A complete production method and system for embodied intelligent micro-factory for customized jeans

Abstract

This invention relates to the field of intelligent manufacturing technology in the apparel industry, and more particularly to a embodied intelligent micro-factory full-process production method and system for customized jeans. The method includes: acquiring user customization requirements; converting these requirements into structured requirement constraints and outputting CAD pattern data; performing process planning on the CAD pattern data and outputting task semantics; generating operation sequences for each piece of equipment and automated guided vehicle (AGV) transport plans, and executing them according to the plans; performing quality inspection; updating process parameters, equipment control parameters, and scheduling results; and triggering rework paths or rescheduling when quality anomalies, equipment malfunctions, or congestion are detected. This invention can quickly transform user requirements into manufacturable patterns and process solutions, and achieve autonomous collaboration and closed-loop optimization of multiple equipment and processes such as cutting, handling, sewing, finishing, and quality inspection within the micro-factory, thereby improving the response efficiency, process stability, and finished product consistency of single-piece or small-batch customization.

Description

Technical Field

[0001] This invention relates to the field of intelligent manufacturing technology for clothing, and in particular to a complete production method and system for a personalized intelligent micro-factory for custom-made jeans. Background Technology

[0002] As a typical garment with a complex structure, jeans typically include multiple components such as the waistband, fly, pockets, legs, and hems, involving numerous sewing and finishing processes. The production process is characterized by many steps, strong process dependence, and a high density of quality control points. For jeans "custom-made to user needs," additional challenges must be addressed, including differences in body size, pattern customization, fabric characteristics, and the rhythm of individual production. This makes it difficult for traditional production methods relying on manual experience to simultaneously meet the requirements of efficiency, cost, and consistency.

[0003] In existing technologies, the apparel industry has widely adopted digital and automated methods such as CAD pattern making, pattern making, automatic cutting, barcode or RFID tracking, MES production management, and AGV logistics handling. Some scenarios have also introduced rule-based scheduling systems or advanced planning and scheduling (APS) systems based on optimization models. However, organizing these dispersed capabilities into an end-to-end closed-loop system "from user demand input to finished trouser output," especially in single-piece or small-batch customized denim scenarios, still faces the following shortcomings: (1) Users’ natural language needs (such as pattern preferences, wearing experience, style elements, etc.) are difficult to be directly mapped into manufacturable size parameters, pattern parameters and process constraints. Designers and pattern makers often need to communicate and confirm repeatedly and make patterns manually, resulting in long cycles, poor reproducibility, and the design information is easily lost or ambiguous when it is transmitted to the production end.

[0004] (2) Even with automated cutting, AGV handling, or partially automated sewing equipment, the connection between processes (such as matching of cut pieces to workstations, organizing sewing sequence, coordinating equipment cycle time, and handling tooling changes and path conflicts) still often relies on manual scheduling or static rules. For high-frequency, single-piece customized orders, static rules are difficult to balance efficiency and stability, and lack the ability to replan and reschedule in real time when faced with disturbances such as equipment congestion, malfunctions, and order insertions.

[0005] (3) The equipment in each process of jeans manufacturing differs significantly in terms of control methods, executable actions, adjustable parameters, tooling constraints, and safety constraints. Existing systems typically use a method of "manual programming or teaching specific equipment" to solidify the process. Once the product style changes or the process parameters change, reconfiguration or even reprogramming is often required, making it difficult to achieve rapid adaptation and autonomous learning for new tasks.

[0006] (4) Existing quality inspections are mostly end-point sampling or phased inspections. The test results lack traceable correlation modeling and automatic feedback mechanisms with the previous version, process parameters, and equipment control parameters. As a result, after a problem is discovered, it mainly relies on manual rework and experience correction, making it difficult to form an adaptive closed loop of "detection-positioning-adjustment-verification", which affects the consistency and delivery stability of customized products.

[0007] (5) Existing digital systems often cover the design end (CAD / PLM), manufacturing end (MES / APS) and equipment end (PLC / motion control) respectively. The semantic gap between systems is large, making it difficult to integrate "design intent, process rules, equipment capabilities, on-site status and quality feedback" into the same decision-making framework. This results in limited end-to-end automation and makes it difficult to form a truly "micro-factory-level" autonomous collaborative production. Summary of the Invention

[0008] The technical problem this invention aims to solve is to provide a fully automated intelligent micro-factory production method and system for customized jeans. This system can automatically transform users' personalized needs into manufacturable patterns and process solutions, and achieve autonomous collaborative execution and closed-loop optimization of multiple processes and equipment within the micro-factory, including cutting, AGV handling, multi-station sewing, ironing and finishing, and quality inspection. This solves the problems existing in current customized production, such as "fragmented links from demand to design to manufacturing, reliance on manual scheduling for cross-process collaboration, high adaptation costs due to differences in equipment capabilities, and lack of closed-loop linkage between quality inspection and process parameters." As a result, it significantly improves the delivery efficiency, process stability, and finished product consistency of single or small-batch customized jeans, while reducing reliance on manual labor and rework rates.

