Automatic production line for underpants

Abstract

The invention relates to an automatic production line for underpants, and aims to solve the problems of high labor intensity, low production efficiency and large quality fluctuation in the traditional production of the underpants. A plurality of process modules such as an automatic cloth roll feeding system, an automatic cutting bed, a cloth piece grabbing and positioning module, a rubber band sewing module, a cloth piece folding and sewing module, an ultrasonic welding module and the like are integrated, and the full-automatic process of underpants production is achieved. Production parameters of the production line can be automatically adjusted according to underpants of different models, accurate butt joint of cloth pieces, rubber bands and stitches is guaranteed, and high consistency and high quality of all finished products are guaranteed. Particularly, ultrasonic-assisted welding fixation is added, so that the edge sealing strength is improved, and the problems of stitch exposure and leakage in a traditional sewing mode are solved. The production line system improves the efficiency, the quality stability and the resource utilization rate of underpants production, is suitable for large-scale underpants manufacturing, and has wide market application prospects.

Description

Technical Field

[0001] The invention belongs to the technical field of automatic production of clothing, in particular to an automatic production line for underwear. Background Art

[0002] Underwear, a basic item of everyday apparel, is widely used in a variety of materials and styles. The traditional underwear production process typically involves multiple steps, including fabric cutting, sewing, rubber banding, folding, and edge banding. Each step requires manual labor, and the processes are often discrete, lacking a unified production rhythm and coordination. This production method is not only labor-intensive but also inefficient, resulting in limited output per unit time and significant quality fluctuations.

[0003] In traditional production processes, fabric cutting and splicing are typically performed manually or using semi-automated equipment. This not only increases labor intensity but also is prone to cutting errors, resulting in inconsistent sizes or uneven splicing. Even with the use of automated cutting equipment, due to the softness and slipperiness of fabric, traditional automated cutting systems still struggle to prevent fabric misalignment or misalignment during transport, impacting the quality of subsequent processes.

[0004] Sewing rubber bands is a crucial step in underwear production. Traditionally, rubber band sewing relies on manual adjustments during the feeding and sewing process. Due to the elastic nature of rubber bands, traditional sewing methods struggle to precisely control their tension, resulting in uneven stitching, loosening, or overtightening, impacting the wearer's experience and the product's lifespan.

[0005] Furthermore, traditional underwear edge-sealing techniques primarily rely on sewing. While this effectively connects fabrics, it can lead to issues like exposed stitches and leakage. This is particularly true for highly elastic and soft fabrics, where this sewing method often results in loose, unsightly edges and compromises comfort. To address these issues, some production lines have begun to incorporate ultrasonic welding technology in recent years. However, in practice, existing ultrasonic welding equipment often suffers from uneven weld edges and unstable welding results, making it difficult to meet the demands of large-scale production.

[0006] Therefore, traditional underwear production technology faces problems such as low production efficiency, large quality fluctuations, and imprecise processes. There is an urgent need for an efficient, stable, and precise production method to improve the production quality, production efficiency, and process stability of underwear. Summary of the Invention

[0007] The purpose of the present invention is to provide an automatic underwear production line. By combining modern automation technology, intelligent control systems and ultrasonic welding technology, the present invention solves the problems existing in the above-mentioned traditional underwear production, provides a fully automatic and intelligent underwear production line, significantly improves production efficiency, product quality and consistency, and provides a new solution for the underwear manufacturing industry.

[0008] The technical solution adopted in the present invention is as follows:

[0009] An automatic underwear production line, comprising:

[0010] Automatic cloth roll feeding system for continuously conveying cloth piece 1 and cloth piece 2;

[0011] An automatic cutting bed for cutting the first and second pieces of fabric;

[0012] The cloth grabbing mechanism includes two sets of manipulators, which respectively grab cloth piece 1 and cloth piece 2 and stack them as required;

[0013] Sewing module 1 and sewing module 2 are used to sew the overlapping cloth pieces together to form a crotch cloth piece unit under the positioning and clamping of the jig;

[0014] The leg rubber band sewing control unit includes a rubber band feeding mechanism and a rubber band sewing mechanism, which sews the leg rubber band to the leg edge of the crotch fabric unit;

[0015] The waist rubber band sewing module is used to sew the waist rubber band to the waist opening of the crotch fabric unit through the waist feeding mechanism;

[0016] The fabric folding and sewing module includes a folding robot and a sewing device, which is used to fold the fabric with the rubber band sewn on in half and seal the edges to form the trouser shape;

[0017] Ultrasonic welding module, used to weld and fix the edges of the folded trouser pieces;

[0018] The edge sewing module is used to sew the excess fabric edges after welding;

[0019] The various workstations are connected in series via synchronous conveyor lines, and their rhythms are coordinated uniformly by a control system.

[0020] Among them, the two sets of manipulators in the cloth grabbing mechanism are both provided with adsorption ends, which can respectively grab the cloth piece 1 and the cloth piece 2 and release them on the fixture synchronously.

[0021] Among them, the sewing module group 1 and the sewing module group 2 are respectively arranged at parallel workstations, and the cloth pieces are driven by the jig to enter the two sewing mechanisms for sewing synchronously.

[0022] The rubber band feeding mechanism can synchronously feed the rubber band according to the beat of the cloth piece, and the rubber band sewing mechanism sews the rubber band and cuts off the excess part of the edge.

[0023] Among them, the folding robot in the cloth folding and sewing module has a multi-joint structure, which cooperates with the folding clamp to fold the trouser piece along the central axis, and the sewing device implements the edge seam processing.

[0024] Among them, the ultrasonic welding module performs auxiliary welding.

[0025] The hemming and sewing module includes an identification module and a cutting head, which performs fine cutting according to the contour of the welding edge.

[0026] Among them, the control system is a centralized PLC controller, which can realize dynamic scheduling and alarm joint control of the entire process modules of cloth grabbing, rubber band feeding, folding and welding, and beat synchronization.

[0027] The beneficial effects of the present invention are as follows:

[0028] This invention relates to an automated underwear production line and its production process. This fully automated system aims to improve the efficiency, quality, and stability of underwear production, resolving numerous challenges associated with traditional manual and semi-automatic production methods, such as high labor intensity, low production efficiency, and difficulty in quality control. By incorporating advanced technologies such as high-precision fabric handling, automated sewing, elastic tension control, and ultrasonic welding, this system automates the entire process, from fabric feeding to finished product output. This system can also be flexibly adjusted to meet the needs of diverse underwear specifications, accommodating diverse production needs.

[0029] The technical solution of this invention includes multiple key links, including an automatic fabric roll feeding system, an automatic cutting table, a fabric grabbing and positioning system, a sewing and rubber band sewing control module, a fabric folding and stitching module, an edge ultrasonic welding process, and hemming. All processes are coordinated by a central control system to ensure the accuracy and efficiency of each link. In particular, the use of ultrasonic welding technology instead of traditional sewing in the welding and sewing process not only improves the strength of the edge banding, but also avoids the problems of exposed stitches and pinhole leakage common in traditional sewing processes, further improving the appearance and comfort of the finished product.

[0030] This invention utilizes an intelligent rubber band feeding and tension control system to ensure uniform stitching and a comfortable fit, avoiding the unpleasant wearing experience caused by overly tight or loose rubber bands in traditional sewing. Furthermore, the folding and sewing module utilizes a high-precision folding robot arm and a multi-axis tracking platform to ensure symmetrical and precise shaping of all parts of the underwear after molding, ensuring high consistency across sizes, including M, L, and XL.

[0031] Furthermore, the invention boasts robust production adaptability, intelligently adjusting production parameters to meet market demand for a wide range of underwear styles and specifications based on the type and material of fabric and rubber band used. Every piece of underwear produced can be recorded and traced via a central control system, ensuring transparency and quality control throughout the production process.

[0032] In general, the automatic underwear production line system of the present invention not only improves production efficiency and reduces labor costs, but also ensures the high quality and high consistency of finished underwear products. It has broad market prospects and practical application value, is suitable for the production needs of large-scale underwear manufacturers, and promotes the development of the field of intelligent clothing manufacturing. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] Figure 1 It is a structural schematic diagram of the production line of the present invention;

[0034] Figure 2 This is a schematic diagram of one type of suction cup array of the present invention;

[0035] Figure 3 is a vertical cross-sectional view of the suction cup of the present invention when the needle is not extended;

[0036] Figure 4 Schematic cross-sectional view of the comb bar of the suction cup of the present invention when the needles are extended;

[0037] Figure 5 Flowchart of the fabric defect prediction and fine-tuning control method of the present invention.

