Energy storage module production line multi-modal interaction control method

Abstract

The application discloses a kind of energy storage module production line multi-modal interaction control methods, belong to energy storage module intelligent manufacturing and laser welding production line control field, to solve the problem that production file, welding program, process parameter, module model and abnormal treatment are dispersed and difficult to real-time joint check, the interface input, visual result, equipment state and personnel trajectory are converted into uniform interaction event in the present application, module model and process state are established, instruction is analyzed and is fused with matching, parameter, license, action and trajectory evidence to carry out risk classification, release instruction is adapted to issue, and after execution and postweld quality feedback calibration, subsequent check, dynamic recalculation operation range are carried out, cross-device instruction consistency control and process switching constraint are realized.

Description

Technical Field

[0001] This invention relates to the field of intelligent manufacturing and laser welding production line control for energy storage modules, and particularly to a multimodal interactive control method for energy storage module production lines. Background Technology

[0002] Energy storage module assembly and laser welding production lines typically include processes such as production document management, welding program invocation, process parameter setting, module model confirmation, equipment license assessment, and post-weld quality inspection. These processes may be handled by different controllers, host computers, vision inspection equipment, and human-machine interfaces. The data fields, parameter units, file indexes, and status expressions of each device are not consistent, requiring operators to select and confirm items one by one across multiple interfaces.

[0003] Existing manual verification methods struggle to establish a unified judgment framework that correlates visually recognized module features, production document metadata, real-time equipment operating status, and personnel operation trajectories. This can lead to issues such as production documents or welding procedures not matching module models, missing or out-of-range required parameters, and unauthorized control commands being issued. Furthermore, the fragmented interfaces make it difficult to maintain a consistent executable scope for operations in the stages of data entry, secondary verification, risk interception, or anomaly recovery.

[0004] Furthermore, when process identifiers, module models, or equipment operating status change, the availability of traditional interfaces and equipment interfaces often relies on manual updates, and it is difficult to reverse-calibrate subsequent instruction judgments based on equipment execution results and post-weld visual quality results. Consequently, the production line lacks a closed-loop control mechanism that can simultaneously constrain the personnel interface and equipment interface under process switching, equipment malfunction, and quality anomaly conditions.

[0005] Therefore, a multimodal interactive control method for energy storage module production lines is needed to address the shortcomings of existing technologies. Summary of the Invention

[0006] One objective of this invention is to propose a multimodal interactive control method for energy storage module production lines. Addressing the problems in existing technologies where production documents, welding procedures, process parameters, module models, and anomaly handling for heterogeneous equipment and interfaces lack unified correlation, manual item-by-item verification cannot integrate visual results, equipment status, and personnel behavior, and the operational scope cannot be updated in a timely manner after process changes, this invention proposes a technical solution that converts multi-source data into unified interactive events, jointly determines the current module model and process status, structurally parses interactive instructions, integrates multiple verification evidences to form a risk score and implements tiered handling, generates control parameters for release instructions and issues them through the equipment adapter, and uses execution feedback and post-weld visual results to update status and calibration quantities. This ensures that document or program matching, parameter boundaries, equipment permissions, process actions, and personnel trajectories are verified within the same control chain, and that the interface controls and equipment interface whitelist are updated synchronously with status changes.

[0007] This invention provides a multimodal interactive control method for an energy storage module production line, comprising: S1. Collect interface inputs, visual inspection results, equipment operating status and personnel operation trajectories, and convert them into a unified set of interactive events according to equipment identifiers, module identifiers, process identifiers and timestamps; S2. Determine the current module model based on the visual inspection results and production file metadata in the set of interactive events, and establish the current process status by combining the process identifier and equipment operating status; S3. Parse the interactive instructions in the current process state to form an instruction structure that includes the instruction intent, target device, file or program identifier, parameter fields, and exception handling actions; S4. According to the verification rules corresponding to the current process status, jointly verify the matching relationship between the current module model and the file or program identifier, the integrity and range of parameters, the equipment permission status, the allowed actions of the process and the personnel operation trajectory, integrate the verification evidence into a risk score and compare it with the classification threshold, and output one of the following verification results: release, supplementary recording or secondary confirmation, and risk interception. S5. Response to release result, generate device control parameters based on the current module model, the file or program identifier and parameter fields, send them to the target device adapter and obtain device execution feedback; S6. Update the current process status based on equipment execution feedback and post-weld visual inspection results, and generate calibration quantities for subsequent verification. Recalculate the executable operation range when the process identifier, current module model, or equipment operating status changes, and refresh the interface control and equipment interface whitelist accordingly.

[0008] Optionally, S1 includes: Establish field mappings between source fields and unified semantic fields for interface input, visual inspection results, equipment operating status, and personnel operation trajectories; Correct the timestamp based on the offset between the source device clock and the production line reference clock; The interactive events are arranged according to the corrected timestamps, using module identifiers and process identifiers as association keys. The personnel operation trajectory consists of control identifiers, operation types, operation values, and operation times.

[0009] Optionally, S2 includes: The module shape features, component position features, and identification character recognition results are extracted from the visual detection results and matched with the visual feature templates corresponding to the candidate module models to obtain the matching value of each candidate module model. Read the applicable module model, process version, and activation status from the production file metadata; Candidate module models whose matching values ​​reach the model determination threshold and are consistent with the applicable module model are determined as the current module model. The model determination threshold is taken from the verification rule corresponding to the current process identifier. If no candidate module model meets the requirements, the current process status will be set to "model pending confirmation".

[0010] Optionally, S3 includes: Based on the instruction dictionary corresponding to the current process identifier, identify the instruction intent and abnormal handling actions; determine the target equipment based on the equipment address mapping table; determine the file or program identifier based on the production file index and welding program index; and extract the parameter fields based on the parameter mapping between the unified semantic field and the equipment parameter field. Generate missing markers for required fields that could not be extracted, and write the parsed fields, field sources, and parsing confidence values ​​into the instruction structure.

[0011] Optionally, S4 includes: The allowed correspondence between module model and production file, welding procedure, required field, numerical range and unit of each parameter field, allowed actions of target equipment in each operating state, allowed action set of current process, preset operation sequence, and non-negative weight that corresponds one-to-one with each verification evidence and sums to one. Map the pass result of each check to zero and the fail result to one; The matching ratio is obtained by dividing the number of operations in the personnel operation trajectory that have the same order as the preset operation sequence and whose operation values ​​satisfy the corresponding field constraints by the total number of operations in the preset operation sequence. The risk score is obtained by multiplying the value obtained by subtracting the matching ratio from one and the mapping value of each verification by their respective weights. Furthermore, the grading threshold includes a first grading threshold and a second grading threshold read from the verification rules corresponding to the current process status, wherein the first grading threshold is less than the second grading threshold; Comparing the risk score with the first grading threshold and the second grading threshold includes: when the risk score is less than the first grading threshold, outputting permission to proceed; When the risk score is greater than or equal to the first grading threshold and less than the second grading threshold, the field to be supplemented is output according to the missing marker, or a secondary confirmation information containing the current module model, file or program identifier and parameter fields is generated, and step S4 is re-executed after supplementation or secondary confirmation. When the risk score is greater than or equal to the second grading threshold, a risk interception is output, and the verification item that caused the failure is used as the reason for the interception.

