A method and system for constructing a dynamic knowledge graph of lacquer decoration process

Abstract

The present application relates to traditional process digitization and virtual simulation data processing technology, and discloses a construction method and system of dynamic knowledge graph of lacquer decoration process. The method comprises the following steps: collecting and standardizing lacquer decoration process data, constructing basic process flow skeleton and operation atlas; establishing the state transition relationship and action constraint relationship between the action unit and the paint layer state; receiving the actual operation event uploaded by the virtual reality terminal and performing action verification; determining the feedback sample according to the virtual reality operation log and the teacher's annotation result, aggregating the feedback samples pointing to the same target relationship and generating candidate update data; verifying by using the historical confirmed sample, updating the operation atlas and generating a new version; monitoring the new version, and restoring to the version before updating when the preset running condition is not met. The present application improves the accuracy and stability of process judgment, error feedback and atlas updating in virtual training.

Description

Technical Field

[0001] This invention relates to the digitization of traditional crafts and virtual simulation data processing technology, and in particular to a method and system for constructing a dynamic knowledge graph of lacquer decoration techniques. Background Technology

[0002] The lacquer finishing process typically includes preparation, mounting with cloth, applying ash, applying the lacquer, air drying, grinding, polishing, and decorative steps such as mother-of-pearl inlay, gilding, and gold plating. These steps are not only sequential but also constrained by factors such as the type of raw lacquer, the state of the lacquer layer, the tools used, the area to be treated, the method of application, and the drying time.

[0003] For example, after application, the paint layer needs to be air-dried to gradually form a film. When a film forms on the paint surface and leaves no obvious mark when lightly touched, it can be considered surface-dry. However, surface-dry does not necessarily mean the paint layer is completely dry from the surface inwards, nor does it necessarily mean the paint layer is ready for polishing. Polishing before the paint layer has reached the appropriate dryness may cause adhesion, paint film scratches, or surface damage. Applying paint directly before the puttying is complete may result in insufficient adhesion or delamination. Excessive paint thickness or unsuitable air-drying conditions may also cause blistering, uneven drying, or loss of gloss.

[0004] Existing virtual reality training systems can typically demonstrate the lacquer application process and determine whether the user has completed the corresponding operation based on pre-set steps. However, these systems usually focus on whether the action occurred or the sequence of procedures was completed, making it difficult to combine the current lacquer layer condition, the tools used, the area of ​​application, and the action parameters to determine whether the current action meets the conditions of a real lacquer application process.

[0005] Existing knowledge graphs can organize knowledge of materials, tools, processes, and techniques into nodes and relationships, but they are mainly used for knowledge retrieval and visualization. For virtual reality training in lacquer decoration, simply knowing that "grinding belongs to a certain process" or "grinding requires the use of grinding tools" is insufficient to determine whether the current lacquer layer can be ground, and it is also difficult to provide corresponding defect risks when operational errors occur.

[0006] Furthermore, the lacquer finishing process is affected by factors such as lacquer batches, lacquer layer thickness, ambient temperature and humidity, and operating habits. The same lacquer layer state may correspond to different operability under different conditions. If the diagram remains unchanged after being put into use, it is easy for some state relationships to be inconsistent with the actual training results; if the diagram is modified directly based on a single student operation, it may write erroneous operations or occasional data into the process relationship.

[0007] Therefore, in the virtual reality training scenario of lacquer decoration, the specific technical problem that needs to be solved is: how to construct the lacquer decoration process, lacquer layer state, operation actions, tool conditions and defect results into a process knowledge graph that can be called in real time by the virtual reality terminal, and how to update the state constraint relationship in the graph in a controlled manner based on the confirmed practical feedback, so that the graph can not only determine whether the current action meets the process conditions, but also be dynamically optimized as training data accumulates. Summary of the Invention

[0008] In view of the shortcomings of the existing technology, the purpose of this invention is to provide a method and system for constructing a dynamic knowledge graph of lacquer decoration techniques.

[0009] To achieve the above objectives, the present invention provides the following technical solution:

[0010] A method for constructing a dynamic knowledge graph of lacquer decoration techniques includes the following steps:

[0011] S1. Collect lacquer decoration process data and standardize the lacquer decoration process data to obtain process sample data;

[0012] S2. Construct the lacquer decoration process entity based on the process sample data, and extract the basic entity relationships from the process sample data;

[0013] S3. The process nodes in the lacquer finishing process body are broken down into action unit nodes, and the state transition relationship between the previous lacquer layer state, the action unit node and the subsequent lacquer layer state is established.

[0014] S4. Establish action constraint relationships and defect mapping relationships for virtual reality practice based on the state transition relationships;

[0015] S5. Write the basic entity relationship, the state transition relationship, the action constraint relationship and the defect mapping relationship into the graph database to generate a lacquer decoration process knowledge graph that can be called by the virtual reality terminal, and perform action verification on the practical events based on the lacquer decoration process knowledge graph to obtain the graph judgment result.

[0016] S6. Based on the virtual reality practice log, the graph judgment result and the teacher annotation result, determine the feedback sample, aggregate the feedback sample to generate candidate update data, and update the lacquer decoration process knowledge graph according to the candidate update data when the candidate update data passes the verification.