[0038] In the figure, 1. Automatic cloth roll feeding system; 2. Automatic cutting bed; 3. Cloth piece grabbing mechanism; 4. Sewing module 1; 5. Sewing module 2; 7. Leg rubber band sewing control unit; 7a. Rubber band feeding mechanism; 7b. Rubber band sewing mechanism; 8. Waist rubber band sewing module; 8a. Waist feeding mechanism; 9. Cloth piece folding and sewing module; 9a. Folding robot; 9b. Sewing device; 10. Ultrasonic welding module; 11. Edge sewing module; 12. Synchronous conveyor line; 13. Suction cup; 131. Suction cup body; 132. Suction cup seat; 133. Suction channel; 14. Telescopic needling mechanism; 141. Needle; 142. Guide ring; 143. S-shaped lever mechanism; 144. Pull rope; 145. Guide channel; 146. Cylinder; 147. Piston; 148. Spring. DETAILED DESCRIPTION

[0039] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0040] See also Figure 1 This embodiment provides a production line system for realizing continuous automated manufacturing of underwear. The production line mainly includes the following stations and modules: automatic cloth roll feeding system 1, automatic cutting table 2, cloth piece grabbing mechanism 3, sewing module 1 4, sewing module 2 5, fixture positioning platform, leg rubber band sewing control unit 7, waist rubber band sewing module 8, cloth piece folding and sewing module 9, ultrasonic welding module 10, and hemming sewing module 11. The entire line process is coordinated by a central control system, and rhythmic transition is achieved through a synchronous conveyor line 12. Specifically:

[0041] Step 1: Fabric supply and cutting

[0042] In this embodiment, the fabric supply and cutting process mainly includes an automatic fabric roll feeding system 1 and an automatic cutting table 2, which are used to realize the automatic loading and cutting of raw materials of the front fabric piece 1 and the crotch fabric piece 2 of the underwear.

[0043] The automatic fabric roll feeding system 1 comprises multiple independent fabric roll holders, a tension adjustment mechanism, a guide roller assembly, and a synchronous servo feed mechanism. Two fabric roll holders correspond to fabric sheet one and sheet two, respectively, allowing for simultaneous loading of fabric rolls of varying materials, colors, or specifications. Each fabric roll holder is equipped with a tension feedback control unit that detects the tension of the fabric roll during feeding. The electronically controlled tension mechanism adjusts the unwinding damping to ensure a smooth fabric surface and balanced tension.

[0044] To prevent jerking, edge collapse, or deviation during the initial unwinding of large rolls, a transition buffer roller set and a fabric edge correction device are installed in the feed path. The latter uses a photoelectric sensor to detect the fabric edge position in real time and fine-tune the angle via guide rollers to ensure centered fabric delivery. An anti-curling tensioning plate is also installed between the guide rollers to smooth out fabric wrinkles and improve downstream cutting accuracy.

[0045] After the fabric is fed through the feeding system and fed into the automatic cutting table 2, the width detection module first identifies the current fabric type (Fabric 1 or Fabric 2) and calls up the preset cutting parameters. The automatic cutting table uses CNC tools and an adjustable cloth-holding beam for cutting. A vacuum plate is installed on the cutting table to stabilize the fabric and prevent displacement during cutting.

[0046] The system control unit dynamically plans the cutting path based on the set model, supporting straight and curved cutting, as well as custom-shaped contours. The CNC cutting machine's blade head uses either a rotatable circular or reciprocating blade, and its cutting speed and blade pressure can be intelligently adjusted to suit the thickness and flexibility of the fabric.

[0047] After cutting is completed, a cloth piece identification and separation mechanism is provided at the outlet of the automatic cutting bed. The mechanism includes an adsorption belt, an identification and scanning device, and a sorting push rod. The identification device can identify the cut cloth piece as cloth piece one or cloth piece two through a QR code, RFID, or color detection, and guide it to the corresponding material receiving area of the cloth piece grabbing mechanism (3), respectively, to ensure that the type and position of the cloth piece in the subsequent grabbing link are accurate.

[0048] To meet the needs of batch alternating production, the automatic cutting bed in this embodiment supports rapid switching of cloth specifications. The model of underwear (such as M, L, XL) can be selected through the operating terminal, and the system automatically calls the matching cutting contour, size and stacking method to achieve seamless switching of cloth pieces of different specifications, avoiding line stops for mold changes.

[0049] Preferably, the cutting positions of cloth piece one and cloth piece two are coordinated with the whole line rhythm, and a dual-channel staggered cutting method is adopted to form a certain spacing between the cloth pieces in the grasping area, which is conducive to the avoidance operation of the grasping robot and avoids adsorption errors caused by excessive density of cloth pieces.

[0050] Through the above structural design, this step realizes the automated and continuous processing from the original cloth roll to the standard cloth piece, providing efficient, accurate and stable cloth source materials for subsequent sewing stations, and improving the front-end stability and compatibility of the entire line's automated production.

[0051] Step 2: Grab and position the fabric

[0052] This step is used to precisely pick up, align, and position the first (front panel) and second (crotch panel) fabric panels, creating the nested structure required for subsequent sewing. This step is accomplished collaboratively by the fabric gripping mechanism 3 and the jig positioning platform. The gripping mechanism includes two independent robotic arms, each responsible for the simultaneous handling of fabric panels 1 and 2.

[0053] 1. Fabric identification and delivery

[0054] Cut pieces 1 and 2 are transported to the pick-up station via a low-speed conveyor belt equipped with a photoelectric sensor and a positioning buffer. Once the pieces arrive at the designated pick-up area, a photoelectric sensor automatically detects their position and sends a signal to the higher-level control system to activate the corresponding robotic arm.

[0055] The end of the output channel is also equipped with an electrostatic elimination module and a slow-drop wind curtain structure to prevent lightweight fabrics from warping, overlapping or drifting due to stacking or wind disturbance, thereby ensuring the consistency of the edges and corners of the fabrics before grabbing.

[0056] 2. Robotic arm structure and picking action

[0057] The robotic arms are all six-degree-of-freedom structures, with a flexible adsorption module at the end. This module uses a multi-point negative pressure suction cup array to adapt to the irregular surface of soft cloth. The suction cup is connected to the lifting rod and can be pressed lightly against the surface of the cloth through an adaptive floating device to form a complete negative pressure adsorption. In addition, see Figures 2 to 4 Since the cloth piece has a porous structure, in order to ensure the stable adsorption and pickup of the suction cup, a telescopic puncture mechanism 14 is respectively provided in the suction cup 13 of the outermost circle of the multi-point negative pressure suction cup array; the telescopic puncture mechanism 14 includes a puncture needle 141, a guide ring 142, an S-shaped lever mechanism 143, a pull rope 144, a guide channel 145 and a piston mechanism; the guide ring 142 and the guide channel 145 are respectively fixed on the inner wall of the suction cup body 131, and the piston mechanism is fixed in the suction channel 133 of the suction cup seat 132, and the diameter of the piston mechanism is smaller than the diameter of the suction channel 133 so as not to affect Negative pressure is generated in the suction cup body 131 through the suction channel 133; the piston mechanism includes a vertically arranged cylinder body 146 and a piston 147 sealingly and slidingly arranged in the cylinder body 146; a spring 148 is connected between the lower end of the piston 147 and the cylinder body 146; the front end of the needle 141 is tilted downward and outward, and the rear end is hinged to one end of the S-shaped lever mechanism 143; the front end of the pull rope 144 is connected to the other end of the S-shaped lever mechanism 143, and the rear end passes through the guide channel 133 and is connected to the lower end of the piston 147; a rotating shaft is provided in the middle of the S-shaped lever mechanism 143, and a torsion spring is sleeved on the rotating shaft.

[0058] The crawling action is divided into the following three steps:

[0059] (1) Positioning: The robotic arm moves to the photoelectric recognition mark and aligns with the edge contour of the cloth;

[0060] (2) Adsorption: The suction cup 13 descends to the surface of the cloth, and the external suction mechanism sucks air outwards through the suction channel 133, so that negative pressure is generated in the suction cup body 131, and the negative pressure is uniformly applied to the cloth for adsorption; in this process, when the air is sucked outwards through the suction channel 133, the piston 147 is simultaneously attracted by the negative pressure, overcomes the pulling force of the spring 148 and moves upwards, and when the piston 147 moves upwards, it pulls the drawstring 144, and the drawstring 144 pulls the S-shaped lever mechanism 143 to rotate, and the other end of the S-shaped lever mechanism 143 pushes the needle 141 to pierce the cloth obliquely outwards and forwards, so that the cloth is stably fixed;

[0061] (3) Lifting and transfer: After the adsorption is completed, the robotic arm is lifted to a safe height and transferred to the fixture positioning platform;

[0062] (4) After reaching the fixture positioning platform, the external suction mechanism cancels the suction, and the suction cup body 131 loses the negative pressure. During this process, the piston 147 loses the negative pressure attraction, and under the action of the restoring force of the spring 148, the piston 147 returns to the initial position; under the action of the restoring force of the torsion spring, the S-shaped lever mechanism 143 rotates, driving the needle 141 to be retracted into the suction cup body 131.

