Liquid packaging equipment maintenance operation instruction book generation method and system

Abstract

A liquid packaging equipment maintenance work instruction generation method and system based on semantic processing and step executable verification. The method obtains maintenance task context, receives oral maintenance steps, analyzes to obtain step elements, step types and source basis, and generates standardized candidate steps containing standard actions, operation objects, required tools, safety prerequisites, acceptance conditions, source basis identification and step sequence information; then the candidate steps are mapped to the data association network, and the source basis verification, measurability verification, operability verification and step sequence constraint verification are performed. When there are unconfirmed model completion items, no source basis, object or tool mismatch, insufficient safety prerequisites, missing acceptance conditions or prohibited condition triggers, the system prohibits automatic writing of the formal instruction. It can reduce the risk of non-executable or non-referenced steps entering the formal instruction.

Description

Technical Field

[0001] This invention relates to the fields of industrial automation, digital manufacturing, natural language processing, and industrial maintenance management, and in particular to a method and system for generating maintenance operation manuals for liquid packaging equipment based on semantic processing modules and step executable verification. Background Technology

[0002] Maintenance of liquid packaging equipment typically involves multiple steps, including shutdown, energy isolation, disassembly and assembly, cleaning, inspection, adjustment, reset, and trial operation. Field engineers often use colloquial expressions when describing these steps, such as "stop the machine," "unscrew the cap," or "check if the belt is loose." Directly incorporating these expressions into the maintenance work instructions may lead to personnel misunderstanding the objects being operated on, the tools used, acceptance criteria, or safety requirements.

[0003] Existing templated document generation systems can generate job files based on fixed fields, but their ability to understand and rewrite unstructured step descriptions is limited. While general natural language processing models can polish text, they are prone to generating step descriptions that do not conform to field conditions if industrial context such as equipment models, tool lists, safety constraints, and operating conditions are lacking. For example, a step may require the use of a specific diagnostic instrument, but the current tool list does not include it; or it may require the disassembly of a component, but the corresponding equipment model does not have that structure.

[0004] Existing solutions often focus on template field filling, video editing, or generative question answering, making it difficult to ensure that each candidate step has a traceable source, and also difficult to determine whether the sequential relationship between steps such as disassembly, cleaning, reassembly, and trial operation satisfies the equipment maintenance logic. If the system only outputs polished natural language text, users will find it difficult to distinguish whether the text comes from the equipment manual, historical records, rule base, or model inference.

[0005] Therefore, a step generation technology solution is needed for maintenance scenarios of liquid packaging equipment, which can enable the system to convert colloquial maintenance steps into standard industrial written language, and perform measurability, operability, relevance and timeliness checks on the converted steps, thereby reducing the number of unclear, unexecutable or mismatched work steps that enter the final instruction manual. Summary of the Invention

[0006] In view of the technical problems existing in the prior art, the present invention aims to provide a system that can solve the problems of oral communication in maintenance steps, unclear action objects, lack of quantitative or observable acceptance standards, mismatch between tools and equipment conditions, and difficulty in constraining model output.

[0007] According to one aspect of the present invention, the present invention provides a method for generating maintenance operation instructions for liquid packaging equipment based on semantic processing and step executable verification, comprising: Obtain the context of the liquid packaging equipment maintenance task; Receive user-inputted, conversational maintenance steps; The colloquial maintenance steps are analyzed using a semantic processing module and a rule base to obtain step elements and step types, and to determine the source basis. Standardized candidate step objects are generated based on step elements. The objects include one or more of the following: standardized step text, action words, operation objects, required tools, safety prerequisites, acceptance conditions, source basis identifiers, step types, prerequisite step identifiers, and prohibition conditions. Standardized candidate step objects are mapped to a data association network, and source basis verification, measurability verification, operability verification, and step sequence constraint verification are performed. When a candidate step object does not match the source basis, contains unconfirmed model completion items, lacks acceptance conditions, the object or tool is mismatched, the security premise is not met, the prerequisite steps are missing, or the prohibition condition is triggered, a supplementary prompt is generated and automatic writing is prohibited. When the above verification passes and there are no unconfirmed model completion items, the standardized candidate step objects and verification results are written into the structured operation step record, and a maintenance operation instruction document is generated based on the record.

[0008] In this solution, standardized candidate steps cannot be viewed merely as polished text. They also need to be mapped to a data association network and satisfy constraints on source basis, equipment and tool matching, and step sequence. The data association network includes at least equipment model nodes, component nodes, tool nodes, operation nodes, safety condition nodes, step type nodes, and source basis nodes. The operation object, required tools, safety prerequisites, acceptance conditions, and step type in the candidate steps are matched with the corresponding nodes or relationships. Here, "standardized candidate steps" refer to the steps to be verified generated by the semantic processing module, rule base, and terminology mapping rules from the colloquial maintenance steps. This step is not yet the final formal step written into the guidance document. Standardized candidate steps can at least include one or more of the following: standard action, operation object, required tools, safety prerequisites, execution content, acceptance conditions, step type, and source basis identifier.

[0009] In this scheme, the "data association network" is a structured data set used to constrain the executability of candidate steps. It can be implemented using relational tables, graph databases, key-value data structures, or other queryable data structures. The "relationship edge" in the data association network refers to the associated record between two nodes, such as the inclusion relationship between a device model node and a component node, the applicability relationship between a component node and an operation node, the requirement relationship between an operation node and a tool node, and the constraint relationship between a step type node and its preconditions.

[0010] In this solution, "source basis" refers to the basis from which a candidate step can be included in the instruction manual, and is not limited to the output of the semantic processing module itself. Source basis can come from text fragments in the user's original input, paragraphs in the equipment manual, clauses in the company's operating procedures, historical maintenance records, terminology mapping rules, or manually confirmed information. Source basis verification is used to determine whether a candidate step matches at least one source basis, and to determine whether the operation object, tool, parameter, and acceptance conditions in the candidate step are consistent with the matched source basis.

