An agent-based textile dyeing and weaving process intelligent design method and system

Abstract

The present application belongs to the field of textile intelligent manufacturing and dyeing and weaving process optimization, and particularly relates to a kind of textile dyeing weaving process intelligent design method and system based on Agent.The method constructs knowledge base containing mechanism model, process template, constraint and quality association information, forms feature input by fusing order and grey cloth, equipment capacity, historical batch and image / video quality data, generates candidate process scheme by using Agent multi-stage reasoning, and obtains executable process that can be directly issued by constraint checking and linkage repair through device window and timing sequence;At the same time, the quality backtracking evidence chain and version iteration mechanism are introduced, and the abnormal attribution, grey verification and rollback update are realized.Compared with the prior art, the present application can improve the process design efficiency, reduce the trial and error and quality fluctuation risk, and improve the adaptation ability to new materials and new equipment.

Description

Technical Field

[0001] This invention relates to the field of intelligent manufacturing and process informatization in textiles, and more specifically, it relates to an agent-based intelligent design method and system for textile dyeing and weaving processes. Background Technology

[0002] The textile industry's orders are characterized by "small batches, multiple varieties, customization, and fast delivery," requiring frequent readjustment of process parameters and placing higher demands on the efficiency and accuracy of process design. In the dyeing process, parameters such as temperature profiles, heating rates, liquor ratios, pH strategies, feeding sequences, and concentration ratios are strongly coupled with material properties such as fiber material, weight, and warp and weft density. Deviations in these parameters can easily lead to quality problems such as color difference and excessive shrinkage. In the weaving process, parameters such as warp and weft density, tension, machine speed, and twist affect indicators such as yarn breakage probability, defect types, and fabric hand feel. Improper combinations of these parameters can easily lead to increased yarn breakage rates or fluctuations in hand feel.

[0003] Existing digital systems (such as ERP / MES) primarily focus on production data collection and traceability, lacking sufficient support for the structured accumulation of process knowledge, parameter reasoning, and solution optimization, making it difficult to achieve rapid generation and adjustment of processes for orders. On the other hand, unstructured quality data generated during production, such as color difference images, fabric defect images, and hand / drape videos, lack an effective correlation mechanism with specific process segments and key process parameters. This leads to the reliance on manual experience for locating and correcting quality anomalies, limiting efficiency and consistency, and making the knowledge difficult to reuse.

[0004] In addition, some process optimization solutions adopt general frameworks such as rule bases or feedback compensation, but they still have at least the following shortcomings in textile dyeing / weaving scenarios:

[0005] (1) There is a lack of mechanism modeling and quality impact assessment mechanism for textile characteristics, making it difficult to effectively predict quality fluctuations caused by changes in key parameters;

[0006] (2) There is a lack of expression and verification mechanism for process segment time-dependent constraints, making it difficult to handle time-series conflicts such as feeding sequence and temperature node interval;

[0007] (3) There is a lack of knowledge expression and reasoning mechanism that links key weaving parameters with quality indicators such as defects, broken yarns, and hand feel, making it difficult to make dynamic trade-offs between objectives such as quality, cost and delivery time.

[0008] Therefore, a technical solution is needed that can support the computable expression of process knowledge, automatically generate executable process schemes that meet equipment and timing constraints, and achieve stable convergence by combining quality backtracking and iterative updates. Summary of the Invention

[0009] In view of the shortcomings of the existing technology, the purpose of this application is to provide an agent-based intelligent design method and system for textile dyeing and weaving processes.

[0010] To achieve the above objectives, this application provides the following technical solution:

[0011] An agent-based intelligent design method for textile dyeing and weaving processes, characterized by the following steps:

[0012] S1, Construct a textile technology knowledge base;

[0013] S2 collects structured production data and unstructured quality data, extracts features from the unstructured quality data and fuses them with the structured production data to generate an input feature vector;

[0014] S3, based on the input feature vector, generates a set of candidate process models through the inference module;

[0015] S4, Perform constraint verification on the candidate process model set; when it is detected that a candidate process model does not meet the preset process constraints, perform parameter adjustment on the candidate process model to obtain an executable process model that meets the preset process constraints;

[0016] S5 outputs the executable process model to the production execution system for production execution;

[0017] S6 updates the textile process knowledge base in a versioned manner based on production feedback for use in subsequent process model generation.

[0018] In a preferred embodiment, the textile process knowledge base includes:

[0019] Mechanism model library, process template library, constraint network, quality correlation graph, repair action library, and version management information;

[0020] The mechanistic model library includes at least fiber thermal shrinkage models and dye uptake kinetic models;

[0021] Version management information should include at least the version identifier, scope of application, triggering reason, and rollback identifier.

[0022] In a preferred embodiment, S2 includes:

[0023] Collect structured production data, which includes order requirements, fabric parameters, equipment parameters, and historical batch data;

[0024] Collect unstructured quality data, including color difference images, defect images, and videos of hand feel and drape.

[0025] Feature extraction is performed on unstructured quality data, including extracting color difference features from color difference images, extracting defect features from defect images, and extracting appearance features from videos of hand feel and drape, in order to obtain structured quality features.

[0026] The structured production data and structured quality features are fused together to obtain the input feature vector.

[0027] In a preferred embodiment, the inference module is an Agent architecture that performs multi-stage inference, including:

[0028] Intent understanding: Transform order requirements into a set of target constraints and target weights;

[0029] Historical reasoning: Retrieve historical batches similar to the input feature vector from the textile process knowledge base and generate risk priors. Risk priors are used to limit the range of values ​​for key parameters of the process template.

[0030] Template instantiation: Call the process template library and parameterize the process template based on risk priors to generate a structured set of candidate process models;

[0031] Multi-objective trade-off: Based on quality indicators, cost indicators and delivery time indicators, a multi-objective evaluation function is constructed to evaluate and screen the set of structured candidate process models to obtain an optimal subset; wherein, the candidate process model includes at least two process parameters from the following: process segment sequence and temperature curve node corresponding to the process segment, heating rate, liquor ratio, pH strategy, feeding sequence and holding time.

[0032] In a preferred embodiment, the constraint verification in S4 includes:

[0033] Equipment capacity window constraint verification: Perform interval verification on at least one of the following parameters of the candidate process model according to the production equipment capacity: temperature range, heating rate, liquor ratio, pH control accuracy, and tension range.

