Fabric defect real-time detection and adaptive correction method based on deep learning

Abstract

The invention discloses a fabric defect real-time detection and self-adaptive correction method based on deep learning, and relates to the technical field of textile production and quality control, and the method comprises the steps: carrying out the structural processing of data from different sources, and forming a unified technology and standard knowledge set; generating corresponding correction constraint conditions or adjustment suggestions; an alternative process scheme is automatically generated and recommended; performing matching verification on the process parameters, the detection items and the quality threshold; outputting defect types, defect positions and defect severity; and predicting the potential quality risk, and generating corresponding risk early warning information. According to the method, the problems of non-standard fabric process parameters, defect identification lagging and difficulty in quality tracing in the prior art are solved, and process optimization, real-time quality monitoring and cross-process tracing are realized through construction of the process and standard knowledge base and process feasibility analysis and defect real-time identification based on deep learning; and the production efficiency and the product quality are improved.

Description

Technical Field

[0001] This invention relates to the field of textile production and quality control technology, specifically to a method for real-time detection and adaptive correction of fabric defects based on deep learning. Background Technology

[0002] In textile production, traditional fabric quality inspection mainly relies on manual visual inspection or simple automated inspection equipment. This typically only identifies obvious defects, lacks adaptability to complex textures or different fabric types, and suffers from low efficiency and poor real-time performance. In recent years, some fabric defect detection systems based on image analysis or automated equipment have emerged; however, these systems still suffer from limited defect identification accuracy, insufficient process and fabric matching analysis, weak quality traceability and cross-process correlation capabilities, and difficulty in automatically verifying environmental and safety standards.

[0003] Therefore, existing technologies are insufficient to achieve high-precision, real-time defect identification and process optimization throughout the entire production process, resulting in fabric defects that cannot be detected and adjusted in a timely manner, increasing rework costs and affecting the quality of the final product. Summary of the Invention

[0004] This application provides a deep learning-based real-time fabric defect detection and adaptive correction method to address the problems in existing technologies where fabric quality inspection relies on post-inspection or human experience, lacks comprehensive analysis of fabric properties, process parameters and equipment capabilities, and is difficult to achieve linkage correction between defect results and process configuration and production batches. As a result, fabric defects cannot be identified and corrected in a timely manner during the production process, leading to the accumulation of quality risks, increased rework costs and difficulties in quality traceability.

[0005] In view of the above problems, this application provides a method for real-time detection and adaptive correction of fabric defects based on deep learning.

[0006] This application provides a deep learning-based method for real-time detection and adaptive correction of fabric defects. The method includes: constructing an enterprise process specification library, a fabric and accessory specification library, and a cost parameter library; and performing structured processing on data from different sources to form a unified set of process and standard knowledge; after receiving demand information related to fabric type, process parameters, and production tasks, performing compliance verification on the demand information based on preset rules and the set of process and standard knowledge; when parameters are detected to be inconsistent with preset standards, generating corresponding correction constraints or adjustment suggestions; based on the process parameters adjusted according to the correction constraints or passed compliance verification, performing a matching degree analysis on the fabric physical properties, process requirements, and production equipment capabilities using a deep learning model to assess the feasibility of process implementation and outputting corresponding process feasibility assessment results; based on the process feasibility assessment results, when a process feasibility score is detected to be lower than a preset threshold or a high-risk process combination exists, automatically generating and recommending alternative process solutions; and determining the feasibility of the process... Based on the evaluation results, the system automatically links to international environmental and safety standards libraries. According to the certification constraints associated with production tasks, it matches and verifies process parameters, testing items, and quality thresholds, and feeds back the certification constraint results to the process parameter configuration process. For fabric image data collected at the production site, it performs defect identification based on a local deep learning inference model, outputting defect type, location, and severity. In cases where the network is unavailable or unstable at the production site, the defect identification results and corresponding image data are stored in a local database for caching, and incremental synchronization is performed after the network is restored. Based on historical quality inspection data accumulated from the defect identification results and synchronously collected production environment data, a time-series analysis model is constructed to predict potential quality risks and generate corresponding risk warning information. The defect identification results and quality risk prediction results are automatically linked to the corresponding process sheets and production batches to achieve cross-process traceability of quality problems, and the analysis results are synchronized to the production management module and trade management module.

[0007] One or more technical solutions provided in this application have at least the following technical effects or advantages: This application constructs an enterprise process specification library, a fabric and accessory specification library, and a cost parameter library, and performs structured processing on multi-source heterogeneous data to form a unified set of process and standard knowledge. This achieves centralized management and standardized expression of process rules, material specifications, and cost parameters, providing a consistent data foundation for subsequent process verification, feasibility analysis, and quality control. Based on the aforementioned process and standard knowledge set, the system performs compliance verification on the fabric type, process parameters, and requirement information input in production tasks. When parameters do not meet preset standards, it generates corrective constraints or adjustment suggestions, enabling unreasonable process configurations to be identified and corrected before production, reducing production risks caused by improper parameter configuration. Based on compliance verification or parameter adjustment, this application uses a deep learning model to model and analyze the matching relationship between fabric physical properties, process requirements, and production equipment capabilities, outputting process feasibility assessment results. This achieves a quantitative assessment of process implementation feasibility, avoiding reliance solely on human experience for judgment. Furthermore, when a process feasibility score is detected to be below a preset threshold or a high-risk process combination exists, the system can automatically generate and recommend alternative process solutions based on the process and standard knowledge set, providing clear data support and an executable path for process optimization. Based on the feasibility assessment results of the process, this application introduces international environmental and safety standards into the production process. According to the certification constraints associated with the production tasks, process parameters, testing items, and quality thresholds are matched and verified, and the certification constraint results are fed back to the process parameter configuration process. This achieves automatic correlation between process configuration and certification requirements, reducing the risk of omissions caused by manual comparison. During production, this application uses fabric image data collected on the production site and a deep learning inference model deployed on an edge computing device to identify fabric defects in real time, outputting the defect type, location, and severity, enabling timely detection and recording of fabric defects during production. For production environments with unavailable or unstable networks, a local database is used to cache the defect identification results and corresponding image data, and incremental synchronization is performed after the network is restored to ensure the continuity and integrity of quality inspection data. Furthermore, this application constructs a time-series analysis model based on historical quality inspection data accumulated from defect identification results and synchronously collected production environment data to predict potential quality risks, thereby achieving early warning of quality problems. By automatically linking defect identification results and quality risk prediction results with the corresponding process sheets and production batches, and synchronizing them to the production management module and trade management module, cross-process traceability of quality problems and sharing among multiple business modules are realized, thereby improving the synergy and response efficiency of production management and quality control. Attached Figure Description

[0008] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0009] Figure 1 This is a schematic diagram of a method for real-time detection and adaptive correction of fabric defects based on deep learning, provided in an embodiment of this application. Detailed Implementation

[0010] This application provides a deep learning-based real-time fabric defect detection and adaptive correction method to address the lack of a comprehensive intelligent analysis and real-time feedback mechanism for process parameter compliance, process feasibility, and production quality in existing fabric production processes. This leads to problems such as delayed fabric defect detection, reliance on manual experience for process adjustments, difficulty in timely prediction of quality risks, and insufficient cross-process quality traceability. This application constructs a unified set of process and standard knowledge by building an enterprise process specification library, a fabric and accessory specification library, and a cost parameter library. Before production tasks are executed, automatic compliance checks are performed on fabric type, process parameters, and certification constraints, reducing the risk of unreasonable process parameters entering the production process from the source. Simultaneously, a deep learning model is used to analyze the matching degree between fabric physical properties, process requirements, and production equipment capabilities, outputting process feasibility assessment results. When high-risk process combinations are detected, alternative process solutions are automatically generated, achieving intelligent optimization and adaptive correction of process solutions. During the production process, this application utilizes local deep learning inference on fabric image data collected at the production site to achieve real-time identification and quantitative description of fabric defects. Under network anomaly conditions, local caching and incremental synchronization mechanisms ensure the integrity and continuity of quality inspection data. Furthermore, by performing time-series analysis on historical quality inspection data and production environment data, potential quality risks are predicted in advance, and risk warning information is generated, realizing a shift from post-event detection to pre-event prevention in quality control. By automatically linking defect identification results and quality risk prediction results with process sheets and production batches, this application achieves cross-process traceability and data linkage for quality problems, and synchronizes the analysis results to the production management module and trade management module, thereby achieving the technical effects of quality controllability, process adjustment efficiency, and overall production management level in the fabric production process.