[0063] The two sets of robotic arms can achieve synchronous grasping and asynchronous release under timing control. The release action of cloth piece two (crotch cloth) lags behind that of cloth piece one, so that it falls on the surface of the main cloth piece.

[0064] 3. Fixture positioning platform design and positioning process

[0065] The fixture positioning platform is set in the center of the cloth release station. Its upper surface is an array of adsorption holes with a multi-zone vacuum zoning function. The adsorption area can be selectively opened according to the specifications of the cloth to improve positioning stability and prevent wrinkles.

[0066] The platform is surrounded by multiple sets of boundary blocks and centering rails. The rails on both sides are adjustable in spacing to accommodate the waist widths of fabrics of varying sizes. Built-in encoders in the rails record the precise X and Y positioning of the fabrics and provide feedback to the control system for verification.

[0067] After the first piece of cloth is released, it is first fixed to the platform surface under the action of vacuum adsorption; then, the robotic arm (3b) slowly releases the second piece of cloth from above, and places it on the crotch area of the first piece of cloth. The release process can be equipped with a visual assistance system, which uses marker points or image recognition technology to identify the edge of the cloth piece and correct the release posture and coordinate deviation in real time.

[0068] In order to ensure the accuracy of the overlap between cloth piece two and cloth piece one, the platform is equipped with a laser alignment auxiliary module, which projects a reference positioning line onto the platform surface to facilitate the visual recognition system or manual debugging to check the overlap of the cloth pieces.

[0069] 4. Compression and temporary fixation

[0070] After the cloth pieces are stacked, the second cloth piece is lightly pressed on the first cloth piece through the liftable pressure beams on both sides of the platform. The pressure beams are flexible silicone edge strip structures that can maintain uniform pressure distribution without damaging the cloth surface.

[0071] In addition, a hot air blowing system or a low-temperature infrared light shaping module can be optionally installed to perform a short-term temperature treatment on the overlapping area to release the stress in the fabric and make the edges fit more tightly, creating an ideal overlapping state for subsequent sewing processes.

[0072] 5. Positioning tolerance and error compensation processing

[0073] Considering the ductility of fabric, especially where rubber bands are sutured, which can cause deformation due to micro-stress and stretching, the platform's control system incorporates correction and compensation logic. When the system detects, through visual feedback, that the fabric overlap deflection exceeds ±1.5mm, it can instantly correct the release trajectory or use the jig's fine-tuning platform to compensate for the coordinates.

[0074] When the system identifies defects such as unflattened corners, overlapping and misplaced cloth pieces, or missing corners, it can trigger an abnormal cloth rejection signal, automatically clear the current platform cloth pieces and wait for the next pair of cloth pieces to be loaded, avoiding defective cloth pieces from being transmitted to the sewing station.

[0075] Step 3: Initial seam of the cloth and the formation of the middle piece

[0076] The goal of this step is to mechanically sew the nested and positioned fabric panels 1 (front panel) and 2 (crotch panel) together to form a structurally complete, integrated crotch panel unit, laying the foundation for subsequent elastic band sewing and three-dimensional shaping. This step is performed by Sewing Module 1 4, Sewing Module 2 5, and the intermediate fixture platform. The system utilizes a distributed collaborative sewing and dynamic workstation switching strategy.

[0077] 1. Sewing module structure and configuration

[0078] Sewing module 1 4 and sewing module 2 5 are respectively arranged on both sides of the positioning platform. They are double-sewing head synchronous sewing structures, suitable for flexible fabrics and have multi-axis motion functions. Each sewing module includes:

[0079] Main sewing unit: adopts servo-driven lockstitch sewing head with automatic thread trimming, automatic presser foot lifting and thread breakage detection functions;

[0080] Synchronous pressing mechanism: equipped with synchronous feeding teeth and flexible upper pressing wheel, which can keep the feeding tension of multiple layers of fabric consistent;

[0081] Thread tension management unit: built-in tension feedback sensor, adapting to dynamic adjustment of different fabrics and thread types;

[0082] Control module: connected to the central PLC system to achieve precise control of sewing speed, presser foot stroke, start and end points.

[0083] 2. Sewing start process

[0084] After the fabric pieces are nested on the center jig platform and the system confirms the "positioning lock signal," the platform automatically enters the sewing preparation state. The sewing modules on both sides extend the presser foot assemblies to press the edges of the fabric pieces, ensuring that the laminated fabric pieces are firmly attached to the positioning platform.

[0085] The control system issues synchronized commands during the sewing process, with two modules sewing along a set trajectory (e.g., the crotch curve of underwear) from the bottom edge to the top edge of the first fabric piece. The stitching can be done with either a herringbone stitch or a three-thread stitch, depending on the product design.

[0086] In order to prevent the edge of the cloth from slipping due to uneven feeding, the platform is equipped with a reverse limit clamp, which applies a constant clamping force to the unsewn end during the movement of the sewing machine to prevent dislocation.

[0087] 3. Suture path and trajectory calibration

[0088] The stitching trajectory can be preset to a standard curve, arc, or S-shape. The sewing module is equipped with a stepper encoder and path calibration sensor. The built-in image recognition module reads the fabric edge reference line (or trimming mark). The system automatically compares the target trajectory with the actual placement trajectory, performing dynamic corrections on the X and Y axes to ensure the stitches precisely fit the edges of the two fabric pieces.

[0089] If an abnormal deviation exceeding ±2mm is detected, the system can interrupt the current sewing action and push the cloth piece to the side line for scrap marking.

[0090] 4. Post-sewing structure maintenance and transportation

[0091] After sewing is completed, the integrated fabric unit is released by the clamping platform, the sewing machine automatically trims the thread, and the presser foot is raised. The platform pushes the finished fabric out of the sewing area and into the transfer buffer area through a rotating mechanism or push rod structure.

[0092] In the transfer area, fabric pieces are slowly conveyed along a conveyor belt equipped with flexible smoothing brushes, ensuring that the seams remain straight and free of curling or shrinkage. This area also features a thread detection and thread trimming mechanism, which uses a micro-suction duct to remove excess thread tails, enhancing the subsequent visual quality.

[0093] 5.Precision management and quality feedback

[0094] This process is embedded with closed-loop monitoring logic throughout the entire process. Signals such as sewing tension, current changes, and feeding synchronization deviation are uploaded to the central control system in real time. The following fault-tolerant mechanisms are also provided:

[0095] If inconsistent feeding tension causes wrinkles in the fabric, the system triggers a "pause + prompt adjustment" signal;

[0096] If the thread breaks, the main sewing unit will automatically stop and go back to the previous piece of fabric number to avoid work delays;

[0097] After stitching, the fabric pieces will be checked for stitch integrity and symmetry by an image detection module, and the qualified / unqualified status will be marked and uploaded to the database.

[0098] Step 4: Leg rubber band suture

[0099] The purpose of this step is to sew the two leg elastic bands (rubber bands) to the leg openings on both sides of the crotch panel unit to form a leg-encircling structure with elastic wrapping function. This step is performed by the leg elastic sewing control unit 7, which mainly includes the elastic feeding mechanism 7a and the elastic sewing mechanism 7b.

[0100] 1. Rubber band feeding mechanism structure and action flow

[0101] The rubber band feeding mechanism 7a consists of two sets of symmetrically arranged servo-driven feeding units. Each set of feeding units includes: an elastic band coil rack, an automatic tension adjustment device, a guide roller group, a programmable fixed-length feeding mechanism (with pulse code feedback) and a rubber band pre-pressing auxiliary device.

[0102] The elastic band is fed from a roll-up rack and guided by guide rollers to a fixed-length feed unit. Before feeding, the system queries the leg length database based on the fabric model, controls the servo motor to output a matching elastic band segment, and uses a tension feedback controller to adjust the tension in real time to prevent the elastic band from sagging or overstretching.

[0103] Preferably, a visual identification point or color ribbon code is installed at the end of the rubber band to detect whether there is any breakage, overlapping or wrong material; if an abnormal state is identified, the feeding mechanism will stop immediately and temporarily store the cloth in the waiting area and will not enter the sewing section to avoid work delays.