[0034] Process timing logic constraint verification: Based on the constraint network, verify the dependencies and execution order between process segments, and verify the timing consistency of feeding order, temperature node interval and heat preservation time.

[0035] In a preferred embodiment, the parameter adjustment in S4 includes:

[0036] Feasible domain pruning: When the key parameters of the candidate process model exceed the equipment capacity window constraints, the excess part is truncated or recycled so that the key parameters fall into the feasible region.

[0037] Timing conflict elimination: When the process timing logic constraints are not met, adjust the feeding time point, temperature node interval or heat preservation time to eliminate timing conflicts;

[0038] Multi-objective conflict resolution: When there is a conflict between quality objectives, cost objectives and delivery objectives, a set of Pareto non-dominant resolution actions is generated, and the resolution action with the smallest evaluation function value is selected and executed according to the multi-objective evaluation function.

[0039] Textile-specific linkage repair: Based on the detected quality anomaly type, the corresponding repair action in the repair action library is called and executed. The repair actions include at least: temperature curve segment repair when shrinkage risk is detected, dye ratio linkage repair when color difference exceeds the standard, and density and tension linkage repair when yarn breakage or defects exceed the standard.

[0040] The types of quality anomalies are determined based on the quality backtracking mechanism and production feedback data from production execution.

[0041] In a preferred embodiment, the quality backtracking mechanism includes: constructing a multi-layered correlation network based on a quality correlation graph, which includes quality characteristics, process segments, key parameters, and repair actions; when production feedback indicates a quality anomaly, generating an evidence chain based on the multi-layered correlation network, the evidence chain including at least the abnormal process segment, the corresponding key parameter, the type of quality anomaly, and the recommended repair action, for attributing the quality anomaly.

[0042] In S4, the key parameters that need to be adjusted and the corresponding repair actions are determined based on the chain of evidence.

[0043] In S6, the chain of evidence is written into the version management information as a version update record.

[0044] In a preferred embodiment, the versioned update in S6 is performed when a textile-specific triggering condition is met:

[0045] The specific triggering conditions for textiles include at least: color number fluctuation triggering, chemical fiber shrinkage triggering, yarn breakage triggering, defect exceeding the standard triggering, new material triggering, and new equipment triggering.

[0046] When any textile-specific triggering condition is met, a version update is performed, a new version is generated and gray-scale verification is carried out. When the quality indicators during the gray-scale verification period deteriorate more than a preset threshold relative to the baseline version, the version is rolled back to the previous stable version.

[0047] The quality indicators include at least color difference indicators, defect indicators, and shrinkage rate indicators.

[0048] An agent-based intelligent design system for textile dyeing and weaving processes, characterized in that it includes:

[0049] The knowledge base module is used to build and maintain a textile technology knowledge base;

[0050] The data processing module is used to collect structured production data and unstructured quality data, and generate input feature vectors;

[0051] The inference module is used to generate a set of candidate process models based on the input feature vector. The inference module adopts an Agent architecture.

[0052] The constraint verification and parameter adjustment module is used to perform constraint verification on the candidate process model set, and when the candidate process model does not meet the preset process constraints, it calls the repair action library to perform parameter adjustment on the candidate process model to obtain an executable process model that meets the preset process constraints.

[0053] The production docking module is used to output the executable process model to the production execution system;

[0054] The closed-loop iteration module is used to update the textile process knowledge base in a versioned manner based on production feedback.

[0055] An agent-based intelligent design system for textile dyeing and weaving processes, characterized in that it includes:

[0056] Memory, used to store computer-executable instructions; and,

[0057] A processor for implementing the steps of the method as described above when executing the computer-executable instructions.

[0058] By adopting the above technical solution, the beneficial effects of the present invention are as follows:

[0059] For small-batch, multi-variety orders, it can quickly generate dyeing / weaving process parameter schemes that meet order requirements and equipment capabilities, reducing the number of manual adjustments and sample tests, and improving process configuration efficiency.

[0060] By unifying quality information such as color difference images, defect images, and hand feel / drape videos with production data such as orders, fabric samples, and equipment in a unified model, process recommendations are made more closely aligned with actual production conditions, improving solution adaptability and hit rate. Equipment windows and process timing constraints are validated for candidate process solutions, enabling early detection of issues such as parameter out-of-bounds errors, unreasonable material feeding sequences, and non-compliant temperature node intervals, ensuring that issued solutions are executable and implementable. In the event of conflicts or quality risks, targeted parameter adjustments and linked repair suggestions (such as temperature curves, formula ratios, density-tension combinations, etc.) are automatically provided to reduce the probability of typical problems such as color difference, excessive shrinkage, yarn breakage, and defects. Abnormal process segments and key parameters are located through a quality backtracking evidence chain, and version iterations and gray-scale verification are performed based on production feedback, promoting the accumulation of process experience and continuous optimization, maintaining quality stability, and reducing iteration risks. Attached Figure Description

[0061] Figure 1 This is a flowchart of a method according to the first embodiment of this application;

[0062] Figure 2 This is a system architecture framework diagram according to the second embodiment of this application;

[0063] Figure 3 This is a schematic diagram of the data structure of the input feature vector and the process model according to the first embodiment of this application;

[0064] Figure 4 This is a flowchart illustrating the constraint verification, conflict resolution, and backtracking iteration according to the first embodiment of this application. Detailed Implementation

[0065] In the following description, many technical details are presented to help the reader better understand this application. However, those skilled in the art will understand that the technical solutions claimed in this application can be implemented even without these technical details and various variations and modifications based on the following embodiments.

[0066] The first embodiment of this application relates to an agent-based intelligent design method for textile dyeing and weaving processes, combined with... Figure 1 , Figure 3 and Figure 4 This method achieves intelligent design and continuous optimization of textile dyeing and weaving processes by constructing a textile process knowledge base, collecting and integrating multi-source data, using agent reasoning to generate candidate process models, performing constraint verification and parameter adjustment, outputting executable process models, and updating them in a versioned manner based on production feedback.