[0011] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the scope of protection of this application.

[0012] It should be noted that any variation of the terms "comprising" and "having" is intended to cover non-exclusive inclusion, for example, a process, method, system, product, or server that includes a series of steps or units is not necessarily limited to those steps or units that are explicitly listed, but may include other steps or modules that are not explicitly listed or that are inherent to such processes, methods, products, or devices.

[0013] Examples, such as Figure 1 As shown, this application provides a deep learning-based method for real-time detection and adaptive correction of fabric defects, the method comprising: S100: Construct enterprise process specification library, raw material and auxiliary material specification library, and cost parameter library, and perform structured processing on data from different sources to form a unified set of process and standard knowledge.

[0014] Furthermore, step S100 of this application also includes: S101: Obtain process specification data from the enterprise's existing process documents, historical production records, and process management system, and split the process specification data according to preset process element fields to generate a set of process parameter items; S102: Obtain fabric and auxiliary material specification data from fabric and auxiliary material supply information, test reports, and material management system, and perform field standardization processing on the material type, physical property indicators, and applicable process range of fabric and auxiliary materials to generate a set of fabric and auxiliary material specification parameters; S103: Obtain cost parameter data from production cost accounting records and procurement records, and perform unified unit conversion and interval processing on the cost parameter data to generate a set of cost parameters; S104: Perform field mapping and semantic alignment processing on the set of process parameter items, the set of fabric and auxiliary material specification parameters, and the set of cost parameters to establish cross-database association identifiers; S105: Write the set of process parameter items after completing field mapping and semantic alignment processing into the process specification library, write the set of fabric and auxiliary material specification parameters into the fabric and auxiliary material specification library, and write the set of cost parameters into the cost parameter library.

[0015] Specifically, process specification data is obtained from the company's existing process documents, historical production records, and process management system. This data is then broken down according to pre-defined process element fields, which include at least the process name, process parameter range, applicable fabric type, and equipment requirements, generating a structured set of process parameter items. This step transforms process experience, originally stored in text or unstructured form, into a calculable and retrievable data format. Fabric and accessory specification data is obtained from supply information, testing reports, and the material management system. The material type, physical property indicators, and applicable process range of the fabrics and accessories are standardized, including but not limited to weight, thickness, density, and shrinkage parameters, generating a unified set of fabric and accessory specification parameters. Standardizing the fields of fabric and accessory data from different sources avoids matching errors caused by inconsistent data formats. Cost parameter data is obtained from production cost accounting records and procurement records. This cost parameter data undergoes unified unit conversion and interval processing, including at least material costs, process costs, and equipment usage costs, forming a set of cost parameters that can be used for subsequent process scheme evaluation. The process parameter set, fabric and accessory specification parameter set, and cost parameter set are processed by field mapping and semantic alignment to establish cross-database association identifiers, thereby realizing logical connections between different data sets. For example, the "applicable fabric type" field in the process specification is semantically mapped to the "material type" field in the fabric and accessory specification parameters, thus forming a data structure that can be directly used for rule verification and model analysis. The process parameter set after field mapping and semantic alignment is written into the process specification library, the fabric and accessory specification parameter set is written into the fabric and accessory specification library, and the cost parameter set is written into the cost parameter library. The knowledge set supports the automatic analysis and retrieval of process compliance, process feasibility, and quality risks in subsequent steps. The cross-database association identifier includes a unified process number, material number, and version identifier used to associate process specifications, fabric and accessory specifications, and cost parameters. Through the above steps, this embodiment realizes unified modeling and structured management of multi-source, heterogeneous process data within an enterprise, providing a stable and reusable data foundation for subsequent deep learning-based process evaluation and fabric defect detection, avoiding the uncertainty caused by manual table lookups or experience-based judgments, and improving the standardization and consistency of process decisions.

[0016] S200: After receiving the demand information related to fabric type, process parameters and production task, the demand information is verified for compliance based on preset rules and the process and standard knowledge set. When the parameters are detected to be inconsistent with the preset standards, corresponding correction constraints or adjustment suggestions are generated.

[0017] Furthermore, step S200 of this application also includes: S201: Parse and process the received demand information, breaking it down into fabric type information, process parameter information, and demand constraint information. Perform format verification and standardization on each type of information to generate a unified data description format. S202: Using the unified data description format, retrieve process specification data, fabric and accessory specification data, and cost parameter data corresponding to the fabric type and production task from the process and standard knowledge set to construct a set of parameters to be verified. S203: Based on preset compliance verification rules, perform item-by-item comparison and analysis on the set of parameters to be verified to identify whether the process parameters meet the corresponding process specifications, fabric and accessory specifications, and cost constraints. S204: When at least one process parameter is detected as not meeting the preset standard, generate a corresponding parameter deviation identifier and determine the target parameter to be adjusted and its constraint range based on the parameter deviation identifier. S205: Based on the target parameter and its constraint range, generate corrective constraints or adjustment suggestions to guide subsequent process parameter adjustments and output the corrective constraints or adjustment suggestions to the subsequent process feasibility analysis process.

