Textile printing intelligent manufacturing production management system based on big data

Abstract

The invention discloses a textile printing intelligent manufacturing production management system based on big data, which relates to the technical field of textile printing intelligent manufacturing and comprises a data acquisition module, a knowledge mining module, a dynamic simulation module, a quality identification module and a feedback correction module. The system forms an empirical rule base containing a specific process mapping relation by collecting whole-process operation state data and mining historical order data, carries out production process dynamic simulation by utilizing the rule base and combining real-time equipment data, and outputs an equipment parameter optimization scheme. And the system identifies quality defects of printed finished products on line and generates a defect position distribution thermodynamic diagram, so that an equipment parameter scheme is fed back and corrected, and a final execution instruction is formed. According to the invention, autonomous extraction of process knowledge and production closed-loop optimization based on quality space information are realized, and the intelligent level and quality control capability of printing production are improved.

Description

Technical Field

[0001] This invention belongs to the field of intelligent manufacturing technology for textile printing, specifically a production management system for intelligent manufacturing of textile printing based on big data. Background Technology

[0002] Currently, the management and control of textile printing production largely rely on fixed process formulas and the experience of operators. Production parameters are often set manually based on limited dimensions such as fabric type and pattern, or matched from static databases, lacking in-depth mining and utilization of the complex relationships hidden within massive amounts of historical production data. This results in insufficient adaptability of process parameters, making it difficult to accurately predict material consumption based on the personalized characteristics of orders, and also hindering the rapid provision of optimized parameters that have been validated in practice when faced with new materials or new pattern combinations, leading to bottlenecks in production efficiency and resource utilization.

[0003] Existing intelligent manufacturing systems typically integrate quality inspection functions, but these often focus on identifying defect types, counting defects, and triggering alarms or sorting. This model treats quality inspection and production control as relatively independent processes, with inspection results serving as the final output, failing to form an effective closed loop with the real-time control of production equipment. Even when defects are detected, it is difficult to pinpoint the specific equipment unit or process parameter anomaly on the production line that caused the defect in a timely and accurate manner, let alone guide the directional adjustment of parameters based on the spatial distribution pattern of defects on the fabric. Optimization of the production process relies on post-production manual analysis and adjustments, resulting in a delayed response and making it difficult to achieve real-time self-correction and quality prevention during the production process. Summary of the Invention

[0004] This invention aims to solve at least one of the technical problems existing in the prior art;

[0005] Therefore, this invention proposes a smart manufacturing production management system for textile printing based on big data, comprising: The data acquisition module is used to collect a set of operational status data for the entire textile printing production process. The set of operational status data includes historical order parameter sequences, real-time equipment operating condition data, and online quality inspection image sequences. The knowledge mining module is used to perform production knowledge mining processing on the historical order parameter sequence to generate a printing process experience rule base. The printing process experience rule base includes the mapping relationship between pattern complexity and ink consumption, the adaptation table between fabric material and drying temperature, and the correlation model between color matching accuracy and equipment adjustment parameters. The dynamic simulation module is used to perform dynamic simulation of the production process based on the real-time equipment operating condition data and the printing process experience rule base, and generate the optimal equipment parameter control scheme for the current production batch. The quality identification module is used to identify quality defect patterns in printed finished products using the online quality inspection image sequence, and generate a statistical report on defect types and a heat map of defect location distribution. The feedback correction module is used to combine the defect type statistical report and the defect location distribution heat map to perform feedback correction on the optimal equipment parameter control scheme and generate the final equipment parameter execution instruction after closed-loop optimization.

[0006] Furthermore, the step of performing production knowledge mining processing on the historical order parameter sequence to generate a printing process experience rule base includes: Extract the order feature vector from the historical order parameter sequence. The order feature vector includes the number of pattern colors, pattern area coverage, fabric weight, order batch, and color fastness grade specified by the customer. The order feature vector is input into the association rule analysis model. Through confidence and support calculations, frequent association patterns between the features of each dimension in the order feature vector and the process parameters in subsequent production records are discovered. For each frequently associated pattern discovered, the corresponding production execution records are retrieved by backtracking. The production execution records include the actual ink formula used, the mesh count of the printing screen, the oven temperature profile, and the equipment operating speed. Perform parameter clustering analysis on the successful records that meet the preset quality standards in the production execution records to summarize the stable process parameter range under the specific order feature vector combination; The frequent correlation patterns and their corresponding stable process parameter ranges are structured and stored to form a mapping relationship between pattern complexity and ink consumption, an adaptation table between fabric material and drying temperature, and a correlation model between color matching accuracy and equipment adjustment parameters.

[0007] Furthermore, the step of performing dynamic simulation of the production process based on the real-time equipment operating condition data and the printing process experience rule base to generate an optimal equipment parameter control scheme for the current production batch includes: Analyze the order parameters of the current production batch to form the feature vector of the current order; Using the current order feature vector as an index, a set of initial recommended process parameters are obtained by matching in the printing process experience rule base; Construct a digital twin model of a virtual production line that includes printing machines, drying rooms, and winding devices; The initial recommended process parameters are used as input to drive the virtual production line digital twin model to perform simulation operation; During the simulation operation, the real-time equipment condition stream data is synchronously accessed. The real-time equipment condition stream data includes the real-time value of the printing squeegee pressure, the actual temperature of each zone of the oven, and the readings of the fabric tension sensor. The deviation data is calculated by comparing the real-time equipment operating condition data with the simulated predicted values ​​of the corresponding nodes in the virtual production line digital twin model. Based on the deviation data, the controllable parameters in the virtual production line digital twin model are dynamically adjusted through a preset model prediction control algorithm, so that the model output tends to be stable and meets the preset quality indicators. Once the model reaches a stable state, the operating parameters of each device in the virtual production line digital twin model are extracted and packaged into the optimal device parameter control scheme.

[0008] Further, the step of comparing the real-time equipment operating condition data with the simulated predicted values ​​of the corresponding nodes in the virtual production line digital twin model to calculate the deviation data includes: The real-time value of the printing squeegee pressure is subjected to time series smoothing filtering to obtain the filtered value of the printing squeegee pressure. The instantaneous difference between the filtered value of the printing squeegee pressure and the squeegee pressure setting value of the printing machine node in the virtual production line digital twin model is calculated as the squeegee pressure deviation; The average temperature of each zone of the oven is calculated by averaging the actual temperatures of each zone. The difference between the average temperature of each zone of the oven and the temperature setpoint of the corresponding temperature zone in the virtual production line digital twin model is calculated as the temperature control deviation. The fluctuation range of the fabric tension sensor readings is monitored in real time, and its standard deviation is calculated; The standard deviation of the fabric tension sensor readings is compared with the tension fluctuation threshold of the winding device node in the virtual production line digital twin model. When the standard deviation exceeds the threshold, a tension stability deviation is generated. The deviation data is formed by summing the scraper pressure deviation, the temperature control deviation, and the tension stability deviation.