[0067] The method includes the following steps:

[0068] Step S10: Construct a textile process knowledge base

[0069] The textile process knowledge base comprises six core components: a mechanism model library, a process template library, a constraint network, a quality correlation graph, a repair action library, and version management information.

[0070] Specifically, the mechanistic model library includes at least fiber thermal shrinkage models and dye uptake kinetic models. These models are based on materials science, chemical kinetics, and heat and mass transfer theories, providing a quantitative mapping relationship between process parameters and quality results. The fiber thermal shrinkage model establishes temperature... heating rate Insulation time A nonlinear relationship model between heat shrinkage and material properties, taking polyester as an example, is expressed as follows:

[0071]

[0072] in, This is the model parameter set, which includes the critical temperature. Shrinkage coefficient Time decay factor The model describes the contraction behavior using a piecewise function: when At that time, the shrinkage rate increases linearly with temperature; when At that time, the shrinkage rate increased exponentially.

[0073] A dye uptake kinetic model was established to model the evolution of dye uptake rate under different temperature curves and pH strategies. Taking polyester as an example, the expression of the dye uptake kinetic model is as follows:

[0074]

[0075] in, The concentration of dye adsorbed on the fabric. This is the saturation adsorption concentration. This is a function of the diffusion / adsorption coefficient.

[0076] Furthermore, the process template library includes dyeing process templates and weaving process templates. The dyeing process templates pre-set process segment sequences (pretreatment → dyeing → fixing → washing → setting), temperature curve nodes, heating rate ranges, liquor ratio ranges, pH strategies, and feed sequence templates. The weaving process templates pre-set warp and weft density ranges, tension / speed strategies, twist strategies, and downtime compensation templates. The process template library supports multi-version management; different fiber materials, dye types, and fabric applications correspond to different template versions. The inference module automatically matches and instantiates the most suitable template based on order characteristics.

[0077] Furthermore, the constraint network includes equipment capability window constraints, process timing logic constraints, and multi-objective conflict constraints, which constitute the hard boundary conditions for process model generation and verification. Among them:

[0078] The equipment capacity window constraints define the operating boundaries such as the upper limit of dyeing vat temperature (≤140℃), the range of heating rate (1-5℃ / min), the pH control accuracy (±0.2), and the range of loom tension (5-30N).

[0079] The process timing logic constraint models the dyeing or weaving process as a timing dependency network with pre- and post-conditions, clarifies the execution order, triggering conditions and time window constraints of each process segment, and represents the dependency relationship between process segments through a directed acyclic graph (DAG), ensuring the timing consistency and logical integrity of the process model.

[0080] Multi-objective conflict constraints define the mutually exclusive relationships and trade-off rules between quality, cost, and delivery time objectives. For example, high-quality objectives (color difference) Production cycles with shrinkage rates ≤2% typically require longer holding times and lower heating rates, leading to longer production cycles and higher energy costs. Low-cost targets (minimizing dye usage and energy consumption) may result in increased color differences or require more precise temperature / pH control, increasing equipment control complexity. Short lead times (production cycle ≤48h) require higher heating rates and shorter holding times, which may increase shrinkage risk and color difference fluctuations.

[0081] Constraint networks identify multi-objective conflicts through Pareto front analysis and provide decision support during the multi-objective trade-off phase.

[0082] Furthermore, the quality correlation map constructs a multi-layered correlation network of "quality characteristics → process segment → key parameters → repair actions". For the weaving process, a correlation map of warp and weft density-tension-defect is constructed to quantify the probability model of yarn breakage. .

[0083] in It is dense (roots / 10cm). The parameters are: weft density (threads / 10cm), weaving tension (N), and speed (revolutions / minute). This model is obtained by fitting historical production data using logistic regression or random forest algorithms and is used to predict the risk of yarn breakage under a given combination of parameters.

[0084] Furthermore, the repair action library includes textile-specific repair actions such as segmented temperature profile repair, material feeding sequence adjustment, ratio linkage repair, and tension-density linkage repair. Each repair action clearly defines its applicable scenario, parameter adjustment range, and expected effect.

[0085] Furthermore, version management information records the version number, reason for change, triggering indicator, scope of application, gray scale ratio, and rollback point of each rule, constraint, and template in the knowledge base, ensuring the traceability and iterability of the knowledge base.

[0086] Through the construction of the aforementioned knowledge base, the system transforms the mechanistic knowledge, practical experience, and constraint rules in the field of textile technology into a structured and computable form, providing knowledge support for the automatic generation of subsequent process models.

[0087] Step S20: Collect data and generate input feature vectors

[0088] The system collects structured production data and unstructured quality data, and merges them into a unified input feature vector.

[0089] Optionally, the collection of structured production data includes order requirements, fabric parameters, equipment parameters, and historical batch data. Order requirements include fiber material / ratio, color code, weight, width, delivery date, and quality grade. Fabric parameters include warp and weft density, weave structure, yarn twist, and fabric moisture content. Equipment parameters include the model of the dyeing vat or loom, control window, and control precision. Historical batch data includes a similar order index, success / failure markers, and feedback summaries.

[0090] Specifically, the acquisition of unstructured quality data includes color difference images, defect images, and videos of hand feel and drape. Feature extraction is then performed on the unstructured quality data to transform it into computable structured quality features.

[0091] Extract color difference features from color difference images, including total color difference. Light and dark deviation Red-green deviation , yellow-blue deviation Extract defect features from defect images, including defect type, number of defects, and defect area ratio. Extract appearance features from feel / sag videos, including feel score (0-10 points) and sag index (0-10 points).

[0092] Furthermore, structured production data and structured quality features are fused together to construct a unified input feature vector. The vector is represented as:

[0093]

[0094] in For a subset of order parameters, These are the core material parameters for the fabric. These are core parameters of equipment capability. This is core information for historical batches. This is a structured quality feature vector. Figure 3 It demonstrates the correspondence between the data structure of the input feature vector and the process model.

[0095] Through the above data collection and feature vector generation process, the system expresses order requirements, material properties, equipment capabilities, historical experience, and quality status in a unified computable vector form, providing complete input information for subsequent agent reasoning.

[0096] Step S30: Agent inference generates candidate process models

[0097] The inference module adopts an Agent architecture to perform multi-stage inference and generate a set of candidate process models based on the input feature vectors.