[0011] In this solution, "Measurability Verification" is used to determine whether candidate steps include quantifiable indicators, observable states, or verifiable acceptance conditions; "Operability Verification" is used to determine whether the operation object belongs to the current equipment model, whether the required tools exist in the tool list, whether the safety prerequisites are met, and whether the current working conditions allow execution; "Step Sequence Constraint Verification" is used to determine whether the current step lacks a prerequisite step or triggers a prohibition condition based on the step type.

[0012] In this solution, the semantic processing module assists in recognizing spoken expressions, ambiguous words, and non-standard terms, and generates candidate standardized steps. The semantic processing module can include one or more of a domain language model, a natural language processing model, or a rule engine. The system does not directly use the output of the semantic processing module as the final step; instead, it constrains and confirms the candidate steps through a rule base, terminology mapping table, data association network, source basis records, step order constraints, measurability rules, and operability rules.

[0013] The source reference record can include one or more of the following: original user input, equipment manual paragraphs, enterprise operating procedures, historical maintenance records, terminology mapping rules, and manual confirmation information. If a candidate step does not match any source reference, or if the operation object, tool, parameter, or acceptance condition in the candidate step is inconsistent with the source reference, the system marks it as pending review and will not include it in the formal instruction manual. Step sequence constraints are used to determine whether preset pre- and post-step conditions are met between steps such as shutdown, energy isolation, disassembly, cleaning, inspection, reassembly, reset, and trial operation.

[0014] In particular, in a scenario involving a write decision relationship, the system can determine the write status based on the following criteria: "Write Allowed = Source Basis Passed + Measurability Passed + Operability Passed + Step Sequence Passed + No Unconfirmed Model Completion Items Exist + No Prohibition Conditions Triggered". Here, "+" indicates that all conditions must be met, not that the values ​​are added together. If there are no basis or unconfirmed model completion items, mismatched equipment objects, missing tools, missing acceptance conditions, or triggered prohibition conditions, the candidate step remains in a state of pending supplementation or review.

[0015] In this solution, the semantic processing module includes one or more of a domain language model, a natural language processing model, or a rule engine; the standardized candidate step object output by the semantic processing module is the object to be verified. If there is content added by the semantic processing module in the standardized candidate step object that does not match the source basis, the content is marked as a model completion item; the standardized candidate step object containing unconfirmed model completion items shall not be automatically written into the structured operation step record or maintenance operation instruction file.

[0016] In this solution, the step elements include one or more of the following: action word, operation object, operation tool, component name, direction, parameter, unit, status word, acceptance condition, time limit, safety prerequisite, step type, and abnormal handling requirements; the source basis includes one or more of the following: user's original input, equipment manual, enterprise operation specifications, historical maintenance records, terminology mapping rules, or manual confirmation information.

[0017] In this solution, the steps for generating standardized candidate actions include: mapping colloquial actions to standard industrial operation actions, mapping colloquial component names to standard component names corresponding to the current equipment model, and completing safety prerequisites or acceptance conditions based on safety constraints.

[0018] In this solution, measurability verification includes: determining whether the standardized candidate steps contain quantitative indicators, observable states, or verifiable acceptance conditions; if quantitative indicators, observable states, or verifiable acceptance conditions are missing, a supplementary prompt requiring supplementary acceptance criteria is generated.

[0019] In this scheme, the operability verification includes: based on the data association network of equipment, tools, operating conditions and safety constraints, determining whether the current equipment model node is connected to the component node corresponding to the operation object of the standardized candidate step object through a valid relationship, whether the tool list contains tool nodes that have a valid relationship with the operation action of the standardized candidate step object, whether the safety premise matches the safety condition node, and whether the current working conditions allow the execution of the standardized candidate step object.

[0020] In this scheme, the data association network includes equipment model nodes, component nodes, tool nodes, operation nodes, safety condition nodes, skill condition nodes, step type nodes, source basis nodes, and relationship edges representing the relationships between equipment and components, components and operations, operations and tools, operations and safety conditions, step types and prerequisites, and candidate steps and source basis.

[0021] In this scheme, determining whether a standardized candidate step should be written into the structured work step record includes generating measurability results, operability results, relevance results, timeliness results, source basis results, step sequence results, prohibition condition results, and manual confirmation results. Among them, the source basis result indicates whether the standardized candidate step matches the source basis and whether the operation object, tool, parameter, or acceptance condition is consistent with the source basis; the step sequence result indicates whether the current step sequence is missing a prerequisite step; and the prohibition condition result indicates whether there are one or more prohibition conditions, such as incomplete energy isolation, unreset protective cover, unconfirmed cleaning agent residue, or missing specified tools from the tool list. Only when all of the above results meet the preset writing conditions will the standardized candidate step be written into the structured work step record or maintenance operation instruction document.

[0022] In this scheme, when the verification fails, the system generates a failure handling record. The failure handling record includes one or more of the following: failure reason, triggering rule, corresponding node identifier, corresponding relationship edge identifier, missing field, suggested supplementary item, whether manual review is allowed, review permission, whether re-verification is required after supplementation, and blocking reason. When the semantic processing module generates multiple standardized candidate step objects, the system determines the candidate step ranking based on the number of source criteria hits, the number of rule hits, the degree of data association network matching, the number of historical confirmations, the number of step order conflicts, and the number of risk flags. The system prioritizes the standardized candidate step objects that are ranked higher and have not triggered blocking conditions.

[0023] According to another aspect of the present invention, the present invention also provides a system for generating maintenance operation instructions for liquid packaging equipment based on semantic processing and step executable verification, characterized in that it includes: The task context acquisition module is used to acquire the task context of liquid packaging equipment maintenance. The step input module is used to receive user-inputted, conversational maintenance steps; The semantic processing module is used to parse colloquial maintenance steps using one or more of the following: a domain language model, a natural language processing model, or a rule engine, along with a rule base, to obtain step elements and step types, and to determine the source basis. The data association network is used to store the relationships between device model nodes, component nodes, tool nodes, operation nodes, security condition nodes, step type nodes, and source basis nodes. The step standardization module is used to generate standardized candidate step objects based on step elements. The step verification module is used to perform source basis verification, measurability verification, operability verification, and step sequence constraint verification on standardized candidate step objects. The feedback and correction module is used to generate supplementary prompts and receive supplementary information from users when the verification fails. The structured record generation module is used to generate structured job step records when the verification passes. The file generation interface is used to generate maintenance operation instruction files based on structured operation step records.

