A method and system for closed-loop recovery of multi-agent group flow planning under constraints

Abstract

The present application relates to the technical field of computer-aided process planning, and particularly relates to a closed-loop recovery method and system for multi-intelligent group process planning under manufacturing constraints, comprising: formalizing manufacturing constraints into quadruples with severity weights, driving the sequential execution and calling of constraint checking of feature extraction intelligent groups, macro process planning intelligent groups, specific process planning intelligent groups and process optimization evaluation intelligent groups; when there is a violation, locating a root cause intelligent group by traversing upstream from a constraint responsibility intelligent group along a directed dependency graph, determining a rollback target and injecting a repair prompt according to a severity weight and a rollback cost heuristic; verifying again after the rollback target intelligent group is regenerated, iteratively triggering root cause tracking and adaptive rollback until passing. The present application realizes accurate attribution and efficient closed-loop recovery under manufacturing constraint violations, improves constraint satisfaction rate and reduces rollback steps.

Description

Technical Field

[0001] This invention relates to the field of computer-aided process planning technology, and in particular to a closed-loop recovery method and system for multi-intelligent group process planning under manufacturing constraints. Background Technology

[0002] Process planning is a crucial link between product design and manufacturing execution, requiring complex decisions under manufacturing constraints such as processing sequence, resource allocation, and precision requirements. In recent years, multi-intelligent collaborative systems driven by large language models have been applied to automatically generate process routes through task decomposition and professional collaboration, significantly improving the intelligence level of computer-aided process planning.

[0003] However, existing methods primarily focus on improving generation quality, implicitly assuming that the pipeline continues execution even in the event of unconstrained failures. This assumption fails in constraint-intensive manufacturing scenarios: simple retries cannot address the attribution of which intelligence group introduced the violation; global regeneration from scratch discards correctly generated intermediate results and risks reintroducing errors into previously successful stages; general reflection mechanisms target open-ended task quality, not structured multi-intelligence pipelines. These shortcomings result in a lack of systematic recovery mechanisms after manufacturing constraint violations, relying solely on manual correction or costly global regeneration.

[0004] Therefore, it is necessary to treat the structured recovery after manufacturing constraint violations as an independent sub-problem. When a constraint violation occurs, the responsible intelligent group should be identified, a rollback target that balances the scope of correction and the cost of regeneration should be selected, and targeted repair guidance should be injected to improve the constraint satisfaction rate and recovery efficiency of the process planning system. Summary of the Invention

[0005] The purpose of this invention is to provide a closed-loop recovery method and system for multi-intelligent group process planning under manufacturing constraints, thereby solving the aforementioned problems existing in the prior art.

[0006] To achieve the above objectives, the present invention provides a closed-loop recovery method for multi-intelligent group process planning under manufacturing constraints, comprising the following steps: S1. Acquire and parse product data files and multi-view images to extract processing feature datasets; formalize manufacturing constraints into quadruplets with severity weights, including: intelligent group for driving feature extraction, intelligent group for macro-process planning, intelligent group for specific process planning, and intelligent group for process optimization evaluation; execute the quadruplets in sequence based on the processing feature dataset to generate process routes and perform constraint verification. S2. When a constraint verification violation occurs, traverse the incoming edges from the directly responsible intelligent group upstream along the directed dependency graph, and perform Boolean matching on the structured fields output by the upstream according to the defect detection rule script to locate the root cause intelligent group. S3. Based on the constraint severity level and rollback cost heuristic, determine the rollback target intelligent group, generate a repair prompt containing the violation constraint identifier, problem description and targeted repair requirements, and inject it into the rollback target intelligent group. S4. The rollback target intelligent group regenerates the process route based on the repair prompts, calls the constraint check function again to verify the constraints, and forwards it to the process optimization evaluation intelligent group to output the comprehensive evaluation result. If it fails, it iteratively triggers root cause tracking and adaptive rollback, and stores the successfully verified feature process routes into the industrial memory module.

[0007] Preferably, S1 includes: Manufacturing constraints are defined as follows: C i =(rule_id i agent_scope i repair_strategy i ,severity i ); Among them, severity i The values ​​∈(0,1) are assigned by domain experts based on the impact of violations on part quality and process feasibility, and are classified as critical severities. i ∈[0.9,1.0], importance / severity i ∈[0.7,0.8], minor severity i =0.4 three levels; The process is executed sequentially, including the feature extraction intelligent group, the macro-process planning intelligent group, the specific process planning intelligent group, and the process optimization evaluation intelligent group, to generate the process route π. The constraint satisfaction is verified by the constraint check function check(π,Ci)∈{pass,fail}.

