An ai intelligent printing and dyeing cloth inspection comprehensive analysis management and control system and method

Abstract

The application discloses an AI intelligent printing and dyeing cloth inspection comprehensive analysis management and control system and method, relates to the field of textile printing and dyeing quality inspection, and comprises the following steps: an image acquisition unit synchronously acquires cloth surface images and binds meter positions and width coordinates through an industrial camera in cooperation with an encoder; an AI recognition unit pre-processes and recognizes defects, and outputs defect categories, confidence, sizes and coordinates; a database and a big data analysis unit aggregate and statistically analyze orders, machines, process parameters and recognition results according to dyeing plants, batches, colors and spatial positions, and automatically recognizes dyeing and finishing process problems according to spatial distribution modes; and a solution recommendation unit matches reasons, investigates projects and proposes processing suggestions, and outputs the same in a sorted manner from an industry knowledge base. The application realizes the correlation and investigation of dyeing plant process problems from defect detection, forms a detection, analysis, recommendation, service, re-inspection and model iteration closed loop.

Description

Technical Field

[0001] This application relates to the field of textile printing and dyeing quality inspection and production management technology, specifically to an AI intelligent printing and dyeing fabric inspection comprehensive analysis and control system and method. Background Technology

[0002] In the fabric inspection process at dyeing plants, fabric traders, and the garment supply chain, ordinary fabric inspection machines are typically used in conjunction with manual visual inspection. The fabric runs continuously through unwinding, guiding, lighting, and rewinding mechanisms. The inspector stands in front of the machine and visually observes the fabric surface, manually judging defects such as holes, creases, color variations, oil stains, weft skew, snags, and broken yarns. The defect name, location, deduction points, and grade are recorded in paper forms, Excel spreadsheets, or ordinary quality inspection systems. Some factories add cameras or photographing devices to the fabric inspection machines, but these are usually only used to store images or generate static reports. Subsequent image screening, classification, and interpretation still rely on manual processing. Some machine vision fabric inspection equipment can also capture fabric images through industrial cameras and use traditional image processing or AI models to identify local defects, outputting inspection reports, defect coordinates, defect images, and quality grades.

[0003] However, current practices mostly remain at the level of discovering defects and generating reports. There is no closed loop between fabric inspection machines, AI recognition results, dyeing plant production problem analysis, mobile visualization, and follow-up service suggestions. Specifically: manual fabric inspection is easily affected by fatigue, experience, and lighting conditions; the judgment standards for the same defect vary among different personnel, easily leading to missed detections, false detections, and inconsistent deductions; traditional recording methods typically only record the defect name and approximate location in meters, lacking structured data such as two-dimensional coordinates of the fabric surface, defect images, area or length, confidence level, and batch trends, making it difficult to review dyeing plant problems; even if existing AI fabric inspection systems can... While capable of identifying defects, the system typically only outputs inspection results without conducting big data analysis on defect types, locations, concentration areas, times, dyeing plants, machines, orders, and process parameters. It cannot automatically identify high-frequency problems, abnormal batches, and potential causes. Customers or managers can only see report documents or tables, and cannot intuitively view defect maps, batch comparisons, dyeing plant rankings, problem trends, risk warnings, and rectification progress within the mini-program. For issues such as uneven coloring, color difference, oil stains, weft skew, holes, snags, creases, and broken yarns, the system cannot automatically match possible causes, troubleshooting paths, and handling suggestions based on AI recognition results. Summary of the Invention

[0004] The purpose of this application is to provide an AI-powered intelligent fabric inspection and control system, comprising: an image acquisition unit for synchronously acquiring fabric images using an industrial camera and encoder, obtaining fabric images bound to the current meter position and width coordinates; an AI recognition unit for preprocessing the fabric images and identifying defects, outputting defect type, confidence level, size, and the meter position and width coordinates of the defects on the entire roll of fabric; a database and big data analysis unit for storing order numbers, customers, dyeing plants, machines, vat numbers, roll numbers, fabric types, colors, process parameters, and recognition results for multiple batches, and aggregating and statistically analyzing them by dyeing plant, machine, batch, color, time period, defect type, and spatial location, automatically identifying dyeing and finishing process problems based on the spatial distribution pattern of defects on multiple batches of fabric; and a solution recommendation unit for matching possible causes, investigation items, and handling suggestions from a pre-set industry knowledge base of defect type-possible causes-investigation items-suggested measures-service cases based on the recognition results of the spatial distribution pattern, and outputting them in priority order.

[0005] By adopting the above technical solution, the image acquisition unit, in conjunction with the encoder, obtains fabric images bound to meter and width coordinates. The AI ​​recognition unit outputs defect information with two-dimensional coordinates. The database and big data analysis unit aggregates and statistically analyzes multiple batches of data according to dyeing plant, machine, batch, color, time period, defect category, and spatial location. Based on the spatial distribution pattern, it automatically identifies dyeing and finishing process problems. The solution recommendation unit matches possible causes, troubleshooting items, and handling suggestions from a pre-set industry knowledge base and outputs them according to priority. This solves the problem that existing AI fabric inspection systems only output detection results and cannot be upgraded to dyeing plant-level process diagnosis. It realizes automatic association and troubleshooting guidance from fabric defect detection to dyeing plant process problems, reduces reliance on human experience judgment, and improves the efficiency of discovering and responding to quality problems in dyeing plants.

[0006] Optionally, the image acquisition unit includes an encoder connected to the guide roller or the main drive shaft; the industrial camera synchronously acquires fabric images according to the fabric running distance based on the signal output by the encoder; each image is bound to the current meter position, width coordinates, acquisition time, and machine information.