[0098] Optionally, the multi-stage reasoning performed by the reasoning module includes four stages: intent understanding, historical reasoning, template instantiation, and multi-objective trade-offs.

[0099] Specifically, the intent understanding phase transforms order requirements into a set of target constraints and target weights. The system converts explicit customer requirements (such as "superior quality, 7-day delivery time, controllable cost") into a computable set of constraints (e.g., ...). Production cycle ≤ 72h, dye / auxiliary agent usage ≤ 10% of industry average) and target weights ( For quality weight, For the purpose of exchanging options, (As a cost weight). At the same time, the system automatically uncovers implicit requirements by analyzing the purpose of the order and the customer's historical needs. For example, high-end clothing fabric orders implicitly include constraints such as "hand feel score ≥ 8 points and color fastness ≥ 3-4 levels".

[0100] Furthermore, in the historical reasoning stage, the system retrieves historical batches similar to the input feature vector from the textile process knowledge base and generates risk priors. The system uses the cosine similarity method to calculate the similarity between the current order and historical orders:

[0101]

[0102] in, This is the input feature vector for the current order. The input feature vector is the historical order data. The vector dot product and norm calculation are based on the normalized numerical features and the one-hot encoded categorical features.

[0103] Extract successful patterns from historical orders with a similarity ≥ 0.8 (e.g., "When the heating rate is 2℃ / min and the liquor ratio is 1:15, the color difference..."). The system identifies risk points and generates risk priors based on failure modes (e.g., "shrinkage rate exceeds the standard at a heating rate of 3.5℃ / min"). This is used to limit the range of values ​​for key parameters in the process template. For example, for polyester, if historical data indicates that a heating rate > 3℃ / min leads to a high risk of shrinkage, then the risk prior limits the heating rate to ≤ 3℃ / min.

[0104] Furthermore, during the template instantiation phase, the system calls the process template library and performs parameterized instantiation of the process templates based on risk priors. According to the order's material, color code, and quality grade, the system matches suitable templates from the template library, performs parameterized instantiation of each template, and generates candidate process models with different parameter combinations. Each candidate process model includes at least a process segment sequence and at least two process parameters from the following: temperature curve node corresponding to the process segment, heating rate, liquor ratio, pH strategy, feeding sequence, and holding time.

[0105] Furthermore, in the multi-objective trade-off stage, a multi-objective evaluation function is constructed based on quality, cost, and delivery time indicators to evaluate and select the optimal subset from the structured set of candidate process models. The system defines quality loss. Periodic loss Cost losses The Pareto screening algorithm is used to eliminate dominated candidate models, forming the Pareto front set. The formula for selecting the optimal model is:

[0106]

[0107] For example, when the target weight When the system selects the candidate model with the lowest overall loss as the optimal model, it outputs the selection criteria (such as "matching historical successful patterns, satisfying contraction risk constraints, and having the lowest overall loss").

[0108] Through the above multi-stage reasoning process, the reasoning module can comprehensively consider order requirements, historical experience, mechanism constraints and multi-objective trade-offs to generate a set of candidate process models that meet both quality requirements and are feasible, providing high-quality candidates for subsequent constraint verification and parameter adjustment.

[0109] Step S4: Constraint verification and parameter adjustment to obtain an executable process

[0110] The system performs constraint verification on the candidate process model set. When a candidate process model is detected as not meeting the preset process constraints, its parameters are adjusted to obtain an executable process model that meets the preset process constraints. This step is a quality assurance step in the process model generation process, ensuring that the generated process model is executable and of stable quality on the production site.

[0111] Optionally, constraint verification includes equipment capability window constraint verification and process timing logic constraint verification. Constraint verification is a prerequisite for parameter adjustment and is responsible for identifying unexecutable parts and logical conflicts in candidate models.

[0112] Specifically, the equipment capability window constraint verification performs interval verification on at least one parameter of the candidate process model, including temperature range, heating rate, liquor ratio, pH control accuracy, and tension range, based on the production equipment capability. The verification logic is to check whether the value of each parameter is within the equipment control window. A verification report is generated based on the verification results, indicating the parameters that do not meet the constraints and the degree of excess (e.g., "temperature exceeds 5℃", "heating rate exceeds 0.5℃ / min"), for subsequent feasible domain trimming.

[0113] Furthermore, the process timing logic constraint verification is based on a constraint network to verify the dependencies and execution order between process segments, and to verify the timing consistency of material feeding sequence, temperature node interval, and holding time. This includes: timing dependency verification, based on the dyeing process segment dependency constraint network (pretreatment → dyeing → fixing → washing → setting), verifying whether the timing logic of the candidate model satisfies the predecessor-successor constraint; material feeding sequence verification, the system verifies whether the feeding sequence of dyes, auxiliaries, alkalis, and fixing agents in the candidate model conforms to the process specifications; and temperature node interval verification, the system verifies whether the temperature node interval in the candidate model meets the process requirements. A timing conflict report is generated based on the verification results, indicating the conflict type (precondition not met, incorrect material feeding sequence, insufficient temperature node interval) and conflict location (specific process segment), for subsequent timing conflict elimination.

[0114] Optionally, parameter adjustment includes feasible region trimming, temporal conflict elimination, multi-objective conflict repair, and textile-specific linkage repair. Parameter adjustment is the repair step after constraint verification, responsible for adjusting candidate models that do not meet the constraints into executable process models.

[0115] Specifically, feasible region trimming refers to truncating or reducing the excess parameters when the key parameters of a candidate process model exceed the equipment capacity window constraints, so that the key parameters fall within the feasible range. Trimming strategies include: truncating parameters exceeding the boundary to their boundary values; scaling out parameter combinations exceeding the boundary to within the feasible region; and after trimming, re-predicting quality indicators using a mechanistic model to assess the impact of trimming on quality. A trimmed candidate model is generated based on the trimming results, and the trimming operation and expected quality impact are labeled (e.g., "Temperature truncation to 140℃, expected dyeing rate decrease of 2%, recommended to extend the holding time + 10min compensation").