[0018] Specifically, the system receives fabric type information, process parameter information, and demand constraint information associated with the production task. The demand constraint information includes delivery requirements, cost limits, and certification constraints. The received information undergoes format validation and field standardization, converting data from different sources and with different expressions into a unified data description format for subsequent automated processing. The fabric type information includes at least: fabric material type (e.g., cotton, polyester, silk, recycled fiber), weave structure (e.g., plain weave, twill, satin), weight (g / m²), width (cm), and nominal shrinkage rate (%). The process parameter information includes at least: sewing method, stitch length, quilting method, cutting shrinkage rate, dyeing and finishing temperature, and drying time. The demand constraint information includes at least: production batch number, delivery cycle requirement, target cost range, and certification constraint identifier. The system performs format validation and field standardization on the above information. For example, it maps fields representing "cutting shrinkage rate" from different sources to the standard field "cutting_shrinkage_rate" and converts the values ​​to percentages, generating a unified data description format. Based on this unified data description format, it retrieves process specification data, fabric and accessory specification data, and cost parameter data matching the fabric type and production task from the process and standard knowledge set. This data is then combined with the process parameter information to construct a set of parameters to be validated. This includes: a set of fabric and accessory specification parameters corresponding to the fabric type, which at least includes the allowable shrinkage rate range, temperature resistance threshold, and a set of compatible process types; a set of process specification parameters corresponding to the process type, which at least includes the recommended stitch length range, available quilting methods, and process prohibition rules; and a set of cost parameters corresponding to the production cost target, which at least includes the upper limit of single-piece cost, the mapping relationship between process complexity level and cost factor. According to preset compliance validation rules, the set of parameters to be validated is compared and analyzed item by item to determine whether the process parameters meet the corresponding process specification requirements, fabric and accessory specification limitations, and cost constraints. The compliance verification rules may include parameter range verification, logical consistency verification, and process adaptability verification. At a minimum, they include: range constraint rules: determining whether process parameters fall within the corresponding specification's allowable range; logical constraint rules: determining whether a specific fabric type allows the corresponding process method; and cost constraint rules: determining whether the selected process combination exceeds the target cost range. For example, when the fabric type is "recycled denim," the system automatically calls its corresponding recommended shrinkage rate range (2.5%–3.5%) and compares the input cutting shrinkage rate parameter with this range. When at least one process parameter is detected as not meeting the preset standard, the system generates a parameter deviation identifier for that parameter. The parameter deviation identifier includes at least: the name of the deviation parameter; the current parameter value; the standard allowable range; and the deviation type (exceeding limits, mismatch, or conflict).In the example above, when the input cutting shrinkage rate is 1.5%, the system identifies it as being below the lower limit of the allowable range and determines "cutting shrinkage rate" as the target adjustment parameter, while simultaneously determining its adjustable constraint range to be 2.5%–3.5%. Based on the target parameter and its constraint range, corrective constraints or adjustment suggestions are generated to guide subsequent process parameter adjustments. These corrective constraints include at least: the direction of parameter adjustment; the recommended parameter range; and an indicator of the adjustment reason. For example, the system generates the following corrective constraint: "When the fabric type is recycled denim, the cutting shrinkage rate should be adjusted to 3% ± 0.5%", and outputs this corrective constraint to the subsequent process feasibility analysis step to constrain the input parameter range of the deep learning model. This allows for structured and rule-based compliance verification of process parameters before production, preventing unreasonable parameters from directly entering the production or model analysis process, thereby reducing process risks and providing a clear data constraint basis for subsequent process feasibility assessment and alternative solution generation.

[0019] S300. Based on the process parameters after compliance verification or adjustment according to the aforementioned modified constraints, a matching degree analysis is performed on the physical properties of the fabric, process requirements and production equipment capabilities using a deep learning model to assess the feasibility of process implementation and output the corresponding process feasibility assessment results.

[0020] Furthermore, step S300 of this application also includes: S301: Obtain the fabric physical property parameters, process requirement parameters, and production equipment capability parameters corresponding to the current production task, and perform parameter type identification and numerical normalization processing on the parameters; S302: Encode the fabric physical property parameters, process requirement parameters, and production equipment capability parameters into a unified-dimensional process feature vector according to a preset feature mapping rule; S303: Input the process feature vector into a trained deep learning model, which is used to characterize the nonlinear correlation between fabric physical properties, process requirements, and production equipment capability; S304: Based on the output of the deep learning model, calculate the feasibility score or risk assessment vector of the current process parameter combination under the corresponding production conditions; S305: Output the feasibility score or risk assessment vector as the process feasibility assessment result.

[0021] Specifically, the process involves acquiring the fabric physical property parameters, process requirement parameters, and production equipment capability parameters corresponding to the current production task. The fabric physical property parameters include, but are not limited to: fabric weight (g / m²), thickness (mm), yarn density (threads / inch), breaking strength (N), elongation (%), and surface friction coefficient. The process requirement parameters include stitch length (mm), sewing speed (needles / minute), quilting type, sewing tension range, and allowable process tolerances. The production equipment capability parameters include equipment model, maximum speed, needle type compatibility range, maximum sewing thickness capacity, and stable operating range. The acquired parameters are identified by type, and based on their attributes, minimum-maximum normalization or interval mapping is used to convert parameters of different dimensions to a unified numerical range. For example, continuous parameters are mapped to the [0,1] interval, and discrete parameters are represented using one-hot encoding. According to preset feature mapping rules, the normalized fabric physical property parameters, process requirement parameters, and production equipment capability parameters are concatenated and encoded into a unified-dimensional process feature vector. The preset feature mapping rules refer to parameter mapping and encoding rules pre-determined based on historical production data statistical analysis results and process expert experience before system deployment. These rules are used to convert process-related parameters from different sources, with different dimensions, and different physical meanings into a feature representation that can be uniformly input into subsequent analysis models. The preset feature mapping rules include combinations of at least one or more of the following rules: Parameter type mapping rules, which classify input parameters into continuous parameters, discrete enumeration parameters, and categorical parameters based on their physical attributes and data types, and adopt corresponding encoding methods for different types of parameters. For example, for continuous parameters, interval normalization or standardization is used for numerical mapping; for discrete enumeration parameters, sequential encoding or interval mapping is used for conversion; and for categorical parameters, one-hot encoding or index encoding is used. Feature order and dimension alignment rules, which pre-determine the arrangement order and corresponding dimension position of each parameter in the process feature vector based on the logical order of process parameters in the production process and their importance to process feasibility, ensuring that process feature vectors generated at different times for the same type of production task have a consistent dimensional structure. The weighting and scaling unification rule adjusts the scale of parameters with significantly different degrees of influence using scaling coefficients or weighting factors to avoid unreasonable amplification or suppression of the feature vector by a single parameter. The weighting factors can be set based on historical process success rate statistics or expert experience.The preset feature mapping rules are obtained in ways including but not limited to: statistically analyzing the correlation between fabric physical property parameters, process parameters, and production results in historical production tasks, combined with the experience judgment of process engineers regarding the influence of key parameters, configuring and storing them during the system initialization phase; and periodically updating or correcting the mapping rules based on new production data during system operation. For example, in one embodiment, fabric physical property parameters include fabric weight (g / m²), thickness (mm), and elasticity level (divided into low, medium, and high levels); process requirement parameters include set tension value (N) and operating speed (m / min); and production equipment capacity parameters include the maximum tension limit and speed adjustment range of the equipment. In the feature mapping process: continuous parameters such as fabric weight, thickness, set tension, and operating speed are normalized according to their historical minimum and maximum values; elasticity level parameters are mapped to corresponding numerical codes, for example, low elasticity is mapped to 0, medium elasticity to 0.5, and high elasticity to 1; and they are concatenated according to a preset order of "fabric physical property parameters - process requirement parameters - equipment capability parameters" to finally generate a fixed-dimensional process feature vector for subsequent process feasibility analysis. Through the above-mentioned preset feature mapping rules, process-related parameters from different sources can be expressed under a unified data structure, ensuring the consistency, comparability, and reusability of the process feature vector, providing a stable data foundation for subsequent process feasibility assessment and risk analysis. The process feature vector is a fixed-length vector, such as 128-dimensional or 256-dimensional, used to fully represent the comprehensive characteristics of the current process scheme in terms of fabric, process, and equipment. The process feature vector is then input into a trained deep learning model for inference analysis. The deep learning model is a multi-layer neural network model trained based on historical production data. Its input layer receives the process feature vector, the intermediate hidden layers learn the non-linear correlation between fabric physical properties, process requirements, and equipment capabilities, and the output layer generates assessment results related to process feasibility. During model training, historical successful and failed process schemes are used as sample data. Through supervised learning, the model can distinguish between high-feasibility and high-risk process combinations. The training process of the deep learning model, a multi-layer neural network model trained based on historical production data, includes the following steps: First, constructing a training dataset.Data samples corresponding to completed production tasks are collected from the company's historical production records. These data samples include at least: fabric physical property parameters, such as weight (g / m²), thickness (mm), fabric density, yarn count, elasticity coefficient, and shrinkage rate; process requirement parameters, such as sewing method, stitch length, tension setting, cutting shrinkage rate, quilting method, and processing temperature range; production equipment capability parameters, such as equipment model, maximum sewing speed, supported stitch length range, tension adjustment accuracy, and equipment stability level; and process execution result labels, used to characterize the implementation results of the corresponding process scheme. These labels include successful process labels and failed process labels, or corresponding quality level evaluation results. Next, the training data is preprocessed. The fabric physical property parameters, process requirement parameters, and equipment capability parameters are numerically encoded. Continuous parameters are processed using normalization or standardization methods, while discrete parameters are processed using one-hot encoding or embedding encoding methods to form a unified-dimensional process feature vector. Then, a multi-layer neural network model is constructed and trained. The multilayer neural network model includes an input layer, at least one hidden layer, and an output layer: the input layer receives the process feature vector; the hidden layer learns the nonlinear mapping relationship between fabric physical properties, process requirements, and production equipment capabilities; and the output layer outputs the assessment results related to process feasibility, including a process feasibility score or a risk assessment vector. During model training, the process feature vector is used as the input sample, and the corresponding process execution result label is used as the supervision signal. The model parameters are iteratively updated using a backpropagation algorithm, enabling the model to gradually learn the discriminative features that distinguish between high-feasibility and high-risk process combinations. Finally, model training and deployment are completed. When the model's prediction results on the validation dataset meet the preset accuracy requirements, training stops, and the trained model parameters are stored and solidified for inference calculations of the current process parameter combinations during subsequent process feasibility assessments. For example, the deep learning model is a multilayer perceptron (MLP) neural network model used to model the nonlinear relationship between fabric physical properties, process requirements, and production equipment capabilities. The model consists of: Input layer: Receives a process feature vector encoded from fabric physical property parameters, process requirement parameters, and production equipment capability parameters. The input dimension is determined by the number of features, for example, 50-200 dimensions. Hidden layer: The model contains 3-5 fully connected hidden layers, each with 128, 256, or 512 nodes. The ReLU activation function is used to learn non-linear feature maps. Dropout (0.2-0.5) can be used in the hidden layers to prevent overfitting. Output layer: Generates a process feasibility score (continuous numerical value) or a risk assessment vector (multi-dimensional classification probability). The output dimension can be set according to the number of scores or risk types, for example, 1-5 dimensions.Training data and sample sources: Sample data comes from the company's historical production records, including successful and unsuccessful process solutions. Each record includes fabric attributes, process parameters, equipment capabilities, production batches, and final quality results. The sample size is recommended to be between 1,000 and 50,000 records or more to ensure the model's generalization ability. Feature preprocessing includes numerical normalization (Min-Max or Z-score), missing value imputation, and outlier filtering. The training process involves dividing historical process data into a training set (70%), a validation set (15%), and a test set (15%). Supervised learning is employed, using process success / failure labels or quality scores as target values. The optimizer is Adam, with an initial learning rate of 0.001~0.01 and a batch size of 32~128. The loss function is selected based on the output type: mean squared error (MSE) is used for continuous scores, and cross-entropy loss is used for classification probabilities. During model training, the validation set loss is monitored, and an early stopping strategy is employed to prevent overfitting. Hyperparameter grid search or Bayesian optimization can be performed to obtain the optimal number of hidden layer nodes, learning rate, and Dropout ratio. After model training, the optimal model weights are deployed to an edge computing device for real-time process feasibility assessment. Successful process solutions are process combinations that have not exhibited major quality defects in actual production and meet quality acceptance standards. Failed process solutions are process combinations that have experienced quality problems such as broken threads, wrinkles, color differences, and structural instability during production. Based on the output of the deep learning model, a feasibility score or risk assessment vector is calculated for the current combination of process parameters under corresponding production conditions. The feasibility score quantifies the overall implementability of the process scheme, for example, ranging from 0 to 1. The risk assessment vector represents the risk weights of each key process factor, indicating potential sources of risk. The feasibility score or risk assessment vector is output as the process feasibility assessment result and correlated with the current production task, process sheet, and fabric type for subsequent alternative process generation, certification constraint verification, and production execution decisions.