[0007] By adopting the above technical solution, the encoder is connected to the guide roller or the main drive shaft, and the industrial camera is triggered by the encoder signal to synchronously collect data according to the running distance of the fabric. Each image is bound to the current meter position, width coordinates, collection time and machine information, so that each frame of fabric image carries a precise physical position and time label. This avoids the problem of spatial distribution pattern recognition being unable to be performed due to missing synchronous collection or incomplete coordinate binding, and ensures the data availability of dyeing plant-level spatial diagnostic analysis.

[0008] Optionally, the AI ​​recognition unit performs preprocessing on the acquired image, including brightness equalization, background correction, texture enhancement, noise reduction, distortion correction, segmentation, and stitching, and calls the defect detection model to identify defects; the output defect information also includes an outer bounding box or segmentation contour used to surround and mark the location of the defect, the dimensions of which include length, width, and area; uncertain defects are marked as pending review.

[0009] By adopting the above technical solutions, the AI ​​recognition unit performs brightness equalization, background correction, texture enhancement, noise reduction, distortion correction, segmentation, and stitching on the acquired images, and outputs bounding boxes or segmented contours, lengths, widths, areas, and two-dimensional coordinates. At the same time, uncertain defects are marked for verification, which reduces interference factors such as fabric texture, shadows, color differences, and vibrations. The geometric information and spatial location of defects are fully structured, and low-confidence results are separated from high-confidence results. This improves the accuracy and reliability of defect data entering the big data analysis unit and reduces the interference caused by false detections or missed detections to subsequent dyeing factory problem analysis.

[0010] Optionally, the statistical results generated by the database and big data analysis unit include defect heatmaps, trend charts, dyeing plant rankings, abnormal fluctuations, and quality scores; the automatic identification of dyeing and finishing process problems based on the spatial distribution pattern of defects on multiple batches of fabric includes: when color spots, watermarks, or weft skew are concentrated in consecutive batches from the same dyeing plant, it is identified as an abnormality in the dyeing and finishing process; when defects are concentrated on the fabric edge or at a fixed width coordinate, it is identified as related to the fabric guide, tension, width control, or camera or light source calibration.

[0011] By adopting the above technical solutions, the database and big data analysis unit generate defect heat maps, trend maps, dyeing plant rankings, abnormal fluctuations, and quality scores. The spatial distribution pattern recognition rules are clarified as follows: when color spots, watermarks, or weft skew are concentrated in consecutive batches, it is identified as an abnormality in the dyeing and finishing process; when defects are concentrated on the fabric edge or at a fixed width coordinate, it is identified as related to fabric guiding, tension, width control, or camera light source calibration. This makes the judgment rules for dyeing and finishing process problems clear and executable. The analysis results are presented in the form of visual indicators, allowing customers and managers to intuitively see the differences and abnormal fluctuations between dyeing plants, batches, and machines, realizing the transformation of dyeing plant problems from invisible to observable and locatable.

[0012] Optionally, the industry knowledge base in the solution recommendation unit includes at least one of the following mappings: color spots or color differences correspond to investigation of dye liquor circulation, temperature control, pH value, auxiliary agent addition sequence, and setting conditions; oil stains correspond to investigation of guide rollers, bearings, lubrication points, and transportation contamination; holes or snagging correspond to investigation of tension, guide roller burrs, needle plates, and fabric friction points; creases or weft skew correspond to investigation of tension control, overfeeding, setting width, fabric speed, and rolling mill pressure; the priority sorting is determined based on the location of defect concentration, frequency of occurrence, and historical cases.

[0013] By adopting the above technical solutions, the recommended results are targeted inspection lists that match specific defect types and spatial distribution characteristics. This can directly transform AI recognition results into on-site actionable guidelines, reducing the time cost for service personnel to analyze images and inquire about processes one by one.

[0014] Optionally, it also includes a mini-program display and service unit, which displays real-time fabric inspection progress, defect list, defect map, dyeing plant problem dashboard, batch comparison, AI recommended solutions, report download, QR code traceability, rectification work order, service personnel comments, and re-inspection results; customers, dyeing plants, quality inspectors, and service personnel can view different data according to their permissions.

[0015] By adopting the above technical solutions, the mini-program display and service unit integrates real-time fabric inspection progress, defect list, defect map, dyeing plant problem dashboard, batch comparison, AI recommended solutions, report download, QR code traceability, rectification work orders, service personnel comments, and re-inspection results into the mobile terminal. The data is displayed according to different role permissions, enabling customers, dyeing plants, quality inspectors, and service personnel to view and communicate in real time on the same data interface. This eliminates the delays caused by the original report file transmission and repeated communication, and extends the visualization of dyeing plant quality problems from static reports on PCs to real-time interactive scenarios on mobile terminals, improving cross-regional collaboration efficiency.

[0016] Optionally, the mini-program display and service unit is also used to receive the reasons, handling measures, customer feedback and re-verification results confirmed by the service personnel, and write the reasons, handling measures, customer feedback and re-verification results back to the industry knowledge base and model training dataset.

[0017] By adopting the above technical solution, the service unit can receive supplementary confirmations from service personnel, including reasons, handling measures, customer feedback, and re-verification results, through a mini-program. These service results are then written back to the industry knowledge base and model training dataset. This allows the judgments and conclusions formed by service personnel during actual investigation and handling to be digitized and automatically fed back to the system. This provides semantic information and samples from real business scenarios for updating the knowledge base and training AI models, forming a business closed loop from detection to service and then to data accumulation. This enables the system's analytical capabilities to continuously improve with the increase in the number of services.

[0018] Optionally, it also includes a model iteration unit, which automatically adds manually verified defect images and service conclusions to the training sample pool, balances the samples according to defect category, fabric type, color and dyeing factory, and trains or fine-tunes the AI ​​model periodically.