[0009] Furthermore, the step of using the online quality inspection image sequence to perform quality defect pattern recognition on the printed finished product and generating a defect type statistical report and a defect location distribution heat map includes: The online quality inspection image sequence is formed by continuously acquiring high-definition images of the surface of the printed finished product using industrial cameras deployed at the end of the production line. Each of the high-definition images is preprocessed, and the preprocessing steps include color space conversion, illumination unevenness correction, and pattern region localization and segmentation. The pre-processed image region is input into a deep learning defect detection network, which is pre-trained using image samples labeled with defect types such as stains, misregistration, white gaps, and color bleeding. The deep learning defect detection network outputs recognition results including defect bounding box coordinates, defect type confidence, and defect category label; Archive the continuous identification results according to the production time sequence and fabric roll number; Statistical report of the defect types is generated by statistically analyzing the frequency and average confidence level of various defects within a production batch or a preset time window. The bounding box coordinates of all defects on the same roll of fabric are mapped to the global length-width coordinate system of the fabric roll, and the defects are colored and rendered according to the defect density to generate a heat map of the defect location distribution.

[0010] Furthermore, the step of combining the defect type statistical report and the defect location distribution heatmap to perform feedback correction on the optimal equipment parameter control scheme and generate the final equipment parameter execution instruction after closed-loop optimization includes: Analyze the defect type statistics report to identify the dominant defect type in the current production batch; The system queries a pre-defined defect root cause analysis knowledge graph, which defines causal relationship chains between different defect types and potential equipment parameter mismatches or improper process conditions. Based on the identified dominant defect type, one or more sets of equipment parameters or process conditions suspected of causing the dominant defect type are derived from the defect root cause analysis knowledge graph. The heat map of the defect location distribution is overlaid with the equipment layout diagram of the production line to locate the physical production links or equipment units corresponding to the high-incidence areas of defects. By combining the equipment parameters or process conditions suspected of causing the defects with the physical production links corresponding to the high-incidence areas of the defects, specific equipment parameter items that need to be adjusted are selected. For the selected specific equipment parameter items, calculate their current set value in the optimal equipment parameter control scheme, and make fine adjustments according to the preset correction rule library, which includes parameter adjustment step size and direction for different defect types and severity. The finely tuned equipment parameters are integrated to form the final equipment parameter execution instructions after closed-loop optimization.

[0011] Furthermore, the query of the preset defect root cause analysis knowledge graph includes: The defect root cause analysis knowledge graph is stored in the form of nodes and edges. Nodes represent production entities or states, including equipment components, process parameters, material properties and defect types, while edges represent causal or influence relationships between entities or states. When a specific defect type node is input, backtracking is performed along the incoming edges pointing to the defect type node; During the backtracking traversal, all upstream nodes that directly or indirectly point to the defect type node are recorded, including equipment parameter nodes and process condition nodes. The upstream nodes are sorted according to the weight of the edges or the causal strength statistically derived from historical data. The upstream nodes with higher weights or stronger causal strength are preferentially identified as the root causes of the association. Output a sorted list of upstream nodes as one or more sets of equipment parameters or process conditions suspected of causing the dominant defect type.

[0012] Furthermore, the method also includes the step of periodically updating the printing process experience rule base: Collect historical order parameter sequences of all production batches within a certain period, the optimal equipment parameter control scheme or final equipment parameter execution instructions adopted, and the corresponding online quality inspection image sequence analysis results; The experience data of successful production batches are extracted. A successful production batch refers to a batch in which the final product defect rate is lower than a predetermined threshold and the production efficiency reaches a predetermined standard. The extracted new experience data, including the combination of new order feature vectors and stable process parameter ranges, will be integrated with the existing rules in the printing process experience rule base. For rules in the printing process experience rule base that have not been called for a long time or that deviate too much from the latest production data statistical trend, their weight will be reduced or they will be archived. Use the integrated and optimized rule data to replace or incrementally update the original printing process experience rule library.

[0013] Furthermore, the refinement of the experience data from successful production batches includes: For a single successful production batch, extract its complete order feature vector and the actual equipment parameter scheme executed. Analysis of the online quality inspection image sequence analysis results confirmed that the products of the successfully produced batch performed excellently in all quality indicators; The order feature vector and the actual executed equipment parameter scheme are considered as a successful "case pair"; Cluster cases with similar order feature vectors from multiple successful "case pairs" sets; For each cluster, calculate the statistical distribution of each parameter value in its equipment parameter scheme, and take the dense distribution interval as the stable process parameter interval in the order feature pattern represented by the cluster. The extracted new experience data is represented as the correspondence between "order characteristic pattern - stable process parameter range".

[0014] Furthermore, the method also includes production anomaly early warning and self-recovery processing: Real-time monitoring of key parameters in the real-time equipment operating condition stream data to determine whether they exceed the safe operating range set by the optimal equipment parameter control scheme or the final equipment parameter execution command; If a key parameter remains abnormal for more than a preset time, a production anomaly alarm will be triggered, and the anomaly root cause diagnosis process will be initiated. The anomaly root cause diagnosis process calls the printing process experience rule base and the defect root cause analysis knowledge graph to quickly infer the main causes of the anomaly and the affected production links. Based on the inference results, generate temporary equipment parameter compensation and adjustment suggestions or maintenance work orders; After automatically performing parameter compensation adjustments or manual intervention maintenance, the equipment operating condition data is re-collected to verify whether the anomaly has been eliminated. The anomaly event and handling measures are recorded in the experience database for optimizing future early warning and diagnostic logic.

[0015] Compared with the prior art, the beneficial effects of the present invention are: By performing data mining on historical order parameter sequences, a printing process experience rule base is constructed, containing specific rules such as the mapping relationship between pattern complexity and ink consumption, and the adaptation table between fabric material and drying temperature. This approach enables the system to automatically invoke quantitative knowledge validated by historical data based on current order characteristics, and to intelligently recommend and predict process parameters. Unlike relying on fixed formulas or fuzzy experience, it autonomously extracts executable knowledge from historical data, directly providing the dynamic simulation module with accurate, data-association-based initial parameter setting basis, improving the scientific nature of parameter setting and the first-time success rate.