[0009] This invention is achieved through the following technical solution: A fully automated intelligent micro-factory production method for custom-made jeans includes the following steps: S1: Obtain user-customized requirements, convert user-customized requirements into structured requirement constraints, generate size parameters and pattern parameters that match the structured requirement constraints in the pattern template library, and perform size adaptation verification, process accessibility verification, and delivery date or cost constraint verification on the size parameters and pattern parameters, and output CAD pattern data. S2: Construct a process planning model based on historical order samples, equipment capacity constraints, process rule constraints, and safety constraints. Train the process planning model under the condition of freezing the parameters of the basic large language model. When processing the current order, call the trained process planning model, combine the denim structure knowledge base and process rule base, perform process planning on the CAD pattern making data, and output task semantics including process route, process-level task tree, and equipment-level action sequence. S3: Generate a digital work order based on the task semantics and bind a unique cut piece identifier to each cut piece. Then, using a collaborative scheduling method, generate the operation sequence of each equipment and the handling plan of the automated guided vehicle based on process sequence constraints, equipment capacity constraints, unique cut piece identifiers and automated guided vehicle path conflict constraints. Compile the task semantic representation into a set of control instructions that each equipment can execute, and control the corresponding equipment to execute according to the operation sequence of each equipment and the handling plan of the automated guided vehicle. S4: Perform quality inspection on cut pieces, semi-finished products and finished products in production. Based on the unique identifier of the cut piece and the associated equipment status receipt and process parameter information, perform root cause localization. Map the defect type or size deviation to the corresponding process and corresponding equipment parameter set, and update the process parameters, equipment control parameters and scheduling results. When a quality abnormality, equipment failure or congestion is detected, trigger the generation of rework path or trigger rescheduling, so that the tasks in the updated digital work order re-enter the instruction compilation, execution and verification closed loop.

[0010] In the optimized step S1, a manual body measurement input method and a two-dimensional human body image estimation method are used to generate the user's body size vector and body shape feature vector in the user's customized requirements. The two-dimensional human body image estimation method includes human body key point detection and human body segmentation, and outputs the circumference and length dimensions based on the calibration scale.

[0011] In the optimized step S1, the generative language model based on the Transformer architecture extracts imperative information and parses user-customized requirements into structured requirement constraints.

[0012] In the optimized step S1, a parametric pattern generation method using template retrieval, rule grading, and constraint solving is adopted. The target template is retrieved from the pattern template library, and grading and local structural adjustment are performed based on the user's human body size vector to obtain size parameters such as waist circumference, hip circumference, crotch depth, pant length, leg circumference, knee circumference, and leg opening size, as well as pattern parameters.

[0013] Furthermore, the process planning model constructed in step S2 is Equation (1), and a data construction method of equipment capability modeling, teaching data collection, and process log archiving is adopted. Process-level professional training sets are established for cutting equipment, automated guided vehicles, sewing equipment, and ironing finishing equipment respectively. The training objective of the process planning model is Equation (2): (1); (2) in: This represents the process planning training set. Indicates the first A fragment of evidence, Indicates the first One output, Indicates the first One constraint This represents the training objective of the process planning model. This indicates the loss from monitoring and fine-tuning. express, This represents the loss due to executability constraints.

[0014] Furthermore, in step S2, the method for combining the denim structure knowledge base and the process rule base to perform process planning on the CAD pattern data and outputting task semantics including process route, process-level task tree, and equipment-level action sequence is as follows: S211: Perform structural analysis on CAD pattern making data to extract information on pattern type, structural relationship, and pattern parameter; S212: Using the cut piece type, structural relationship and pattern parameter information as the search key, match the corresponding structural unit in the jeans structure knowledge base; S213: Retrieve the process constraint rules corresponding to the structural unit from the process rule base, including process sequence constraints, equipment capability constraints, parameter range constraints, and quality control constraints; input the retrieved process constraint rules as constraint conditions into the process planning model, and generate process routes, process-level task trees, and equipment-level action sequences through constraint guidance.

[0015] Furthermore, in step S3, a collaborative scheduling method is then employed to generate the operation sequence for each piece of equipment and the handling plan for the automated guided vehicle (AGV) based on process sequence constraints, equipment capacity constraints, unique identifiers of cut pieces, and path conflict constraints of the AGV. The method is as follows: S311: Using process start time, completion time, equipment allocation variables, and automated guided vehicle (AGV) task allocation variables as decision variables, and using process sequence constraints, equipment capacity constraints, tooling or line change constraints, AGV travel time constraints, and path conflict constraints as linear constraints, a scheduling model is constructed. S312: The optimization objective of the scheduling model is established using a weighted objective function construction method. The optimization objective includes at least minimizing the total completion time, minimizing the work-in-process waiting time, minimizing the number of tooling or line changes, and minimizing energy consumption. S313: Solve the scheduling model using the branch and bound algorithm or the cutting plane algorithm, and output the operation sequence of each device and the handling plan of the automated guided vehicle.

