A tire building machine sop retrieval enhancement generation question and answer method based on finite state machine constraint

Abstract

The application discloses a tire building machine SOP retrieval enhancement generation question and answer method based on finite state machine constraints. The application pre-constructs a finite state machine model of the SOP flow, determines the standard step boundary, execution sequence and safety condition before answer generation, and checks the legality, integrity, sequence legality and safety constraint satisfaction of the steps after answer generation, effectively solving the problems of step omission, unstable sequence and insufficient safety confirmation in the ordinary RAG in the industrial SOP flow question and answer.

Description

Technical Field

[0001] This invention relates to the field of industrial equipment operation and maintenance and intelligent question answering technology, specifically to a method for enhancing the generation of question answers based on SOP retrieval for tire forming machines using finite state machine constraints. Background Technology

[0002] In the on-site operation and maintenance of tire forming machines, operations such as equipment start-up, shutdown, air supply, calibration, adjustment, repair and replacement, and reset checks typically rely on equipment manuals, maintenance data, and Standard Operating Procedures (SOPs). Large Language Models (LLMs) and Retrieval-Augmented Generation (RAGs) offer new approaches to knowledge-based question answering for industrial equipment. By introducing external knowledge retrieval before generation, RAGs can use equipment manuals, maintenance data, and operation document fragments as contextual input models, thereby improving the consistency between the answer and the original data. They are currently used for tasks such as question answering in the industrial field. However, answers to SOP process-related questions not only need to be relevant but also need to conform to the step boundaries, execution order, and safety constraints of the original SOP. Ordinary RAGs are easily affected by the scope of the retrieval fragments and the model organization, leading to problems such as missing steps, unstable order, or insufficient safety confirmation. Therefore, it is necessary to introduce explicit process constraints and compliance verification mechanisms during the question-answering process to improve the completeness, sequential consistency, and safety verifiability of the process answers.

[0003] Existing RAG improvements primarily enhance response quality through retrieval quality, generative reflection, retrieval error correction, long document organization, and graph-structured knowledge organization. However, these methods mainly target knowledge retrieval, factual consistency, or complex corpus organization, and cannot directly guarantee the compliance of industrial SOP process answers. For SOP process question answering, textual evidence alone does not equate to process compliance. Ordinary RAGs focus more on the semantic relevance of the retrieved fragments to the question. When the retrieved content only covers partial steps or multiple fragments are out of order, the model may only answer significant operations such as disassembly and installation, while omitting necessary steps such as shutdown, power outage, reset, and inspection. Even with prompts requiring the model to "answer step by step" and "pay attention to safety," it is difficult to determine whether all necessary steps are covered, whether the step order conforms to the original SOP, and whether safety conditions such as shutdown and power outage are explicitly preserved. Structured output can improve the parsability of the answer format, but a fixed template itself cannot determine whether a certain step belongs to the current SOP, nor can it detect operations outside the process or missing key steps.

[0004] Industrial Standard Operating Procedures (SOPs) have clearly defined process attributes. Tasks such as equipment start-up and shutdown, maintenance and replacement, calibration and alignment, and mechanism adjustment are usually organized in a fixed sequence, and status confirmation, safety protection, and reset check requirements are given at key stages. Finite State Machines (FSMs) are suitable for describing discrete states and their transition relationships. They can abstract key operations in SOPs into state nodes, represent the sequential dependencies of steps as state transitions, and record requirements such as shutdown, power failure, pressure relief, protection reset, and trial operation checks as safety constraints.