[0104] 2. Positioning guide rail and fabric joint structure

[0105] After the crotch cloth unit enters the sewing station through the synchronous conveyor line 12, a pair of flexible adsorption positioning clamping mechanisms will first stretch and flatten the leg opening area of the cloth edge and send it into the rubber band sewing mechanism 7b.

[0106] 3. Sewing process and path control

[0107] The rubber band sewing mechanism 7b features a double-needle overlock stitching mechanism, supporting both herringbone and three-thread overlock stitching styles. Stitch density and thread tension can be adjusted based on fabric thickness and rubber band width. The sewing head is equipped with an electric XY platform that automatically compensates for curved edges by following the guide rail path.

[0108] The synchronous sewing process of the rubber band and the cloth piece is coordinated by the central control system. The tension change and sewing speed are recorded at each needle placement point. By adjusting the feed and presser foot pressure in real time, the seams are ensured to be tight, the rubber band is even, and the cloth surface is smooth.

[0109] During the sewing process, in order to prevent the rubber band from shrinking or the cloth from turning inside out, a set of small "cloth edge guide rollers" are provided at the front end of the sewing head, which can fold the edge of the cloth downward by 5° to 10° to form a stable wrapping state.

[0110] 4. Treatment of excess rubber band ends

[0111] After sewing is completed, the rubber band sewing mechanism 7b automatically cuts the thread, and the excess rubber band at the end is automatically cut off by the rubber band cutting mechanism. The cutting head is an electric circular knife or a hot melt knife, and the cutting edge is neat and without burrs.

[0112] The cut rubber band heads are automatically recycled into the waste bin through a micro adsorption channel to prevent the remaining segments from entering the next process area.

[0113] 5. Post-sewing output and quality assurance

[0114] The stitched cloth pieces (including the leg bands on both sides) are sent out of the band stitching station by the synchronous conveying mechanism 12 and enter the next process. During the conveying process, an infrared probe is provided to detect whether the band is completely stitched to its full length, and a camera system is used to detect the stitch distribution, stitch density and symmetry of the band.

[0115] If the rubber band position deviation is detected to be more than ±2mm, or the stitching density is lower than the set threshold, the system will automatically mark the fabric piece as "needs re-inspection" and divert it to the side line for manual review or rework.

[0116] Through the above-mentioned structural configuration and closed-loop control strategy, this step can achieve high-precision leg elastic stitching while maintaining high-speed operation, ensuring uniform elastic tension, regular seams, and natural fabric contours, providing a reliable hemming foundation for subsequent molding processes.

[0117] Step 5: Suture the waist rubber band

[0118] This step is used to sew the elastic bands on the front and back waist edges to form a waist structure with elastic closing function. This step is achieved by the waist rubber band sewing module 8, which mainly includes a waist feeding mechanism 8a and a sewing mechanism.

[0119] The waist feeding mechanism 8a includes two symmetrically arranged feeding mechanisms, which correspond to the stitching areas of the front waist and the back waist in the cloth piece respectively. Each feeding mechanism includes: a rubber band reel and tension controller, an electronically controlled servo feeding roller, a guide channel with deviation correction function and an end cloth pressing and shaping roller.

[0120] The system dynamically calls preset rubber band length parameters based on the fabric size and waist curvature, driving the feed roller to precisely deliver the rubber band. During the conveyor path, a tension controller applies micro-tension to the rubber band, slightly stretching it and helping to maintain the fabric's hem during subsequent stitching.

[0121] To prevent the rubber band from deflecting, twisting or kinking during high-speed feeding, a rotating anti-twist mechanism and an infrared recognition point corrector are installed at the end of each feeding guide to ensure that the rubber band is always at the correct angle and facing the fabric.

[0122] Before sewing, the fabric pieces are already sewn with leg elastic bands, and there is a certain tension difference at the edges. To ensure that the fabric pieces do not slip during the waist sewing process, the waist sewing station is equipped with a fabric width-fixing and guiding assembly, which mainly consists of: front and rear synchronous clamping rails, a flexible centering slide, and a central tension release bridge. This assembly can fix and guide the fabric pieces along the waistline, while adapting to the width requirements of fabric pieces of different sizes and realizing dynamic clamping adjustment. "Multi-point tension sensors" are installed on the edges of the clamping rails to monitor the tension state of the fabric pieces before sewing, and feedback is provided to the feeding system to adjust the elastic band expansion and contraction.

[0123] The sewing mechanism features a high-precision single-needle chainstitch mechanism with programmable switching of multiple needle positions, making it suitable for sewing wide and medium-width rubber bands. The sewing stitch can be configured as chain, herringbone, or hemming, with automatic switching via user-defined parameters.

[0124] The sewing module is mounted on a two-axis dynamic track. During the sewing process, it automatically follows the curve of the fabric waistline to prevent deviation caused by elastic tension. A presser foot vibration suppressor is installed at the front end to prevent "skipped stitches" or "floating seams" when the elastic and fabric are superimposed.

[0125] The sewing action is precisely controlled by the central control system to start and stop rhythm, and cooperates with the feeding mechanism to complete synchronous sewing, ensuring uniform stitch density and consistent stitch length to avoid shrinkage or wrinkling.

[0126] After sewing is complete, the ends of the rubber band are cut by a hot knife device, forming a sealed edge. A tension release mechanism is also installed here, using "local heating + breeze cooling" to deal with the rubber band's shrinkage stress, preventing the edges of the fabric from curling or rebounding during the subsequent folding process.

[0127] The slow-release mechanism can also detect the integrity of the seal at the rubber band cut point through optical fiber sensing. If an abnormality is detected, the corresponding cloth piece can be automatically removed and sent to the rework channel.

[0128] To ensure the quality of waistband stitching, the module's exit is equipped with a stitch integrity image recognition module, an infrared rubber band path tracker, and a unit for calculating left and right waistband position deviations. If skipped stitches, uneven stitches, or left-right deviations exceeding ±1.8mm are detected, the system marks the fabric piece as an "abnormal workpiece" and uses a laser dot to indicate rework.

[0129] In addition, this workstation supports an automatic re-sewing mechanism for broken threads. Once the sewing thread breaks, the system can identify the breakpoint and stop the machine, automatically returning the current piece of fabric to the waiting area to ensure the stable quality of subsequent products.

[0130] Through the above structure and control process, the elastic band at the waist edge of the cloth piece can be accurately sewn under high-speed operation, maintaining the natural transition of the edge shape of the cloth piece, balanced tension, and regular seams, ensuring that the cloth piece structure will not be deformed or erroneous due to stress deviation in the subsequent folding process.

[0131] Step 6: Fold the fabric and sew the pants together

[0132] The goal of this step is to fold the fabric unit with the waist and leg elastic bands sewn into a trouser-shaped structure and form a stable three-dimensional contour by edge-sealing. This step is performed by the fabric folding and sewing module 9, which mainly includes: a multi-joint folding robot 9a and a sewing device 9b.

[0133] The finished fabric sheet is fed by conveyor line 12 into the folding station. Initially, it is an unfolded, symmetrical strip, with leg openings on either side, a crotch area in the middle, and a waist elastic band at the top. The sheet is laid flat on a horizontal support platform, which has low-pressure suction holes and a non-slip surface to prevent slippage during folding.

[0134] Paired flexible clamps and slide rails flank the folding station, guiding the edges of the fabric into the designated folding path. An alignment grid aligns the front of the fabric with the centerline of the fabric at the preceding station, ensuring symmetry along the central axis.

[0135] The folding robot 9a features two symmetrical arms with six degrees of freedom (DOF) joints, each equipped with a flexible flattening plate and suction device. The control system sets the folding angle and path based on the size and type of fabric, supporting multiple fabric sizes.

[0136] The folding process is as follows:

[0137] (1) Grab the waist openings on both sides of the cloth piece with both arms simultaneously from the outside of the edge of the cloth piece;

[0138] (2) folding along the set path toward the center axis of the cloth to form a symmetrical overlap;

[0139] (3) The pressing plate gently presses the folded area to flatten it;

[0140] (4) The fixed clamping claw turns the crotch area toward the waist to achieve the bottom crotch encapsulation;

[0141] (5) The three-dimensional outline of the pants is finally formed (in an inverted "U" shape) and the edge is ready for sewing.

[0142] The entire folding process is completed within 4-5 seconds, with a speed-position linkage curve to ensure that no edge misalignment or centerline deviation occurs. The control system has a "folding trajectory learning" function, which allows the path to be set through manual teaching mode and automatically recorded as a standard program.

[0143] During the folding process, in order to ensure that the central axis fold is centered and the edges overlap accurately, a sewing device 9b is provided in the center of the platform, including: a liftable central line boss, left and right slide guide slopes and a crotch downward pressure bending arm.

