An industrial virtual-real symbiosis practical training data processing method based on space calculation

Abstract

The application discloses an industrial virtual-real symbiosis practical training data processing method based on space calculation, and particularly relates to the technical field of virtual-real fusion practical training data processing, and is used for solving the problems that real operation behaviors and virtual interaction behaviors are difficult to be uniformly associated, and process attribution and object relationship are difficult to be accurately identified in the existing virtual-real fusion practical training process. By constructing a space-time semantic base model corresponding to a target practical training scene, the association relationship of personnel entities, equipment entities, tool entities, material entities, process entities and region entities is uniformly defined, and real operation data and virtual interaction data are subjected to event anchor point extraction, state slice generation and semantic homing processing in combination with process constraint rules, so that the effect that the dispersed heterogeneous field behaviors and virtual interaction behaviors are uniformly mapped to the same process semantic framework is achieved, and then the problems that object relationship is unclear, process attribution is inaccurate, and process state is difficult to be consistently expressed in the existing virtual-real fusion practical training are solved.

Description

Technical Field

[0001] This invention relates to the field of virtual-real fusion training data processing technology, specifically a method for processing industrial virtual-real symbiotic training data based on spatial computing. Background Technology

[0002] Existing industrial assembly training systems are typically capable of collecting and displaying some of the trainees' operational behaviors during training by integrating on-site equipment, sensor terminals, display terminals, or virtual interactive interfaces. For example, information such as personnel movement, equipment start / stop, tool usage, and step submission can be obtained through location sensing, equipment status collection, tool recognition, and interface click recording. Some systems also combine on-site operations with virtual teaching interfaces to provide step prompts, result verification, process playback, or training assessments. In terms of application, these technologies primarily serve scenarios such as assembly teaching, skills training, on-the-job drills, and process monitoring, and have already established a certain foundation in improving the visualization of training, enhancing interactivity, and accumulating training data.

[0003] However, in practical applications, existing technologies mostly process data from a single data source, a single object type, or a single process separately. Information collected on-site, such as personnel location, equipment status, tool contact, and material changes, is often stored independently according to its own data format. Interactive information in virtual terminals, such as confirmation, grabbing, placing, and step submission, is also usually recorded separately as interface events or log events. Although data from different sources are all related to the same training process, inconsistencies in object naming conventions, spatial boundary conventions, process division conventions, and time reference conventions make it difficult to establish stable, direct, and continuous correspondences between different records. Especially in assembly-type training scenarios, the same process often involves simultaneous changes in personnel, equipment, tools, materials, and work areas. Without a unified definition of object relationships and semantic constraints for processes, relying solely on discrete events or single-point states often makes it difficult to accurately determine which process a certain behavior belongs to, which object relationship it corresponds to, and what process state it is in, and it is also difficult to consistently express behaviors on both the virtual and physical sides.

[0004] Based on this, a core problem prevalent in existing technologies is that, in the process of virtual-real integrated training, the lack of a unified semantic organizational basis for real operational behaviors and virtual interactive behaviors makes it difficult to accurately correspond and continuously represent scattered and heterogeneous process data within the same semantic framework. This, in turn, affects the identification, understanding, recording, and subsequent analysis of the training process. In other words, existing technologies can often "see" on-site actions and virtual operations separately, but lack a unified way of expressing the object relationships, process attribution, spatial positioning, and constraint states between these behaviors, thus remaining insufficient in process reconstruction, state determination, and structured recording. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides a spatial computing-based method for processing industrial virtual-real coexistence training data, thereby resolving the problems mentioned in the background section.

[0006] To achieve the above objectives, the present invention provides the following technical solution: a method for processing industrial virtual-real coexistence training data based on spatial computing, comprising: S1. Construct a spatiotemporal semantic base model corresponding to the target training scenario, define the relationships between personnel entities, equipment entities, tool entities, material entities, process entities, and area entities, and configure process constraint rules; S2. Collect real operation data and virtual interaction data during the target training process, and extract event anchor points based on changes in equipment status, changes in tool contact, changes in area entry and exit, changes in process status, and changes in constraint triggering. S3. Retrieve the corresponding previous state record, action record, and subsequent state record around the event anchor point to generate a virtual-real coexistence state slice containing entity association, spatial association, process state, and constraint state; S4. Based on the spatiotemporal semantic base model and process constraint rules, perform semantic positioning processing on the virtual and real coexistence state slices to obtain process attribution results, object binding results, spatial station location results, and constraint violation results. S5. Generate a compliant counterfactual trajectory based on process attribution results, object binding results, spatial station location results, and constraint violation results, and extract the causal chain of deviation between the actual trajectory and the compliant counterfactual trajectory; S6. Generate training data atoms based on event anchors, virtual-real coexistence state slices, compliance counterfactual trajectories, and deviation causal chains, and revise the spatiotemporal semantic base model based on the training data atoms.

[0007] Furthermore, S1 includes: Based on equipment ledgers, tooling lists, material master data, process cards, work instructions, workstation layout diagrams, 3D structural data, and on-site survey records, entity master records, relationship records, and process constraint rule records are generated. The object records are aligned according to the equipment ledger number, tooling number, material code, process number and area number. After removing duplicates and filling in static fields that can be uniquely identified by the same source file, the object records are written into the spatiotemporal semantic base model. Configure process constraint rules for process entities. Process constraint rules include allowed areas, allowed equipment access, allowed tool use, allowed material retrieval, prerequisite completion requirements, and prohibited action boundaries.

[0008] Furthermore, S2 includes: After the current version of the spatiotemporal semantic base model and process constraint rules are submitted and take effect, the acquisition channel will be time-corrected according to the unified time synchronization benchmark. The acquired records are grouped by entity identifier and sorted by acquisition time, then aligned and denoised according to a unified time synchronization standard, and the records corresponding to the continuous records are supplemented.

[0009] Furthermore, S2 also includes: Extract event anchor points based on changes in equipment status, tool contact, area entry / exit, process status, and constraint triggering. For each extracted and established event anchor point, an event anchor point record is generated. The event anchor point record includes the event anchor point identifier, event anchor point type, corresponding entity identifier, corresponding process entity, corresponding region entity, start time, end time, source of establishment basis, and current batch version number.

[0010] Furthermore, S3 includes: For each event anchor record, retrieve the preceding state record, action record, and following state record; Grouped by current batch version number, corresponding process entity, and corresponding entity identifier, and aligned by collection time; The event anchor points are used to record the start and end times to divide the pre-state interval, action interval, and post-state interval; Based on the entity master record and relationship record, as well as the previous state record, action record and subsequent state record in the spatiotemporal semantic base model, a virtual-real coexistence state slice containing entity relationship, spatial relationship, process state and constraint state is generated.

[0011] Furthermore, S4 includes: When performing semantic relocation processing on virtual-real coexistence state slices, the slices are first grouped and organized according to the current batch version number, the corresponding process entity, and the corresponding entity identifier. Then, the entity master record, the association record, and the process constraint rules are retrieved, and the unique relocation is completed in the order of process attribution result, object binding result, spatial station location result, and constraint violation result.

[0012] Furthermore, S4 also includes: The object binding result is determined based on the consistency, missing, and misalignment of the object relationship group and the effective association record; The location of the space station is determined based on the location coordinates and regional affiliation, combined with the standard placement location, the stable process action surface, and the standard execution boundary. Constraint violation results are recorded based on the triggering and release status of each constraint in the preceding, following, and subsequent states.

[0013] Furthermore, S5 includes: Based on the process attribution results, object binding results, spatial station location results, and constraint violation results, a compliance counterfactual trajectory is generated in the priority order of constraint violation correction, spatial station location correction, restoration of object binding, and process advancement. The actual trajectory is then aligned with the compliant counterfactual trajectory to extract the deviation causal chain, which includes the deviation start point, deviation trigger, deviation transmission segment, and deviation end point.

[0014] Furthermore, S6 includes: Group the data according to the current batch version number, the corresponding process entity, the corresponding entity identifier, and the trajectory comparison identifier. Organize the event anchor points, virtual and real coexistence state slices, compliance counterfactual trajectories, and deviation causal chains according to the event sequence. Verify the correspondence between the event anchor point identifier, slice identifier, return result identifier, and trajectory comparison identifier to generate training data atoms. The atomic starting boundary is determined based on the event anchor point, and the atomic ending boundary is determined based on the event anchor point ending time and the deviation causal chain ending time. The standard compliance evolution chain is written based on the compliance counterfactual trajectory, and the deviation starting point, deviation cause, deviation transmission segment and deviation ending point are written based on the deviation causal chain. When the preset correction threshold is met, based on the abnormal object binding result, abnormal space station result, abnormal area entry boundary and continuous triggering of constraint items corresponding to the deviation cause, the correction object in the association record, area entity boundary record, standard placement position record, stable process action surface record and process constraint rule record is determined and corrected.

[0015] Compared with the prior art, the present invention has the following beneficial effects: 1. By constructing a spatiotemporal semantic foundation model corresponding to the target training scenario, the relationships between personnel entities, equipment entities, tool entities, material entities, process entities, and area entities are uniformly defined. Combined with process constraint rules, event anchor points are extracted, state slices are generated, and semantic positioning is performed on real operation data and virtual interaction data. This achieves the effect of uniformly mapping the scattered, heterogeneous, and difficult-to-correspond on-site behaviors and virtual interaction behaviors in the training process to the same process semantic framework. This solves the technical problems in existing virtual-real fusion training where the object relationships are unclear, the process attribution is inaccurate, and the process state is difficult to express consistently, making it difficult to accurately identify, continuously understand, and structurally represent the training process.