[0016] This system utilizes online quality inspection image sequences to not only identify defect types but also generate a heat map reflecting the spatial distribution of defects, feeding this information back to the parameter control process. This approach transforms image-level spatial quality information into guidance for adjusting equipment parameters. The system can analyze the heat map to determine whether defects are random or systematically distributed, thus linking them to abnormal operating conditions or parameter mismatches in specific equipment units such as printing heads, conveyor belts, and drying ovens. Based on this analysis, the optimal equipment parameter control scheme generated by simulation is spatially targeted and corrected, achieving closed-loop control from "seeing defects" to "locating the cause and adjusting corresponding parameters," enabling the production process to self-optimize based on real-time quality feedback. Attached Figure Description

[0017] Figure 1 This is a sequence diagram of the intelligent manufacturing production management system for textile printing based on big data as described in this invention. Figure 2 A flowchart for generating a printing process experience rule base; Figure 3 A flowchart for generating the optimal equipment parameter control scheme; Figure 4 A bar chart showing the statistical types of defects in textile printing production; Figure 5 This is a performance evaluation diagram for handling abnormalities in a textile printing production line. Detailed Implementation

[0018] The technical solution of the present invention will be clearly and completely described below with reference to the embodiments. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0019] See Figure 1 The data acquisition module is responsible for collecting operational status data sets of the entire textile printing production process. This set includes historical order parameter sequences, real-time equipment operating condition data, and online quality inspection image sequences. The knowledge mining module performs production knowledge mining processing on the historical order parameter sequences to generate a printing process experience rule base. This rule base includes the mapping relationship between pattern complexity and ink consumption, the adaptation table between fabric material and drying temperature, and the correlation model between color registration accuracy and equipment adjustment parameters. The dynamic simulation module performs dynamic simulation of the production process based on real-time equipment operating condition data and the printing process experience rule base, generating the optimal equipment parameter control scheme for the current production batch. The quality identification module uses online quality inspection image sequences to perform quality defect pattern recognition on printed products, generating a defect type statistical report and a defect location distribution heat map. The feedback correction module combines the defect type statistical report and the defect location distribution heat map to provide feedback correction to the optimal equipment parameter control scheme, generating the final equipment parameter execution instructions after closed-loop optimization. All modules work together to form a closed-loop control from data acquisition to decision execution, improving the intelligence level of printing production.

[0020] In one embodiment of the present invention, see [reference] Figure 2The system extracts order feature vectors from historical order parameter sequences. These vectors include the number of pattern colors, pattern area coverage, fabric weight, order batch size, and customer-specified colorfastness grade. The order feature vectors are then input into an association rule analysis model. Through confidence and support calculations, frequent correlation patterns between the features of each dimension of the order feature vector and process parameters in subsequent production records are identified. For each discovered frequent correlation pattern, the corresponding production execution record is extracted. This record includes the actual ink formula used, printing screen mesh count, drying oven temperature profile, and equipment operating speed. Successful records that meet preset quality standards are subjected to parameter clustering analysis to summarize stable process parameter ranges under specific order feature vector combinations. The frequent correlation patterns and their corresponding stable process parameter ranges are then structured and stored to form a mapping relationship between pattern complexity and ink consumption, a matching table between fabric material and drying temperature, and a correlation model between color registration accuracy and equipment adjustment parameters.

[0021] In practical implementation, a knowledge mining module of a big data-based intelligent manufacturing production management system for textile printing performs production knowledge mining processing on historical order parameter sequences and generates a printing process experience rule base. The knowledge mining module extracts order feature vectors from the stored historical order parameter sequences. These feature vectors include the number of pattern colors, pattern area coverage, fabric weight, order batch size, and customer-specified colorfastness grade. Specifically, the knowledge mining module automatically obtains the number of pattern colors for each completed order by parsing structured records in the historical order database, the pattern area coverage percentage calculated through image analysis, the fabric weight in grams recorded in the fabric specification table, the order batch size in meters recorded in the order document, and the customer-specified colorfastness grade symbol specified in the contract document.

[0022] In some embodiments, order feature vectors are input into an association rule analysis model. Through confidence and support calculations, frequent association patterns between the features of each dimension in the order feature vector and process parameters in subsequent production records are discovered. The association rule analysis model scans historical datasets to calculate the frequency of different combinations of order features occurring simultaneously with specific process parameter values. The association rule analysis model adopts the form of… The rules indicate that, among them It is a conditional combination of one or more dimensions of the order feature vector. These are one or more process parameters in the production execution record. Association rule analysis models assess the importance of rules by calculating support and confidence. Support represents the frequency of a rule's occurrence in historical data, and confidence represents the frequency of a rule's occurrence under certain conditions. Conclusion when it appears The probability that both are true. Support and confidence are calculated using the following formulas:

[0023] in: The score represents the importance of the rules and is used to filter frequently associated patterns. Representative itemset and The frequency of occurrence of parameters in historical order parameters and corresponding production execution records; Represents order feature conditions In the event of this, process parameters The conditional probability also appears accordingly; and These are preset weighting coefficients used to balance the contributions of support and confidence in importance assessment.

[0024] In practice, for each frequently associated pattern discovered, the knowledge mining module backtracks and extracts its corresponding production execution records. Based on the historical order number matched by the association rules, the knowledge mining module retrieves from the production execution database the actual ink formula composition and ratio, printing screen mesh count, drying room temperature curve data, and equipment operating speed values ​​used in the production process of that order.

[0025] In some embodiments, parametric clustering analysis is performed on successful production execution records that meet preset quality standards to summarize stable process parameter ranges under specific order feature vector combinations. The preset quality standard refers to batch records where the defect rate of the final product is below a set threshold and the production efficiency index meets the standard. The knowledge mining module filters out all successful production execution records that meet the quality standard. For each frequent association pattern defined by a specific order feature vector combination, the actual set of process parameters corresponding to all successful records under that pattern is used as the input dataset for clustering analysis. It can be understood that parametric clustering analysis uses unsupervised learning algorithms, such as density-based clustering methods, to identify densely distributed point sets in the process parameter space as a cluster. It can be understood that the central region of each cluster represents the set of process parameter values ​​that have been practically verified under that order feature pattern and can stably produce qualified products, and the cluster boundary defines the range of stable process parameter intervals.

[0026] In practical implementation, frequent association patterns and corresponding stable process parameter ranges are structured and stored to form specific knowledge entries in the printing process experience rule base. The knowledge mining module encapsulates the conditional combination of order feature vectors, associated stable process parameter ranges, and related statistical confidence measures into a queryable rule. Optionally, the mapping relationship between pattern complexity and ink consumption is composed of a set of rules that include the number of pattern colors and pattern area coverage as conditions, and ink formulation and unit consumption as conclusions. Optionally, the adaptation table of fabric material and drying temperature is composed of a set of rules that include fabric basis weight and color fastness grade as conditions, and drying room temperature curve parameters as conclusions. In practical implementation, the association model between color matching accuracy and equipment adjustment parameters is composed of a set of rules that include the number of pattern colors and order batch as conditions, and the matching range of printing screen mesh number and equipment operating speed as conclusions. Finally, all refined and verified rules are systematically organized and stored in the printing process experience rule base for querying and calling by the dynamic simulation module.

