An intelligent management system and method applied to clothing flexible customization

Abstract

The application discloses an intelligent management system and method applied to clothing flexible customization, relates to the technical field of clothing intelligent manufacturing, and comprises a multi-source color data acquisition module, a color space preprocessing module, a dynamic color space mapping module, an edge real-time calibration module, a device state monitoring module and a whole-process data closed loop module. The multi-source color data acquisition module constructs a cross-device color benchmark library, the color space preprocessing module standardizes color data and unifies a color gamut, the dynamic color space mapping module generates and optimizes a dynamic conversion matrix, the edge real-time calibration module updates device parameters in combination with fabric characteristics, the device state monitoring module constructs a correlation model to analyze device states and color deviations, and the whole-process data closed loop module stores data through a block chain and optimizes algorithms. The system realizes a complete process from fabric color data acquisition, cross-device mapping and calibration, device state monitoring to whole-process data closed loop management through cooperation of the modules.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent clothing manufacturing, and in particular to an intelligent management system and method for flexible clothing customization. Background Art

[0002] In the field of flexible clothing customization, color management faces significant challenges in cross-device color space differences. In traditional processes, there is a significant difference in the coverage of the RGB color gamut of the cutting machine and the CMYK color gamut of the printing and dyeing machine. The lack of a dynamic cross-device mapping mechanism has led to a high rate of complaints about cross-device color differences in small-batch orders. The color cast problem is prominent when processing high-saturation colors, and multiple trial and error adjustments are required. At the same time, the existing static ICC profiles are unable to adapt to individual differences in equipment, aging, and changes in fabric properties. Changes in equipment status can easily lead to invalidation of calibration parameters, and manual recalibration is time-consuming. In addition, the light reflection characteristics of different fabrics vary significantly, making it difficult for static profiles to dynamically compensate. As a result, color differences generally exceed the visually perceptible threshold, affecting product quality and production efficiency.

[0003] Furthermore, existing technologies rely on independent data from cutting, printing, dyeing, and quality inspection processes, lacking real-time synchronization and reliable evidence storage. Tracing quality issues requires manual verification of multiple data sources, resulting in lengthy locating times. Traditional databases are susceptible to tampering, and responsibility lines are unclear. More critically, existing solutions fail to establish a closed-loop system of "data collection - dynamic calibration - quality traceability - algorithm evolution," lack multimodal data fusion capabilities, and fail to incorporate technologies like edge computing and blockchain to address industrial scenarios. Algorithm optimization relies on manual experience and cannot achieve adaptive evolution using historical data, making it difficult to cope with the frequent equipment adjustments and fabric changes required for flexible customization.

[0004] To solve the above problems, the present invention proposes an intelligent management system and method for flexible customization of clothing. Summary of the Invention

[0005] The purpose of the present invention is to provide an intelligent management system and method for flexible customization of clothing to solve the problems raised in the prior art.

[0006] To achieve the above object, the present invention provides the following technical solutions:

[0007] An intelligent management system for flexible clothing customization includes a multi-source color data acquisition module, a color space preprocessing module, a dynamic color space mapping module, an edge real-time calibration module, an equipment status monitoring module, and a full-process data closed-loop module; the multi-source color data acquisition module uses a multi-spectral sensor and an industrial-grade RGB sensor to synchronously collect the spectral reflectance, RGB value, and ambient light parameters of the standard color card when the fabric is put into storage, and build a cross-device color reference library; the color space preprocessing module is used to standardize the original color data and use CIE The XYZ space unifies the color gamut range of the cutting machine RGB and the printing and dyeing machine CMYK to unify the color space of multiple devices; the dynamic color space mapping module generates the dynamic conversion matrix of the device based on the iterative optimization algorithm, converts the cutting machine RGB data into the CMYK theoretical value of the printing and dyeing machine, and dynamically updates the matrix parameters by comparing the actual output error to achieve cross-device color space mapping; the edge real-time calibration module is used to deploy edge computing nodes on the device side, and update the device parameters in real time based on the physical properties of the fabric; the equipment status monitoring module is used to collect equipment operating parameters in real time, and combine the real-time detection data of the spectrophotometer to build an equipment status-color parameter association model; the full-process data closed-loop module is used to store the full-link color data through the blockchain, and through the adaptive learning engine optimization algorithm, based on the association model data of the equipment status monitoring module and the historical data stored in the blockchain, optimize the dynamic conversion matrix and compensation matrix generation algorithm to form a closed-loop management of quality traceability and continuous evolution.

[0008] The multi-source color data acquisition module includes a sensor array unit and a reference library construction unit;

[0009] The sensor array unit is used to establish a multi-dimensional color data source during the fabric warehousing process, providing an original reference for color matching throughout the entire process. Specifically, a multispectral sensor array is first used to obtain the spectral reflectance curve S(λ) of the standard color card, and then an industrial-grade RGB sensor is used to synchronously collect the three stimulus values ​​R, G, and B of the color card. The three stimulus values ​​directly correspond to the input color gamut space of the cutting machine, making the front-end data compatible with the device interface. Finally, the ambient light perception sensor collects the ambient illumination and correlated color temperature in real time to form an environmental parameter vector L.

[0010] The reference library construction unit is used to establish a cross-device color reference library with a timestamp and device identification; specifically: using the fabric batch unique identification ID as the index, S(λ), R, G, B, L and the acquisition timestamp are encapsulated into a reference data structure D base ; Then, a hash algorithm is used to generate a data fingerprint, the hash value is stored through a hardware security module, and the benchmark library is deployed on a distributed file system, supporting the corresponding benchmark data to be retrieved through ID in the cutting and printing and dyeing links to establish a cross-device color benchmark.

[0011] The color space preprocessing module includes a data standardization unit and a color gamut unification unit;

[0012] The data normalization unit is used to remove environmental noise and device deviation from the original data to form standardized color data that can be processed across platforms. Specifically, illumination normalization correction is performed on the RGB data, and the formula is as follows:

[0013] ;

[0014] in, are the normalized tristimulus values; E ref is the illuminance of the standard lighting object, E is the ambient illuminance collected in real time, and the color deviation under different lighting conditions is eliminated through division operation;

[0015] Then, the spectral data S(λ) is normalized by extreme values. By calculating the historical minimum and maximum response values ​​of the sensor at this wavelength, the real-time collected spectral reflectance S(λ) is mapped to the [0,1] interval to obtain the normalized spectral data. ;

[0016] The color gamut unification unit unifies the color gamut description of the cutting machine RGB and the printing and dyeing machine CMYK through the CIE XYZ standard space, eliminating the incompatibility problem of cross-device color space; specifically: first, the RGB value output by the cutting machine is converted through the device's conversion matrix M RGB Mapped to CIE XYZ space; this matrix is ​​pre-calibrated based on the spectral tristimulus values ​​of the CIE 1931 standard observer and contains 3×3 linear conversion coefficients. The matrix is ​​as follows:

[0017] ;

[0018] Among them, the vector [X, Y, Z] represents the tristimulus value vector of the CIE XYZ color space, representing the coordinates of the color in the standardized color space;

[0019] Then, based on the ink overprinting principle, the lightness component Y is first calculated through the ink overprinting model. The formula is as follows:

[0020] ;

[0021] Among them, C, M, Y k , K are the concentrations of cyan, magenta, yellow, and black inks, respectively, and their value range is [0, 1]. Combined with the chromaticity coordinates (x, y), the X and Z values ​​are inferred to obtain the CIE XYZ coordinates corresponding to CMYK.