[0016] 2. By generating compliant counterfactual trajectories based on process attribution results, object binding results, spatial station location results, and constraint violation results, the deviation causal chain between the actual trajectory and the compliant counterfactual trajectory is extracted, and further training data atoms are formed to correct the spatiotemporal semantic base model. This achieves the effect of traceable analysis and continuous optimization of the causes, transmission process, and correction direction of abnormal deviations in the training process, thereby improving the accuracy and continuity of training process evaluation, problem location, and model adaptive correction. Attached Figure Description

[0017] Figure 1 This is a schematic diagram of the overall process of a spatial computing-based industrial virtual-real symbiotic training data processing method; Figure 2 This diagram illustrates the collection of real operational data and virtual interactive data, as well as the extraction of event anchor points. Figure 3 A schematic diagram of generating virtual-real coexistence state slices driven by event anchors; Figure 4 A schematic diagram of semantic repositioning processing for slices in a virtual-real coexistence state. Detailed Implementation

[0018] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0019] Example: Combined with Appendix Figures 1-4 This embodiment provides a method for processing industrial virtual-real coexistence training data based on spatial computing, including: S1. Construct a spatiotemporal semantic foundation model corresponding to the target training scenario, define the relationships between personnel entities, equipment entities, tool entities, material entities, process entities, and area entities, and configure process constraint rules. The specific implementation is as follows: First, a spatiotemporal semantic base model is established on the actual assembly line corresponding to the target training scenario. The target training scenario is preferably taken from an industrial training unit with a stable workstation layout, a fixed equipment list, and a clear assembly sequence, such as the loading station, pre-assembly station, fastening station, inspection station, and off-line station in the reducer assembly training line. The execution is carried out by the modeling and maintenance personnel at the training site in cooperation with the process personnel during the production line downtime for cleanup. The cleanup interval is preferably the data freeze window from the most recent process version finalization to the start of the current training batch. The spatiotemporal semantic base model consists of entity master records, relational records, and process constraint rule records. The unique object definition is used to define personnel entities, equipment entities, tool entities, material entities, process entities, and area entities. The association record is used to define the effective correspondence between each entity. The process constraint rule record is used to define the boundary conditions that each process entity is allowed to execute and the conditions that are prohibited from exceeding the boundary within the current training batch. All three types of records include version number, source number, submission time, record number, and effective status. The system is considered complete when all six types of entities under the current effective version have formed effective entity master records, all necessary associations have formed effective association records, and the corresponding process entities have formed process constraint rule records. If any type is missing, the current effective version cannot be submitted.

[0020] The spatiotemporal semantic foundation model is not simply a 3D display file, but a recordable model built around the on-site objects, object locations, object contact boundaries, object action order, and process execution boundaries required for subsequent practical training. Its establishment is based on basic records obtained item by item from existing on-site data. Data sources include equipment ledgers, tooling lists, material master data, process cards, work instructions, workstation layout diagrams, 3D structural data, and on-site surveying records. Names are uniformly adopted using names already in use on-site, length units are uniformly in millimeters, angle units are uniformly in degrees, and time units are uniformly... The position is uniformly measured in seconds, and the location information is uniformly placed under the same scene coordinate system. The same scene coordinate system is preferably recorded with the ground reference point on the entrance side of the target training scene as the origin, the production line extension direction as the longitudinal direction, the workstation lateral expansion direction as the lateral direction, and the ground normal as the height direction. The version system is uniformly locked by the combination of "production line number + training unit number + process version number + modeling version number". In case of data conflict, the signed on-site survey record takes precedence over the signed drawing, and the signed drawing takes precedence over the historical ledger. Unsigned data shall not be used as the basis for writing.

[0021] Before any object record enters the spatiotemporal semantic base model, it is first aligned by the unique identifier field, then duplicates are removed, and finally, default completion is performed within the allowed boundaries. Alignment uses equipment ledger number, tooling number, material code, process number, and area number as primary keys; the name field is only used for post-alignment verification and does not participate in primary key matching. Duplicates are removed based on the primary key being consistent with the installation location, workstation, and source version. Records with consistent primary keys but inconsistent locations or source versions are transferred to pending verification. Completion is limited to static fields that can be directly and uniquely identified by the same source file and do not involve permissions, order, boundaries, or manual confirmation. Direct unique identification refers to the same field of the same object. It can be independently determined by a single signed document and does not need to be inferred from other documents. For example, the equipment installation reference surface number, workstation boundary line, and standard material placement position can be supplemented, but personnel entity identification, process entity sequence number, area entity safety boundary, and previous completion requirements cannot be automatically supplemented. In cases of the same name but different code, the same code but different position, the deviation between the drawing size and the on-site measurement exceeds 3 mm, or the process card and the corresponding process in the work instruction are inconsistent, the current object should be stopped from being processed and transferred to the pending verification record. The current object should not be written into the entity master record, while other confirmed objects can continue to be processed. The pending verification record is stored separately in the verification area of ​​the current effective version. After manual review and correction, it should be resubmitted in the original alignment, deduplication, and supplementation order.

[0022] Based on the definition of personnel entities, this step establishes personnel entity records according to their actual on-duty roles. Each personnel entity corresponds to a natural person who enters the target training scenario to participate in operation, supervision, and instruction. The on-site sources are the shift schedule, training list, and on-duty identification record. Personnel entity filing is completed 15 minutes before the start of the current training batch. The record content includes at least the personnel entity identifier, job category, permitted entry area, permitted equipment access range, and permitted work procedure range. The permitted scope is uniformly defined as the access boundary under the condition that the current effective version, the current training batch, and the corresponding job authorization are all met simultaneously. The permitted entry area is obtained by taking the intersection of on-site access control permissions and workstation operation lines. The permitted equipment access range is obtained by taking the intersection of the job authorization table and the safety briefing table. The permitted work procedure range is obtained by associating the corresponding page number of the shift training plan and the work instruction.

[0023] Based on the definition of equipment entities, each equipment entity corresponds to a real equipment object that performs a technological function within the target training scenario. These include assembly tables, press machines, tightening machines, testing tables, and conveying units. The data is obtained from equipment ledgers, inspection records, and on-site nameplates. During compilation, equipment that has passed the most recent inspection and is in a "ready to operate" state is considered a valid object. "Ready to operate" means that before the start of the current batch, the equipment is powered on and operational, not tagged as out of service, and not in a maintenance isolation state. The equipment entity identifier, installation location, functional area, contact surface location, start / stop boundary, and associated workstation are recorded. The installation location... The unified coordinates corresponding to the center of the equipment installation reference plane are used for entry. The area of ​​action is determined by the maximum controlled motion envelope of the equipment actuator and the affected range of the workpiece. It does not include the maintenance deployment position and the manual contact area in the power-off static state. The contact surface position is recorded by the stable process action surface corresponding to the reference plane of the drawing. If the drawing is missing, the boundary of the force surface measured on site is recorded. The protective cover surface and decorative surface are not included. The stable process action surface corresponds to the actual action surface boundary of the equipment entity or tool entity that continuously performs process action under the current process version and does not change due to changes in protective components, decorative components or maintenance deployment position.

[0024] Regarding the definition of tool entities, a tool entity corresponds to a controlled tool object held and used by personnel entities to act on equipment entities or material entities during the training process, such as a torque wrench, locating pin, or assembly fixture. The source of the tool is the tooling list and requisition record. When organizing, first match the tool number with the workstation, and then write the tool entity identifier, the workstation it belongs to, the standard placement position, the allowed contact objects, and the preconditions for use. The standard placement position is uniquely determined by the tooling plate hole position, limit block, material box boundary, or ground baseline. Non-fixed temporary placement positions are not included. Allowed contact objects are merged item by item according to the corresponding steps of the tool in the work instruction. The main tool and the spare tools that have been signed in the current process version can be written at the same time. Unsigned cross-workstation borrowed tools are not included in the current effective version.

[0025] Based on the definition of material entities, each material entity corresponds to a part, semi-finished product, or auxiliary material that enters the training assembly process and participates in process changes. The source of the material entity is the material master data, material requisition form, and process card. First, the material code is aligned with the process number, and then the material entity identifier, material type, entry station, exit station, standard placement position, and allowed assembly orientation are written. The allowed assembly orientation is determined by the reference plane of the drawing and the direction requirements in the process card. When multiple specifications with the same code are mixed together on site, the writing of the material entity is suspended and it is put into pending entry. Pending entry will only be resumed after the specification correspondence is completed in the same batch.

[0026] Based on the definition of process entities, each process entity corresponds to a field operation unit that must be completed in a predetermined order during training. The process entity is not replaced by the training chapter name. The source of the process entity is process card, work instruction book and cycle time table. When organizing, a unique process entity record is established according to the process card order. The process entity identifier, preceding process, following process, execution station, associated equipment, associated tools, associated materials and completion confirmation criteria are written. The completion confirmation criteria only record the on-site confirmation basis that is visible, verifiable and traceable when the current process is completed. The preceding completion requirements only record the completion conditions of the preceding process that must be called when the subsequent process is admitted. The two are recorded separately and must not be mixed.

[0027] Based on the definition of regional entities, regional entities correspond to work areas, waiting areas, equipment operation areas, personnel passage areas, and safety isolation areas within the target training scenario that have fixed spatial boundaries and technological meanings. The sources are workstation layout diagrams, on-site landmark lines, and surveying records. When organizing, first determine the regional boundaries according to ground markings and equipment outlines, and then write in the regional entity identifier, boundary coordinates, allowed entry objects, prohibited stay objects, and associated workstations. Regional entity boundaries must not overlap within the same version. If the markings are unclear, they should be verified with the most recently signed layout diagram and on-site photos before being written in.

[0028] After completing the above object organization, the relationships are defined according to the relationships in the independent claims. The relationships include at least the entry correspondence between personnel entities and area entities, the holding correspondence between personnel entities and tool entities, the contact correspondence between tool entities and equipment entities, the assembly correspondence between material entities and process entities, the function correspondence between equipment entities and process entities, and the execution correspondence between area entities and process entities. Among them, the entry correspondence is confirmed by the simultaneous establishment of job authorization and area access; the holding correspondence is confirmed by the simultaneous establishment of requisition records and work instructions; the contact correspondence is confirmed by the actual correspondence of the stable process action surface; the assembly correspondence is confirmed by the fact that the material is necessary for the completion of the process and is listed in the process card; the function correspondence is confirmed by the fact that the equipment directly participates in the process change of the process; and the execution correspondence is the unique corresponding boundary of the actual area where the process occurs. In the current effective version, one process entity is only allowed to correspond to one main execution area.

[0029] Next, process constraint rules are configured. These rules correspond to the boundary conditions that each process entity is allowed to execute and the conditions that prohibit exceeding those boundaries in the actual training environment. The configuration sources are process cards, work instructions, safety operating procedures, and quality control point cards. Within the current step, only the static and sequential constraints required to complete the process entity are written. Each process constraint rule must specify at least the corresponding process entity, allowed entry area, allowed contact with equipment, allowed use of tools, allowed material retrieval, prerequisite completion requirements, prohibited action boundaries, and submission effective period. Prohibited action boundaries include prohibited entry boundaries, prohibited contact boundaries, prohibited crossing boundaries, and prohibited premature action boundaries. In case of conflict between the safety operating procedures and the process card, the safety operating procedures shall prevail. The submission effective period starts from the start time of the current training batch and ends at the end time of the current training batch. It cannot be retrieved before the start time and automatically becomes invalid and reserved after the end time.