[0116] Furthermore, timing conflict elimination refers to adjusting the material feeding time, temperature node interval, or holding time to eliminate timing conflicts when the process timing logic constraints are not met. Elimination strategies include: if the temperature node interval is insufficient, extending the holding time to 15 minutes; if the material feeding sequence is incorrect, adjusting the feeding sequence to feed dye first and then alkali; if the preconditions are not met, inserting a heating node before the dyeing section; if the process segment sequence is incorrect, reordering the process segments according to the dependency constraint network.

[0117] The candidate model after the timing conflict is eliminated is generated based on the elimination results, and the adjustment operation is marked (e.g., "the heat preservation time is extended from 5 min to 15 min, and the feeding order is adjusted to dye first and then alkali").

[0118] Furthermore, multi-objective conflict remediation refers to generating a Pareto-non-dominant set of remediation actions when there is a conflict between quality, cost, and delivery time objectives, and selecting the remediation action with the smallest evaluation function value based on the multi-objective evaluation function. The system predicts the multi-objective indicators (quality, cost, and delivery time) of candidate models through a mechanistic model, and identifies conflict types and generates candidate remediation actions as follows:

[0119] Quality-Delivery Conflict: Color Difference Meets Standards ( However, the production cycle exceeded the standard (>72 hours). Reason: A low heating rate (1.5℃ / min) and a long holding time (60 minutes) ensured the color difference met the standard, but led to a longer cycle. The proposed repair action is to "increase the heating rate to 2.5℃ / min + shorten the holding time to 45 minutes" (shortening the cycle by 10 hours, expected to improve color difference). Increase by 0.2).

[0120] Quality-Cost Conflict: Color difference meets standards, but dye usage exceeds standards (>10% above industry average), meaning high dye concentration (+15%) is compensating for insufficient dye uptake due to low-temperature dyeing. The proposed remedial action is to "reduce dye concentration to +8% + extend heat treatment time to 50 minutes" (reducing costs by 3%, expected to improve color difference). Increase by 0.1).

[0121] Delivery time-quality conflict: Production cycle meets standard (≤72h) but shrinkage rate exceeds standard (>2.5%). This is because a high heating rate (4℃ / min) shortens the cycle but increases the risk of fiber heat shrinkage. The solution is to "reduce the heating rate to 2℃ / min + staged heating strategy" (extend the cycle by 5h, and the expected shrinkage rate will decrease to 2.2%).

[0122] Furthermore, the system calculates the overall damage of each repair action.

[0123] (in (For the repaired model), weighted by order Select the repair action that minimizes overall loss and execute it.

[0124] Based on the repair results, generate a candidate model after repair and output the basis for repair (e.g., "select the repair action 'reduce the heating rate to 2℃ / min' according to the order weight [0.6, 0.3, 0.1], and the overall loss is reduced from 0.55 to 0.42").

[0125] Furthermore, the textile-specific linkage repair function executes corresponding repair actions from the repair action library based on the detected quality anomaly type. This step is the core innovation of parameter adjustment, enabling precise linkage repair of specific quality problems (shrinkage, color difference, yarn breakage, defects) in textile dyeing and weaving.

[0126] The repair actions should include at least: segmented temperature profile repair when shrinkage risk is detected; dye ratio-linked repair when color difference exceeds the standard; and density and tension-linked repair when yarn breakage or defects exceed the standard. Among these:

[0127] Segmented temperature profile repair (for shrinkage risk): at the critical temperature The conventional heating rate (2-3℃ / min) was used previously. Then, a low heating rate (1-1.5℃ / min) was adopted, and the dyeing rate loss was calculated by using a dyeing kinetic model. The holding time was extended or the dye concentration was increased to compensate for the dyeing rate.

[0128] Dye ratio linkage repair (for excessive color difference): based on , , The direction of deviation determines the complementary color strategy, and the amount of fixing agent and pH value are adjusted in conjunction with the adjustment.

[0129] Density and tension linkage repair (for yarn breakage or excessive defects): Within the allowable range of fabric strength constraints, weft density, tension, machine speed, and twist are adjusted in a linked manner. Furthermore, the type of quality anomaly is determined based on the quality backtracking mechanism and production feedback data from production execution. The quality backtracking mechanism constructs a multi-layered correlation network based on a quality correlation graph, encompassing quality characteristics, process segments, key parameters, and repair actions.

[0130] When production feedback indicates a quality abnormality (e.g., color difference is found during finished product inspection) (Exceeding the standard), the system uses quality correlation graphs to perform reverse reasoning and generates the following chain of evidence:

[0131] Quality Feature Node Localization: Identifying Abnormal Quality Feature "Color Difference" "Exceeding the standard";

[0132] Process segment tracing: By using the edge of the pattern "color difference → dyeing segment", the abnormal process segment is located as "dyeing segment";

[0133] Key parameter identification: By identifying the nodes of "staining segment → temperature curve", "staining segment → pH value", and "staining segment → dye concentration" on the graph, the key parameter is "low dye concentration".

[0134] Recommended repair actions: Based on the graph's edge indicators "Dye concentration too low → increase dye concentration" and "Dye concentration too low → extend insulation time", the recommended repair action is "increase dye concentration + extend insulation time".

[0135] The chain of evidence should include at least the abnormal process segment, the corresponding key parameters, the type of quality anomaly, and the recommended corrective action, which is used to attribute the quality anomaly.

[0136] Through the above constraint verification and parameter adjustment process, the system can ensure that the generated process model meets the constraints of equipment capacity, timing logic and multi-objective trade-offs, improve the executability and quality stability of the process plan, and reduce the number of parameter adjustments and the occurrence rate of quality anomalies in the production process.

[0137] Step S5: Output the executable process model to the production execution system.

[0138] Optionally, the executable process model is output in the form of structured data, including complete process parameters such as process segment sequence, temperature curve nodes, heating rate, liquor ratio, pH strategy, feeding sequence, and holding time, as well as expected quality indicators (such as...). The system includes risk warning rules (such as shrinkage rate and defect rate) and risk warning rules (warning trigger conditions, risk type, and recommended actions). Specifically, after receiving the executable process model, the Production Execution System (MES) controls the actual operation of the dyeing vat or weaving machine according to the preset process segment sequence and parameter settings.