[0022] The dynamic color space mapping module includes a matrix generation unit and an error feedback unit;

[0023] The matrix generation unit is used to generate a dynamic conversion matrix exclusive to each device; specifically: first, the RGB vector set {C RGB,i} and printing machine theory CMYK output vector set , where i is the sample index; then the iterative least squares optimization algorithm is used to solve the optimal transformation matrix T ∗ , the formula is as follows:

[0024] ;

[0025] Among them, T is the 4×3 transformation matrix to be optimized, and the matrix elements T are adjusted iteratively. mn , minimize the Euclidean distance deviation between RGB and CMYK vectors; T mn Represents the conversion coefficients of CMYK channels and RGB channels in the dynamic conversion matrix T;

[0026] Finally, the regularization term is introduced Suppress overfitting, where λ is a regularization parameter in a preset range, is the Frobenius norm of the matrix, which enables the model to generalize across different devices;

[0027] The error feedback unit dynamically adjusts the conversion matrix parameters based on the difference between the actual output data and the theoretical value; specifically: first, the actual color data of the printed and dyed fabric collected by the spectrophotometer is converted into the lightness component of the CIELAB space using a nonlinear transformation method Chroma component ; Then the k-th iteration matrix T is generated by the matrix generation unit k , convert the cutting machine RGB data into the printing and dyeing machine theoretical CMYK value, and then convert it into the theoretical component through CIELAB ; Then calculate the color difference between the measured value and the theoretical value based on the above conversion value. The formula is as follows:

[0028] ;

[0029] When ΔE ab When the preset threshold is exceeded, the gradient descent algorithm is triggered to update the matrix to the k+1th version. The formula is as follows:

[0030] ;

[0031] Among them, T k is the dynamic transformation matrix of the kth iteration, η is the learning rate, ∇ is the gradient operator, and the matrix elements are adjusted by backpropagation until the color difference converges.

[0032] The edge real-time calibration module includes an edge computing node unit and a fabric property compensation unit;

[0033] The edge computing node unit is used to complete real-time parsing of the dynamic conversion matrix and update device parameters at the device site. Specifically, a real-time operating system is deployed on the edge server, and the dynamic conversion matrix T is subscribed to through the industrial Internet of Things protocol and parsed into a device executable parameter set. The matrix is ​​then parsed into a parameter set executable by the device controller according to the implementation method described in the matrix generation unit, ensuring that the parameter format is compatible with the device interface.

[0034] Then, a priority scheduling mechanism is adopted to assign the highest execution priority to the matrix parsing and parameter update tasks. Specifically, the FreeRTOS real-time operating system is selected, and a hierarchical priority architecture is adopted. Matrix parameter update, matrix parsing and parameter generation, and communication protocol processing are set as tasks of different levels of priority. The priority inheritance mechanism is used to avoid priority inversion problems. The interrupt service routine is optimized, and the bottom half mechanism is used to handle time-consuming tasks. Dedicated memory resources are allocated to high-level tasks and bound to CPU cores to achieve resource isolation. Finally, parameter writing is achieved by directly accessing device registers through memory-mapped IO, and a double buffer mechanism is used to achieve lock-free parameter updates.

[0035] The fabric property compensation unit generates dynamic compensation parameters based on the physical properties of different fabrics to correct for variations in light reflection characteristics caused by material differences. Specifically, a sensor first collects fabric physical property data in real time and constructs a feature vector F = [d, g, w] containing fabric thickness d, gloss g, and weight w. Thickness and weight can be collected using a weighing sensor or a laser ranging sensor, while gloss is measured using a gloss meter. The sensor type must be compatible with the sensor array of the multi-source color data acquisition module.

[0036] A fabric property-compensation parameter mapping model is then established using a long short-term memory (LSTM) neural network. The input is the feature vector F, and the output is the compensation matrix K, which is used to adjust the equipment output curve. The model training data comes from historical fabric property data and corresponding color deviation data stored in the full-process data closed-loop module.

[0037] Specifically: First, a three-layer architecture is adopted, including an input layer, two hidden layers with 64 neurons each and using the ReLU activation function, and an output layer. The continuously collected feature vectors of fabric thickness, glossiness, and gram weight are organized as time series inputs. The input layer receives a feature vector with a dimension of 3, and the output layer outputs a compensation matrix with a dimension of 4×4 corresponding to the CMYK color space conversion parameters through a fully connected layer; the fabric characteristic data collected in real time are cleaned, standardized, and a time series is constructed. The thickness is collected and filtered by a laser rangefinder, and the gram weight is collected by a weighing sensor and filtered by a Kalman filter; then, the fabric characteristic sequence and the target compensation corresponding to the color deviation conversion are extracted from the full-process data closed-loop module. The matrix is ​​used as training data, and the function containing mean square error and CIEDE2000 color difference is used as the loss function. The Adam optimizer and learning rate scheduling strategy are used for training. During model inference, the predicted compensation matrix is ​​denormalized and reversible constraints are processed, and the output 4×4 compensation matrix is ​​used to adjust the device output curve. The sensor collects data at a preset frequency. After collecting a preset number of data points, the feature sequence is updated and input into the model to generate the compensation matrix. The matrix is ​​written to the device controller through the edge computing node, and the historical window is updated using the sliding window mechanism. When the change rate of the compensation matrix predicted multiple times exceeds the threshold, the model is triggered to fine-tune online, update the normalization parameters, and adjust the loss function weight.

[0038] The equipment status monitoring module includes an operating parameter acquisition unit and a correlation model construction unit;

[0039] The operating parameter acquisition unit is used to obtain the equipment operating status and color output data in real time, providing a multi-dimensional evidence chain for quality abnormality analysis; specifically: the printing and dyeing machine parameters, including but not limited to temperature, pressure, speed, and ink flow, are read in real time through the PLC data interface, and the acquisition frequency is synchronized with the industrial production rhythm; according to the online spectrophotometer installed at the discharge end of the printing and dyeing equipment, the output fabric is scanned at preset intervals to obtain the real-time spectral reflectance S meas (λ), and retrieve the standard spectral reflectance S of the corresponding fabric batch from the cross-device color reference library of the full-process data closed-loop module base (λ), which is used to calculate the spectral difference vector ΔS(λ) to quantify the reflectance deviation at each wavelength. The formula is as follows:

[0040] ;

[0041] Then, the equipment parameters and spectral data are aligned according to the acquisition timestamp to generate a dataset with time series tags, in which the benchmark library data is stored in a distributed file system with the unique identification ID of the fabric batch as the index;

[0042] The correlation model building unit is used to identify the causal relationship between device state fluctuation and color deviation; specifically: first, the CIEDE2000 color difference formula is used to calculate the comprehensive color deviation ΔE ab The dependent variable ΔC;, which is the output of the model, is calculated as follows:

[0043] ;

[0044] in, is the component difference between the measured color and the reference color in the CIELAB space

[0045] Then, a multivariate linear regression model was constructed, with the independent variables including the equipment status parameters and the spectral deviation characteristics ΔS (λ j ); Equipment status parameters include temperature T m , pressure P m , speed V m and ink flow Q i , the model structure is as follows:

[0046] ;

[0047] Among them, β0 is the intercept term, β1-β 8+n is the regression coefficient of each variable, reflecting its contribution to color deviation;

[0048] Finally, the data including equipment status parameters, spectral deviation data and corresponding ΔE are obtained from the full process data closed-loop module. ab The least squares method is used to solve the regression coefficient for a historical data set, and the t-test is used to screen the independent variables that significantly affect color deviation. When the ΔC predicted by the model exceeds the preset threshold, an early warning is triggered, and the priority of device parameter adjustment is determined based on the absolute value of the regression coefficient, where the regression coefficient reflects the contribution of the independent variable to color deviation.

[0049] The full-process data closed-loop module includes a blockchain evidence storage unit and an adaptive learning unit;

[0050] The blockchain evidence storage unit is responsible for establishing an unalterable, full-link data ledger to achieve quality traceability and responsibility definition. Specifically, a blockchain is first constructed, with each block containing a data layer and a cryptographic layer. The data layer stores the conversion matrix, calibration timestamps, fabric characteristics, and full-process operation records of equipment status. The cryptographic layer ensures data integrity through a hash algorithm. Then, smart contracts are used to automatically upload data to the chain, solidifying operation records from fabric storage to finished product inspection in chronological order. It also supports retrieval of full-process color data through fabric batch identification to locate problem nodes.

[0051] The adaptive learning unit is responsible for using historical data to drive continuous optimization of the algorithm to form a closed-loop evolution capability; specifically: establishing a data storage architecture to integrate historical data from the entire process, and extracting effective samples through feature engineering; effective samples refer to data samples that can truly reflect the relationship between equipment status, fabric characteristics and color deviation after data cleaning and standardization preprocessing; based on the adaptive learning engine to optimize algorithm parameters, by learning the relationship between equipment status, fabric characteristics and color deviation in historical data, the algorithm model is updated to improve the accuracy of cross-device color mapping and fabric characteristic compensation, forming a closed-loop management of collection-analysis-evolution.

[0052] An intelligent management method for flexible clothing customization includes the following steps:

[0053] S1. Using multispectral sensors and industrial-grade RGB sensors, we simultaneously collect the spectral reflectance, RGB values, and ambient light parameters of standard color cards when fabrics enter the warehouse. This builds a cross-device color reference library with timestamps and device identification, providing an original reference for color matching throughout the entire process.

[0054] S2. Standardize the original color data, eliminating environmental noise and device deviations through illumination normalization correction and extreme value normalization. Then, unify the color gamut descriptions of the cutting machine RGB and the printing machine CMYK using the CIE XYZ standard space to eliminate incompatibilities in cross-device color spaces.

[0055] S3. Generate a device-specific dynamic conversion matrix based on an iterative least squares optimization algorithm to convert the cutting machine's RGB data into the printing and dyeing machine's CMYK theoretical values. Comparing the actual output errors collected by the spectrophotometer, the gradient descent algorithm is used to dynamically update the matrix parameters to achieve cross-device color space mapping.

[0056] S4. Deploy edge computing nodes on the device side, use the real-time operating system to analyze the dynamic conversion matrix and update the device parameters. Combined with the physical properties of the fabric such as thickness, glossiness, and weight collected in real time by the sensor, the LSTM neural network generates a compensation matrix to correct for changes in light reflection characteristics caused by material differences, thus achieving real-time updates of device parameters.

[0057] S5. Real-time collection of operating parameters such as temperature, pressure, speed, and ink flow of the printing and dyeing machine. Combined with the real-time spectral reflectance obtained by the spectrophotometer scanning output of the fabric, the difference vector is calculated with the reference library spectrum, and a multivariate linear regression model is constructed to identify the causal relationship between equipment status fluctuations and color deviations, providing data support for quality anomaly analysis.

[0058] S6. Use blockchain to store full-link color data, including operation records such as conversion matrix version, calibration timestamp, fabric characteristics, and equipment status, to form an unalterable traceability ledger. At the same time, use historical data to optimize the algorithm model, and use the adaptive learning engine to analyze the relationship between equipment status, fabric characteristics and color deviation, improve the accuracy of cross-device color mapping and fabric characteristic compensation, and form a closed-loop management of quality traceability and algorithm evolution.

[0059] Compared with the prior art, the present invention has the following beneficial effects:

[0060] 1. Significantly Improved Cross-Device Color Consistency: This system builds a cross-device color benchmark library by collecting multi-source color data. Combined with the iterative optimization algorithm of the dynamic color space mapping module, this system achieves dynamic mapping of the color spaces between cutting and printing machines. Compared to traditional static calibration solutions, this system significantly reduces cross-device color differences, meeting the high-precision color matching requirements of custom clothing and effectively reducing color difference complaints for small-volume orders.

[0061] 2. A dynamic calibration mechanism addresses equipment and fabric changes, improving production flexibility: The edge-based real-time calibration module uses edge computing nodes to implement real-time analysis and parameter updates of the dynamic conversion matrix. It also utilizes a neural network model to construct a fabric property-compensation model, effectively correcting for light reflection deviations caused by material differences. Compared to traditional static calibration, this significantly improves equipment calibration efficiency and can adapt to changes in the physical properties of various fabrics, significantly enhancing flexible production's ability to adapt to varying equipment conditions and fabric types.

[0062] 3. A trusted closed-loop data system for the entire process enables rapid tracing and accountability for quality issues: The blockchain evidence storage unit automatically solidifies data across the entire process chain through smart contracts, utilizing encryption algorithms to ensure data immutability. It also supports rapid retrieval of full-process color data via unique identifiers. Combined with a device status-color correlation model, this system accurately locates key links and device parameters that influence color deviation, enabling rapid tracing of quality issues and precise accountability, thereby enhancing the transparency and reliability of the production process. BRIEF DESCRIPTION OF THE DRAWINGS

[0063] Figure 1 This is an organizational structure diagram of an intelligent management system for flexible clothing customization according to the present invention;

[0064] Figure 2 This is a workflow diagram of an intelligent management system for flexible clothing customization according to the present invention. DETAILED DESCRIPTION

[0065] 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.

[0066] Example: Figure 1-Figure 2 As shown, the present invention provides a technical solution.