[0030] Upon completion of this step, a master record of the spatiotemporal semantic base model and an auxiliary record of process constraint rules are formed. These are uniformly stored in the main database of the training scenario as structured record files. Only the currently effective version is made available to the next stage for use. The minimum set of fields for the submission message includes scenario number, object identifier, association type, rule identifier, version number, and submission time. The minimum set of fields for the returned result includes reception status, effectiveness status, and trace number. The reception status only indicates whether the main database has received the submission content, and the effectiveness status only indicates whether the content is allowed to be used by the next stage. The maximum submission delay is preferably 30 seconds. Concurrent submissions are written serially in groups by scenario number. The number of retries is preferably no more than 2, and the retry interval is preferably 10 seconds. Old versions must not be overwritten or deleted before the current training batch ends.

[0031] In this step, the safety and compliance boundaries are limited to not writing permissions for personnel beyond the scope of on-site authorization, not writing unsigned process versions, not collecting sensitive personal information unrelated to the training, and not solidifying temporary detour practices on the official production line into process constraint rules.

[0032] In a preferred embodiment, taking a reducer assembly training line as an example, the scenario can be set to 5 workstations, 12 pieces of equipment, 18 tools, 26 kinds of materials, 9 processes, and 11 areas. The coordinate mapping error is controlled within ±2 mm. A total of 43 process constraint rules are formed. Among them, the fastening process limits the allowed entry area to the inner boundary of the fastening workstation, the allowed contact equipment to be the electric tightening machine body and the clamping platform, the allowed tool to be the corresponding numbered sleeve, the prerequisite completion requirement is that the housing positioning pin is in place, and the prohibited action boundary is that the personnel entity crosses the tightening machine's rotation coverage area. When an alternative approach is adopted, the boundary of the area entity does not depend on the fixed landmark line, but is formed by the equipment outline plus a safety distance to form the boundary record. As long as the spatiotemporal semantic base model is still written with the same version caliber, and the names, formation order, and record boundaries of personnel entities, equipment entities, tool entities, material entities, process entities, area entities, and process constraint rules are kept consistent, the implementation result consistent with the technical concept of this step can be obtained.

[0033] S2. Collect real operation data and virtual interaction data during the target training process, and extract event anchor points based on changes in equipment status, tool contact, area entry and exit, process status, and constraint triggering. The specific implementation is as follows: After the current version of the spatiotemporal semantic base model and process constraint rules have been submitted and implemented, real operation data and virtual interaction data are collected item by item around the target training process. The collection locations correspond to each execution station, equipment action area, personnel passage area, and safety isolation area within the target training scenario. The execution entity is the on-site data collection and maintenance personnel. After the channel verification is completed before the start of the training batch, the data collection system continuously executes the data collection. The collection interval starts from the start time of the current training batch and stops at the end time of the current training batch. In the meantime, only when the equipment is powered off, the sensor channel is disconnected, the virtual terminal exits the current training scenario, or the unified time synchronization benchmark is mismatched or exceeds the limit, the corresponding channel will be transferred to the pending verification record, without changing the continuous collection of other channels. Among them, the real operation data corresponds to the observable changes of personnel entities, equipment entities, tool entities, material entities, process entities, and area entities on site, including at least equipment operation status records, equipment execution component position records, tool holding records, and tool contact records. Records include personnel location records, area entry and exit records, process execution confirmation records, and constraint trigger records. Equipment operation status records preferably use equipment controller status words as the primary record source, with start / stop receipts, position switches, and position changes as verification sources. Tool holding records preferably use the continuous correlation between the tool position record showing the tool leaving the standard placement position and the personnel position record showing the corresponding personnel entering the allowed entry area as the primary record source. Tool contact records preferably use contact signals collected by contact sensing terminals as the primary record source. Personnel location records preferably use position sensing terminals as the primary record source. Process execution confirmation records preferably use the corresponding records of start confirmation, process arrival confirmation, step submission confirmation, and completion confirmation formed by the current process entity during on-site execution as the primary record source. Among them, the completion confirmation record is only used as the basis for determining whether the process has been completed. Constraint trigger records preferably use the out-of-bounds results after comparing each constraint rule of the current process.

[0034] Virtual interaction data corresponds to records of personnel entities' confirmations, pointing, grabbing, placement, step submissions, alarm responses, and interface switching within the virtual terminal. Virtual interaction records are timed according to the moment of event occurrence. The virtual interaction acquisition rhythm is only used for continuous validity determination. Confirmation is established when the virtual confirmation control is continuously triggered for one full virtual interaction acquisition rhythm. Pointing is established when the line of sight or gesture continuously falls on the same virtual object for two full virtual interaction acquisition rhythms. Grabbing is established when the virtual object is selected and continuously follows the current personnel entity's operation trajectory for two full virtual interaction acquisition rhythms. Placement is established when the grabbed virtual object enters the target virtual area and is released from following. Step submission is established when the virtual submission control of the corresponding process entity is triggered and not withdrawn. Alarm response is established when the virtual alarm is confirmed closed or the required corrective action is registered. Interface switching is established when the virtual terminal's display page continuously switches and remains unchanged for one full virtual interaction acquisition rhythm. The names and definitions of virtual interaction records maintain a one-to-one correspondence with personnel entity identifiers, tool entity identifiers, process entity identifiers, and area entity identifiers, and the names and definitions will not be changed.

[0035] To ensure continuous comparison of real operational data and virtual interactive data under the same time reference, the device controller, position sensing terminal, contact sensing terminal, and virtual terminal are time-calibrated using a unified time reference before data collection begins. The unified time reference is preferably taken from the master clock source that has been verified within the target training scenario. The first calibration is completed before the batch begins, and the data is checked at fixed intervals during batch execution. Channels with a time deviation exceeding 10 milliseconds are not included in the current batch's effective data collection. If a channel that has already entered effective data collection exceeds the limit during the batch's check, only the writing of new records for that channel is paused and the data is transferred to the pending verification record, while the remaining channels continue to collect data. After calibration, the records of each channel are uniformly written to the collection time. Preferably, the data collection rhythm of device status quantities can be set to 20 milliseconds to 100 milliseconds, the data collection rhythm of position quantities can be set to 20 milliseconds to 50 milliseconds, and the virtual interactive records are timed immediately upon the occurrence of the event.

[0036] The acquired records are first grouped by entity identifier and sorted by acquisition time, then aligned according to a unified timing benchmark, and then denoised for obvious spikes and momentary data loss. Finally, continuous records are supplemented within the allowable boundaries. The denoising order is fixed after alignment and before supplementation. Among them, when three consecutive sampling points of the position quantity show a sudden jump to one side and then return to the original stable range, and the displacement amplitude in the middle exceeds the maximum allowable movement distance of the current object under the corresponding acquisition rhythm, it is judged as a spike. The maximum allowable movement distance is calculated according to the preset movement upper limit of personnel entities, tool entities, and material entities and the corresponding acquisition rhythm. The preset movement upper limit is preferably taken from the object movement boundary configuration record in the current effective version. Records of equipment status quantities that return to the original state before the single flip is judged as momentary jitter and are not considered as valid changes. Denoising only cancels their participation in event anchor point extraction, without deleting the original record. Supplementation is limited to the same channel before and after the same time. For continuously held fields where the duration of the gap does not exceed 3 collection cycles, the continuously held fields are limited to equipment stable status fields, fixed start / stop fields, and non-position status fields that have not changed after being established. Personnel position records, tool contact records, process execution confirmation records, constraint trigger records, and all virtual interaction records must not be automatically filled in, and cross-channel results must not be borrowed to fill in the gap. Among them, the preset priority of equipment operation status records is preferably that the equipment controller status word takes precedence over start / stop receipts, start / stop receipts take precedence over position switches, position switches take precedence over point changes, and the main record source of other records takes precedence over the verification source. In cases where entity identifiers are missing, multiple entities correspond to the same time, collection time is reversed, the main record source conflicts with the verification source and cannot be resolved according to the preset priority, or the virtual terminal record is inconsistent with the current personnel entity, the record will be stopped from being processed and transferred to the pending verification record, and will not be filled in by presumption.

[0037] After continuous collection of real operation data and virtual interaction data, event anchor points are extracted based on changes in equipment status, object contact, area entry / exit, process status, and constraint triggering. Each event anchor point corresponds to a single established event formed after continuous condition confirmation within the target training process. These anchor points are used to mark process advancement boundaries, operation switching boundaries, or constraint boundary crossings. Their start time is the time when the establishment condition is first met, and their end time is the time when the corresponding established state is released. For instantaneous events, the start and end times are allowed to be the same. Equipment status changes are recorded as continuous changes from shutdown to operation, from operation to completion, from preparation to execution, and from execution to completion. As a condition for establishment, the aforementioned state name is based on the set of valid states configured for the current equipment entity under the current process version. If a certain state does not exist, this state change type is not applicable. The changed state must be maintained continuously for at least two equipment state acquisition cycles to be considered valid. Object contact change is based on the tool contact record showing contact establishment, contact disengagement, or contact object switching. Contact establishment is defined as the contact signal corresponding to the stable process action surface remaining active for two acquisition cycles; contact disengagement is defined as the original contact signal disappearing for two acquisition cycles; contact object switching is defined as the same tool entity disengaging from the original contact object and then re-establishing contact with a new contact object, with different object identifiers before and after. (Slippage / rubbing) Instantaneous passage is not counted; changes in area entry and exit are considered valid only when the boundary judgment point of a personnel entity, tool entity, or material entity crosses the boundary of the area entity. Specifically, the human body center judgment point is preferably the two-dimensional projection point corresponding to the center of the human pelvis, and the geometric center judgment point is preferably the geometric center point of the object's outer boundary. For personnel entities, the human body center judgment point is preferably used as the boundary judgment point; for tool and material entities, the geometric center point is preferably used as the boundary judgment point. Continuous entry into the inner boundary of the area while maintaining rhythmic data collection at at least three positions is considered entry; continuous departure from the outer boundary of the area while maintaining rhythmic data collection at at least three positions is considered exit. Back-and-forth oscillations within the boundary tolerance zone are not counted. The boundary tolerance zone is preferably... The allowable fluctuation range is determined by the superposition of regional boundary mapping error and location perception error; the change of process status is based on the condition that the process execution confirmation record changes from not started to in progress, and from in progress to completed. Not started corresponds to the current process entity not having any valid start confirmation, process in place confirmation, step submission confirmation, or completion confirmation record, and the preceding process has met the previous completion requirements. In progress corresponds to the current process entity having the first valid start confirmation, process in place confirmation, or step submission confirmation record, but not yet having a completion confirmation record. Completed corresponds to the state where the completion confirmation record has appeared and the preceding process is not in an incomplete state.Constraint trigger changes are established when a constraint trigger record changes from non-triggered to triggered or from triggered to released. Constraint trigger records are obtained by comparing each item in the process constraint rules against allowed entry areas, allowed equipment contact, allowed tool use, allowed material retrieval, prerequisite completion requirements, and prohibited action boundaries. If any item is not met, a corresponding constraint trigger record is created. If multiple items are not met simultaneously, multiple constraint trigger records are created. When the corresponding items are restored to their original state, corresponding constraint release records are created.