[0019] By adopting the above technical solution, the defect images and service conclusions confirmed by humans are automatically added to the training sample pool through the model iteration unit. The samples are balanced according to defect type, fabric type, color and dyeing factory. The AI ​​model is trained or fine-tuned regularly, so that the differences in defect appearance caused by different fabric types, colors and dyeing factory processes are balanced in the training data. This reduces the model overfitting or recognition bias caused by uneven sample distribution. At the same time, the human confirmation results are fed back through image and semantic dual channels, which improves the recognition stability and generalization ability of the AI ​​model under changing production conditions and realizes the continuous iteration of AI recognition ability.

[0020] Optionally, it also includes a record creation and scoring report unit, which is used to enter or scan the order number, customer, dyeing plant, dyeing batch number, roll number, fabric type, color, width, target length, inspection standards and deduction rules before fabric inspection; and convert the defect length, area, quantity, meter position and severity into deductions, grades and release or rework suggestions according to a four-point system, a ten-point system or customer-defined rules, and generate an AIQC inspection report with a QR code.

[0021] By adopting the above technical solutions, the testing process of each roll of fabric is associated with a specific business context, the deduction and grade calculation are supported by unified customer standards, and the report can be traced and verified through QR codes. This solves the problems of inconsistent manual recording standards and the inability to trace reports, and improves the standardization and credibility of the test results.

[0022] The second objective of this application is to provide an AI-powered intelligent fabric inspection and comprehensive analysis and control system, comprising: a system for order creation and parameter input, whereby order number, customer, dyeing plant, dyeing vat number, roll number, fabric type, color, width, target length, inspection standards, and deduction rules are entered or obtained by scanning a code before fabric inspection; a synchronous acquisition step, where after the fabric inspection machine is started, the encoder outputs the meter position and speed in real time, and the camera synchronously acquires fabric images according to the fabric running distance, with each image bound to the current meter position, width coordinates, time, and machine information; an image preprocessing step, where the acquired images undergo brightness equalization, background correction, texture enhancement, noise reduction, distortion correction, segmentation, and stitching; and an AI recognition and classification step, where the AI ​​model identifies defects in each image block and outputs the defect category, confidence level, bounding box or segmentation contour, length, width, area, and other parameters. The meter and width coordinates on the entire roll of fabric are marked as pending verification for uncertain defects; the scoring and report generation steps automatically calculate the number of defects, deductions, deductions per 100 square meters, grade, and release recommendations according to customer standards, generating an AIQC inspection report with a QR code; the dyeing plant problem analysis step uses a big data analysis unit to statistically analyze defect distribution by dyeing plant, machine, batch, color, and process dimensions. When color spots, watermarks, or weft skew are concentrated in consecutive batches from the same dyeing plant, it is automatically identified as an abnormality in the dyeing and finishing process. When defects are concentrated on the fabric edge or fixed width coordinates, it is determined to be related to the fabric guide, tension, width control, or camera or light source calibration; the solution recommendation step uses a solution recommendation unit to generate a checklist, a ranking of possible causes, and handling suggestions based on defect type, location distribution, frequency, historical cases, and industry knowledge base. Attached Figure Description

[0023] Figure 1 This is a system block diagram of the AI-powered intelligent textile dyeing and inspection comprehensive analysis and control system of this application.

[0024] Figure 2 This is a flowchart illustrating the AI-powered intelligent fabric inspection and control method for this application. Detailed Implementation

[0025] To make the objectives, technical solutions, and advantages of this application clearer, the technical solutions in the embodiments of this application will be clearly and completely described below in conjunction with the embodiments of this application. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the scope of protection of this application.

[0026] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.

[0027] In the description of this application, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of this application, unless otherwise stated, "a plurality of" means two or more.

[0028] The technical problem addressed in this application stems from the actual situation in the fabric inspection process of the textile printing and dyeing industry. In the usual fabric inspection process, the fabric will continuously pass through unwinding, guiding, lighting and rewinding mechanisms. Personnel will conduct visual inspection in front of the fabric, or machine vision equipment will collect images of the fabric surface and output inspection reports. If the system only stops at the level of identifying defects and generating reports, then the inspection results themselves cannot indicate whether there are systemic problems in the production process or equipment status of the dyeing plant. Following this direction, this application considers treating each defect generated during the fabric inspection process as structured data carrying two-dimensional physical location and business context: In the acquisition stage, the images captured by the camera are synchronized with the meter signals given by the encoder and associated with the fabric width coordinates; in the identification stage, the defect category, confidence level, size, and two-dimensional coordinates are output; in the analysis stage, data from the same dyeing plant, the same machine, the same batch, or even multiple batches are aggregated, and the spatial distribution pattern of defects is used to determine whether there are abnormalities in the dyeing and finishing process; further, the solution recommendation unit matches troubleshooting items and handling suggestions based on the identified process problems, and connects the analysis results, recommended solutions, and subsequent service processes through a mini-program. In this way, the detection data first comes from the machine, then is analyzed to form problem judgments, and finally transformed into actionable troubleshooting guidelines, with the service results returning to the knowledge base and training data.

[0029] like Figure 1 As shown in the figure, this application provides an AI intelligent textile inspection and control system. The system includes an image acquisition unit, an AI recognition unit, a database and big data analysis unit, and a solution recommendation unit. The system can be integrated into a new textile inspection machine or used to modify an existing textile inspection machine.

[0030] It can be understood that the mechanism of the cloth inspecting machine may include an unwinding mechanism, cloth guide rollers, a tension control mechanism, an illumination box, a winding mechanism, a PLC controller, an encoder (also called a meter counter) and an operation console; wherein the encoder is connected to the cloth guide roller or the main transmission shaft, and is used to obtain the running speed of the cloth, the current meter position and the trigger signal. After being unwound by the unwinding mechanism, the cloth passes through the cloth guide rollers and the illumination box, and is wound up by the winding mechanism. The PLC controller is used to control the speed, start and stop of the cloth inspecting machine. It should be noted that the above-mentioned mechanical components of the cloth inspecting machine can adopt the structure of a common cloth inspecting machine in the art, and the main improvement of the present application lies in using it as a data acquisition front end to connect with the identification, analysis and recommendation links at the back end.