[0008] Preferably, in S2, the root cause intelligence group includes: Construct a directed dependency graph D=(A,E), where A is the set of intelligent groups, and the edge (ai,aj)∈E indicates that the output of ai is the direct input of aj. The execution depths of the feature extraction intelligent group, the macro process planning intelligent group, the specific process planning intelligent group, and the process optimization evaluation intelligent group are set to 1, 2, 3, and 4 respectively. From the constraint quadruple, agent_scope i Determine the directly responsible intelligent group, traverse the incoming edges layer by layer upstream, and use the defect detection rule script to perform Boolean matching on the structured fields output by the upstream. The upstreammost node that meets the defect conditions is the root cause intelligent group. When multiple nodes meet the conditions, select the one with the smallest depth.

[0009] Preferably, in S3, the rollback target intelligent group is determined heuristically based on the constraint severity level and rollback cost, including: Calculate rollback costs: ; in, The cost is used as a heuristic metric to measure the regeneration range, based on the execution depth in the dependency graph. Initialize the cumulative penalty vector P as a zero vector, the remaining rollback budget Brem as a preset value B, the penalty increment α = 0.3 and the penalty increment threshold θ = 0.8. When the cumulative penalty of the same smart group exceeds the threshold θ, the penalty is incremented and rollback to the upstream is considered. Prioritize handling high-severity constraint violations. Critical levels are non-negotiable, important levels must be met, and minor levels can be relaxed when the budget is exhausted. Choose rollback targets that balance the scope of correction and the cost of regeneration.

[0010] Preferably, in step S3, the step of generating a repair prompt containing a violation identifier, a problem description, and a repair strategy and injecting it into the target includes: Append a repair instruction to the end of the user prompt for the target smart group to be rolled back. The repair instruction includes the identifier and location of the violation constraint, a description of the problem, and instructions based on the repair strategy. i Targeted repair requirements for field derivation; Targeted remediation requirements include reordering, supplementing available machine tools and cutting tools, reselecting process chains or adding auxiliary processes, to clearly guide the rollback target intelligent group to avoid corresponding constraint violations during regeneration, improve constraint satisfaction rate and reduce invalid rollbacks.

[0011] Preferably, in S4, the step of regenerating the process route based on the repair prompts and performing closed-loop verification includes: The rollback target intelligent group re-executes the process planning and generates new intermediate outputs based on the repair prompts, and the downstream intelligent group regenerates the complete process route based on the updated inputs. The constraint check function check(π,C) is called again on the regenerated process route. i Perform constraint-by-constraint verification. If all constraints pass, forward the results to the process optimization evaluation intelligent group for comprehensive evaluation. If constraint violations still exist, the iteration triggers root cause tracing in S2 and adaptive rollback in S3 until all constraints are satisfied or the remaining rollback budget is exhausted.

[0012] Preferably, S1 further includes building a shared industrial memory module before the sequential execution of the multi-intelligent group, specifically: Collect feature pairs and process routes from historical successful process planning cases, index them according to feature type similarity, and persistently store them in the shared industrial memory module; Before the multi-intelligent group executes process planning, related cases are extracted from the industrial memory module through feature type similarity retrieval and injected into the prompt words of the corresponding intelligent group to improve the constraint satisfaction rate of the initial process route. The industrial memory module continuously stores successfully verified feature process route pairs in S4 for reference in the process planning of similar parts.

[0013] Preferably, a closed-loop recovery system for multi-intelligent group process planning under manufacturing constraints includes: The quadruple construction module is used to acquire and parse product data files and multi-view images, extract processing feature datasets, and formalize manufacturing constraints into quadruples with severity weights. The quadruples include: intelligent group for driving feature extraction, intelligent group for macro process planning, intelligent group for specific process planning, and intelligent group for process optimization evaluation. The quadruples are executed sequentially based on the processing feature dataset to generate process routes and perform constraint verification. The first verification module is used to traverse the incoming edges from the directly responsible intelligent group upstream along the directed dependency graph when a violation is found in the constraint verification. Based on the defect detection rule script, it performs Boolean matching on the structured fields output by the upstream to locate the root cause intelligent group. The repair injection module is used to heuristically determine the rollback target smart group based on the constraint severity level and rollback cost, generate a repair prompt containing the violation constraint identifier, problem description and targeted repair requirements and inject it into the rollback target smart group. The secondary verification module is used to roll back the target intelligent group to regenerate the process route based on the repair prompts, and call the constraint check function again to perform constraint verification. If it passes, it forwards the result to the process optimization evaluation intelligent group to output the comprehensive evaluation result. If it fails, it iteratively triggers root cause tracking and adaptive rollback, and stores the successfully verified feature process routes into the industrial memory module.