[0038] Each event anchor point, once established, generates an event anchor point record. This record includes at least the event anchor point identifier, event anchor point type, corresponding entity identifier, corresponding process entity, corresponding region entity, start time, end time, source of establishment basis, and current batch version number. The current batch version number preferably uses a combination of the process version number, modeling version number, and rule version number corresponding to the current training batch. The event anchor point identifier is generated as "current training batch + process entity identifier + event sequence number." The event sequence number is ordered according to the start time of each event anchor point under the same process entity. The records are arranged in the following order: equipment status change, object contact change, area entry / exit change, process status change, and constraint trigger change. For records with the same corresponding entity, the same event anchor point type, the same start time and end time, the source of the basis for establishment, and the current batch version number, only one effective record is retained. When re-extracting records, records with completely identical fields are retained as idempotent records, while those with inconsistencies are transferred to records pending verification and are not overwritten. Event anchor point records are written to the event anchor point record area in a structured record file for the next step to retrieve previous state records, action records, and subsequent state records around the event anchor point.

[0039] In a preferred embodiment, taking the fastening station of the reducer assembly training line as an example, the equipment status acquisition rhythm is 50 milliseconds, the position acquisition rhythm is 20 milliseconds, and the virtual interaction confirmation rhythm is 50 milliseconds. After a person enters the inner area of ​​the fastening station and remains there for 60 milliseconds, an area entry event anchor point is formed. Subsequently, after the tool establishes contact with the head of the fastener and remains there for 100 milliseconds, an object contact event anchor point is formed. After the tightening machine changes from ready to execute and remains there for 100 milliseconds, an equipment status change event anchor point is formed. After the process execution confirmation record shows a specified number of cycles and a qualified prompt, a process status change event anchor point is formed. Alternatively, the area entry and exit changes can be recorded by a fixed visual acquisition device combined with the area boundary line, in addition to using a position sensing terminal. As long as the same scene coordinate caliber, the same entity identification caliber, the same timing reference, and the same continuous judgment boundary are used to extract the event anchor points, the implementation result consistent with the technical concept of this step can be obtained.

[0040] S3. Retrieve the corresponding previous state record, action record, and subsequent state record around the event anchor point to generate a virtual-real coexistence state slice containing entity association, spatial association, process state, and constraint state. Specifically, this is implemented as follows: After the event anchor record has been written into the event anchor record area, the corresponding preceding state record, action record, and following state record are retrieved for each event anchor record. Under the constraints of the spatiotemporal semantic base model and process constraint rules of the currently effective version, a virtual-real coexistence state slice is generated. The execution location corresponds to the execution station, equipment action area, and virtual terminal session area associated with the event anchor record within the target training scenario. The execution subject is the state slice generation process, which is continuously completed during the current training batch. The action period starts after the corresponding event anchor record is formed and ends when the event anchor record completes the slice and is recorded. Among them, the preceding state record corresponds to the continuous record before the start time of the event anchor record, used to represent the stable field state before the occurrence of the event anchor record. The stable field state corresponds to the process state, constraint state, holding correspondence, and contact pair of the core corresponding entity under the process entity to which the current event anchor record belongs. The data retrieved is based on real operational data, including equipment operation status records, equipment execution component position records, tool holding records, tool contact records, personnel position records, area entry and exit records, process execution confirmation records, constraint trigger records, and virtual interaction data associated with corresponding personnel entities. The retrieval interval is fixed by tracing back from the start time of the event anchor record. The pre-state interval is preferably determined by five times the length of the acquisition rhythm corresponding to the main source channel of the current event anchor record. The main source channel of the current event anchor record corresponds to the channel to which the main judgment record of the event anchor record belongs. The starting point of the pre-state record is determined by the start time of the last continuous record that satisfies the stable field state within the pre-state interval. If there is no continuous record that satisfies the stable field state within the pre-state interval, the starting point of the pre-state interval is used as the starting point of the pre-state record.

[0041] Action logs are continuous records between the start and end times of the corresponding event anchor record, used to characterize the process of establishing the event anchor record. When retrieving them, the corresponding entity identifier, corresponding process entity, and corresponding area entity in the event anchor record are used as the search criteria. Records that are consistent with or directly related to the current event anchor record type are continuously extracted from real operation data and virtual interaction data according to the collection time. Direct association is limited to records that participate in the determination of the establishment of the current event anchor record, records that are consistent with the entity identifier corresponding to the current event anchor record, and records that are consistent with the process entity corresponding to the current event anchor record and have on-site constraints on the establishment of the event anchor record.

[0042] The subsequent state record is a continuous record that follows the end time of the corresponding event anchor record and is used to characterize the stable field state after the completion of the event anchor record. The retrieval source is consistent with the previous state record. The retrieval interval is a fixed subsequent state interval extended from the end time of the event anchor record. The subsequent state interval is preferably determined by five times the length of the acquisition rhythm corresponding to the main source channel of the current event anchor record. The end point of the subsequent state record is determined by the end time of the first continuous record that meets the stable field state in the extended interval. If there is no continuous record that meets the stable field state in the extended interval and no subsequent similar event anchor record appears, then the end point of the subsequent state interval is taken as the end point of the subsequent state record. Among them, the similar event anchor record corresponds to the event anchor record with the same event anchor type, the same corresponding process entity, and the same corresponding entity identifier. If a subsequent similar event anchor record is formed again under the same corresponding entity, the same event anchor type, and the same corresponding process entity in the extended interval, then the start time of the subsequent similar event anchor record is taken as the end boundary of this subsequent state record.

[0043] Based on the order of organizing the pre-state records, action records, and post-state records, first group them by the current batch version number, the corresponding process entity, and the corresponding entity identifier, then align them in ascending order by the collection time, and then divide the pre-state interval, action interval, and post-state interval by the start and end times of the event anchor point record. After alignment, only the continuous fields that have been allowed to be filled in the previous step are allowed to use the original filling results in the same channel, and no new filling is added in this step.

[0044] After retrieving the corresponding pre-state record, action record, and post-state record, a virtual-real coexistence state slice is generated. Each virtual-real coexistence state slice corresponds to a unique fragment record of the same event anchor record within the pre-state interval, action interval, and post-state interval. Only one virtual-real coexistence state slice is generated for each event anchor record. Virtual-real coexistence state slices are allowed to have missing virtual interaction data. As long as the main record sources necessary for forming entity associations, spatial associations, process states, and constraint states under the corresponding event anchor record type are complete, the virtual-real coexistence state slice will still be generated. Its recording scope is limited to the associated equipment, associated tools, associated materials, associated areas, and associated personnel under the corresponding process entity of the event anchor record.

[0045] During generation, the entity master record and related relationship record corresponding to the current event anchor point record are first extracted based on the spatiotemporal semantic base model. Then, the corresponding entity identifiers appearing in the previous state record, action record, and subsequent state record are attached to the same segment one by one to form entity relationship. The entity relationship includes at least the holding correspondence between personnel entity and tool entity, the contact correspondence between tool entity and equipment entity or material entity, the assembly correspondence between material entity and process entity, the function correspondence between equipment entity and process entity, and the execution correspondence between area entity and process entity. The confirmation of entity relationship is based on the effective relationship record in the spatiotemporal semantic base model. The actual occurrence in the previous state record, action record, and subsequent state record is used as the presence confirmation in the current virtual-real coexistence state slice. The presence confirmation corresponds to the relevant entity having at least one direct relationship record in any of the previous state interval, action interval, and subsequent state interval that has passed the noise reduction in the previous step, has not been transferred to pending verification, and has not been judged as conflicting.

[0046] Subsequently, spatial relationships are generated. These spatial relationships correspond to the positional relationships, entry relationships, adjacency relationships, and contact relationships of corresponding entities within the current virtual-real coexistence slice under the same scene coordinate caliber. Among them, positional relationships are recorded first when the corresponding entity has positional coordinate records, and when positional coordinates are missing but the region entry and exit records are complete, the region affiliation is recorded. Entry relationships correspond to the relationship of related entities relative to the region entities, such as entering the inner side, being located on the inner side, and exiting the outer side. Adjacency relationships correspond to the close neighbor relationships formed between related entities within the preset adjacency distance boundary of the current process version. Contact relationships correspond to the establishment, maintenance, or dissolution of contact relationships on the stable process action surface.

[0047] Continue generating process states. The process state corresponds to the state progress result of the process entity to which the current event anchor record belongs in the previous state interval, action interval, and subsequent state interval. All are limited to the three state categories of not started, in execution, and completed. If mutually exclusive states appear in the same interval, they are merged in the order of completed before in execution, and in execution before not started. If the priorities are the same, the one with the longer duration is taken as the process state of the interval. If the durations are the same, the one that was last established in the interval is taken as the process state of the interval.

[0048] Regenerate constraint states. Each constraint state corresponds to the current event anchor point record and the satisfaction status of each constraint item of the process constraint rule by the process entity within the pre-state interval, action interval, and post-state interval. Records are made separately for each constraint item and are not merged into a single comprehensive state. Only three recording criteria are used: satisfied, triggered, and released. The pre-state constraint state is determined based on the constraint trigger records corresponding to each constraint item within the pre-state interval. The action constraint state is determined based on the constraint trigger records corresponding to each constraint item within the action interval. The post-state constraint state is determined based on the constraint trigger records corresponding to each constraint item within the post-state interval.

