Full-process fabric shrinkage dynamic calibration method and system based on edge computing

Abstract

The present invention relates to the technical field of dynamic process calibration, specifically a full-process dynamic calibration method and system for fabric shrinkage based on edge computing. The method includes the following steps: setting and sensing: collecting sample information and processing requirement instructions of the fabric to be tested; decision basis generation: identifying the fabric to be tested to obtain initial characteristic parameters; control instruction generation: obtaining decision basis parameters based on the sample information and initial characteristic parameters; obtaining process control parameters based on the decision basis parameters and the processing requirement instructions; process monitoring: processing the fabric to be tested based on the process control parameters, and collecting test environment data and test dynamic parameters; identifying the test dynamic parameters to obtain fabric shrinkage data; and feedback and calibration: identifying and calibrating the process control parameters based on the obtained data to obtain calibration process parameters. The present invention achieves dynamic process calibration by identifying fabric shrinkage.

Description

Technical Field

[0001] The present invention relates to the technical field of process dynamic calibration, and specifically to a full-process fabric shrinkage dynamic calibration method and system based on edge computing. Background Art

[0002] The control of shrinkage during the processing of textiles directly affects the processing quality of the products.

[0003] The control systems currently used in this scenario generally lack an effective process model that can accurately describe the complex and nonlinear relationship between numerous process parameters and the final shrinkage rate; their control strategies generally remain in open-loop mode, that is, setting a fixed set of process parameters for the entire processing flow based on offline experience or test results.

[0004] Lacking a reliable predictive model, control systems are unable to predict the impact of parameter adjustments on the results, and are therefore forced to rely on fixed, static instructions. More importantly, these open-loop systems typically lack the ability to perceive the key characteristics of the controlled object in real time, making it difficult to effectively identify material fluctuations or environmental disturbances that occur during the process.

[0005] Since the current system cannot form an effective real-time closed-loop feedback, there is no real-time deviation signal and adjustment means, and it is impossible to actively compensate for various disturbances, which ultimately leads to low control accuracy and poor consistency of fabric shrinkage rate.

[0006] To this end, a full-process fabric shrinkage dynamic calibration method and system based on edge computing is proposed. Summary of the Invention

[0007] The purpose of the present invention is to provide a full-process fabric shrinkage dynamic calibration method and system based on edge computing, which can realize dynamic calibration of the processing technology by identifying the fabric shrinkage rate.

[0008] To achieve the above object, the present invention provides the following technical solutions:

[0009] The full-process dynamic calibration method and system for fabric shrinkage based on edge computing include:

[0010] Setting and sensing: Collect sample information and processing requirements of the fabric to be tested;

[0011] Decision basis generation: Identify the fabric to be tested and obtain initial characteristic parameters; identify based on sample information and initial characteristic parameters to obtain decision basis parameters;

[0012] Control instruction generation: Based on the decision-making parameters and processing requirement instructions, process control parameters are obtained; the process control parameters include at least one operation flow of the full-process processing solution and the corresponding standard environment data;

[0013] Process monitoring: The fabric to be tested is processed based on process control parameters, and the test environment data and test dynamic parameters are collected using sensors. The test dynamic parameters are characterized by edge computing to obtain fabric shrinkage data.

[0014] Feedback and calibration: Identification is performed based on process control parameters, fabric shrinkage data, decision-making parameters and processing requirement instructions, and the identification process is calibrated through the difference between test environment data and standard environment data to obtain calibrated process parameters.

[0015] The sample information includes physical information, component information, appearance information and structural information;

[0016] The physical information includes fabric shape, fabric weight, fabric thickness, and fabric density; the component information includes fiber type; the appearance information includes color distribution; and the structural information includes weaving method and stitch density.

[0017] The processing requirement instructions are the condition restrictions and quality requirements for processing the fabric to be tested;

[0018] The initial characteristic parameters include fabric texture data.

[0019] A fabric safety boundary recognition model is constructed based on a gradient boosting decision tree, and the fabric safety boundary recognition model is trained based on a fabric damage dataset; the fabric damage dataset includes fabric sample information, fabric initial characteristic parameters, fabric operation process parameters, fabric processing environment parameters, and fabric damage parameters;

[0020] The fabric operation process parameters are a combination of operation steps for processing a fabric sample, wherein the operation steps include processing content, processing equipment type, processing equipment parameters, and chemical additive information; the fabric damage parameters include a shrinkage rate label of the fabric sample after processing according to the fabric operation process parameters;

[0021] The sample information and initial characteristic parameters of the fabric to be tested are input into the fabric safety boundary recognition model to identify the fabric damage prediction parameters under different fabric operation process parameters. Threshold screening is performed based on the fabric damage prediction parameters, and the fabric operation process parameters that meet the screening conditions and the fabric operation process parameters that do not meet the screening conditions are used as decision-making parameters.

[0022] A process parameter generation model is constructed based on the process path planning method of constraint satisfaction and multi-objective optimization;

[0023] The process parameter generation model includes a demand instruction decomposition layer and a process parameter generation layer;

[0024] The demand instruction decomposition layer obtains processing rules based on the processing conditions and restrictions for the fabric to be tested in the processing demand instruction; and obtains processing targets based on the quality requirements for the fabric to be tested in the processing demand instruction;

[0025] The process parameter generation layer uses the processing target as the optimization target and the processing rules as the constraint conditions, and performs identification based on the decision basis parameters to obtain process control parameters.