[0016] In the optimized version, step S4 uses a parameter adaptive optimization method to update the process parameters, equipment control parameters, and scheduling results.

[0017] Furthermore, the parameter adaptive optimization method is one or more combinations of Bayesian optimization method, gradient-based parameter optimization method, and feedback-based reinforcement learning optimization method.

[0018] A fully automated intelligent micro-factory production system for custom jeans, used to execute a fully automated intelligent micro-factory production method for custom jeans as described above, includes a user customization demand information acquisition module, a CAD pattern making data output module, a process planning model training module, a task semantic output module, an equipment operation sequence and automated guided vehicle handling plan generation module, a control module, and a quality closed loop and event-triggered rescheduling module. The user customization requirement information acquisition module is used to acquire user customization requirement information; The CAD pattern making data output module is used to convert user-customized requirements into structured requirement constraints, generate size parameters and pattern parameters that match the structured requirement constraints in the pattern template library, and perform size adaptation verification, process accessibility verification, and delivery date or cost constraint verification on the size parameters and pattern parameters, and output CAD pattern making data. The process planning model training module constructs a process planning model based on historical order samples, equipment capacity constraints, process rule constraints, and safety constraints, and trains the process planning model under the condition of freezing the parameters of the basic large language model. The task semantic output module is used to call the trained process planning model when processing the current order, and combine the denim structure knowledge base and process rule base to perform process planning on the CAD pattern data, and output task semantics including process route, process-level task tree and equipment-level action sequence. The equipment operation sequence and automated guided vehicle (AGV) handling plan generation module is used to generate digital work orders based on the task semantics, bind a unique cut piece identifier to each cut piece, and then use a collaborative scheduling method to generate each equipment operation sequence and AGV handling plan based on process sequence constraints, equipment capacity constraints, cut piece unique identifiers and AGV path conflict constraints. The control module is used to compile the task semantic representation into a set of control instructions that can be executed by each device, and control the corresponding devices to execute according to the work sequence of each device and the handling plan of the automated guided vehicle. The quality closed-loop and event-triggered rescheduling module is used to perform quality inspection on cut pieces, semi-finished products and finished products in production. It performs root cause localization based on the equipment status feedback and process parameter information associated with the unique identifier of the cut piece, maps the defect type or size deviation to the corresponding process and corresponding equipment parameter set, and updates the process parameters, equipment control parameters and scheduling results. When a quality abnormality, equipment failure or congestion is detected, it triggers the generation of rework path or event-triggered rescheduling, so that the tasks in the updated digital work order re-enter the instruction compilation, execution and verification closed loop.

[0019] Beneficial effects of the invention: The present invention provides a fully automated intelligent micro-factory production method and system for customized jeans, which has the following advantages: (1) Automatically structuring user needs and mapping them to pattern parameters, fabric schemes and CAD data reduces the information gap between the design and manufacturing ends and repeated manual communication, and realizes the end-to-end closed loop from demand to finished pants.

[0020] (2). By scheduling and event-triggered rescheduling, combined with AGV handling plans and process sequence constraints, the scheduling can be quickly rescheduled when there are order insertions, congestion or equipment abnormalities, thereby improving the overall throughput and delivery stability and realizing autonomous collaboration and dynamic optimization of multiple equipment and multiple processes.

[0021] (3) By generating process routes, process-level task trees and equipment-level action sequences through process planning models, and combining task semantic representation with the compilation and issuance mechanism of equipment control instructions, unified arrangement and skill call for "what to do, how to do it, and when to coordinate" of different processes are realized, reducing the cost of traditional manual programming or teaching, and improving the equipment's ability to generalize and quickly adapt to new orders or new models.

[0022] (4) By linking the visual or dimensional inspection results with process parameters, equipment control parameters and scheduling strategies, a closed-loop optimization of "inspection-positioning-adjustment-verification" is achieved, which reduces the rework rate and improves the consistency of finished products, forming a quality-driven adaptive closed loop.

[0023] (5) Through the unique identification of cut pieces, digital work orders and process status feedback, the whole-chain tracking from cut pieces to finished products is realized, which facilitates quality traceability, process review and continuous improvement, and enhances the traceability and manageability of the whole process.