[0022] S400. Based on the process feasibility assessment results, when a process feasibility score is detected to be lower than a preset threshold or a high-risk process combination exists, an alternative process solution is automatically generated and recommended.

[0023] Furthermore, step S400 of this application also includes: S401: Obtain the process feasibility assessment result output by the process feasibility assessment step, the process feasibility assessment result including a process feasibility score and a corresponding risk assessment vector; S402: Compare the process feasibility score with a preset feasibility threshold, or identify high-risk process factors identified in the risk assessment vector, to determine whether the current process scheme belongs to a high-risk process combination; S403: When the current process scheme is determined to be a high-risk process combination, extract a set of process correction constraints for generating alternative processes based on the process and standard knowledge set and the process correction constraints generated by the previous steps; S404: Under the constraints of the process correction constraint set, adjust, replace, or reorganize the process parameters of the current process scheme based on the process and standard knowledge set to generate at least one alternative process scheme; S405: Output the generated alternative process scheme as a process recommendation result for subsequent process parameter configuration and production execution.

[0024] Specifically, the process feasibility assessment results output from the process feasibility assessment step are obtained, including a process feasibility score (ranging from 0 to 1, with higher scores indicating stronger process feasibility) and a risk assessment vector (the number of dimensions matches the number of risk factors, for example, 10 to 20). (Each dimension corresponds to a risk score for specific process parameters); the process feasibility score is compared with a preset feasibility threshold (e.g., 0.7), or high-risk process factors in the risk assessment vector that exceed the threshold are identified; if the score is below the threshold or a high-risk process factor exists, the current process scheme is determined to be a high-risk process combination; after determining the current process scheme to be a high-risk process combination, based on the process and standard knowledge set and the correction constraints generated in the previous steps (e.g., specific parameters such as cutting shrinkage rate, seam density, stitch length range, pressure range, etc.), a set of process correction constraints for generating alternative processes is extracted; at the same time, the constraints may include material compatibility limitations, upper and lower limits of equipment capabilities, and special requirements of production batches; under the constraints of the set of process correction constraints, based on the process and standard knowledge set, the process parameters of the current process scheme are adjusted, replaced, or reorganized to generate at least one alternative process scheme; for example: adjusting the cutting shrinkage rate by ±1%~3%; replacing double-needle quilting with blind stitching or single-needle quilting; optimizing the sewing machine stitch length from the range of 2.5 mm~3.5 mm to the recommended value of 3.0. mm; Replace incompatible auxiliary materials, such as replacing polyester lining with cotton lining; Alternative process solutions can generate 1 to 5 candidate solutions, which are sorted according to process feasibility score, and the corresponding risk assessment vector is retained. The generated alternative process solutions are output as process recommendation results to the process parameter configuration and production execution module, and the alternative solutions are associated with the original process solutions for subsequent quality traceability and risk analysis.

[0025] S500: Based on the determination of the process feasibility assessment results, automatically associate with the international environmental protection and safety standard library, match and verify the process parameters, testing items and quality thresholds according to the certification constraint information associated with the production task, and feed back the certification constraint results to the process parameter configuration process.

