A template machine sewing process flaw identification method and system thereof

Abstract

The application discloses a kind of flaw identification method and system in template machine sewing process, belong to garment manufacturing production technical field.The core to be solved technical problem is that when existing template machine high-speed nonlinear sewing, quality detection and production rhythm are difficult to take into account, and detection data cannot support the closed-loop management needs of downstream automated sorting.With spindle encoder signal as space-time reference, needle plate area dynamic image stream and spindle mechanical data stream are synchronously collected, and are mapped to unified sewing trajectory coordinate system;sewing quality state is determined by multi-modal feature fusion;when abnormal, the vector deviation coordinates of defect points relative to U-shaped template are calculated, and the pose compensation is performed by follow-up marking module, to complete accurate physical marking under non-stop state;at the same time, the vector deviation coordinates are written into the RFID electronic tag of U-shaped template, for downstream six-axis robot to directly read and realize "zero search" automatic sorting and repair, to realize production whole-process digital closed-loop control.

Description

Technical Field

[0001] This invention relates to the field of garment manufacturing technology, and in particular to a method and system for identifying defects during the sewing process of a template machine. Background Technology

[0002] Template sewing machines, as core equipment in modern intelligent textile and apparel manufacturing, use specialized U-shaped template clamps to fix fabric and perform high-speed, complex nonlinear trajectory movements in the XY plane, greatly improving the production efficiency and consistency of processes such as patch pockets, plackets, and decorative stitches. However, during high-speed mass production, quality defects such as skipped stitches, broken threads, abnormal stitch density, and backstitches still occur frequently due to factors such as equipment vibration, thread tension fluctuations, and fabric deformation. Current quality control methods mainly rely on post-production manual sampling or manual shutdown inspection after anomalies are discovered. This "separation of production and inspection" model has significant lag, and once the machine is stopped for inspection, it directly interrupts the production cycle, severely sacrificing the high-speed production capacity advantage of template sewing machines. In addition, existing automated inspection solutions are mostly single-point applications, failing to establish a digital link from the sewing end to the downstream sorting end, resulting in downstream robots being unable to obtain accurate defect coordinate information, making it difficult to achieve unmanned closed-loop management of the entire process.

[0003] Chinese patent CN117147582B discloses a fabric defect identification device and its control method. This solution identifies defects through an image acquisition mechanism and uses a preset inertial movement amount to control a marking mechanism to physically mark the defects after the machine stops or decelerates. However, this patent has significant technical limitations: First, its marking logic is designed based on a scenario of unidirectional linear fabric transport, which cannot adapt to the working conditions of template sewing machines performing complex curves, turns, and speed changes in the XY plane. If simply transplanted, its linear guide marking mechanism cannot track the rapidly changing sewing trajectory. Second, the execution of this solution depends on the stopping or inertial sliding of the fabric conveying mechanism, essentially remaining an "interrupted" detection method, unable to meet the cycle time requirements of continuous production. Third, this solution relies solely on visual recognition, making it difficult to detect hidden defects such as abnormal bottom thread tension. Furthermore, the chalk physical markings used can only be recognized by the human eye and cannot carry digital coordinate information and type codes, causing subsequent automated sorting robots to be unable to directly read and utilize this information for targeted grasping, resulting in a break in the production data chain at the marking stage. Summary of the Invention

[0004] The purpose of this invention is to address the technical challenge of existing template sewing machines in high-speed, non-linear trajectory operations, which cannot simultaneously achieve quality inspection and production cycle time. It provides a quality closed-loop control method and system based on digital twin mapping. By constructing a spatiotemporal coordinate system based on the sewing trajectory and utilizing follow-up pose compensation technology, real-time capture and non-contact precise marking of quality anomalies on the fabric surface and internally can be achieved without stopping the high-speed needle movement. This allows for full inspection while ensuring production efficiency.