[0026] The method for obtaining the fabric shrinkage data is as follows:

[0027] Identify the process of fabric processing to be tested based on process control parameters to obtain test environment data and test dynamic parameters; the test dynamic parameters include fabric shape dynamic parameters, fabric appearance dynamic parameters, fabric texture dynamic parameters, and fabric temperature dynamic parameters;

[0028] Based on the time distribution of each operation step in the processing operation process, the test dynamic parameters and test environment data are divided to obtain the operation step dynamic parameters and operation step environment parameters;

[0029] Extracting features of the dynamic parameters of the operation steps to obtain dynamic feature parameters; the dynamic feature parameters include dynamic features of fabric shape, dynamic features of fabric appearance, dynamic features of fabric texture and dynamic features of fabric temperature;

[0030] Inputting the dynamic feature parameters and corresponding operation step environment parameters into a shrinkage feature recognition model deployed on an edge computing node to calculate and output structured fabric shrinkage data;

[0031] The fabric shrinkage data includes the instantaneous shrinkage rate and the cumulative shrinkage rate of the fabric at different processing time points.

[0032] The process of obtaining the calibration process parameters includes: correcting the fabric shrinkage data based on the difference between the standard environment data of the process control parameters and the test environment data in actual processing to obtain the calibration shrinkage data;

[0033] Based on the calibration shrinkage data, identification is performed to obtain the operation abnormality point; based on the decision basis parameters and the processing requirement instructions, the process control parameters at the operation abnormality point are calibrated to obtain the calibrated process parameters.

[0034] The full-process fabric shrinkage dynamic calibration system based on edge computing includes:

[0035] The setting and perception module collects sample information and processing requirements of the fabric to be tested;

[0036] The decision basis generation module identifies the fabric to be tested and obtains initial characteristic parameters; the initial characteristic parameters include fabric texture data; based on the sample information and the initial characteristic parameters, identification is performed to obtain decision basis parameters;

[0037] A control instruction generation module identifies the decision-making basis parameters and the processing requirement instructions to obtain process control parameters; the process control parameters include at least one operation flow of a full-process processing solution and corresponding standard environment data;

[0038] The processing monitoring module processes the fabric to be tested based on process control parameters and collects test environment data and test dynamic parameters based on sensors. It also performs feature recognition on the test dynamic parameters based on edge computing to obtain fabric shrinkage data.

[0039] The feedback and calibration module performs identification based on process control parameters, fabric shrinkage data, decision-making parameters and processing requirement instructions. It calibrates the identification process by the difference between the test environment data and the standard environment data to obtain the calibrated process parameters.

[0040] Compared with the prior art, the present invention has the following beneficial effects:

[0041] 1. The present invention obtains a fabric safety boundary recognition model based on gradient boosting decision tree training. The fabric safety boundary recognition model is used to identify sample information and initial characteristic parameters of the fabric to be tested, thereby realizing the prediction of process risks and proactively avoiding a large number of unreasonable process settings before processing occurs, greatly improving the safety and effectiveness of the process plan.

[0042] 2. The present invention identifies the processing rules and processing targets based on the conditional restrictions and quality requirements of the fabric to be tested in the processing demand instructions; then uses the processing targets as optimization targets and the processing rules as constraints to identify the process control parameters based on the decision-making parameters; and realizes the identification of the optimal solution within the safety range of the decision-making parameters.

[0043] 3. The present invention constructs a shrinkage feature recognition model based on a convolutional neural network-long short-term memory network hybrid model. By deploying the shrinkage feature recognition model at the edge, it solves the delay bottleneck of massive data processing in industrial sites, realizes sub-second real-time analysis of fabric shrinkage status, and converts invisible internal dimensional changes into continuous, quantified data curves, accurately and quickly identifying fabric shrinkage data.

[0044] 4. The present invention corrects the fabric shrinkage data based on the degree of difference between the standard environmental data of the process control parameters and the test environmental data in actual processing to obtain calibrated shrinkage data; identifies the operation abnormality point based on the calibrated shrinkage data; calibrates the process control parameters at the operation abnormality point based on the decision basis parameters and the processing requirement instructions to obtain calibrated process parameters; and achieves further optimization of the process control parameters. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] Figure 1 Schematic diagram of the process of the full-process dynamic calibration method of fabric shrinkage based on edge computing of the present invention;

[0046] Figure 2 A schematic diagram of the structure of the process parameter generation model of the present invention;

[0047] Figure 3 This is a schematic diagram of the fabric shrinkage data acquisition process of the present invention;

[0048] Figure 4 This is a structural schematic diagram of the full-process fabric shrinkage dynamic calibration system based on edge computing of the present invention. DETAILED DESCRIPTION

[0049] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0050] During the dyeing and finishing process of textiles, fabrics typically undergo multiple steps, including desizing, scouring, bleaching, mercerizing, dyeing, printing, and shaping. During these wet and hot processes, the fabric inevitably undergoes dimensional changes, known as shrinkage, due to the moisture absorption and expansion of fibers, the disentanglement and rearrangement of molecular chains, and the mechanical tension applied during processing. Controlling shrinkage is a core indicator of fabric quality, especially for high-value-added apparel fabrics, and directly affects the subsequent cutting and sewing precision, as well as the dimensional stability of the final garment.