[0026] Furthermore, step S500 of this application also includes: S501: Based on the determined process feasibility assessment results, obtain certification constraint information associated with the current production task. The certification constraint information includes certification type identifiers corresponding to at least one international environmental protection standard and / or safety standard. S502: Based on the certification constraint information, automatically call the international environmental protection standard library and safety standard library corresponding to the certification type identifier, and extract the certification rule set matching the production task. The certification rule set includes process parameter restriction rules, testing item configuration rules, and quality threshold rules. S503: Perform item-by-item matching and verification between the process parameters corresponding to the current production task and the process parameter restriction rules in the certification rule set to determine whether the process parameters meet the corresponding certification requirements. S504: Based on the testing item configuration rules and quality threshold rules in the certification rule set, determine the set of testing items corresponding to the certification requirements and the quality judgment threshold for each testing item. S505: When it is detected that the process parameters or testing items do not meet the corresponding certification requirements, generate certification constraint verification results and feed them back to the process parameter configuration process for constraint adjustment or correction of the process parameters.

[0027] Specifically, based on the results of the process feasibility assessment, step S500 of this application automatically associates with the international environmental protection and safety standards library, matches and verifies the process parameters, testing items and quality thresholds according to the certification constraint information associated with the production task, and feeds back the certification constraint results to the process parameter configuration process, thereby ensuring that the production process meets international certification requirements and optimizing the quality control process. Obtain certification constraint information associated with the current production task. This information includes certification type identifiers corresponding to at least one international environmental standard (such as OEKO-TEX, GOTS) and / or safety standard (such as ISO 9001, ISO 14001), and standardizes these identifiers. Based on this certification constraint information, automatically call the international environmental standard library and safety standard library corresponding to the certification type identifier, and extract a certification rule set matching the production task. This rule set includes: process parameter restriction rules (e.g., dyeing temperature 20~60°C, sewing stitch length 2.0~3.5 mm, pressing time 5~15 s, and accessory material type restrictions (cotton, polyester, etc.); testing item configuration rules (e.g., formaldehyde content testing, pH value testing, color fastness testing, tensile strength testing, etc.); and quality threshold rules (e.g., formaldehyde content ≤ 75 mg / kg, pH value 4.0~7.5, color fastness ≥ 3~4 grade, tensile strength ≥ 300 N). The process parameters corresponding to the current production task are matched and verified item by item with the process parameter restriction rules in the certification rule set to determine whether the process parameters meet the certification requirements and generate a process parameter compliance identifier. Based on the testing item configuration rules and quality threshold rules in the certification rule set, a set of testing items corresponding to the certification requirements is determined, and a specific judgment threshold is set for each testing item. When a process parameter or testing item is detected as not meeting the corresponding certification requirements, a certification constraint verification result is generated, including the non-compliant parameters, the deviation range, and the corresponding suggested adjustment value. This verification result is fed back to the process parameter configuration process to guide subsequent process parameter adjustments, corrections, or the generation of alternative solutions. This system enables automatic matching and verification of process parameters with international environmental and safety standards, ensuring that the production process meets certification requirements. It also provides accurate data support for process parameter adjustments, reduces production risks, improves product quality consistency, and supports traceable certification compliance management.

[0028] S600 performs defect identification based on local deep learning inference models on fabric image data collected from the production site, and outputs the defect type, defect location and defect severity.

[0029] Furthermore, step S600 of this application also includes: S601: Acquire the fabric image data and preprocess the image data, including noise suppression, image enhancement, and standardization, to generate input data suitable for deep learning inference; S602: Input the preprocessed image data into a trained deep learning inference model deployed on an edge computing device. The deep learning inference model is used to identify the type, spatial location, and severity of fabric defects; S603: Based on the inference output of the deep learning inference model, generate fabric defect information, including defect category identifier, spatial location information, and severity score, and store the fabric defect information in a local database; S604: When a fabric defect is detected, generate defect labeling data based on the fabric defect information for production quality analysis and process parameter adjustment; S605: Associate the fabric defect information with the corresponding production batch information and process sheet to form a data record that can be used for subsequent quality traceability, risk prediction, and process optimization, and synchronize the data to a remote management system when the network is available.

[0030] Specifically, for fabric image data collected at the production site, defect identification is performed based on a deep learning inference model deployed on an edge computing device, outputting the defect type, spatial location, and severity, thereby achieving real-time quality monitoring and process parameter optimization during production. Fabric image data collected by industrial cameras or intelligent vision sensors at the production site is acquired. The image data resolution is 1920×1080 or 2048×2048 pixels, with a frame rate of 10~30fps. The image data is preprocessed, including: noise suppression: applying a Gaussian filter or median filter to remove acquisition noise while preserving fabric texture details; image enhancement: performing histogram equalization, contrast stretching, or adaptive gamma correction to improve the distinguishability of defect features; and normalization: normalizing image pixel values ​​to the range of [0,1] or [-1,1] to adapt to the input requirements of the deep learning model. Preprocessed image data is input into a trained deep learning inference model deployed on an edge computing device. This model, either a multi-layer convolutional neural network (CNN) or a visual Transformer based on an attention mechanism, is used to identify the category, spatial location, and severity of fabric defects. The model input is a size-normalized image tensor, and the output includes a defect category probability distribution, defect bounding box coordinates (x, y, width, height), and a severity score (e.g., a 0-1 scale). Based on the inference output of the deep learning model, fabric defect information is generated, including: defect category identifiers (e.g., loose threads, stains, uneven stitch spacing, color difference); spatial location information (coordinates and bounding boxes of the defect in the image); and a severity score (calculated using probability values ​​from the model output or a custom scoring function). The generated fabric defect information is written to a local database. Simultaneously, when a fabric defect is detected, corresponding defect annotation data is generated for production quality analysis and process parameter adjustment, such as adjusting weaving speed, tension, or dyeing and finishing process parameters. The fabric defect information is associated with corresponding production batch information and process sheets to form data records that can be used for quality traceability, risk prediction, and process optimization. When network access is available at the production site, fabric defect information and related data incrementally synchronized from the local database to the remote management system to support end-to-end quality management and historical data accumulation and analysis. This enables real-time, high-precision identification and spatial positioning of fabric defects, generating structured fabric defect information and automatically associating it with production batches and process sheets, supporting subsequent process optimization and quality traceability. Simultaneously, local inference based on edge computing devices reduces network dependence, ensuring continuous monitoring and data availability at the production site, thereby improving the automation, accuracy, and real-time nature of production quality management.

[0031] S700: In the event that the network is unavailable or unstable at the production site, the defect identification results and corresponding image data are stored in the local database for caching, and incremental synchronization is performed after the network is restored.

[0032] Furthermore, step S700 of this application also includes: S701: Acquire fabric defect information and corresponding image data output from the real-time defect identification step at the production site, and record the batch number and process sheet information; S702: Detect the network status at the production site. When the network is unavailable or unstable, store the fabric defect information and corresponding image data, along with the batch number and process sheet information, in the local database for caching; S703: After the network is restored, determine the synchronization status between the data cached in the local database and the data in the remote management system based on the fabric defect information and the association identifier of the batch number and process sheet; S704: Perform incremental synchronization on the data that has not yet been synchronized in the local database, including uploading fabric defect information, image data, and related batch and process sheet association information; S705: During the synchronization process, automatically verify and process data conflicts or duplicate records. The verification strategy includes unique identifier comparison, timestamp comparison, and data integrity verification.