[0017] In a preferred embodiment, the lacquer finishing process data includes process text data, process image data, demonstration operation video data, and virtual reality operation log data;

[0018] The standardization process includes terminology unification, synonym merging, process numbering, tool numbering, operation area numbering, and paint layer status numbering.

[0019] The process sample data includes process data, action unit data, paint layer condition data, tool data, material data, operation area data, and defect data.

[0020] In a preferred embodiment, step S2 includes:

[0021] The lacquer finishing process body is constructed based on the process sample data. The lacquer finishing process body includes process categories, action unit categories, lacquer layer state categories, tool categories, material categories, operation area categories, and defect categories.

[0022] Basic entities and entity relationships are extracted from the process sample data to obtain basic knowledge triples;

[0023] Based on the aforementioned basic knowledge triplet, a basic process flow framework is formed, which includes the process sequence relationship and the confirmed core process relationship.

[0024] The knowledge graph of lacquer finishing process also includes a dynamically updatable operation graph, which includes action unit nodes, lacquer layer state nodes, tool nodes, material nodes, operation area nodes and defect nodes, as well as state transition relationships, action constraint relationships and defect mapping relationships formed by the nodes.

[0025] In a preferred embodiment, step S3 includes:

[0026] Configure the preceding paint layer state, allowed tools, allowed action area, action parameter range, and subsequent paint layer state for each of the aforementioned action unit nodes;

[0027] The state transition relationship is established based on the state of the preceding paint layer, the action unit node, and the state of the subsequent paint layer.

[0028] The paint layer state includes at least one of the following: wet state, surface dry state, fully dry state, grindable state, ground state, polishable state, and polished state.

[0029] In a preferred embodiment, the practical event includes action type, tool identifier, area of ​​action, action parameters, action duration, and paint layer state before the action is performed;

[0030] The action constraints include state constraints, tool constraints, region constraints, and time constraints;

[0031] In step S5, the practical event is subjected to action verification, including determining whether the practical event meets the constraints of the preceding paint layer state, allowed tools, allowed action area and time of the corresponding action unit, and determining whether the practical event is legal based on the judgment result.

[0032] In a preferred embodiment, the defect mapping relationship is used to associate practical events that do not satisfy the action constraint relationship with the corresponding defect nodes;

[0033] Based on the graph judgment result, the following information is returned to the virtual reality terminal: the legality result of the action, the next action allowed to be executed under the current paint layer state, the unmet action constraints, the corresponding defect risks and correction suggestions;

[0034] The defect nodes include at least one of the following: blistering, cracking, delamination, adhesion, paint film scratches, surface scratches, loss of gloss, uneven drying, and pattern misalignment.

[0035] In a preferred embodiment, step S6 includes:

[0036] The practical event in which the atlas judgment result is inconsistent with the teacher confirmation result in the teacher annotation result is identified as the feedback sample;

[0037] Based on the target state node, action unit, tool, area of ​​action, and teacher confirmation result, multiple feedback samples pointing to the same target relationship are aggregated.

[0038] When the number of feedback samples reaches a preset minimum number and the proportion of feedback samples with the same teacher confirmation result reaches a preset ratio, it is determined that the preset update conditions are met, and the candidate update data is generated.

[0039] The candidate update data includes at least one of the following: state constraint relationship weight adjustment data, defect mapping relationship addition or deletion data, defect risk level adjustment data, feedback text modification data, and paint layer state node refinement data.

[0040] In a preferred embodiment, step S6, verifying the candidate update data, includes:

[0041] An updated graph relationship is formed based on the candidate update data;

[0042] Using historical confirmed samples, the action judgment results after adopting the updated graph relationship and the action judgment results before the update are obtained respectively;

[0043] When the consistency between the action judgment result after adopting the updated graph relationship and the teacher's confirmation result is higher than that before the update, and the basic process sequence has not changed, the candidate updated data is determined to pass the verification.

[0044] The knowledge graph of lacquer decoration techniques is updated based on the verified candidate update data, and a new version of the graph is generated.

[0045] The action misjudgment rate and query anomaly rate of the new map version are monitored. When the action misjudgment rate of the new map version is higher than the action misjudgment rate before the update, or the query anomaly rate exceeds a preset anomaly threshold, the map version before the update is restored.

[0046] In a preferred embodiment, when the feedback sample indicates that the original paint layer state node cannot distinguish different action execution conditions, refined paint layer state node data is generated based on surface drying time, actual drying time, shade drying time, paint layer thickness and ambient humidity. The original paint layer state node is subdivided into surface-dry but not fully dried state, fully dried but not reaching grinding conditions state, and grindable state. Action constraint relationship and defect mapping relationship between the subdivided paint layer state node and grinding action are established respectively.

[0047] Candidate update data that are verified and do not involve changes in the sequence relationship of basic processes in the basic process flow framework are used to update the operation map;

[0048] When the candidate update data involves changes to the basic process sequence relationship, the candidate update data is manually reviewed, and after the manual review is passed, the basic process skeleton is updated according to the approved candidate update data.

[0049] A system for constructing a dynamic knowledge graph of lacquer decoration techniques includes:

[0050] The data acquisition and standardization module is used to collect data on lacquer decoration techniques and to standardize the data to obtain process sample data.