[0014] The advantages and beneficial effects of this invention compared to the prior art are: 1. The manufacturing constraints are formalized into quadruples with severity weights, and the root cause is traced through the intelligent group dependency graph to achieve accurate attribution and responsibility positioning of constraint violations, avoiding blind retries; 2. Employ severity-driven hierarchical recovery control, heuristically determine rollback targets based on constraint severity levels and rollback costs, prioritize critical and important constraints, and improve constraint satisfaction rate; 3. By using a repair prompt injection mechanism, targeted repair instructions are added to the prompt words of the target smart group to achieve efficient closed-loop recovery with fewer rollback steps and reduce invalid rollback overhead; 4. Successful cases of building a shared industrial memory module for persistent storage, providing reference for intelligent groups through feature type similarity retrieval, improving the first pass rate of the initial process route, and reducing the cost of regeneration.

[0015] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0016] Figure 1 This is a flowchart of a closed-loop recovery method for multi-intelligent group process planning under manufacturing constraints, as described in an embodiment of the present invention. Figure 2 This is a structural diagram of a closed-loop recovery system for multi-intelligent group process planning under manufacturing constraints, according to an embodiment of the present invention. Detailed Implementation

[0017] In the description of this invention, it should be noted that the terms "upper," "lower," "inner," and "outer," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings, or the orientation or positional relationship commonly used when the product is in use. They are used only for the convenience of describing the invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on the invention. In the description of this invention, it should also be noted that, unless otherwise explicitly specified and limited, the terms "set," "install," and "connect" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal communication between two components. Those skilled in the art can understand the specific meaning of the above terms in this invention based on the specific circumstances.

[0018] The embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0019] like Figure 1 As shown, this invention provides a closed-loop recovery method for multi-intelligent group process planning under manufacturing constraints, comprising the following steps: S1. Acquire and parse product data files and multi-view images to extract processing feature datasets; formalize manufacturing constraints into quadruplets with severity weights, including: intelligent group for driving feature extraction, intelligent group for macro-process planning, intelligent group for specific process planning, and intelligent group for process optimization evaluation; execute the quadruplets in sequence based on the processing feature dataset to generate process routes and perform constraint verification. S2. When a constraint verification violation occurs, traverse the incoming edges from the directly responsible intelligent group upstream along the directed dependency graph, and perform Boolean matching on the structured fields output by the upstream according to the defect detection rule script to locate the root cause intelligent group. S3. Based on the constraint severity level and rollback cost heuristic, determine the rollback target intelligent group, generate a repair prompt containing the violation constraint identifier, problem description and targeted repair requirements, and inject it into the rollback target intelligent group. S4. The rollback target intelligent group regenerates the process route based on the repair prompts, calls the constraint check function again to verify the constraints, and forwards it to the process optimization evaluation intelligent group to output the comprehensive evaluation result. If it fails, it iteratively triggers root cause tracking and adaptive rollback, and stores the successfully verified feature process routes into the industrial memory module.

[0020] Preferably, S1 includes: Manufacturing constraints are defined as follows: C i =(rule_id i agent_scope i repair_strategy i ,severity i ); Among them, severity i The values ​​∈(0,1) are assigned by domain experts based on the impact of violations on part quality and process feasibility, and are classified as critical severities. i ∈[0.9,1.0], importance / severity i ∈[0.7,0.8], minor severity i =0.4 three levels; The process is executed sequentially, including the feature extraction intelligent group, the macro-process planning intelligent group, the specific process planning intelligent group, and the process optimization evaluation intelligent group, to generate the process route π. The constraint satisfaction is verified by the constraint check function check(π,Ci)∈{pass,fail}.