[0033] Specifically, in cases where the production site network is unavailable or unstable, the fabric defect information and corresponding image data output from the real-time defect identification step are locally cached, and incremental synchronization is performed after the network is restored to ensure the integrity and continuity of production data. The fabric defect information and corresponding image data output from the real-time defect identification step are acquired at the production site. The fabric defect information includes defect category identifiers, spatial location information, and severity scores. The image data resolution is 1920×1080 pixels or 2048×2048 pixels. The production batch number, process order number, and acquisition timestamp corresponding to each piece of fabric defect information are recorded synchronously. The network status at the production site is monitored in real time. When network unavailability or network quality below a preset threshold (e.g., packet loss rate > 5%, latency > 200 ms) is detected, the fabric defect information and corresponding image data, along with the batch number, process order number, and timestamp, are stored in a local database for caching. The database can be SQLite, PostgreSQL, or other lightweight embedded databases. After network recovery, based on fabric defect information and its associated batch number and process order number, the synchronization status between the data cached in the local database and the remote management system is determined, and the set of incremental data to be uploaded is identified. This determination includes checking whether the unique identifier (UUID), timestamp, and checksum are consistent with the remote data. Incremental synchronization is performed on data that has not yet been synchronized in the local database, including uploading fabric defect information, image data, and related batch and process order information. During the upload process, data can be uploaded in batches and concurrently, with a limit on the size of each data entry (e.g., no more than 5 MB per entry) to ensure network transmission stability and data integrity of the remote management system. During synchronization, data conflicts or duplicate records are automatically checked and processed. The check strategies include: unique identifier comparison: ensuring that the same fabric defect information is not uploaded repeatedly; timestamp comparison: using the latest timestamp as the standard; data integrity verification: using MD5 or SHA-256 to calculate the data digest to ensure that image data and fabric defect information are not damaged during transmission. It can continuously save fabric defect information and image data when the network is unstable or unavailable, and efficiently and securely synchronize incremental data to the remote management system after the network is restored, ensuring the integrity, consistency and traceability of production data, while reducing the impact of network fluctuations on production monitoring and quality analysis, and achieving continuous and stable production quality management.

[0034] S800: Based on the historical quality inspection data accumulated from the defect identification results and the synchronously collected production environment data, a time series analysis model is constructed to predict potential quality risks and generate corresponding risk warning information.

[0035] Furthermore, step S800 of this application also includes: S801: Obtain the defect identification results output by the real-time defect identification step and accumulate them to form a historical quality inspection dataset; S802: Collect production environment data corresponding to the historical quality inspection data, including temperature, humidity and other production process-related parameters, and associate the production environment data with the historical quality inspection data based on the timestamp or batch number of the defect record; S803: Perform data preprocessing on the historical quality inspection data and production environment data, including missing value processing, outlier filtering and numerical normalization; S804: Input the preprocessed historical quality inspection data and production environment data into a pre-trained time series analysis model to generate a corresponding quality risk prediction vector; S805: Output the quality risk prediction vector and record the association information with the corresponding batch and process sheet.

[0036] Specifically, to address potential quality risks in the fabric production process, a time-series analysis model is constructed by analyzing historical quality inspection data accumulated from real-time defect identification results and synchronously collected production environment data. This model predicts potential quality problems and generates corresponding risk warning information to assist in production management and process optimization. Fabric defect information output from the real-time defect identification step is acquired, including defect category, spatial location information, severity score, and corresponding image data. This data is accumulated according to production batch number, process order number, and collection timestamp to form a historical quality inspection dataset. The historical quality inspection dataset can contain defect records from the most recent 30 to 180 days, with fields including batch number, process order number, defect category ID (e.g., 1=pinhole, 2=stain, 3=weave abnormality), severity score (range 0~1), etc. Simultaneously, production environment data corresponding to the historical quality inspection data is collected, including but not limited to process-related parameters such as temperature (°C), humidity (%), fabric rolling speed (m / min), equipment temperature (°C), and rolling pressure (kPa). Based on the timestamp or batch number of defect records, the production environment data is associated with historical quality inspection data to form a time-series feature matrix containing process conditions. The historical quality inspection data and production environment data undergo data preprocessing, including missing value handling (e.g., forward imputation or mean imputation), outlier filtering (e.g., Z-score > 3 is considered anomaly), and numerical normalization (e.g., min-max normalization to the 0~1 range) to ensure the data is suitable for training and inference of the time-series analysis model. The preprocessed historical quality inspection data and production environment data are input into a pre-trained time-series analysis model, which can be a Long Short-Term Memory Network (LSTM), a Gated Recurrent Unit (GRU), or a Temporal Convolutional Network (TCN), to capture the time-dependent impact of production environment changes on fabric defect occurrence. The model outputs a corresponding quality risk prediction vector, containing the predicted probability, potential severity, and risk level score for each type of defect. The quality risk prediction vector is output and its association with the corresponding batch number and process document number is recorded for subsequent quality traceability, process optimization, and risk warning systems. The output data fields include the predicted batch number, process order number, defect type probability (0~1), comprehensive risk score (0~1), and recommended warning level (e.g., low, medium, high). It enables efficient correlation and processing of production site defect data with environmental parameters, accurately predicts potential quality risks using time-series analysis models, provides forward-looking risk warning information for production management, assists in process adjustment and quality optimization, achieves refined and controllable management of the production process, and reduces fabric defect rates and production costs.

[0037] S900 automatically associates defect identification results and quality risk prediction results with the corresponding process sheets and production batches to achieve cross-process traceability of quality problems and synchronizes the analysis results to the production management module and trade management module.

[0038] Furthermore, step S900 of this application also includes: S901: Obtain the defect identification results output from the real-time defect identification step and the quality risk prediction results generated based on historical quality inspection data and production environment data; S902: Automatically associate the defect identification results and the quality risk prediction results with the corresponding process sheets and production batches according to the batch information and process sheet identifier in the defect identification results and the quality risk prediction results; S903: Synchronize the associated defect identification results and quality risk prediction results to the production management module and the trade management module; S904: During the synchronization process, automatically verify and process data conflicts or duplicate records. The verification strategy includes unique identifier comparison, timestamp comparison, and data integrity verification.

[0039] Specifically, for quality data in the fabric production process, the system automatically associates defect identification results and quality risk prediction results with corresponding process orders and production batches, and synchronizes the analysis results to the production management module and trade management module to achieve cross-process traceability of quality problems, production management optimization, and trade information integration. It acquires fabric defect information output from real-time defect identification step S600, including defect category, spatial location information, severity score, and corresponding image data; simultaneously, it acquires quality risk prediction results output from step S800, including predicted probabilities for each type of defect, comprehensive risk scores, and warning levels. The defect identification results and quality risk prediction results form a dataset according to production batch number, process order number, and timestamp. Examples of fields include: Batch Number (Batch_ID), Process Order Number (WorkOrder_ID), Defect Type ID (Defect_Type_ID), Defect Severity (Severity_Score, range 0~1), Predicted Defect Probability (Probability, range 0~1), Comprehensive Risk Score (Risk_Score, range 0~1), and Warning Level (Risk_Level, low / medium / high). Based on the batch information and process sheet identifiers in the defect identification and quality risk prediction results, the data is automatically associated with the process sheet records and production batch information in the enterprise production management system. The association method can be based on a one-to-one correspondence between batch number and process sheet number, while simultaneously verifying the data collection timestamp to ensure data integrity and temporal consistency. The records formed through association include fabric defect information, risk prediction information, process sheet parameters, and production batch parameters to support subsequent cross-process quality analysis. The associated defect identification and quality risk prediction results are synchronized to the production management module and the trade management module. The synchronization process can use incremental data updates, uploading only newly added or updated data to reduce network load and data redundancy. Synchronized data fields include batch number, process sheet number, defect category, defect severity, prediction probability, comprehensive risk score, warning level, and related process parameters. During the synchronization process, data conflicts or duplicate records are automatically checked and processed. The verification strategy includes: unique identifier comparison: checking whether the combination of batch number and process order number is unique; timestamp comparison: comparing data collection time and prioritizing the retention of the latest record; data integrity verification: verifying whether key fields (such as defect category, risk score, process parameters) have null or outlier values, and correcting or marking the anomalies. This enables cross-process traceability and automatic synchronization management of fabric defect information and quality risk prediction data, improving data consistency and traceability in production and trade management, supporting process optimization and risk control based on historical data, reducing quality accidents and trade disputes, and improving production efficiency and fabric quality reliability.