[0012] Optionally, S5 includes: Based on the current module model, file or program identifier and parameter fields, select the parameter version that is in effect from the parameter version library, use the parameter mapping to convert the parameter fields into target device fields, and perform unit conversion and numerical discretization according to the dimensions and resolution of the target device to generate the device control parameters; Before issuing the command, the device adapter verifies the target device identifier, parameter version, and unique instruction identifier corresponding to the current interaction instruction. If the unique instruction identifier is received repeatedly, the adapter returns the previous device execution feedback instead of executing it again.

[0013] Optionally, S6 includes: The modification records of the instruction structure by personnel during supplementary recording or secondary confirmation are used as personnel correction samples. The deviation between the execution success flag and measured parameters in the equipment execution feedback and the equipment control parameters generated according to the current module model, file or program identifier and parameter fields are used as execution samples. The weld defect category and defect size in the post-weld visual inspection results are used as quality samples. Read the personnel correction field, execution failure flag, allowable deviation of measured parameters, quality limit of each defect category, target abnormal sample ratio, threshold sensitivity coefficient and threshold boundary from the valid and quality valid rules, and mark abnormal samples accordingly; Within a preset sample window, the proportion of abnormal samples corresponding to the released instructions is counted, and the difference between the proportion of abnormal samples and the target proportion of abnormal samples is used as the calibration value. Subtract the product of the calibration amount and the corresponding threshold sensitivity coefficient from the first and second grade thresholds respectively, and limit the corrected thresholds within the corresponding threshold boundaries, while keeping the first grade threshold less than the second grade threshold; Furthermore, when the device execution feedback indicates that the distribution has failed, the device refuses to execute, or the measured parameters exceed the value range, the current process status is set to an execution abnormal state, and subsequent release operations related to the file or program identifier are frozen. The abnormal handling action is issued through the target device adapter. The execution result of the abnormal handling action is determined according to the returned execution feedback and the verification rules corresponding to the current process status. The freeze is lifted after the device license status is regained, the module model is reconfirmed, and the execution result is passed. Read the quality limit value corresponding to the defect category from the quality verification rule that corresponds to the current module model, the file or program identifier and the current process identifier and is in effect. When the defect size in the post-weld visual inspection result exceeds the quality limit value, add the rework program to the executable operation range and remove the start interface of the next process from the equipment interface whitelist. Furthermore, a traceability record is established for each interactive instruction. The traceability record includes a unified interactive event, a version identifier of the current process status, an instruction structure, version identifiers of the verification rules and quality verification rules used, various verification results, risk scores, grading thresholds, verification results, equipment control parameters generated based on the current module model, file or program identifier and parameter fields, equipment adapter return information, equipment execution feedback, post-weld visual inspection results, calibration quantities and version identifiers of the executable operation range, and each field in the traceability record is associated with the unique instruction identifier.

[0014] The beneficial effects of this invention are: 1. By converting interface inputs, visual inspection results, equipment operating status, and personnel operation trajectories into interactive events with unified association identifiers and time bases, consistent data objects can be formed across devices, reducing association deviations caused by differences in field formats, time bases, and device indexes.

[0015] 2. By jointly verifying the module model and document or program identifier, parameter integrity and range, equipment license, process actions and personnel trajectories based on the current process status, and using risk scoring to output release, supplementary entry or secondary confirmation and risk interception results, mismatch, omission, boundary violation and cross-process operation risks can be identified before control parameters are issued.

[0016] 3. By using equipment execution feedback, personnel correction records, and post-weld visual quality results to calibrate grading thresholds, and recalculating the executable operation range when the process, model, or equipment status changes, subsequent verification and interface controls, equipment interface whitelists can be linked with the actual production line status, while retaining the traceability relationship of instruction parsing, verification, issuance, and feedback. Attached Figure Description

[0017] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings: Figure 1 This is a flowchart of the multimodal interactive control method for the energy storage module production line of the present invention.

[0018] Figure 2 This is a flowchart of the S4 joint verification and risk scoring of the present invention. Detailed Implementation

[0019] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, illustrating only the basic structure of the invention, and therefore only show the components relevant to the invention.

[0020] refer to Figures 1-2 A multimodal interactive control method for an energy storage module production line, comprising: S1. Collect interface inputs, visual inspection results, equipment operating status and personnel operation trajectories, and convert them into a unified set of interactive events according to equipment identifiers, module identifiers, process identifiers and timestamps; S2. Determine the current module model based on the visual inspection results and production file metadata in the set of interactive events, and establish the current process status by combining the process identifier and equipment operating status; S3. Parse the interactive instructions in the current process state to form an instruction structure that includes the instruction intent, target device, file or program identifier, parameter fields, and exception handling actions; S4. According to the verification rules corresponding to the current process status, jointly verify the matching relationship between the current module model and the file or program identifier, the integrity and range of parameters, the equipment permission status, the allowed actions of the process and the personnel operation trajectory, integrate the verification evidence into a risk score and compare it with the classification threshold, and output one of the following verification results: release, supplementary recording or secondary confirmation, and risk interception. S5. Response to release result, generate device control parameters based on the current module model, the file or program identifier and parameter fields, send them to the target device adapter and obtain device execution feedback; S6. Update the current process status based on equipment execution feedback and post-weld visual inspection results, and generate calibration quantities for subsequent verification. Recalculate the executable operation range when the process identifier, current module model, or equipment operating status changes, and refresh the interface control and equipment interface whitelist accordingly.

[0021] In this specific embodiment, S1 includes: The process is executed by the production line data acquisition gateway, which is connected to the human-machine interface, the vision inspection device, the equipment controller, and the operation log module. The human-machine interface reports the interface input when selecting files, welding programs, inputting parameters, and confirming submission. The vision inspection device reports the vision inspection result after completing a module photo recognition. The equipment controller reports the equipment operating status at a fixed acquisition cycle. The operation log module reports the personnel's operation trajectory each time the personnel trigger the interface controls. A field mapping table is pre-established within the production line acquisition gateway. The field mapping table includes source device identifier, source field name, source field data type, source field unit, unified semantic field name, and field conversion method. The unified semantic field includes device identifier, module identifier, process identifier, timestamp, event source, event type, and event payload. For UI input, the field mapping table maps UI number, control identifier, input value, submission action, and UI time to event payload, event type, and original timestamp. For visual inspection results, the field mapping table maps camera identifiers, image identifiers, module shape inspection results, component position inspection results, identifier character recognition results, and inspection completion time to device identifiers, event payloads, and original timestamps. For device operating status, the field mapping table maps device address, operating status code, license status, alarm code, measured parameters and status acquisition time to device identifier, event payload and original timestamp; For personnel operation trajectories, each trajectory record output by the operation log module consists of control identifier, operation type, operation value, and operation time. The control identifier, operation type, and operation value are written into the event payload, and the operation time is used as the original timestamp to participate in time correction and is synchronously saved in the event payload. When a source data does not directly carry a module identifier or process identifier, the production line acquisition gateway reads the module identifier corresponding to the pallet or fixture according to the current workstation binding table, and reads the process identifier according to the current workstation process configuration table, so that the interface input, visual inspection results, equipment operating status and personnel operation trajectory generated by the same energy storage module in the same process have the same association key; The production line data acquisition gateway uses the production line reference clock as a unified time reference. It reads the clocks of each source device when the production line starts and at fixed synchronization cycles to obtain the time offset between the source device clocks and the production line reference clock, and generates a unified set of interactive events in the following format: ; in, Represents a unified set of interactive events. This represents the total number of uniform interaction events. Indicates the sequence number of the unified interactive event. Indicates the first A unified interactive event Indicates the first A device identifier corresponding to a unified interactive event. Indicates the first A module identifier corresponding to a unified interactive event. Indicates the first A unified interactive event corresponds to a process identifier. Indicates the first A unified timestamp after interaction event correction Indicates the first The unified interactive event source can be selected from one of the following: interface input, visual inspection results, device operating status, and personnel operation trajectory. Indicates the first A unified interactive event type, Indicates the first Event payload of a unified interactive event Indicates the generation of the first The original timestamp provided by the source device of a unified interactive event. Indicates the generation of the first The source device for a unified interactive event. Indicates source device Time offset relative to the production line reference clock; After time correction, the production line data acquisition gateway groups unified interactive events using module identifiers and process identifiers as association keys. Within each group, events are arranged in ascending order of their corrected timestamps. When two unified interactive events have the same corrected timestamp, they are arranged according to the order in which they were written to the data acquisition gateway's cache, thus obtaining the data for the next step. A unified set of interactive events invoked.