[0031] The image acquisition unit is arranged above or on both sides of the observation area of the cloth inspecting machine, and includes an industrial camera, a line-scan camera or a high-definition camera, and is equipped with a strip light source, a backlight source, a polarized light source or a multi-angle light source. The camera is mounted on an adjustable support to make its field of view cover the effective width of the cloth surface, and the camera collects images at a fixed distance or a fixed time through an encoder or a PLC signal; wherein collecting at a fixed distance refers to triggering the camera to shoot according to the cloth running distance signal output by the encoder, and collecting at a fixed time refers to triggering the camera to shoot according to a preset time interval, and which triggering mode is specifically adopted can be determined by system parameter configuration.

[0032] The AI identification unit includes an industrial personal computer, a GPU server or an edge AI box. The AI identification unit receives the camera image and the meter position signal, preprocesses the image, and then calls a defect detection model to identify defects.

[0033] It can be understood that the embodiment of the present application may further include a defect knowledge base and scoring unit preset with common defect categories and codes. The common defect categories include, for example, stain, oil stain, hole, crease, paper crease, broken warp, broken yarn, line mark, white spot, thick yarn, skewed weft, water mark, water stamp, water splash, clip mark, nep, color thread, needle mark, press wheel mark, loose crease, blanket mark, anti-dyeing, transverse damage, etc.; the scoring unit converts the length, area, quantity, meter position and severity of the defects into deduction scores, grades, and pass or rework suggestions according to the four-point system, ten-point system or customer-defined rules.

[0034] The database stores order number, customer, dyeing mill, machine station, vat number, roll number, batch, cloth type, color, process parameters, collected images, identification results, reports, rectification records and reinspection results. The big data analysis unit performs statistics by dyeing mill, machine station, batch, color, time period, defect category and spatial position, and generates a defect heat map, trend chart, ranking, abnormal fluctuation and quality score. The big data analysis unit is used to aggregate the detection results of a single roll of cloth to the dimensions of the dyeing mill and the batch, and make a process-level judgment in combination with the spatial position distribution of the defects, and the specific judgment manner will be described in detail in subsequent embodiments.

[0035] The solution recommendation unit establishes an industry knowledge base of "defect category - possible causes - investigation items - suggested measures - service cases" and generates suggestions by combining AI big data models or rule engines. The solution recommendation unit prioritizes the defects based on their concentrated location, frequency of occurrence and historical cases, and outputs an investigation list and handling suggestions.

[0036] In terms of connectivity, industrial cameras connect to the AI ​​recognition unit via acquisition cards, Ethernet ports, or USB interfaces. Encoders and PLCs connect to the AI ​​recognition unit via I / O, RS485, Ethernet, or industrial buses. The AI ​​recognition unit uploads detection results through a local database or cloud API.

[0037] Understandably, during the operation of the fabric inspection machine, the fabric continuously passes through the observation area covered by the image acquisition unit. The encoder is connected to the guide roller or main drive shaft, and outputs meter and speed signals in real time as the fabric moves. The camera in the image acquisition unit synchronously acquires fabric images according to the fabric's running distance based on the signals output by the encoder, or it can acquire images at fixed intervals based on PLC signals. In this way, each acquired fabric image is bound to the current meter position, width coordinates, acquisition time, and machine information.

[0038] Among them, the meter position refers to the position along the length of the fabric, usually expressed as the length value of the fabric starting from the beginning. The width coordinate refers to the position along the width of the fabric, also known as the width coordinate. The width coordinate means that each identified defect has not only meter position information along the length of the fabric, but also position information along the width of the fabric.

[0039] The camera continuously captures images of the fabric under a stable light source. This light source can be a strip light source, backlight source, polarized light source, or multi-angle light source. Its function is to provide relatively stable lighting conditions as the fabric passes through the observation area, reducing the impact of ambient light and fabric reflections on image quality. The captured fabric images, along with synchronously acquired meter coordinates, width coordinates, time, and machine information, are sent to the AI ​​recognition unit via a capture card, network port, or USB interface.

[0040] In one embodiment, the camera mounting position and bracket in the image acquisition unit are set to an adjustable structure, so that the camera's field of view can cover the effective width of the fabric. The effective width can be determined according to the width parameters entered each time an order is placed. In this way, during the subsequent recognition process, the system can map the position of the defect in the image to the width direction coordinates based on the effective width. It should be noted that this application does not limit the specific calibration algorithm for mapping the image pixel position to the width coordinates; any conventional calibration method that can achieve this mapping is applicable. This application focuses on how the two-dimensional defect data formed after mapping the position of the identified defect to the length direction and width direction of the entire roll of fabric can be used in subsequent analysis.

[0041] After receiving the camera image and position signal, the AI ​​recognition unit first preprocesses the image to reduce the impact of fabric texture, shadows, color difference, and vibration on recognition. The preprocessing process includes brightness equalization, background correction, texture enhancement, denoising, distortion correction, segmentation, and stitching of the acquired image. It may also include cropping and standardization. Brightness equalization reduces the impact of uneven lighting on the image; background correction weakens interference from the fabric texture background; texture enhancement highlights abnormal texture changes on the fabric surface; denoising removes random noise from the image; distortion correction corrects geometric distortion caused by the camera lens; and segmentation and stitching divide the large fabric image into image blocks suitable for model processing and stitch them together as needed.

[0042] After preprocessing, the AI ​​recognition unit calls the defect detection model to identify each image block. The AI ​​model may include a classification model, an object detection model, a semantic or instance segmentation model, and an anomaly detection model. In actual use, different model combinations can be selected according to different fabric types, colors, or detection requirements. This application does not limit which type of model to use or the specific structure of the model.