[0005] To address the aforementioned issues, a method is needed that integrates process model constraints and compliance verification into the enhanced retrieval generation process. This method involves pre-constructing a finite state machine model of the Standard Operating Procedure (SOP) process to determine the boundaries, execution order, and safety conditions of standard steps before answer generation. After answer generation, the legality, completeness, sequential legality, and satisfaction of safety constraints of the steps are verified. If the verification fails, controlled regeneration or fallback processing is performed. When the operation object is unclear, a clarifying answer is output. When the retrieval evidence is insufficient or there is a high risk tendency, a rejection statement is output. This ensures the completeness, sequential consistency, and safety checkability of the steps in the answer to the industrial SOP process. Summary of the Invention

[0006] To address the technical problems mentioned above, this invention provides a method for enhancing the generation of question-and-answer functions for tire forming machine SOP retrieval based on finite state machine constraints, comprising the following steps: S1. Based on the industry standard operating procedures (SOPs), obtain the FSM process library and problem triage results; S2. Based on the intent index of the process library and user questions, obtain a reliable SOP match; S3. Based on reliable SOP matching and knowledge base evidence, obtain the constrained generated answer and verification result; S4. Based on the compliance verification results and the fallback mechanism, obtain the final controlled process answer.

[0007] Preferably, S1 includes: organizing SOPs into an external SOP-FSM process library, the SOP-FSM process library consisting of two parts: SOP intent index and FSM process constraints; when handling problems, identifying the problem type and making a risk judgment, non-process problems enter the search enhancement generation question and answer channel, problems with unclear objects or non-unique candidates output clarification answers, and high-risk problems are directly rejected.

[0008] Preferably, in the SOP-FSM process library, the intent index records the process name, category, keywords, aliases, operation objects, and step summaries for matching issues to SOPs; The FSM process constraint record includes state nodes, state transitions, standard step sequences, required step sets, and security constraint sets.

[0009] Preferably, in the SOP-FSM process library, the first i The FSM process instance corresponding to each SOP is defined as follows: in, Si A set of state nodes; Ei It is a set of state transitions; and These represent the start and end states. Li This is a standard sequence of steps; Pi A set of necessary steps; Ci The set of safety constraints includes recording shutdowns, power outages, pressure relief, status confirmations, protection resets, and trial operation checks; This allows natural language SOPs to be organized into structured process data.

[0010] Preferably, S2 includes: for process-type questions, performing candidate matching based on the SOP intent index, and the matching score consists of three parts: semantic similarity, keyword hit and rule correction; When the highest matching score reaches the matching threshold, and the difference between the highest matching score and the second highest matching score reaches a set value, the candidate with the highest score is determined to be a reliable match. When the difference between candidate SOPs does not reach the set value, or when the operation object and limiting information are insufficient, a clarification response is output, requiring supplementary operation object or limiting information; When no reliable SOP match is found and no further clarification is required, output a reference process answer or refuse to answer based on the sufficiency of evidence retrieved from the knowledge base.

[0011] Preferably, the semantic similarity in the matching score is used to measure the semantic similarity between the question and the candidate SOP retrieval text, the keyword hit is used to measure the hit score of the operation object, action word and limiting word, and the rule correction is used to correct the rules for similar parts, similar actions and insufficient object limitation.

[0012] Preferably, S3 includes: after obtaining a reliable matching SOP-FSM process instance, reading the standard step sequence, the set of required steps, and the set of safety constraints to form a process skeleton that limits the scope, sequence, and safety conditions of the steps; By retrieving relevant instruction manual fragments, maintenance instructions, and operation descriptions from the knowledge base, user questions, matched SOP information, FSM step skeleton, retrieval evidence, and structured output requirements are input into the large language model to guide the model to generate process answers step by step. The structured answer is mapped to a state sequence, compared with the standard step sequence, the set of required steps, and the set of safety constraints, and a hard constraint check is performed; the answer that passes the check is output as the FSM-SOP process answer.

[0013] Preferably, the hard constraint check includes illegal step check, repeated step check, required step coverage check, sequence legality check, and safety constraint check; The illegal step check is used to determine whether there are states in the generated state sequence that do not belong to the current SOP-FSM; The repeated step check is used to determine whether the same state is repeatedly output; The required step coverage check is used to determine whether all the key states in the required step set have appeared; the sequence validity check is used to determine whether the relative order of states conforms to the standard step sequence. The safety constraint check is used to determine whether the requirements for shutdown, power outage, pressure relief, status confirmation, protection reset, and trial operation check are fully retained.