[0005] The purpose of this invention is to solve the problem that a single visual inspection method cannot identify hidden sewing defects. By establishing a spatiotemporal synchronization mechanism driven by a spindle encoder, the visual flow data of the needle plate area is fused with the mechanical data flow such as tension and torque of the spindle in multiple dimensions. Multimodal data is used to determine visually invisible internal quality hazards such as loose bobbin thread and early signs of skipped stitches, which significantly improves the anomaly detection rate and judgment confidence in complex sewing processes.

[0006] The purpose of this invention is to overcome the data barrier that existing physical markers cannot be directly accessed by automated production lines. By converting the identified physical defects into digital vector coordinates relative to the template positioning reference and writing them into RFID tags that move with the workpiece, a digital twin file that can be directly parsed by downstream six-axis robots is constructed. This supports subsequent processes for "zero-search" automatic sorting and fixed-point rework based on coordinate data, realizing digital closed-loop management of the entire production process.

[0007] This invention proposes a defect identification method during the sewing process of a template machine. The method includes: synchronously triggering the acquisition of dynamic image streams of the needle plate area and mechanical data streams of the template machine spindle and sewing thread through the spindle encoder signal; mapping the dynamic image streams and mechanical data streams to a unified sewing trajectory coordinate system; and determining the sewing quality status based on the fusion result of the mapped multi-dimensional feature data. When the judgment result is abnormal, the vector deviation coordinates of the defect point relative to the positioning reference of the U-shaped template are calculated according to the current sewing trajectory position; based on the vector deviation coordinates and the real-time sewing speed, the follow-up marking module is controlled to perform pose compensation and generate a physical mark on the moving workpiece surface; the vector deviation coordinates are written into the electronic tag fixed to the U-shaped template through the radio frequency read / write node.

[0008] Preferably, this method uses the pulse signal of the spindle encoder as a spatiotemporal reference, triggers the visual acquisition unit to capture the region of interest image at a preset needle position phase point; simultaneously captures the spindle torque waveform and the surface tension waveform within the time window corresponding to the phase point; and binds the region of interest image and waveform data through a timestamp alignment mechanism to construct a multimodal sewing fingerprint corresponding to the current sewing trajectory coordinate point.

[0009] Preferably, this method extracts the line edge features and texture features of the dynamic image stream; extracts the tension peak stability features and torque fluctuation features of the mechanical data stream; inputs the above features into the anomaly scoring model for weighted calculation to obtain a comprehensive anomaly score; and compares the comprehensive anomaly score with an adaptive judgment threshold to output the judgment result.

[0010] Preferably, the method calculates the process complexity coefficient of the current sewing trajectory segment in real time; combines the distribution trend of historical quality data with the process complexity coefficient to dynamically adjust the judgment threshold; and automatically relaxes the threshold of visual feature weights and tightens the threshold of mechanical feature weights in the corners or speed change areas of the sewing trajectory.

[0011] As a preferred method, the method calculates the process complexity coefficient by linearly combining the root mean square curvature of the sewing path, the coefficient of variation of the sewing machine spindle speed, and the range of line tension fluctuation.

[0012] Preferably, the method obtains the absolute position of the sewing machine needle hole in the machine tool coordinate system at the moment of the abnormality; establishes a local coordinate system of the U-shaped template by reading the positioning feature points on the U-shaped template; and uses a coordinate transformation matrix to convert the absolute position in the machine tool coordinate system into a vector deviation value relative to the origin of the U-shaped template.

[0013] Preferably, the method calculates the displacement vector of the machine head movement within the time delay from the time of abnormality determination to the time of marking execution by a follow-up marking module integrated on the side of the extendable head structure of the template sewing machine and moving synchronously with the needle bar; and controls the galvanometer deflection angle of the follow-up marking module according to the displacement vector.

[0014] Preferably, after the sewing task is completed, the method uses a six-axis robot to read the vector deviation coordinates in the electronic tag; the six-axis robot directly drives the end effector to move to the physical location of the defect on the workpiece surface to grasp it based on the vector deviation coordinates.

[0015] Preferably, the follow-up marking module uses a laser thermochromic marking unit or a high-speed micro-volume inkjet unit, including a coded pattern corresponding to the anomaly type or a color-printing mark under a specific wavelength.