[0043] After the AI ​​model identifies defects in each image block, it outputs the defect category, confidence level, bounding box or segmentation contour, length, width, area, and the meter position and width coordinates on the entire roll of fabric. In addition, the AI ​​recognition unit can also output the original image and the labeled image. The labeled image is an image that marks the location of the defect or the segmentation contour on the original image, which is convenient for subsequent manual review and report display.

[0044] For the results output by the AI ​​model, the system marks uncertain defects as pending review. Specifically, when the AI ​​model's recognition result for a certain defect is not reliable enough, or when the matching relationship between the category output by the model and common defect categories is unclear, the system will mark the defect as pending review.

[0045] Before the fabric inspection begins, the system can provide a form creation interface. Operators can enter or obtain information such as order number, customer, dyeing factory, dyeing batch number, roll number, fabric type, color, width, target length, inspection standards, and deduction rules in the form creation interface. These parameters constitute the business context of this fabric inspection task. Among them, the inspection standards and deduction rules are used for subsequent scoring, and can be based on a four-point system, a ten-point system, or customer-defined rules. These parameters, after being entered, are associated with the single roll of fabric for this inspection and are consistently referenced in subsequent identification results, analysis results, and reports.

[0046] After the AI ​​recognition unit outputs defect information, the defect knowledge base and scoring unit convert the defect length, area, quantity, meter value, and severity into deductions according to the scoring rules determined when the order was created. They further derive deductions per 100 square meters, fabric grade, and release or rework recommendations. The specific calculation method of the scoring unit corresponds to the four-point system, ten-point system, or customer-defined rules used. This application does not limit the specific scoring formula; it only needs to be able to automatically calculate the number of defects, deductions, and grades according to customer standards.

[0047] After scoring is completed, the system automatically generates an AIQC inspection report. The AIQC inspection report includes a QR code and contains the inspection order number, inspection machine, length, width, total number of defects, deductions, fabric grade, and defect distribution map. It can also include a defect list, defect pictures, scatter plot, bar chart, and pie chart. The QR code is used for report traceability. For example, the corresponding inspection records and re-inspection records can be retrieved later by scanning the code through a mini-program.

[0048] It should be noted that the order creation and scoring report functions in this embodiment can be implemented as independent functional modules, or they can be integrated into the database and big data analysis unit. This application does not limit the physical deployment relationship between the scoring unit, the report generation unit, and the production management database, as long as the above-mentioned parameter input, deduction calculation, and report output can be achieved.

[0049] The database stores raw data including order number, customer, dyeing plant, machine, vat number, roll number, batch, fabric type, color, process parameters, captured images, recognition results, reports, rectification records, and re-inspection results. After the inspection of multiple rolls of fabric is completed, the system accumulates a large amount of structured defect data, encompassing multiple dyeing plants, machines, batches, fabric types, and colors. Each defect record in this data includes defect category, confidence level, bounding box or segmentation outline, length, width, area, meter coordinates (i.e., the defect's coordinates along the fabric's length), width coordinates (i.e., the defect's coordinates along the fabric's width), and business fields such as the batch, fabric type, and color.

[0050] The big data analytics unit aggregates and statistically analyzes data by dyeing plant, machine, batch, color, time period, defect category, and spatial location, including meter and fabric width coordinates. After aggregation and statistical analysis, the big data analytics unit generates a defect heatmap, trend chart, dyeing plant ranking, abnormal fluctuations, and quality score. The defect heatmap displays the distribution density of defects along the fabric length and width. The trend chart shows the changes of a certain type of defect or a particular dyeing plant over a period of time. The dyeing plant ranking allows for horizontal comparison of the quality performance between different dyeing plants. Abnormal fluctuations mark batches or time periods that deviate significantly from the historical normal range. The quality score provides a comprehensive quantitative assessment based on the above statistical results.

[0051] Based on aggregated statistics, the big data analysis unit automatically identifies dyeing and finishing process problems according to the spatial distribution pattern of defects on multiple batches of fabric. The spatial distribution pattern refers to the clustering characteristics of defects in the meter and width directions after the defect data of multiple rolls of fabric belonging to the same dyeing plant, the same machine or the same batch are summarized.

[0052] One identification rule is: when color spots, watermarks, or weft skew appear in concentrated batches in the same dyeing factory, the system automatically identifies it as an abnormality in the dyeing and finishing process. The so-called concentrated appearance in consecutive batches means that the same or the same type of defect appears repeatedly in multiple adjacent batches and shows a relatively concentrated distribution in spatial location. In this case, the system's identification result is: there may be an abnormality in the dyeing and finishing process of the dyeing factory, and the corresponding process needs to be further investigated.

[0053] Another identification rule is: when defects are concentrated at the fabric edge or at a fixed width coordinate, the system determines that they are related to the fabric guide, tension, width control, or camera or light source calibration. The "fabric edge" or "fixed width coordinate" refers to the fact that defects in multiple rolls of fabric maintain a consistent position or are concentrated within the same width range in the fabric width direction, rather than being randomly scattered with the meter length. This concentration at a fixed width position means that the source of the defects is more likely a point of action related to the equipment position, rather than a random raw material or operational problem. It should be noted that when the system gives the judgment result of "related to the fabric guide, tension, width control, or camera or light source calibration," it does not directly determine the final cause, but rather narrows the investigation to the range related to the above-mentioned aspects.

[0054] Beyond spatial distribution pattern recognition, the big data analysis unit can further determine whether defects exhibit periodic repetition, batch fluctuations, or historical recurrence. Periodic repetition refers to defects appearing repeatedly at certain intervals along the meter direction; this pattern is usually related to rotating parts or periodic movements. Batch fluctuations refer to significant fluctuations in the quantity or type distribution of defects among different rolls of fabric within the same batch. Historical recurrence refers to the repeated occurrence of the same type of problem in historical records. These judgment conditions can be used individually or in combination with the aforementioned two recognition rules to improve the reliability of the recognition results.