[0027] In one embodiment of the present invention, see [reference] Figure 3The system analyzes the order parameters of the current production batch to form a feature vector for the current order. Using this feature vector as an index, a set of initial recommended process parameters is obtained by matching them in the printing process experience rule base. A digital twin model of the virtual production line, including the printing machine, drying room, and winding device, is constructed. The initial recommended process parameters are used as input to drive the virtual production line digital twin model for simulation. During the simulation, real-time equipment condition data is synchronously accessed, including the real-time value of the printing squeegee pressure, the actual temperature of each zone in the drying room, and the readings of the fabric tension sensor. The real-time equipment condition data is compared with the simulated predicted values ​​of the corresponding nodes in the virtual production line digital twin model to calculate the deviation data. Based on the deviation data, the controllable parameters in the virtual production line digital twin model are dynamically adjusted through a preset model predictive control algorithm to make the model output tend to be stable and meet the preset quality indicators. When the model reaches a stable state, the operating parameters of each device in the virtual production line digital twin model are extracted and packaged into an optimal equipment parameter control scheme. When calculating the deviation data, the real-time value of the printing squeegee pressure is subjected to time series smoothing filtering to obtain the filtered value of the printing squeegee pressure. The instantaneous difference between the squeegee pressure filter value and the squeegee pressure setpoint of the printing machine node in the virtual production line digital twin model is calculated as the squeegee pressure deviation. The average temperature of each zone in the drying oven is calculated by averaging the actual temperatures of each zone. The difference between the average temperature of each zone in the drying oven and the temperature setpoint of the corresponding temperature zone in the drying chamber node of the virtual production line digital twin model is calculated as the temperature control deviation. The fluctuation range of the fabric tension sensor readings is monitored in real time, and its standard deviation is calculated. The standard deviation of the fabric tension sensor readings is compared with the tension fluctuation threshold of the winding device node in the virtual production line digital twin model. When the standard deviation exceeds the threshold, a tension stability deviation is generated. The squeegee pressure deviation, temperature control deviation, and tension stability deviation are summarized to form the deviation data.

[0028] In practical implementation, the dynamic simulation module performs dynamic simulation of the production process based on real-time equipment operating condition data and a printing process experience rule base, generating an optimal equipment parameter control scheme for the current production batch. The module analyzes the order parameters of the current production batch, forming a current order feature vector. This vector includes the number of pattern colors, pattern area coverage, fabric weight, order batch size, and the colorfastness grade specified by the customer. Using this feature vector as an index, the module matches a set of initial recommended process parameters from the printing process experience rule base. These parameters include recommended values ​​for ink formulation, screen mesh count, oven temperature profile, and equipment operating speed. In practical implementation, the dynamic simulation module constructs a virtual production line digital twin model including the printing machine, oven, and winding device. This model simulates the dynamic behavior of actual production equipment using physical equations and machine learning models. The initial recommended process parameters are used as input to drive the virtual production line digital twin model's simulation, outputting predicted states of the printing process, including predicted values ​​for squeegee pressure, oven temperature, and fabric tension.

[0029] In some embodiments, during simulation operation, the dynamic simulation module synchronously accesses real-time equipment condition stream data, which includes the real-time value of the printing squeegee pressure, the actual temperature of each zone in the oven, and the fabric tension sensor readings. The dynamic simulation module compares the real-time equipment condition stream data with the simulated predicted values ​​of the corresponding nodes in the virtual production line digital twin model, calculating the deviation data. The deviation data calculation includes performing time-series smoothing filtering on the real-time value of the printing squeegee pressure to obtain a filtered value. The instantaneous difference between the filtered value and the squeegee pressure setpoint of the printing machine node in the virtual production line digital twin model is calculated as the squeegee pressure deviation. The average temperature of each zone in the oven is calculated by averaging the actual temperatures of each zone. The difference between the average temperature of each zone in the oven and the temperature setpoint of the corresponding temperature zone in the drying room node of the virtual production line digital twin model is calculated as the temperature control deviation. The fluctuation range of the fabric tension sensor readings is monitored in real time, and its standard deviation is calculated. The standard deviation of the fabric tension sensor readings is compared with the tension fluctuation threshold of the winding device node in the digital twin model of the virtual production line. When the standard deviation exceeds the threshold, a tension stability deviation is generated. The deviation data is then compiled from the doctor blade pressure deviation, temperature control deviation, and tension stability deviation.

[0030] In practical implementation, based on deviation data, the dynamic simulation module dynamically adjusts the controllable parameters in the virtual production line digital twin model through a preset model predictive control algorithm. Within each simulation time step, the model predictive control algorithm solves an optimization problem to minimize the prediction deviation over a future period. The adjustment amount of the controllable parameters is calculated using the following formula:

[0031] in: Representative at the The adjustment vector of controllable parameters within each time step; Representative at the The deviation data vector calculated within each time step includes scraper pressure deviation, temperature control deviation, and tension stability deviation. and These are preset control gain matrices, corresponding to proportional and integral actions respectively; Represents the period from the start of simulation to the [number]th [stage]. The cumulative deviation of each step. By dynamically adjusting controllable parameters, the output of the virtual production line digital twin model tends to be stable and meets the preset quality indicators, including printing uniformity, drying degree, and fabric flatness.

[0032] It is understandable that once the virtual production line digital twin model reaches a stable state, the dynamic simulation module extracts the operating parameters of each device in the virtual production line digital twin model at this time. These operating parameters include the final set value of the printing squeegee pressure, the final temperature set curves for each temperature zone of the oven, and the final tension control parameters of the winding device. The dynamic simulation module packages these operating parameters into an optimal equipment parameter control scheme. Optionally, when calculating the standard deviation of the fabric tension sensor readings, a reading sequence within a rolling time window is used. Optionally, the time series smoothing filtering process uses an exponentially weighted moving average method. It is understandable that the stable state of the virtual production line digital twin model is defined as the fluctuation amplitude of key output variables consistently being below a preset threshold. In some embodiments, the optimal equipment parameter control scheme is output to the production execution system in the form of digital instructions to drive the actual equipment to make adjustments.

[0033] In one embodiment of the present invention, high-definition images of the surface of printed finished products are continuously acquired by an industrial camera deployed at the end of the production line, forming an online quality inspection image sequence. Each high-definition image is preprocessed, including color space conversion, illumination unevenness correction, and pattern area localization and segmentation. The preprocessed image area is input into a deep learning defect detection network, which is pre-trained using image samples labeled with defect types such as stains, misregistration, white showing, and bleeding. The deep learning defect detection network outputs recognition results including defect bounding box coordinates, defect type confidence, and defect category labels. The continuous recognition results are archived according to production time sequence and fabric roll number. The frequency and average confidence of various defects within a production batch or a preset time window are statistically analyzed to generate a defect type statistical report. The bounding box coordinates of all defects on the same roll of fabric are mapped to the global length-width coordinate system of the fabric roll, and color rendering is performed according to the defect density to generate a defect location distribution heatmap.