[0016] This invention proposes a defect identification system for the template sewing process. This system is applied to the aforementioned defect identification method for the template sewing process. The system includes: a spatiotemporal synchronization sensing module, whose signal triggering terminal is electrically connected to the main shaft encoder of the template sewing machine; a digital twin mapping engine receiving the output data of the spatiotemporal synchronization sensing module through a data bus and establishing a spatiotemporal reference mapping channel with the main shaft encoder; a follow-up marking module mechanically fixed to the side of the extendable head structure of the template sewing machine; a posture compensation controller bidirectionally connected to the main shaft encoder and the digital twin mapping engine; and a radio frequency data closed loop point communicatively coupled to the posture compensation controller.

[0017] The present invention has the following beneficial effects: 1. This invention achieves uninterrupted and precise marking on complex nonlinear sewing trajectories, resolving the conflict between production cycle time and quality control. By physically connecting a non-contact marking module to the sewing machine's extendable head structure and combining it with posture compensation control based on spindle speed, the marking action can synchronously follow the high-speed, variable-speed, and curved motion trajectories of the sewing machine head. This follow-up compensation mechanism effectively overcomes the shortcomings of existing linear marking mechanisms, which cannot adapt to complex two-dimensional sewing paths and must rely on machine stops or inertial gliding for positioning. It ensures quality marking is completed simultaneously with high-speed sewing operations, significantly improving the overall utilization rate of the equipment and production continuity.

[0018] 2. This invention overcomes the limitations of single-vision inspection in identifying latent defects and improves detection confidence under variable speed conditions. It constructs a spatiotemporal synchronization mechanism based on the spindle encoder signal, forcibly aligning the visual image stream of the needle plate area with the mechanical data streams such as thread dynamic tension and torque within the same sewing trajectory coordinate system. This multimodal fusion detection method can not only identify visible defects such as broken threads and skipped stitches, but also keenly capture visually invisible latent quality hazards such as abnormal bottom thread tension and loose stitches through mechanical waveform characteristics. Simultaneously, the sampling mechanism based on physical position triggering eliminates detection distortion caused by speed fluctuations during sewing machine start-stop and speed changes, ensuring detection stability across the entire speed range.

[0019] 3. A digital twin archive that can be directly parsed by automated production lines has been constructed, enabling "zero-search" operations in downstream processes. This invention abandons the traditional physical trace marking or simple "present / absent" information recording mode, and innovatively calculates physical defect points as vector deviation coordinates relative to the template positioning reference, and writes them into electronic tags that travel with the workpiece. This method breaks down the data barrier from the inspection end to the sorting end, allowing the subsequent six-axis robot to directly read the coordinate data in the tag and drive the end effector to accurately locate the defect point without secondary visual scanning or global search, thereby significantly improving the response speed and operational accuracy of automated sorting and rework processes.

[0020] 4. A dynamic judgment logic based on trajectory complexity was established, effectively reducing the false alarm rate in complex process sections. This invention introduces a process complexity assessment mechanism, which can automatically adjust the judgment weights and thresholds of multimodal data according to the curvature changes and speed fluctuations of the sewing path. In complex areas with large natural tension fluctuations, such as corners and reinforcement seams, the system automatically adapts the judgment logic, avoiding false alarms easily generated by traditional fixed threshold detection methods. While ensuring "zero missed detections" of real defects, it minimizes "false detections" caused by process characteristics, improving the system's environmental adaptability. Attached Figure Description

[0021] Figure 1 This is a schematic diagram of the method flow of the present invention.