[0014] Preferably, S4 includes: If the initial structured generation result fails the validation, the reason for the failure will be fed back to the large language model, and a controlled regeneration will be performed within the original SOP-FSM process instance and the original retrieval evidence. If the regenerated result still fails the verification, the FSM fallback mechanism is triggered to generate a deterministic process answer based on the standard step sequence, state description and security constraints; Based on the question triage results, SOP matching results, knowledge base retrieval evidence, and FSM compliance verification results, the final output type is determined. The final output type includes general Q&A, reference process Q&A, FSM-SOP process answer, clarification answer, and rejection answer. Specifically, the process answer will be output as the FSM-SOP process answer only if the generated process answer passes the backward compliance check of the corresponding SOP-FSM process instance.

[0015] Compared with the prior art, the beneficial effects of the present invention are as follows: This invention pre-constructs a finite state machine model of the SOP process, determines the standard step boundaries, execution order, and safety conditions before the answer is generated, and verifies the legality, completeness, order legality, and safety constraint satisfaction of the steps after the answer is generated. This effectively solves the problems of missing steps, unstable order, and insufficient safety confirmation that are prone to occur in ordinary RAG in industrial SOP process Q&A. Attached Figure Description

[0016] To more clearly illustrate the technical solution of the present invention, the drawings used in the embodiments are briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0017] Figure 1 This is a schematic diagram of the overall process of the industrial SOP process question and answer method based on finite state machines according to the present invention. Detailed Implementation

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

[0019] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0020] Example 1 In this embodiment of the invention, an SOP-FSM process library is constructed outside the existing AnalyticDB-based retrieval-enhanced question-answering system (hereinafter referred to as ADB-RAG). Let the user question be q, the AnalyticDB knowledge base be D, and the external SOP-FSM process library be F. D stores device manuals, maintenance data, and operation document fragments to provide domain knowledge evidence; F is derived from a structured, organized industrial SOP and is used to describe process steps, state transitions, necessary steps, and safety constraints. The overall system flow is as follows: Figure 1 As shown, the method includes the following steps: S1. Based on the industrial SOP, obtain the FSM process library and problem triage results.

[0021] The Standard Operating Procedures (SOPs) related to tire forming machines have been compiled into an external SOP-FSM process library, covering process types such as equipment start-up and shutdown, ventilation, alarm handling, calibration, maintenance and replacement, accuracy alignment, and mechanism adjustment. This library consists of two parts: an SOP intent index and FSM process constraints. The intent index records the process name, category, keywords, aliases, operation objects, and step summaries for matching problems to SOPs. The FSM process constraints record state nodes, state transitions, standard step sequences, required step sets, and safety constraint sets for forward planning and backward compliance verification. Taking the equipment ventilation SOP as an example, pre-ventilation status confirmation, opening the air source, phased pressure adjustment, operating pressure check, and reset confirmation are mapped to state nodes, and status confirmation and pressure check are included in the safety constraint set. i The FSM process instance corresponding to each SOP is defined as follows: in, Si For a set of state nodes, Ei For the set of state transitions, and For the start and end states, Li This is a standard sequence of steps. Pi For the set of necessary steps, Ci This is a set of safety constraints to record requirements such as shutdown, power outage, pressure relief, status confirmation, protection reset, and trial operation checks. Therefore, the natural language SOPs are organized into structured process data.

[0022] When handling issues, the system identifies the issue type and makes a risk assessment: non-process issues enter the search-enhanced question-and-answer generation channel. In this embodiment, the knowledge base is constructed using AnalyticDB, hereinafter referred to as the ADB-RAG channel; issues with unclear objects or multiple unique candidates output clarification answers; high-risk issues are directly rejected.

[0023] S2. Based on the intent index of the process library and user questions, obtain a reliable SOP match.