[0051] The graph construction module is used to construct the lacquer decoration process ontology based on the process sample data, extract basic entity relationships, establish state transition relationships, action constraint relationships, and defect mapping relationships between action units and lacquer layer states, and write the basic entity relationships, state transition relationships, action constraint relationships, and defect mapping relationships into the graph database to generate a lacquer decoration process knowledge graph.

[0052] The virtual reality verification module is used to receive the practical events uploaded by the virtual reality terminal, call the knowledge graph of lacquer decoration process to verify the actions of the practical events, and obtain the graph judgment result;

[0053] The feedback update module is used to determine feedback samples based on the virtual reality practice log, the graph judgment results, and the teacher annotation results, aggregate the feedback samples to generate candidate update data, and update the lacquer decoration process knowledge graph based on the candidate update data when the candidate update data passes the verification.

[0054] Compared with the shortcomings of the prior art, the beneficial effects of the present invention are as follows:

[0055] This invention breaks down the lacquer application process into action units that can be recognized by virtual reality terminals, and establishes state transition relationships between the previous lacquer layer state, action units, and subsequent lacquer layer states. This allows the knowledge graph to go beyond simply displaying static knowledge such as processes, materials, and tools, and to perform real-time verification of user actions based on the current lacquer layer state.

[0056] Furthermore, by associating the lacquer layer states such as surface dry, fully dry, grindable, and ground with action units, tool conditions, action areas, and action parameters, it is possible to distinguish between "action has occurred" and "action meets process conditions," thereby improving the process authenticity of virtual reality training in lacquer decoration.

[0057] Furthermore, by establishing a mapping relationship between illegal actions and defects such as blistering, cracking, delamination, adhesion, surface scratches, and loss of gloss, the system can output the reason for the violation and the possible consequences of the defect while judging the illegality of the action, making the error feedback interpretable.

[0058] Furthermore, by collecting virtual reality practice logs and teacher annotation results, recurring misjudgment relationships are aggregated, and after verification, state constraint relationships, defect mapping relationships, and risk levels are updated, enabling the knowledge graph to be dynamically optimized based on actual training results.

[0059] Furthermore, when the original paint layer state cannot accurately represent the operability under different process conditions, the original state nodes can be further refined based on surface drying time, actual drying time, ambient humidity, paint layer thickness, and shade drying time, thereby reducing misjudgment of actions caused by overly coarse state division.

[0060] Furthermore, by storing the basic process flow skeleton separately from the updatable operational relationships, and by monitoring and rolling back new graph versions, the impact of single misoperations or abnormal data on traditional core processes can be reduced, thereby improving the stability of the continuous operation of the dynamic knowledge graph. Attached Figure Description

[0061] Figure 1 This is a flowchart illustrating the method for constructing a dynamic knowledge graph of lacquer decoration techniques according to the present invention.

[0062] Figure 2This is a schematic diagram of the structure of a system for constructing a dynamic knowledge graph of lacquer decoration techniques according to the present invention. Detailed Implementation

[0063] In the following description, to enable those skilled in the art to better understand the present invention, specific implementation methods are provided, including lacquer finishing process data, virtual reality operation events, atlas relationships, and atlas update processes. Those skilled in the art will understand that corresponding adjustments can be made to the following implementation methods without departing from the technical concept of the present invention.

[0064] Explanation of some concepts:

[0065] A knowledge graph is a data structure consisting of multiple entity nodes and the edges connecting these entities. The knowledge graph in this invention is not a simple listing of lacquer terminology, but rather uses processes, action units, lacquer layer states, tools, materials, operating areas, and defects as entity nodes, and process sequences, state transitions, tool adaptations, area adaptations, and defect mappings as relational edges. This allows the virtual reality terminal to use these relational edges to make action judgments.

[0066] Dynamic knowledge graphs refer to knowledge graphs that can update certain node attributes or relationship edges and form new versions based on verified feedback from virtual reality practices. It's important to note that dynamic updates do not directly modify the graph based on arbitrary single practice results; rather, they are controlled updates implemented after aggregating and verifying the feedback data.

[0067] The basic process flow framework refers to the diagrammatic part composed of the sequence of processes and the confirmed core process relationships. For example, the slagging process precedes the coating process, the air-drying process follows the coating process, and the grinding process begins after reaching the corresponding dry state. The practical logs of ordinary trainees do not directly alter the basic process flow framework.

[0068] The surface-dry state refers to the state where, after raw lacquer or refined lacquer is applied to the surface of an object, the lacquer layer has formed a film on the surface, and lightly touching the lacquer surface will not leave obvious marks.

[0069] Surface drying time refers to the time it takes for paint to change from a liquid state to the formation of a paint film on the surface.

[0070] The "fully dried" state refers to the state in which the raw lacquer layer has dried from the surface to the inside and formed a solid lacquer film.

[0071] The drying time refers to the time it takes for the raw lacquer layer to change from a liquid state to the formation of a solid lacquer film from the surface to the interior.

[0072] It should be noted that surface-dry and fully-dry states are different lacquer layer states, and surface-dry state is not necessarily equivalent to a polishable state. The polishable state is an operational condition state set by this invention according to specific lacquer finishing process requirements, and can be determined based on the degree of fully-drying, lacquer layer thickness, ambient humidity, air-drying time, and teacher confirmation. The meanings of the terms surface-dry, surface-drying time, fully-dry, and fully-drying time mentioned above correspond to the lacquer finishing technique terminology standards provided.