[0024] (6) By introducing executability constraint loss into the training objective of the process planning model, and adding instruction compilation and executability verification in the process of converting task semantic representation into equipment control instructions, the constraints of equipment capability, process sequence, piece_id matching, parameter range and AGV path conflict are moved to the process planning stage, so that the generated results meet the requirements of compilability and executability when outputting, reducing invalid process plans, manual correction and on-site trial and error, improving the scheduling success rate, equipment coordination efficiency and quality stability of single or small batch denim customization production, and improving the executability of process planning. Attached Figure Description

[0025] Picture 1 This is a schematic diagram of the overall process of the present invention. Detailed Implementation

[0026] A fully automated intelligent micro-factory production method for custom-made jeans, the overall process is shown in Figure 1, and specifically includes the following steps: S1: Obtain user-customized requirements, convert user-customized requirements into structured requirement constraints, generate size parameters and pattern parameters that match the structured requirement constraints in the pattern template library, and perform size adaptation verification, process accessibility verification, and delivery date or cost constraint verification on the size parameters and pattern parameters, and output CAD pattern data. This step is for customized design generation.

[0027] Specifically, the user's body size vector and body shape feature vector in the user's customized requirements can be generated by manual body measurement and two-dimensional human image estimation methods. The two-dimensional human image estimation method includes human body key point detection and human body segmentation, and outputs the circumference and length dimensions based on the calibration scale.

[0028] The optimized approach can extract instructional information based on a generative language model with a Transformer architecture, parsing user-customized requirements from natural language into structured requirement constraints. These structured requirement constraints are output in a data format with a preset field structure, which includes at least: pattern constraints, wearing experience constraints, appearance and craftsmanship constraints, fabric or accessory constraints, and delivery date or cost constraints. Furthermore, the structured requirement constraints are subjected to field integrity checks, unit normalization checks, numerical range checks, and enumeration validity checks.

[0029] A parametric pattern generation method can be adopted, which involves template retrieval, rule grading, and constraint solving. The target template is retrieved from the pattern template library, and grading and local structural adjustments are performed based on the user's human body size vector to obtain size parameters such as waist circumference, hip circumference, crotch depth, pant length, leg circumference, knee circumference, and leg opening size, as well as pattern parameters.

[0030] The AHP-TOPSIS multi-attribute decision method can be used to filter and sort the fabric attribute library and output fabric or accessory selection schemes; and the CAD pattern generation method can be used to output the paper pattern outline, seam allowance, alignment mark and process annotation; the layout data is generated by the layout method based on No-Fit Polygon, and the cutting path data is generated, while a unique identifier and tracking information are generated for each piece.

[0031] By automatically structuring user needs and mapping them to pattern parameters, fabric schemes, and CAD data, information gaps between the design and manufacturing ends and repetitive manual communication are reduced, achieving an end-to-end closed loop from demand to finished trousers.

[0032] S2: Construct a process planning model based on historical order samples, equipment capacity constraints, process rule constraints, and safety constraints. Train the process planning model while freezing the parameters of the basic large language model. When processing the current order, call the trained process planning model and combine it with the jeans structure knowledge base and process rule base to perform process planning on the CAD pattern making data, and output task semantics including process route, process-level task tree, and equipment-level action sequence. This step can be summarized as process planning and task decomposition.

[0033] Specifically, data construction methods such as equipment capability modeling, teaching data collection, and process log archiving can be adopted to establish process-level professional training for cutting equipment, automated guided vehicles, sewing equipment, and ironing finishing equipment.

[0034] Furthermore, the process planning model constructed in step S2 is Equation (1), and the training objective of the process planning model is Equation (2): (1); (2) in: This represents the process planning training set. Indicates the first A fragment of evidence, Indicates the first One output, Indicates the first One constraint This represents the training objective of the process planning model. This indicates the loss from monitoring and fine-tuning. express, This represents the loss due to executability constraints.

[0035] Specifically, a fusion fine-tuning method incorporating supervised fine-tuning loss and executability constraint loss can be used to train the process planning model. The executability constraint loss is obtained by inputting the task semantic representation output by the process planning model into an instruction compiler and performing executability checks on the compilation results. Violations generated by these executability checks are then incorporated back into the training objective, ensuring that the output of the process planning model tends towards compilability and executability.

[0036] The specific implementation is as follows: Freeze the parameters of the basic large language model, and introduce LoRA low-rank adaptation only in the attention layer or feedforward layer to satisfy... W’ = W + W , W=BA ,in A ∈ R r×d , B ∈R d×r , r d During training, only the low-rank fitness matrix is ​​updated. A , B， To reduce training and deployment costs in customized scenarios. Specifically: W’ This represents the updated weight matrix after introducing low-rank adaptation. ，W This represents the original weight matrix of the corresponding layer in the basic large language model. ， W This represents the low-rank incremental update matrix of the original weight matrix. ，B This represents the second low-rank fitting matrix that maps the low-dimensional subspace back to the original high-dimensional parameter space. ，A This represents the first low-rank fitting matrix that maps the original high-dimensional parameter space to a low-dimensional subspace. ，R Representing the real number field ，r Represents the rank of the low-rank fit ，d This indicates the dimension of the original weight matrix.

