surface treatment consistency mark between different titanium dioxide from the compound combination layer, extract system context constraints from the ink system layer, and encode the monomer index sequences, inter-component interaction features and system context constraints into a compound fingerprint according to a preset field order. The matching fingerprint corresponding to the same historical matching record is simultaneously associated with the ink performance result, detection condition marker and matching record source identifier, and a gap marker is generated for historical matching records that lack any component interaction field.

4. The ink multi-component titanium dioxide compounding system based on big data analysis according to claim 1, characterized in that, The composite index module is used to establish a three-level retrieval path for the compound fingerprint. The rule index is used to filter historical compound records according to ink system compatibility conditions and target application performance. The inverted index is used to recall candidate records according to titanium dioxide model, surface treatment type, key indicator range and performance label. The vector index is used to recall compound records according to the joint similarity of monomer raw material fingerprint, component interaction fingerprint and system context fingerprint. The search results of the three-level search path are merged by matching record identifiers, and the hit fields corresponding to each search path are retained.

5. The ink multi-component titanium dioxide compounding system based on big data analysis according to claim 1, characterized in that, The candidate output module is used to perform conflict sample resolution on the historical compound records recalled by the composite index module. The conflict sample resolution includes classifying historical compound records with similar titanium dioxide raw material indicators but inconsistent ink performance results into a first conflict group, classifying historical compound records with similar ink performance results but different raw material indicators exceeding the historical distribution of the same type into a second conflict group, and generating sorting information, key difference fields and compatibility confidence levels of candidate titanium dioxide compound records based on detection condition markers, system context distance and performance deviation fields, and writing the conflict group type into the retrieval log of candidate titanium dioxide compound records.

6. The ink multi-component titanium dioxide compounding system based on big data analysis according to claim 1, characterized in that, The hierarchical modeling module is also used to establish a detection condition transformation graph, wherein the nodes of the detection condition transformation graph correspond to the detection methods in the detection method coding table, and the edges correspond to the comparable fields generated by the same batch of titanium dioxide under different detection methods. When new raw material inspection data is received, the fields that can be directly compared, the fields that need to be converted, and the fields that are not comparable are determined according to the test condition conversion diagram. The fields that need to be converted are written into the converted index set, and the fields that are not comparable are written into the isolation index set. The composite fingerprint construction module references the transformed indicator set when generating composite fingerprints, and only generates record labels that do not involve similarity for the isolated indicator set.

7. The ink multi-component titanium dioxide compounding system based on big data analysis according to claim 1, characterized in that, The compound fingerprint construction module is also used to establish a directed interaction code for the interaction features between the components. The directed interaction code takes the main contributing titanium dioxide in the target ink system as the starting point and the auxiliary compensating titanium dioxide as the ending point, and records the relative change direction of the masking related index, the dispersion related index and the rheological related index respectively. When the compound ratio changes, the directed interactive code is regenerated according to the order of components before and after the ratio change, and the codes corresponding to different component orders are written into the compound index module as different compound fingerprints. For the same two-component combination, the corresponding codes are retained when the main and auxiliary orders are reversed.

8. The ink multi-component titanium dioxide compounding system based on big data analysis according to claim 5, characterized in that, The composite index module is also used to establish a composite record relationship graph. The nodes of the composite record relationship graph are historical composite records, and the edges are the raw material similarity relationship, performance similarity relationship or system context similarity relationship between composite fingerprints. The graph also records the corresponding set of similar fields and the source of detection conditions for each edge. When performing conflict sample resolution, the candidate output module locates the intermediary record that connects the first conflict group and the second conflict group along the re-matching record relationship graph, and adds the detection condition mark, re-matching ratio difference and performance deviation fields of the intermediary record to the candidate sorting information.

9. The ink multi-component titanium dioxide compounding system based on big data analysis according to claim 1, characterized in that, It also includes an index incremental update module, which is used to determine whether the new compounding experimental data contains existing titanium dioxide models, existing ink system conditions and newly added component interaction characteristics after the new compounding experimental data is written. When new component interaction features are included, a temporary compound fingerprint is generated for the new compound experimental data and written into the verification index area; When the new composite experimental data and the records in the index area to be verified form a stable correspondence in the ink forming performance results, the temporary composite fingerprint is converted into a formal composite fingerprint, and the detection condition transformation graph or directed interactive coding is updated simultaneously.

10. The ink multi-component titanium dioxide compounding system based on big data analysis according to claim 8, characterized in that, It also includes a retrieval request correction module, which is used to receive the target ink performance constraints and the list of available titanium dioxide, determine the fields in the target ink performance constraints that lack reference bases according to the compounding record relationship diagram, formal compounding fingerprint and candidate sorting information, convert the fields that lack reference bases into fields to be confirmed, and generate a retrieval request containing fields to be confirmed, comparable fields and prohibited extrapolation fields before rule index filtering; The composite index module performs similarity retrieval only on comparable fields and uses fields that are prohibited from extrapolation as filtering conditions.

Images