[0034] In specific implementation, the quality identification module performs quality defect pattern recognition on printed finished products using online quality inspection image sequences, generating defect type statistical reports and defect location distribution heatmaps. The quality identification module continuously acquires high-definition images of the printed finished product surface using industrial cameras deployed at the end of the production line, forming an online quality inspection image sequence. The acquisition frequency of the industrial cameras is synchronized with the conveyor speed of the production line to ensure continuous image coverage of the entire fabric surface. In specific implementation, each high-definition image undergoes preprocessing. Preprocessing steps include color space conversion, illumination unevenness correction, and pattern region localization and segmentation. Color space conversion converts the image from RGB color space to LAB color space to better separate color and brightness information. Illumination unevenness correction uses an algorithm based on Retinex theory to compensate for changes in light intensity. Pattern region localization and segmentation uses edge detection and contour extraction algorithms to determine the precise boundary regions of the printed pattern. In some embodiments, the preprocessed image regions are input into a deep learning defect detection network. The deep learning defect detection network is pre-trained using image samples labeled with defect types such as stains, misregistration, white gaps, and bleeding. The training samples contain thousands of defect images with precise bounding boxes and category labels. The output of a deep learning defect detection network includes the defect bounding box coordinates, defect type confidence score, and defect category label. The defect bounding box coordinates represent the location of the defect in the image in pixel coordinates, and the defect type confidence score is a value between 0 and 1 that represents the network's certainty about the recognition result.

[0035] It is understandable that the continuous identification results are archived according to the production time sequence and fabric roll number. The system creates an independent database record for each roll of fabric, storing the defect identification results corresponding to each frame of image in timestamp order. In specific implementation, the frequency and average confidence level of various defects within a production batch or a preset time window are statistically analyzed to generate a defect type statistical report. For a preset time window lasting one hour, the system may count stain defects appearing 15 times with an average confidence level of 0.92, misregistration defects appearing 8 times with an average confidence level of 0.87, white showing defects appearing 5 times with an average confidence level of 0.85, and color bleeding defects appearing 3 times with an average confidence level of 0.89. The defect type statistical report is presented in tabular form, including fields such as defect category, frequency of occurrence, average confidence level, and percentage of total defects. In some embodiments, the bounding box coordinates of all defects on the same roll of fabric are mapped to the global length-width coordinate system of the fabric roll. The mapping process uses encoder signals from the industrial camera and the fabric conveyor shaft to perform coordinate transformation, converting image pixel coordinates into actual fabric position coordinates in meters. Optionally, color rendering is performed based on defect density to generate a heatmap of defect location distribution. The defect density is calculated using a kernel density estimation method, and the color intensity of the heatmap is proportional to the local defect density. At the fabric coordinate point The calculation formula at this location is:

[0036] in: Represents the point in the global length-width coordinate system of the fabric. Defect density value at; This represents the total number of defects identified on the roll of fabric. Representing the The coordinates of the center point of the defect in the global coordinate system; Representing the The weight of each defect can be assigned by its defect type confidence level; Represents kernel functions, such as the Gaussian kernel function; The bandwidth parameter controls the smoothness of the density estimation. The heatmap uses a gradient from blue to red, with blue areas representing low defect density and red areas representing high defect density. Optionally, the defect location distribution heatmap can be overlaid with the actual fabric pattern design to visually demonstrate the relationship between defect distribution and pattern features. In practice, after generating the defect type statistical report and the defect location distribution heatmap, the quality identification module transmits the results to the feedback correction module for subsequent analysis.

[0037] In one embodiment of the invention, a defect type statistical report is analyzed to identify the dominant defect type in the current production batch. A preset defect root cause analysis knowledge graph is queried, which defines the causal relationship chain between different defect types and potential equipment parameter mismatches or improper process conditions. Based on the identified dominant defect type, one or more sets of equipment parameters or process conditions suspected of causing the dominant defect type are derived from the defect root cause analysis knowledge graph. The defect location distribution heatmap is overlaid with the equipment layout diagram of the production line to locate the physical production links or equipment units corresponding to the high-incidence areas of defects. Combining the suspected defect-causing equipment parameters or process conditions with the physical production links corresponding to the high-incidence areas of defects, specific equipment parameter items that need to be adjusted are screened out. For the screened specific equipment parameter items, their current set values ​​in the optimal equipment parameter control scheme are calculated, and fine-tuning is performed according to a preset correction rule library, which contains parameter adjustment step sizes and directions for different defect types and severity. The fine-tuned equipment parameters are integrated to form the final equipment parameter execution instructions after closed-loop optimization. When querying a pre-defined defect root cause analysis knowledge graph, the graph is stored as nodes and edges. Nodes represent production entities or states, including equipment components, process parameters, material properties, and defect types. Edges represent causal or influence relationships between entities or states. When a specific defect type node is input, a backtracking traversal is performed along the incoming edges pointing to that node. During the backtracking traversal, all upstream nodes that directly or indirectly point to the defect type node are recorded. Upstream nodes include equipment parameter nodes and process condition nodes. Based on edge weights or causal strength statistically derived from historical data, the upstream nodes are sorted. Upstream nodes with higher weights or stronger causal strength are prioritized as associated root causes. The sorted list of upstream nodes is output as one or more sets of equipment parameters or process conditions suspected of causing the dominant defect type.

[0038] In practical implementation, the feedback correction module performs feedback correction on the optimal equipment parameter control scheme by combining the defect type statistical report and the defect location distribution heatmap, and generates the final equipment parameter execution instruction after closed-loop optimization. The feedback correction module analyzes the defect type statistical report generated by the quality identification module to identify the dominant defect type in the current production batch. The defect type statistical report lists the frequency and average confidence of each type of defect in tabular form. In a specific scenario, referring to Table 1, the defect type statistical report may contain data as shown in Table 1. The report shows that the frequency of the "color mismatch" defect is significantly higher than that of other defect types and the average confidence is high, therefore it is identified as the dominant defect type.

[0039] Table 1: Defect Type Statistics Table

[0040] In practice, the feedback correction module queries a pre-defined defect root cause analysis knowledge graph to deduce the root cause. This knowledge graph is stored as nodes and edges. Nodes represent production entities or states, including equipment components, process parameters, material properties, and defect types. Edges represent causal or influence relationships between entities or states. When the specific defect type node "color misregistration" is input, the feedback correction module backtracks along the incoming edges pointing to this node. During this backtracking process, it records all upstream nodes that directly or indirectly point to the "color misregistration" node. Upstream nodes include equipment parameter nodes and process condition nodes, such as nodes like "printing screen alignment accuracy," "squeegee pressure stability," and "fabric tension fluctuation." Based on edge weights or the causal strength statistically derived from historical data, the upstream nodes are sorted. Upstream nodes with higher weights or stronger causal strength are prioritized as associated root causes. The sorted list of upstream nodes is output as one or more sets of suspected equipment parameters or process conditions leading to the dominant defect type.

[0041] In some embodiments, the feedback correction module overlays and analyzes the defect location distribution heatmap with the equipment layout diagram of the production line to locate the physical production links or equipment units corresponding to the high-incidence areas of defects. The red high-density areas shown on the defect location distribution heatmap are mapped to specific units of the printing machine or specific heating sections of the drying oven in the equipment layout diagram. By combining the equipment parameters or process conditions suspected of causing the defects with the physical production links corresponding to the high-incidence areas of defects, specific equipment parameter items that need to be adjusted are selected. For example, when the high-incidence area of ​​defects corresponds to the third printing unit of the printing machine, and the suspected root cause includes "squeegee pressure stability", then "squeegee pressure of the third printing unit" is selected as the specific equipment parameter item that needs to be adjusted.