[0037] Construction of the fusion loss function: The total loss is defined as L total = L sft + λ exec L exec The monitoring fine-tuning loss is defined as ; among which, L total L represents the total loss. sft Indicates the monitoring and fine-tuning loss, λ exec L represents the fusion weighting coefficient for the executability constraint loss. exec Indicates the loss due to executability constraints. This indicates the number of training samples in the process planning training set. This represents the conditional probability distribution determined by the trainable parameters of the process planning model. The trainable parameters of the process planning model are represented. This indicates the first section of the process planning training set. The nth training sample target output sequence Target markers at each location, This indicates the first section of the process planning training set. The input of each training sample, Representation and Process Planning Training Collection, Chapter The set of constraints corresponding to each training sample This indicates the first section of the process planning training set. The target output sequence of each training sample. This represents the position index in the target output sequence. This indicates the first section of the process planning training set. The training sample at the th ... The target prefix sequence generated before each position. This indicates the first section of the process planning training set. The length of the target output sequence for each training sample.

[0038] Calculation of executability constraint loss: The task semantic representation output from the process planning model is input into the instruction compiler to generate the intermediate representation (IR) of the equipment instructions, and the constraint checker is called to check each item in the constraint set C. The executability constraint loss is defined as follows: L exec = w 1 v id ( z , C ) + w 2 v order ( z , C ) + w 3 v cap ( z , C ) + w 4 v param ( z , C ) + w 5 v path ( z , C ) + M 1 compile_fail ,in, v id ( z , C This indicates the mismatch between the product's unique identifier and the target process / station. v order ( z , C This indicates a violation of the correct order of operations; v cap ( z , C This indicates a violation of the target device's ability to not support the action sequence; v param ( z , C This indicates a deviation of process parameters from the allowable or safe range of the equipment. v path ( z , C) represents the AGV transport path conflict or unreachability violation; w1~w5 are the weight coefficients of the corresponding constraints; M is the compilation failure penalty constant; 1 compile_fail This is a function to indicate compilation failure. When an instruction fails to compile, an unexecutable action occurs, or a security conflict occurs, the corresponding violation value will be set to a preset large value, thereby making the model output more compilable and executable.

[0039] In one embodiment, v param ( z , C This can be expressed as the cumulative sum of the excesses of each process parameter. Specifically, it involves calculating the portion of each parameter that exceeds the upper limit and the portion that falls below the lower limit, and then summing them up. v cap ( z , C This can be represented as the number of actions in the action sequence that do not belong to the set of actions allowed by the target device; v path ( z , C This can be represented as the number of conflicting or unreachable path segments in the AGV planning path; when the compiler returns a failure code, let 1... compile_fail Select 1 if the value is 1, otherwise select 0.

[0040] Furthermore, in step S2, the method for combining the denim structure knowledge base and the process rule base to perform process planning on the CAD pattern data and outputting task semantics including process route, process-level task tree, and equipment-level action sequence is as follows: S211: Perform structural analysis on CAD pattern making data to extract information on pattern type, structural relationship, and pattern parameter; S212: Using the cut piece type, structural relationship and pattern parameter information as the search key, match the corresponding structural unit in the jeans structure knowledge base; S213: Retrieve the process constraint rules corresponding to the structural unit from the process rule base, including process sequence constraints, equipment capability constraints, parameter range constraints, and quality control constraints; input the retrieved process constraint rules as constraint conditions into the process planning model, and generate process routes, process-level task trees, and equipment-level action sequences through constraint guidance.

[0041] S3: Generate a digital work order based on the task semantics and bind a unique piece identifier (piece_id) to each piece. Then, using a collaborative scheduling method, generate the operation sequence of each device and the handling plan of the automated guided vehicle based on process sequence constraints, equipment capacity constraints, unique piece identifiers, and automated guided vehicle path conflict constraints. Compile the task semantic representation into a set of control instructions that each device can execute, and control the corresponding devices to execute according to the operation sequence of each device and the handling plan of the automated guided vehicle. This step can be summarized as multi-device collaborative execution.

[0042] Furthermore, in step S3, a collaborative scheduling method is then employed to generate the operation sequence for each piece of equipment and the handling plan for the automated guided vehicle (AGV) based on process sequence constraints, equipment capacity constraints, unique identifiers of cut pieces, and path conflict constraints of the AGV. The method is as follows: S311: Using process start time, completion time, equipment allocation variables, and automated guided vehicle (AGV) task allocation variables as decision variables, and using process sequence constraints, equipment capacity constraints, tooling or line change constraints, AGV travel time constraints, and path conflict constraints as linear constraints, a scheduling model is constructed. S312: The optimization objective of the scheduling model is established using a weighted objective function construction method. The optimization objective includes at least minimizing the total completion time, minimizing the work-in-process waiting time, minimizing the number of tooling or line changes, and minimizing energy consumption. S313: Solve the scheduling model using the branch and bound algorithm or the cutting plane algorithm, and output the operation sequence of each device and the handling plan of the automated guided vehicle.