[0022] In this specific embodiment, S2 includes: This is jointly executed by the production line data acquisition gateway and the process status service. The production line data acquisition gateway generates a unified set of interactive events from step S1. In the process, interactive events under the same association key are extracted according to the current module identifier and the current process identifier, and the unified interactive event with the latest corrected timestamp as the event source is the visual detection result is selected as the model determination event. The process status service reads the module shape detection result, component position detection result and identification character recognition result from the event payload of the model determination event. The module shape detection results are converted into a module shape feature vector, which is composed of the aspect ratio of the module's outer rectangle, the area ratio of the pole region, and the normalized value of the end plate height. The component position detection result is converted into a component position feature vector. The component position feature vector consists of the normalized coordinates of the positive terminal center, negative terminal center, busbar reference point, and end plate reference point in the module local coordinate system. The module local coordinate system takes the lower left corner of the module's circumscribed rectangle in the visual image as the origin, the module length direction as the horizontal axis, and the module width direction as the vertical axis. Each coordinate is divided by the length and width of the module's circumscribed rectangle to obtain the normalized coordinates. The identification character recognition result is converted into a standardized character result, which is obtained by removing spaces and separators, uniformly converting to uppercase characters, and retaining the model field. The process status service pre-establishes a visual feature template library corresponding to the current process identifier. The visual feature template library is not a training model, but is generated by ten qualified samples of each candidate module model in the first piece confirmation process. The average value of the module shape feature vectors of the ten qualified samples is used to obtain the module shape feature template of the candidate module model. The average value of the component position feature vectors of the ten qualified samples is used to obtain the component position feature template of the candidate module model. The model character registered in the product master data is used as the identification character template of the candidate module model. The process status service calculates the matching value for each candidate module model using the following formula: ; in, This indicates the candidate module model number corresponding to the current process identifier. Indicates the candidate module model The matching value, This represents the weight of the module's shape features and its value is... The value represents the weight of the component position feature and takes the value of . This represents the weight of the identifier character recognition result and its value is... This represents the module shape feature vector extracted from this model determination event. Indicates the candidate module model Module shape feature template, Denotes the Euclidean norm. This represents the normalized tolerance of the module's external shape features, and its value is [value missing]. This represents the component location feature vector extracted from this model determination event. Indicates the candidate module model Component position feature template, This represents the normalized tolerance of component position features and takes the value of . This indicates that the upper limit of the value within the parentheses is set to 1. This represents the standardized character result extracted from this model determination event. Indicates the candidate module model Identifier character template, Represents the result of normalized characters With identifier character template The character-by-character similarity is calculated by dividing the number of identical characters at the same position by the number of characters in the identifier character template. Total number of characters, standardized character result Positions with insufficient length will be treated as different. The process status service reads the current production file identifier from the production plan binding table based on the current module identifier and the current process identifier, and reads the applicable module model set, process version and effective status corresponding to the current production file identifier from the production file metadata table. Each record in the production file metadata table includes the production file identifier, applicable module model set, process version, effective status, effective time, file verification code and file type. The effective status is either effective or ineffective. The process status service reads the model judgment threshold from the verification rules corresponding to the current process identifier. ,in, This indicates the model determination threshold corresponding to the current process identifier, with a value of 0.85. This applies when the production file metadata is in an active state and a matching value exists. Reaching the model determination threshold Candidate module models When a candidate module model belongs to the set of applicable module models and has the largest matching value, the process status service determines the candidate module model as the current module model and sets the model confirmation flag to confirmed. When there is no candidate module model that meets the above conditions, or when there are two or more candidate module models with the largest matching value, the process status service will not write the current module model, and will set the current process status to model pending confirmation, while writing the reason for not meeting the conditions into the status reason field. Process status service also comes from a unified set of interactive events Extract equipment status events whose source is the equipment operating status and which are related to the current process identifier. For each equipment identifier, retain the latest corrected timestamp of the operating status code, permission status, alarm code, and measured parameters. Combine the current process identifier, current module model, process version, effective status, equipment operating status, model confirmation flag, and status reason fields to form the current process status. The version identifier of the current process status is obtained by concatenating the above fields in a fixed order and calculating the hash digest. The current process status is used for parsing interactive instructions in step S3.