[0042] In practice, for each selected specific equipment parameter, the feedback correction module calculates its current setpoint within the optimal equipment parameter control scheme and makes fine adjustments based on a pre-defined correction rule library. The correction rule library contains parameter adjustment steps and directions for different defect types and severity levels, stored in a structured data table format. Parameter adjustment amount The calculation follows the formula:

[0043] in: This represents the amount of adjustment to a specific equipment parameter item; This represents the gain coefficient determined based on the severity of the defect; This represents the baseline adjustment step size defined for this defect type in the rule base. This represents the adjustment direction function, the value of which is determined by the causal relationship between the defect type and the parameters. A value of +1 could indicate that pressure needs to be increased, or -1 could indicate that pressure needs to be decreased.

[0044] It is understandable that all fine-tuned equipment parameters are integrated to form a closed-loop optimized final equipment parameter execution instruction. In some embodiments, the final equipment parameter execution instruction is sent to the central control system of the production line in the form of a structured data packet. Optionally, the baseline adjustment step size in the rule base is corrected. It can be dynamically updated based on historical optimization results. Optional, defect severity gain coefficient. Quantitative calculations can be performed based on the size of high-density areas in the defect location heatmap. In practice, after the feedback correction module completes parameter fine-tuning and command generation, it stores the defect type, root cause analysis results, and parameter adjustment records used for this correction in the log for periodic updates to the printing process experience rule base.

[0045] See Figure 4 This is a bar chart statistically analyzing defect types in textile printing production. It quantifies the occurrence of four main printing defects from two dimensions: frequency of occurrence and average confidence level. "Inaccurate color registration," as the dominant defect, should be the primary target for production optimization. Priority should be given to checking core parameters such as printing screen alignment accuracy and squeegee pressure stability, using a defect location heatmap to pinpoint high-incidence areas for targeted adjustments. The low confidence level of "bleed-through" indicates insufficient detection algorithm capability for this type of defect, requiring more samples for model optimization. High-frequency, high-confidence defect data can be directly added to the printing process experience rule base, providing negative examples for parameter recommendations in subsequent orders and preventing the recurrence of similar problems. These statistical results can directly drive the feedback correction module to generate equipment parameter adjustment instructions for "inaccurate color registration," achieving closed-loop control from defect identification to parameter optimization.

[0046] In one embodiment of the present invention, historical order parameter sequences, optimal equipment parameter control schemes or final equipment parameter execution instructions, and corresponding online quality inspection image sequence analysis results for all production batches within a certain period are collected. The experience data of successful production batches is refined; a successful production batch refers to a batch whose final product defect rate is below a predetermined threshold and whose production efficiency reaches a predetermined standard. The refined new experience data, including combinations of new order feature vectors and stable process parameter ranges, is merged with existing rules in the printing process experience rule base. Rules in the printing process experience rule base that have not been called for a long time or deviate significantly from the latest production data statistical trend are downgraded or archived. The merged and optimized rule data is used to replace or incrementally update the original printing process experience rule base. When refining the experience data of successful production batches, for a single successful production batch, its complete order feature vector and the actual executed equipment parameter scheme are extracted. The online quality inspection image sequence analysis results are analyzed to confirm that the products of successful production batches perform excellently in all quality indicators. The order feature vector and the actual executed equipment parameter scheme are considered as a successful case pair. In a set of multiple successful case pairs, cases with similar order feature vectors are clustered. The system calculates the statistical distribution of each parameter value in the equipment parameter scheme for each cluster, and takes the densely distributed interval as the stable process parameter interval in the order feature pattern represented by the cluster. The extracted new experience data is represented as the correspondence between order feature pattern and stable process parameter interval. The system also includes production anomaly early warning and self-recovery processing, which monitors key parameters in real-time equipment operating condition data and determines whether they exceed the safe operating range set by the optimal equipment parameter control scheme or the final equipment parameter execution instruction. If the key parameter remains abnormal for more than a preset time, a production anomaly alarm is triggered, and the anomaly root cause diagnosis process is initiated. The anomaly root cause diagnosis process calls the printing process experience rule base and defect root cause analysis knowledge graph to quickly infer the main cause of the anomaly and the affected production links. Based on the inference results, temporary equipment parameter compensation adjustment suggestions or maintenance work orders are generated. After automatic parameter compensation adjustment or manual intervention maintenance, equipment operating condition data is re-collected to verify whether the anomaly has been eliminated, and the anomaly event and handling measures are recorded in the experience base for optimizing future early warning and diagnosis logic.

[0047] In practice, the system periodically updates the printing process experience rule base. This involves collecting historical order parameter sequences, optimal equipment parameter control schemes, or final equipment parameter execution instructions for all production batches within a certain period, along with corresponding online quality inspection image sequence analysis results. A collection period can be set to one month or a cumulative total of one hundred production batches. The experience data from successful production batches is refined. A successful production batch is defined as one where the final product defect rate is below a predetermined threshold and production efficiency meets a predetermined standard. The predetermined threshold is, for example, a defect rate of 0.5%, and the predetermined standard is, for example, achieving a target output per unit time. The refined new experience data, including combinations of new order feature vectors and stable process parameter ranges, is then integrated with existing rules in the printing process experience rule base. This integration process involves a weighted average of the confidence levels of the new and old rules and arbitration of conflicting rules. Rules in the printing process experience rule base that have not been called for a long time or deviate significantly from the latest production data statistical trends are downgraded or archived; for example, rules that have not been matched by any orders within a year are marked as dormant rules. The integrated and optimized rule data is then used to replace or incrementally update the original printing process experience rule base.

[0048] In some embodiments, the process of refining the experience data from successful production batches includes extracting the complete order feature vector and the actual executed equipment parameter scheme for a successful production batch. The order feature vector includes the number of pattern colors, pattern area coverage, fabric weight, order batch size, and customer-specified colorfastness grade. The actual executed equipment parameter scheme originates from the final equipment parameter execution instructions actually used in that batch. Analysis of online quality inspection image sequence analysis results confirms that the products in the successful production batch perform excellently in all quality indicators, including the number of defects and defect severity. The order feature vector and the actual executed equipment parameter scheme are stored as a successful case pair. In multiple successful case pair sets, cases with similar order feature vectors are clustered, and Euclidean distance is used to measure the similarity between order feature vectors. For each cluster, the statistical distribution of each parameter value in its equipment parameter scheme is calculated. The densely distributed interval is taken as the stable process parameter interval in the order feature pattern represented by the cluster. The densely distributed interval can be determined by calculating the kernel density estimation of parameter values ​​and selecting the region with the highest probability density. The refined new experience data is represented as the correspondence between order feature patterns and stable process parameter intervals, and an initial confidence level is assigned. New experience data confidence level The calculation follows the formula:

[0049] in: The initial confidence level represents the correspondence between the newly extracted order feature patterns and stable process parameter ranges; This represents a preset baseline credibility coefficient; This represents the number of times the order's characteristic pattern occurred within the period covered by the extracted data; This represents the total number of all successfully produced batches within that period.