[0073] like Figure 1 As shown, this embodiment relates to a method for constructing a dynamic knowledge graph of lacquer decoration techniques, including steps S1 to S6.

[0074] S1, collect and standardize data on lacquer finishing techniques.

[0075] In step S1, process text data, process image data, demonstration operation video data, and virtual reality operation log data are collected, and the collected data are standardized.

[0076] Generally, textual data on crafts can include "The Record of Lacquer Decoration" and its annotations, lacquer art textbooks, intangible cultural heritage application materials, process regulations, restoration cases, teaching notes, and workshop experience records.

[0077] Process image data can include images of the object's surface during various process stages such as preparation, mounting, ash application, lacquering, air drying, grinding, polishing, and decoration.

[0078] The demonstration operation video data can include video and motion trajectory data formed by the inheritor or teacher performing operations such as brushing, scraping, dry sanding, wet sanding, polishing, inlaying, gilding and painting.

[0079] Virtual reality operation log data can include action type, tool identifier, area of ​​action, duration of action, trajectory of action, speed of action, angle of action, and status of virtual workpiece, collected by virtual reality controller, gesture recognition device, or motion capture device.

[0080] Furthermore, process data related to the paint layer condition can also be collected. For example, but not limited to: surface dry state, surface dry time, fully dry state and fully dry time; air drying start time, air drying time, air drying ambient temperature and relative humidity; paint layer thickness, number of coats and time interval between two adjacent coats; defects such as adhesion, blistering, cracking, delamination, surface scratches, loss of gloss and uneven drying; material identification such as raw lacquer, refined lacquer, ash, pigments, mother-of-pearl flakes and metal powder; tool identification such as paint brushes, scrapers, grinding tools, polishing tools, inlay tools and painting tools.

[0081] Specifically, this involves unifying synonyms from different sources. For example, "lacquering," "applying paint," and "brushing paint" are uniformly mapped to the action of brushing; "drying in the shade" and "air drying" are uniformly mapped to the process of air drying.

[0082] The terms "surface dry," "semi-dry," and "not sticky" are not directly considered to represent the same state. Instead, they are mapped based on the corresponding data source and labeling results. Data that meets the conditions of surface film formation and leaves no obvious mark when lightly touched can be labeled as surface dry; data that has formed a solid paint film from the surface to the interior can be labeled as fully dry; and data that meets the conditions for subsequent grinding can be further labeled as grindable.

[0083] Optionally, object codes can be configured for various types of objects. For example, process objects include process identifier and process name; motion unit objects include motion identifier, the process to which it belongs, and motion parameters; paint layer status objects include status identifier, status name, status source, and applicable conditions; tool objects include tool identifier and tool category; and defect objects include defect identifier, defect name, and corresponding cause.

[0084] Through the above processing, process sample data that can be uniformly called upon in subsequent map construction processes are obtained. Then proceed to step S2.

[0085] S2, construct the lacquer decoration process entity and extract the basic entity relationships.

[0086] In step S2, the lacquer decoration process body is constructed based on the process sample data.

[0087] The process body includes process categories, action unit categories, paint layer state categories, tool categories, material categories, operation area categories, and defect categories.

[0088] The process category can include processes such as preparing the base, mounting the cloth, applying putty, painting, air drying, grinding, polishing, and decoration; the action unit category can include actions such as brushing, scraping putty, rough grinding, fine grinding, polishing, pasting, and painting; the paint layer state category can include untreated state, primed state, wet state, surface dry state, fully dry state, grindable state, ground state, polishable state, and polished state.

[0089] Materials include, but are not limited to, raw lacquer, refined lacquer, ash, pigments, mother-of-pearl flakes, and metal powder; defects include, but are not limited to, blistering, cracking, delamination, adhesion, paint film scratches, surface scratches, loss of gloss, and pattern misalignment.

[0090] Furthermore, rule extraction, manual annotation, or model-assisted extraction methods can be used to extract entities and entity relationships from process sample data to obtain basic knowledge triples.

[0091] The basic knowledge triad includes, for example: spare tire process—precedes—putting process; puttying process—precedes—coating process; coating process—subsequent process—air drying process; coating process—includes—brushing action; brushing action—using tool—paint brush; surface dry state—distinguished from—fully dry state; fully dry state—transforms into—grindable state after meeting preset grinding conditions; grinding process—includes—coarse grinding action and fine grinding action; polishing process—using tool—polishing tool.

[0092] It should be noted that this step is not merely about increasing the number of technical terms, but rather about converting these technical terms into data objects with unified coding and relational orientation, so that they can participate in subsequent virtual reality practical verification.

[0093] Through the above processing, the main body and basic entity relationship of the lacquer decoration technique are obtained. Then proceed to step S3.

[0094] S3, construct the state transition relationship between the action unit and the paint layer state.

[0095] In step S3, corresponding action units are set for each process node, and each action unit is configured with the previous paint layer state, allowed tools, allowed action area, action parameter range and subsequent paint layer state.

[0096] The state transition relationship can be represented as: previous paint layer state — action unit — subsequent paint layer state.