[0051] Therefore, it is necessary to monitor and optimize the shrinkage during fabric processing.

[0052] Example 1:

[0053] The present invention proposes a full-process dynamic calibration method for fabric shrinkage based on edge computing, the process of which is as follows: Figure 1 As shown, including:

[0054] Setting and sensing: Collect sample information and processing requirements of the fabric to be tested;

[0055] Decision basis generation: Identify the fabric to be tested and obtain initial characteristic parameters; identify based on sample information and initial characteristic parameters to obtain decision basis parameters;

[0056] Control instruction generation: Based on the decision-making parameters and processing requirement instructions, process control parameters are obtained; the process control parameters include at least one operation flow of the full-process processing solution and the corresponding standard environment data;

[0057] Process monitoring: The fabric to be tested is processed based on process control parameters, and the test environment data and test dynamic parameters are collected using sensors. The test dynamic parameters are characterized by edge computing to obtain fabric shrinkage data.

[0058] Feedback and calibration: Identification is performed based on process control parameters, fabric shrinkage data, decision-making parameters and processing requirement instructions, and the identification process is calibrated through the difference between test environment data and standard environment data to obtain calibrated process parameters.

[0059] Preferably, the sample information includes physical information, component information, appearance information and structural information;

[0060] The physical information includes fabric shape, fabric weight, fabric thickness, and fabric density; the component information includes fiber type; the appearance information includes color distribution; and the structural information includes weaving method and stitch density.

[0061] The processing requirement instructions are the condition restrictions and quality requirements for processing the fabric to be tested;

[0062] The initial characteristic parameters include fabric texture data, which is obtained based on sensor detection and analysis. Furthermore, in order to improve the understanding of the initial characteristics of the fabric, spectral data and infrared data of the fabric can also be collected.

[0063] The aforementioned conditions and restrictions are the restrictive conditions imposed by the production entity on the fabric processing process:

[0064] For example, the same operation step can be achieved through a variety of different processing equipment. If the production entity is equipped with only a specific type of processing equipment, then it can be specified as a certain type of equipment that must be used as a conditional restriction. In the subsequent process combination and optimization, the specified equipment must be used as one of the constraints. Conditional restrictions also include prohibited chemical additives, the upper limit of the total processing time, etc.

[0065] The quality requirements are determined based on the testing standards of the fabric to be tested after processing, including the final shrinkage of the fabric; further, color fastness grade requirements, processing costs, etc. can also be set as quality requirements.

[0066] The present invention provides rich, reliable and multi-dimensional input features for subsequent intelligent decision-making by comprehensively, hierarchically and structuredly defining and acquiring sample information, processing requirements and initial characteristic parameters collected online.

[0067] A fabric safety boundary recognition model is constructed based on a gradient boosting decision tree, and the fabric safety boundary recognition model is trained based on a fabric damage dataset; the fabric damage dataset includes fabric sample information, fabric initial characteristic parameters, fabric operation process parameters, fabric processing environment parameters, and fabric damage parameters;

[0068] The fabric operation process parameters are a combination of operation steps for processing a fabric sample, wherein the operation steps include processing content, processing equipment type, processing equipment parameters, and chemical additive information. The fabric damage parameters include a shrinkage rate label of the fabric sample after processing using the fabric operation process parameters. Furthermore, the fabric damage parameters can dynamically select data dimensions based on the quality requirements in the processing demand instructions.

[0069] The sample information and initial characteristic parameters of the fabric to be tested are input into the fabric safety boundary recognition model to identify the fabric damage prediction parameters under different fabric operation process parameters. Threshold screening is performed based on the fabric damage prediction parameters, and the fabric operation process parameters that meet the screening conditions and the fabric operation process parameters that do not meet the screening conditions are used as decision-making parameters.

[0070] Gradient boosting decision trees iteratively train a series of weak learners, with each new tree fitting the residual between the previous tree's prediction and the true value. The final model is a weighted sum of the predictions from all trees, resulting in a powerful ensemble model. It excels at processing heterogeneous data containing both numerical and categorical features, effectively capturing complex nonlinear relationships between features and targets.

[0071] The process of constructing a fabric safety boundary recognition model based on the gradient boosting decision tree includes:

[0072] Data preprocessing: Standardize the numerical features in the fabric damage dataset, such as fabric weight, fabric thickness, and fabric density; and perform one-hot encoding on categorical features, such as weaving method and fiber type.

[0073] Dataset partitioning: Divide the dataset into training set, validation set, and test set in a ratio of 8:1:1.

[0074] Model training and tuning: On the training set, the model is trained using mean squared error as the loss function. During training, the validation set is used to tune key hyperparameters of GBDT, such as the learning rate, number of trees, maximum depth of each tree, and subsampling ratio, through grid search or Bayesian optimization to prevent overfitting and find the optimal performance point.

[0075] Model evaluation: Evaluate the performance of the final model on the test set, using metrics such as R² score and mean absolute error (MAE) to ensure its generalization ability.

[0076] The model uses the sample information and initial characteristic parameters of the fabric to be tested as input, programmatically generating a large number of fabric operation process parameters covering the typical processing range. These combined parameters, along with the fabric information, are then fed into the fabric safety boundary identification model. For each combination, the model outputs a fabric damage prediction parameter, namely the predicted shrinkage rate. When the fabric damage parameter includes multiple indicators, multiple corresponding indicators are also predicted simultaneously.