[0043] Specifically, the output of each equipment operation sequence includes the operation sequences of cutting equipment, automated guided vehicles (AGVs), multi-station sewing equipment, and ironing and finishing equipment.

[0044] This invention generates process routes, process-level task trees, and equipment-level action sequences through process planning models. Combined with the compilation and issuance mechanism from task semantic representation to equipment control instructions, it achieves unified arrangement and skill invocation of "what to do, how to do it, and when to coordinate" for different processes. This reduces the cost of traditional manual programming or teaching and improves the equipment's ability to generalize and quickly adapt to new orders or new models.

[0045] Furthermore, by introducing executability constraint loss into the training objective of the process planning model, and adding instruction compilation and executability verification during the conversion from task semantic representation to equipment control instructions, the constraints of equipment capacity, process sequence, piece_id matching, parameter range, and AGV path conflict are moved forward to the process planning stage. This ensures that the generated results meet the requirements of compilability and executability at the time of output, reducing invalid process plans, manual corrections, and on-site trial and error. This improves the scheduling success rate, equipment coordination efficiency, and quality stability of single-piece or small-batch custom denim production, and enhances the executability of process planning.

[0046] S4: Perform quality inspection on cut pieces, semi-finished products, and finished products during production. Based on the unique identifier of each cut piece and the associated equipment status feedback and process parameter information, perform root cause localization, mapping defect categories or dimensional deviations to corresponding processes and equipment parameter sets. Update process parameters, equipment control parameters, and scheduling results. When quality anomalies, equipment malfunctions, or congestion are detected, trigger rework path generation or rescheduling, ensuring that tasks in the updated digital work order re-enter the instruction compilation, execution, and verification closed loop. This step can be summarized as quality inspection and closed-loop optimization.

[0047] In the optimized version, step S4 uses a parameter adaptive optimization method to update the process parameters, equipment control parameters, and scheduling results.

[0048] Furthermore, the parameter adaptive optimization method is one or more combinations of Bayesian optimization method, gradient-based parameter optimization method, and feedback-based reinforcement learning optimization method.

[0049] Quality inspection of cut pieces, semi-finished products, and finished products during production includes the following steps: S411: Appearance Defect Detection and Location: A deep learning-based visual detection method is used to identify defects such as stitch defects, alignment deviations, stains, holes, and wrinkles, and outputs the defect category and location. S412: Critical Dimension Inspection: Key dimensions such as waist circumference, hip circumference, pant length, inner length, and leg opening are obtained by using dimensional measurement methods and compared with the target pattern parameters to obtain the dimensional deviation vector.

[0050] This invention links visual or dimensional inspection results with process parameters, equipment control parameters, and scheduling strategies, achieving closed-loop optimization of "inspection-positioning-adjustment-verification," reducing rework rates and improving finished product consistency.

[0051] By scheduling and event-triggered rescheduling, combined with AGV handling plans and process sequence constraints, rapid rescheduling can be achieved when orders are interrupted, congestion occurs, or equipment malfunctions occur, thereby improving overall throughput and delivery stability and realizing autonomous collaboration and dynamic optimization of multiple devices and processes.

[0052] Furthermore, by using unique identifiers for cut pieces, digital work orders, and process status feedback, the entire chain of tracking from cut pieces to finished products is achieved, facilitating quality traceability, process review, and continuous improvement, thereby enhancing the traceability and manageability of the entire process.

[0053] A fully automated intelligent micro-factory production system for custom jeans, used to execute a fully automated intelligent micro-factory production method for custom jeans as described above, includes a user customization demand information acquisition module, a CAD pattern making data output module, a process planning model training module, a task semantic output module, an equipment operation sequence and automated guided vehicle handling plan generation module, a control module, and a quality closed loop and event-triggered rescheduling module. The user customization requirement information acquisition module is used to acquire user customization requirement information; The CAD pattern making data output module is used to convert user-customized requirements into structured requirement constraints, generate size parameters and pattern parameters that match the structured requirement constraints in the pattern template library, and perform size adaptation verification, process accessibility verification, and delivery date or cost constraint verification on the size parameters and pattern parameters, and output CAD pattern making data. The process planning model training module constructs a process planning model based on historical order samples, equipment capacity constraints, process rule constraints, and safety constraints, and trains the process planning model under the condition of freezing the parameters of the basic large language model. The task semantic output module is used to call the trained process planning model when processing the current order, and combine the denim structure knowledge base and process rule base to perform process planning on the CAD pattern data, and output task semantics including process route, process-level task tree and equipment-level action sequence. The equipment operation sequence and automated guided vehicle (AGV) handling plan generation module is used to generate digital work orders based on the task semantics, bind a unique cut piece identifier to each cut piece, and then use a collaborative scheduling method to generate each equipment operation sequence and AGV handling plan based on process sequence constraints, equipment capacity constraints, cut piece unique identifiers and AGV path conflict constraints. The control module is used to compile the task semantic representation into a set of control instructions that can be executed by each device, and control the corresponding devices to execute according to the work sequence of each device and the handling plan of the automated guided vehicle. The quality closed-loop and event-triggered rescheduling module is used to perform quality inspection on cut pieces, semi-finished products and finished products in production. It performs root cause localization based on the equipment status feedback and process parameter information associated with the unique identifier of the cut piece, maps the defect type or size deviation to the corresponding process and corresponding equipment parameter set, and updates the process parameters, equipment control parameters and scheduling results. When a quality abnormality, equipment failure or congestion is detected, it triggers the generation of rework path or event-triggered rescheduling, so that the tasks in the updated digital work order re-enter the instruction compilation, execution and verification closed loop.