[0067] An intelligent management system for flexible clothing customization includes a multi-source color data acquisition module, a color space preprocessing module, a dynamic color space mapping module, an edge real-time calibration module, an equipment status monitoring module, and a full-process data closed-loop module; the multi-source color data acquisition module uses a multi-spectral sensor and an industrial-grade RGB sensor to synchronously collect the spectral reflectance, RGB value, and ambient light parameters of the standard color card when the fabric is put into storage, and build a cross-device color reference library; the color space preprocessing module is used to standardize the original color data and use CIE The XYZ space unifies the color gamut range of the cutting machine RGB and the printing and dyeing machine CMYK to unify the color space of multiple devices; the dynamic color space mapping module generates the dynamic conversion matrix of the device based on the iterative optimization algorithm, converts the cutting machine RGB data into the CMYK theoretical value of the printing and dyeing machine, and dynamically updates the matrix parameters by comparing the actual output error to achieve cross-device color space mapping; the edge real-time calibration module is used to deploy edge computing nodes on the device side, and update the device parameters in real time based on the physical properties of the fabric; the equipment status monitoring module is used to collect equipment operating parameters in real time, and combine the real-time detection data of the spectrophotometer to build an equipment status-color parameter association model; the full-process data closed-loop module is used to store the full-link color data through the blockchain, and through the adaptive learning engine optimization algorithm, based on the association model data of the equipment status monitoring module and the historical data stored in the blockchain, optimize the dynamic conversion matrix and compensation matrix generation algorithm to form a closed-loop management of quality traceability and continuous evolution.

[0068] The multi-source color data acquisition module includes a sensor array unit and a reference library construction unit;

[0069] The sensor array unit is used to establish a multi-dimensional color data source during the fabric warehousing process, providing an original reference for color matching throughout the entire process. Specifically, a multispectral sensor array is first used to obtain the spectral reflectance curve S(λ) of the standard color card, and then an industrial-grade RGB sensor is used to synchronously collect the three stimulus values ​​R, G, and B of the color card. The three stimulus values ​​directly correspond to the input color gamut space of the cutting machine, making the front-end data compatible with the device interface. Finally, the ambient light perception sensor collects the ambient illumination and correlated color temperature in real time to form an environmental parameter vector L.

[0070] The reference library construction unit is used to establish a cross-device color reference library with a timestamp and device identification; specifically: using the fabric batch unique identification ID as the index, S(λ), R, G, B, L and the acquisition timestamp are encapsulated into a reference data structure D base ; Then, a hash algorithm is used to generate a data fingerprint, the hash value is stored through a hardware security module, and the benchmark library is deployed on a distributed file system, supporting the corresponding benchmark data to be retrieved through ID in the cutting and printing and dyeing links to establish a cross-device color benchmark.

[0071] The color space preprocessing module includes a data standardization unit and a color gamut unification unit;

[0072] The data normalization unit is used to remove environmental noise and device deviation from the original data to form standardized color data that can be processed across platforms. Specifically, illumination normalization correction is performed on the RGB data, and the formula is as follows:

[0073] ;

[0074] in, are the normalized tristimulus values; E ref is the illuminance of the standard lighting object, E is the ambient illuminance collected in real time, and the color deviation under different lighting conditions is eliminated through division operation;

[0075] Then, the spectral data S(λ) is normalized by extreme values. By calculating the historical minimum and maximum response values ​​of the sensor at this wavelength, the real-time collected spectral reflectance S(λ) is mapped to the [0,1] interval to obtain the normalized spectral data. ;

[0076] The color gamut unification unit unifies the color gamut description of the cutting machine RGB and the printing and dyeing machine CMYK through the CIE XYZ standard space, eliminating the incompatibility problem of cross-device color space; specifically: first, the RGB value output by the cutting machine is converted through the device's conversion matrix M RGB Mapped to CIE XYZ space; this matrix is ​​pre-calibrated based on the spectral tristimulus values ​​of the CIE 1931 standard observer and contains 3×3 linear conversion coefficients. The matrix is ​​as follows:

[0077] ;

[0078] Among them, the vector [X, Y, Z] represents the tristimulus value vector of the CIE XYZ color space, representing the coordinates of the color in the standardized color space;

[0079] Then, based on the ink overprinting principle, the lightness component Y is first calculated through the ink overprinting model. The formula is as follows:

[0080] ;

[0081] Among them, C, M, Y k , K are the concentrations of cyan, magenta, yellow, and black inks, respectively, and their value range is [0, 1]. Combined with the chromaticity coordinates (x, y), the X and Z values ​​are inferred to obtain the CIE XYZ coordinates corresponding to CMYK.

[0082] The dynamic color space mapping module includes a matrix generation unit and an error feedback unit;

[0083] The matrix generation unit is used to generate a dynamic conversion matrix exclusive to each device; specifically: first, the RGB vector set {C RGB,i} and printing machine theory CMYK output vector set , where i is the sample index; then the iterative least squares optimization algorithm is used to solve the optimal transformation matrix T ∗ , the formula is as follows:

[0084] ;

[0085] Among them, T is the 4×3 transformation matrix to be optimized, and the matrix elements T are adjusted iteratively. mn , minimize the Euclidean distance deviation between RGB and CMYK vectors; T mn Represents the conversion coefficients of CMYK channels and RGB channels in the dynamic conversion matrix T;

[0086] Finally, the regularization term is introduced Suppress overfitting, where λ is a regularization parameter in a preset range, is the Frobenius norm of the matrix, which enables the model to generalize across different devices;

[0087] The error feedback unit dynamically adjusts the conversion matrix parameters based on the difference between the actual output data and the theoretical value; specifically: first, the actual color data of the printed and dyed fabric collected by the spectrophotometer is converted into the lightness component of the CIELAB space using a nonlinear transformation method Chroma component ; Then the k-th iteration matrix T is generated by the matrix generation unit k , convert the cutting machine RGB data into the printing and dyeing machine theoretical CMYK value, and then convert it into the theoretical component through CIELAB ; Then calculate the color difference between the measured value and the theoretical value based on the above conversion value. The formula is as follows:

[0088] ;

[0089] When ΔE abWhen the preset threshold is exceeded, the gradient descent algorithm is triggered to update the matrix to the k+1th version. The formula is as follows:

[0090] ;

[0091] Among them, T k is the dynamic transformation matrix of the kth iteration, η is the learning rate, ∇ is the gradient operator, and the matrix elements are adjusted by backpropagation until the color difference converges.

[0092] The edge real-time calibration module includes an edge computing node unit and a fabric property compensation unit;

[0093] The edge computing node unit is used to complete real-time parsing of the dynamic conversion matrix and update device parameters at the device site. Specifically, a real-time operating system is deployed on the edge server, and the dynamic conversion matrix T is subscribed to through the industrial Internet of Things protocol and parsed into a device executable parameter set. The matrix is ​​then parsed into a parameter set executable by the device controller according to the implementation method described in the matrix generation unit, ensuring that the parameter format is compatible with the device interface.

[0094] Then, a priority scheduling mechanism is adopted to assign the highest execution priority to the matrix parsing and parameter update tasks. Specifically, the FreeRTOS real-time operating system is selected, and a hierarchical priority architecture is adopted. Matrix parameter update, matrix parsing and parameter generation, and communication protocol processing are set as tasks of different levels of priority. The priority inheritance mechanism is used to avoid priority inversion problems. The interrupt service routine is optimized, and the bottom half mechanism is used to handle time-consuming tasks. Dedicated memory resources are allocated to high-level tasks and bound to CPU cores to achieve resource isolation. Finally, parameter writing is achieved by directly accessing device registers through memory-mapped IO, and a double buffer mechanism is used to achieve lock-free parameter updates.