[0055] It is understandable that the specific judgment methods for concepts such as "continuous batches," "fixed width," and "concentration" involved in the above judgment conditions can be preset by the system based on the historical data distribution in implementation. This application does not limit specific threshold values. The key point is that the system can summarize single-roll detection data into statistical views across rolls and batches, and use width coordinates to incorporate spatial distribution characteristics into the judgment criteria.

[0056] The solution recommendation unit has a pre-built industry knowledge base. In this embodiment, the structure of the industry knowledge base can be "defect category - possible causes - investigation items - suggested measures - service cases". The content in the industry knowledge base is compiled from the experience of the dyeing and finishing industry and is associated with the results identified by the AI ​​recognition unit and the big data analysis unit. The solution recommendation unit can generate suggestions by combining the AI ​​big model or the rule engine, or it can combine the retrieval results of the rule engine with the semantic generation capabilities of the AI ​​big model.

[0057] In the industry knowledge base, color variations or color differences correspond to checks on dye liquor circulation, temperature control, pH value, auxiliary agent addition sequence, and setting conditions; oil stains correspond to checks on guide rollers, bearings, lubrication points, and transportation contamination; holes or snagging correspond to checks on tension, guide roller burrs, needle plates, and fabric friction points; creases or weft skew correspond to checks on tension control, overfeeding, setting width, fabric speed, and rolling mill pressure. Each pair of "correspondences" in the above mapping relationship indicates that when the AI ​​recognition results and the statistical results of the big data analysis unit show that this type of defect has occurred, the system will prioritize recommending checks on the relevant links.

[0058] The solution recommendation unit prioritizes defects based on their location, frequency, and historical cases. For example, when the big data analysis unit identifies oil stains concentrated on the fabric edge or at a fixed width, the solution recommendation unit suggests prioritizing checks on the corresponding guide rollers, bearing lubrication points, fabric contact parts, and cleaning records. Similarly, when the big data analysis unit identifies a continuous increase in weft skew, the solution recommendation unit suggests prioritizing checks on the setting machine's width, overfeed, tension, and fabric speed matching. The recommended checklist corresponds to the spatial distribution pattern recognition results, enabling service personnel to check items sequentially from highest to lowest priority.

[0059] In one implementation, the solution recommendation unit generates suggestions when the big data analysis unit outputs spatial distribution pattern recognition results. In other words, the solution recommendation unit takes "defect type," "spatial distribution pattern," "occurrence frequency," and "historical cases" as input. Taking oil stains as an example, oil stains can come from transportation pollution or equipment oil leaks. Without spatial distribution information, the category of "oil stains" alone cannot distinguish the priority direction for investigation. However, when the analysis results further indicate that the oil stains are concentrated in a fixed width area, the system can prioritize the corresponding guide rollers, bearing lubrication points, and other location-related links, making the recommendation results more targeted.

[0060] In this embodiment, a mini-program display and service unit may also be included, implemented in the form of a WeChat mini-program or a mobile terminal page. The mini-program is connected to the AI ​​analysis platform, reads reports, images, statistical charts and recommendations through an interface, and displays them on the mobile terminal interface. The displayed content includes real-time fabric inspection progress, defect list, defect map, dyeing factory problem dashboard, batch comparison, AI recommended solutions, report download, QR code traceability, rectification work order, service personnel comments and re-inspection results.

[0061] The defect map is constructed based on the meter and width coordinates of the fabric roll. After unfolding the entire roll into a planar view, each defect is mapped to its corresponding position in the view according to its two-dimensional coordinates. In this way, customers, dyeing plants, quality inspectors and service personnel can intuitively see the distribution of defects on the defect map, and understand quality issues by combining the dyeing plant's problem board and batch comparison.

[0062] Customers, dyeing plants, quality inspectors, and service personnel can view different data according to their permissions. After logging into the mini-program, the system displays the corresponding data range and function access based on their permissions. For example, customers can view reports and suggestions, while service personnel can view all test data and AI-recommended solutions and initiate rectification work orders. Different roles can see different interface content.

[0063] After viewing defect images, reports, and recommendations in the mini-program, service personnel combine their on-site experience to provide on-site or remote services. They can then supplement the mini-program with confirmations of causes, corrective measures, customer feedback, and re-inspection results. The mini-program writes these service results back to the industry knowledge base and model training dataset. Thus, the conclusions reached by service personnel during actual service are digitally fed back into the system, forming a closed loop of "detection—analysis—recommendation—service—re-inspection—model iteration."

[0064] In terms of the mini-program interface, service personnel can click on a defect in the defect map to view the original image, labeled image, identification category, and recommended suggestions for that defect. After on-site verification, they can enter the confirmed cause and corrective measures. Once the re-inspection results are entered, the corresponding rectification work order status is updated synchronously. All of the above interface interaction methods can be implemented through the data interface provided by the mini-program.

[0065] In this embodiment, the system may further include a model iteration unit, which is used to automatically add manually confirmed defect images and service conclusions to the training sample pool. Here, manual confirmation includes both the manual confirmation results of defects "to be reviewed" in the AI ​​recognition stage and the processing results and re-inspection results supplemented by service personnel in the service closed-loop stage. The service conclusions include the reasons for the supplementary confirmation by service personnel, the handling measures, customer feedback, and the re-inspection results.

[0066] Before training, the model iteration unit balances the samples in the training sample pool according to defect type, fabric type, color, and dyeing factory. This ensures that the number of samples of different defect types, fabric types, colors, and dyeing factories remains relatively balanced, so that the AI ​​model obtained through subsequent training or fine-tuning can maintain stable recognition capabilities under diverse production conditions.