[0040] In summary, the embodiments of this application have at least the following technical effects: This application constructs an enterprise process specification library, a fabric and accessory specification library, and a cost parameter library, and structures data from different sources to form a unified set of process and standard knowledge, improving the uniformity, manageability, and reusability of enterprise process data. After receiving demand information related to fabric type, process parameters, and production tasks, it performs compliance verification based on preset rules and the aforementioned process and standard knowledge set, generating corrective constraints or adjustment suggestions, thus achieving standardization and automatic optimization of process parameters. Through process feasibility analysis based on a multi-layer deep learning model, it assesses the matching degree between fabric physical properties, process requirements, and production equipment capabilities, effectively identifying high-risk process combinations and generating alternative process solutions, thereby improving the feasibility and reliability of process implementation. Combined with an international environmental protection and safety standard library, it achieves automatic processing of process parameters, testing items, and quality thresholds. Matching and verification ensure that production tasks meet relevant certification requirements, improving production compliance. Industrial cameras and edge computing devices are deployed on the production floor, using deep learning inference models to achieve real-time fabric defect identification, outputting defect type, location, and severity, and linking it to production batches and process sheets for real-time quality monitoring. For situations where the network is unavailable or unstable, defect data is cached in a local database and incrementally synchronized after network recovery, ensuring data integrity and continuity. Based on accumulated historical quality inspection data and synchronously collected production environment data, a time-series analysis model is constructed to predict and warn of potential quality risks, assisting in process optimization decisions. Simultaneously, defect identification results and risk prediction results are automatically linked to process sheets and production batches, enabling cross-process quality traceability and synchronizing with production management and trade management modules, improving data consistency and traceability. In summary, this application, through data-driven process optimization, real-time defect detection, and risk prediction, effectively solves the problems of non-standard processes, lagging quality monitoring, difficulties in defect traceability, and insufficient certification compliance in existing technologies, achieving significant improvements in production efficiency, product quality, and management level.

[0041] It should be noted that the order of the embodiments described above is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. Furthermore, the above description focuses on specific embodiments of this specification. The processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired results. In some implementations, multitasking and parallel processing are possible or may be advantageous.

[0042] The above description is only a preferred embodiment of this application and is not intended to limit this application. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.

[0043] This specification and accompanying drawings are merely illustrative examples of this application and are intended to cover any and all modifications, variations, combinations, or equivalents within the scope of this application. Clearly, those skilled in the art can make various alterations and modifications to this application without departing from its scope. Therefore, if such modifications and modifications fall within the scope of this application and its equivalents, this application intends to include such modifications and modifications.

Claims

1. A method for real-time detection and adaptive correction of fabric defects based on deep learning, characterized in that, include: S100. Construct an enterprise process specification library, a raw material and auxiliary material specification library, and a cost parameter library, and perform structured processing on data from different sources to form a unified set of process and standard knowledge. S200. After receiving the demand information related to fabric type, process parameters and production task, the demand information is verified for compliance based on preset rules and the process and standard knowledge set. When the parameters are detected to be inconsistent with the preset standards, corresponding correction constraints or adjustment suggestions are generated. S300. Based on the process parameters after compliance verification or adjustment according to the modified constraints, a matching degree analysis is performed on the physical properties of the fabric, process requirements and production equipment capabilities using a deep learning model to assess the feasibility of the process implementation and output the corresponding process feasibility assessment results. S400. Based on the process feasibility assessment results, when the process feasibility score is detected to be lower than a preset threshold or there is a high-risk process combination, an alternative process solution is automatically generated and recommended. S500. Based on the determination of the process feasibility assessment results, automatically associate with the international environmental protection and safety standard library, match and verify the process parameters, testing items and quality thresholds according to the certification constraint information associated in the production task, and feed back the certification constraint results to the process parameter configuration process. S600: Based on the fabric image data collected at the production site, it performs defect identification and outputs the defect type, defect location and defect severity. S700: In the event that the network is unavailable or unstable at the production site, the defect identification results and corresponding image data are stored in the local database for caching, and incremental synchronization is performed after the network is restored. S800: Based on the historical quality inspection data accumulated from the defect identification results and the synchronously collected production environment data, a time series analysis model is constructed to predict potential quality risks and generate corresponding risk warning information. S900 automatically associates defect identification results and quality risk prediction results with the corresponding process sheets and production batches to achieve cross-process traceability of quality problems and synchronizes the analysis results to the production management module and trade management module.

2. The method for real-time detection and adaptive correction of fabric defects based on deep learning as described in claim 1, characterized in that, Construct an enterprise process specification library, a raw material and auxiliary material specification library, and a cost parameter library, and perform structured processing on data from different sources to form a unified set of process and standard knowledge, including: S101. Obtain process specification data from the company's existing process documents, historical production records and process management system, and split the process specification data according to the preset process element fields to generate a set of process parameter items. S102. Obtain fabric and accessory specification data from fabric and accessory supply information, test reports and material management system, and perform field standardization processing on the material type, physical property index and applicable process range of fabric and accessory to generate a set of fabric and accessory specification parameters. S103. Obtain cost parameter data from production cost accounting records and procurement records, and perform unified unit conversion and interval processing on the cost parameter data to generate a cost parameter set. S104. Perform field mapping and semantic alignment processing on the set of process parameters, the set of raw and auxiliary material specifications, and the set of cost parameters, and establish a cross-database association identifier. S105. Write the set of process parameter items after completing field mapping and semantic alignment processing into the process specification library, write the set of fabric and auxiliary material specification parameters into the fabric and auxiliary material specification library, and write the set of cost parameters into the cost parameter library.

3. The method for real-time detection and adaptive correction of fabric defects based on deep learning as described in claim 1, characterized in that, After receiving the demand information related to fabric type, process parameters, and production tasks, the system performs compliance verification on the demand information based on preset rules and the aforementioned process and standard knowledge set. When parameters are detected to be inconsistent with preset standards, corresponding correction constraints or adjustment suggestions are generated, including: S201. The received demand information is parsed and processed, and the demand information is split into fabric type information, process parameter information and demand constraint information. The format verification and standardization processing of each type of information after splitting are performed to generate a unified data description format. S202. Using the unified data description format, retrieve the process specification data, fabric and accessory specification data, and cost parameter data corresponding to the fabric type and production task from the process and standard knowledge set to construct a set of parameters to be verified. S203. Based on the preset compliance verification rules, the set of parameters to be verified is compared and analyzed item by item to identify whether the process parameters meet the corresponding process specifications, raw material specifications and cost constraints. S204. When at least one process parameter is detected to be non-compliant with the preset standard, a corresponding parameter deviation identifier is generated, and the target parameter to be adjusted and its constraint range are determined based on the parameter deviation identifier. S205. Based on the target parameters and their constraint range, generate corrective constraints or adjustment suggestions to guide subsequent process parameter adjustments, and output the corrective constraints or adjustment suggestions to the subsequent process feasibility analysis process.