[0095] The fabric property compensation unit generates dynamic compensation parameters based on the physical properties of different fabrics to correct for variations in light reflection characteristics caused by material differences. Specifically, a sensor first collects fabric physical property data in real time and constructs a feature vector F = [d, g, w] containing fabric thickness d, gloss g, and weight w. Thickness and weight can be collected using a weighing sensor or a laser ranging sensor, while gloss is measured using a gloss meter. The sensor type must be compatible with the sensor array of the multi-source color data acquisition module.

[0096] A fabric property-compensation parameter mapping model is then established using a long short-term memory (LSTM) neural network. The input is the feature vector F, and the output is the compensation matrix K, which is used to adjust the equipment output curve. The model training data comes from historical fabric property data and corresponding color deviation data stored in the full-process data closed-loop module.

[0097] Specifically: First, a three-layer architecture is adopted, including an input layer, two hidden layers with 64 neurons each and using the ReLU activation function, and an output layer. The continuously collected feature vectors of fabric thickness, glossiness, and gram weight are organized as time series inputs. The input layer receives a feature vector with a dimension of 3, and the output layer outputs a compensation matrix with a dimension of 4×4 corresponding to the CMYK color space conversion parameters through a fully connected layer; the fabric characteristic data collected in real time are cleaned, standardized, and a time series is constructed. The thickness is collected and filtered by a laser rangefinder, and the gram weight is collected by a weighing sensor and filtered by a Kalman filter; then, the fabric characteristic sequence and the target compensation corresponding to the color deviation conversion are extracted from the full-process data closed-loop module. The matrix is ​​used as training data, and the function containing mean square error and CIEDE2000 color difference is used as the loss function. The Adam optimizer and learning rate scheduling strategy are used for training. During model inference, the predicted compensation matrix is ​​denormalized and reversible constraints are processed, and the output 4×4 compensation matrix is ​​used to adjust the device output curve. The sensor collects data at a preset frequency. After collecting a preset number of data points, the feature sequence is updated and input into the model to generate the compensation matrix. The matrix is ​​written to the device controller through the edge computing node, and the historical window is updated using the sliding window mechanism. When the change rate of the compensation matrix predicted multiple times exceeds the threshold, the model is triggered to fine-tune online, update the normalization parameters, and adjust the loss function weight.

[0098] The equipment status monitoring module includes an operating parameter acquisition unit and a correlation model construction unit;

[0099] The operating parameter acquisition unit is used to obtain the equipment operating status and color output data in real time, providing a multi-dimensional evidence chain for quality abnormality analysis; specifically: the printing and dyeing machine parameters, including but not limited to temperature, pressure, speed, and ink flow, are read in real time through the PLC data interface, and the acquisition frequency is synchronized with the industrial production rhythm; according to the online spectrophotometer installed at the discharge end of the printing and dyeing equipment, the output fabric is scanned at preset intervals to obtain the real-time spectral reflectance S meas (λ), and retrieve the standard spectral reflectance S of the corresponding fabric batch from the cross-device color reference library of the full-process data closed-loop module base (λ), which is used to calculate the spectral difference vector ΔS(λ) to quantify the reflectance deviation at each wavelength. The formula is as follows:

[0100] ;

[0101] Then, the equipment parameters and spectral data are aligned according to the acquisition timestamp to generate a dataset with time series tags, in which the benchmark library data is stored in a distributed file system with the unique identification ID of the fabric batch as the index;

[0102] The correlation model building unit is used to identify the causal relationship between device state fluctuation and color deviation; specifically: first, the CIEDE2000 color difference formula is used to calculate the comprehensive color deviation ΔE ab The dependent variable ΔC;, which is the output of the model, is calculated as follows:

[0103] ;

[0104] in, is the component difference between the measured color and the reference color in the CIELAB space

[0105] Then, a multivariate linear regression model was constructed, with the independent variables including the equipment status parameters and the spectral deviation characteristics ΔS (λ j ); Equipment status parameters include temperature T m , pressure P m , speed V m and ink flow Q i , the model structure is as follows:

[0106] ;

[0107] Among them, β0 is the intercept term, β1-β 8+n is the regression coefficient of each variable, reflecting its contribution to color deviation;

[0108] Finally, the data including equipment status parameters, spectral deviation data and corresponding ΔE are obtained from the full process data closed-loop module. ab The least squares method is used to solve the regression coefficient for a historical data set, and the t-test is used to screen the independent variables that significantly affect color deviation. When the ΔC predicted by the model exceeds the preset threshold, an early warning is triggered, and the priority of device parameter adjustment is determined based on the absolute value of the regression coefficient, where the regression coefficient reflects the contribution of the independent variable to color deviation.

[0109] The full-process data closed-loop module includes a blockchain evidence storage unit and an adaptive learning unit;

[0110] The blockchain evidence storage unit is responsible for establishing an unalterable, full-link data ledger to achieve quality traceability and responsibility definition. Specifically, a blockchain is first constructed, with each block containing a data layer and a cryptographic layer. The data layer stores the conversion matrix, calibration timestamps, fabric characteristics, and full-process operation records of equipment status. The cryptographic layer ensures data integrity through a hash algorithm. Then, smart contracts are used to automatically upload data to the chain, solidifying operation records from fabric storage to finished product inspection in chronological order. It also supports retrieval of full-process color data through fabric batch identification to locate problem nodes.

[0111] The adaptive learning unit is responsible for using historical data to drive continuous optimization of the algorithm to form a closed-loop evolution capability; specifically: establishing a data storage architecture to integrate historical data from the entire process, and extracting effective samples through feature engineering; effective samples refer to data samples that can truly reflect the relationship between equipment status, fabric characteristics and color deviation after data cleaning and standardization preprocessing; based on the adaptive learning engine to optimize algorithm parameters, by learning the relationship between equipment status, fabric characteristics and color deviation in historical data, the algorithm model is updated to improve the accuracy of cross-device color mapping and fabric characteristic compensation, forming a closed-loop management of collection-analysis-evolution.

[0112] An intelligent management method for flexible clothing customization includes the following steps:

[0113] S1. Using multispectral sensors and industrial-grade RGB sensors, we simultaneously collect the spectral reflectance, RGB values, and ambient light parameters of standard color cards when fabrics enter the warehouse. This builds a cross-device color reference library with timestamps and device identification, providing an original reference for color matching throughout the entire process.

[0114] S2. Standardize the original color data, eliminating environmental noise and device deviations through illumination normalization correction and extreme value normalization. Then, unify the color gamut descriptions of the cutting machine RGB and the printing machine CMYK using the CIE XYZ standard space to eliminate incompatibilities in cross-device color spaces.

[0115] S3. Generate a device-specific dynamic conversion matrix based on an iterative least squares optimization algorithm to convert the cutting machine's RGB data into the printing and dyeing machine's CMYK theoretical values. Comparing the actual output errors collected by the spectrophotometer, the gradient descent algorithm is used to dynamically update the matrix parameters to achieve cross-device color space mapping.