[0022] Figure 2 This is a system module architecture diagram of the present invention. Detailed Implementation

[0023] Example 1 according to Figure 1 As shown, this invention provides a closed-loop quality control method for template sewing based on digital twin mapping, which operates within a highly integrated intelligent flexible sewing unit. The core hardware of this unit includes a high-speed template sewing machine with multi-axis linkage capabilities, a dedicated U-shaped template for holding the fabric to be sewn, and a digital twin monitoring system built upon the physical equipment. Before the sewing task begins, the operator first clamps the fabric in the U-shaped template, which has an RFID tag embedded with product identification information. Once the U-shaped template is conveyed to the sewing machine's work platform and locked by an electromagnetic positioning pin, the system automatically initializes. At this time, the follow-up marking module installed on the side of the sewing machine's extendable head structure completes self-checking and returns to zero. The integrated coaxial vision acquisition unit adjusts to the optimal focal length, and simultaneously, the RFID reader / writer node reads the tag information to load the sewing trajectory data and quality judgment criteria for the current process, preparing for subsequent synchronous monitoring.

[0024] After the sewing process officially starts, the template sewing machine drives the U-shaped template to move at high speed along a preset nonlinear trajectory in the XY plane. To solve the data distortion problem caused by the traditional time-triggered mode in variable-speed sewing (such as cornering and start-stop phases), this embodiment adopts a spatiotemporal synchronization sensing mechanism based on physical position. Specifically, the system receives the high-frequency pulse signal from the sewing machine spindle encoder in real time as a global spatiotemporal reference. Whenever the spindle rotates through a specific phase angle (for example, the moment when the needle tip is about to pierce the fabric), the encoder triggers a synchronous acquisition command. In response to this command, the coaxial vision acquisition unit immediately performs millisecond-level exposure shooting on the needle plate area to capture the instantaneous image of the current needle position; at the same time, the mechanical sensor integrated into the spindle drive chain synchronously captures the surface tension waveform and spindle torque data within this time window. Through this hard-connected triggering method, whether the sewing machine is in high-speed straight sewing or low-speed precision corner sewing, each frame of image and each segment of mechanical waveform acquired by the system can strictly correspond to the specific physical needle position on the sewing trajectory, thereby ensuring a high degree of alignment of multi-source data in the spatial and temporal dimensions.

[0025] The collected raw data is then fed into the digital twin mapping engine for processing. The system first maps the discrete visual and mechanical data streams to a unified sewing trajectory coordinate system, constructing a multi-dimensional feature curve that extends over time. In this process, the system does not rely solely on a single data source but performs multi-modal fusion judgment. The visual algorithm module is responsible for extracting appearance features such as stitch edge continuity and texture density distribution from the image to identify visible defects such as skipped stitches and broken threads. Simultaneously, the mechanical analysis module deeply analyzes the peak stability of the tension waveform and abnormal fluctuations in torque to detect hidden problems such as excessively loose bottom thread tension and thread tangling. To further improve the accuracy of the judgment, the system introduces a dynamic process complexity evaluation mechanism. This mechanism calculates the local geometric curvature and speed variation of the current sewing path in real time. For areas with complex geometry or drastic speed fluctuations (such as pocket corners), the system automatically adjusts the judgment logic, appropriately relaxing the sensitivity to minor visual deformations to prevent false alarms, while tightening the threshold for mechanical features to prevent missed detections, thus achieving a dynamic balance between "over-detection" and "missed detection."

[0026] When the result of multi-dimensional feature data fusion calculation shows that the comprehensive anomaly score exceeds the current adaptive threshold, the system determines that there is a quality defect at the current needle position. At this time, the traditional stop-and-inspection mode would interrupt the production cycle, while this embodiment adopts a dynamic processing strategy that does not stop the machine. The system first accurately locks the position of the sewing machine needle hole in the absolute coordinate system of the machine tool based on the encoder feedback at the moment the anomaly is captured. Then, using the U-shaped template geometric parameters pre-stored in the digital twin model, the system converts the absolute position into vector deviation coordinates relative to the positioning reference point of the U-shaped template through a coordinate transformation algorithm. This vector deviation coordinate represents the inherent physical position of the defect point on the fabric, which does not change with the movement of the U-shaped template on the machine, laying the data foundation for subsequent "what you see is what you get" marking.