[0067] The model iteration unit periodically trains or fine-tunes the AI ​​model. After training or fine-tuning, the AI ​​model is redeployed to the AI ​​recognition unit for subsequent verification and recognition. Through this process of continuous feedback of results and service conclusions by human verification, followed by sample balancing and training fine-tuning, the recognition accuracy and suggestion matching degree of the AI ​​model can be continuously improved with the accumulation of business experience.

[0068] It should be noted that the sample source, sample balancing dimension, and training triggering method of the model iteration unit can all be configured according to the specific implementation scenario. This application does not limit the specific type of training algorithm, nor does it limit the number of samples used in each training session, as long as the process of "manually confirmed defective images and service conclusions enter the training sample pool - sample balancing is performed according to defect category, fabric type, color, and dyeing factory - periodic training or fine-tuning of the AI ​​model" can be achieved.

[0069] like Figure 2 As shown in the embodiments of this application, an AI-powered intelligent fabric inspection and comprehensive analysis and control method is also disclosed. The method includes steps such as order creation and parameter input, synchronous data acquisition, image preprocessing, AI recognition and classification, scoring and report generation, dyeing plant problem analysis, and recommended solutions.

[0070] In the order creation and parameter entry steps, operators enter or scan codes to obtain the order number, customer, dyeing plant, dyeing vat number, roll number, fabric type, color, width, target length, inspection standards, and deduction rules before fabric inspection. These parameters are uniformly referenced as business context in subsequent steps.

[0071] In the synchronous acquisition step, after the fabric inspection machine is started, the encoder outputs the meter position and speed in real time. The camera synchronously acquires images of the fabric surface according to the fabric running distance, or it can acquire images at fixed times. Each image is bound to the current meter position, width coordinates, time and machine information, and the shooting is completed under stable light source conditions.

[0072] In the image preprocessing step, the acquired image is subjected to brightness equalization, background correction, texture enhancement, noise reduction, distortion correction, segmentation and stitching, and may also include cropping and standardization. The preprocessed image is then sent to the AI ​​recognition and classification step.

[0073] In the AI ​​recognition and classification step, the AI ​​model identifies defects in each image block and outputs the defect category, confidence level, bounding box or segmentation contour, length, width, area, and meter and width coordinates on the whole roll of fabric. For uncertain defects, the system marks them as pending review and allows manual confirmation in the mini-program or management terminal.

[0074] In the scoring and report generation process, the system automatically calculates the number of defects, deductions, deductions per 100 square meters, grade, and release recommendations according to customer standards, generates an AIQC inspection report with a QR code, and outputs a defect list, defect images, scatter plots, bar charts, pie charts, and fabric grades.

[0075] In the dyeing plant problem analysis step, the big data analysis unit statistically analyzes the distribution of defects according to the dimensions of dyeing plant, machine, batch, color and process. When color spots, watermarks or weft skew are concentrated in consecutive batches in the same dyeing plant, the system automatically identifies it as an abnormality in the dyeing and finishing process. When defects are concentrated on the fabric edge or fixed width coordinates, the system judges that it is related to the guide fabric, tension, width control or camera or light source calibration.

[0076] In the solution recommendation step, the solution recommendation unit generates a checklist, a ranking of possible causes, and handling suggestions based on defect type, location distribution, frequency, historical cases, and industry knowledge base, and displays them in the mini-program.

[0077] It is understood that in other embodiments of this application, there may also be a course registration service closed loop and model iteration steps.

[0078] The checklist, possible cause ranking, and handling suggestions generated in the recommended solution step can be displayed in the mini-program. In the service loop step, service personnel can supplement and confirm the causes, handling measures, customer feedback, and re-verification results in the mini-program. The system writes the service results back to the knowledge base, forming a closed loop of detection, analysis, recommendation, service, re-verification, and model iteration.

[0079] During the model iteration process, defect images and service conclusions confirmed by humans are automatically entered into the training sample pool. The samples are balanced according to defect type, fabric type, color and dyeing factory, and the AI ​​model is trained or fine-tuned periodically.

[0080] In the above embodiments, the descriptions of each embodiment have their own emphasis. Parts not described in detail in a particular embodiment can be referred to in the relevant descriptions of other embodiments. It is understood that related features in the above systems and methods can be referenced interchangeably. Furthermore, the terms "first," "second," etc., in the above embodiments are used to distinguish between different embodiments or features, and do not indicate superiority or inferiority of the embodiments or features, or their order of priority.

[0081] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product implemented on one or more computer-usable storage media containing computer-usable program code.

[0082] The functional units in the above embodiments can be implemented in hardware, in software modules executed by a processor, or in a combination of hardware and software. The software modules can be located in any other form of storage medium, such as random access memory, main memory, read-only memory, electrically programmable ROM, electrically erasable programmable ROM, registers, hard disks, removable disks, or CD-ROMs.

[0083] Obviously, those skilled in the art can make various modifications and variations to this application without departing from the spirit and scope of this application. Therefore, if such modifications and variations fall within the scope of the claims of this application and their equivalents, this application also intends to include such modifications and variations.

[0084] The above description is merely a preferred embodiment of this application and is not intended to limit this application. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.

Claims

1. An AI-powered intelligent fabric inspection and control system, characterized in that, include: The image acquisition unit is used to synchronously acquire images of the fabric surface using an industrial camera and an encoder, thereby obtaining an image of the fabric surface bound to the current meter position and width coordinates. The AI ​​recognition unit is used to preprocess the fabric image and identify defects, and output the defect type, confidence level, size, and the meter position and width coordinates of the defect on the whole roll of fabric; The database and big data analysis unit is used to store order numbers, customers, dyeing plants, machines, vats, roll numbers, fabric types, colors, process parameters, and identification results for multiple batches. It also aggregates and statistically analyzes data by dyeing plant, machine, batch, color, time period, defect type, and spatial location, and automatically identifies dyeing and finishing process problems based on the spatial distribution pattern of defects on fabrics in multiple batches. The solution recommendation unit is used to match possible causes, investigation items and handling suggestions from a preset industry knowledge base of defect categories, possible causes, investigation items, suggested measures and service cases based on the identification results of the spatial distribution pattern, and output them in order of priority.