[0024] According to another aspect of the present invention, an electronic device is provided, characterized in that it includes a processor and a memory, wherein the memory stores a computer program, which, when executed by the processor, implements the above-described method. A computer-readable storage medium is also provided, characterized in that it stores a computer program, which, when executed by a processor, implements the above-described method.

[0025] According to the present invention, at least the following technical effects are achieved: It can convert the natural language descriptions of field engineers into structured candidate step objects with action words, operation objects, tools, safety prerequisites, acceptance conditions, source basis, and step type; it can identify steps lacking quantitative indicators or observable states; it can identify unexecutable steps based on equipment model, tool list, and operating conditions; it can reduce the risk of unfounded content entering the formal instruction manual by recording source basis and marking model completion items; and it can also identify step sequence problems that lack prerequisite steps or trigger prohibition conditions, improving the traceability and executability of the maintenance instruction manual generation process. Attached Figure Description

[0026] Figure 1 This is a flowchart illustrating a method for generating maintenance operation instructions for liquid packaging equipment based on semantic processing and step-executable verification, according to an embodiment of the present invention.

[0027] Figure 2 This is a structural block diagram illustrating a liquid packaging equipment maintenance operation instruction generation system based on semantic processing and step executable verification, according to an embodiment of the present invention.

[0028] Figure 3 The flowchart illustrates the colloquialization step analysis, source basis determination, and candidate step generation of an embodiment of the present invention.

[0029] Figure 4 This is a schematic diagram illustrating a data association network of devices, tools, step types, source basis, and prohibition conditions according to an embodiment of the present invention.

[0030] Figure 5 The flowchart illustrates the step verification feedback, failure handling record, supplementary review, and re-verification process according to an embodiment of the present invention. Detailed Implementation

[0031] The present invention will now be described in detail with reference to the accompanying drawings and specific embodiments. However, this description is exemplary and the present invention is not limited to these specific embodiments.

[0032] Figure 1 This is a flowchart of a maintenance work instruction generation method based on semantic processing and step executable verification, which includes candidate step object generation, verification blocking, and controlled writing, according to one embodiment. Figure 2 This is a system structure diagram. Figure 3 Flowcharts are generated for analyzing conversational steps, determining source evidence, and generating candidate steps. Figure 4 A diagram illustrating the network association between equipment, tools, step types, source basis, and prohibition conditions. Figure 5 This is a flowchart for step verification feedback, failure handling records, supplementary review, and re-verification.

[0033] like Figure 1 As shown, the maintenance work instruction generation system (hereinafter referred to as the system) Figure 2 (Detailed explanation follows) First, the maintenance task context of the liquid packaging equipment is obtained. This maintenance task context may include one or more of the following: maintenance task name, equipment name, equipment model, unit to which it belongs, maintenance content, work area, required tools, shutdown status, energy isolation tagging and locking method, safety constraints, pre-operation preparation requirements, and operating procedures. The maintenance task context is used to constrain the semantic normalization and executability judgment of subsequent steps.

[0034] The system receives user-inputted, conversational maintenance steps. Users can input multiple steps at once or one step at a time. For example, a user might input, "First, stop the machine, then unscrew that big cover and check if the belt inside is loose." The system breaks down this input into multiple raw step units and generates a step identifier for each unit.

[0035] Based on the aforementioned step elements, a standardized candidate step object is generated. The standardized candidate step object includes one or more of the following: candidate step identifier, original text fragment, standardized step text, action word, operation object, required tools, safety prerequisites, acceptance conditions, source basis identifier, step type, prerequisite step identifier, prohibition conditions, and manual confirmation status.

[0036] The standardized candidate step objects are matched with the component nodes, tool nodes, safety prerequisites, acceptance conditions, source basis identifiers, and step types in the data association network, respectively. The standardized candidate step objects are then subjected to source basis verification, measurability verification, operability verification, and step sequence constraint verification. The source basis verification includes determining whether the operation objects, tools, parameters, or acceptance conditions in the candidate step objects match at least one of the following: user original input, equipment manual, enterprise operation specifications, historical maintenance records, or terminology mapping rules. The step sequence constraint verification includes determining the prerequisite steps and prohibition conditions based on the step type node.

[0037] When a candidate step does not match the source basis, contains model completion items without source support and without manual confirmation, the operation object does not match the current equipment model, the required tools are missing, the safety prerequisites are not met, the acceptance conditions are missing, the prerequisite steps are missing, or the prohibition conditions are triggered, supplementary prompts are generated and automatic writing to the structured operation step record is prohibited.

[0038] When all the above verifications pass and there are no unconfirmed model completion items, the standardized candidate step object, source basis result, step order result and verification result are written into the structured operation step record, and a maintenance operation instruction document is generated based on the structured operation step record.

[0039] The semantic processing module includes one or more of a domain language model, a natural language processing model, or a rule engine; the standardized candidate step object output by the semantic processing module is the object to be verified. If there is content added by the semantic processing module in the standardized candidate step object that does not match the source basis, the content is marked as a model completion item; the standardized candidate step object containing unconfirmed model completion items shall not be automatically written into the structured operation step record or maintenance operation instruction file.

[0040] The steps include one or more of the following: action words, operation objects, operation tools, component names, directions, parameters, units, status words, acceptance conditions, time limits, safety prerequisites, step types, and abnormal handling requirements; the sources include one or more of the following: user original input, equipment manuals, enterprise operating procedures, historical maintenance records, terminology mapping rules, or manual confirmation information.

[0041] The steps for generating standardized candidate actions include: mapping colloquial actions to standard industrial operation actions, mapping colloquial component names to standard component names corresponding to the current equipment model, and completing safety prerequisites or acceptance conditions based on safety constraints.

[0042] Measurability verification includes: determining whether the standardized candidate steps contain quantitative indicators, observable states, or verifiable acceptance conditions; if quantitative indicators, observable states, or verifiable acceptance conditions are missing, a supplementary prompt requiring supplementary acceptance criteria is generated.