[0139] During production, the system collects actual process curves (temperature, pH value, tension, etc.) and quality data in real time, and compares them with the preset curves of the candidate model, including:

[0140] Temperature curve comparison: Real-time acquisition of dye vat temperature (sampling frequency 1 time / minute), plotting a comparison graph of the actual temperature curve and the preset curve, and calculating the deviation. , if the deviation If the condition persists for more than 5 minutes, an alarm will be triggered.

[0141] pH curve comparison: Real-time pH value acquisition (sampling frequency 1 time / 5 minutes), comparing the actual pH with the preset pH target value, if deviation... If so, an alarm will be triggered;

[0142] Tension curve comparison: Real-time acquisition of loom tension (sampling frequency 1 time / second), comparison of actual tension with preset tension, if deviation... If so, an alarm will be triggered;

[0143] When the actual parameters enter the risk zone (e.g., polyester temperature > 125℃ and heating rate > 2.5℃ / min enters the shrinkage risk zone), the system immediately issues an early warning and automatically triggers local repair actions.

[0144] Through the aforementioned process model output and dynamic adjustment mechanism, the system achieves seamless integration from process design to production execution, and provides real-time risk warning and dynamic repair capabilities during production, thereby improving the adaptability of the process scheme and the stability of the production process.

[0145] Step S6: Update the textile process knowledge base based on production feedback.

[0146] The system updates the textile process knowledge base based on production feedback for subsequent process model generation. This step is a closed-loop iterative process of the intelligent design system, ensuring that the knowledge base can continuously evolve with production practice and adapt to changes in materials, equipment upgrades, and quality fluctuations.

[0147] Optionally, version updates are performed when textile-specific trigger conditions are met. Textile-specific trigger conditions include at least: color number fluctuation trigger, chemical fiber shrinkage trigger, yarn breakage trigger, defect exceeding the standard trigger, new material trigger, and new equipment trigger.

[0148] Specifically, color number fluctuation triggering refers to the color difference in three consecutive batches of the same color number. The variance is >0.5 (significant fluctuation), and external factors such as equipment failure and batch differences in raw materials are excluded. After triggering, the system updates the mixing template (adjusting the dye concentration vector) and sensitive parameter constraints for that color number.

[0149] Specifically, chemical fiber shrinkage triggering refers to two consecutive batches of chemical fiber (polyester, acrylic) fabrics having a shrinkage rate ≥3% (exceeding the standard), and the process parameters meeting the requirements of the current model. After triggering, the system refits the heat shrinkage model parameters. , , Update the temperature curve segment constraints and repair action priority.

[0150] Specifically, yarn breakage / defect triggering refers to a yarn breakage rate ≥0.08 or a defect rate ≥1.5% within a statistical period (e.g., 7 days), exceeding the historical average by 50%. After triggering, the system updates the edge weights and thresholds of the density-tension-defect graph and adjusts the tension limit and speed limit of the weaving template.

[0151] Specifically, new material triggers refer to the introduction of new fiber materials, dye types, or fabric structures where no corresponding templates or mechanism model parameters exist in the knowledge base. New equipment triggers refer to the introduction of new dyeing vat models, loom models, or control systems, requiring updates to equipment capability window constraints. New material updates involve creating new process templates for the new material (initializing based on similar material templates), supplementing dyeing kinetic model parameters for the new dye type (determined through small-scale experiments), and supplementing density-tension correlation rules for the new fabric structure. New equipment updates involve creating equipment capability window constraints for the new equipment (based on equipment manuals and measured data), updating the mapping relationship between equipment models and constraints, and ensuring that subsequent orders correctly match the constraints of the new equipment.

[0152] Further, when any one of the textile-specific triggering conditions is met, versioned updates are performed, including updating the version number, triggering reason, modified object, scope of application, and expected effects, and recording version logs; gray-box verification is carried out, where the new version is first applied to a small proportion of orders (such as 20% of orders with the same material / color number), and key quality indicators (shrinkage rate, color difference, yarn breakage rate) are continuously monitored.

[0153] Further, when the quality indicators during gray-box verification deteriorate beyond a preset threshold relative to the baseline version, roll back to the previous stable version. The quality indicators at least include color difference indicators, defect indicators, and shrinkage rate indicators.

[0154] Through the above versioned update mechanism, the system realizes the continuous optimization and iteration of the knowledge base with production feedback, enabling the mechanism model, association graph, rule thresholds, and process templates to adapt to material changes, equipment upgrades, and quality fluctuations, ensuring the long-term stability and accuracy of the process plan, and reducing the risk of model drift and the probability of rule failure.

[0155] In this embodiment, by constructing a textile-specific knowledge base, integrating structured and unstructured data, using Agent multi-stage reasoning to generate candidate process models, performing multi-level constraint verification and linkage repair, establishing a quality traceability mechanism, and implementing versioned closed-loop iteration, the intelligent design of textile dyeing and weaving processes is achieved. Compared with the prior art:

[0156] The executability and consistency of the process plan are significantly improved: the non-executability rate of the process plan is reduced, and the quality fluctuations across devices and batches are reduced, ensuring the quality consistency of the same color number on different batches and different devices;

[0157] Computable associations between unstructured quality data and process parameters are realized: through the feature extraction of color difference images, defect images, and handfeel videos and the construction of a quality association graph, unstructured quality data is transformed into computable structured features, shortening the time for locating quality anomalies and improving the repair accuracy;

[0158] The knowledge base is continuously optimized with production feedback: through the versioned update and gray-box verification mechanisms, the mechanism model parameters, association graph edge weights, and rule thresholds of the knowledge base are continuously calibrated with production data, reducing the risk of model drift and the probability of rule failure;

[0159] The process design cycle is significantly shortened: the process design cycle is shortened from 4 - 6 hours to within 30 minutes, effectively improving the first-class product rate and reducing production costs.

[0160] The second embodiment of this application relates to an intelligent design system for textile dyeing and weaving processes based on Agent, combined with Figure 2Through modular design, the system achieves full-process automation from data acquisition, knowledge management, reasoning and decision-making, constraint verification, production integration to closed-loop iteration.

[0161] Optionally, the system architecture modules include a data processing module, a knowledge base module, an inference module, a constraint verification and parameter adjustment module, a production docking module, and a closed-loop iteration module.