2. The AI ​​intelligent printing and dyeing fabric inspection comprehensive analysis and control system according to claim 1, characterized in that, The image acquisition unit includes an encoder connected to the guide roller or the main drive shaft; the industrial camera synchronously acquires fabric images according to the fabric running distance based on the signal output by the encoder; each image is bound to the current meter position, width coordinates, acquisition time, and machine information.

3. The AI ​​intelligent printing and dyeing fabric inspection comprehensive analysis and control system according to claim 1, characterized in that, The AI ​​recognition unit performs preprocessing on the acquired images, including brightness equalization, background correction, texture enhancement, noise reduction, distortion correction, segmentation, and stitching, and calls the defect detection model to identify defects. The output defect information also includes an outer bounding box or segmentation contour used to surround and mark the location of the defect, and the dimensions include length, width, and area. Uncertain defects are marked as pending review.

4. The AI ​​intelligent printing and dyeing fabric inspection comprehensive analysis and control system according to claim 1, characterized in that, The statistical results generated by the database and big data analysis unit include defect heatmaps, trend charts, dyeing plant rankings, abnormal fluctuations, and quality scores. The automatic identification of dyeing and finishing process problems based on the spatial distribution pattern of defects on multiple batches of fabric includes: when color spots, watermarks, or weft skew are concentrated in consecutive batches from the same dyeing plant, it is identified as an abnormality in the dyeing and finishing process; when defects are concentrated on the fabric edge or at a fixed width coordinate, it is identified as related to the fabric guide, tension, width control, or camera or light source calibration.

5. The AI ​​intelligent printing and dyeing fabric inspection comprehensive analysis and control system according to claim 1, characterized in that, The industry knowledge base in the solution recommendation unit includes at least one of the following mappings: color spots or color differences correspond to investigation of dye liquor circulation, temperature control, pH value, auxiliary agent addition sequence, and setting conditions; oil stains correspond to investigation of guide rollers, bearings, lubrication points, and transportation contamination; holes or snagging correspond to investigation of tension, guide roller burrs, needle plates, and fabric friction points; creases or weft skew correspond to investigation of tension control, overfeeding, setting width, fabric speed, and rolling mill pressure; the priority sorting is determined based on the location of defect concentration, frequency of occurrence, and historical cases.

6. The AI-powered intelligent fabric inspection and control system according to claim 1, characterized in that, It also includes a mini-program display and service unit, which displays real-time fabric inspection progress, defect list, defect map, dyeing plant problem dashboard, batch comparison, AI recommended solutions, report download, QR code traceability, rectification work order, service personnel comments, and re-inspection results; customers, dyeing plants, quality inspectors, and service personnel can view different data according to their permissions.

7. The AI ​​intelligent printing and dyeing fabric inspection comprehensive analysis and control system according to claim 6, characterized in that, The mini-program display and service unit is also used to receive supplementary confirmation from service personnel regarding the reasons, handling measures, customer feedback, and re-verification results, and to write the reasons, handling measures, customer feedback, and re-verification results back to the industry knowledge base and model training dataset.

8. The AI ​​intelligent printing and dyeing fabric inspection comprehensive analysis and control system according to claim 1, characterized in that, It also includes a model iteration unit, which automatically adds manually confirmed defect images and service conclusions to the training sample pool, balances the samples according to defect type, fabric type, color and dyeing factory, and trains or fine-tunes the AI ​​model periodically.

9. The AI-powered intelligent fabric inspection and control system according to claim 1, characterized in that, It also includes a record creation and scoring report unit, which is used to enter or scan codes to obtain the order number, customer, dyeing plant, dyeing batch number, roll number, fabric type, color, width, target length, inspection standards and deduction rules before fabric inspection; and convert the defect length, area, quantity, meter position and severity into deductions, grades and release or rework suggestions according to the four-point system, ten-point system or customer-defined rules, and generate an AIQC inspection report with a QR code.

10. An AI-powered intelligent fabric inspection and control method, based on the AI-powered intelligent fabric inspection and control system according to any one of claims 1 to 9, characterized in that, include: The order creation and parameter entry steps involve entering or scanning the code to obtain the order number, customer, dyeing plant, dyeing batch number, roll number, fabric type, color, width, target length, inspection standards, and deduction rules before fabric inspection. In the synchronous acquisition process, after starting the fabric inspection machine, the encoder outputs the meter position and speed in real time, and the camera synchronously acquires fabric images according to the fabric running distance. Each image is bound to the current meter position, width coordinates, time and machine information. The image preprocessing steps include brightness equalization, background correction, texture enhancement, noise reduction, distortion correction, segmentation, and stitching of the acquired images. The AI ​​recognition and classification steps involve the AI ​​model identifying defects in each image block and outputting the defect category, confidence level, bounding box or segmentation contour, length, width, area, and meter and width coordinates on the entire roll of fabric. Uncertain defects are marked as pending review. The scoring and report generation process automatically calculates the number of defects, deductions, deductions per 100 square meters, grade, and release recommendations according to customer standards, generating an AIQC inspection report with a QR code. The dyeing plant problem analysis steps involve a big data analysis unit that statistically analyzes the distribution of defects by dyeing plant, machine, batch, color, and process. When color spots, watermarks, or weft skew are concentrated in consecutive batches from the same dyeing plant, it is automatically identified as an abnormality in the dyeing and finishing process. When defects are concentrated on the fabric edge or fixed width coordinates, it is determined to be related to the fabric guide, tension, width control, or camera or light source calibration. The recommended solution steps are generated by the solution recommendation unit based on defect type, location distribution, frequency, historical cases and industry knowledge base, as well as troubleshooting list, possible cause ranking and handling suggestions.

Images