[0043] Operability verification includes: based on the data association network of equipment, tools, operating conditions and safety constraints, determining whether the current equipment model node is connected to the component node corresponding to the operation object of the standardized candidate step object through a valid relationship, whether the tool list contains tool nodes that have a valid relationship with the operation action of the standardized candidate step object, whether the safety prerequisite matches the safety condition node, and whether the current working conditions allow the execution of the standardized candidate step object.

[0044] The data association network includes equipment model nodes, component nodes, tool nodes, operation nodes, safety condition nodes, skill condition nodes, step type nodes, source basis nodes, and relationship edges representing the relationships between equipment and components, components and operations, operations and tools, operations and safety conditions, step types and prerequisites, and candidate steps and source basis.

[0045] Determining whether a standardized candidate step should be written into the structured work procedure record includes generating measurable results, operability results, relevance results, timeliness results, source basis results, step sequence results, prohibition condition results, and manual confirmation results. Among these, the source basis result indicates whether the standardized candidate step matches the source basis and whether the operation object, tool, parameter, or acceptance condition is consistent with the source basis; the step sequence result indicates whether the current step sequence is missing a prerequisite step; and the prohibition condition result indicates whether there are one or more prohibition conditions, such as incomplete energy isolation, unreset protective shield, unconfirmed cleaning agent residue, or missing specified tools from the tool list. The standardized candidate step will only be written into the structured work procedure record or maintenance operation instruction document if all of the above results meet the preset writing conditions.

[0046] When the verification fails, the system generates a failure handling record, which includes one or more of the following: failure reason, triggering rule, corresponding node identifier, corresponding relationship edge identifier, missing field, suggested supplementary item, whether manual review is allowed, review permission, whether re-verification is required after supplementation, and blocking reason. When the semantic processing module generates multiple standardized candidate step objects, the system determines the order of candidate steps based on the number of source criteria hits, the number of rule hits, the degree of data association network matching, the number of historical confirmations, the number of step order conflicts, and the number of risk flags, and prioritizes the standardized candidate step objects that are ranked higher and have not triggered blocking conditions.

[0047] like Figure 2As shown, in a preferred embodiment, the liquid packaging equipment maintenance operation instruction generation system based on semantic processing and step executable verification includes: a task context acquisition module for acquiring the liquid packaging equipment maintenance task context; a step input module for receiving user-inputted colloquial maintenance steps; a semantic processing module for parsing the colloquial maintenance steps using one or more of a domain language model, natural language processing model, or rule engine, along with a rule base, to obtain step elements and step types, and determine the source basis; a data association network for storing the relationships between equipment model nodes, component nodes, tool nodes, operation nodes, safety condition nodes, step type nodes, and source basis nodes; a step standardization module for generating standardized candidate step objects based on step elements. Specifically, the semantic processing module identifies actions / objects and step types, and the step standardization module standardizes candidate steps, standard actions / objects, and acceptance conditions; and a step verification module performs source basis verification, measurability verification, operability verification, and step sequence constraint verification on the standardized candidate step objects. If the verification fails, the feedback correction module generates supplementary prompts and receives supplementary information from the user. If the verification passes, the structured record generation module generates structured work step records, and the maintenance work instruction document is generated based on the structured work step records and output through the file generation interface.

[0048] In the minimum feasible system embodiment, the system includes at least a spoken step input interface, a terminology mapping table, a data association network, a standardized candidate step object table, and a write control module. The spoken step input interface receives on-site descriptions; the terminology mapping table converts spoken actions and component aliases into standard fields; the data association network stores the relationships between device models, components, tools, safety conditions, step types, and source bases; the standardized candidate step object table stores candidate texts and their verification status; and the write control module determines whether to write to the formal instruction manual based on source bases, measurability, operability, step order, and prohibition conditions. Even without using a large model, the core controlled write process of this invention can be implemented using only a rule base and a terminology mapping table.

[0049] like Figure 3 As shown, the semantic processing module and rule base parse the original step units to obtain step elements and step types, and determine the source basis. Step elements may include action words, operation objects, operation tools, component names, directions, parameters, units, status words, acceptance conditions, time limits, safety prerequisites, and abnormal handling requirements. Step types may include shutdown, isolation, disassembly, cleaning, inspection, adjustment, reassembly, reset, trial operation, and record confirmation. The source basis may come from the user's original input, equipment manual, enterprise operating specifications, historical maintenance records, or terminology mapping rules.

[0050] The semantic processing module may include a domain language model trained or fine-tuned on maintenance texts, equipment manuals, historical maintenance records, and expert-annotated data for liquid packaging equipment, or it may include a natural language processing model or a rule engine. The rule base may include terminology mapping rules, verb normalization rules, component alias tables, safety phrase completion rules, unit identification rules, and rules for disabling fuzzy words. For example, "stop" may be mapped to "implement shutdown procedures and perform energy isolation tag locking," "lid" may be mapped to "equipment top cover" under the current equipment model, and "look" may be mapped to "check and confirm."

[0051] The system generates standardized candidate steps based on the parsed step elements. Standardized candidate steps can include operation number, standard action, operation object, required tools, safety prerequisites, execution content, acceptance conditions, and exception handling prompts. For example, "stop the machine first" is transformed into "implement the shutdown procedure and perform energy isolation tag locking, confirming the equipment is in a zero-energy state"; "take a wrench to unscrew that large cover" is transformed into "use a special wrench to loosen counterclockwise and remove the equipment top cover."

[0052] like Figure 4 As shown, the system also performs operability checks, source basis checks, and step sequence constraint checks on standardized candidate steps. These checks are based on a data association network of equipment, tools, operating conditions, safety constraints, step types, and source basis. The data association network can include equipment model nodes, component nodes, tool nodes, operating nodes, safety condition nodes, skill condition nodes, step type nodes, source basis nodes, and relational edges. The system determines whether the current equipment model contains the object to be operated on, whether the tool list contains the required tools, whether the safety prerequisites are met, whether the source basis is correct, and whether the step type meets the preconditions and prohibition conditions.