[0023] In this specific embodiment, S3 includes: Executed by the instruction parsing service, the instruction parsing service receives the current process status output in step S2, and reads the current process identifier, current module model, equipment operating status, model confirmation flag, and status reason field from the current process status, while also retrieving the unified interactive event set generated in step S1. Extract the interactive events that are identical to the current module identifier and the current process identifier; The instruction parsing service identifies unified interactive events whose source is interface input and whose event type is submission confirmation, program call, file selection, parameter submission, start execution, stop execution, pause execution, reset execution, or exception handling submission as interactive instruction trigger events, and generates an instruction structure for each interactive instruction trigger event. The instruction dictionary is established according to the current process identifier. The instruction dictionary is not a training model, but a deterministic mapping table configured by process engineers before production line commissioning. Each dictionary record includes process identifier, control identifier set, event type, operation value keyword, instruction intent code, abnormal handling action code, and applicable state set. The instruction intent code is used to represent one of the following: file selection, welding program call, parameter setting, start welding, stop welding, pause welding, equipment reset, rework start, and abnormal recovery. The abnormal handling action code is used to represent one of the following: reset, clear alarm, return to origin, stop laser, release fixture, reposition, and rework call. When the control identifier, event type, and operation value keyword of the interactive instruction trigger event match the dictionary record at the same time, and the current process state belongs to the applicable state set of the dictionary record, the instruction parsing service writes the instruction intent code in the dictionary record into the instruction intent and writes the abnormal handling action code in the dictionary record into the abnormal handling action. The device address mapping table is established according to the process identifier. Each record in the device address mapping table includes the process identifier, control identifier, source device address, target device identifier, device type, device adapter identifier, and communication endpoint. The instruction parsing service first uses the control identifier and the current process identifier in the interactive instruction trigger event to query the device address mapping table. When a unique record is found, the target device identifier in the unique record is identified as the target device. When the control identifier fails to uniquely locate the target device, the source device address and the current process identifier in the event payload of the interactive instruction trigger event are used for a second query. If the second query still cannot find a unique record, the target device field is written as null and a target device missing flag is generated. Each record in the production document index includes a production document identifier, document name, document type, set of applicable module models, process version, effective status, and document verification code. Each record in the welding program index includes a welding program identifier, program name, set of applicable module models, target equipment identifier, parameter version, effective status, and program verification code. The instruction parsing service queries the production document index and welding program index based on the document name, program name, document number, or program number carried in the event payload of the interactive instruction triggering event, and performs consistency filtering based on the current module model, current process identifier, and target equipment. When the query result is a production document record, the production document identifier is written to the document or program identifier; when the query result is a welding program record, the welding program identifier is written to... When neither the production file record nor the welding program record is uniquely determined, the file or program identifier is written as a null value and a file or program missing flag is generated. The parameter mapping is established according to the current process identifier and the target equipment identifier. Each record of the parameter mapping includes a unified semantic field name, an equipment parameter field name, a field data type, a field unit, a value source priority, and a parameter group name. The unified semantic field name is consistent with the unified semantic field in step S1, and the equipment parameter field name is consistent with the parameter field that the target equipment adapter can recognize. The value source priority is determined according to the order of interactive instruction trigger event, the most recent interface input under the same module identifier and the same process identifier, the equipment running status in the current process status, and the production file metadata. The instruction parsing service extracts parameter fields item by item according to the parameter mapping. When the same unified semantic field has a value in multiple unified interaction events, the value whose corrected timestamp is no later than the interaction instruction trigger event and is closest to the interaction instruction trigger event is selected. When two values ​​have the same corrected timestamp, the value whose writing order to the production line acquisition gateway is later is selected. The extracted parameter fields are saved in the form of unified semantic field name, device parameter field name, parameter value, field unit, field data type and parameter group name. The instruction parsing service reads the list of required fields from the verification rules corresponding to the current process status. Each item in the list of required fields includes a unified semantic field name, a required field flag, and the instruction intent to which it belongs. When a required field corresponding to the instruction intent of the current interactive instruction fails to extract a non-empty value from the value source specified by the parameter mapping, a missing field flag is generated for the required field. The missing field flag includes the unified semantic field name, the instruction intent to which it belongs, the target device identifier, and the reason for the missing field. The instruction parsing service writes the field source for each parsed field. The field source includes the device identifier, module identifier, process identifier, corrected timestamp, event source, and event type of the unified interaction event. For fields from the current process status or index table, the field source is written with the version identifier of the current process status or the version identifier of the index table record. The instruction parsing service also writes a parsing confidence value for each parsed field. The parsing confidence value is a deterministic value. Specifically, a field where the control identifier exactly matches the instruction dictionary record has a confidence value of 1.00; a field where the operation value keyword matches the instruction dictionary record but the control identifier is indirectly located by the same parameter group name has a confidence value of 0.90; a field uniquely located by the production file index or welding program index has a confidence value of 1.00; a field uniquely located after filtering by the current module model, current process identifier, and target equipment has a confidence value of 0.95; a parameter field obtained by supplementing the input under the most recent associated key has a confidence value of 0.90; and a parameter field obtained by supplementing the input by supplementing the equipment running status in the current process status has a confidence value of 0.85. Each instruction structure is based on… Write to the instruction cache, where, Indicates the first The instruction structure corresponding to each interactive instruction. Indicates the sequence number of the interactive command. Indicates the first The instruction intent of each interactive command. Indicates the first The target device of the interactive command Indicates the first The file or program identifier for the interactive command. Indicates the first The set of parameter fields for each interactive command. Indicates the first Abnormal handling actions for interactive commands, Indicates the first A set of missing markers for interactive instructions. Indicates the first The set of fields from which the interactive instructions originate. Indicates the first The set of confidence values ​​for the parsed interactive instructions, the instruction structure, the version identifier of the current process status, and the corrected timestamp of the interactive instruction triggering event are saved together and used as input for joint verification in step S4.