[0097] For example, in the coating process, the pre-treatment state of the paint layer before the brushing action can be the pre-treatment state, the allowed tool is a paint brush, the allowed action area is the area to be coated, and the subsequent paint layer state is a wet state.

[0098] After the wet paint layer enters the air drying process, it can be transformed into a surface-dry state according to the air drying time and corresponding state data; after continuing to air dry and reaching the condition of forming a solid paint film from the surface to the inside, it can be transformed into a fully-dry state.

[0099] It should be noted that the fully dry state can be used as a basis for judging subsequent operating conditions, but not all fully dry states necessarily meet the same grinding conditions. Optionally, a grinding-ready state can be set after the fully dry state, depending on the paint layer thickness, ambient humidity, and actual process requirements.

[0100] For example, in the grinding process, the pre-grind paint layer is in a grindable state, the permissible tool is a coarse grinding tool, the permissible action area is the already coated area, and the subsequent paint layer is in a state where coarse grinding is completed; the pre-grind paint layer is in a state where coarse grinding is completed, and the subsequent paint layer is in a state where it has been ground.

[0101] Coarse grinding is not permitted when the paint layer is currently wet or has only reached surface dryness but has not yet reached the corresponding grinding conditions.

[0102] For the polishing process, the initial polishing action is preceded by a polished or polishable paint layer, and the permitted tool is a polishing tool. The subsequent paint layer is the state after the initial polishing is completed. The final polishing action is preceded by the state after the initial polishing is completed, and the subsequent state is the state after polishing.

[0103] Through the above processing, the state transition relationship between the action unit and the paint layer state is obtained. Then proceed to step S4.

[0104] S4, construct motion constraint models and defect mapping relationships for virtual reality practice.

[0105] In step S4, an action constraint model is established for the practical events uploaded by the virtual reality terminal.

[0106] In one embodiment, the practical event E can be represented as:

[0107]

[0108] Where A represents the action type, T represents the tool identifier, Z represents the area of ​​action, P represents the action parameter, D represents the action duration, and S represents the paint layer state before the action is executed.

[0109] Perform state constraint verification, tool constraint verification, area constraint verification, and time constraint verification on the practical event E.

[0110] Optionally, the legality of an action can be determined in the following ways:

[0111]

[0112] in, This indicates whether the current paint layer condition meets the requirements of the previous condition; Indicate whether the tool used is a permitted tool; Indicates whether the area of ​​effect is within an allowed area; This indicates whether the duration or timing of the action meets the time constraints. A value of 1 indicates that the constraints are met, and a value of 0 indicates that the constraints are not met.

[0113] when When, determine that the practical event meets the execution conditions of the corresponding action unit; when At that time, it is determined that at least one constraint is not satisfied in the practical event.

[0114] Furthermore, a defect mapping relationship is established for practical events that do not meet the constraints. For example: wet state or surface dryness not reaching the grinding condition - perform grinding action - corresponding to adhesion, paint film scratches, or surface damage; incomplete puttying - perform paint application action - corresponding to insufficient adhesion or delamination; paint thickness exceeding the corresponding range - enter the air drying process - corresponding to blistering or uneven drying; un-grinded state - perform polishing action - corresponding to surface scratches or uneven gloss; mismatched polishing tools - perform polishing action - corresponding to surface scratches or loss of gloss.

[0115] Correspondingly, the defect mapping relationship can also store risk levels and feedback text. For example, when the virtual reality terminal uploads "the current state is surface dry, the user selects a coarse grinding tool and performs the grinding action", the system can output: the current action does not meet the grinding conditions; surface dry state is not the same as a grindable state; continuing the operation may cause adhesion or paint film damage; it is recommended to continue air drying or wait for the teacher to confirm the paint layer status.

[0116] Through the above processing, the action constraint relationship and defect mapping relationship are obtained. Then proceed to step S5.

[0117] S5 generates a knowledge graph of lacquer decoration techniques that can be accessed by virtual reality terminals.

[0118] In step S5, the basic entity relationships, state transition relationships, action constraint relationships, and defect mapping relationships are written into the graph database to generate a knowledge graph of lacquer decoration techniques.

[0119] Generally, each node stores the node type, node code, node name, process stage, associated parameters, and display description.

[0120] Each relation edge stores the edge type, starting node, target node, preconditions, successor conditions, constraint rules, risk level, and feedback text.

[0121] Optionally, the graph database can be an attribute graph database or a resource description framework database. When using an attribute graph database, processes, action units, paint layer states, tools, materials, areas of action, and defects are used as nodes, and various process relationships are used as edges with direction and attributes.

[0122] The generated knowledge graph connects to the virtual reality terminal via a query interface. When a user performs an action, the virtual reality terminal uploads the actual action event. The graph query interface retrieves the corresponding action unit and its constraints based on the current paint layer state and returns: whether the current action is allowed to be executed; the next action allowed in the current state; unmet state, tool, region, or time constraints; potential defects; and corresponding corrective suggestions.

[0123] For example, when the current paint layer is wet, if the user selects a polishing tool and performs a polishing action, the system returns an invalid action and prompts that the current paint layer has not yet reached a polishable state, which may cause adhesion or surface damage. It suggests entering the air-drying waiting step.