[0024] For process-oriented questions, the system performs candidate matching based on the SOP intent index. Let the user question be q, and the candidate SOPs be... s i The matching score consists of three parts: semantic similarity, keyword hit rate, and rule correction. in, S sem ( q , s i The semantic similarity between the question and the candidate SOP search text is used. S key ( q, s i () is the score for the hit on the object of operation, action word, and qualifier; S rule ( q , s i (This is to correct the score for rules that address similar components, similar actions, and insufficient object definition;) α , β , γ Let be the weighting coefficients. Let the highest matching score and the second highest matching score be respectively... Score 1 and Score 2, when Score 1. When the matching threshold is reached and the difference between the two reaches a set value, it is considered a reliable match; otherwise, a clarifying response is output, requesting additional information about the target operation or limiting information. If no reliable match is obtained, an indicative process response or a rejection is determined based on evidence retrieved from AnalyticDB.

[0025] S3. Based on reliable SOP matching and knowledge base evidence, obtain the constrained generated answer and verification result.

[0026] Once a reliable matching SOP-FSM process instance is obtained, the system reads the standard step sequence, the set of required steps, and the set of safety constraints to form a process skeleton that defines the scope, sequence, and safety conditions of the steps. Utilizing the AnalyticDB knowledge base, the system constructs search expressions by comprehensively using user questions, SOP titles, target objects, action words, and step summaries to retrieve relevant instruction manual fragments, maintenance instructions, and operation descriptions, ensuring that the retrieved results closely match the current process object.

[0027] The user's question, the matched SOP information, the FSM step skeleton, RAG retrieval evidence, and the structured output requirements are all input into the LLM to guide the model to generate the process answer step by step. The step order must be consistent with the FSM standard sequence. No steps other than the current SOP should be added, and no necessary steps should be omitted. Steps involving safety constraints should retain corresponding prompts.

[0028] After LLM output, the structured answer is mapped to a sequence of states. , idk For the first k The FSM status identifier for each step. G With standard procedure sequence Li Required Steps Set Pi and safety constraint set Ci The comparison involves five types of hard constraint checks: the first is illegal step checks, which determine... G The checks include: 1) whether there are states that do not belong to the current SOP-FSM; 2) duplicate step checks to determine if the same state is repeatedly output; and 3) mandatory step coverage checks to determine... Pi The fourth step is to check the order validity of the states, determining whether their relative order is correct. Li Fifth, safety constraint checks are conducted to determine whether requirements for shutdown, power outage, pressure relief, status confirmation, protection reset, and trial operation checks are fully retained. Illegal and repetitive steps are subject to a zero-tolerance rule; the coverage rate and sequence legality rate of necessary steps must reach 1; and the safety constraint satisfaction rate must reach a threshold. τs =0.8. If any condition is not met, the generated answer must not be output as the answer in the FSM-SOP process.

[0029] S4. Based on the compliance verification results and the fallback mechanism, obtain the final controlled process answer.

[0030] Answers that pass S3 validation are directly output as FSM-SOP process answers. If the initial structured generation result fails validation, the system reports the reason for the failure to the LLM, and performs a controlled regeneration within the original SOP-FSM process instance and the original retrieved evidence. If the regeneration result still fails validation, the FSM fallback mechanism is triggered, generating a deterministic process answer based on the standard step sequence, state description, and security constraints. This fallback answer must pass the backward compliance validation described in S3 again; if it fails, the system outputs a clarification or rejection statement. Through the above process, the system ultimately generates five types of outputs: general questions and answers, reference process questions and answers, FSM-SOP process answers, clarification answers, and rejection answers.

[0031] Example 2 To verify the superiority of this invention compared to existing technologies, this embodiment is provided for illustration. This embodiment constructs an experimental dataset based on the tire forming machine's instruction manual, maintenance data, and operating procedures. The SOP-FSM process library contains 55 SOPs, each organized into state nodes, state transitions, standard step sequences, required step sets, and safety constraints.

[0032] A total of 175 test questions were conducted, of which 115 were process questions covered by the process library, and the remainder were used to evaluate the system's processing capabilities under reference processes, general knowledge, fuzzy input, and high-risk requests. The questions were presented in a direct question format, paraphrased, and colloquial style, and underwent deduplication and manual review. The specific distribution is shown in Table 1.