[0053] For example, if a standardized candidate step requires "using a dedicated diagnostic tool to read servo parameters," but the current tools list does not contain such a tool, the system will generate a message stating "The dedicated diagnostic tool required to perform this step is not in the tools list." If a candidate step requires disassembling a valve assembly, but the valve assembly is not listed in the parts list for the current device model, the system will generate a message stating "The current device model does not match the operation object."

[0054] The system can further perform relevance and timeliness checks. Relevance checks determine if the steps are consistent with the maintenance task name, maintenance content, and equipment components; timeliness checks determine if the steps include specific operation times, waiting times, pressure holding times, or completion deadlines. If a check fails, the system generates a correction prompt and retains the original steps, candidate steps, and the reason for failure.

[0055] Once all checks pass, the system generates a structured work step record. This record can include step identifiers, original step text, standardized step text, action words, operation objects, required tools, quantifiable indicators, observable status, safety prerequisites, measurability check results, operability check results, relevance check results, timeliness check results, supplementary prompts, and manual confirmation status.

[0056] In one embodiment, the structured job step record also includes a source basis identifier, a source basis type, a source basis fragment, a step sequence number, a preceding step identifier, a following step identifier, a sequence constraint result, and a blocking reason. Through these records, the system can explain what inputs or rules support the candidate step, and can reconstruct why the step was written, rejected, or required for manual review during subsequent audits.

[0057] The system records multiple structured operation steps in the order of the operation and writes them into the maintenance operation instruction data structure. It then calls the file generation interface to generate the maintenance operation instruction file. The file generation interface can generate spreadsheet files, word processing files, or portable document files based on preset templates, and writes standardized steps, verification status, supplementary tips, and illustrated links into the corresponding areas.

[0058] This application primarily controls the generation, source verification, and executability verification of the transformation from colloquial maintenance steps to structured work steps. The integrity of the safety association between energy type, hazard source, and protective measures can be performed by the safety association verification module; even if the safety association verification module outputs safety items, whether the step text can be written into the formal instruction manual is still jointly determined by the standardized candidate step objects, data association network, source verification, measurability verification, operability verification, and step sequence constraint verification.

[0059] Reference Figure 5 The system performs measurability checks on standardized candidate steps. Measurability checks determine whether a step contains quantifiable metrics, observable states, or verifiable acceptance criteria. The system can match numbers, units, range values, torque values, pressure values, temperature values, time values, and status words using regular expressions, or it can use a rule base to determine whether observable states such as "qualified," "green," "no leakage," "no abnormal noise," and "tension meets standard requirements" exist.

[0060] When a step lacks measurable information, the system generates supplementary prompts. For example, for "Check the tension of the internal drive belt," if no acceptance criteria are included, the system prompts, "Please supplement the tension criteria, such as specifying the tension range or confirming that it conforms to the equipment manual standards." After the user supplements the information, the system regenerates the standardized candidate steps and performs the verification again.

[0061] Back Figure 2In one specific embodiment, the user inputs the maintenance steps for the "TSC-235 filling and capping machine maintenance": "First, stop the machine, use a wrench to unscrew the large cap, and check if the internal belt is loose." The system identifies the equipment model TSC-235 and the dedicated wrench kit in the tool list based on the task context. The system standardizes the first step as "Implement the shutdown procedure and perform energy isolation tag locking, confirming the equipment is in a zero-energy state," and prompts the user to confirm the main power is disconnected, the tag is locked, and the residual energy is released. The system standardizes the second step as "Use a dedicated wrench to loosen and remove the equipment top cover counterclockwise," and determines the tool is available based on the tool list. The system standardizes the third step as "Check the tension of the internal drive belt," and prompts the user to supplement the tension standard or equipment manual acceptance conditions.

[0062] After the user adds "the tension meets the range specified in the equipment manual", the system re-executes the measurability and operability verification. After the verification passes, a structured operation step record is generated and the step is written into the final maintenance operation manual.

[0063] In one embodiment, the semantic processing module can be implemented in different ways. The first implementation is a domain language model, which can be trained or fine-tuned based on maintenance texts for liquid packaging equipment, equipment manuals, historical maintenance records, and expert-annotated samples. The second implementation is a natural language processing model, which can perform word segmentation, entity recognition, action recognition, dependency parsing, or text similarity calculation. The third implementation is a rule engine, which can transform colloquialization steps based on a terminology mapping table, a verb normalization table, a component alias table, and rules for disabling fuzzy words. These implementations can be used individually or in combination, with the system ultimately using successful verification and manual confirmation as the conditions for writing.

[0064] In a data structure example, the original step unit may include the original step identifier, original text, input order, and inputter identifier fields; the step element object may include the action word, operation object, operation tool, component name, direction, parameter value, unit, status word, acceptance condition, time limit, safety prerequisite, and exception handling requirements fields; the standardized candidate step may include the candidate step identifier, standard action, standard object, required tool, safety prerequisite, operation content, acceptance condition, precautions, and candidate source fields; the structured operation step record may include the step identifier, original text, candidate text, final text, measurable result, operable result, relevant result, time-limited result, feedback record, and manual confirmation status field.

[0065] Measurability verification can be implemented using a combination of regular expressions, unit dictionaries, status dictionaries, and acceptance condition rules. The system can recognize units such as Newtons, Newton-meters, millimeters, degrees Celsius, megapascals, minutes, and seconds, as well as observable states such as "no leakage," "no abnormal noise," "indicator light is green," and "tension meets the equipment manual range." When a candidate step only contains vague expressions such as "check," "take a look," or "handled," without specifying the object to be checked, the quantification range, or the acceptance status, the system will generate supplementary prompts and retain the reason for failure. After the user adds acceptance conditions, the system regenerates candidate steps and performs verification again.