[0027] Next, the system enters the crucial follow-up vector marking stage. Due to the continuous high-speed relative motion between the sewing machine head and the U-shaped template, an unavoidable time delay exists between anomaly detection and marking execution. To eliminate positional drift caused by this delay, the system activates a pose compensation controller. This controller reads the instantaneous velocity vector of the sewing machine spindle in real time and calculates the displacement of the machine head relative to the defect point within the delay time. Based on this calculation, the controller sends a trigger command with offset compensation to the follow-up marking module (such as a high-speed inkjet head or laser galvanometer) physically fixed to the extendable machine head structure. While following the movement of the machine head, the marking module precisely projects the physical marker onto the defect coordinates on the moving workpiece surface by adjusting the spray angle or beam deflection angle. The entire process does not require reducing the sewing speed, achieving precise "accompanying" marking under non-linear trajectory operations, completely solving the marking misalignment problem caused by motion ambiguity or positional lag in existing technologies.

[0028] While completing the physical marking, the system performs a data closed-loop operation. The RFID reader / writer node, via contactless communication, writes the defect type code (such as skipped stitches, loose thread, etc.) and the calculated millimeter-level vector deviation coordinates into the data storage area of ​​the RFID electronic tag that moves with the U-shaped template. This not only records the "product defective" status but also embeds a "digital map" of the defect into the product itself. When the U-shaped template flows to the sewing unit's exit, subsequent six-axis robots or sorting equipment do not need to perform a global visual scan to find the defect again. They only need to read the vector coordinate data within the tag to drive the end effector to directly and quickly locate the defect point for grasping, sorting, or targeted rework. This data-driven back-end processing method eliminates the time loss of traditional manual searching or secondary machine vision searches, achieving full-process, coordinate-level, unmanned closed-loop management from front-end sewing inspection to back-end automated sorting.

[0029] Example 2 according toFigure 2 As shown, this invention details the specific hardware architecture and logical connection relationships of a template sewing quality closed-loop control system based on digital twin mapping.

[0030] The control system constructed in this embodiment is mainly deployed in an intelligent flexible sewing production line. Its physical foundation includes three parallel template sewing machines, a U-shaped template conveyor platform for carrying fabric, and a six-axis industrial robot responsible for material flow. At the sensing front end of the system, a spatiotemporal synchronization sensing module is installed. This module is not an independent acquisition device but is deeply embedded in the control loop of the sewing machine through a specific electrical interface. Specifically, the signal triggering end of the spatiotemporal synchronization sensing module establishes a direct electrical connection with the main shaft encoder of the template sewing machine. This hard connection ensures that the sensing module no longer relies on an unstable timer but strictly responds to the phase pulse signal emitted by the main shaft encoder. Whenever the sewing machine main shaft rotates through a preset angular interval, the encoder generates a synchronous trigger pulse. This pulse signal is simultaneously transmitted via a high-speed signal line to a coaxial vision acquisition unit mounted above the needle plate, a thread tension sensor integrated into the sewing machine's thread path, and a torque monitoring unit within the main shaft motor driver. This parallel trigger connection structure enables strict synchronization of visual data (such as stitch images) and mechanical data (i.e., the tension waveforms of the top / bottom threads and the instantaneous torque when the spindle pierces the fabric) at the source of physical acquisition, based on the "needle position." Regardless of whether the sewing machine is accelerating or decelerating, the system can obtain multi-dimensional raw data with precise spatial positioning, laying a solid physical foundation for subsequent accurate analysis. The core processing hub of the system is the digital twin mapping engine, which is connected to the aforementioned spatiotemporal synchronization sensing module via a high-bandwidth industrial data bus. This engine internally constructs a virtual model that is a complete mirror image of the physical sewing machine and establishes a real-time spatiotemporal reference mapping channel with the spindle encoder through a dedicated communication protocol. Through this channel, the digital twin mapping engine can continuously receive real-time XY axis coordinate streams from the encoder and dynamically project the discrete image frames received from the sensing module, along with continuous thread tension and spindle torque waveforms, onto a unified sewing trajectory coordinate system. Based on this, the multi-dimensional feature fusion module inside the engine performs real-time calculations on the projected data and, combined with a pre-set process complexity model, comprehensively determines the quality status of the current trajectory point. This deep data coupling between modules enables the system to effectively distinguish between normal tension / torque fluctuations caused by process actions (such as backstitching) and actual sewing defects, thereby achieving high-confidence anomaly detection in complex nonlinear sewing processes.