[0024] In this specific embodiment, S4 includes: This is performed by the risk verification service, which receives the instruction cache written in step S3. Instruction structure ,in, Indicates the sequence number of the interactive command. Indicates the first The instruction structure corresponding to each interactive instruction. Indicates the first The instruction intent of each interactive command. Indicates the first The target device of the interactive command Indicates the first The file or program identifier for the interactive command. Indicates the first The set of parameter fields for each interactive command. Indicates the first Abnormal handling actions for interactive commands, Indicates the first A set of missing markers for interactive commands; The risk verification service reads the current process identifier, current module model, process version, effective status, equipment operating status, model confirmation flag, and version identifier of the current process status based on the current process status output in step S2, and calls the verification rule that corresponds to the current process status and is in effect. The verification rules are deterministic rule tables pre-stored in the rule version library, not training models. Each rule record of the verification rule includes process identifier, process version, rule version identifier, rule effective status, allowed correspondence between module model and production document, allowed correspondence between module model and welding program, mandatory field markers for each parameter field, field data type, numerical range, unit, permitted actions of target equipment in each operating state, set of permitted actions for the current process, preset operation sequence, first grade threshold, second grade threshold, and non-negative weights corresponding to each verification evidence. In this embodiment, the first graded threshold is denoted as The second-level threshold is denoted as This represents the boundary value between output release and output supplementation or secondary confirmation, and its value is [value missing]. This represents the boundary value between output supplementary recording or secondary confirmation and output risk interception, and its value is 0.50, and it satisfies the following conditions: ; The risk verification service first performs a matching verification between the module model and the file or program identifier. If the model confirmation flag is "confirmed," the current module model exists, and the file or program identifier... The existing production document or welding procedure is in an active state, and the current module model matches the document or procedure identifier. If the combination exists in the allowed correspondence, the validation passes; otherwise, the validation fails. The risk verification service then performs parameter integrity verification, according to the... The command intent of the interactive command Read the list of required fields from the validation rules and check the parameter field set item by item. Does the database contain parameter fields with identical field names, data types, and non-empty values? If all required fields meet these conditions and the missing marker set is complete... If the value is empty, the validation passes; otherwise, the validation fails. The risk verification service then performs parameter range and unit verification for the parameter field set. For each numeric parameter field, a unit consistency check is first performed based on the units registered in the validation rules. When the unit of the parameter field is inconsistent with the unit of the validation rules, a unit conversion table is called to perform a deterministic conversion. Each record in the unit conversion table includes the source unit, target unit, scaling factor, and bias. The converted parameter value is equal to the source parameter value multiplied by the scaling factor and then the bias. After the unit consistency is completed, it is determined whether the parameter value falls within the closed interval given by the validation rules. When all numeric parameter fields have valid units and all fall within the corresponding closed interval, the validation passes; otherwise, the validation fails. The risk verification service then performs equipment license status verification, reading the target equipment from the equipment operating status in the current process status. The corresponding latest operating status code, license status, and alarm code, when the target device The device exists, its permission status is allowed, its alarm code is empty, and it is the target device in the verification rules. The license action under the current running status code contains instruction intent. Or abnormal handling actions If the condition is met, the check passes; otherwise, the check fails. The risk verification service then performs a process permission action verification, reading the set of permitted actions for the current process from the verification rules, when the instruction intent... Actions that belong to the set of allowed actions, or actions that handle exceptions when the current process state is in an abnormal execution state. If the action belongs to the set of allowed actions, the verification passes; otherwise, the verification fails. The risk verification service then performs personnel operation trajectory verification from the unified interaction event set generated in step S1. Extract those that are identical to the current module identifier and the current process identifier, whose event source is personnel operation trajectory, and whose corrected timestamp is no later than the [number missing]. The personnel operation trajectory triggered by the interactive command is arranged in ascending order according to the corrected timestamp. The preset operation sequence is given by the verification rules. Each operation node in the preset operation sequence includes control identifier, operation type, operation value constraint and node sequence number. The risk verification service scans sequentially starting from the first personnel operation trajectory. When the control identifier of the current personnel operation trajectory is consistent with the control identifier of the current operation node, the operation type is consistent, and the operation value meets the operation value constraint of the operation node, the operation node is recorded as a match and the next operation node is matched. When it is not satisfied, the next personnel operation trajectory is scanned until the personnel operation trajectory is scanned. Thus, the number of operations that are in the same order as the preset operation sequence and whose operation values ​​meet the corresponding field constraints is obtained. The risk scoring adopts a deterministic weighted summation model. All model parameters are derived from the verification rules corresponding to the current process status. In this embodiment, the weight of module model matching with file or program identifier verification is 0.25, the weight of parameter integrity verification is 0.15, the weight of parameter range and unit verification is 0.20, the weight of equipment permission status verification is 0.15, the weight of process permission action verification is 0.10, and the weight of personnel operation trajectory verification is 0.15. The sum of the above non-negative weights is 1.00. The pass result of each verification is mapped to 0, and the failure result is mapped to 1. The risk score is calculated according to the following formula: ; in, Indicates the first Risk score for each interactive instruction This represents the weight of the verification process where the module model matches the file or program identifier, and its value is [value missing]. Indicates the first The module model of each interactive command is matched with the file or program identifier to verify the mapping value. This represents the weight for parameter integrity verification and its value is... Indicates the first The parameter integrity verification mapping value of the interactive command. The weights representing the parameter range and unit verification are given, and their values ​​are... Indicates the first The parameter range and unit verification mapping value of each interactive command. This represents the weight of the device license status verification and its value is... Indicates the first The device license status verification mapping value of the interactive command. This represents the weight of the allowed actions for the process and its value is [value]. Indicates the first The process of this interactive instruction allows the action to verify the mapping value. This represents the weight of the personnel operation trajectory verification and its value is [value]. Indicates the first The proportion of personnel operation trajectories matching the interactive commands is as follows: Indicates the first The number of operations in the personnel operation trajectory corresponding to each interactive instruction that have the same order as the preset operation sequence and whose operation values ​​satisfy the constraints of the corresponding fields. This indicates the total number of operations in the preset operation sequence corresponding to the current process identifier; Risk verification service will score risks With the first grade threshold Second grade threshold When comparing, Output the release result in time, and set the first The verification result of the interactive command is marked as allowed; when Output supplementary or secondary confirmation results in real time; if a missing set of markers is missing. If not empty, then based on the missing marker set. Generate fields to be added, including the unified semantic field name, the corresponding instruction intent, the target device identifier, and the reason for the absence. (If a set of missing tags is also included...) If empty, a secondary confirmation message will be generated, which includes the current module model and file or program identifier. Parameter field collection Target equipment Instruction intent and risk score After personnel complete the supplementary data entry or secondary confirmation, the instruction parsing service updates the instruction structure based on the supplementary data entry or confirmation result. The risk verification service verifies the updated instruction structure. Re-execute steps when The system outputs risk interception results in real time, and matches the verification items with a mapping value of 1 with the personnel operation trajectory matching ratio. For personnel operation trajectory verification items that do not meet version 1.00, write the blocking reason. The blocking reason includes the name of the failed verification item, the corresponding field name, the rule version identifier, the current field value, the rule allowed value, and the version identifier of the current process status. The risk verification service will match the verification results, mapping values, and personnel operation trajectories. Risk Score First-level threshold Second-level threshold The verification result and the reason for blocking are written to the verification result cache, and the instruction structure corresponding to the release result is written to the cache. Proceed to step S5.