[0021] Preferably, in S2, the root cause intelligence group includes: Construct a directed dependency graph D=(A,E), where A is the set of intelligent groups, and the edge (ai,aj)∈E indicates that the output of ai is the direct input of aj. The execution depths of the feature extraction intelligent group, the macro process planning intelligent group, the specific process planning intelligent group, and the process optimization evaluation intelligent group are set to 1, 2, 3, and 4 respectively. From the constraint quadruple, agent_scope i Determine the directly responsible intelligent group, traverse the incoming edges layer by layer upstream, and use the defect detection rule script to perform Boolean matching on the structured fields output by the upstream. The upstreammost node that meets the defect conditions is the root cause intelligent group. When multiple nodes meet the conditions, select the one with the smallest depth.

[0022] Preferably, in S3, the rollback target intelligent group is determined heuristically based on the constraint severity level and rollback cost, including: Calculate rollback costs: ; in, The cost is used as a heuristic metric to measure the regeneration range, based on the execution depth in the dependency graph. Initialize the cumulative penalty vector P as a zero vector, the remaining rollback budget Brem as a preset value B, the penalty increment α = 0.3 and the penalty increment threshold θ = 0.8. When the cumulative penalty of the same smart group exceeds the threshold θ, the penalty is incremented and rollback to the upstream is considered. Prioritize handling high-severity constraint violations. Critical levels are non-negotiable, important levels must be met, and minor levels can be relaxed when the budget is exhausted. Choose rollback targets that balance the scope of correction and the cost of regeneration.

[0023] Preferably, in step S3, the step of generating a repair prompt containing a violation identifier, a problem description, and a repair strategy and injecting it into the target includes: Append a repair instruction to the end of the user prompt for the target smart group to be rolled back. The repair instruction includes the identifier and location of the violation constraint, a description of the problem, and instructions based on the repair strategy. i Targeted repair requirements for field derivation; Targeted remediation requirements include reordering, supplementing available machine tools and cutting tools, reselecting process chains or adding auxiliary processes, to clearly guide the rollback target intelligent group to avoid corresponding constraint violations during regeneration, improve constraint satisfaction rate and reduce invalid rollbacks.

[0024] Preferably, in S4, the step of regenerating the process route based on the repair prompts and performing closed-loop verification includes: The rollback target intelligent group re-executes the process planning and generates new intermediate outputs based on the repair prompts, and the downstream intelligent group regenerates the complete process route based on the updated inputs. The constraint check function check(π,C) is called again on the regenerated process route. i Perform constraint-by-constraint verification. If all constraints pass, forward the results to the process optimization evaluation intelligent group for comprehensive evaluation. If constraint violations still exist, the iteration triggers root cause tracing in S2 and adaptive rollback in S3 until all constraints are satisfied or the remaining rollback budget is exhausted.

[0025] Preferably, S1 further includes building a shared industrial memory module before the sequential execution of the multi-intelligent group, specifically: Collect feature pairs and process routes from historical successful process planning cases, index them according to feature type similarity, and persistently store them in the shared industrial memory module; Before the multi-intelligent group executes process planning, related cases are extracted from the industrial memory module through feature type similarity retrieval and injected into the prompt words of the corresponding intelligent group to improve the constraint satisfaction rate of the initial process route. The industrial memory module continuously stores successfully verified feature process route pairs in S4 for reference in the process planning of similar parts.

[0026] Preferred, such as Figure 2 As shown, a closed-loop recovery system for multi-intelligent group process planning under manufacturing constraints includes: The quadruple construction module is used to acquire and parse product data files and multi-view images, extract processing feature datasets, and formalize manufacturing constraints into quadruples with severity weights. The quadruples include: intelligent group for driving feature extraction, intelligent group for macro process planning, intelligent group for specific process planning, and intelligent group for process optimization evaluation. The quadruples are executed sequentially based on the processing feature dataset to generate process routes and perform constraint verification. The first verification module is used to traverse the incoming edges from the directly responsible intelligent group upstream along the directed dependency graph when a violation is found in the constraint verification. Based on the defect detection rule script, it performs Boolean matching on the structured fields output by the upstream to locate the root cause intelligent group. The repair injection module is used to heuristically determine the rollback target smart group based on the constraint severity level and rollback cost, generate a repair prompt containing the violation constraint identifier, problem description and targeted repair requirements and inject it into the rollback target smart group. The secondary verification module is used to roll back the target intelligent group to regenerate the process route based on the repair prompts, and call the constraint check function again to perform constraint verification. If it passes, it forwards the result to the process optimization evaluation intelligent group to output the comprehensive evaluation result. If it fails, it iteratively triggers root cause tracking and adaptive rollback, and stores the successfully verified feature process routes into the industrial memory module.