[0077] Finally, a safety threshold is set. Any combination of operating process parameters whose predicted values ​​fall within this range is considered to meet the screening conditions; otherwise, it is considered to not meet the screening conditions. The combination of these two types of parameters together constitutes the decision-making basis parameters for the next step.

[0078] The present invention obtains a fabric safety boundary recognition model based on gradient boosting decision tree training. The fabric safety boundary recognition model is used to identify sample information and initial characteristic parameters of the fabric to be tested, thereby realizing the prediction of process risks and actively avoiding a large number of unreasonable process settings before processing occurs, greatly improving the safety and effectiveness of the process plan.

[0079] A process parameter generation model is constructed based on the process path planning method of constraint satisfaction and multi-objective optimization;

[0080] The structure of the process parameter generation model is as follows Figure 2 As shown, it includes the demand instruction decomposition layer and the process parameter generation layer;

[0081] The demand instruction decomposition layer obtains processing rules based on the processing conditions and restrictions for the fabric to be tested in the processing demand instruction; and obtains processing targets based on the quality requirements for the fabric to be tested in the processing demand instruction;

[0082] This layer is a parsing engine responsible for converting semi-structured or natural language instructions into a machine-understandable form; for example, "specify the use of a certain model of equipment" will be converted into an equality constraint; "the upper limit of the total processing time" will be converted into an inequality constraint.

[0083] The process parameter generation layer uses the processing target as the optimization target and the processing rules as the constraint conditions, and performs identification based on the decision basis parameters to obtain process control parameters.

[0084] The process parameter generation layer uses a non-dominated sorting genetic algorithm with an elitist strategy as a solver. This algorithm simulates biological evolution and iteratively searches within the feasible solution space through selection, crossover, and mutation operations. Its core advantages are:

[0085] Non-dominated sorting: It can handle multiple objectives simultaneously and find a set of Pareto optimal solutions in which no single objective can be improved without sacrificing other objectives.

[0086] Crowding calculation: ensures that the optimal solution set is evenly distributed in the target space, providing decision makers with diverse options.

[0087] Algorithm process: Initialize the population: Randomly generate a set of initial process plans (chromosomes) within the safety range defined by the decision-making parameters. Each plan represents a complete processing flow. Fitness evaluation: For each plan, calculate the function value of each corresponding processing target (such as cost, predicted shrinkage rate). Selection, crossover, mutation: Through tournament selection, simulated binary crossover and polynomial mutation and other genetic operators, simulate the biological evolution process and generate a new offspring population. Elite strategy and sorting: Merge the parent and offspring generations, perform non-dominated sorting and crowding calculations, and select the best individuals to enter the next generation. Iteration: Repeat the identification until the preset number of iterations is reached or the solution set converges.

[0088] After the algorithm runs, it outputs a Pareto front. For example, a two-dimensional graph illustrates the trade-offs between quality deviation and processing cost among different processing options. The optimal option can be selected based on the current production task priority. For example, this order requires extreme quality regardless of cost, or this order requires fast delivery with satisfactory quality. This solution serves as the final process control parameters, including detailed operating procedures and standard environmental data for each process step.

[0089] The present invention identifies processing rules and processing targets based on the conditional restrictions and quality requirements of the fabric to be tested according to the processing demand instructions; then uses the processing targets as optimization targets and the processing rules as constraints to identify process control parameters based on decision-making parameters, thereby realizing the identification of the optimal solution within the safety range of the decision-making parameters.

[0090] The process of obtaining the fabric shrinkage data is as follows: Figure 3 As shown:

[0091] Identify the process of fabric processing to be tested based on process control parameters to obtain test environment data and test dynamic parameters; the test dynamic parameters include fabric shape dynamic parameters, fabric appearance dynamic parameters, fabric texture dynamic parameters, and fabric temperature dynamic parameters;

[0092] Based on the time distribution of each operation step in the processing operation process, the test dynamic parameters and test environment data are divided to obtain the operation step dynamic parameters and operation step environment parameters;

[0093] Extracting features of the dynamic parameters of the operation steps to obtain dynamic feature parameters; the dynamic feature parameters include dynamic features of fabric shape, dynamic features of fabric appearance, dynamic features of fabric texture and dynamic features of fabric temperature;

[0094] Inputting the dynamic feature parameters and corresponding operation step environment parameters into a shrinkage feature recognition model deployed on an edge computing node to calculate and output structured fabric shrinkage data;

[0095] The fabric shrinkage data includes the instantaneous shrinkage rate and the cumulative shrinkage rate of the fabric at different processing time points.

[0096] In order to realize the rapid identification of data during the processing of the fabric to be tested, this application deploys visual, infrared and other sensors at the entrances and exits of key equipment to collect raw data streams.

[0097] Collect and test dynamic parameters: including fabric shape dynamic parameters, continuous width and length changes captured by a line array camera; fabric appearance dynamic parameters, color distribution and changes on the fabric surface; fabric texture dynamic parameters, continuous surface images captured; fabric temperature dynamic parameters, surface temperature distribution captured by an infrared thermal imager.

[0098] Test environment data, including temperature, humidity, wind speed, and pH during processing.

[0099] Data segmentation and feature extraction: The edge computing node divides the continuous data stream into segments corresponding to the operation steps based on the time distribution of the operation process; then, the feature extraction of the dynamic parameters of each segment is performed to obtain dynamic feature parameters.