[0116] S4. Deploy edge computing nodes on the device side, use the real-time operating system to analyze the dynamic conversion matrix and update the device parameters. Combined with the physical properties of the fabric such as thickness, glossiness, and weight collected in real time by the sensor, the LSTM neural network generates a compensation matrix to correct for changes in light reflection characteristics caused by material differences, thus achieving real-time updates of device parameters.

[0117] S5. Real-time collection of operating parameters such as temperature, pressure, speed, and ink flow of the printing and dyeing machine. Combined with the real-time spectral reflectance obtained by the spectrophotometer scanning output of the fabric, the difference vector is calculated with the reference library spectrum, and a multivariate linear regression model is constructed to identify the causal relationship between equipment status fluctuations and color deviations, providing data support for quality anomaly analysis.

[0118] S6. Use blockchain to store full-link color data, including operation records such as conversion matrix version, calibration timestamp, fabric characteristics, and equipment status, to form an unalterable traceability ledger. At the same time, use historical data to optimize the algorithm model, and use the adaptive learning engine to analyze the relationship between equipment status, fabric characteristics and color deviation, improve the accuracy of cross-device color mapping and fabric characteristic compensation, and form a closed-loop management of quality traceability and algorithm evolution.

[0119] In flexible, customized clothing production, the system first constructs a cross-device color benchmark library using a multi-source color data acquisition module. For example, a silk fabric with a thickness of 0.15mm, a glossiness of 92GU, and a weight of 80g / ㎡ is used. A multispectral sensor array covering the 400-700nm wavelength range with an accuracy of ±0.5% is used to collect the spectral reflectance curve of a standard color chart, for example, 68% reflectance at 550nm. Simultaneously, an industrial-grade RGB sensor model IMX571, with a color gamut covering 95% of sRGB, acquires tristimulus values ​​of R=185, G=210, and B=190. An ambient light sensor also collects illuminance of 5000lux and color temperature of 6500K. This data is processed by the benchmark library construction unit, indexed by the fabric batch ID "F20230516-001," to generate a timestamp-containing benchmark data structure. This data is tamper-proofed using the SHA-256 hash algorithm and stored in a distributed file system. The color space preprocessing module performs illumination normalization and extreme value normalization on the raw data, converting the RGB values ​​into CIE XYZ space coordinates X=45.2, Y=52.8, Z=39.5, and calculates the corresponding CMYK coordinates through the ink overprint model, achieving the unification of the RGB color gamut of the cutting machine and the CMYK color gamut of the printing and dyeing machine, and the overlap is increased from 62% to 89%.

[0120] The dynamic color space mapping module generates a device-specific conversion matrix based on the iterative least squares algorithm, converting the RGB data output by the cutting machine, such as R=200, G=150, B=120, into the theoretical CMYK values ​​of the printing and dyeing machine, with the initial values ​​being C=30%, M=45%, Y=50%, K=0%, and comparing them with the measured data ΔE of the spectrophotometer. ab =2.7, after 5 iterations of the gradient descent algorithm, the error between the theoretical value and the measured value converges to ΔE ab =0.9. The edge real-time calibration module deploys the FreeRTOS system through the edge computing node with a response time of ≤100ms. It analyzes the dynamic matrix and writes it into the printing and dyeing machine controller. At the same time, it uses the three-layer LSTM neural network architecture with 64 neurons in each hidden layer to analyze the real-time collected fabric characteristic vector d=0.15mm, g=92GU, w=80g / ㎡, generate a 4×4 compensation matrix, and adjust the ink flow parameters, such as increasing the cyan flow from 22mL / min to 25mL / min, to achieve the actual output color difference ΔE*ab The equipment status monitoring module simultaneously collects parameters such as the printing and dyeing machine temperature (60°C ± 1°C) and pressure (0.3MPa ± 0.02MPa), constructs a multivariate linear regression model, identifies that temperature contributes 42% to color deviation, and automatically adjusts the temperature control system when the warning threshold is triggered.

[0121] The full-process data closed-loop module stores data from the entire chain on the blockchain. Each block contains the conversion matrix version (e.g., V3.2), the calibration timestamp (2025-05-16 14:30:05), fabric characteristics, and device status. This data is automatically uploaded to the blockchain via smart contracts, improving data storage efficiency by 70%. The adaptive learning engine extracts over 100,000 historical data items, including 2,000 fabric characteristics and 5,000 matrix update records. Through feature engineering, it identifies glossiness (weighted at 35%) and temperature (weighted at 28%) as key influencing factors. This optimizes the LSTM model parameters, improving the accuracy of compensation matrix generation by 30%. When the rate of change in the compensation matrix exceeds 5% over 10 consecutive predictions, online fine-tuning is triggered, updating the normalization parameters and adjusting the loss function weights, forming a closed loop of collection, calibration, learning, and evolution. In practical applications, the system has increased the cross-device color consistency pass rate from 75% to 98%, reduced device calibration time from 120 minutes per session to 15 minutes per session, and reduced fabric waste by 22%, enabling precise color control for small custom orders, such as 50 pieces per order.

[0122] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above and that the invention can be embodied in other specific forms without departing from the spirit or essential characteristics of the invention. Therefore, the embodiments should be considered in all respects as illustrative and non-restrictive, and the scope of the invention is defined by the appended claims, not the foregoing description, and all variations within the meaning and range of equivalents of the claims are intended to be included therein. Any reference sign in a claim should not be construed as limiting the claim to which it relates.