[0050] Understandably, the system also includes production anomaly early warning and self-recovery processing, real-time monitoring of key parameters in the real-time equipment operating data stream. These key parameters include the real-time value of the printing squeegee pressure, the actual temperature of each zone in the oven, and the fabric tension sensor readings. It determines whether these key parameters exceed the safe operating range set by the optimal equipment parameter control scheme or the final equipment parameter execution command. The safe operating range is typically set to ±5% of the parameter set value. If a key parameter remains abnormal for more than a preset duration, a production anomaly alarm is triggered, and the anomaly root cause diagnosis process is initiated. The preset duration is, for example, 30 consecutive seconds. The anomaly root cause diagnosis process calls upon the printing process experience rule base and defect root cause analysis knowledge graph to quickly infer the main cause of the anomaly and the affected production process. For example, it infers that a persistently low temperature in the third zone of the oven may lead to insufficient drying. Based on the inference results, temporary equipment parameter compensation adjustment suggestions or maintenance work orders are generated. Parameter compensation adjustment suggestions, for example, temporarily increasing the temperature set value of the third zone of the oven by five degrees Celsius. After automatically executing parameter compensation adjustments or manual intervention maintenance, equipment operating data is re-collected to verify whether the anomaly has been eliminated. Record this abnormal event and the handling measures in the experience base to optimize future early warning and diagnosis logic, such as updating the judgment threshold for specific parameter anomalies.

[0051] Optionally, when merging new and old rules, rules describing the same order characteristic pattern are merged using a confidence-weighted average. Optionally, rules that have not been called for a long time are moved to the archive library instead of being directly deleted if their confidence level falls below a certain threshold after being downweighted. In some embodiments, the defect rate threshold and production efficiency standard in the criteria for judging successful production batches can be dynamically configured according to customer requirements or production plans. It is understood that when the anomaly root cause diagnosis process calls the defect root cause analysis knowledge graph, it uses the same backtracking traversal and causal strength sorting logic as the knowledge graph query in the feedback correction module. In specific implementations, the anomaly event information recorded in the experience base includes anomaly parameters, duration, inferred root cause, measures taken, and parameter recovery status after the measures take effect. This information is used to train a more accurate anomaly prediction model.

[0052] See Figure 5This is a performance evaluation chart for handling anomalies in a textile printing production line. It quantifies the handling efficiency of five typical production anomalies from three dimensions: anomaly detection time, recovery time, and recovery success rate. Anomalies with longer recovery times (such as ink flow anomalies) tend to have lower recovery success rates, indicating higher complexity and uncertainty in handling such anomalies. Anomalies with shorter detection times usually have more obvious characteristics, resulting in higher algorithm recognition efficiency. "Ink flow anomalies" and "oven temperature fluctuations" are currently the main bottlenecks in production, requiring priority optimization of their handling processes and algorithm models. For high-risk anomalies like "ink flow anomalies," equipment layout can be optimized, and online flow monitoring sensors can be added to improve detection accuracy; simultaneously, more mature parameter adjustment rules can be developed to shorten recovery time. For the low success rate of "ink flow anomalies," more anomaly samples can be added, and the feature extraction capability of the detection algorithm can be optimized to improve the accuracy of early warnings.

[0053] The above embodiments are only used to illustrate the technical methods of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical methods of the present invention without departing from the spirit and scope of the technical methods of the present invention.

Claims

1. A smart manufacturing production management system for textile printing based on big data, characterized in that, The system includes: The data acquisition module is used to collect a set of operational status data for the entire textile printing production process. The set of operational status data includes historical order parameter sequences, real-time equipment operating condition data, and online quality inspection image sequences. The knowledge mining module is used to perform production knowledge mining processing on the historical order parameter sequence to generate a printing process experience rule base. The printing process experience rule base includes the mapping relationship between pattern complexity and ink consumption, the adaptation table between fabric material and drying temperature, and the correlation model between color matching accuracy and equipment adjustment parameters. The dynamic simulation module is used to perform dynamic simulation of the production process based on the real-time equipment operating condition data and the printing process experience rule base, and generate the optimal equipment parameter control scheme for the current production batch. The quality identification module is used to identify quality defect patterns in printed finished products using the online quality inspection image sequence, and generate a statistical report on defect types and a heat map of defect location distribution. The feedback correction module is used to combine the defect type statistical report and the defect location distribution heat map to perform feedback correction on the optimal equipment parameter control scheme and generate the final equipment parameter execution instruction after closed-loop optimization.

2. The intelligent manufacturing production management system for textile printing based on big data as described in claim 1, characterized in that, The step of performing production knowledge mining processing on the historical order parameter sequence to generate a printing process experience rule base includes: Extract the order feature vector from the historical order parameter sequence. The order feature vector includes the number of pattern colors, pattern area coverage, fabric weight, order batch, and color fastness grade specified by the customer. The order feature vector is input into the association rule analysis model. Through confidence and support calculations, frequent association patterns between the features of each dimension in the order feature vector and the process parameters in subsequent production records are discovered. For each frequently associated pattern discovered, the corresponding production execution records are retrieved by backtracking. The production execution records include the actual ink formula used, the mesh count of the printing screen, the oven temperature profile, and the equipment operating speed. Perform parameter clustering analysis on the successful records that meet the preset quality standards in the production execution records to summarize the stable process parameter range under the specific order feature vector combination; The frequent correlation patterns and their corresponding stable process parameter ranges are structured and stored to form a mapping relationship between pattern complexity and ink consumption, an adaptation table between fabric material and drying temperature, and a correlation model between color matching accuracy and equipment adjustment parameters.

3. The intelligent manufacturing production management system for textile printing based on big data as described in claim 1, characterized in that, The process of dynamically simulating the production process based on the real-time equipment operating condition data and the printing process experience rule base to generate an optimal equipment parameter control scheme for the current production batch includes: Analyze the order parameters of the current production batch to form the feature vector of the current order; Using the current order feature vector as an index, a set of initial recommended process parameters are obtained by matching in the printing process experience rule base; Construct a digital twin model of a virtual production line that includes printing machines, drying rooms, and winding devices; The initial recommended process parameters are used as input to drive the virtual production line digital twin model to perform simulation operation; During the simulation operation, the real-time equipment condition stream data is synchronously accessed. The real-time equipment condition stream data includes the real-time value of the printing squeegee pressure, the actual temperature of each zone of the oven, and the readings of the fabric tension sensor. The deviation data is calculated by comparing the real-time equipment operating condition data with the simulated predicted values ​​of the corresponding nodes in the virtual production line digital twin model. Based on the deviation data, the controllable parameters in the virtual production line digital twin model are dynamically adjusted through a preset model prediction control algorithm, so that the model output tends to be stable and meets the preset quality indicators. Once the model reaches a stable state, the operating parameters of each device in the virtual production line digital twin model are extracted and packaged into the optimal device parameter control scheme.