[0144] The centerline boss automatically rises before the fabric is folded, guiding the two side panels symmetrically onto it to form a "mountain-shaped" hem. The slope of the slide allows the fabric to form a symmetrical waveform under natural tension, with a symmetry error of less than ±1.2mm. In the final stage, the crotch-pressing bending arm pushes the lower portion of the fabric inward, forming a multi-layered bond between the front, back, and crotch panels.

[0145] After folding to form the trouser-shaped structure, the sewing device 9b performs edge sealing along the left and right trouser legs. The sewing assembly is an integrated double-head sewing module that acts on both sides of the trouser seams at the same time, adopts a chain-type hemming line, and has functions such as automatic thread cutting, thread breakage detection, and dynamic tension compensation.

[0146] The sewing head sews by following the edge contour of the cloth. The built-in optical contour recognition module can automatically identify the folding edge of the cloth and make real-time adjustments even if there is a slight offset.

[0147] To prevent the folding layer from slipping, auxiliary positioning and stabilization mechanisms are used during the suturing process, including: a suspended folding fixing clamp, a programmable side pressure arm and a double-sided pressure-relieving adsorption belt.

[0148] This mechanism applies flexible clamping to the folded edge of the cloth before and during sewing, ensuring that there is no "sliding zone" between the sewing foot and the cloth, and ensuring that the seam and folded edge are highly consistent.

[0149] This workstation is equipped with a barcode scanner or RFID identifier, which can synchronously record the process number of each piece of fabric and the completion time of folding and sewing, realizing production rhythm traceability.

[0150] Finished underwear is automatically transported to the next workstation (ultrasonic welding) where it is gripped and transferred by a high-precision mechanical gripper. If an incomplete edge banding or folding anomaly is detected, the platform will automatically alarm, block the flow, and send the underwear to a sideline for rework.

[0151] Through the design of this step, the key transformation from a flat piece of cloth to a three-dimensional trouser-shaped structure is completed, ensuring that the underwear product has neat folds, symmetrical seams, and a natural fit, laying the foundation for structural integrity for the final edge sealing and packaging processes.

[0152] Step 7: Ultrasonic welding of edges

[0153] This step ultrasonically welds the side seams of the folded, panty-shaped panty panels to achieve a stable, non-sewing edge seal, enhancing product integrity and comfort. This step is performed by the edge ultrasonic welding module 10, which primarily includes a welding base, an ultrasonic transducer assembly, a welding head tracking device, a contour recognition module, and a post-weld stabilization output module.

[0154] The underwear pieces entering the welding station are delivered to the welding base by the synchronous conveyor line 12. The base is a grooved support platform with an elastic coating layer and a multi-point negative pressure adsorption hole array on its surface to stabilize the contour of the folded underwear pieces, especially the side overlap area.

[0155] There are liftable clamping blocks and support rails on both sides of the welding area, which are used to dynamically press the seams of the cloth pieces during the welding process to prevent the cloth from slipping or warping due to pressure from the welding head.

[0156] Before welding, the base will automatically adjust the clamping block opening distance and adsorption area distribution according to the underwear model (for example: M, L, XL) to meet the welding needs of fabrics of different waist sizes.

[0157] The ultrasonic welding device consists of a transducer assembly and a vibration system. The transducer is an industrial-grade high-frequency piezoelectric crystal component, which converts input electrical energy into longitudinal mechanical vibration through electrical-to-acoustic energy conversion. The frequency is set between 20 and 35 kHz, the maximum stroke of the welding head is 8 mm, and the adjustable welding head end pressure is 30 to 120 N.

[0158] The front end of the welding head features a tapered roller structure with a wave-shaped outer circumference. This ensures that when the material in the weld zone is compressed, a localized high-frequency molecular friction-fusion-solidification welding process occurs. This structure supports continuous track welding and is suitable for stable sealing of flexible fabric edges.

[0159] In order to ensure that the welding trajectory follows the edge of the cloth, the welding module is equipped with a welding head following trajectory device, which includes: a multi-axis slide support platform, an XY electric drive module, an edge recognition camera (or laser locator) and a roller-type dynamic pressure control feedback system.

[0160] Before welding, the system uses a contour recognition module to capture the contour of the overlapping edge of the fabric in real time, forming a visual welding path. During welding, the welding head's front-end recognition continuously monitors the actual direction of the fabric edge within a ±5mm area of the welding front. It feeds back trajectory deviations in real time to the XY platform, driving the welding head for dynamic compensation to ensure that it always aligns with the overlapping seam.

[0161] The system supports edge tracking error ≤1mm, and the weld width is stable between 2.5 and 3.0mm, meeting the mechanical strength and comfort requirements of the side seam sealing structure of underwear.

[0162] During welding, the welding head's pressurization is controlled by a solenoid valve or servo drive, with preload pressure set based on the fabric thickness, number of layers, and overlap width. To prevent localized overpressure and cracking in thin fabric areas, the welding head features a pressure buffer and micro-amplitude modulation mechanism, allowing for fine-tuning of the head's stroke to ensure uniform pressure distribution.

[0163] Welding energy (acoustic energy) output is monitored in real time by a central controller and integrated synchronously with the weld length, ensuring consistent welding energy per unit length for each weld. If frictional resistance or variations in fabric edge thickness are detected within a particular weld section, resulting in insufficient weld temperature rise, the system automatically reduces feed speed and increases instantaneous energy density to ensure weld quality.

[0164] After welding is completed, the welding head withdraws and the air cooling / compressed air channel is automatically activated in the welding base area to implement targeted rapid cooling of the welding area to prevent material rebound or adhesion.

[0165] After welding, the fabric is automatically fed into the post-weld stabilization output module, which features a lightweight smoothing brush and infrared sensor to check the weld line continuity and hem closure quality. If infrared recognition detects a broken or incomplete weld, the fabric is laser-marked and sent to the sideline for re-inspection.

[0166] The system can be equipped with an optional data acquisition terminal to record process parameters such as welding trajectory, pressure, energy, cooling time, etc. for each welded product for traceability and quality analysis.

[0167] This step achieves efficient, uniform and intelligent side seam welding and sealing of underwear, replacing the traditional stitch sewing process. It not only improves the production rhythm and appearance consistency, but also significantly improves the product's softness and wearing safety in the skin-contact area.

[0168] Step 8: Edge sewing, de-sewing and trimming

[0169] This step is used to hem the ultrasonically welded underwear pieces, trim the edges, and remove thread ends to ensure a complete appearance, neat edges, and burr-free edges, improving overall quality and wearing comfort. This step is performed by the hem sewing module 11, which primarily includes a recognition and positioning module, a sewing machine, a precision trimming device, a multi-axis tracking platform, a thread suction and collection device, and a residual material discharge mechanism.

[0170] After the panty sheet is hemmed and sewn, it is fed to the sewing station via conveyor line 12. The recognition and positioning module first captures the actual contour of the welded side seam area. This module includes a high-resolution industrial camera, a stripe laser projector, and an image processing and edge extraction unit.

[0171] The system automatically extracts the starting and ending points of the weld line, curvature changes, and boundary distances by collecting and analyzing images of the side seam area in real time, and establishes a deviation comparison chart with the standard weld template.

[0172] The module's recognition accuracy reaches ±0.3mm, and it can accommodate slight changes in weld contours under different sizes and fabric tensions, providing a dynamic basis for cutting path generation.

[0173] The trimming device, mounted on a multi-axis follower platform, utilizes a high-speed electric circular blade or ultrasonic cutter with a blade diameter of 30 to 60 mm and a cutting speed of 3,000 to 8,000 rpm. The device supports micro-cutting depth control, enabling precise removal of 1.5 to 2.5 mm of excess material outside the weld edge.

[0174] The system dynamically generates a cutting trajectory based on weld contour recognition data, controlling a multi-axis platform to drive the cutting head along the weld contour, ensuring the cut remains parallel to the edge. The platform utilizes a dual-servo XY track linkage drive mechanism to ensure that the cutting head movement matches the sheet transport speed in real time, ensuring a continuous cut without skipping.

[0175] To prevent the fabric from wrinkling when the cutter head is pressed down, a synchronous cloth pressing wheel and a side smoothing air nozzle are provided in front of the cutter to stabilize the fabric surface.

[0176] Residual thread, partially broken rubber bands, and burrs from the welding and pre-stitching processes are handled in this step using a thread suction and collection device. This device consists of a directional suction channel and a sliding thread comb. The channel ports cover both sides of the fabric and are positioned at the start and end points of the weld, creating a negative pressure suction zone.