[0027] The following specific embodiments will be used for verification.

[0028] Example 1: Step 1: Experiment Setup Dataset and Partitioning: The dataset originates from the internal process database of a precision aerospace manufacturing company, containing 100 precision parts across four categories: housings (30 pieces), shafts (28 pieces), discs (22 pieces), and brackets (20 pieces), with precision levels ranging from IT6 to IT10. The dataset was stratified by part category and divided into three completely independent subsets (61+13+26=100): a training reference set of 61 examples (used for industrial memory and process knowledge graph construction); a validation set of 13 examples (used for hyperparameter tuning); and a test set of 26 examples (used for final performance evaluation).

[0029] Evaluation indicators: (1) Constraint satisfaction rate (CSR, main indicator): the proportion of parts that pass all critical (severity∈[0.9,1.0]) and important (severity∈[0.7,0.8]) constraint checks; (2) First pass rate (FPR): the proportion of parts that pass constraint checks without any rollback; (3) Average number of rollbacks; (4) Processing time; (5) TOPSIS comprehensive score as a supplementary quality indicator.

[0030] Baseline methods: (1) Single intelligent group: The entire pipeline is completed by a single LLM; (2) Multi-intelligent group - open loop: a four-stage multi-intelligent group framework with no recovery mechanism; (3) Multi-intelligent group - retry: rollback to the directly responsible intelligent group and retry as is (maximum 3 times); (4) Multi-intelligent group - full restart: global restart from FEA (maximum 2 times); (5) Feedback retry: convert violation constraints into natural language feedback without dependency graph tracing. All baselines share the same multi-intelligent group skeleton, constraint validator, and knowledge conditions.

[0031] Step 2 Comparison with the baseline method: In the current enterprise dataset, CAAR's CSR (84.6%) is statistically significantly higher than most baselines (p<0.05). CAAR achieves an additional 7.7 percentage points improvement over Feedback-Retry (76.9%), demonstrating the practical value of combining structured root cause attribution with hierarchical control. CAAR achieves this improvement with fewer rollback steps (1.73 vs. 2.03), proving that its gain comes from more accurate rollback target selection rather than additional retries. The global restart method (78.5%) incurs the highest time overhead (12.8 minutes) while having a lower CSR than CAAR, confirming the dual disadvantages of the global strategy in terms of cost and quality. The performance comparison results on the test set are shown in Table 1.

[0032] Table 1. Performance Comparison Results of the Test Set

[0033] Step 3 Ablation Study: Ablation results revealed the specific contribution mechanisms of each component: Repair hint injection (variant E) had the greatest impact on CSR (-10.4 percentage points, p<0.05); Root cause tracking (variant B) ranked second (-7.4 percentage points, p<0.05); Variant C (without cumulative penalty) had a CSR of 82.3%, and the difference from the full CAAR was not statistically significant (p=0.09>0.05), indicating that local loops are rarely triggered under the current test set size—the value of cumulative penalty is mainly to prevent extreme scenarios; Variant F (without industrial memory) reduced FPR to 65.4% (p<0.05), confirming that industrial memory mainly affects the initial generation quality, rather than the recovery capability itself. Ablation results are shown in Table 2.

[0034] Table 2 Ablation Results

[0035] Step 4: Rollback Behavior and Attribution Case Analysis Root Cause Attribution Case Study: Taking the process planning of a certain shaft-type part as an example, constraint C3 violation was detected (direct responsibility intelligent group: SPPA). Root cause tracing traversed upstream along the dependency graph and ultimately attributed the root cause to FEA (which recorded the machining method of this feature as a mixture of 'milling' and 'turning', causing an abnormal process chain selection in SPPA). The system was rolled back to FEA and a repair suggestion was injected, and recovery was completed with only one accurate rollback. However, when using the retry baseline (rollback only to SPPA), since the root cause was in FEA, SPPA regeneration would still trigger the same constraint violation, requiring two additional rollbacks. The cross-model CSR comparison results are shown in Table 3.