[0100] The shrinkage feature recognition model is deployed on edge computing nodes close to the sensors, such as the NVIDIA Jetson series embedded GPU platform. Dynamic feature parameters and synchronously collected operating step environmental parameters are calculated locally without uploading to the cloud. The model then outputs structured fabric shrinkage data that is then uploaded to a central system.

[0101] This embodiment uses a lightweight convolutional neural network-long short-term memory network (CNN-LSTM) hybrid model to construct a shrinkage feature recognition model.

[0102] The input of the model is a sequence of time steps. The data of each time step is composed of two parts: the CNN feature map or flattened feature vector extracted from the image at that moment; and the vector composed of sensor readings at that moment, including temperature, humidity, etc.

[0103] CNN module: This module processes image data. It consists of convolutional layers (Conv2D, using the ReLU activation function) and pooling layers (MaxPooling2D), which are responsible for automatically extracting spatial features in the image, such as changes in texture and yarn structure.

[0104] LSTM module: The spatial features extracted by the CNN are combined with other sensor data and fed into the LSTM layer, which consists of a two-layer 128-unit LSTM network. LSTM is a special type of recurrent neural network that can learn long-term dependencies in time series data and understand dynamic processes such as how temperature increases and then decreases while textures become denser.

[0105] Output layer: The output of the LSTM layer passes through one or more fully connected layers and finally a single-neuron linear activation output layer, which directly regresses and predicts the instantaneous shrinkage rate at that moment. By accumulating the instantaneous shrinkage rates, the cumulative shrinkage rate can be obtained.

[0106] The model's training dataset is obtained by accurately measuring the actual shrinkage of fabric at each point in the fabric processing process under experimental conditions using a high-precision laser length gauge and other equipment. This data is then paired with synchronously collected sensor data sequences to form data pairs. Training is performed end-to-end using optimizers such as Adam, with mean squared error as the loss function.

[0107] The present invention constructs a shrinkage feature recognition model based on a convolutional neural network-long short-term memory network hybrid model. By deploying the shrinkage feature recognition model at the edge, it solves the delay bottleneck of massive data processing in industrial sites, realizes sub-second real-time analysis of fabric shrinkage status, and converts invisible internal dimensional changes into continuous, quantified data curves, thus accurately and quickly identifying fabric shrinkage data.

[0108] The process of obtaining the calibration process parameters includes: correcting the fabric shrinkage data based on the difference between the standard environment data of the process control parameters and the test environment data in actual processing to obtain the calibration shrinkage data;

[0109] Based on the calibration shrinkage data, identification is performed to obtain the operation abnormality point; based on the decision basis parameters and the processing requirement instructions, the process control parameters at the operation abnormality point are calibrated to obtain the calibrated process parameters.

[0110] During the processing according to the process control parameters, due to the equipment wear of the production entity, the quality of the chemical additives used, etc., there will be differences between the actual processing effect of the process control parameters and the theoretical target. Therefore, it is necessary to adjust the process in combination with the dynamic parameters in the actual processing process.

[0111] The specific calibration process includes:

[0112] Identifying operational anomalies: First, due to environmental factors and human intervention during the processing process, test environment data may fluctuate. Fabric shrinkage data needs to be corrected based on the difference between the test environment data and the standard environment data. Training data can be collected and used to train a neural network model for data correction. The calibrated shrinkage data is compared in real time with the target data set based on process control parameter identification. When the data deviation exceeds a preset tolerance threshold, the moment is identified as an operational anomaly.

[0113] Local parameter adjustment: Identify the degree of abnormality and cause of abnormal operation points. The degree of abnormality is determined based on the calibration shrinkage data, and the cause of the abnormality is identified based on the data characteristics of process control parameters and test dynamic parameters. Adjust the optimization goal of this local planning based on the degree of abnormality and cause of the abnormal operation point, such as minimizing the deviation between the future shrinkage trajectory and the original target trajectory. Further determine the range of parameters that need to be adjusted. For example, if the shrinkage is found to be too fast halfway through the shaping process, then the adjustable variables are the vehicle speed, overfeed rate, wind speed and other parameters of the subsequent shaping process.

[0114] Local constraints: Call the decision basis generation module to load the pre-calculated decision basis parameters for the current fabric to be processed, and clarify the safe operating range of various process parameters under the current process; in the subsequent optimization calculation, the value range of all parameters to be adjusted is strictly limited within the safe operating range.

[0115] Local optimization: Use process parameters to generate re-optimization of the solution local area in the model.

[0116] Furthermore, causal inference techniques, such as Bayesian networks or Do-calculus, can be incorporated into the identification of operational anomalies. This approach leverages high-dimensional time-series data from the processing process, including all dynamic test parameters and test environment data, to construct a dynamic causal relationship diagram. When an anomaly occurs, the system goes beyond simply comparing data differences and instead uses causal inference algorithms to calculate the causal contribution of each potential factor to the current shrinkage anomaly. Ultimately, the system can pinpoint the root cause with a high probability. For example, a report might output: "Warning: The current weft shrinkage exceeds the standard by 0.5%. Causal analysis indicates an 85% probability that the actual oven temperature is 5°C lower than the setpoint." This root cause diagnosis not only guides immediate corrections but, more importantly, automatically stores the "abnormal event-root cause-solution" case study in a structured process knowledge base. This allows for long-term optimization of equipment maintenance strategies and raw material incoming standards, and can also continuously optimize the fabric safety boundary identification model above, achieving knowledge accumulation and system self-evolution.