[0031] When the digital twin mapping engine detects a sewing quality anomaly, a signal is transmitted to the follow-up vector marking subsystem. This subsystem features a significantly innovative mechanical design: its core execution component—the follow-up marking module (e.g., a miniature laser emitter or a high-speed inkjet head)—is not statically mounted but mechanically fixed to the side of the template sewing machine's extendable head structure via a precision bracket. This means the marking module and the sewing machine needle bar maintain a rigid, synchronized motion relationship in the XY plane, sharing the same motion vector. To address the physical position deviation caused by the time difference between anomaly detection and marking execution, the subsystem also includes a pose compensation controller with bidirectional data connection to the spindle encoder and the digital twin mapping engine. This controller receives the anomaly trigger signal from the engine and simultaneously reads the instantaneous head velocity vector fed back by the encoder in real time, calculating the minute displacement of the head movement within the response delay time using an internal algorithm. Based on this displacement, the controller dynamically adjusts the action parameters of the marking module (such as adjusting the deflection angle of the laser galvanometer or delaying the triggering time of the inkjet), thereby achieving accurate "accompanying" physical marking of defects on the fabric surface without stopping the high-speed movement of the machine head.

[0032] To facilitate cross-process quality data flow, this system also integrates an RFID data closed-loop node. This node is coupled with the pose compensation controller via an internal communication protocol and is equipped with a non-contact read / write antenna installed under the sewing machine table. After the pose compensation controller completes the calculation of the marking action, it sends the digital vector deviation coordinates of the defect point relative to the positioning reference hole of the U-shaped template to the RFID data closed-loop node. The node is then activated and writes this coordinate data containing precise location information and the defect type code into the RFID electronic tag fixed to the current U-shaped template in real time via wireless RFID signal. This connection transforms the physical marking action into a permanent record at the data level, making the U-shaped template an intelligent carrier of "self-diagnostic medical records." When the template flows to subsequent processes, downstream six-axis robots or sorting equipment do not need to rely on complex vision search systems. They only need to establish a connection with the electronic tag through their built-in reading end to directly parse the vector coordinates of the defect, thereby driving the robotic arm to reach the defect location directly via the shortest path for grasping or sorting, forming a closed-loop control system that highly integrates the physical world and digital information.

[0033] Example 3 The fabric defect detection system constructed in this embodiment is first established on the pre-analysis and dynamic early warning mechanism of the sewing process. Before the U-shaped template enters the sewing station, the system does not passively wait, but actively reads the sewing trajectory instruction code pre-stored in the electronic tag through the radio frequency read / write node. The system's process complexity assessment module immediately performs geometric and kinematic analysis on the trajectory code, calculating the local radius of curvature and the estimated principal axis acceleration at each point on the entire sewing path. Based on these parameters, the system divides the sewing path into a "smooth sewing zone" and a "high-risk complex zone". For the smooth zone with straight lines or large radii of curvature, the system automatically sets a lower mechanical sensitivity threshold and assigns a higher visual judgment weight; while for the high-risk zone with sharp turns, backstitches, or multiple layers of fabric, the system automatically switches strategies, reducing the weight of visual features to avoid false alarms caused by presser foot obstruction or fabric wrinkles, while tightening the judgment thresholds for tension and torque, and activating the micro-waveform analysis algorithm to capture subtle resistance anomalies. This adaptive configuration based on "process feedforward" allows the system to make logical preparations to deal with the detection difficulties of different road sections before sewing even begins.