[0162] Specifically, the data processing module is responsible for the fusion of structured and unstructured data. This module receives three types of data sources from the data input layer, cleans, normalizes, and extracts features from the structured data, and extracts features (color difference, defect, and appearance features) from the unstructured quality data, transforming the unstructured data into structured quality features. Finally, the data processing module fuses the structured production data with the structured quality features to generate a unified input feature vector, which is then output to the inference module and the constraint verification and parameter adjustment module.

[0163] Specifically, the knowledge base module is responsible for building and maintaining the textile process knowledge base, including a mechanism model library, a process template library, a constraint network, a quality correlation graph, a repair action library, and version management information. The knowledge base module provides the inference module with mechanism models (fiber heat shrinkage models, dye uptake kinetic models, etc.) and process templates for the generation and parameter instantiation of candidate process models. The knowledge base module also provides the constraint verification and parameter adjustment module with constraint networks (equipment capacity window constraints, process timing logic constraints) and a repair action library for constraint verification and parameter adjustment. Furthermore, the knowledge base module receives version update instructions from the closed-loop iteration module and iteratively optimizes mechanism model parameters, graph parameters, rule thresholds, and process templates based on production feedback.

[0164] Specifically, the inference module employs an Agent architecture, executing multi-stage inference. The inference module receives input feature vectors from the data processing module, calls upon the mechanism models and process templates in the knowledge base module, and generates a set of candidate process models following a four-stage inference process: intent understanding → historical inference → template instantiation → multi-objective trade-offs. In the intent understanding stage, order requirements are converted into a set of target constraints and target weights. In the historical inference stage, similar historical batches are retrieved from the knowledge base, and risk priors are generated to limit the value range of key parameters for the process templates. In the template instantiation stage, the process template library is called, and parameterized instantiation is performed based on the risk priors, generating a structured set of candidate process models. In the multi-objective trade-offs stage, a multi-objective evaluation function is constructed based on quality indicators, cost indicators, and delivery time indicators, and Pareto selection is performed on the candidate process model set to obtain a preferred subset. The inference module outputs the generated set of candidate process models to the constraint verification and parameter adjustment module.

[0165] Specifically, the constraint verification and parameter adjustment module performs constraint verification on the candidate process model set. When a candidate process model does not meet the preset process constraints, it calls the repair action library to adjust the parameters of the candidate process model, resulting in an executable process model that meets the preset process constraints. This module calls the constraint network and repair action library in the knowledge base module to perform equipment capability window constraint verification and process timing logic constraint verification. When constraint conflicts are detected, the module performs parameter adjustments in the order of feasible domain trimming → timing conflict elimination → multi-objective conflict repair → textile-specific linkage repair. Textile-specific linkage repair determines the type of quality anomaly based on the quality backtracking mechanism and calls the corresponding repair actions (temperature curve segment repair, dye ratio linkage repair, density and tension linkage repair, etc.). The constraint verification and parameter adjustment module outputs the final executable process model to the production docking module.

[0166] Specifically, the production integration module is responsible for outputting the executable process model to the Manufacturing Execution System (MES) and establishing an interface with the MES system. The production integration module sends the executable process model to the MES system in structured data format (including complete process parameters such as process segment sequences, temperature curve nodes, heating rates, liquor ratios, pH strategies, feeding sequences, and holding times) through standard interfaces (such as OPCUA and RESTful API). After receiving the process model, the MES system sends the parameter settings to the PLC controllers of the dyeing vats or looms on-site, realizing the automatic loading and execution of process parameters. The production integration module is also responsible for collecting real-time production process data (actual temperature curves, actual tension curves, real-time quality inspection data, etc.) from the MES system and sending the production feedback data back to the closed-loop iteration module.

[0167] Specifically, the closed-loop iteration module is responsible for versioning the textile process knowledge base based on production feedback, thus achieving version management. The closed-loop iteration module receives production feedback data from the production docking module, including batch quality inspection results (color difference, shrinkage rate, defect rate, etc.), yield, rework records, etc. The module determines whether a version update needs to be triggered based on preset textile-specific trigger conditions (color number fluctuation trigger, chemical fiber shrinkage trigger, yarn breakage trigger, defect exceeding standard trigger, new material trigger, new equipment trigger). When the trigger conditions are met, the module generates a new version and performs gray-scale verification. The closed-loop iteration module sends update instructions to the knowledge base module, modifies the mechanism model parameters, graph parameters, rule thresholds, or process templates, and records the version log. During gray-scale verification, the module continuously monitors key quality indicators. When quality indicators stabilize or improve, the application ratio is gradually expanded; when quality indicators deteriorate beyond a preset threshold, it immediately rolls back to the previous stable version.

[0168] Optionally, the system output layer includes four types of output: executable process model, early warning rules, quality backtracking evidence chain, and version update record.

[0169] Specifically, the executable process model output includes a dyeing process route model (process sequence, temperature curve nodes / rates, pH strategy, feeding sequence, concentration vector) or a weaving parameter model (warp and weft density, tension / speed strategy, twist strategy). The early warning rule output includes early warning trigger conditions (e.g., temperature > 125℃ and heating rate > 2.5℃ / min), risk type (shrinkage risk / color difference risk / yarn breakage risk), and suggested actions (parameter adjustment direction and range). The quality backtracking evidence chain output includes a complete mapping relationship between quality characteristics → process segment → key parameters → repair strategy, as well as supporting evidence (historical cases, mechanism prediction). The version update record output includes version number, trigger reason, modified object, applicable scope, grayscale verification results, and rollback information.

[0170] Furthermore, the system output layer transmits the above four types of outputs to the Manufacturing Execution System (MES), Quality Management System (QMS), and Enterprise Resource Planning System (ERP) through system interfaces, achieving multi-system collaboration. Executable process models are sent to the MES system for production execution, early warning rules are sent to the MES system for real-time monitoring and alarms, quality backtracking evidence chains are uploaded to the QMS system for quality traceability and analysis, and version update records are uploaded to the ERP system for knowledge asset management and auditing.

[0171] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any ordinary changes and substitutions made by those skilled in the art within the scope of the technical solution of the present invention should be included within the protection scope of the present invention.