[0117] Furthermore, the current calibration scheme mainly focuses on the adjustment of physical parameters such as temperature and speed. However, the type and concentration of chemical additives are also key factors affecting shrinkage and even other fabric qualities. This innovative design extends the control loop to the chemical reaction level. In specific implementation, the concentration changes of key chemical components in the liquid environment and the absorption of additives by the fabric surface are monitored in real time. The shrinkage feature recognition model will be expanded to increase its sensitivity to chemical parameters so that it can predict the reaction trend of the fabric under the current chemical environment. At the same time, the system adds a "chemical additive control module", which can accurately control the start and stop and flow rate of the titration pump in real time. When the system predicts that abnormal shrinkage or color difference may occur due to excessive consumption of additives, it will automatically issue instructions to replenish the corresponding additives to maintain the dynamic balance of the chemical environment.

[0118] The present invention corrects fabric shrinkage data based on the degree of difference between standard environmental data of process control parameters and test environmental data in actual processing to obtain calibrated shrinkage data; identifies operation abnormal points based on the calibrated shrinkage data; calibrates the process control parameters at the operation abnormal points based on decision-making basis parameters and processing requirement instructions to obtain calibrated process parameters; and achieves further optimization of the process control parameters.

[0119] The present invention also proposes a full-process fabric shrinkage dynamic calibration system based on edge computing, the structure of which is as follows: Figure 4 As shown, including:

[0120] The setting and perception module collects sample information and processing requirements of the fabric to be tested;

[0121] The decision basis generation module identifies the fabric to be tested and obtains initial characteristic parameters; the initial characteristic parameters include fabric texture data; based on the sample information and the initial characteristic parameters, identification is performed to obtain decision basis parameters;

[0122] A control instruction generation module identifies the decision-making basis parameters and the processing requirement instructions to obtain process control parameters; the process control parameters include the operation process and standard environment data of at least one full-process processing solution;

[0123] The processing monitoring module processes the fabric to be tested based on process control parameters and collects test environment data and test dynamic parameters based on sensors. It also performs feature recognition on the test dynamic parameters based on edge computing to obtain fabric shrinkage data.

[0124] The feedback and calibration module performs identification based on process control parameters, fabric shrinkage data, decision-making parameters and processing requirement instructions. It calibrates the identification process by the difference between the test environment data and the standard environment data to obtain the calibrated process parameters.

[0125] The present invention collects sample information and processing requirement instructions of a fabric to be tested; identifies the fabric to be tested to obtain initial characteristic parameters; obtains decision-making basis parameters based on the sample information and initial characteristic parameters; obtains process control parameters based on the decision-making basis parameters and the processing requirement instructions; processes the fabric to be tested based on the process control parameters, and collects test environment data and test dynamic parameters; identifies the test dynamic parameters to obtain fabric shrinkage data; and identifies and calibrates the process control parameters based on the obtained data to obtain calibrated process parameters. The present invention achieves dynamic process calibration by identifying fabric shrinkage.

[0126] Furthermore, model training requires a large amount of data. However, the limited size and diversity of fabric damage datasets from individual textile factories or testing units restricts the generalization capabilities of fabric safety boundary recognition models and shrinkage feature recognition models. However, production data between different factories is considered core commercial confidentiality and cannot be directly shared.

[0127] The present invention is designed to use a federated learning framework to solve this problem. Multiple factories that have deployed the present invention can form a "learning alliance". During model training, the central server is only responsible for distributing the initial model and aggregating the updated model parameters. The original data of each factory will never leave the local edge node or server. Each factory uses its own data to train the model locally, and then only uploads the encrypted parameter updates of the model to the central server. The central server securely aggregates the parameter updates from multiple factories to generate a more powerful global model, and then distributes this model to all participants. In this way, each factory contributes to the optimization of the model and can enjoy a more robust model trained with massive and diverse data without exposing any sensitive production data, realizing "data available but invisible" and resolving the contradiction between multi-agent collaboration and data privacy protection.

[0128] Example 2:

[0129] The present invention proposes a full-process dynamic calibration method for fabric shrinkage based on edge computing, including:

[0130] Setting and sensing: Collect sample information and processing requirements of the fabric to be tested;

[0131] Decision basis generation: Identify the fabric to be tested and obtain initial characteristic parameters; identify based on sample information and initial characteristic parameters to obtain decision basis parameters;

[0132] Control instruction generation: Based on the decision-making basis parameters and processing requirement instructions, process control parameters are obtained; the process control parameters include the operation process and standard environment data of at least one full-process processing solution;

[0133] Process monitoring: The fabric to be tested is processed based on process control parameters, and the test environment data and test dynamic parameters are collected using sensors. The test dynamic parameters are characterized by edge computing to obtain fabric shrinkage data.

[0134] Feedback and calibration: Identification is performed based on process control parameters, fabric shrinkage data, decision-making parameters and processing requirement instructions, and the identification process is calibrated through the difference between test environment data and standard environment data to obtain calibrated process parameters.

[0135] The sample information includes physical information, component information, appearance information and structural information;

[0136] The physical information includes fabric shape, fabric weight, fabric thickness, and fabric density; the component information includes fiber type; the appearance information includes color distribution; and the structural information includes weaving method and stitch density.