[0034] Upon entering the real-time operation phase, this embodiment employs a "full-element spatiotemporal alignment" data acquisition strategy. Beyond simply synchronizing vision and tension, the system also incorporates auxiliary data such as the height and position of the sewing machine presser foot and the real-time current load of the XY motor into the monitoring scope. Data sampling from all sensors is rigorously controlled by the phase pulses of the spindle encoder. The digital twin mapping engine internally runs a high-frequency feature extraction algorithm: at the visual level, the algorithm not only compares the geometric positional deviations of the stitches but also infers the thread tension by calculating changes in the gloss of the stitch surface through texture analysis; at the mechanical level, the algorithm dynamically regularizes and compares the acquired tension waveform with the standard "gold sample" waveform, calculating the waveform area difference and peak phase difference. Subsequently, the multi-dimensional fusion judgment engine inputs all the above features into the anomaly scoring model, which incorporates the aforementioned "process complexity coefficient" as a dynamic correction factor. If the current area is a high-risk, complex zone, the model automatically offsets reasonable tension fluctuations caused by mechanical inertia; only when the score significantly exceeds the dynamic threshold after complexity correction is it confirmed as a genuine defect. This mechanism greatly improves the robustness of the system under non-steady-state conditions and effectively solves the industry pain point of frequent false alarms at corners in traditional equipment.

[0035] For points identified as defects, this embodiment details the complete logic of "vector coordinate calculation and prediction compensation." When an anomaly is triggered, the system first records the instantaneous position of the sewing machine needle tip in the machine tool's absolute coordinate system. However, since the U-shaped template is in continuous high-speed motion with the XY platform, this absolute coordinate is a meaningless variable for subsequent processes. Therefore, the coordinate mapping and calibration module immediately incorporates the real-time pose matrix of the U-shaped template and converts the machine tool's absolute coordinates into relative vector coordinates relative to a specific positioning reference hole on the U-shaped template through inverse kinematics calculation. Next, to achieve non-stop marking, the pose compensation controller intervenes. It calculates the time lag between the "anomaly determination moment" and the "marking module execution moment" in real time, and, combined with the current real-time velocity vector of the XY axes, predicts the displacement deviation of the defect point relative to the nozzle or beam center of the marking module at the instant the marking action occurs. Based on this, the controller generates a drive command with phase lead, controlling the follow-up marking module to emit laser pulses or ink droplets to the predicted physical position within a very short time window. This process not only compensates for mechanical transmission delays but also offsets nonlinear errors caused by changes in movement speed, ensuring that the physical markers accurately land on fabric defects.

[0036] Finally, this embodiment details the RFID-based data closed-loop and automated sorting logic. The RFID data closed-loop not only writes the defect coordinates but also the specific classification code of the anomaly (such as "broken thread," "skipped stitch," "oil stain," etc.) and a snapshot of the process parameters at the time of the anomaly (such as spindle speed and tension peak) into the RFID tag. This complete data package constitutes the product's "digital birth certificate." When the product flows to the later stage, the six-axis robot of the intelligent sorting unit reads the tag. If the tag shows "no anomaly," the robot executes the standard palletizing procedure; if an anomaly is detected, the robot parses the anomaly coordinates and type. For "repairable" defects (such as localized skipped stitches), the robot directly plans the optimal path based on the vector coordinates and sends the workpiece to the needle of the automatic rework machine; for "unrepairable" defects (such as fabric damage), they are directly placed in the scrap bin. Furthermore, this process data stored in the RFID is ultimately uploaded to the central production management system for reverse training of the front-end anomaly scoring model, correcting process complexity assessment parameters, thereby achieving self-evolution and accuracy iteration of the entire manufacturing system.

Claims

1. A method for identifying defects during the sewing process of a template machine, characterized in that, The method includes: The dynamic image stream of the needle plate area and the mechanical data stream of the template machine spindle and sewing thread are synchronously triggered by the main shaft encoder signal, and the dynamic image stream and mechanical data stream are mapped to a unified sewing trajectory coordinate system. The sewing quality status is determined based on the fusion result of the mapped multidimensional feature data. When the judgment result is abnormal, calculate the vector deviation coordinates of the defect point relative to the U-shaped template positioning reference based on the current sewing trajectory position; Based on the vector deviation coordinates and real-time sewing speed, the follow-up marking module is controlled to perform pose compensation and generate physical markings on the moving workpiece surface. The vector deviation coordinates are written into the electronic tag fixed to the U-shaped template via an RFID reader / writer node.