[0033] Table 1 .

[0034] This embodiment compares Baseline RAG, RAG+Prompt, RAG-Struct, and FB-FSM-RAG, and sets up an ablation method that removes the FSM fallback mechanism. Baseline RAG does not use process constraints; RAG+Prompt adds step-by-step responses and safety prompts; RAG-Struct uses fixed-field output; FB-FSM-RAG further introduces SOP matching, FSM skeleton planning, RAG evidence alignment, backward compliance verification, and FSM fallback. FB-FSM-RAG (without FSM fallback) removes the FSM fallback mechanism from the complete method to analyze its impact on the effective SOP output rate.

[0035] The experiment was conducted under a unified configuration. The AnalyticDB knowledge base was used to store equipment manuals, maintenance materials, and fragments of operation documents. An external SOP-FSM process library was used to store structured process constraints. The original AnalyticDB knowledge base was not modified during the experiment. SOP semantic representation used text-embedding-v4, the main generation model was qwen-max, the generation temperature was set to 0.01, and top_p was set to 0.8. The SOP candidate recall Top-K was 3, and the intent matching weight was set to... , , The highest matching score threshold was set to 0.35, the candidate score difference threshold to 0.05, the security constraint satisfaction rate threshold to 0.8, and the AnalyticDB retrieval return count to 12. All methods used the same retrieval scope, generative model, generation parameters, and evaluation script.

[0036] Evaluation metrics include routing accuracy, SOP Top-1 matching accuracy, step coverage, sequence validity, security constraint satisfaction rate, evidence support rate, comprehensive compliance score, and end-to-end effective SOP output rate. Routing accuracy measures whether the system output type matches the manually labeled type; SOP Top-1 matching accuracy evaluates whether process issues are matched with the correct SOP.

[0037] For process compliance evaluation, this embodiment is based on the generated state sequence in Embodiment 1. G Standard Step Sequence Required Steps Set and safety constraint set Calculate the relevant indicators.

[0038] (1) The steps coverage rate, the order validity rate, and the safety constraint satisfaction rate are defined as follows: in, This indicates the degree to which the generated answer covers the necessary steps; This indicates the proportion of generation steps whose adjacent order conforms to a standard step sequence. This indicates the degree to which safety constraints are satisfied. This represents the number of adjacent steps in the generated sequence that conform to the FSM transition relationship or standard order. This represents the total number of adjacent steps in the generated sequence. This indicates the number of security constraints that have been satisfied.

[0039] (2) Evidence support rate Evidence support rate is used to measure whether the generation step can find corresponding operation objects, main actions, or state conditions to support it in the AnalyticDB retrieval fragment. All methods use the same step parsing and evidence matching rules, defined as: in, This indicates the number of generation steps supported by retrieval evidence. This indicates the total number of generation steps. The SOP-FSM process library is mainly used to define step boundaries, execution order, and safety constraints, and is not directly included in the evidence support rate.

[0040] (3) Comprehensive compliance score The overall compliance score is defined as follows: The comprehensive compliance score is primarily used to help compare the overall quality of process answers generated by different methods. Considering the SOP process Q&A's focus on sequence, completeness of steps, safety constraints, and evidence support, the weights of the four indicators are set to 0.4, 0.3, 0.2, and 0.1, respectively.

[0041] (4) End-to-end effective SOP output rate The end-to-end valid SOP output rate measures the overall availability of the system from process matching to final compliant output. A valid output is defined as follows: The system matches the correct target SOP, and the final result passes the FSM hard constraint check corresponding to that SOP. in, This represents the number of questions where the target SOP is correctly matched and the final answer passes the hard constraint check. This indicates the total number of issues covered by the SOP-FSM process.

[0042] Step coverage, sequence validity, security constraint satisfaction, evidence support, and overall compliance score are calculated on parsable candidate answers, including results that ultimately fail the hard constraints; end-to-end effective SOP output rate is calculated based on the number of answers that correctly match the target SOP and ultimately pass the validation.