[0137] The processing requirement instructions are the condition restrictions and quality requirements for processing the fabric to be tested;

[0138] The initial characteristic parameters include fabric texture data.

[0139] The present invention provides rich, reliable and multi-dimensional input features for subsequent intelligent decision-making by comprehensively, hierarchically and structuredly defining and acquiring sample information, processing requirements and initial characteristic parameters collected online.

[0140] A fabric safety boundary recognition model is constructed based on a gradient boosting decision tree, and the fabric safety boundary recognition model is trained based on a fabric damage dataset; the fabric damage dataset includes fabric sample information, fabric initial characteristic parameters, fabric operation process parameters, fabric processing environment parameters, and fabric damage parameters;

[0141] The fabric operation process parameters are a combination of operation steps for processing a fabric sample, wherein the operation steps include processing content, processing equipment type, processing equipment parameters, and chemical additive information; the fabric damage parameters include a shrinkage rate label of the fabric sample after processing according to the fabric operation process parameters;

[0142] The sample information and initial characteristic parameters of the fabric to be tested are input into the fabric safety boundary recognition model to identify the fabric damage prediction parameters under different fabric operation process parameters. Threshold screening is performed based on the fabric damage prediction parameters, and the fabric operation process parameters that meet the screening conditions and the fabric operation process parameters that do not meet the screening conditions are used as decision-making parameters.

[0143] The present invention obtains a fabric safety boundary recognition model based on gradient boosting decision tree training. The fabric safety boundary recognition model is used to identify sample information and initial characteristic parameters of the fabric to be tested, thereby realizing the prediction of process risks and actively avoiding a large number of unreasonable process settings before processing occurs, greatly improving the safety and effectiveness of the process plan.

[0144] A process parameter generation model is constructed based on the process path planning method of constraint satisfaction and multi-objective optimization;

[0145] The process parameter generation model includes a demand instruction decomposition layer and a process parameter generation layer;

[0146] The demand instruction decomposition layer obtains processing rules based on the processing conditions and restrictions for the fabric to be tested in the processing demand instruction; and obtains processing targets based on the quality requirements for the fabric to be tested in the processing demand instruction;

[0147] The process parameter generation layer uses the processing target as the optimization target and the processing rules as the constraint conditions, and performs identification based on the decision basis parameters to obtain process control parameters.

[0148] The present invention identifies processing rules and processing targets based on the conditional restrictions and quality requirements of the fabric to be tested according to the processing demand instructions; then uses the processing targets as optimization targets and the processing rules as constraints to identify process control parameters based on decision-making parameters, thereby realizing the identification of the optimal solution within the safety range of the decision-making parameters.

[0149] The method for obtaining the fabric shrinkage data is as follows:

[0150] Identify the process of fabric processing to be tested based on process control parameters to obtain test environment data and test dynamic parameters; the test dynamic parameters include fabric shape dynamic parameters, fabric appearance dynamic parameters, fabric texture dynamic parameters, and fabric temperature dynamic parameters;

[0151] Based on the time distribution of each operation step in the processing operation process, the test dynamic parameters and test environment data are divided to obtain the operation step dynamic parameters and operation step environment parameters;

[0152] Extracting features of the dynamic parameters of the operation steps to obtain dynamic feature parameters; the dynamic feature parameters include dynamic features of fabric shape, dynamic features of fabric appearance, dynamic features of fabric texture and dynamic features of fabric temperature;

[0153] Inputting the dynamic feature parameters and corresponding operation step environment parameters into a shrinkage feature recognition model deployed on an edge computing node to calculate and output structured fabric shrinkage data;

[0154] The present invention constructs a shrinkage feature recognition model based on a convolutional neural network-long short-term memory network hybrid model. By deploying the shrinkage feature recognition model at the edge, it solves the delay bottleneck of massive data processing in industrial sites, realizes sub-second real-time analysis of fabric shrinkage status, and converts invisible internal dimensional changes into continuous, quantified data curves, thus accurately and quickly identifying fabric shrinkage data.

[0155] The fabric shrinkage data includes the instantaneous shrinkage rate and the cumulative shrinkage rate of the fabric at different processing time points.

[0156] The process of obtaining the calibration process parameters includes: correcting the fabric shrinkage data based on the difference between the standard environment data of the process control parameters and the test environment data in actual processing to obtain the calibration shrinkage data;

[0157] Based on the calibration shrinkage data, identification is performed to obtain the operation abnormality point; based on the decision basis parameters and the processing requirement instructions, the process control parameters at the operation abnormality point are calibrated to obtain the calibrated process parameters.

[0158] The present invention corrects fabric shrinkage data based on the degree of difference between standard environmental data of process control parameters and test environmental data in actual processing to obtain calibrated shrinkage data; identifies operation abnormal points based on the calibrated shrinkage data; calibrates the process control parameters at the operation abnormal points based on decision-making basis parameters and processing requirement instructions to obtain calibrated process parameters; and achieves further optimization of the process control parameters.