[0036] Table 3. Cross-model CSR comparison results

[0037] Step 5 Conclusion: This invention explicitly defines "structured recovery after manufacturing constraint violations" as an independent sub-problem and proposes a CAAR mechanism, contributing at three levels: (i) Problem definition—modeling recovery as a severity-level closed loop of constraint violation, attribution, rollback, and remediation; (ii) Decision mechanism—unifying responsibility attribution and heuristic rollback target selection into a single control loop; (iii) Control strategy—a three-layer hierarchical logic combining severity-driven priority ranking, incremental cumulative penalties, and gradual relaxation. CAAR is an interpretable heuristic recovery controller; its root cause tracing serves as an operational attribution for effective rollback selection, rather than strict causal inference, and does not provide optimality guarantees.

[0038] With the current enterprise dataset (100 cases, 26 cases in the test set) and experimental configuration, CAAR achieved a CSR of 84.6% (±2.1%), which is statistically significantly better than most baselines (Wilcoxonp<0.05). Ablation experiments confirmed that repair cue injection and root cause tracing are the two components that contribute the most to CSR. Variant C (without cumulative penalty) did not establish a statistically significant benefit (p=0.09); its value lies mainly in preventing rare local loop scenarios (6.8% of cases in the full log analysis).

[0039] In this application, unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains. In case of any inconsistency, the meaning as set forth in this specification or derived from the content described herein shall prevail. Furthermore, the terminology used in this invention is for the purpose of describing embodiments of this application only and is not intended to limit this application.

[0040] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the technical solutions of the present invention, and these modifications or equivalent substitutions cannot cause the modified technical solutions to deviate from the spirit and scope of the technical solutions of the present invention.

Claims

1. A closed-loop recovery method for multi-intelligent group process planning under manufacturing constraints, characterized in that, Includes the following steps: S1. Obtain and parse product data files and multi-view images to extract processing feature datasets; Manufacturing constraints are formalized as quadruples with severity weights. The quadruples include: a driving feature extraction intelligent group, a macro process planning intelligent group, a specific process planning intelligent group, and a process optimization evaluation intelligent group. The quadruples are executed sequentially based on the processing feature dataset to generate a process route and perform constraint verification. S2. When a constraint verification violation occurs, traverse the incoming edges from the directly responsible intelligent group upstream along the directed dependency graph, and perform Boolean matching on the structured fields output by the upstream according to the defect detection rule script to locate the root cause intelligent group. S3. Based on the constraint severity level and rollback cost heuristic, determine the rollback target intelligent group, generate a repair prompt containing the violation constraint identifier, problem description and targeted repair requirements, and inject it into the rollback target intelligent group. S4. The rollback target intelligent group regenerates the process route based on the repair prompts, calls the constraint check function again to verify the constraints, and forwards it to the process optimization evaluation intelligent group to output the comprehensive evaluation result. If it fails, it iteratively triggers root cause tracking and adaptive rollback, and stores the successfully verified feature process routes into the industrial memory module.

2. The closed-loop recovery method for multi-intelligent group process planning under manufacturing constraints according to claim 1, characterized in that, S1 includes: Manufacturing constraints are defined as follows: C i =(rule_id i ,agent_scope i ,repair_strategy i ,severity i ); Among them, severity i The values ​​∈(0,1) are assigned by domain experts based on the impact of violations on part quality and process feasibility, and are classified as critical severities. i ∈[0.9,1.0], importance / severity i ∈[0.7,0.8], minor severity i =0.4 three levels; The process is executed sequentially, including the feature extraction intelligent group, the macro-process planning intelligent group, the specific process planning intelligent group, and the process optimization evaluation intelligent group, to generate the process route π. The constraint satisfaction is verified by the constraint check function check(π,Ci)∈{pass,fail}.

3. The closed-loop recovery method for multi-intelligent group process planning under manufacturing constraints according to claim 2, characterized in that, In S2, the root cause intelligence group includes: Construct a directed dependency graph D=(A,E), where A is the set of intelligent groups, and the edge (ai,aj)∈E indicates that the output of ai is the direct input of aj. The execution depths of the feature extraction intelligent group, the macro process planning intelligent group, the specific process planning intelligent group, and the process optimization evaluation intelligent group are set to 1, 2, 3, and 4 respectively. From the constraint quadruple, agent_scope i Determine the directly responsible intelligent group, traverse the incoming edges layer by layer upstream, and use the defect detection rule script to perform Boolean matching on the structured fields output by the upstream. The upstreammost node that meets the defect conditions is the root cause intelligent group. When multiple nodes meet the conditions, select the one with the smallest execution depth.