[0177] The lint comb uses electrostatic fiber bristles that gently contact the fabric to remove the lint, improving lint collection efficiency. Collected material is introduced into the filter cartridge through the air duct and stored centrally for subsequent regular cleaning.

[0178] The cut fabric scraps are blown to the recovery trough by the scrap discharge mechanism through airflow, and a weight sensor is provided to monitor whether the amount of single piece scrap exceeds the limit, and to assist in identifying the risk of blade bluntness or cutting deviation.

[0179] After trimming is complete, a laser rangefinder and infrared light transmission verification module are installed at the end of the workstation to assess the trim width, flatness, and edge finish. If any signs of severe sawing, deviation from the weld seam, or residual thread are detected, the system will automatically sort the fabric piece and send it to a sideline for repair, preventing it from entering the packaging section.

[0180] The hemming sewing module operates in conjunction with the central control system. Parameters such as the hemming sewing path, cutting path, cutter speed, and airflow intensity are automatically associated with the current product model and recorded in real time in the process traceability system.

[0181] The system supports "workstation-level anomaly traceability". The image and inspection data of each piece of cloth after sewing are stored as a separate record, which is convenient for the quality control department to conduct batch review.

[0182] This step completes the edge forming and finishing of the underwear product, effectively improving the structural integrity and consumer acceptance of the product, and creating a good foundation for the next folding and packaging process.

[0183] Control system description

[0184] The entire production line is coordinated by a centralized PLC control system. Fabric picking rhythm, sewing path, elastic tension, welding trajectory, and trimming accuracy are all controlled by the system. The control system features real-time monitoring and fault self-diagnosis, providing alarms and interlocking shutdowns for conditions such as material shortages, thread breaks, and abnormal tension.

[0185] Furthermore, defect detection in current automated underwear production lines mostly relies on image-based static vision inspection systems, which primarily analyze two-dimensional images of the fabric surface to identify obvious defects such as uneven stitching, wrinkled fabric, or misaligned seams. However, this type of technology has the following problems:

[0186] It can only identify defects that have occurred, but cannot predict the location and trend of future defects.

[0187] The continuous deformation and subtle tension changes of cloth cannot be modeled;

[0188] In high-beat, high-speed continuous processing scenarios, missing the best time for correction will lead to increased rework and scrap rates.

[0189] To solve the above problems, see Figure 5 The present invention also includes a method for predicting and fine-tuning fabric defects. This method treats the fabric during the production process as a continuously evolving stress state system. By modeling the fabric's motion trends and local strain changes in image sequences, it forms a high-dimensional tensor flow field structure covering time and space. By extracting features and performing deep learning predictions on the perturbed evolution of this flow field, the system provides early warning of minor defects and implements active adjustments through a feedback control mechanism.

[0190] The system mainly consists of the following parts:

[0191] Industrial camera image acquisition module: deployed above and to the side of the fabric feeding path, used to continuously and comprehensively capture the surface image of the fabric during movement;

[0192] Image processing and edge computing unit: uses a GPU acceleration architecture to run the image optical flow extraction algorithm (multi-layer Lucas-Kanade in this embodiment), the surface morphology tensor construction module, and the inertial flow spectrum fitting module;

[0193] Stepper motor control feedback module: provides the speed, acceleration and real-time position status of the motor in the feeding mechanism to support dynamic modeling;

[0194] Central controller and depth prediction model: The central control unit integrates multimodal data from vision and motion, predicts the disturbance evolution trend based on LSTM or GRU networks, and provides judgment signals;

[0195] Execution control unit (PLC): According to the strategy signal output by the central controller, it dynamically adjusts the feeding rate, the tension of the cloth pressing mechanism and the trajectory of the sewing unit to achieve local fine-tuning;

[0196] Model training and learning module: Continuously receives detection results and correction feedback results, optimizes the prediction model based on differential learning, and improves the robustness and adaptability of the system.

[0197] The control steps include:

[0198] Step 1: Fabric unwinding and feeding: The fabric is fed from the fabric roll via a servo feeder, passes through a constant tension assembly, guide rollers, and a web-correcting unit, and is then fed into the image acquisition area to ensure the fabric surface is flat and maintain constant tension.

[0199] Step 2: Image acquisition: The industrial camera captures images of the fabric surface at 30 to 60 frames per second, and the edge computing unit processes them in real time to generate a complete image sequence.

[0200] The following are the specific implementation details of the image acquisition process:

[0201] Industrial camera layout: Multiple industrial cameras are installed above and on the sides of the fabric feeding area, usually high-definition color cameras or high-resolution industrial cameras, to ensure that all angles of the fabric can be captured. The cameras are usually arranged at a 45-degree staggered angle, so that the entire picture of the fabric can be fully acquired from different perspectives. As the fabric passes through the image acquisition area, the camera can simultaneously capture the details of the fabric surface, including the texture, color, gloss, and possible defects such as folds, wrinkles, and seam errors. The shooting frame rate of the industrial camera is set to 30 to 60 frames per second to ensure that the dynamic changes of the fabric can be accurately captured and every subtle change of the fabric can be reflected in real time.

[0202] Image Data Acquisition and Synchronization: Image data captured by each camera is transmitted in real time to the edge computing unit, using efficient transmission protocols to ensure efficient transmission and processing of image data. To prevent image blur or motion distortion caused by the rapid movement of the fabric, the system uses a multi-camera synchronization system to ensure that image data captured from different angles remains consistent in time, achieving precise pairing for subsequent processing and analysis.

[0203] Lighting and environmental adjustment: A stable, uniform light source (such as an LED) is used in the image acquisition area, and reflectors or diffuse reflectors are used to eliminate any possible shadows or light spots, ensuring that surface details of the fabric are clearly and evenly presented. The image acquisition system can automatically adjust the light intensity based on the fabric's material, color, or glossiness, ensuring that even fabrics with strong gloss or complex patterns can be accurately captured.

[0204] Image denoising and preprocessing: After image acquisition, the data undergoes denoising to remove noise caused by fabric surface imperfections, uneven lighting, or motion. Common denoising methods include Gaussian filtering and median filtering. Image preprocessing also includes contrast enhancement and edge enhancement to highlight surface details, particularly folds, wrinkles, and seams, making these defects more visible.

[0205] Image Sequence Construction: All collected image data is arranged in chronological order to form image sequences. These image sequences provide the basis for subsequent processing, such as optical flow analysis and motion trajectory prediction. Each frame is captured at predetermined intervals to ensure the coherence of the image sequence. This allows the system to capture dynamic changes in the fabric, ensuring that no potential anomalies or defects are missed.

[0206] Image Storage and Management: All captured image data is stored in real-time in a high-speed storage unit for subsequent image processing and analysis. This storage unit should have a large capacity to handle the large amounts of data collected on high-speed production lines. Furthermore, the system associates image data with information such as fabric position and speed based on the image acquisition timestamp, allowing subsequent analysis to accurately locate the fabric state corresponding to each image frame.

[0207] Data backup and fault tolerance: To prevent data loss due to equipment failure, image data will be regularly backed up to a backup storage device. The system should also have fault tolerance mechanisms, such as automatically switching to a backup camera or data path in the event of a network or camera failure, to ensure continuous and stable operation of the production line.

[0208] Through these implementation details, the system can capture the fabric surface state in real time during high-speed production, providing high-quality, accurate image data for subsequent image analysis and defect detection. This process is crucial for ensuring the accuracy of subsequent defect prediction and fine-tuning control.

[0209] Step 3: Optical Flow and Tensor Modeling: During the image acquisition phase, the fabric surface is captured at high speed and precision by an industrial camera. Next, the system uses the image processing unit to perform optical flow analysis and tensor modeling to extract the fabric's motion trajectory, deformation trends, and surface dynamic features. This process provides critical dynamic data support for subsequent defect prediction and fine-tuning control. The following are the specific implementation details of this step:

[0210] Optical Flow Analysis Overview: The optical flow method calculates pixel changes between adjacent frames in an image sequence to estimate the motion vector of each pixel in the image, that is, the movement trend of the object's surface. This method, based on the Lucas-Kanade optical flow algorithm, can track the movement of objects in consecutive image frames and extract displacement information on the fabric surface. It uses a multi-scale optical flow method, calculating motion vectors at different image scales and processing fabric deformations of different sizes to capture dynamic changes from microscopic to large scales. This enables the system to accurately identify local movement, wrinkles, stretching, and dynamic changes at seams in the fabric.