[0049] After the virtual-real coexistence state slice is generated, a slice record is formed. The slice record includes at least the slice identifier, event anchor point identifier, corresponding entity identifier, corresponding process entity, corresponding region entity, previous state record boundary, action record boundary, subsequent state record boundary, entity association relationship, spatial association relationship, process status, previous state constraint status, action constraint status, subsequent state constraint status corresponding to each constraint item, and the current batch version number. The slice identifier is preferably generated according to "current training batch + event anchor point identifier + slice sequence number". The slice sequence number is arranged in ascending order according to the event sequence number corresponding to the event anchor point identifier. The slice record is written to the state slice record area in a structured record file for direct retrieval in the next stage.

[0050] In a preferred embodiment, taking the object contact change event anchor point record of the fastening station in the reducer assembly assembly training line as an example, the pre-state interval can be taken as 250 milliseconds before the start time of the event anchor point record, the action interval can be taken as 100 milliseconds from the start time to the end time of the event anchor point record, and the post-state interval can be taken as 250 milliseconds after the end time of the event anchor point record. Within this segment, a virtual-real coexistence state slice can be formed, which includes the holding correspondence between the personnel entity and the tool entity, the contact correspondence between the tool entity and the fastener, the adjacency relationship between the tool entity and the tightening machine workstation, the fastening process being in execution, and the prohibition of crossing the boundary. Alternatively, the pre-state interval and the post-state interval can also be determined according to a fixed ratio of the corresponding workstation cycle length. As long as the corresponding pre-state record, action record, and post-state record are still retrieved around the same event anchor point record, and entity association relationship, spatial association relationship, process state, and constraint state are generated according to the same process entity caliber, the implementation result consistent with the technical concept of this step can be obtained.

[0051] S4. Based on the spatiotemporal semantic base model and process constraint rules, semantic positioning processing is performed on the virtual-real coexistence state slices to obtain process attribution results, object binding results, spatial station location results, and constraint violation results. The specific implementation is as follows: After the virtual-real coexistence state slice has been written into the state slice record area, based on the spatiotemporal semantic base model and process constraint rules of the currently effective version, semantic relocation processing is performed on each virtual-real coexistence state slice. The execution location corresponds to the execution station and virtual terminal session area associated with the virtual-real coexistence state slice within the target training scenario. The execution subject is the on-site relocation program, which is continuously completed during the current training batch. The effective period starts from the formation of the corresponding virtual-real coexistence state slice and ends when the relocation result of the virtual-real coexistence state slice is recorded. Among them, semantic relocation processing corresponds to the processing of the virtual-real coexistence state slice that has been formed within the same virtual-real coexistence state slice. The entity associations, spatial associations, process statuses, and constraint statuses formed are compared item by item with the entity master records, association records, and process constraint rules in the spatiotemporal semantic base model for each corresponding entity, process entity, and region entity within the current virtual-real coexistence state slice. The unique placement is completed in the order of first determining the process attribution result, then the object binding result, then the spatial location result, and finally the constraint violation result. Under the current process version, only one set of process attribution result, object binding result, spatial location result, and constraint violation result is allowed to be formed in the same virtual-real coexistence state slice.

[0052] Before repositioning, virtual-physical coexistence state slices are grouped according to the current batch version number, the corresponding process entity, and the corresponding entity identifier. Then, they are sorted in ascending order according to the event sequence number corresponding to the slice identifier. When multiple consecutive virtual-physical coexistence state slices appear under the same corresponding entity and the same process entity, the previous effective repositioning result is used as the precedent reference for the repositioning process of the next slice. The precedent reference is only used to verify the process continuity, object continuity, and spatial continuity between adjacent virtual-physical coexistence state slices of the same corresponding entity. It does not replace the entity association relationship, spatial association relationship, process status, and constraint status recorded in the current virtual-physical coexistence state slice itself. The first virtual-physical coexistence state slice, virtual-physical coexistence state slices that switch across process entities, and virtual-physical coexistence state slices whose previous repositioning result is in the pending verification state do not call the precedent reference.

[0053] If the current batch version number is missing, the corresponding process entity is missing, the personnel entity and tool entity holding correspondence is missing, and there are tool entities in the current virtual-physical coexistence state slice, the tool entity and equipment entity or material entity contact correspondence is missing, and there are contact relationships, spatial relationships, positional relationships and regional affiliations in the current virtual-physical coexistence state slice are missing at the same time, the process status exceeds the three status calibers of not started, in execution, and completed, and the constraint status is not recorded according to the constraint items, then stop the further processing of this virtual-physical coexistence state slice and transfer it to the pending verification record.

[0054] Subsequently, the entity master record, association record, and process constraint rules in the spatiotemporal semantic base model are retrieved. The entity master record is used to confirm the unique object boundary of the relevant entity under the current process version. The unique object boundary corresponds to the unique object identity and unique physical affiliation of the same entity identifier under the current process version. The association record is used to confirm the standard correspondence relationships of holding correspondence, contact correspondence, assembly correspondence, function correspondence, and execution correspondence. The process constraint rules are used to confirm the standard constraint boundaries of allowed entry areas, allowed contact with equipment, allowed use of tools, allowed retrieval of materials, prerequisite completion requirements, and prohibited action boundaries.

[0055] When resetting, the entity relationships are checked item by item, only the object relationship groups that have appeared in the current virtual-physical coexistence state slice are checked; if the object relationship group is consistent with the effective relationship record, it remains unchanged; if the standard corresponding object has not completed the on-site confirmation and no other substitute object is on-site, it is recorded as binding missing; if a non-standard object has arrived and its identifier is inconsistent with the standard corresponding object, it is recorded as binding misalignment; if the standard object has not arrived and a non-standard object has arrived, both binding missing and binding misalignment are retained.

[0056] Next, a step-by-step comparison of spatial relationships is performed. Spatial relationships correspond to the positional relationships, entry relationships, adjacency relationships, and contact relationships of corresponding entities within the current virtual-real coexistence slice under the same scene coordinate system. Among them, the standard execution boundary corresponds to the range of permitted positions of the entity jointly defined by the execution station boundary, allowed entry area boundary, standard placement position, stable process action surface, and prohibited action boundary under the current process entity; positional relationships prioritize recording position coordinates when available, and record area affiliation when position coordinates are missing but area affiliation is complete; entry relationships correspond to the relationship of related entities relative to the area entity as entering the inner side, located in the inner side, and exiting the outer side; adjacency relationships correspond to the close neighbor relationship formed between related entities within the preset adjacency distance boundary of the current process version; contact relationships correspond to the establishment, maintenance, or dissolution of contact relationships on the stable process action surface.

[0057] Subsequently, semantic positioning processing is performed on the virtual-real coexistence state slices to obtain the process assignment results. The process assignment result corresponds to the unique process entity to which the current virtual-real coexistence state slice belongs under the current process version. When the process status is not started, the process assignment result is the process entity corresponding to the first standard execution boundary entered by the entity association relationship and spatial association relationship within the current virtual-real coexistence state slice. When the process status is in execution or completed, the process entity to which the standard execution boundary corresponding to the entity association relationship, spatial association relationship and prior completion requirement belongs is the process assignment result. Adjacent process entities are limited to the direct predecessor process entity or direct successor process entity of the process entity marked by the current slice. If it falls into the standard execution boundary of two process entities at the same time, the unique process assignment result is determined in the following order: process status matching takes precedence over prior completion requirement satisfaction, prior completion requirement satisfaction takes precedence over object action relationship consistency, and object action relationship consistency takes precedence over the degree of conformity between spatial position and standard execution boundary.

[0058] Among them, an object relationship group corresponds to two or more entity combinations with standard correspondences under the same process entity, including holding object relationship groups, contact object relationship groups, assembly object relationship groups, action object relationship groups, and execution object relationship groups. Object binding results are recorded separately for object relationship groups and are not merged into a single comprehensive result. For each group of personnel entities, tool entities, equipment entities, material entities, and area entities, the binding consistency, binding missing, and binding misalignment are recorded item by item.

[0059] The spatial station positioning results are then obtained. These results are recorded separately for each relevant entity. A valid station position indicates that the current location and area affiliation are both within the standard execution boundary. A station position offset indicates that the station is still within the allowed entry area but deviates from the standard placement position or the stable process action surface beyond the allowable deviation boundary. A station position exceeding the boundary indicates that the station has crossed the prohibited action boundary, exited the standard execution station, or entered a non-allowed area. The allowable deviation boundary is preferably taken from the station position deviation configuration record in the currently effective version. The area entity itself is not written separately into the spatial station positioning results.

[0060] The constraint violation results are then obtained. These results are used to record the non-compliance of each constraint item in the preceding, following, and subsequent states. They are written separately for each constraint item and are not merged into a single comprehensive result. If the same constraint item is triggered only in a single state, the trigger record for that state is retained. If the same constraint item is triggered consecutively across multiple states, one result for the same constraint item is written and the corresponding trigger state range is retained. If the preceding state is a trigger and the following state is a release, both the trigger trace and the release mark are retained under the same constraint item.

[0061] After semantic relocation processing is completed, a relocation result record is generated. The relocation result record includes at least the relocation result identifier, slice identifier, corresponding process entity, process attribution result, object binding results recorded by object relationship group, spatial station location results recorded by related entity, constraint violation results recorded by constraint item, and the current batch version number. The relocation result record is written to the relocation result record area in a structured record file for direct retrieval in the next stage.

[0062] In a preferred embodiment, taking the virtual-real coexistence state slice of the object contact change at the fastening station of the reducer assembly assembly training line as an example, if the personnel entity and the sleeve tool form a holding correspondence, the sleeve tool and the fastener form a contact correspondence, and the tightening machine and the fastening process form an action correspondence within the current virtual-real coexistence state slice, and the personnel entity is located inside the fastening station, the tool entity is located near the stable process action surface, and triggering occurs within the action range while crossing the boundary is prohibited, then the fastening process can be obtained as the process attribution result, the sleeve tool and the fastener being bound together as the object binding result, the personnel entity being in a valid position but the tool entity being offset as the spatial position result, and triggering by crossing the boundary is prohibited as the constraint violation result. Alternatively, the spatial position result can also be determined entirely based on the region attribution, entry relationship, and adjacency relationship when the position coordinates are missing. As long as it is still based on the spatiotemporal semantic base model and process constraint rules, semantic positioning processing is performed on the virtual-real coexistence state slice and the process attribution result, object binding result, spatial position result, and constraint violation result are output, the implementation result consistent with the technical concept of this step can be obtained.