[0043] For comparison methods that lack online FSM verification capabilities, their output is parsed into a sequence of steps, and offline evaluation is performed based on the same reference SOP-FSM.

[0044] In 175 comprehensive test questions, the system's strict routing accuracy was 93.14%; in 115 SOP-FSM coverage process questions, the SOPTop-1 matching accuracy and end-to-end effective SOP output rate were both 94.78%. Samples that did not produce effective output were mainly concentrated in boundary scenarios such as similar candidate processes, insufficient limitation of operation objects, and insufficient retrieval evidence.

[0045] To compare the generation effects of different methods in SOP process Q&A, this embodiment conducts a comparative experiment on 115 SOP-FSM covered process questions, comparing four methods: basic RAG, prompt-enhanced RAG, structured output RAG, and FB-FSM-RAG.

[0046] Table 2 .

[0047] As shown in Table 2, FB-FSM-RAG achieved the highest scores in all indicators except for the evidence support rate. Compared to cue constraints and fixed structured output, FSM skeleton planning and backward compliance verification significantly improve step completeness, execution order, and security constraints. The relatively low evidence support rate of the method in this embodiment is mainly because some fallback steps are derived from the audited SOP-FSM process description and were not included in the AnalyticDB retrieval evidence.

[0048] To further illustrate the manifestation of the above-mentioned differences in specific question-and-answer scenarios, we selected explicit process questions, vague process questions, and high-risk questions as typical cases for analysis, and the results are shown in Table 3.

[0049] Table 3 .

[0050] These cases demonstrate that the system can choose to answer, clarify, or refuse to answer based on the clarity of the target and the risk of the problem.

[0051] To analyze the impact of the FSM fallback mechanism on the usability of process question answering, this paper compares the results of the complete method with those after removing the FSM fallback on 115 process questions, while keeping the other modules consistent. The experimental results are shown in Table 4.

[0052] Table 4 .

[0053] As shown in Table 4, after disabling FSM fallback, the effective SOP output rate of the system decreased from 94.78% to 72.17%, and the number of rejected responses increased from 6 to 32. This indicates that FSM fallback helps reduce rejected responses caused by structured generation failing validation and improves the usability of process answers.

[0054] The embodiments described above are merely preferred embodiments of the present invention and are not intended to limit the scope of the present invention. Various modifications and improvements made by those skilled in the art to the technical solutions of the present invention without departing from the spirit of the present invention should fall within the protection scope defined by the claims of the present invention.

Claims

1. A method for enhancing the generation of question-and-answer functions for tire forming machine SOP retrieval based on finite state machine constraints, characterized in that, Includes the following steps: S1. Based on the industry standard operating procedures (SOPs), obtain the FSM process library and problem triage results; S2. Based on the intent index of the process library and user questions, obtain a reliable SOP match; S3. Based on reliable SOP matching and knowledge base evidence, obtain the constrained generated answer and verification result; S4. Based on the compliance verification results and the fallback mechanism, obtain the final controlled process answer.

2. The method for enhancing the generation of question-and-answer functions for tire forming machine SOP retrieval based on finite state machine constraints according to claim 1, characterized in that, S1 includes: organizing SOPs into an external SOP-FSM process library, which consists of two parts: SOP intent index and FSM process constraints; when handling problems, identifying the problem type and making a risk judgment, non-process problems enter the search enhancement question and answer generation channel, problems with unclear objects or non-unique candidates output clarification answers, and high-risk problems are directly rejected.

3. The method for enhancing SOP retrieval and generating questions and answers for tire forming machines based on finite state machine constraints according to claim 2, characterized in that, In the SOP-FSM process library, the intent index records the process name, category, keywords, aliases, operation objects, and step summaries, which are used for matching issues to SOPs; The FSM process constraint record includes state nodes, state transitions, standard step sequences, required step sets, and security constraint sets.