2. The defect identification method during template sewing process according to claim 1, characterized in that, The method uses the pulse signal of the spindle encoder as a spatiotemporal reference and triggers the vision acquisition unit to capture the region of interest image at a preset needle phase point. Simultaneously capture the spindle torque waveform and surface tension waveform within the time window corresponding to the phase point; By binding the region of interest image with waveform data through a timestamp alignment mechanism, a multimodal sewing fingerprint corresponding to the current sewing trajectory coordinate point is constructed.

3. A method for identifying defects during the sewing process of a template machine according to claim 1 or 2, characterized in that, The method extracts the line edge features and texture features of a dynamic image stream; Extract the peak tension stability characteristics and torque fluctuation characteristics of the mechanical data stream; The above features are input into the anomaly scoring model for weighted calculation to obtain a comprehensive anomaly score, and the comprehensive anomaly score is compared with an adaptive judgment threshold to output the judgment result.

4. The defect identification method during template sewing process according to claim 3, characterized in that, The method calculates the process complexity coefficient of the current sewing trajectory segment in real time; By combining the distribution trend of historical quality data with the process complexity coefficient, the judgment threshold is dynamically adjusted; In the corners or speed-changing areas of the sewing trajectory, the threshold for visual feature weights is automatically relaxed while the threshold for mechanical feature weights is tightened.

5. A method for identifying defects during the sewing process of a template machine according to claim 1 or 4, characterized in that, The method calculates the process complexity coefficient by linearly combining the root mean square curvature of the sewing path, the coefficient of variation of the sewing machine spindle speed, and the range of thread tension fluctuation.

6. The defect identification method in the template sewing process according to claim 1, characterized in that, The method obtains the absolute position of the sewing machine needle hole in the machine tool coordinate system at the moment of the abnormality. By reading the positioning feature points on the U-shaped template, a local coordinate system for the U-shaped template is established; The absolute position in the machine tool coordinate system is converted into a vector deviation value relative to the origin of the U-shaped template using a coordinate transformation matrix.

7. The defect identification method during template sewing process according to claim 1, characterized in that, The method calculates the displacement vector of the machine head movement within the time delay from the time of anomaly determination to the time of marking execution by a follow-up marking module integrated into the side of the extendable head structure of the template sewing machine and moving synchronously with the needle bar. The displacement vector controls the mirror deflection angle or inkjet triggering timing of the follow-up marking module.

8. The defect identification method in the template sewing process according to claim 1, characterized in that, After the sewing task is completed, the method uses a six-axis robot to read the vector deviation coordinates within the electronic tag; The six-axis robot directly drives the end effector to move to the physical location of the defect on the workpiece surface based on the vector deviation coordinates to grasp it.

9. A method for identifying defects during the sewing process of a template machine according to claim 1 or 7, characterized in that, The follow-up marking module uses a laser thermochromic marking unit or a high-speed micro-volume inkjet unit, including a coding pattern corresponding to the anomaly type or a color-printing mark under a specific wavelength.

10. A defect identification system during the sewing process of a template machine, the system being applied to the defect identification method during the sewing process of a template machine according to any one of claims 1 to 9, the system comprising: The spatiotemporal synchronization sensing module has its signal triggering terminal electrically connected to the main shaft encoder of the template sewing machine; The digital twin mapping engine receives the output data of the spatiotemporal synchronization sensing module through the data bus and establishes a spatiotemporal reference mapping channel with the spindle encoder. The follow-up marking module is mechanically fixed to the side of the extension arm type head structure of the template sewing machine, and the posture compensation controller is bidirectionally connected to the main shaft encoder and digital twin mapping engine; The radio frequency data closing point is communicatively coupled to the pose compensation controller.

Images