4. The method for real-time detection and adaptive correction of fabric defects based on deep learning as described in claim 1, characterized in that, Based on the process parameters adjusted according to the aforementioned modified constraints or compliance verification, a deep learning model is used to analyze the matching degree between the fabric physical properties, process requirements, and production equipment capabilities. The feasibility of the process implementation is assessed, and the corresponding process feasibility assessment results are output, including: S301. Obtain the physical property parameters of the fabric, the process requirement parameters, and the production equipment capacity parameters corresponding to the current production task, and perform parameter type identification and numerical normalization processing on the parameters. S302. According to the preset feature mapping rules, the physical property parameters of the fabric, the process requirement parameters, and the production equipment capability parameters are encoded into a process feature vector of a unified dimension. S303. Input the process feature vector into the trained deep learning model, the deep learning model being used to characterize the nonlinear relationship between fabric physical properties, process requirements and production equipment capabilities. S304. Based on the output of the deep learning model, calculate the feasibility score or risk assessment vector of the current combination of process parameters under the corresponding production conditions. S305. Output the feasibility score or risk assessment vector as the process feasibility assessment result.

5. The method for real-time detection and adaptive correction of fabric defects based on deep learning as described in claim 1, characterized in that, Based on the process feasibility assessment results, when a process feasibility score is detected to be lower than a preset threshold or a high-risk process combination exists, alternative process solutions are automatically generated and recommended, including: S401. Obtain the process feasibility assessment result output by the process feasibility assessment step, wherein the process feasibility assessment result includes a process feasibility score and a corresponding risk assessment vector. S402. Compare the process feasibility score with a preset feasibility threshold, or identify the high-risk process factors marked in the risk assessment vector to determine whether the current process scheme belongs to a high-risk process combination. S403. When the current process scheme is determined to be a high-risk process combination, extract the set of process correction constraints for replacing the process based on the process and standard knowledge set and the process correction constraints generated by the previous steps. S404. Under the constraints of the process modification constraint set, based on the process and standard knowledge set, adjust, replace or reorganize the process parameters of the current process scheme to generate at least one alternative process scheme. S405. Output the generated alternative process solutions as process recommendation results for subsequent process parameter configuration and production execution.

6. The method for real-time detection and adaptive correction of fabric defects based on deep learning as described in claim 1, characterized in that, Based on the determined process feasibility assessment results, the system automatically links to international environmental and safety standard libraries. According to the certification constraint information associated with the production task, it matches and verifies process parameters, testing items, and quality thresholds, and feeds back the certification constraint results to the process parameter configuration process, including: S501. Based on the determination of the process feasibility assessment results, obtain certification constraint information associated with the current production task. The certification constraint information includes certification type identifiers corresponding to at least one international environmental protection standard and / or safety standard. S502. Based on the certification constraint information, automatically call the international environmental protection standard library and safety standard library corresponding to the certification type identifier, and extract the certification rule set that matches the production task. The certification rule set includes process parameter restriction rules, testing item configuration rules and quality threshold rules. S503. Match and verify the process parameters corresponding to the current production task with the process parameter restriction rules in the certification rule set item by item to determine whether the process parameters meet the corresponding certification requirements. S504. Based on the testing item configuration rules and quality threshold rules in the certification rule set, determine the set of testing items corresponding to the certification requirements and the quality judgment threshold of each testing item. S505. When it is detected that the process parameters or test items do not meet the corresponding certification requirements, a certification constraint verification result is generated, and the certification constraint verification result is fed back to the process parameter configuration process for constraint adjustment or correction of the process parameters.

7. The method for real-time detection and adaptive correction of fabric defects based on deep learning as described in claim 1, characterized in that, For fabric image data collected from the production site, defect identification is performed based on a local deep learning inference model, outputting defect type, defect location, and defect severity, including: S601. Obtain the fabric image data and preprocess the image data, including noise suppression, image enhancement and standardization, to generate input data suitable for deep learning inference. S602. Input the preprocessed image data into a trained deep learning inference model deployed on an edge computing device. The deep learning inference model is used to identify the type, spatial location and severity of fabric defects. S603. Based on the inference output of the deep learning inference model, generate fabric defect information, including defect category identifier, spatial location information and severity score, and store the fabric defect information in a local database. S604. When a fabric defect is detected, defect labeling data is generated based on the fabric defect information for production quality analysis and process parameter adjustment. S605. Associate the fabric defect information with the corresponding production batch information and process sheet to form a data record that can be used for subsequent quality traceability, risk prediction and process optimization, and synchronize the data to the remote management system when the network is available.

8. The method for real-time detection and adaptive correction of fabric defects based on deep learning as described in claim 1, characterized in that, In the event of network unavailability or instability at the production site, defect identification results and corresponding image data are stored in a local database for caching, and incremental synchronization is performed after the network is restored, including: S701. Obtain fabric defect information and corresponding image data output by the real-time defect identification step at the production site, and record the batch number and process sheet information. S702. Detect the network status at the production site. When the network is unavailable or unstable, store the fabric defect information and corresponding image data, along with the batch number and process sheet information, in the local database for caching. S703. After the network is restored, determine the synchronization status between the data cached in the local database and the data in the remote management system based on the fabric defect information, batch number, and process sheet association identifier. S704. Perform incremental synchronization on data that has not yet been synchronized in the local database, including uploading fabric defect information, image data, and related batch and process sheet information. S705. During the synchronization process, data conflicts or duplicate records are automatically checked and processed. The check strategy includes unique identifier comparison, timestamp comparison and data integrity check.

9. The method for real-time detection and adaptive correction of fabric defects based on deep learning as described in claim 1, characterized in that, Based on historical quality inspection data accumulated from defect identification results and synchronously collected production environment data, a time-series analysis model is constructed to predict potential quality risks and generate corresponding risk warning information, including: S801. Obtain the defect identification results output by the real-time defect identification step and accumulate them to form a historical quality inspection dataset; S802. Collect production environment data corresponding to the historical quality inspection data, including temperature, humidity and other production process-related parameters, and associate the production environment data with the historical quality inspection data based on the timestamp or batch number of the defect record. S803. Perform data preprocessing on the historical quality inspection data and production environment data, including missing value processing, outlier filtering and numerical normalization. S804. Input the preprocessed historical quality inspection data and production environment data into a pre-trained time series analysis model to generate a corresponding quality risk prediction vector. S805. Output the quality risk prediction vector and record the association information with the corresponding batch and process sheet.

10. The method for real-time detection and adaptive correction of fabric defects based on deep learning as described in claim 1, characterized in that, The defect identification results and quality risk prediction results are automatically associated with the corresponding process sheets and production batches to achieve cross-process traceability of quality problems. The analysis results are also synchronized to the production management module and the trade management module, including: S901. Obtain the defect identification results output by the real-time defect identification step and the quality risk prediction results generated based on historical quality inspection data and production environment data. S902. Based on the batch information and process sheet identifier in the defect identification result and the quality risk prediction result, automatically associate the defect identification result and the quality risk prediction result with the corresponding process sheet and production batch. S903. Synchronize the associated defect identification results and quality risk prediction results to the production management module and trade management module; S904. During the synchronization process, data conflicts or duplicate records are automatically checked and processed. The check strategy includes unique identifier comparison, timestamp comparison and data integrity check.

Images