[0124] When the current paint layer is in a grindable state, and the user selects the corresponding grinding tool to perform a rough grinding action in the allowed area, and the action parameters are within the preset range, the system returns that the action is valid and updates the virtual workpiece status to the rough grinding completed state.

[0125] Through the above processing, a knowledge graph of lacquer decoration techniques that can be accessed in real time by a virtual reality terminal is obtained. Then proceed to step S6.

[0126] S6 dynamically updates the knowledge graph based on virtual reality hands-on feedback.

[0127] In step S6, virtual reality practice logs and teacher annotation results are collected.

[0128] Virtual reality operation logs include action sequences, graph judgment results, violation types, repeated error nodes, correction actions, and correction results.

[0129] Teacher annotation results include the teacher's confirmation, correction, or supplementation of the graph judgment results by the annotation terminal. For example, the teacher can correct actions judged as legal by the system to be illegal, or confirm actions judged as illegal by the system as permissible under specific state conditions.

[0130] Specifically, the practical events where the results of the graph judgment are inconsistent with the results of the teacher confirmation are identified as feedback samples and aggregated according to the target state node, action unit, tool, area of ​​action and confirmation result.

[0131] In one embodiment, candidate update data is generated when the number of valid feedback samples corresponding to the same target relationship reaches a preset minimum, and the proportion of samples with the same teacher confirmation result reaches a preset percentage. The preset minimum number can be, for example, 10, and the preset percentage can be, for example, 80%. These values ​​are merely examples and can be adjusted according to the amount of training data.

[0132] Candidate update data may include: increasing or decreasing the weight of state constraint edges; adding or deleting defect mapping relationships; adjusting defect risk levels; modifying the feedback text returned to the virtual reality terminal; and subdividing the original paint layer state node into multiple state nodes.

[0133] Furthermore, the historical confirmation samples corresponding to the candidate updated data are used for verification. If the consistency between the action judgment result after adopting the candidate updated data and the teacher's confirmation result is higher than before the update, and the basic procedure sequence has not changed, then the candidate updated data is determined to have passed the verification, and a new map version is generated.

[0134] It should be noted that routine practical feedback only updates the runtime state constraints, defect mapping relationships, risk levels, and feedback text; it does not directly change the sequence of basic processes such as spare tire preparation, puttying, coating, air drying, grinding, and polishing. Data involving changes to the relationships of basic processes is transferred to manual review.

[0135] Optionally, after a new map version is put into operation, its action misjudgment rate and query anomaly rate are monitored. When the new map version meets the preset operating conditions, the version is retained; when the action misjudgment rate increases or the query anomaly rate exceeds the preset threshold, the map version before the update is restored.

[0136] In one specific embodiment, the initial knowledge graph is set with: practical state - allowed execution - grinding action.

[0137] During the virtual reality training, the system collected multiple sets of practical records of grinding actions performed under actual conditions. Some records were deemed valid by the system, but the corresponding physical test records or teacher confirmation results showed that the paint layer adhered or the paint film was scratched.

[0138] Polymer analysis revealed that the defective samples mainly corresponded to those with short surface drying times, thick paint layers, or high humidity during air drying. Although the paint layer had formed a film on its surface and left no obvious marks when lightly touched, it had not yet reached a state of complete drying from the surface inwards.

[0139] Therefore, the system generates detailed status node data, subdividing the original surface-drying state into: surface-drying not fully dried state; fully dried but not meeting grinding conditions state; and grinding-ready state.

[0140] The following settings are established: Surface dry but not fully dry – grinding action prohibited; Surface dry but not fully dry – may lead to adhesion or paint film damage; Fully dry but not meeting grinding conditions – grinding action prohibited; Grinding is permitted.

[0141] Accordingly, the above-mentioned status nodes are configured with status attributes such as surface drying time, actual drying time, shade drying time, paint layer thickness, and ambient humidity.

[0142] The refined state relationships are verified using historical confirmation samples. When the consistency between the action judgment result and the teacher confirmation result increases and the basic process sequence remains unchanged, a new version of the knowledge graph is generated and made available for use by the virtual reality terminal.

[0143] Through the above process, the knowledge graph can adjust the state division and constraint relationship based on the confirmed practical feedback, forming a dynamic closed loop of "practical verification - feedback collection - relationship update - re-verification".

[0144] like Figure 2 As shown, this embodiment provides a system for constructing a dynamic knowledge graph of lacquer decoration techniques, including a data acquisition and standardization module, a graph construction module, a virtual reality verification interface module, and a feedback and update module; wherein, the graph construction module includes a process ontology construction module, an action state mapping module, a constraint defect modeling module, and a graph generation and storage module; the virtual reality verification module is implemented by the virtual reality verification interface module.

[0145] The data acquisition standardization module is used to collect process text data, process image data, demonstration operation video data, and virtual reality operation log data, and to standardize terminology, merge synonyms, and assign process numbers, tool numbers, area numbers, and paint layer status numbers.

[0146] Optionally, the data acquisition standardization module is also used to convert professional data such as surface drying, surface drying time, actual drying, actual drying time, shade drying time, ambient humidity, and paint layer thickness into unified data fields.