[0211] Implementation steps of the optical flow method: Preprocessing and image alignment: Before performing the optical flow calculation, the collected image is first denoised, and Gaussian filtering or median filtering is used to remove interference caused by lighting or equipment noise. In addition, an edge enhancement algorithm (Sobel operator) is used to improve the recognizability of fabric surface features. Based on the pixel grayscale changes between each two frames, the Lucas-Kanade method is used to calculate the motion vector of each pixel. The core of the optical flow method is to calculate the displacement of image points within a time interval based on the assumption of constant image brightness. By downsampling the image to different resolutions, multi-scale optical flow calculations are performed to capture motion patterns from subtle to large scales. In this way, the system can effectively handle changes in local details and overall structure of the fabric to ensure the comprehensiveness and accuracy of motion features.

[0212] Tensor modeling and feature extraction: Optical flow analysis provides information about the motion of the fabric surface in the temporal dimension, but motion vectors alone cannot fully describe the complex changes in the fabric surface morphology. Therefore, the system combines optical flow information with the spatial characteristics of the fabric and uses tensor flow modeling technology to further extract the deformation characteristics of the fabric surface in space and time. The tensor flow algorithm is used to model the fabric surface, combining the optical flow vectors with the spatial geometric characteristics of the fabric to form a high-dimensional tensor that describes the fabric's motion, deformation, and tension. This tensor contains dynamic information in multiple dimensions such as velocity, direction, acceleration, and curvature, and can fully express the spatiotemporal characteristics of the fabric. The tensor flow calculation uses a nonlinear least squares method based on the optical flow field, combined with the spatial distribution of the tensor, to describe the stress and deformation state of the fabric in different regions and at different time points.

[0213] Dynamic Change and Stress Analysis: After establishing the spatiotemporal tensor flow of the fabric, the system further identifies possible abnormal areas by performing stress analysis on local areas of the fabric. For example, by calculating the degree of deformation of the fabric (such as curvature and stretch ratio) and the difference in the velocity field, it can detect in advance areas of fabric that may be wrinkled, misaligned, or stretched. By comparing real-time image data with the tensor flow model, the system can accurately calculate the stress and deformation state of the fabric at a specific moment. These stress changes will provide basic data for subsequent defect prediction and fine-tuning control.

[0214] Fabric deformation trend prediction: Using the obtained optical flow vectors and tensor information, combined with a deep learning model (LSTM or GRU), the system performs time-series predictions of the fabric's future deformation trends. The system combines the optical flow features in the current image frame with historical motion patterns to predict the fabric's trajectory and potential deformation trends over the next few seconds. Through regression analysis of multiple frames of optical flow data, the system can accurately predict the fabric's future behavior. For example, it can predict whether the fabric will break due to overstretching or wrinkle due to uneven pressure.

[0215] Real-time processing and feedback of dynamic tensor data: During real-time analysis, the system compares the optical flow tensor of each frame with a standard inertial atlas, tracking changes in fabric surface details in real time. If an anomaly is detected, the system corrects the fabric's motion and issues appropriate intervention instructions. Data processing is entirely performed by the edge computing unit, ensuring real-time data transmission and processing. All data is immediately transmitted back to the central control unit for subsequent defect prediction and fine-tuning.

[0216] By combining optical flow analysis with tensor modeling, this step accurately captures the dynamic changes of fabric and extracts high-dimensional features of the fabric surface in both time and space, enabling the system to achieve early warning and precise control of fabric defects. This provides extremely precise input data for subsequent defect detection and dynamic adjustment, and is one of the core technologies of this invention.

[0217] Step 4: Inertia Atlas Construction: After extracting the dynamic features and deformation trends of the fabric surface during the optical flow and tensor modeling phase, the next key step is to construct an inertia atlas based on this data. The inertia atlas is a dynamic representation of the fabric's motion and stress state during the production process. It integrates the fabric's time-series motion data and spatial deformation characteristics to provide a comprehensive model of fabric behavior. This atlas provides the basis for subsequent defect prediction and fine-tuning control, ensuring precise adjustment of the fabric during production. The following are the specific implementation details of the inertia atlas construction process:

[0218] Inertial Atlas Overview: An inertial atlas is a high-dimensional data structure that describes the dynamic behavior of fabric in both time and space. It integrates fabric motion, stress, and deformation data to form a reference map reflecting the fabric's dynamic behavior. This map serves as a standard state in the model for subsequent comparison with real-time image data, anomaly detection, and defect prediction. The inertial atlas not only captures the fabric's motion trajectory but also takes into account its physical properties (such as tension, elasticity, and shape changes). In this way, the atlas comprehensively reflects the fabric's state changes during the production process.

[0219] Data fusion and feature extraction: The motion vectors (including speed, direction, acceleration, etc.) obtained by the optical flow method are combined with the displacement information of the fabric to construct the spatial dynamic data of the fabric. Each data point represents the motion state of the fabric surface at a specific point in time. Through the stepper motor control feedback module, the system obtains the actual tension change and position offset data of the fabric during the transmission process. Tension change data, as an important feature of the fabric stress state, can reflect the deformation that may occur in the fabric during stretching or compression. The multi-dimensional data such as motion, tension, and optical flow are fused, and the data is compressed into multiple key dimensions through feature extraction algorithms (such as principal component analysis PCA), reducing redundant data and retaining important dynamic features. These features will constitute the baseline structure of the inertial map.

[0220] Establishing a Steady-State Inertia Map: Initially, as fabric enters the production line, the system samples the fabric's dynamic behavior multiple times, collecting motion data under different operating conditions. This data is standardized and preprocessed to generate a "steady-state inertia map" for the fabric under ideal conditions. A steady-state map is a standard reference map of fabric surface motion when the fabric is free of any anomalies, normal load conditions, deformation, or defects. This map provides a benchmark for subsequent image data comparison and disturbance detection, serving as a reference for assessing fabric anomalies.

[0221] Dynamic inertia map update: As the production line runs, the dynamic state of the fabric may change. The system will adjust the inertia map in real time based on each frame of image and the corresponding motion data to make it more suitable for the fabric state in the current production environment. The dynamically updated inertia map can reflect the latest motion trends and morphological changes on the fabric surface, ensuring that subsequent defect prediction and correction are more accurate. In order to cope with different fabric types (such as fabrics with different elasticity, thickness, and materials), the system adjusts the inertia map through real-time data feedback. When the fabric type changes, the system will rebuild the inertia map according to the characteristics of the new fabric and quickly adapt to the operating parameters on the production line without the need to retrain the entire model, greatly improving the adaptability of the system.

[0222] Comparison of the inertial map with real-time data: The fabric image and motion data collected in real time are compared with the inertial map. This comparison not only detects static position changes in the fabric but also analyzes dynamic stress changes, morphological deformation, and other factors to see if they deviate significantly from the steady-state map. If the real-time data's motion trajectory, degree of deformation, or tension distribution differs significantly from the standard values in the inertial map (e.g., excessive acceleration or uneven deformation in a local area), the system identifies this as a potential fabric disturbance or defect. The system then marks the area as a "potential defect point" and proceeds to the next step of defect trend prediction.

[0223] Multi-level decomposition of the inertial map: In order to capture local defects more accurately, the inertial map can be further decomposed into multiple levels of map structures. For example, a local map is established separately in a specific area of the fabric (such as seams, folds, and curved areas) to capture the specific movement and deformation patterns of that area. Based on the global map, the local map can more accurately track small-scale local defects and minor deformations. The global map represents the overall behavior of the fabric, while the local map reflects subtle changes in the fabric in specific areas. The combination of the two ensures that the system can effectively balance the overall state of the fabric and local changes, and detect possible production problems in real time.

[0224] The role of inertial mapping in defect warning: By monitoring the movement and stress state of fabric in real time, inertial mapping can issue early warning signals before defects fully develop. For example, when fabric is overstretched or locally deformed, the system will identify and predict in advance the potential for defects such as cracks, misalignment, or folds in that area, facilitating timely correction. If the inertial mapping comparison detects an anomaly, the system immediately activates a fine-tuning control mechanism to adjust the fabric feed speed, pressure strength, and stitching path to prevent defects from forming.

[0225] By constructing and updating the inertial map in real time, the system can accurately capture the dynamic changes and deformation trends of the fabric surface, providing high-precision data support for defect prediction and fine-tuning control. This process plays a vital role in ensuring precise adjustment and defect prevention during the fabric production process.