[0025] In this specific embodiment, S5 includes: This process is jointly executed by the parameter distribution service and the target device adapter. The parameter distribution service only receives the first parameter from step S4 when the output is a release result. Instruction structure And from the first Instruction structure Read target device File or program identifier Parameter field collection Instruction intent and abnormal handling actions Simultaneously, the current module model, current process identifier, process version, and version identifier of the current process status are read from the current process status output in step S2; The parameter distribution service accesses the parameter version library, which is a pre-configured deterministic version table. Each parameter version record includes the current module model, current process identifier, file or program identifier, target device identifier, parameter version identifier, parameter version effective status, effective time, expiration time, parameter template, parameter check code, and device adapter identifier. The parameter version effective status can be either effective or ineffective. The parameter delivery service uses the current module model, current process identifier, and file or program identifier. and target equipment As a query key, select a unique parameter version from the parameter version library whose effective status is active and whose current time is between the effective time and the expiration time. If no unique parameter version exists, do not generate device control parameters and write the parameter version error into the device adaptation error information. When a unique parameter version exists, the parameter delivery service reads the parameter template corresponding to that parameter version. The parameter template includes the device parameter field name, field data type, target unit, target dimension, device resolution, device minimum value, device maximum value, default write order, and write confirmation method. The parameter delivery service simultaneously reads the current process identifier and the target equipment. The corresponding parameter mapping, the parameter mapping and steps The parameter mappings are kept consistent, and each mapping record includes a unified semantic field name, device parameter field name, field data type, field unit, target unit, scaling factor, bias, and parameter group name; The parameter delivery service processes device parameter fields one by one according to the default writing order in the parameter template. For each device parameter field, it first determines the corresponding unified semantic field name through parameter mapping, and then selects the parameter field set... The parameter value and field unit corresponding to the unified semantic field name are read. When the field unit is inconsistent with the target unit, the unit is converted according to the proportional coefficient and bias in the parameter mapping. When the field unit is consistent with the target unit, the proportional coefficient is 1 and the bias is 0. After the unit conversion is completed, the numerical discretization is performed according to the device resolution of the target device to obtain the device control parameters that the target device can recognize. The equipment control parameters are generated as follows: , ; in, Indicates the first The set of device control parameters corresponding to each interactive command. Indicates the target device A device parameter field name, Indicates the first The parameter template corresponding to each interactive command needs to be written to the target device. The set of device parameter field names, Indicates the first Device parameter field name in the interactive command The corresponding discretized parameter values, This indicates the field name of the device parameter as determined by the parameter mapping. The corresponding unified semantic field name, Represents a set of parameter fields Uniform Semantic Field Name The corresponding parameter values, Indicates the name of the device parameter field. The corresponding unit conversion ratio factor, Indicates the name of the device parameter field. The corresponding unit conversion bias, Indicates the name of the device parameter field. Corresponding device resolution, This represents the rounding function; The parameter distribution service generates a set of device control parameters. Then, for each device parameter field, a target device boundary check is performed again. This is done after discretization of the parameter values. If the value is less than the minimum device value in the parameter template or greater than the maximum device value in the parameter template, the distribution will not be executed, and the corresponding device parameter field name, discretized parameter value, and device allowable boundary will be written into the device adaptation error information. When all equipment control parameters pass the target equipment boundary check, the parameter distribution service generates a unique instruction identifier corresponding to this interaction instruction. The unique instruction identifier consists of the current module identifier, the current process identifier, the version identifier of the current process status, and the instruction structure. Cache sequence number, file or program identifier Target equipment Parameter version identifier and device control parameter set Calculate after concatenating according to a fixed field order. The abstract shows that the fixed field order remains unchanged in the parameter distribution service configuration file; The parameter delivery service will use a unique command identifier and target device. Device adapter identifier, parameter version identifier, file or program identifier Instruction intent Abnormal handling actions Equipment control parameter set The confirmation method and the version identifier of the current process status are encapsulated into a delivery request and sent to the target device adapter; After receiving the distribution request, the target device adapter first queries the local instruction execution registration table. The local instruction execution registration table includes a unique instruction identifier, a target device identifier, a parameter version identifier, a set of device control parameters, a distribution start time, a distribution end time, an execution status, and the previous device execution feedback. When the same unique instruction identifier already exists in the local instruction execution registration table, the target device adapter directly returns the previous device execution feedback corresponding to that unique instruction identifier and does not write the device control parameters to the target device again. When no identical unique instruction identifier exists in the local instruction execution register, the target device adapter verifies the target device in the issuance request. Check if the target device identifier matches the one bound to the adapter; verify if the parameter version identifier matches the parameter version identifier that is currently active in the parameter version library; verify the set of device control parameters. Check whether all device parameter field names exist in the target device communication protocol field table, and verify whether the unique instruction identifier can be recalculated from the fields in the issued request in a fixed field order; After all the above verifications pass, the target device adapter will set the device control parameters according to the target device communication protocol field table. The target device communication protocol field table includes device parameter field names, register addresses or command words, data length, data type, byte order, write order, and response code definitions. The target device adapter writes communication packets sequentially according to the default writing order in the parameter template. After each write, it reads the response code and readback parameter value returned by the target device. When the response code indicates success and the readback parameter value is consistent with the discretized parameter value, the device parameter field is marked as written successfully. When the response code indicates failure, communication timeout, or inconsistent readback parameter value, the device parameter field is marked as written failed. Once all device parameter fields have been successfully written, the target device adapter will execute the command. Send execution commands corresponding to startup, invocation, parameter loading, or exception handling, and receive device execution feedback returned by the target device. The device execution feedback includes a unique instruction identifier, target device identifier, parameter version identifier, execution success flag, device return code, measured parameters, device operating status code, alarm code, feedback timestamp, and adapter return information. When there is a failure to write device parameter fields, the execution command is rejected by the target device, communication timeout, or the target device returns an alarm code, the target device adapter will set the execution success flag to failure and write the failure field, device return code, alarm code, and adapter return information into the device execution feedback. The target device adapter writes the device execution feedback into the local instruction execution register and returns it to the parameter distribution service. The parameter distribution service then sends the set of device control parameters. The parameter version identifier, unique instruction identifier, adapter return information, and device execution feedback are written into the distribution result cache, and the device execution feedback is passed to step S6.