Claims

1. An intelligent management system for flexible clothing customization, characterized by: It includes a multi-source color data acquisition module, a color space preprocessing module, a dynamic color space mapping module, a real-time edge calibration module, an equipment status monitoring module and a full-process data closed-loop module; the multi-source color data acquisition module uses a multi-spectral sensor and an industrial-grade RGB sensor to synchronously collect the spectral reflectance, RGB value and ambient light parameters of the standard color card when the fabric is put into storage, and build a cross-device color reference library; the color space preprocessing module is used to standardize the original color data, unify the color gamut range of the cutting machine RGB and the printing and dyeing machine CMYK color space through the CIE XYZ space; the dynamic color space mapping module generates a dynamic conversion matrix for the device based on an iterative optimization algorithm, converts the cutting machine RGB data into the printing and dyeing machine CMYK theoretical value, and dynamically updates the matrix parameters by comparing the actual output error to achieve cross-device color space mapping; The edge real-time calibration module is used to deploy edge computing nodes on the device side and update device parameters in real time based on the physical properties of the fabric. The device status monitoring module is used to collect device operating parameters in real time and, combined with the real-time detection data of the spectrophotometer, build a device status-color parameter correlation model. The full-process data closed-loop module is used to store full-link color data through blockchain and, through an adaptive learning engine optimization algorithm, optimize the dynamic conversion matrix and compensation matrix generation algorithm based on the correlation model data of the device status monitoring module and the historical data stored on the blockchain, forming a closed-loop management of quality traceability and continuous evolution. The edge real-time calibration module includes an edge computing node unit and a fabric property compensation unit; The edge computing node unit is used to complete real-time parsing of the dynamic conversion matrix and update device parameters at the device site. Specifically, a real-time operating system is deployed on the edge server, and the dynamic conversion matrix T is subscribed to through the industrial Internet of Things protocol and parsed into a device executable parameter set. The matrix is ​​then parsed into a parameter set executable by the device controller according to the implementation method described in the matrix generation unit, ensuring that the parameter format is compatible with the device interface. Then, a priority scheduling mechanism is adopted to assign the highest execution priority to the matrix parsing and parameter update tasks. Specifically, the FreeRTOS real-time operating system is selected, and a hierarchical priority architecture is adopted. Matrix parameter update, matrix parsing and parameter generation, and communication protocol processing are set as tasks of different levels of priority. The priority inheritance mechanism is used to avoid priority inversion problems. The interrupt service routine is optimized, and the bottom half mechanism is used to handle time-consuming tasks. Dedicated memory resources are allocated to high-level tasks and bound to CPU cores to achieve resource isolation. Finally, parameter writing is achieved by directly accessing device registers through memory-mapped IO, and a double buffer mechanism is used to achieve lock-free parameter updates. The fabric property compensation unit generates dynamic compensation parameters based on the physical properties of different fabrics to correct for variations in light reflection characteristics caused by material differences. Specifically, a sensor first collects fabric physical property data in real time and constructs a feature vector F = [d, g, w] containing fabric thickness d, gloss g, and weight w. Thickness and weight can be collected using a weighing sensor or a laser ranging sensor, while gloss is measured using a gloss meter. The sensor type must be compatible with the sensor array of the multi-source color data acquisition module. Then, a long short-term memory (LSTM) neural network is used to establish a fabric property-compensation parameter mapping model. The input is the feature vector F, and the output is the compensation matrix K, which is used to adjust the equipment output curve. The model training data comes from the historical fabric property data and corresponding color deviation data stored in the full-process data closed-loop module.

2. The intelligent management system for flexible clothing customization according to claim 1, characterized in that: The multi-source color data acquisition module includes a sensor array unit and a reference library construction unit; The sensor array unit is used to establish a multi-dimensional color data source during the fabric warehousing process, providing an original reference for color matching throughout the entire process. Specifically, a multispectral sensor array is first used to obtain the spectral reflectance curve S(λ) of the standard color card, and then an industrial-grade RGB sensor is used to synchronously collect the three stimulus values ​​R, G, and B of the color card. The three stimulus values ​​directly correspond to the input color gamut space of the cutting machine, making the front-end data compatible with the device interface. Finally, the ambient light perception sensor collects the ambient illumination and correlated color temperature in real time to form an environmental parameter vector L. The reference library construction unit is used to establish a cross-device color reference library with a timestamp and device identification; specifically: using the fabric batch unique identification ID as the index, S(λ), R, G, B, L and the acquisition timestamp are encapsulated into a reference data structure D base ; Then, a hash algorithm is used to generate a data fingerprint, the hash value is stored through a hardware security module, and the benchmark library is deployed on a distributed file system, supporting the corresponding benchmark data to be retrieved through ID in the cutting and printing and dyeing links to establish a cross-device color benchmark.

3. The intelligent management system for flexible clothing customization according to claim 1 is characterized in that: The color space preprocessing module includes a data standardization unit and a color gamut unification unit; The data normalization unit is used to remove environmental noise and device deviation from the original data to form standardized color data that can be processed across platforms. Specifically, illumination normalization correction is performed on the RGB data, and the formula is as follows: ； in, are the normalized tristimulus values; E ref is the illuminance of the standard lighting object, E is the ambient illuminance collected in real time, and the color deviation under different lighting conditions is eliminated through division operation; Then, the spectral data S(λ) is normalized by extreme values. By calculating the historical minimum and maximum response values ​​of the sensor at this wavelength, the real-time collected spectral reflectance S(λ) is mapped to the [0,1] interval to obtain the normalized spectral data. ; The color gamut unification unit unifies the color gamut description of the cutting machine RGB and the printing and dyeing machine CMYK through the CIE XYZ standard space, eliminating the incompatibility problem of cross-device color space; specifically: first, the RGB value output by the cutting machine is converted through the device's conversion matrix M RGB Mapped to CIE XYZ space; this matrix is ​​pre-calibrated based on the spectral tristimulus values ​​of the CIE 1931 standard observer and contains 3×3 linear conversion coefficients. The matrix is ​​as follows: ； Among them, the vector [X, Y, Z] represents the tristimulus value vector of the CIE XYZ color space, representing the coordinates of the color in the standardized color space; Then, based on the ink overprinting principle, the lightness component Y is first calculated through the ink overprinting model. The formula is as follows: ； Among them, C, M, Y k , K are the concentrations of cyan, magenta, yellow, and black inks, respectively, and their value range is [0, 1]. Combined with the chromaticity coordinates (x, y), the X and Z values ​​are inferred to obtain the CIE XYZ coordinates corresponding to CMYK.

4. The intelligent management system for flexible clothing customization according to claim 1, characterized in that: The dynamic color space mapping module includes a matrix generation unit and an error feedback unit; The matrix generation unit is used to generate a dynamic conversion matrix exclusive to each device; specifically: first, the RGB vector set {C RGB,i } and printing machine theory CMYK output vector set , where i is the sample index; then the iterative least squares optimization algorithm is used to solve the optimal transformation matrix T ∗ , the formula is as follows: ； Among them, T is the 4×3 transformation matrix to be optimized, and the matrix elements T are adjusted iteratively. mn , minimize the Euclidean distance deviation between RGB and CMYK vectors; T mn Represents the conversion coefficients of CMYK channels and RGB channels in the dynamic conversion matrix T; Finally, the regularization term is introduced Suppress overfitting, where λ is a regularization parameter in a preset range, is the Frobenius norm of the matrix, which enables the model to generalize across different devices; The error feedback unit dynamically adjusts the conversion matrix parameters based on the difference between the actual output data and the theoretical value; specifically: first, the actual color data of the printed and dyed fabric collected by the spectrophotometer is converted into the lightness component of the CIELAB space using a nonlinear transformation method Chroma component ; Then the k-th iteration matrix T is generated by the matrix generation unit k , convert the cutting machine RGB data into the printing and dyeing machine theoretical CMYK value, and then convert it into the theoretical component through CIELAB ; Then calculate the color difference between the measured value and the theoretical value based on the above conversion value. The formula is as follows: ； When ΔE ab When the preset threshold is exceeded, the gradient descent algorithm is triggered to update the matrix to the k+1th version. The formula is as follows: ； Among them, T k is the dynamic transformation matrix of the kth iteration, η is the learning rate, ∇ is the gradient operator, and the matrix elements are adjusted by backpropagation until the color difference converges.