[0063] S5. Based on the process attribution results, object binding results, spatial station location results, and constraint violation results, generate a compliant counterfactual trajectory and extract the causal chain of the deviation between the actual trajectory and the compliant counterfactual trajectory. The specific implementation is as follows: After the relocation result record has been written into the relocation result record area, a compliance counterfactual trajectory is generated based on the process attribution result, object binding result, spatial station position result, and constraint violation result. The deviation causal chain between the actual trajectory and the compliance counterfactual trajectory is extracted. The execution location corresponds to the execution station, equipment action area, and virtual terminal session area associated with the relocation result record within the target training scenario. The execution subject is the on-site trajectory generation program, which continuously completes the process during the current training batch. The action period begins after the corresponding relocation result record is formed and ends when the corresponding deviation causal chain is completed. The compliance counterfactual trajectory corresponds to the standard compliance evolution trajectory from the current starting state to the compliance completion state of the current process entity, under the condition that the current process version, current batch version number, current process entity, current corresponding entity combination, and current execution boundary remain unchanged. The current starting state corresponds to the immediate on-site state reflected in the current relocation result record, i.e., the relocation result. The result record records the combined status of object binding results, spatial station location results, constraint violation results, and process attribution results at the corresponding moment. The compliant completion status corresponds to the current process entity having its object binding result restored to consistent binding, its spatial station location result restored to valid location, and the constraint violation result no longer having any triggering items, and the process status advancing to the completed state. When generating the record, the attribution result records are first grouped by the current batch version number, the corresponding process entity, and the corresponding entity identifier, and then sorted in ascending order by the event sequence number corresponding to the attribution result identifier. When multiple consecutive attribution result records appear under the same corresponding entity and the same process entity, the first attribution result record with a constraint violation result or an abnormal spatial station location result is taken as the starting point for the current trajectory comparison. Continuous attribution result records correspond to a sequence of attribution result records that are consistent in the current batch version number, consistent in the corresponding process entity, and continuous in the corresponding entity identifier, and that do not have any records to be verified or records with reversed order in the middle.

[0064] Subsequently, the process attribution result, object binding result, spatial station location result, and constraint violation result corresponding to the relocation result record are retrieved. The entity master record, association record, process constraint rules, and standard execution boundary record corresponding to the process entity in the spatiotemporal semantic base model are also retrieved. The process entity, object relationship group, allowed entry area, allowed equipment contact, allowed tool use, allowed material retrieval, prerequisite completion requirements, and prohibited action boundaries are fixed item by item. Within the aforementioned fixed boundaries, a unique compliance path is generated. The unique compliance path corresponds to a unique correction order determined according to a preset priority order under the same current starting point state. If multiple paths exist, all can achieve the compliant completion state. The candidate paths for a state are determined in the following order: safety boundary constraints take precedence over process boundary constraints, process boundary constraints take precedence over sequence boundary constraints, constraint removal takes precedence over correcting spatial positions, correcting spatial positions takes precedence over restoring object bindings, and restoring object bindings takes precedence over process advancement. A lower priority level cannot be entered until the previous priority level is completed. When there are multiple candidate changes within the same priority level, the unique change order is determined by the standard execution order of the corresponding entity identifier under the current process entity. The standard execution order is taken from the order of object action jointly defined by the process card, work instructions, and process constraint rules under the current process version.

[0065] If the object binding result shows both missing binding and misaligned binding, a compliant counterfactual trajectory is generated in the order of first restoring the standard object to be present, and then unbinding the erroneous object. If the space station position result shows both station position offset and station position out of bounds, a compliant counterfactual trajectory is generated in the order of first exiting the prohibited action boundary, and then returning to the standard placement position or stable process action surface. If the constraint violation result involves multiple constraint items, they are released one by one in the order of priority of the aforementioned constraint items.

[0066] The trajectory nodes uniformly adopt four types of node definitions: process status nodes, object binding nodes, space station nodes, and constraint status nodes. Each trajectory node records only one type of status change, and only one constraint item is allowed to change within a constraint status node. If on-site linkage changes cause multiple types of status changes to occur at the same time, they are split into multiple adjacent trajectory nodes and recorded sequentially according to the priority order of the aforementioned unique compliant path.

[0067] Only one type of state change is allowed between adjacent nodes, and the change must not introduce new constraint violation results. If a change will cause other constraint items to be triggered, the current change will be abandoned, and the next change will be selected according to the established priority order of the remaining candidate changes. If all remaining changes will cause new constraint violation results, the current compliance counterfactual trajectory will be stopped from being generated and transferred to the pending verification record.

[0068] After generating the compliance counterfactual trajectory, the causal chain of the deviation between the actual trajectory and the compliance counterfactual trajectory is extracted. The actual trajectory corresponds to the current repositioning result record and its subsequent continuous repositioning result records forming the true state evolution chain under the same process entity. Its cutoff boundary is based on the first of the following three occurrences: the actual trajectory is re-aligned with the compliance counterfactual trajectory, the end of this process entity, or the start of the next process entity. During extraction, the actual trajectory and the compliance counterfactual trajectory are first aligned segment by segment according to the event sequence number, corresponding entity identifier, process status node, object binding node, spatial station node, and constraint status node. The alignment primary key is preferably the event sequence number and the corresponding entity identifier. If the event sequence number is missing, the process status node and object binding node of the adjacent node are used for joint alignment. If alignment is still not possible, the record is stopped from being processed and transferred to the record to be verified.

[0069] The first inconsistency node is recorded as the deviation starting point. Inconsistency corresponds to any one of the following: process status, object binding result, spatial station position result, or constraint violation result, which is inconsistent with the node corresponding to the compliance counterfactual trajectory. If multiple inconsistencies exist within the same node, the deviation starting point is determined in the following order: constraint violation result takes precedence over spatial station position result, spatial station position result takes precedence over object binding result, and object binding result takes precedence over process status.

[0070] The object binding result abnormality, space station location result abnormality, or constraint violation result triggering the deviation starting point is recorded as the deviation cause. Deviation causes are allowed to be single or multiple in parallel. When multiple are in parallel, they are arranged in the aforementioned priority order. If the latter abnormality is directly caused by the former abnormality, only the upstream abnormality is retained as the deviation cause.

[0071] The subsequent process status lag, object relationship misalignment, continuous spatial station position offset or continuous constraint triggering caused by the deviation cause are continuously expanded by nodes to form a deviation transmission segment. When the same type of continuous abnormality persists in adjacent nodes, it is merged and recorded as 1 segment. The recurrence after recovery is recorded as a new deviation transmission segment.

[0072] Until the actual trajectory returns to the corresponding node of the compliant counterfactual trajectory or the entity of this process ends; the four items of the corresponding process status, object binding result, spatial station position result and constraint violation result are realigned to be consistent with the corresponding node of the compliant counterfactual trajectory.

[0073] The deviation causal chain includes at least the deviation start point, the deviation cause, the deviation transmission segment, and the deviation end point. The deviation end point corresponds to the node where the actual trajectory is realigned with the compliant counterfactual trajectory. If the alignment is not achieved by the end of the current process entity, the end point of the current process entity is taken as the deviation end point.

[0074] If the current batch version number is missing, the process attribution result is missing, the object binding result is missing, the spatial station location result is missing to the point that the current starting point status cannot be formed, or the order of the attribution result records for the same corresponding entity under the same process entity is reversed, then the attribution result record will be stopped from being processed and transferred to the record to be verified.

[0075] After the compliance counterfactual trajectory and deviation causal chain are formed, a trajectory comparison record is generated. The trajectory comparison record includes at least the trajectory comparison identifier, the return result identifier, the corresponding process entity, the compliance counterfactual trajectory, the deviation start point, the deviation cause, the deviation transmission segment, the deviation end point, and the current batch version number. The trajectory comparison record is written into the trajectory comparison record area as a structured record file for direct retrieval in the next stage.

[0076] In a preferred embodiment, taking the fastening station of the reducer assembly training line as an example, when the positioning result record shows that the sleeve tool is consistently bound to the fastener, the personnel's physical position is valid, the tool's physical position is offset, and the prohibited boundary crossing is triggered in the action range, a compliant counterfactual trajectory of "exiting the prohibited action boundary - returning to the stable process action surface - maintaining consistent binding - completing the fastening process" can be generated first. Then, the continuous nodes of "continuously crossing the boundary - the tool not returning to the stable process action surface in time - the fastening action being delayed" in the actual trajectory can be extracted as a deviation causal chain. Alternatively, the compliant counterfactual trajectory can also be generated in the same priority order based on the region attribution, object binding result, and constraint violation result when the position coordinates are missing. As long as the compliant counterfactual trajectory is still generated based on the process attribution result, object binding result, spatial position result, and constraint violation result, and the deviation causal chain between the actual trajectory and the compliant counterfactual trajectory is extracted, the implementation result consistent with the technical concept of this step can be obtained.

[0077] S6. Based on event anchors, virtual-real coexistence state slices, compliance counterfactual trajectories, and deviation causal chains, generate training data atoms, and revise the spatiotemporal semantic base model based on the training data atoms. The specific implementation is as follows: After the trajectory comparison record has been written into the trajectory comparison record area, training data atoms are generated based on event anchors, virtual-real coexistence state slices, compliant counterfactual trajectories, and deviation causal chains. The spatiotemporal semantic base model is then corrected based on these training data atoms. The execution location corresponds to the execution station, equipment action area, and virtual terminal session area associated with the trajectory comparison record within the target training scenario. The execution entity is the on-site atom generation program, which continuously completes this process before the end of the current training batch. The action period begins after the corresponding trajectory comparison record is formed and ends when the training data atom is finalized and the corresponding spatiotemporal semantic base model correction result is submitted. Each training data atom corresponds to the smallest reusable training segment record formed around the same process entity, the same corresponding entity combination, and the same deviation causal chain. The smallest reusable training segment record corresponds to the ability to independently reproduce the formation of the deviation starting point once. The generation of individual atomic records for the deviation propagation and deviation endpoint convergence processes involves first grouping event anchors, virtual-real coexistence state slices, compliant counterfactual trajectories, and deviation causal chains according to the current batch version number, corresponding process entity, corresponding entity identifier, and trajectory comparison identifier. Then, these records are sorted in ascending order by event sequence number, and the correspondence between event anchor identifiers, slice identifiers, relocation result identifiers, and trajectory comparison identifiers within the same group is verified. Among these, the same deviation causal chain is limited to a causal chain with the same deviation start type, the same deviation cause type, the same corresponding entity identifier, and the same deviation endpoint type. The corresponding entity combination is based on the personnel entity, tool entity, equipment entity, material entity, and area entity that actually appear and participate in the object binding result, spatial station result, or constraint violation result under the current process entity, and it is not required to include all entity categories.