[0054] In summary, the embodied intelligent micro-factory full-process production method and system for customized jeans provided by this invention can automatically transform users' personalized needs into manufacturable patterns and process solutions. Within the micro-factory, it enables autonomous collaborative execution and closed-loop optimization of multiple processes and equipment, including cutting, AGV handling, multi-station sewing, ironing and finishing, and quality inspection. This solves the problems existing in current customized production, such as "fragmented links from demand to design to manufacturing, reliance on manual scheduling for cross-process collaboration, high adaptation costs due to differences in equipment capabilities, and lack of closed-loop linkage between quality inspection and process parameters." As a result, it significantly improves the delivery efficiency, process stability, and finished product consistency of single or small-batch customized jeans, while reducing reliance on manual labor and rework rates.

[0055] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A fully automated intelligent micro-factory production method for customized jeans, characterized by: Includes the following steps: S1: Obtain user-customized requirements, convert user-customized requirements into structured requirement constraints, generate size parameters and pattern parameters that match the structured requirement constraints in the pattern template library, and perform size adaptation verification, process accessibility verification, and delivery date or cost constraint verification on the size parameters and pattern parameters, and output CAD pattern data. S2: Construct a process planning model based on historical order samples, equipment capacity constraints, process rule constraints, and safety constraints. Train the process planning model under the condition of freezing the parameters of the basic large language model. When processing the current order, call the trained process planning model, combine the denim structure knowledge base and process rule base, perform process planning on the CAD pattern making data, and output task semantics including process route, process-level task tree, and equipment-level action sequence. S3: Generate a digital work order based on the task semantics and bind a unique cut piece identifier to each cut piece. Then, using a collaborative scheduling method, generate the operation sequence of each equipment and the handling plan of the automated guided vehicle based on process sequence constraints, equipment capacity constraints, unique cut piece identifiers and automated guided vehicle path conflict constraints. Compile the task semantic representation into a set of control instructions that each equipment can execute, and control the corresponding equipment to execute according to the operation sequence of each equipment and the handling plan of the automated guided vehicle. S4: Perform quality inspection on cut pieces, semi-finished products and finished products in production. Based on the unique identifier of the cut piece and the associated equipment status receipt and process parameter information, perform root cause localization. Map the defect type or size deviation to the corresponding process and corresponding equipment parameter set, and update the process parameters, equipment control parameters and scheduling results. When a quality abnormality, equipment failure or congestion is detected, trigger the generation of rework path or trigger rescheduling, so that the tasks in the updated digital work order re-enter the instruction compilation, execution and verification closed loop.

2. The full-process production method of a personalized intelligent micro-factory for customized jeans as described in claim 1, characterized in that: In step S1, a manual body measurement input method and a two-dimensional human body image estimation method are used to generate the user's body size vector and body shape feature vector in the user's customized requirements. The two-dimensional human body image estimation method includes human body key point detection and human body segmentation, and outputs the circumference and length dimensions based on the calibration scale.

3. The end-to-end production method of a personalized intelligent micro-factory for customized jeans as described in claim 1, characterized in that: In step S1, the generative language model based on the Transformer architecture extracts imperative information and parses user-customized requirements into structured requirement constraints.

4. The end-to-end production method of a personalized intelligent micro-factory for customized jeans as described in claim 2, characterized in that: In step S1, a parametric pattern generation method using template retrieval, rule grading, and constraint solving is adopted. The target template is retrieved from the pattern template library, and grading and local structural adjustment are performed based on the user's human body size vector to obtain size parameters such as waist circumference, hip circumference, crotch depth, pant length, leg circumference, knee circumference, and leg opening size, as well as pattern parameters.

5. The full-process production method of a personalized intelligent micro-factory for customized jeans as described in claim 1, characterized in that: The process planning model constructed in step S2 is Equation (1), and the data construction method of equipment capability modeling, teaching acquisition, and process log archiving is adopted. Process-level professional training sets are established for cutting equipment, automatic guided transport vehicles, sewing equipment, and ironing finishing equipment respectively. The training objective of the process planning model is Equation (2): （1）； （2） in: This represents the process planning training set. Indicates the first A fragment of evidence, Indicates the first One output, Indicates the first One constraint This represents the training objective of the process planning model. This indicates the loss from monitoring and fine-tuning. express, This represents the loss due to executability constraints.