4. The closed-loop recovery method for multi-intelligent group process planning under manufacturing constraints according to claim 3, characterized in that, In step S3, the rollback target intelligent group is determined heuristically based on constraint severity level and rollback cost, including: Calculate rollback costs: ； in, The cost is used as a heuristic metric to measure the regeneration range, based on the execution depth in the dependency graph. Initialize the cumulative penalty vector P as a zero vector, the remaining rollback budget Brem as a preset value B, the penalty increment α = 0.3 and the penalty increment threshold θ = 0.

8. When the cumulative penalty of the same smart group exceeds the threshold θ, the penalty is incremented and rollback to the upstream is considered. Prioritize handling high-severity constraint violations. Critical levels are non-negotiable, important levels must be met, and minor levels can be relaxed when the budget is exhausted. Choose rollback targets that balance the scope of correction and the cost of regeneration.

5. The closed-loop recovery method for multi-intelligent group process planning under manufacturing constraints according to claim 4, characterized in that, In step S3, the step of generating a repair prompt containing a violation identifier, a problem description, and a repair strategy, and injecting it into the target, includes: Append a repair instruction to the end of the user prompt for the target smart group to be rolled back. The repair instruction includes the identifier and location of the violation constraint, a description of the problem, and instructions based on the repair strategy. i Targeted repair requirements for field derivation; Targeted remediation requirements include reordering, supplementing available machine tools and cutting tools, reselecting process chains or adding auxiliary processes, to clearly guide the rollback target intelligent group to avoid corresponding constraint violations during regeneration, improve constraint satisfaction rate and reduce invalid rollbacks.

6. The closed-loop recovery method for multi-intelligent group process planning under manufacturing constraints according to claim 5, characterized in that, In step S4, the steps of regenerating the process route based on the repair prompts and performing closed-loop verification include: The rollback target intelligent group re-executes the process planning and generates new intermediate outputs based on the repair prompts, and the downstream intelligent group regenerates the complete process route based on the updated inputs. The constraint check function check(π,C) is called again on the regenerated process route. i Perform constraint-by-constraint verification. If all constraints pass, forward the results to the process optimization evaluation intelligent group for comprehensive evaluation. If constraint violations still exist, the iteration triggers root cause tracing in S2 and adaptive rollback in S3 until all constraints are satisfied or the remaining rollback budget is exhausted.

7. The closed-loop recovery method for multi-intelligent group process planning under manufacturing constraints according to claim 6, characterized in that, S1 also includes constructing a shared industrial memory module before the sequential execution of the multi-intelligent group, specifically: Collect feature pairs and process routes from historical successful process planning cases, index them according to feature type similarity, and persistently store them in the shared industrial memory module; Before the multi-intelligent group executes process planning, related cases are extracted from the industrial memory module through feature type similarity retrieval and injected into the prompt words of the corresponding intelligent group to improve the constraint satisfaction rate of the initial process route. The industrial memory module continuously stores successfully verified feature process route pairs in S4 for reference in the process planning of similar parts.

8. A closed-loop recovery system for multi-intelligent group process planning under manufacturing constraints, characterized in that, include: The quadruple construction module is used to acquire and parse product data files and multi-view images, and extract processing feature datasets. Manufacturing constraints are formalized as quadruples with severity weights. The quadruples include: a driving feature extraction intelligent group, a macro process planning intelligent group, a specific process planning intelligent group, and a process optimization evaluation intelligent group. The quadruples are executed sequentially based on the processing feature dataset to generate a process route and perform constraint verification. The first verification module is used to traverse the incoming edges from the directly responsible intelligent group upstream along the directed dependency graph when a violation is found in the constraint verification. Based on the defect detection rule script, it performs Boolean matching on the structured fields output by the upstream to locate the root cause intelligent group. The repair injection module is used to heuristically determine the rollback target smart group based on the constraint severity level and rollback cost, generate a repair prompt containing the violation constraint identifier, problem description and targeted repair requirements and inject it into the rollback target smart group. The secondary verification module is used to roll back the target intelligent group to regenerate the process route based on the repair prompts, and call the constraint check function again to perform constraint verification. If it passes, it forwards the result to the process optimization evaluation intelligent group to output the comprehensive evaluation result. If it fails, it iteratively triggers root cause tracking and adaptive rollback, and stores the successfully verified feature process routes into the industrial memory module.

Images