[0078] If the current batch version number is missing, the corresponding process entity is missing, the event anchor point and the virtual-real coexistence state slice cannot be matched one by one, the compliance counterfactual trajectory is missing, the deviation causal chain is missing, or the order of records in the same group is reversed, then the processing of records in that group will be stopped and the records will be transferred to the pending verification records; if the records in the same group do not form a deviation causal chain, then no training data atom will be generated.

[0079] After verification, firstly, atomic start boundaries and atomic end boundaries are generated based on event anchor points. The atomic start boundary corresponds to the start time of the first event anchor point in the same group, and the atomic end boundary corresponds to the end time of the last event anchor point in the same group or the end time of the deviation causal chain, whichever is later. Then, based on the virtual-real coexistence state slice, the entity association, spatial association, process status, and constraint status within the current atom are written. The entity association, spatial association, process status, and constraint status all use the effective record caliber of the corresponding virtual-real coexistence state slice and are not rewritten in this step. Subsequently, based on the compliance counterfactual trajectory, the standard compliance evolution chain corresponding to the atom is written. The standard compliance evolution chain corresponds to the trajectory segment of the compliance counterfactual trajectory in the previous step between the current atom start boundary and the atomic end boundary. Finally, based on the deviation causal chain, the deviation start point, deviation cause, deviation transmission segment, and deviation end point of the current atom are written. The deviation cause, deviation transmission segment, and deviation end point are kept consistent with the trajectory comparison record and no new higher-level summary descriptions are added.

[0080] After the training data atoms are generated, atomic records are formed. An atomic record includes at least an atomic identifier, a corresponding process entity, a corresponding entity identifier, an atomic start boundary, an atomic end boundary, a set of event anchor points arranged in event order, a set of virtual and real coexistence state slices arranged in event order, a standard compliance evolution chain, a deviation start point, a deviation cause, a deviation transmission segment, a deviation end point, and the current batch version number. Among them, the atomic identifier is preferably generated according to "current training batch + corresponding process entity + atomic sequence number". Only one effective atomic record is retained for the same corresponding entity combination, the same deviation causal chain, the same atomic start boundary, and the same atomic end boundary. Completely identical duplicate submissions are retained as idempotent records, while inconsistent ones are transferred to records pending verification.

[0081] After the training data atoms are written into the records, the spatiotemporal semantic base model is corrected based on the training data atoms. The correction is limited to the records of association relationships, regional entity boundary records, standard placement position records, stable process action surface records, and process constraint rule records that are directly related to the corresponding process entity of the current atom under the current process version. The correction does not cross process entities or the current process version. During the correction, it is first determined whether the deviation cause in the current atom appears continuously under the same process entity and the same corresponding entity combination and reaches the preset correction threshold. The continuous appearance corresponds to the absence of a continuous sequence of atoms to be checked in the most recently formed atomic records. The preset correction threshold is preferably determined by the occurrence of the same deviation cause in 3 or more consecutive training data atoms and the consistency of the corresponding deviation endpoint. Among them, the same deviation cause corresponds to the same deviation cause type, the same entity identifier, and the same constraint item or object relationship group. The same deviation endpoint corresponds to the same deviation endpoint node type and the same process status. If the preset correction threshold is not reached, the current training data atom is only written to the atom record area and the spatiotemporal semantic base model correction is not initiated.

[0082] If the binding result of the object corresponding to the deviation cause continues to be misaligned after reaching the preset correction threshold, the standard corresponding object boundary in the association record is corrected; if the spatial station position result corresponding to the deviation cause continues to shift and the shift direction is consistent, the standard placement position record or the stable process action surface record is corrected respectively, depending on whether the shift occurs at the standard placement position or the stable process action surface; if the region entry boundary corresponding to the deviation cause continues to be triggered prematurely or released late, the region entity boundary record is corrected; if the same constraint item corresponding to the deviation cause is repeatedly triggered in multiple atoms and the actual compliance counterfactual trajectories all point to the same release sequence, the boundary of the corresponding constraint item in the process constraint rule record is corrected; when multiple correction objects simultaneously meet the correction conditions, the unique correction object is determined in the order that safety boundary-related correction takes precedence over spatial location-related correction, spatial location-related correction takes precedence over object correspondence correction, and object correspondence correction takes precedence over sequential constraint correction.

[0083] During correction, only the numerical boundary, positional boundary, or object-corresponding boundary in the currently effective version is allowed to be fine-tuned in the same direction. Fine-tuning in the same direction corresponds to a single small step adjustment along the same deviation direction indicated by the continuous deviation cause. The adjustment range shall not exceed the preset fine-tuning upper limit of the current boundary value. Preferably, the single adjustment range of the positional boundary can be set to 1 mm to 5 mm, and the single adjustment range of the numerical boundary can be set to 1% to 5% of the current boundary value. Correction is carried out by adding a new record after correction and retaining the invalid state of the record before correction. Existing original records are not overwritten or deleted. If the corresponding deviation cause no longer meets the preset correction threshold in subsequent consecutive atoms after a single correction, fine-tuning along that direction will stop.

[0084] Each correction generates a correction trace record, which includes at least the correction object, correction basis type, content before correction, content after correction, set of triggering atom identifiers, and correction submission time. The correction basis type corresponds to the category of basis that triggered the current correction, including basis for persistent misalignment of object binding results, basis for persistent offset of space station results, basis for persistent anomalies in region entry boundaries, and basis for persistent repetitive triggering of constraint items. This data is also bound to the current process version, modeling version, and rule version. Subsequently, a model correction result record is generated, which includes at least a correction result identifier, atom identifier, correction object, correction content, and the current batch version number. The model correction result record and the correction trace record are associated through the same correction result identifier. The correction trace record is written to the correction trace record area as a structured record file, and the model correction result record is written to the model correction result record area as a structured record file, for direct retrieval in the next stage.

[0085] In a preferred embodiment, taking the fastening station of the reducer assembly training line as an example, when three consecutive training data atoms all show that the same sleeve tool can only form compliant contact after shifting inwards towards the station before fastening begins, and the deviation causal chain is all expressed as "tool entity position shift - prohibited boundary crossing trigger - return to stable process action surface after exiting the boundary - process completion", then the stable process action surface record corresponding to the fastening station can be finely adjusted inwards towards the station in a unified direction based on the three training data atoms, and a new stable process action surface record can be written into the spatiotemporal semantic base model. Alternatively, when the position coordinates are missing, the correction direction can also be determined based on the region affiliation, adjacency relationship and consistency of continuous deviation endpoints. As long as the training data atoms are still generated based on the event anchor point, virtual and real coexistence state slice, compliant counterfactual trajectory and deviation causal chain, and the spatiotemporal semantic base model is corrected based on the training data atoms, the implementation result consistent with the technical concept of this step can be obtained.

[0086] In the operational scenario shown in this embodiment: the target training scenario is selected as the fastening training unit in the reducer assembly training line. The fastening training unit is sequentially connected to the material loading station, pre-assembly station, fastening station, inspection station, and off-line station along the production line direction. This embodiment focuses on the fastening station. Before the current training batch begins, the modeling and maintenance personnel and process personnel first construct the spatiotemporal semantic base model corresponding to the target training scenario based on the equipment ledger, tooling list, material master data, process card, work instructions, station layout diagram, 3D structural data, and on-site survey records, and complete the personnel entity, equipment entity, and process entity. The system includes entity master record archiving, relationship definition, and process constraint rule configuration for entities, materials, processes, and regions. For example, the system registers the on-duty trainee as a personnel entity, the electric tightening machine, clamping platform, and testing table as equipment entities, the sleeve tool and locating pin as tool entities, the housing, fasteners, and accessories as material entities, "perform tightening after housing positioning" as a process entity, and the inner area of ​​the tightening station, the equipment working area, and the safety isolation area as region entities. It also includes the rule that "the tightening process only allows the holder to enter the inner boundary of the tightening station with the corresponding numbered sleeve tool, and must not cross the tightening..." The process constraint rules for the "tightening machine rotation coverage area" will be implemented after the current effective version is submitted. At the start of the current training batch, the data acquisition system will continuously collect real operation data and virtual interaction data. Specifically, the position sensing terminal will continuously record the position changes of personnel and tool entities, the contact sensing terminal will continuously record the contact changes between the socket tool and the fastener head, the equipment controller will continuously record the state changes of the tightening machine from preparation to execution to completion, and the virtual terminal will synchronously record the trainee's step submissions and alarm responses in the virtual interface. Then, alignment, noise reduction, and completion within allowable boundaries will be completed according to a unified time reference, and based on the equipment... Event anchors are extracted from state changes, object contact changes, area entry / exit changes, process state changes, and constraint triggering changes. For example, when a person enters the inner area of ​​the fastening station and maintains a preset position acquisition rhythm, an area entry event anchor is formed. When the sleeve tool establishes stable contact with the fastener head and maintains a preset acquisition rhythm, an object contact event anchor is formed. When the tightening machine changes from ready to execute and maintains a preset equipment state acquisition rhythm, an equipment state change event anchor is formed. When a process execution confirmation record appears with a record corresponding to the completion confirmation caliber, a process state change event anchor is formed.

[0087] After each event anchor point is formed, the system retrieves the previous state record, the action record, and the subsequent state record for each event anchor point, and generates a corresponding virtual-real coexistence state slice accordingly. For example, in the virtual-real coexistence state slice corresponding to the object contact change event anchor point, the system can record the holding correspondence between the personnel entity and the socket tool, the contact correspondence between the socket tool and the fastener, the entry relationship between the tool entity and the fastening station, the fastening process from not started to being executed, and the constraint states that prohibit crossing the boundary constraint are satisfied in the previous state interval and triggered in the action interval.