[0026] In this specific embodiment, S6 includes: The process is jointly executed by the feedback calibration service, the status recalculation service, and the traceability service. The feedback calibration service receives the device execution feedback written into the distribution result cache in step S5, and associates it with the first result formed in step S3 using a unique instruction identifier. Instruction structure The verification result cache formed in step S4 and the set of device control parameters formed in step S5 ; The feedback calibration service reads the execution success flag, device return code, measured parameters, device operating status code, alarm code, feedback timestamp, and adapter return information from the device execution feedback, and also reads the unified interactive event set generated in step S1. The system reads the post-weld visual inspection results that share the same current module identifier, the same current process identifier, and whose corrected timestamp is later than the feedback timestamp. The post-weld visual inspection results include weld identifier, defect category, defect size, inspection completion time, and image identifier. Feedback calibration service will allow personnel to verify the instruction structure during supplementary data entry or secondary confirmation. The field modification record is used as a personnel correction sample. The field modification record includes the modified unified semantic field name, the field value before modification, the field value after modification, the modification time, the operator's identifier, and the modification source. When the modified unified semantic field name belongs to the personnel correction field registered in the verification rule corresponding to the current process status, the personnel correction sample is marked as abnormal. The feedback calibration service combines the execution success flags from the device's execution feedback with the set of measured parameters and device control parameters. The deviation between them is used as an execution sample. The measured parameters are first converted to the set of equipment control parameters according to the parameter mapping in step S5. A consistent target unit is used, and then compared with the discretized parameter values ​​of the corresponding device parameter fields. Successful execution is marked as failure, a device return code indicates rejection, an alarm code is not empty, and the relative deviation of the laser power exceeds [a certain threshold]. The relative deviation of welding speed exceeds If the welding position deviation exceeds 0.10mm or the fixture pressure deviation exceeds 0.05MPa, the sample will be marked as abnormal. The feedback calibration service uses the defect category and defect size in the post-weld visual inspection results as a quality sample, and compares them with the current module model, document, or program identifier. The quality verification rules corresponding to the current process identifier and in an effective state read the quality limit value corresponding to the defect category. The quality verification rules are a deterministic rule table. Each rule record includes the current module model, current process identifier, file or program identifier, quality rule version identifier, effective state, defect category, quality limit value, rework program identifier, and next process start interface identifier. In this embodiment, the porosity limit value is 0.20mm, the undercut depth limit value is 0.10mm, the weld offset limit value is 0.15mm, and the burn-through size limit value is 0.00mm. When the defect size in the post-weld visual inspection result exceeds the corresponding quality limit value, the quality sample is marked as abnormal. Feedback calibration service was finally released and the process was completed. Interactive commands are used as statistical objects, within the same current module model, file, or program identifier. Select the nearest one under the same current process identifier The granted permission command forms a preset sample window. The value is 100. When the number of released instructions is less than 100, only samples are recorded without correcting the grading threshold. When the number of released instructions reaches 100, the calibration value is generated and the first and second grading thresholds are corrected as follows: , , ; in, Indicates the first The calibration amount for subsequent verification corresponding to each interactive instruction. This indicates the number of allowed instructions within the preset sample window, and its value is... This indicates the sequence number of the allowed instruction within the preset sample window. Indicates the first The first preset sample window corresponding to the interactive command An abnormal sample of a released instruction is marked when at least one of the personnel correction sample, execution sample, and quality sample is marked as abnormal. ,otherwise This represents the cumulative value of the abnormal sample markers within the preset sample window. Represents the proportion of target abnormal samples and takes the value of This indicates the first grading threshold used in step S4. This indicates the second grading threshold used in step S4. This represents the corrected first-level threshold. This represents the corrected second-level threshold. This represents the threshold sensitivity coefficient for the first grade threshold, and its value is... This represents the threshold sensitivity coefficient for the second-level threshold and takes a value of [value missing]. This represents the lower limit of the first-level threshold boundary and has a value of 0.10. This represents the upper limit of the first-level threshold boundary and its value is [value missing]. This represents the lower limit of the second-level threshold boundary and its value is... This represents the upper limit of the second-level threshold boundary and its value is... This represents the minimum interval between the first and second level thresholds, and its value is... Indicates the numerical value Limit to the lower limit and upper limit The truncation function between them This indicates that the maximum value within the parentheses is taken; when A value greater than 0 indicates that the proportion of abnormal samples within the preset sample window is higher than the target proportion of abnormal samples. The feedback calibration service lowers the first and second level thresholds used in subsequent verifications to improve interception sensitivity. When the value is less than 0, it means that the proportion of abnormal samples in the preset sample window is lower than the target proportion of abnormal samples. The feedback calibration service increases the first and second grade thresholds used for subsequent verification within the threshold boundary. The corrected first and second grade thresholds are written into the runtime threshold table corresponding to the current process status, and the original verification rule version identifier is retained unchanged. When the device execution feedback indicates failure to send a message, device refusal to execute, communication timeout, alarm code not being empty, or measured parameters exceeding the value range in the verification rules used in step S4, the status recalculation service sets the current process status to an execution abnormal state and establishes a connection with the current module model, file, or program identifier. Current process identification and target equipment The bound freeze record includes the freeze start time, freeze reason, associated unique instruction identifier, and the identifier of the frozen file or program. The set of subsequent release actions following the freeze, and the risk verification service identifying the same file or program during the freeze period. The risk interception is directly output for the initiation of welding, welding program call, parameter loading, and the start of the next process; The status recalculation service reads the verification rules corresponding to the current process status and the equipment return code, alarm code, or executes the exception handling action corresponding to the abnormal status, when the instruction structure... Abnormal handling actions have already been taken. This abnormal handling procedure will be used in the future. When the instruction structure There were no abnormal handling actions. The exception handling actions in the verification rules are written into the instruction structure. And as an abnormal handling action Subsequently, an exception handling action was issued through the target device adapter. After the target device adapter returns the abnormal handling execution feedback, the status recalculation service checks the execution success flag, device return code, alarm code, device running status code and permission status in the abnormal handling execution feedback according to the verification rules corresponding to the current process status. When the execution success flag is successful, the device return code belongs to the pass return code set, the alarm code is empty and the permission status is allowed, the execution result of the abnormal handling action is determined to be pass. The state recalculation service reacquires the target device. The equipment license status, the current module model confirmation is completed again according to step S2, and the execution result of the abnormal handling action is passed, the freeze record is lifted, and the current process status is updated from execution abnormal status to pending execution status or the continued processing status corresponding to the current process identifier; When the defect size in the post-weld visual inspection result exceeds the quality limit in the quality verification rules, the status recalculation service sets the current process status to a quality abnormality status, reads the rework program identifier from the quality verification rules and adds the rework program to the executable operation scope, and removes the next process start interface identifier from the equipment interface whitelist, so that the human-machine interface displays the rework start control and hides or disables the next process start control, and the equipment interface gateway refuses the next process start interface call. When the process identifier, current module model, equipment operating status, freeze record, quality anomaly status, or running threshold table changes, the status recalculation service recalculates the range of executable operations. The range of executable operations is jointly determined by the set of allowed actions for the current process, the permitted actions of the target equipment in the current operating status, the set of restricted actions in the freeze record, and the set of additional actions for quality anomalies. During the calculation, the set of allowed actions for the current process is read first, then the intersection with the permitted actions of the target equipment is calculated, then the actions in the set of restricted actions in the freeze record are deleted, and finally, when there is a quality anomaly status, the rework initiation action corresponding to the rework procedure is added. The status recalculation service converts the scope of executable operations into the enabled, disabled, shown, and hidden states of interface controls based on the interface control mapping table, and converts the scope of executable operations into a device interface whitelist based on the device interface mapping table. Each record in the interface control mapping table includes a process identifier, action code, control identifier, and control status. Each record in the device interface mapping table includes a process identifier, action code, device interface name, target device identifier, and interface permission status. The refreshed interface control status is sent to the human-machine interface, and the refreshed device interface whitelist is sent to the device interface gateway. The version identifier of the scope of executable operations is calculated by concatenating the current process identifier, current module model, device operating status, frozen record identifier, quality anomaly status, rework program identifier, and device interface whitelist in a fixed order. Abstract obtained; The traceability service establishes a traceability record for each interactive instruction. For interactive instructions that have been released and completed step S5, the unique instruction identifier generated in step S5 is used as the traceability primary key. Intermediate records formed during the supplementary recording, secondary confirmation, and risk interception processes are merged using the unique instruction identifier of the final release. For interactive instructions that have not been finally released, a unique instruction identifier is generated using the instruction structure cache sequence number and the empty device control parameter set, which is only used for traceability. The traceability record includes the unified interaction event generated in step S1, the version identifier of the current process status generated in step S2, and the instruction structure generated in step S3. ,step Version identifier of the verification rules used, version identifier of the quality verification rules, verification results for each item, risk score, first-level threshold, second-level threshold, verification results, and steps. The generated set of device control parameters Equipment adapter return information, equipment execution feedback, post-weld visual inspection results, calibration quantities The traceability service associates the above fields with a unique instruction identifier and saves them by appending them, including freeze records, abnormal handling action execution results, rework procedure addition results, equipment interface whitelist refresh results, and version identifiers of the executable operation range. This allows the entire process of the same interactive instruction, from the generation of a unified interactive event, the establishment of the current process status, instruction parsing, risk verification, equipment adaptation and distribution, execution feedback, post-weld quality inspection, threshold calibration to operation range refresh, to be fully verified.

[0027] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

[0028] This invention unifies multimodal events, establishes model and process status, structures interactive instructions, performs risk classification verification, and distributes equipment adaptation information into a continuous processing chain. This allows visual evidence, equipment status, and personnel operation evidence to jointly participate in the judgment of whether control instructions are executable, and connects each step through clear intermediate results.

[0029] This invention utilizes personnel correction samples, equipment execution samples, and post-weld quality samples to generate calibration quantities, which are then used to make bounded corrections to subsequent grading thresholds. Furthermore, state changes are used as conditions for recalculating the operating range, transforming the method from static manual confirmation to a closed-loop interactive control structure that updates with production line status and execution quality.