[0147] The process ontology construction module is used to construct the lacquer decoration process ontology, which includes process class, action unit class, paint layer state class, tool class, material class, operation area class and defect class, and extracts basic entity relationships from process sample data.

[0148] Furthermore, the process ontology construction module is also used to construct a basic process flow skeleton and set access permissions for the basic process flow skeleton that are different from those for the updatable and runnable relationships.

[0149] The action state mapping module is used to break down process nodes into action unit nodes, configure the preceding paint layer state, allowed tools, allowed action area, action parameter range and subsequent paint layer state for each action unit node, and generate state transition relationships.

[0150] The constraint defect modeling module is used to establish state constraints, tool constraints, region constraints and time constraints based on state transition relationships, and to establish defect mapping relationships for practical events that do not meet the corresponding constraints.

[0151] Correspondingly, the constraint defect modeling module is also used to generate action legality results, reasons for violations, defect risk levels, and corresponding feedback text.

[0152] The graph generation and storage module is used to write process nodes, action unit nodes, paint layer state nodes, tool nodes, material nodes, operation area nodes, and defect nodes, as well as the corresponding basic entity relationships, state transition relationships, action constraint relationships, and defect mapping relationships into the graph database.

[0153] Furthermore, the map generation and storage module is used to save the current map version, historical map versions, and update records for each version.

[0154] The virtual reality verification interface module is used to receive the practical events uploaded by the virtual reality terminal, call the knowledge graph of lacquer decoration techniques for querying and verification, and return the legality result of the action, the allowed next action, the reason for the violation, the defect risk, and the correction suggestions to the virtual reality terminal.

[0155] Optionally, the virtual reality verification interface module can be deployed in a server or edge computing device that communicates with the virtual reality terminal.

[0156] The feedback update module is used to obtain the operation logs output by the virtual reality verification interface module and the confirmation, correction or supplement results output by the teacher's annotation terminal.

[0157] The feedback update module aggregates feedback data corresponding to the same target state relationship or defect mapping relationship, generates candidate update data when the preset update conditions are met, and verifies it using historical confirmation samples.

[0158] When candidate update data passes verification, the feedback update module sends it to the map generation and storage module to generate a new map version; when candidate update data fails verification, the current map version remains unchanged.

[0159] Optionally, the feedback update module is also used to monitor new map versions and, if the new map version does not meet the preset operating conditions, control the map generation and storage module to restore the map version before the update.

[0160] Specifically, the data flow between modules is as follows: the data acquisition and standardization module sends process sample data to the process ontology construction module; the process ontology construction module sends basic entities and relationships to the action state mapping module; the action state mapping module sends state transition relationships to the constraint defect modeling module; the graph generation and storage module receives and stores various nodes and relationships; the virtual reality verification interface module calls the current graph version from the graph generation and storage module and receives practical events uploaded by the virtual reality terminal; the feedback update module receives practical logs and teacher annotation results, and performs controlled updates to the graph relationships, thereby forming a dynamic knowledge graph system.

[0161] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any ordinary changes and substitutions made by those skilled in the art within the scope of the technical solution of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for constructing a dynamic knowledge graph of lacquer decoration techniques, characterized in that, Includes the following steps: S1. Collect lacquer decoration process data and standardize the lacquer decoration process data to obtain process sample data; S2. Construct the lacquer decoration process entity based on the process sample data, and extract the basic entity relationships from the process sample data; S3. The process nodes in the lacquer finishing process body are broken down into action unit nodes, and the state transition relationship between the previous lacquer layer state, the action unit node and the subsequent lacquer layer state is established. S4. Establish action constraint relationships and defect mapping relationships for virtual reality practice based on the state transition relationships; S5. Write the basic entity relationship, the state transition relationship, the action constraint relationship and the defect mapping relationship into the graph database to generate a lacquer decoration process knowledge graph that can be called by the virtual reality terminal, and perform action verification on the practical events based on the lacquer decoration process knowledge graph to obtain the graph judgment result. S6. Based on the virtual reality practice log, the graph judgment result and the teacher annotation result, determine the feedback sample, aggregate the feedback sample to generate candidate update data, and update the lacquer decoration process knowledge graph according to the candidate update data when the candidate update data passes the verification.

2. The method for constructing a dynamic knowledge graph of lacquer decoration techniques according to claim 1, characterized in that, The lacquer finishing process data includes process text data, process image data, demonstration operation video data, and virtual reality operation log data; The standardization process includes terminology unification, synonym merging, process numbering, tool numbering, operation area numbering, and paint layer status numbering. The process sample data includes process data, action unit data, paint layer condition data, tool data, material data, operation area data, and defect data.

3. The method for constructing a dynamic knowledge graph of lacquer decoration techniques according to claim 1, characterized in that, Step S2 includes: The lacquer finishing process body is constructed based on the process sample data. The lacquer finishing process body includes process categories, action unit categories, lacquer layer state categories, tool categories, material categories, operation area categories, and defect categories. Basic entities and entity relationships are extracted from the process sample data to obtain basic knowledge triples; Based on the aforementioned basic knowledge triplet, a basic process flow framework is formed, which includes the process sequence relationship and the confirmed core process relationship. The knowledge graph of lacquer finishing process also includes a dynamically updatable operation graph, which includes action unit nodes, lacquer layer state nodes, tool nodes, material nodes, operation area nodes and defect nodes, as well as state transition relationships, action constraint relationships and defect mapping relationships formed by the nodes.