[0159] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. A full-process dynamic calibration method for fabric shrinkage based on edge computing, characterized in that: include: Setting and sensing: Collect sample information and processing requirements of the fabric to be tested; Decision basis generation: Identify the fabric to be tested and obtain initial characteristic parameters; Identify based on sample information and initial characteristic parameters to obtain decision-making parameters; The specific process includes: A fabric safety boundary recognition model is constructed based on a gradient boosting decision tree, and the fabric safety boundary recognition model is trained based on a fabric damage dataset; the fabric damage dataset includes fabric sample information, fabric initial characteristic parameters, fabric operation process parameters, fabric processing environment parameters, and fabric damage parameters; The fabric operation process parameters are a combination of operation steps for processing a fabric sample, wherein the operation steps include processing content, processing equipment type, processing equipment parameters, and chemical additive information; the fabric damage parameters include a shrinkage rate label of the fabric sample after processing according to the fabric operation process parameters; The sample information and initial characteristic parameters of the fabric to be tested are input into the fabric safety boundary recognition model to identify the fabric damage prediction parameters under different fabric operation process parameters; threshold screening is performed based on the fabric damage prediction parameters, and the fabric operation process parameters that meet the screening conditions and the fabric operation process parameters that do not meet the screening conditions are used as decision-making parameters; Control instruction generation: Based on the decision-making parameters and processing requirement instructions, process control parameters are obtained; the process control parameters include at least one operation flow of the full-process processing solution and the corresponding standard environment data; Process monitoring: The fabric to be tested is processed based on process control parameters, and the test environment data and test dynamic parameters are collected using sensors. The test dynamic parameters are characterized by edge computing to obtain fabric shrinkage data. Feedback and calibration: Identification is performed based on process control parameters, fabric shrinkage data, decision-making parameters and processing requirement instructions, and the identification process is calibrated through the difference between test environment data and standard environment data to obtain calibrated process parameters.

2. The full-process dynamic calibration method for fabric shrinkage based on edge computing according to claim 1 is characterized in that: The sample information includes physical information, component information, appearance information and structural information; The physical information includes fabric shape, fabric weight, fabric thickness, and fabric density; the component information includes fiber type; the appearance information includes color distribution; and the structural information includes weaving method and stitch density. The processing requirement instructions are the condition restrictions and quality requirements for processing the fabric to be tested; The initial characteristic parameters include fabric texture data.

3. The full-process dynamic calibration method for fabric shrinkage based on edge computing according to claim 1 is characterized in that: A process parameter generation model is constructed based on the process path planning method of constraint satisfaction and multi-objective optimization; The process parameter generation model includes a demand instruction decomposition layer and a process parameter generation layer; The demand instruction decomposition layer obtains processing rules based on the processing conditions and restrictions for the fabric to be tested in the processing demand instruction; and obtains processing targets based on the quality requirements for the fabric to be tested in the processing demand instruction; The process parameter generation layer uses the processing target as the optimization target and the processing rules as the constraint conditions, and performs identification based on the decision basis parameters to obtain process control parameters.

4. The full-process dynamic calibration method for fabric shrinkage based on edge computing according to claim 1 is characterized in that: The method for obtaining the fabric shrinkage data is as follows: Identify the process of fabric processing to be tested based on process control parameters to obtain test environment data and test dynamic parameters; the test dynamic parameters include fabric shape dynamic parameters, fabric appearance dynamic parameters, fabric texture dynamic parameters, and fabric temperature dynamic parameters; Based on the time distribution of each operation step in the processing operation process, the test dynamic parameters and test environment data are divided to obtain the operation step dynamic parameters and operation step environment parameters; Extracting features of the dynamic parameters of the operation steps to obtain dynamic feature parameters; the dynamic feature parameters include dynamic features of fabric shape, dynamic features of fabric appearance, dynamic features of fabric texture and dynamic features of fabric temperature; Inputting the dynamic feature parameters and corresponding operation step environment parameters into a shrinkage feature recognition model deployed on an edge computing node to calculate and output structured fabric shrinkage data; The fabric shrinkage data includes the instantaneous shrinkage rate and the cumulative shrinkage rate of the fabric at different processing time points.

5. The full-process dynamic calibration method for fabric shrinkage based on edge computing according to claim 1 is characterized in that: The process of obtaining the calibration process parameters includes: correcting the fabric shrinkage data based on the difference between the standard environment data of the process control parameters and the test environment data in actual processing to obtain the calibration shrinkage data; Based on the calibration shrinkage data, identification is performed to obtain the operation abnormality point; based on the decision basis parameters and the processing requirement instructions, the process control parameters at the operation abnormality point are calibrated to obtain the calibrated process parameters.

6. The full-process fabric shrinkage dynamic calibration system based on edge computing is characterized by: The method for dynamic calibration of fabric shrinkage rate based on edge computing in the entire process as claimed in claim 1 comprises: The setting and perception module collects sample information and processing requirements of the fabric to be tested; The decision basis generation module identifies the fabric to be tested and obtains initial characteristic parameters; the initial characteristic parameters include fabric texture data; based on the sample information and the initial characteristic parameters, identification is performed to obtain decision basis parameters; A control instruction generation module identifies the decision-making basis parameters and the processing requirement instructions to obtain process control parameters; the process control parameters include at least one operation flow of a full-process processing solution and corresponding standard environment data; The processing monitoring module processes the fabric to be tested based on process control parameters and collects test environment data and test dynamic parameters based on sensors. It also performs feature recognition on the test dynamic parameters based on edge computing to obtain fabric shrinkage data. The feedback and calibration module performs identification based on process control parameters, fabric shrinkage data, decision-making parameters and processing requirement instructions. It calibrates the identification process by the difference between the test environment data and the standard environment data to obtain the calibrated process parameters.

Images