Claims

1. A multimodal interactive control method for an energy storage module production line, characterized in that, include: S1. Collect interface inputs, visual inspection results, equipment operating status and personnel operation trajectories, and convert them into a unified set of interactive events according to equipment identifiers, module identifiers, process identifiers and timestamps; S2. Determine the current module model based on the visual inspection results and production file metadata in the set of interactive events, and establish the current process status by combining the process identifier and equipment operating status; S3. Parse the interactive instructions in the current process state to form an instruction structure that includes the instruction intent, target device, file or program identifier, parameter fields, and exception handling actions; S4. According to the verification rules corresponding to the current process status, jointly verify the matching relationship between the current module model and the file or program identifier, the integrity and range of parameters, the equipment permission status, the allowed actions of the process and the personnel operation trajectory, integrate the verification evidence into a risk score and compare it with the classification threshold, and output one of the following verification results: release, supplementary recording or secondary confirmation, and risk interception. S5. Response release result: Generate device control parameters based on the current module model, file or program identifier and parameter fields, send them to the target device adapter and obtain device execution feedback; S6. Update the current process status based on equipment execution feedback and post-weld visual inspection results, and generate calibration quantities for subsequent verification. Recalculate the executable operation range when the process identifier, current module model, or equipment operating status changes, and refresh the interface control and equipment interface whitelist accordingly.

2. The multimodal interactive control method for energy storage module production lines according to claim 1, characterized in that, S1 includes: Establish field mappings between source fields and unified semantic fields for interface input, visual inspection results, equipment operating status, and personnel operation trajectories; Correct the timestamp based on the offset between the source device clock and the production line reference clock; The interactive events are arranged according to the corrected timestamps, using module identifiers and process identifiers as association keys. The personnel operation trajectory consists of control identifiers, operation types, operation values, and operation times.

3. The multimodal interactive control method for energy storage module production lines according to claim 2, characterized in that, S2 includes: The module shape features, component position features, and identification character recognition results are extracted from the visual inspection results and matched with the visual feature templates corresponding to the candidate module models to obtain the matching value of each candidate module model. The applicable module model, process version, and effective status are read from the production file metadata. The candidate module model whose matching value reaches the model determination threshold and is consistent with the applicable module model is determined as the current module model. The model determination threshold is taken from the verification rule corresponding to the current process identifier. When there is no candidate module model that meets the conditions, the current process status is set to the model pending confirmation status.

4. The multimodal interactive control method for energy storage module production lines according to claim 3, characterized in that, S3 includes: identifying the instruction intent and abnormal handling action based on the instruction dictionary corresponding to the current process identifier; determining the target equipment based on the equipment address mapping table; determining the file or program identifier based on the production file index and welding program index; extracting parameter fields based on the parameter mapping between unified semantic fields and equipment parameter fields; generating missing markers for required fields that could not be extracted; and writing the parsed fields, field sources, and parsing confidence values ​​into the instruction structure.

5. The multimodal interactive control method for energy storage module production lines according to claim 4, characterized in that, S4 includes: reading from the verification rules the permissible correspondence between the module model and the production file, welding program, the mandatory field markers, numerical ranges and units of each parameter field, the permissible actions of the target equipment in each operating state, the set of permissible actions for the current process, the preset operation sequence, and the non-negative weights that correspond one-to-one with each verification evidence and sum to one; mapping the pass result of each verification to zero and the fail result to one; dividing the number of operations in the personnel operation trajectory that are in the same order as the preset operation sequence and whose operation values ​​satisfy the corresponding field constraints by the total number of operations in the preset operation sequence to obtain the consistency ratio; multiplying the value obtained by subtracting the consistency ratio from one and the mapping value of each verification by the corresponding weight and summing them to obtain the risk score.

6. The multimodal interactive control method for energy storage module production lines according to claim 5, characterized in that, The grading threshold includes a first grading threshold and a second grading threshold read from the verification rules corresponding to the current process status, wherein the first grading threshold is less than the second grading threshold; Comparing the risk score with the first grading threshold and the second grading threshold includes: when the risk score is less than the first grading threshold, outputting permission to proceed; When the risk score is greater than or equal to the first grading threshold and less than the second grading threshold, the field to be supplemented is output according to the missing marker, or a secondary confirmation information containing the current module model, file or program identifier and parameter fields is generated, and step S4 is re-executed after supplementation or secondary confirmation; when the risk score is greater than or equal to the second grading threshold, risk interception is output, and the verification item that caused the failure result is used as the reason for interception.

7. The multimodal interactive control method for energy storage module production lines according to claim 6, characterized in that, S5 includes: Based on the current module model, file or program identifier, and parameter fields, the parameter version that is in effect is selected from the parameter version library. The parameter fields are converted into target device fields using the parameter mapping. The unit conversion and numerical discretization are performed according to the dimensions and resolution of the target device to generate the device control parameters. Before issuing the command, the device adapter verifies the target device identifier, parameter version, and unique instruction identifier corresponding to the current interaction command. If the unique instruction identifier is received repeatedly, the adapter returns the previous device execution feedback without repeating the execution.

8. The multimodal interactive control method for energy storage module production line according to claim 7, characterized in that, S6 include: The modification records of the instruction structure by personnel during supplementary recording or secondary confirmation are used as personnel correction samples. The deviation between the execution success flag and measured parameters in the equipment execution feedback and the equipment control parameters generated according to the current module model, file or program identifier and parameter fields are used as execution samples. The weld defect category and defect size in the post-weld visual inspection results are used as quality samples. The system reads the personnel correction field, execution failure flag, allowable deviation of measured parameters, quality limit value of each defect category, target abnormal sample ratio, threshold sensitivity coefficient and threshold boundary from the valid and quality valid rules, and marks abnormal samples accordingly; it counts the abnormal sample ratio corresponding to the released instruction within the preset sample window, and uses the difference between the abnormal sample ratio and the target abnormal sample ratio as the calibration amount; it subtracts the product of the calibration amount and the corresponding threshold sensitivity coefficient from the first grade threshold and the second grade threshold respectively, and restricts the corrected threshold within the corresponding threshold boundary, while keeping the first grade threshold less than the second grade threshold.

9. The multimodal interactive control method for energy storage module production lines according to claim 8, characterized in that, When the device execution feedback indicates that the delivery failed, the device refused to execute, or the measured parameters exceeded the specified value range, the current process status is set to an execution abnormal state, and subsequent release operations related to the file or program identifier are frozen. The abnormal handling action is issued through the target device adapter, and the execution result of the abnormal handling action is determined according to the returned execution feedback and the verification rules corresponding to the current process status. The freeze is lifted after the device permission status is regained, the module model confirmation is completed again, and the execution result is passed. Read the quality limit corresponding to the defect category from the quality verification rule that corresponds to the current module model, the file or program identifier and the current process identifier and is in effect. When the defect size in the post-weld visual inspection result exceeds the quality limit, add the rework program to the executable operation range and remove the start interface of the next process from the equipment interface whitelist.

10. The multimodal interactive control method for energy storage module production lines according to claim 9, characterized in that, A traceability record is established for each interactive instruction. The traceability record includes a unified interactive event, a version identifier of the current process status, an instruction structure, version identifiers of the verification rules and quality verification rules used, various verification results, risk scores, grading thresholds, verification results, equipment control parameters generated based on the current module model, file or program identifier and parameter fields, equipment adapter return information, equipment execution feedback, post-weld visual inspection results, calibration quantities and version identifiers of the executable operation range, and each field in the traceability record is associated with the unique instruction identifier.

Images