[0066] Operability verification can be implemented based on a data association network. This network can include nodes for equipment model, components, tools, operations, safety conditions, and skill conditions. The system determines whether the operation object in a candidate step belongs to the component list of the current equipment model, whether the required tool exists in the tool list, whether the safety prerequisites meet LOTO or shutdown requirements, and whether the operation conditions conflict with the current work area, cleaning status, or shutdown status. If a candidate step requires disassembling a valve assembly that is not in the equipment register, the system generates an object mismatch warning; if a candidate step requires using a dedicated diagnostic tool that is not included in the tool list, the system generates a tool missing warning.

[0067] Step sequence constraint verification can be implemented based on step type and preconditions / preconditions. The system categorizes standardized candidate step objects into step types such as shutdown, isolation, disassembly, cleaning, inspection, adjustment, reassembly, reset, trial operation, and record confirmation, and configures preconditions and prohibition conditions for different step types. For example, the preconditions for the disassembly step may include shutdown and confirmed energy isolation; the preconditions for the trial operation step may include reassembly completion, protective cover reset, and personnel evacuation. Prohibition conditions may include prohibiting the writing of disassembly, cleaning, or entry into the equipment interior steps if energy isolation is not completed; prohibiting the writing of trial operation steps if the protective cover is not reset; prohibiting the writing of steps to open pipelines or disassemble nozzles if the residual state of cleaning agent is not confirmed; and prohibiting the writing of steps requiring a specific tool if the tool list is missing. If a candidate step object triggers a prohibition condition, the system generates a step sequence conflict prompt and blocks the writing.

[0068] In the exception handling embodiment, when the semantic processing module outputs multiple candidate steps, the system can sort the candidate steps according to the number of rule base hits, the degree of matching of data association network, and historical confirmation records. When a candidate step conflicts with a security constraint, the system does not write it into the final guidance document, but generates a security review prompt. When the user's supplementary information still cannot make the candidate step pass the verification, the system retains the original step, the candidate step, the reason for failure, and the user's supplementary record, and marks the step as a manual review status. When the manual confirmation status meets the preset writing conditions, the system writes the final confirmation text into the structured job step record.

[0069] In another specific embodiment, a user inputs "After stopping the machine, disassemble the capping head assembly and check if it's loose" for the TSC-235 filling and capping machine. The system first breaks down this input into three basic steps: stopping the machine, disassembling the capping head assembly, and checking the internal connection status. The semantic processing module maps "stopping the machine" to "implementing the shutdown procedure and performing energy isolation tag locking to confirm the equipment is in a zero-energy state"; maps "capping head assembly" to "capping head assembly" under the current equipment model; and maps "whether it's loose" to "check the fastener connection status". The system then prompts the user to supplement acceptance criteria, such as the tightening torque range, whether looseness is allowed, or whether the equipment manual standard should be followed. After the user adds "the tightening status meets the torque range specified in the equipment manual", the system generates standardized candidate steps and verifies them for measurability and operability.

[0070] In another embodiment, the system can adjust the feedback method according to different user roles. For ordinary maintenance personnel, the system outputs supplementary prompts related to the operation, such as "Please supplement the inspection object and acceptance criteria"; for safety supervisors, the system outputs safety prerequisites and risk items, such as "This step involves pneumatic residual pressure release, please confirm LOTO and residual pressure release status"; for document administrators, the system outputs terminology mappings and template fields, such as "This step will be written into the work step field and acceptance criteria field." The prompts for different roles do not directly change the final instruction manual; the final writing is still based on the verification results and manual confirmation status.

[0071] The technical effects of this invention can be verified through step logs and generation records. For example, the system can count the number of conversational steps broken down into original step units, the initial pass rate of candidate steps, the number of times supplementary prompts are triggered due to a lack of acceptance criteria, the number of steps blocked due to incompatible tool or equipment models, and the number of steps written into the instruction manual after manual confirmation. Through the above records, it can be demonstrated that the system does not merely polish the text, but rather computerizes the measurability, operability, relevance, and timeliness of the steps before generation.

[0072] In the semantic processing flow embodiment, the system can first perform step boundary identification on the user input, and then perform action identification, object identification, tool identification, and condition identification on each step unit. Action identification is used to map spoken actions such as "fiddle with it," "take a look," and "disassemble" to standard actions such as inspection, disassembly, installation, cleaning, adjustment, or reset; object identification is used to determine the name of standard parts by combining the equipment model and part alias table; tool identification is used to determine whether the step includes a wrench, diagnostic instrument, measuring tool, or cleaning tool; condition identification is used to extract shutdown, power failure, pressure release, waiting time, and acceptance status. The above identification results together form a step element object, which is used for subsequent candidate step generation and verification.

[0073] In the model output constraint implementation, if the candidate step generated by the semantic processing module contains a component name that does not exist in the equipment ledger, the system marks the candidate step as an object mismatch; if the candidate step contains a tool that does not exist in the tool list, the system marks it as a missing tool; if the candidate step omits LOTO, safety protection, or acceptance conditions, the system marks it as insufficient or unmeasurable safety prerequisites. The marked candidate steps are not directly written into the final guidance document but instead enter the feedback correction module. The feedback correction module generates targeted questions based on the reasons for failure, allowing the user to supplement equipment objects, tools, quantification standards, or manual confirmation information.

[0074] In the source basis binding embodiment, the system records at least one source basis for each standardized candidate step object. Source basis can come from step fragments in the user's original input, component descriptions in equipment manuals, safety clauses in enterprise operating procedures, confirmation steps in historical maintenance records, or terminology mapping rules in a rule base. The system performs source basis matching on the operation objects, tools, parameters, and acceptance conditions in the candidate step objects; if a certain content is added by the semantic processing module but does not match a source basis, the system marks that content as a model completion item. Candidate step objects containing unconfirmed model completion items cannot be automatically written into the structured operation step record and require user-added basis or confirmation by personnel with review permissions.

[0075] In the failure feedback embodiment, when the source basis verification, measurability verification, operability verification, or step sequence constraint verification fails, the system generates a failure handling record. The failure handling record may include the failure reason, triggering rule, corresponding node identifier, corresponding relationship edge identifier, missing fields, suggested additions, whether manual review is allowed, review role, whether re-verification is required after addition, and the reason for blocking. After the user adds acceptance conditions, tool information, source basis, or manual confirmation information, the system regenerates standardized candidate step objects and re-executes all verifications.