4. The method for enhancing the generation of question-and-answer functions for tire forming machine SOP retrieval based on finite state machine constraints according to claim 3, characterized in that, In the SOP-FSM process library, the first i The FSM process instance corresponding to each SOP is defined as follows: in, Si A set of state nodes; Ei It is a set of state transitions; and These represent the start and end states; Li This is a standard sequence of steps; Pi A set of necessary steps; Ci The set of safety constraints includes recording shutdowns, power outages, pressure relief, status confirmations, protection resets, and trial operation checks; This allows natural language SOPs to be organized into structured process data.

5. The method for enhancing SOP retrieval and generating questions and answers for tire forming machines based on finite state machine constraints according to claim 1, characterized in that, S2 includes: for process-oriented questions, candidate matching is performed based on the SOP intent index, and the matching score consists of three parts: semantic similarity, keyword hit and rule correction. When the highest matching score reaches the matching threshold, and the difference between the highest matching score and the second highest matching score reaches a set value, the candidate with the highest score is determined to be a reliable match. When the difference between candidate SOPs does not reach the set value, or when the operation object and limiting information are insufficient, a clarification response is output, requiring supplementary operation object or limiting information; When no reliable SOP match is found and no further clarification is required, output a reference process answer or refuse to answer based on the sufficiency of evidence retrieved from the knowledge base.

6. The method for enhancing SOP retrieval and generating questions and answers for tire forming machines based on finite state machine constraints according to claim 5, characterized in that, The semantic similarity in the matching score is used to measure the semantic similarity between the question and the candidate SOP retrieval text. The keyword hit is used to measure the hit score of the operation object, action word and limiting word. The rule correction is used to correct the rules for similar parts, similar actions and insufficient object limitation.

7. The method for enhancing the generation of question-and-answer functions for tire forming machine SOP retrieval based on finite state machine constraints according to claim 1, characterized in that, The S3 includes: after obtaining a reliable matching SOP-FSM process instance, reading the standard step sequence, the set of necessary steps, and the set of safety constraints to form a process skeleton that limits the scope of steps, their sequence, and safety conditions; By retrieving relevant instruction manual fragments, maintenance instructions, and operation descriptions from the knowledge base, user questions, matched SOP information, FSM step skeleton, retrieval evidence, and structured output requirements are input into the large language model to guide the model to generate process answers step by step. The structured answer is mapped to a state sequence, compared with the standard step sequence, the set of required steps, and the set of safety constraints, and a hard constraint check is performed; the answer that passes the check is output as the FSM-SOP process answer.

8. The method for enhancing the generation of question-and-answer functions for tire forming machine SOP retrieval based on finite state machine constraints according to claim 7, characterized in that, The hard constraint checks include illegal step checks, duplicate step checks, required step coverage checks, sequence legality checks, and safety constraint checks. The illegal step check is used to determine whether there are states in the generated state sequence that do not belong to the current SOP-FSM; The repeated step check is used to determine whether the same state is repeatedly output; The required step coverage check is used to determine whether all the key states in the required step set have appeared; the sequence validity check is used to determine whether the relative order of states conforms to the standard step sequence. The safety constraint check is used to determine whether the requirements for shutdown, power outage, pressure relief, status confirmation, protection reset, and trial operation check are fully retained.

9. The method for enhancing the generation of question-and-answer functions for tire forming machine SOP retrieval based on finite state machine constraints according to claim 1, characterized in that, S4 includes: If the initial structured generation result fails the validation, the reason for the failure will be fed back to the large language model, and a controlled regeneration will be performed within the original SOP-FSM process instance and the original retrieval evidence. If the regenerated result still fails the verification, the FSM fallback mechanism is triggered to generate a deterministic process answer based on the standard step sequence, state description and security constraints; Based on the question triage results, SOP matching results, knowledge base retrieval evidence, and FSM compliance verification results, the final output type is determined. The final output type includes general Q&A, reference process Q&A, FSM-SOP process answer, clarification answer, and rejection answer. Specifically, the generated process answer will be output as the FSM-SOP process answer only if the backward compliance verification of the corresponding SOP-FSM process instance is passed.

Images