6. The end-to-end production method of a personalized intelligent micro-factory for customized jeans as described in claim 1, characterized in that: In step S2, the method for combining the denim structure knowledge base and the process rule base to perform process planning on the CAD pattern data and outputting task semantics including process route, process-level task tree, and equipment-level action sequence is as follows: S211: Perform structural analysis on CAD pattern making data to extract information on pattern type, structural relationship, and pattern parameter; S212: Using the cut piece type, structural relationship and pattern parameter information as the search key, match the corresponding structural unit in the jeans structure knowledge base; S213: Retrieve the process constraint rules corresponding to the structural unit from the process rule base, including process sequence constraints, equipment capability constraints, parameter range constraints, and quality control constraints; input the retrieved process constraint rules as constraint conditions into the process planning model, and generate process routes, process-level task trees, and equipment-level action sequences through constraint guidance.

7. The end-to-end production method of a personalized intelligent micro-factory for customized jeans as described in claim 1, characterized in that: In step S3, a collaborative scheduling method is then used to generate the operation sequence of each piece of equipment and the handling plan of the automated guided vehicle based on process sequence constraints, equipment capacity constraints, unique identifiers of cut pieces, and path conflict constraints of the automated guided vehicle. The method is as follows: S311: Using process start time, completion time, equipment allocation variables, and automated guided vehicle (AGV) task allocation variables as decision variables, and using process sequence constraints, equipment capacity constraints, tooling or line change constraints, AGV travel time constraints, and path conflict constraints as linear constraints, a scheduling model is constructed. S312: The optimization objective of the scheduling model is established using a weighted objective function construction method. The optimization objective includes at least minimizing the total completion time, minimizing the work-in-process waiting time, minimizing the number of tooling or line changes, and minimizing energy consumption. S313: Solve the scheduling model using the branch and bound algorithm or the cutting plane algorithm, and output the operation sequence of each device and the handling plan of the automated guided vehicle.

8. The end-to-end production method of a personalized intelligent micro-factory for customized jeans as described in claim 1, characterized in that: In step S4, the process parameters, equipment control parameters, and scheduling results are updated using a parameter adaptive optimization method.

9. The end-to-end production method of a personalized intelligent micro-factory for customized jeans as described in claim 8, characterized in that: The parameter adaptive optimization method is one or more combinations of Bayesian optimization method, gradient-based parameter optimization method, and feedback-based reinforcement learning optimization method.

10. A fully automated intelligent micro-factory production system for customized jeans, characterized in that: The method for implementing the full-process production method of a smart micro-factory for jeans customization as described in any one of claims 1 to 9 includes a user customization demand information acquisition module, a CAD pattern making data output module, a process planning model training module, a task semantic output module, an equipment operation sequence and automatic guided vehicle handling plan generation module, a control module, and a quality closed loop and event-triggered rescheduling module. The user customization requirement information acquisition module is used to acquire user customization requirement information; The CAD pattern making data output module is used to convert user-customized requirements into structured requirement constraints, generate size parameters and pattern parameters that match the structured requirement constraints in the pattern template library, and perform size adaptation verification, process accessibility verification, and delivery date or cost constraint verification on the size parameters and pattern parameters, and output CAD pattern making data. The process planning model training module constructs a process planning model based on historical order samples, equipment capacity constraints, process rule constraints, and safety constraints, and trains the process planning model under the condition of freezing the parameters of the basic large language model. The task semantic output module is used to call the trained process planning model when processing the current order, and combine the denim structure knowledge base and process rule base to perform process planning on the CAD pattern data, and output task semantics including process route, process-level task tree and equipment-level action sequence. The equipment operation sequence and automated guided vehicle (AGV) handling plan generation module is used to generate digital work orders based on the task semantics, bind a unique cut piece identifier to each cut piece, and then use a collaborative scheduling method to generate each equipment operation sequence and AGV handling plan based on process sequence constraints, equipment capacity constraints, cut piece unique identifiers and AGV path conflict constraints. The control module is used to compile the task semantic representation into a set of control instructions that can be executed by each device, and control the corresponding devices to execute according to the work sequence of each device and the handling plan of the automated guided vehicle. The quality closed-loop and event-triggered rescheduling module is used to perform quality inspection on cut pieces, semi-finished products and finished products in production. It performs root cause localization based on the equipment status feedback and process parameter information associated with the unique identifier of the cut piece, maps the defect type or size deviation to the corresponding process and corresponding equipment parameter set, and updates the process parameters, equipment control parameters and scheduling results. When a quality abnormality, equipment failure or congestion is detected, it triggers the generation of rework path or event-triggered rescheduling, so that the tasks in the updated digital work order re-enter the instruction compilation, execution and verification closed loop.

Images