5. The intelligent management system for flexible clothing customization according to claim 4 is characterized in that: Specifically, a three-layer architecture consisting of an input layer, two hidden layers with 64 neurons each and using the ReLU activation function, and an output layer is adopted. The continuously collected feature vectors of fabric thickness, glossiness, and weight are organized into time series inputs. The input layer receives feature vectors of dimension 3, and the output layer outputs a compensation matrix of dimension 4×4 corresponding to the CMYK color space conversion parameters through a fully connected layer. The fabric characteristic data collected in real time is cleaned, standardized, and constructed into a time series. The thickness is collected and filtered by a laser rangefinder, and the weight is collected by a weighing sensor and filtered by a Kalman filter. The fabric feature sequence and the target compensation matrix corresponding to color deviation conversion are then extracted from the full-process data closed-loop module as training data. A function containing mean square error and CIEDE2000 color difference is used as the loss function, and the training is performed using the Adam optimizer and a learning rate scheduling strategy. During model inference, the predicted compensation matrix is ​​denormalized and reversibility constrained. The output 4×4 compensation matrix is ​​then used to adjust the device output curve. The sensor collects data at a preset frequency. Every time a preset number of data points are collected, the feature sequence is updated and input into the model to generate a compensation matrix. The data is written to the device controller through the edge computing node, and the historical window is updated using a sliding window mechanism. When the change rate of the compensation matrix predicted for multiple consecutive times exceeds the threshold, the model is triggered to fine-tune online, update the normalization parameters, and adjust the loss function weight.

6. The intelligent management system for flexible clothing customization according to claim 1, characterized in that: The equipment status monitoring module includes an operating parameter acquisition unit and a correlation model construction unit; The operating parameter acquisition unit is used to obtain the equipment operating status and color output data in real time, providing a multi-dimensional evidence chain for quality abnormality analysis; specifically: the printing and dyeing machine parameters, including but not limited to temperature, pressure, speed, and ink flow, are read in real time through the PLC data interface, and the acquisition frequency is synchronized with the industrial production rhythm; according to the online spectrophotometer installed at the discharge end of the printing and dyeing equipment, the output fabric is scanned at preset intervals to obtain the real-time spectral reflectance S meas (λ), and retrieve the standard spectral reflectance S of the corresponding fabric batch from the cross-device color reference library of the full-process data closed-loop module base (λ), which is used to calculate the spectral difference vector ΔS(λ) to quantify the reflectance deviation at each wavelength. The formula is as follows: ； Then, the equipment parameters and spectral data are aligned according to the acquisition timestamp to generate a dataset with time series tags, in which the benchmark library data is stored in a distributed file system with the unique identification ID of the fabric batch as the index; The correlation model building unit is used to identify the causal relationship between device state fluctuation and color deviation; specifically: first, the CIEDE2000 color difference formula is used to calculate the comprehensive color deviation ΔE ab The dependent variable ΔC;, which is the output of the model, is calculated as follows: ； in, is the component difference between the measured color and the reference color in the CIELAB space Then, a multivariate linear regression model was constructed, with the independent variables including the equipment status parameters and the spectral deviation characteristics ΔS (λ j ); Equipment status parameters include temperature T m , pressure P m , speed V m and ink flow Q i , the model structure is as follows: ； Among them, β0 is the intercept term, β1-β 8+n is the regression coefficient of each variable, reflecting its contribution to color deviation; Finally, the data including equipment status parameters, spectral deviation data and corresponding ΔE are obtained from the full process data closed-loop module. ab The least squares method is used to solve the regression coefficient for a historical data set, and the t-test is used to screen the independent variables that significantly affect color deviation. When the ΔC predicted by the model exceeds the preset threshold, an early warning is triggered, and the priority of device parameter adjustment is determined based on the absolute value of the regression coefficient, where the regression coefficient reflects the contribution of the independent variable to color deviation.

7. The intelligent management system for flexible clothing customization according to claim 1, characterized in that: The full-process data closed-loop module includes a blockchain evidence storage unit and an adaptive learning unit; The blockchain evidence storage unit is responsible for establishing an unalterable, full-link data ledger to achieve quality traceability and responsibility definition. Specifically, a blockchain is first constructed, with each block containing a data layer and a cryptographic layer. The data layer stores the conversion matrix, calibration timestamps, fabric characteristics, and full-process operation records of equipment status. The cryptographic layer ensures data integrity through a hash algorithm. Then, data is automatically uploaded to the blockchain through smart contracts, solidifying the operation records from fabric storage to finished product inspection in chronological order. It also supports retrieval of full-process color data through fabric batch identification to locate problem nodes. The adaptive learning unit is responsible for using historical data to drive continuous optimization of the algorithm to form a closed-loop evolutionary capability. Specifically, it establishes a data storage architecture to integrate historical data from the entire process and extracts valid samples through feature engineering. Valid samples are data samples that, after data cleaning and standardization preprocessing, can truly reflect the relationship between equipment status, fabric characteristics, and color deviation. Based on the adaptive learning engine to optimize algorithm parameters, by learning the correlation between device status, fabric characteristics and color deviation in historical data, the algorithm model is updated to improve the accuracy of cross-device color mapping and fabric characteristic compensation, forming a closed-loop management of collection-analysis-evolution.

8. An intelligent management method for flexible clothing customization, applied to the intelligent management system for flexible clothing customization according to claims 1-7, characterized in that: The following steps are involved: S1. Using multispectral sensors and industrial-grade RGB sensors, we simultaneously collect the spectral reflectance, RGB values, and ambient light parameters of standard color cards when fabrics enter the warehouse. This builds a cross-device color reference library with timestamps and device identification, providing an original reference for color matching throughout the entire process. S2. Standardize the original color data, eliminating environmental noise and device deviations through illumination normalization correction and extreme value normalization. Then, unify the color gamut descriptions of the cutting machine RGB and the printing machine CMYK using the CIE XYZ standard space to eliminate incompatibilities in cross-device color spaces. S3. Generate a device-specific dynamic conversion matrix based on an iterative least squares optimization algorithm to convert the cutting machine's RGB data into the printing and dyeing machine's CMYK theoretical values. Comparing the actual output errors collected by the spectrophotometer, the gradient descent algorithm is used to dynamically update the matrix parameters to achieve cross-device color space mapping. S4. Deploy edge computing nodes on the device side, use the real-time operating system to analyze the dynamic conversion matrix and update the device parameters. Combined with the physical properties of the fabric such as thickness, glossiness, and weight collected in real time by the sensor, the LSTM neural network generates a compensation matrix to correct for changes in light reflection characteristics caused by material differences, thus achieving real-time updates of device parameters. S5. Real-time collection of operating parameters such as temperature, pressure, speed, and ink flow of the printing and dyeing machine. Combined with the real-time spectral reflectance obtained by the spectrophotometer scanning output of the fabric, the difference vector is calculated with the reference library spectrum, and a multivariate linear regression model is constructed to identify the causal relationship between equipment status fluctuations and color deviations, providing data support for quality anomaly analysis. S6. Use blockchain to store full-link color data, including operation records such as conversion matrix version, calibration timestamp, fabric characteristics, and equipment status, to form an unalterable traceability ledger. At the same time, use historical data to optimize the algorithm model, and use the adaptive learning engine to analyze the relationship between equipment status, fabric characteristics and color deviation, improve the accuracy of cross-device color mapping and fabric characteristic compensation, and form a closed-loop management of quality traceability and algorithm evolution.

Images