[0088] Subsequently, based on the spatiotemporal semantic base model and process constraint rules, the system performs semantic relocation processing on the virtual-real coexistence state slice. It compares the entity associations, spatial associations, process states, and constraint states in the current virtual-real coexistence state slice with the effective entity master record, association record, and process constraint rules item by item to obtain the process attribution result, object binding result, spatial positioning result, and constraint violation result. For example, the system can uniquely assign the current virtual-real coexistence state slice to the fastening process, confirm that the object binding result between the sleeve tool and the fastener is consistent, confirm that the personnel entity position is within the inner boundary of the fastening station and is therefore valid, and confirm that the tool entity deviates from the stable process action surface and is therefore a positioning offset. It also confirms that the constraint item prohibiting crossing the boundary is triggered within the action range, thereby forming the corresponding relocation result record.

[0089] After the relocation result record is generated, the system further generates a compliance counterfactual trajectory based on the process attribution result, object binding result, spatial position result, and constraint violation result, and extracts the deviation causal chain between the actual trajectory and the compliance counterfactual trajectory. For example, for the current on-site status in the relocation result record of "sleeve tool and fastener are bound consistently, personnel physical position is valid, tool physical position is offset, and boundary crossing is prohibited", under the condition that the current process version and the current execution boundary remain unchanged, the system generates "exit prohibited" according to the order of constraint violation removal taking precedence over spatial position correction, spatial position correction taking precedence over object binding restoration, and object binding restoration taking precedence over process advancement. The compliance counterfactual trajectory is as follows: "stop action boundary - return to stable process surface - maintain binding consistency - complete tightening process". At the same time, the continuous repositioning result record sequence that occurs in the actual operation on site, such as "personnel entity continuously crosses the rotation coverage area - tool entity fails to return to stable process surface in time - tightening action is delayed", is aligned with the compliance counterfactual trajectory segment by segment. The deviation starting point is extracted from the first inconsistency node. "Tool entity position deviation and triggering the prohibited crossing boundary trigger" is identified as the deviation cause. The subsequent process status lag and continuous triggering of constraint items are merged into the deviation transmission segment. The realignment node or the end node of the current process entity is used as the deviation endpoint to form a complete deviation causal chain.

[0090] After the trajectory comparison record is formed, the system generates training data atoms based on event anchor points, virtual-real coexistence state slices, compliant counterfactual trajectories, and deviation causal chains. For example, around the same process entity, the same corresponding entity combination, and the same deviation causal chain, the first region entry event anchor point is taken as the atom starting boundary, and the end point of the deviation causal chain or the end time of the last event anchor point is taken as the atom ending boundary. The set of event anchor points, the set of virtual-real coexistence state slices, the standard compliance evolution chain, the deviation starting point, the deviation cause, the deviation transmission segment, and the deviation ending point within the boundary range are all written into the atom record, thus forming the smallest reusable training segment record that can independently reproduce "entering the fastening station—establishing contact—occurring station position deviation—triggering the prohibited crossing boundary—exiting the boundary and returning to the stable process action surface—completing the process". When multiple subsequent training data atoms repeatedly indicate the same process entity and the same corresponding entity combination, Before tightening begins, the sleeve tool needs to be offset inwards from the workstation to achieve stable and compliant contact. When the corresponding deviation endpoints are consistent, the system reaches the preset correction threshold. Then, based on the training data atoms, it performs correction on the spatiotemporal semantic base model, making a fine adjustment in the same direction to the stable process action surface record under the tightening workstation inwards. It adds a record of the effect after correction and retains the failure state of the record before correction. At the same time, it generates correction trace records and model correction result records for subsequent batches to continue to use. Through the above continuous operation, the training system can not only complete the complete closed loop from object modeling, event extraction, state slicing, semantic positioning, trajectory comparison to atomic precipitation and model correction in a single training process, but also enable the process boundaries, object correspondence, spatial station boundaries and process constraint rules in subsequent similar tightening training processes to gradually conform to the real field performance, thereby obtaining implementation results consistent with the technical concept of this invention.

[0091] All calculations involved in the embodiments are dimensionless numerical calculations, and the preset parameters and thresholds in the calculations are set by those skilled in the art according to the actual situation.

[0092] It should be noted that this invention can be deployed on the device itself to realize embedded applications, or it can run on a PC or other terminal with a user interface, thereby meeting various hardware environments and usage requirements.

[0093] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, as a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, all or part of the processes or functions described in the embodiments of this application are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wireless or wired transmission; wired transmission methods include optical fiber, twisted pair, coaxial cable, etc.; wireless transmission includes infrared, microwave, etc. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center containing one or more sets of available media. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium. A semiconductor medium can be a solid-state drive.

[0094] Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and modules described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0095] In the several embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of modules is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple modules or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between apparatuses or modules may be electrical, mechanical, or other forms.

[0096] The modules described as separate components may or may not be physically separate. The components shown as modules may or may not be physical modules; they may be located in one place or distributed across multiple network modules. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0097] In addition, the functional modules in the various embodiments of this application can be integrated into one processing module, or each module can exist physically separately, or two or more modules can be integrated into one module.

[0098] If the aforementioned functions are implemented as software functional modules and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0099] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0100] In conclusion, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for processing industrial virtual-real co-existence training data based on spatial computing, characterized in that, include: S1. Construct a spatiotemporal semantic base model corresponding to the target training scenario, define the relationships between personnel entities, equipment entities, tool entities, material entities, process entities, and area entities, and configure process constraint rules; based on equipment ledgers, tooling lists, material master data, process cards, work instructions, workstation layout diagrams, 3D structural data, and on-site survey records, form entity master records, relationship records, and process constraint rule records; align object records according to equipment ledger numbers, tooling numbers, material codes, process numbers, and area numbers, remove duplicates, and fill in static fields that can be uniquely identified by the same source file, then write the object records into the spatiotemporal semantic base model; configure process constraint rules for process entities, including allowed entry areas, allowed contact with equipment, allowed use of tools, allowed material retrieval, prerequisite completion requirements, and prohibited action boundaries; S2. Collect real operation data and virtual interaction data during the target training process, and extract event anchor points based on changes in equipment status, changes in tool contact, changes in area entry and exit, changes in process status, and changes in constraint triggering. S3. Retrieve the corresponding previous state record, action record, and subsequent state record around the event anchor point to generate a virtual-real coexistence state slice containing entity association, spatial association, process state, and constraint state; S4. Based on the spatiotemporal semantic base model and process constraint rules, perform semantic positioning processing on the virtual and real coexistence state slices to obtain process attribution results, object binding results, spatial station location results, and constraint violation results. S5. Generate a compliant counterfactual trajectory based on process attribution results, object binding results, spatial station location results, and constraint violation results, and extract the causal chain of deviation between the actual trajectory and the compliant counterfactual trajectory; S6. Generate training data atoms based on event anchors, virtual-real coexistence state slices, compliance counterfactual trajectories, and deviation causal chains, and revise the spatiotemporal semantic base model based on the training data atoms; group by current batch version number, corresponding process entity, corresponding entity identifier, and trajectory comparison identifier, and organize event anchors, virtual-real coexistence state slices, compliance counterfactual trajectories, and deviation causal chains according to event sequence, verify the correspondence between event anchor identifiers, slice identifiers, relocation result identifiers, and trajectory comparison identifiers, and generate training data atoms; determine the atom starting boundary based on event anchors, and determine the atom ending boundary based on the event anchor end time and the deviation causal chain ending time, write the standard compliance evolution chain based on the compliance counterfactual trajectory, and write the deviation starting point, deviation cause, deviation transmission segment, and deviation ending point based on the deviation causal chain; when the preset correction threshold is met, determine the correction objects in the association record, regional entity boundary record, standard placement position record, stable process action surface record, and process constraint rule record based on the abnormal object binding result, abnormal space station position result, abnormal area entry boundary, and continuous triggering of constraint items corresponding to the deviation cause, and perform correction.

2. The industrial virtual-real symbiotic training data processing method based on spatial computing according to claim 1, characterized in that, S2 include: After the current version of the spatiotemporal semantic base model and process constraint rules are submitted and take effect, the acquisition channel will be time-corrected according to the unified time synchronization benchmark. The acquired records are grouped by entity identifier and sorted by acquisition time, then aligned and denoised according to a unified time synchronization standard, and the records corresponding to the continuous records are supplemented.

3. The method for processing industrial virtual-real coexistence training data based on spatial computing according to claim 2, characterized in that, S2 also includes: Extract event anchor points based on changes in equipment status, tool contact, area entry / exit, process status, and constraint triggering. For each extracted and established event anchor point, an event anchor point record is generated. The event anchor point record includes the event anchor point identifier, event anchor point type, corresponding entity identifier, corresponding process entity, corresponding region entity, start time, end time, source of establishment basis, and current batch version number.

4. The industrial virtual-real symbiotic training data processing method based on spatial computing according to claim 1, characterized in that, S3 include: For each event anchor record, retrieve the preceding state record, action record, and following state record; Grouped by current batch version number, corresponding process entity, and corresponding entity identifier, and aligned by collection time; The event anchor points are used to record the start and end times to divide the pre-state interval, action interval, and post-state interval; Based on the entity master record and relationship record, as well as the previous state record, action record and subsequent state record in the spatiotemporal semantic base model, a virtual-real coexistence state slice containing entity relationship, spatial relationship, process state and constraint state is generated.

5. The industrial virtual-real symbiotic training data processing method based on spatial computing according to claim 1, characterized in that, S4 include: When performing semantic relocation processing on virtual-real coexistence state slices, the slices are first grouped and organized according to the current batch version number, the corresponding process entity, and the corresponding entity identifier. Then, the entity master record, the association record, and the process constraint rules are retrieved, and the unique relocation is completed in the order of process attribution result, object binding result, spatial station location result, and constraint violation result.

6. The method for processing industrial virtual-real coexistence training data based on spatial computing according to claim 5, characterized in that, S4 also includes: The object binding result is determined based on the consistency, missing, and misalignment of the object relationship group and the effective association record; The location of the space station is determined based on the location coordinates and regional affiliation, combined with the standard placement location, the stable process action surface, and the standard execution boundary. Constraint violation results are recorded based on the triggering and release status of each constraint in the preceding, following, and subsequent states.

7. The industrial virtual-real symbiotic training data processing method based on spatial computing according to claim 1, characterized in that, S5 include: Based on the process attribution results, object binding results, spatial station location results, and constraint violation results, a compliance counterfactual trajectory is generated in the priority order of constraint violation correction, spatial station location correction, restoration of object binding, and process advancement. The actual trajectory is then aligned with the compliant counterfactual trajectory to extract the deviation causal chain, which includes the deviation start point, deviation trigger, deviation transmission segment, and deviation end point.

Images