4. The method for constructing a dynamic knowledge graph of lacquer decoration techniques according to claim 1, characterized in that, Step S3 includes: Configure the preceding paint layer state, allowed tools, allowed action area, action parameter range, and subsequent paint layer state for each of the aforementioned action unit nodes; The state transition relationship is established based on the state of the preceding paint layer, the action unit node, and the state of the subsequent paint layer. The paint layer state includes at least one of the following: wet state, surface dry state, fully dry state, grindable state, ground state, polishable state, and polished state.

5. The method for constructing a dynamic knowledge graph of lacquer decoration techniques according to claim 1, characterized in that: The practical event includes action type, tool identifier, area of ​​action, action parameters, action duration, and paint layer condition before the action is performed; The action constraints include state constraints, tool constraints, region constraints, and time constraints; In step S5, the practical event is subjected to action verification, including determining whether the practical event meets the constraints of the preceding paint layer state, allowed tools, allowed action area and time of the corresponding action unit, and determining whether the practical event is legal based on the judgment result.

6. The method for constructing a dynamic knowledge graph of lacquer decoration techniques according to claim 5, characterized in that: The defect mapping relationship is used to associate practical events that do not satisfy the action constraint relationship with the corresponding defect nodes; Based on the graph judgment result, the following information is returned to the virtual reality terminal: the legality result of the action, the next action allowed to be executed under the current paint layer state, the unmet action constraints, the corresponding defect risks and correction suggestions; The defect nodes include at least one of the following: blistering, cracking, delamination, adhesion, paint film scratches, surface scratches, loss of gloss, uneven drying, and pattern misalignment.

7. The method for constructing a dynamic knowledge graph of lacquer decoration techniques according to claim 1, characterized in that, Step S6 includes: The practical event in which the atlas judgment result is inconsistent with the teacher confirmation result in the teacher annotation result is identified as the feedback sample; Based on the target state node, action unit, tool, area of ​​action, and teacher confirmation result, multiple feedback samples pointing to the same target relationship are aggregated. When the number of feedback samples reaches a preset minimum number and the proportion of feedback samples with the same teacher confirmation result reaches a preset ratio, it is determined that the preset update conditions are met, and the candidate update data is generated. The candidate update data includes at least one of the following: state constraint relationship weight adjustment data, defect mapping relationship addition or deletion data, defect risk level adjustment data, feedback text modification data, and paint layer state node refinement data.

8. The method for constructing a dynamic knowledge graph of lacquer decoration techniques according to claim 1, characterized in that, In step S6, the candidate update data is validated, including: An updated graph relationship is formed based on the candidate update data; Using historical confirmed samples, the action judgment results after adopting the updated graph relationship and the action judgment results before the update are obtained respectively; When the consistency between the action judgment result after adopting the updated graph relationship and the teacher's confirmation result is higher than that before the update, and the basic process sequence has not changed, the candidate updated data is determined to pass the verification. The knowledge graph of lacquer decoration techniques is updated based on the verified candidate update data, and a new version of the graph is generated. The action misjudgment rate and query anomaly rate of the new map version are monitored. When the action misjudgment rate of the new map version is higher than the action misjudgment rate before the update, or the query anomaly rate exceeds a preset anomaly threshold, the map version before the update is restored.

9. The method for constructing a dynamic knowledge graph of lacquer decoration techniques according to claim 3, characterized in that: When the feedback sample indicates that the original paint layer state node cannot distinguish different action execution conditions, the paint layer state node refinement data is generated based on the surface drying time, actual drying time, shade drying time, paint layer thickness and environmental humidity. The original paint layer state node is subdivided into surface-dry but not fully dried state, fully dried but not reaching grinding conditions state and grinding state, and action constraint relationship and defect mapping relationship between the subdivided paint layer state node and grinding action are established respectively. Candidate update data that are verified and do not involve changes in the sequence relationship of basic processes in the basic process flow framework are used to update the operation map; When the candidate update data involves changes to the basic process sequence relationship, the candidate update data is manually reviewed, and after the manual review is passed, the basic process skeleton is updated according to the approved candidate update data.

10. A system for constructing a dynamic knowledge graph of lacquer decoration techniques, characterized in that, include: The data acquisition and standardization module is used to collect data on lacquer decoration techniques and to standardize the data to obtain process sample data. The graph construction module is used to construct the lacquer decoration process ontology based on the process sample data, extract basic entity relationships, establish state transition relationships, action constraint relationships, and defect mapping relationships between action units and lacquer layer states, and write the basic entity relationships, state transition relationships, action constraint relationships, and defect mapping relationships into the graph database to generate a lacquer decoration process knowledge graph. The virtual reality verification module is used to receive the practical events uploaded by the virtual reality terminal, call the knowledge graph of lacquer decoration process to verify the actions of the practical events, and obtain the graph judgment result; The feedback update module is used to determine feedback samples based on the virtual reality practice log, the graph judgment results, and the teacher annotation results, aggregate the feedback samples to generate candidate update data, and update the lacquer decoration process knowledge graph based on the candidate update data when the candidate update data passes the verification.

Images