[0076] The failure closure loop can be executed in the following order: record the reason for failure, locate the triggering rule, write the involved nodes, relationship edges or fields, execute the action to prohibit automatic writing, generate supplementary prompts, receive user supplements or authorized personnel for review, and regenerate standardized candidate step objects after supplementation. If, after re-verification, there are still contents without source basis, inoperable objects, missing acceptance conditions or prohibition conditions, the system remains in a pending review state and the candidate step is not written into the formal guidance document.

[0077] In a counterexample embodiment, if the user inputs "Open the pressure cap to check if the machine is not stopped," the system recognizes the disassembly step, but the step sequence constraint check finds that the shutdown and energy isolation prerequisite steps are missing. Therefore, a blocking reason is generated and automatic writing is prohibited. If the user inputs "Read servo parameters with a dedicated diagnostic tool," but the tool list does not include a dedicated diagnostic tool, the system marks the candidate step object as a missing tool. If the semantic processing module expands "Check for leaks" to "Hold pressure at 0.3 MPa for 5 minutes without leaks," but the equipment manual, enterprise specifications, historical records, and user input do not support this parameter, the system marks the parameter as a model completion item and requires supplementary source evidence or manual confirmation.

[0078] In the template mapping embodiment, structured work step records can be mapped to different fields of the maintenance work instruction template. For example, standard actions are written to the "Work Action" field, standard objects to the "Operation Object" field, required tools to the "Required Tools" field, safety prerequisites to the "Safety Prerequisites" field, acceptance conditions to the "Acceptance Criteria" field, precautions to the "Precautions" field, and manual confirmation status to the "Confirmation Status" field. The document generation interface checks whether required fields are empty before writing; if a work action, operation object, or acceptance criterion is empty, it refuses to generate the official document and returns a supplementary prompt.

[0079] In the tabular record implementation, the system can save spoken input, semantic processing results, candidate steps, verification results, and final write status as a structured record. Table 1 shows an example of step element fields.

[0080] Table 1 Example of step element fields

[0081] Table 2 shows an example of a standardized candidate step record. This table illustrates how the output of a model, natural language processing model, or rule engine is constrained to a standardized candidate step.

[0082] Table 2. Examples of Standardized Candidate Step Records

[0083] Table 3 shows an example of a step verification record. The system can use this record to decide whether to write it into the structured job step record or generate supplementary prompts.

[0084] Table 3 Example of Step Verification Record

[0085] In the write condition determination embodiment, the system can generate measurability results, operability results, relevance results, time-limited results, and manual confirmation results for standardized candidate steps, respectively.

[0086] When the measurability, operability, relevance, timeliness, and manual verification results are all passed, the system determines that the candidate step meets the basic writing conditions. Only candidate steps that meet the basic writing conditions are allowed to enter the writing process of structured work step records or maintenance work instructions.

[0087] In the enhanced write condition determination embodiment, the write condition may further include source basis approval and step sequence approval. Source basis approval means that the candidate step at least matches one of the following: user input, equipment manual, enterprise operating specifications, historical maintenance records, terminology mapping rules, or manual confirmation information, and the operation object, tool, parameter, or acceptance condition in the candidate step does not conflict with the matched source basis; step sequence approval means that the candidate step does not have any preset conflicts with its predecessor and successor steps. If any result is not met, the system will retain the candidate step as pending supplementation or pending manual review.

[0088] In the candidate step ranking embodiment, the system can rank multiple candidate steps based on the number of rule hits, the number of source basis hits, the degree of data association network matching, the number of historical confirmations, the number of step order conflicts, and risk indicators. During ranking, the system can increase the priority of candidate steps with clear source basis, high network matching degree, and no order conflicts, while decreasing the priority of candidate steps with risk indicators or lacking source basis. This relationship describes the priority display method for candidate steps; in actual implementation, different weights can be set, or rule priority or manual confirmation priority can be used.

[0089] In another specific embodiment, a user inputs the following for the changeover cleaning operation of the TSC-235 filling and capping machine: "Remove the feed pipe and rinse it, then reinstall it and check for leaks." The system breaks down this input into four basic step units: disassembling the feed pipe, rinsing the feed pipe, installing the feed pipe, and checking for leaks. The semantic processing module maps "rinse it" to "rinse the inner wall of the feed pipe using the specified cleaning medium," and "check for leaks" to "check for leaks at the feed pipe connection during trial operation."

[0090] In the above-described replacement cleaning embodiment, when the system performs measurability verification on the candidate steps, it finds that the "rinsing" step lacks rinsing time or cleaning status, and generates a prompt "Please add rinsing time or cleaning completion standard". When the system performs operability verification on the "reinstall" step, it checks whether the tool list contains the corresponding quick-connect tool and determines whether the current equipment is in a stopped or LOTO state. After the user adds "rinse for no less than 3 minutes until the drain is clear", the system regenerates the candidate steps and passes the verification.

[0091] The structured work procedure record ultimately generated by the system can include four steps: disassembling the feed pipe, rinsing the feed pipe, reassembling the feed pipe, and trial run leak check. Each step includes the object of operation, required tools, safety prerequisites, acceptance conditions, and manual confirmation status. If any step lacks acceptance conditions or safety prerequisites, the document generation interface will refuse to generate a formal instruction manual and will leave that step in a pending status.

Claims

1. A method for generating maintenance operation instructions for liquid packaging equipment based on semantic processing and step-executability verification, characterized in that, include: Obtain the context of the liquid packaging equipment maintenance task; Receive user-inputted, conversational maintenance steps; The colloquial maintenance steps are analyzed using a semantic processing module and a rule base to obtain step elements and step types, and to determine the source basis. Generate standardized candidate step objects based on the aforementioned step elements; The operation objects, required tools, safety prerequisites, acceptance conditions, source basis identifiers, and step types in the standardized candidate step objects are matched with the component nodes, tool nodes, safety condition nodes, source basis nodes, and step type nodes in the data association network, respectively. The standardized candidate steps are subjected to source basis verification, measurability verification, operability verification, and step sequence constraint verification. When a candidate step object does not match the source basis, contains model completion items without source support and not manually confirmed, the operation object does not match the current equipment model, the required tools are missing, the safety prerequisites are not met, the acceptance conditions are missing, the prerequisite steps are missing, or the prohibition conditions are triggered, a supplementary prompt is generated and automatic writing to the structured operation step record is prohibited. When all the above verifications pass and there are no unconfirmed model completion items, the standardized candidate step object, source basis result, step order result and verification result are written into the structured operation step record, and a maintenance operation instruction document is generated based on the structured operation step record.