4. The intelligent manufacturing production management system for textile printing based on big data according to claim 3, characterized in that, The step of comparing the real-time equipment operating condition data with the simulated predicted values ​​of the corresponding nodes in the virtual production line digital twin model to calculate the deviation data includes: The real-time value of the printing squeegee pressure is subjected to time series smoothing filtering to obtain the filtered value of the printing squeegee pressure. The instantaneous difference between the filtered value of the printing squeegee pressure and the squeegee pressure setting value of the printing machine node in the virtual production line digital twin model is calculated as the squeegee pressure deviation; The average temperature of each zone of the oven is calculated by averaging the actual temperatures of each zone. The difference between the average temperature of each zone of the oven and the temperature setpoint of the corresponding temperature zone in the virtual production line digital twin model is calculated as the temperature control deviation. The fluctuation range of the fabric tension sensor readings is monitored in real time, and its standard deviation is calculated; The standard deviation of the fabric tension sensor readings is compared with the tension fluctuation threshold of the winding device node in the virtual production line digital twin model. When the standard deviation exceeds the threshold, a tension stability deviation is generated. The deviation data is formed by summing the scraper pressure deviation, the temperature control deviation, and the tension stability deviation.

5. The intelligent manufacturing production management system for textile printing based on big data according to claim 1, characterized in that, The step of using the online quality inspection image sequence to perform quality defect pattern recognition on the printed finished product, and generating a defect type statistical report and a defect location distribution heat map, includes: The online quality inspection image sequence is formed by continuously acquiring high-definition images of the surface of the printed finished product using industrial cameras deployed at the end of the production line. Each of the high-definition images is preprocessed, and the preprocessing steps include color space conversion, illumination unevenness correction, and pattern region localization and segmentation. The pre-processed image region is input into a deep learning defect detection network, which is pre-trained using image samples labeled with defect types such as stains, misregistration, white gaps, and color bleeding. The deep learning defect detection network outputs recognition results including defect bounding box coordinates, defect type confidence, and defect category label; Archive the continuous identification results according to the production time sequence and fabric roll number; Statistical report of the defect types is generated by statistically analyzing the frequency and average confidence level of various defects within a production batch or a preset time window. The bounding box coordinates of all defects on the same roll of fabric are mapped to the global length-width coordinate system of the fabric roll, and the defects are colored and rendered according to the defect density to generate a heat map of the defect location distribution.

6. The intelligent manufacturing production management system for textile printing based on big data according to claim 1, characterized in that, The step of combining the defect type statistical report and the defect location distribution heatmap to provide feedback correction to the optimal equipment parameter control scheme, and generating the final equipment parameter execution instruction after closed-loop optimization, includes: Analyze the defect type statistics report to identify the dominant defect type in the current production batch; The system queries a pre-defined defect root cause analysis knowledge graph, which defines causal relationship chains between different defect types and potential equipment parameter mismatches or improper process conditions. Based on the identified dominant defect type, one or more sets of equipment parameters or process conditions suspected of causing the dominant defect type are derived from the defect root cause analysis knowledge graph. The heat map of the defect location distribution is overlaid with the equipment layout diagram of the production line to locate the physical production links or equipment units corresponding to the high-incidence areas of defects. By combining the equipment parameters or process conditions suspected of causing the defects with the physical production links corresponding to the high-incidence areas of the defects, specific equipment parameter items that need to be adjusted are selected. For the selected specific equipment parameter items, calculate their current set value in the optimal equipment parameter control scheme, and make fine adjustments according to the preset correction rule library, which includes parameter adjustment step size and direction for different defect types and severity. The finely tuned equipment parameters are integrated to form the final equipment parameter execution instructions after closed-loop optimization.

7. The intelligent manufacturing production management system for textile printing based on big data according to claim 6, characterized in that, The query preset defect root cause analysis knowledge graph includes: The defect root cause analysis knowledge graph is stored in the form of nodes and edges. Nodes represent production entities or states, including equipment components, process parameters, material properties and defect types, while edges represent causal or influence relationships between entities or states. When a specific defect type node is input, backtracking is performed along the incoming edges pointing to the defect type node; During the backtracking traversal, all upstream nodes that directly or indirectly point to the defect type node are recorded, including equipment parameter nodes and process condition nodes. The upstream nodes are sorted according to the weight of the edges or the causal strength statistically derived from historical data. The upstream nodes with higher weights or stronger causal strength are preferentially identified as the root causes of the association. Output a sorted list of upstream nodes as one or more sets of equipment parameters or process conditions suspected of causing the dominant defect type.

8. The intelligent manufacturing production management system for textile printing based on big data according to claim 1, characterized in that, The method also includes the step of periodically updating the printing process experience rule base: Collect historical order parameter sequences of all production batches within a certain period, the optimal equipment parameter control scheme or final equipment parameter execution instructions adopted, and the corresponding online quality inspection image sequence analysis results; The experience data of successful production batches are extracted. A successful production batch refers to a batch in which the final product defect rate is lower than a predetermined threshold and the production efficiency reaches a predetermined standard. The extracted new experience data, including the combination of new order feature vectors and stable process parameter ranges, will be integrated with the existing rules in the printing process experience rule base. For rules in the printing process experience rule base that have not been called for a long time or that deviate too much from the latest production data statistical trend, their weight will be reduced or they will be archived. Use the integrated and optimized rule data to replace or incrementally update the original printing process experience rule library.

9. A smart manufacturing production management system for textile printing based on big data as described in claim 8, characterized in that, The process of extracting experience data from successful production batches includes: For a single successful production batch, extract its complete order feature vector and the actual equipment parameter scheme executed. Analysis of the online quality inspection image sequence analysis results confirmed that the products of the successfully produced batch performed excellently in all quality indicators; The order feature vector and the actual executed equipment parameter scheme are considered as a successful "case pair"; Cluster cases with similar order feature vectors from multiple successful "case pairs" sets; For each cluster, calculate the statistical distribution of each parameter value in its equipment parameter scheme, and take the dense distribution interval as the stable process parameter interval in the order feature pattern represented by the cluster. The extracted new experience data is represented as the correspondence between "order characteristic pattern - stable process parameter range".

10. The intelligent manufacturing production management system for textile printing based on big data according to claim 1, characterized in that, The method also includes production anomaly early warning and self-recovery processing: Real-time monitoring of key parameters in the real-time equipment operating condition stream data to determine whether they exceed the safe operating range set by the optimal equipment parameter control scheme or the final equipment parameter execution command; If a key parameter remains abnormal for more than a preset time, a production anomaly alarm will be triggered, and the anomaly root cause diagnosis process will be initiated. The anomaly root cause diagnosis process calls the printing process experience rule base and the defect root cause analysis knowledge graph to quickly infer the main causes of the anomaly and the affected production links. Based on the inference results, generate temporary equipment parameter compensation and adjustment suggestions or maintenance work orders; After automatically performing parameter compensation adjustments or manual intervention maintenance, the equipment operating condition data is re-collected to verify whether the anomaly has been eliminated. The anomaly event and handling measures are recorded in the experience database for optimizing future early warning and diagnostic logic.

Images