Claims

1. An agent-based intelligent design method for textile dyeing and weaving processes, characterized in that, Includes the following steps: S1, Construct a textile technology knowledge base; S2 collects structured production data and unstructured quality data, extracts features from the unstructured quality data and fuses them with the structured production data to generate an input feature vector; S3, based on the input feature vector, generates a set of candidate process models through the inference module; S4, Perform constraint verification on the candidate process model set; When a candidate process model is detected as not meeting the preset process constraints, the parameters of the candidate process model are adjusted to obtain an executable process model that meets the preset process constraints. S5 outputs the executable process model to the production execution system for production execution; S6 updates the textile process knowledge base in a versioned manner based on production feedback for use in subsequent process model generation.

2. The agent-based intelligent design method for textile dyeing and weaving processes according to claim 1, characterized in that, The textile technology knowledge base includes: Mechanism model library, process template library, constraint network, quality correlation graph, repair action library, and version management information; The mechanistic model library includes at least fiber thermal shrinkage models and dye uptake kinetic models; Version management information should include at least the version identifier, scope of application, triggering reason, and rollback identifier.

3. The agent-based intelligent design method for textile dyeing and weaving processes according to claim 2, characterized in that, S2 includes: Collect structured production data, which includes order requirements, fabric parameters, equipment parameters, and historical batch data; Collect unstructured quality data, including color difference images, defect images, and videos of hand feel and drape. Feature extraction is performed on unstructured quality data, including extracting color difference features from color difference images, extracting defect features from defect images, and extracting appearance features from videos of hand feel and drape, in order to obtain structured quality features. The structured production data and structured quality features are fused together to obtain the input feature vector.

4. The agent-based intelligent design method for textile dyeing and weaving processes according to claim 3, characterized in that, The inference module uses an agent architecture and performs multi-stage inference, including: Intent understanding: Transform order requirements into a set of target constraints and target weights; Historical reasoning: Retrieve historical batches similar to the input feature vector from the textile process knowledge base and generate risk priors. Risk priors are used to limit the range of values ​​for key parameters of the process template. Template instantiation: Call the process template library and parameterize the process template based on risk priors to generate a structured set of candidate process models; Multi-objective trade-off: Based on quality indicators, cost indicators, and delivery time indicators, a multi-objective evaluation function is constructed to evaluate and select the preferred subset of the structured candidate process model set; The candidate process model includes at least two process parameters from the following: process segment sequence, temperature curve node corresponding to the process segment, heating rate, bath ratio, pH strategy, feeding sequence, and holding time.

5. The agent-based intelligent design method for textile dyeing and weaving processes according to claim 4, characterized in that, The constraint verification in S4 includes: Equipment capacity window constraint verification: Perform interval verification on at least one of the following parameters of the candidate process model according to the production equipment capacity: temperature range, heating rate, liquor ratio, pH control accuracy, and tension range. Process timing logic constraint verification: Based on the constraint network, verify the dependencies and execution order between process segments, and verify the timing consistency of feeding order, temperature node interval and heat preservation time.

6. The agent-based intelligent design method for textile dyeing and weaving processes according to claim 5, characterized in that, The parameter adjustment in S4 includes: Feasible domain pruning: When the key parameters of the candidate process model exceed the equipment capacity window constraints, the excess part is truncated or recycled so that the key parameters fall into the feasible region. Timing conflict elimination: When the process timing logic constraints are not met, adjust the feeding time point, temperature node interval or heat preservation time to eliminate timing conflicts; Multi-objective conflict resolution: When there is a conflict between quality objectives, cost objectives and delivery objectives, a set of Pareto non-dominant resolution actions is generated, and the resolution action with the smallest evaluation function value is selected and executed according to the multi-objective evaluation function. Textile-specific linkage repair: Based on the detected quality anomaly type, the corresponding repair action in the repair action library is called and executed. The repair actions include at least: temperature curve segment repair when shrinkage risk is detected, dye ratio linkage repair when color difference exceeds the standard, and density and tension linkage repair when yarn breakage or defects exceed the standard. The types of quality anomalies are determined based on the quality backtracking mechanism and production feedback data from production execution.

7. The agent-based intelligent design method for textile dyeing and weaving processes according to claim 6, characterized in that, The quality backtracking mechanism includes: constructing a multi-layered correlation network based on quality correlation graphs, encompassing quality characteristics, process segments, key parameters, and repair actions; when production feedback indicates a quality anomaly, generating an evidence chain based on the multi-layered correlation network, the evidence chain including at least the abnormal process segment, corresponding key parameters, quality anomaly type, and recommended repair actions, for attributing the quality anomaly to its cause; In S4, the key parameters that need to be adjusted and the corresponding repair actions are determined based on the chain of evidence. In S6, the chain of evidence is written into the version management information as a version update record.

8. The agent-based intelligent design method for textile dyeing and weaving processes according to claim 7, characterized in that, The versioned update in S6 is executed when the textile-specific triggering condition is met: The specific triggering conditions for textiles include at least: color number fluctuation triggering, chemical fiber shrinkage triggering, yarn breakage triggering, defect exceeding the standard triggering, new material triggering, and new equipment triggering. When any textile-specific triggering condition is met, a version update is performed, a new version is generated and gray-scale verification is carried out. When the quality indicators during the gray-scale verification period deteriorate more than a preset threshold relative to the baseline version, the version is rolled back to the previous stable version. The quality indicators include at least color difference indicators, defect indicators, and shrinkage rate indicators.

9. An agent-based intelligent design system for textile dyeing and weaving processes, characterized in that, include: The knowledge base module is used to build and maintain a textile technology knowledge base; The data processing module is used to collect structured production data and unstructured quality data, and generate input feature vectors; The inference module is used to generate a set of candidate process models based on the input feature vector. The inference module adopts an Agent architecture. The constraint verification and parameter adjustment module is used to perform constraint verification on the candidate process model set, and when the candidate process model does not meet the preset process constraints, it calls the repair action library to perform parameter adjustment on the candidate process model to obtain an executable process model that meets the preset process constraints. The production docking module is used to output the executable process model to the production execution system; The closed-loop iteration module is used to update the textile process knowledge base in a versioned manner based on production feedback.

10. An agent-based intelligent design system for textile dyeing and weaving processes, characterized in that, include: Memory is used to store executable instructions for a computer; as well as, A processor configured to implement the steps of the method as described in any one of claims 1 to 8 when executing the computer-executable instructions.

Images