2. The method according to claim 1, characterized in that, The maintenance task context includes one or more of the following: equipment model, maintenance content, required tools, safety constraints, and pre-operation preparation requirements. The semantic processing module includes one or more of the following: domain language model, natural language processing model, or rule engine. The standardized candidate step object includes one or more of the following: candidate step identifier, original text fragment, standardized step text, action word, operation object, required tools, safety prerequisites, acceptance conditions, source basis identifier, step type, prerequisite step identifier, prohibition conditions, and manual confirmation status. Source basis verification includes determining whether the operation object, tool, parameter, or acceptance condition in the candidate step object matches at least one of the following: user's original input, equipment manual, enterprise operation specifications, historical maintenance records, or terminology mapping rules. Step sequence constraint verification includes determining the prerequisite steps and prohibition conditions based on the step type node. The standardized candidate step object output by the semantic processing module is the object to be verified. If there is content added by the semantic processing module that does not match the source basis in the standardized candidate step object, then the content is marked as a model completion item. Standardized candidate step objects containing unconfirmed model completion items shall not be automatically written into the structured operation step record or maintenance operation instruction file.

3. The method according to claim 1, characterized in that, The steps include one or more of the following: action words, operation objects, operation tools, component names, directions, parameters, units, status words, acceptance conditions, time limits, safety prerequisites, step types, and abnormal handling requirements; the sources include one or more of the following: user original input, equipment manuals, enterprise operating procedures, historical maintenance records, terminology mapping rules, or manual confirmation information.

4. The method according to claim 1, characterized in that, The steps for generating standardized candidate actions include: mapping colloquial actions to standard industrial operation actions, mapping colloquial component names to standard component names corresponding to the current equipment model, and completing safety prerequisites or acceptance conditions based on safety constraints.

5. The method according to claim 1, characterized in that, Measurability verification includes: determining whether the standardized candidate steps contain quantitative indicators, observable states, or verifiable acceptance conditions; if quantitative indicators, observable states, or verifiable acceptance conditions are missing, a supplementary prompt requiring supplementary acceptance criteria is generated.

6. The method according to claim 1, characterized in that, Operability verification includes: based on the data association network of equipment, tools, operating conditions and safety constraints, determining whether the current equipment model node is connected to the component node corresponding to the operation object of the standardized candidate step object through a valid relationship, whether the tool list contains tool nodes that have a valid relationship with the operation action of the standardized candidate step object, whether the safety prerequisite matches the safety condition node, and whether the current working conditions allow the execution of the standardized candidate step object.

7. The method according to claim 6, characterized in that, The data association network includes equipment model nodes, component nodes, tool nodes, operation nodes, safety condition nodes, skill condition nodes, step type nodes, source basis nodes, and relationship edges representing the relationships between equipment and components, components and operations, operations and tools, operations and safety conditions, step types and prerequisites, and candidate steps and source basis.

8. The method according to claim 1, characterized in that, Determining whether a standardized candidate step should be written into the structured work procedure record includes generating measurable results, operability results, relevance results, timeliness results, source basis results, step sequence results, prohibition condition results, and manual confirmation results. Among these, the source basis result indicates whether the standardized candidate step matches the source basis and whether the operation object, tool, parameter, or acceptance condition is consistent with the source basis; the step sequence result indicates whether the current step sequence is missing a prerequisite step; and the prohibition condition result indicates whether there are one or more prohibition conditions, such as incomplete energy isolation, unreset protective shield, unconfirmed cleaning agent residue, or missing specified tools from the tool list. The standardized candidate step will only be written into the structured work procedure record or maintenance operation instruction document if all of the above results meet the preset writing conditions.

9. The method according to claim 1, characterized in that, When the verification fails, the system generates a failure handling record, which includes one or more of the following: failure reason, triggering rule, corresponding node identifier, corresponding relationship edge identifier, missing field, suggested supplementary item, whether manual review is allowed, review permission, whether re-verification is required after supplementation, and blocking reason. When the semantic processing module generates multiple standardized candidate step objects, the system determines the order of candidate steps based on the number of source criteria hits, the number of rule hits, the degree of data association network matching, the number of historical confirmations, the number of step order conflicts, and the number of risk flags, and prioritizes the standardized candidate step objects that are ranked higher and have not triggered blocking conditions.

10. A system for generating maintenance operation instructions for liquid packaging equipment based on semantic processing and step-executable verification, characterized in that, include: The task context acquisition module is used to acquire the task context of liquid packaging equipment maintenance. The step input module is used to receive user-inputted, conversational maintenance steps; The semantic processing module is used to parse colloquial maintenance steps using one or more of the following: a domain language model, a natural language processing model, or a rule engine, along with a rule base, to obtain step elements and step types, and to determine the source basis. The data association network is used to store the relationships between device model nodes, component nodes, tool nodes, operation nodes, security condition nodes, step type nodes, and source basis nodes. The step standardization module is used to generate standardized candidate step objects based on step elements. The step verification module is used to perform source basis verification, measurability verification, operability verification, and step sequence constraint verification on standardized candidate step objects. The feedback and correction module is used to generate supplementary prompts and receive supplementary information from users when the verification fails. The structured record generation module is used to generate structured job step records when the verification passes. The file generation interface is used to generate maintenance operation instruction files based on structured operation step records.

11. An electronic device, characterized in that, It includes a processor and a memory, wherein the memory stores a computer program that, when executed by the processor, implements the method according to any one of claims 1 to 9.

12. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, implements the method according to any one of claims 1 to 9.

Images