[0226] Step 5: Real-time Comparison and Disturbance Detection: After completing fabric image acquisition, optical flow analysis, and inertial map construction, the system compares each frame of image data with the standard inertial map in real time. This step is the core of the fabric defect prediction and fine-tuning control system. It aims to detect deviations in the fabric's motion, shape, tension, and stress state to promptly identify potential disturbance areas and provide decision support for subsequent defect correction and fine-tuning control. The following is the detailed implementation details of Step 5:

[0227] Comparison of image data and inertial atlas: Through optical flow analysis and tensor modeling, the system has obtained the motion vector and deformation tensor of the fabric at each time point. Each frame of image data is converted into high-dimensional tensor data containing information such as motion, acceleration, and deformation. The system uses the pre-built inertial atlas (containing the normal motion pattern and morphological characteristics of the fabric) as a standard reference atlas for subsequent comparison of real-time data. After each new frame of image is collected, the system will use a local difference detection algorithm (such as correlation calculation, differential calculation, cross-correlation, etc.) to compare the characteristics of the current image data with the inertial atlas and calculate the deviation value. The larger the deviation, the more abnormal changes have occurred on the surface of the fabric, and potential defective areas may appear.

[0228] Disturbance Detection and Abnormal Area Marking: For each frame of image, the system not only compares the overall fabric movement trend but also analyzes changes in local areas. For example, if the optical flow direction in a certain area of the fabric is abnormal (such as a sudden change in speed or a reverse change), or if the edge changes in the local image exceed a predetermined threshold, the system will mark the area as a "potential disturbance area." The system also detects whether the local tension and deformation of the fabric match the expectations in the standard atlas. If the tension distribution, cloth pressing force, or folding in a local area exceeds the set threshold, the system will automatically identify it as a potential deformation area. For example, abnormal tension at a fabric seam may indicate the risk of seam misalignment or uneven stitching. In addition to real-time comparison, the system also dynamically tracks deviation areas based on historical data and analyzes their changing trends. For example, if a certain area experiences continuous slight deviations, the system will track its movement trajectory and predict whether the area is likely to become more serious in the future, thereby issuing a timely alarm.

[0229] Defect warning mechanism: Once a potential disturbance area is detected, the system will perform a time series analysis of the area through a trained deep learning prediction model (such as LSTM, GRU, etc.) to predict whether the disturbance will develop into an actual defect (such as cracks, wrinkles, seam dislocation, etc.) in the next few seconds. Based on the characteristics of the disturbance area, the system can not only determine whether there is a defect in the area, but also predict the type and severity of the defect. For example, if a local area of the fabric is overstretched, the system will predict that the area may break; if a local wrinkle appears in a certain area, the system will identify and determine the possible location of the defect in advance. When the system determines that the disturbance in a certain area is likely to develop into a defect, the system will issue a warning signal. At this time, the system will not only mark the area, but also generate relevant defect type, prediction time, defect level and other information for subsequent intervention and processing.

[0230] Real-time feedback and optimization: Once a potential defect or abnormal area is detected, the system will convert the prediction result into a control instruction and send it to the execution control unit (PLC). For example, if the system predicts that the deviation in the seam area may cause misalignment, the system will automatically adjust the trajectory of the sewing head, the pressure of the cloth pressing roller, or fine-tune the feed speed to avoid the occurrence of defects. Each abnormality detection and defect correction process will be recorded and fed back to the model training and learning module. By comparing the deviation between the prediction and the actual results, the system can continuously optimize the model in subsequent iterations and improve the accuracy of future predictions. As the production line continues to operate, the system will automatically adjust the reference standard of the inertial map according to the actual situation, so that the system's defect detection and correction capabilities will gradually improve. This adaptive mechanism enables the system to cope with changes in different fabrics and different production environments.

[0231] Visual Interface and Manual Intervention: To enable operators to monitor the status of fabric production in real time, the system displays each image frame and the corresponding defect detection results on the user interface. The system highlights potential defect areas and provides detailed defect information (such as type, location, and severity), facilitating quick decision-making. When the system detects a serious defect or anomaly, in addition to automatic control intervention, operators can also trigger manual intervention through the interface. For example, if the system detects an imminent irreversible defect in a specific location on the fabric, the operator can pause the production line and make manual adjustments to ensure production quality.

[0232] Through real-time comparison and disturbance detection, the system can quickly and accurately identify potential problems in the fabric production process. Through early warning, fine-tuning control, and manual intervention mechanisms, it significantly reduces the incidence of fabric defects, improving production line efficiency and product quality. This step is a core component of the entire fabric defect prediction and fine-tuning control system, ensuring the system can be precisely controlled in a dynamic production environment.

[0233] Step 6: Defect Trend Prediction: In the previous steps, the system has identified potential fabric defect areas and marked the dynamic changes in these areas through image acquisition, optical flow analysis, inertial atlas modeling, and disturbance detection. Next, defect trend prediction is a key step in the system. It aims to accurately predict the evolutionary trends of fabric defects based on historical and real-time data using deep learning models, allowing preventive measures to be taken in advance to prevent the expansion or worsening of potential defects. The following are the specific implementation details of defect trend prediction:

[0234] Defect Prediction Overview: Defect trend prediction primarily uses deep learning models (such as LSTM, GRU, and CNN-LSTM) to analyze real-time disturbance data and predict the evolution of potential defects over the next few seconds. The system utilizes a time series regression algorithm to learn the fabric's motion characteristics, deformation patterns, and stress trends in both historical and real-time states, inferring whether defects will manifest in the future. The core goal of this process is to predict whether a "potential defect area" will develop into an actual defect and determine the defect's type, severity, and timing.

[0235] Historical Data Accumulation and Model Training: The system continuously records disturbance data, defect data, and corresponding corrective actions during each production cycle, feeding this data into the model training and learning module. Through extensive historical data training, the deep learning model can be continuously optimized to accurately predict defects for different fabrics and under different production conditions. Initially, the system is trained using sample data to identify typical "defect patterns" for fabrics in various production processes. Over time, the system continuously accumulates data and gradually refines its defect trend prediction model, achieving increasingly higher prediction accuracy.

[0236] Real-time defect trend prediction: During real-time production, when the system detects a disturbance or potential defect, it feeds real-time image data, optical flow analysis results, tension change data, and historical motion pattern data into a predictive model for analysis. Deep learning models (such as LSTM) process the input time series data and, combined with the dynamic changes in the fabric over the past few seconds, make predictions for the short future (typically 3-5 seconds). These predictions can include determining whether the disturbed area will develop into specific defects such as cracks, misalignments, or wrinkles. They can also predict the morphological development of the defect, for example, whether it will become more severe or automatically correct itself over time. Accurately predicting that a defect will occur at a certain moment or time in the future.

[0237] Defect Level and Risk Assessment: The prediction model not only determines the presence of a defect but also assesses its severity based on factors such as the degree of deformation, stress changes, and optical flow anomalies in the area. For example, the system may categorize defects as "minor deviation," "medium risk," and "high-risk defect" for subsequent processing. By analyzing the prediction results, the system calculates the risk level of the defective area and, based on this level, determines whether immediate corrective action is necessary. For example, for high-risk defects, the system will automatically make fine-tuning adjustments to prevent further expansion; for minor risks, the system may implement a delayed adjustment or continue monitoring strategy.

[0238] Defect Evolution Path and Timeliness: The system maps the evolutionary path of defect areas over time, helping operators or line managers understand how defects develop. For example, they can determine how a potential wrinkle area will expand over time, and whether it will become more pronounced as production rates increase. The accuracy and timeliness of defect trend predictions are crucial. The system makes real-time adjustments and optimizations based on dynamic factors such as the fabric's actual speed, stretching state, and stitching path, ensuring that predictions accurately reflect defect evolution in the shortest possible time.

[0239] Feedback mechanism for prediction results: When the system predicts that a potential defect will develop into a serious defect within a short period of time, the control system immediately generates an intervention signal and automatically adjusts relevant production line parameters. Specific measures may include: If the system predicts excessive fabric tension, it will reduce the feed speed to slow fabric movement and prevent breakage caused by rapid movement. If the system predicts that the sewing path may deviate, it will fine-tune the trajectory of the sewing head to prevent seam misalignment. If the system predicts that the fabric surface may wrinkle, it will adjust the pressure of the pressing roller to maintain the fabric surface flatness.

[0240] Manual Intervention and Recommendations: When the system predicts complex defect trends or that automated fine-tuning cannot fully resolve them, it provides an early warning to the operator and offers suggestions for manual intervention. These suggestions include the location, steps, and estimated time required for possible manual intervention. The system provides an intuitive interface that displays the predictions and allows operators to make decisions based on real-time data to minimize production disruptions.

[0241] Model Optimization and Self-Learning Mechanism: The predictive model is continuously optimized during operation based on new real-time data and intervention results. Each comparison between predicted results and actual defects provides new data samples for model learning and updating, helping the system to more accurately identify future defect trends. The system uses adaptive optimization algorithms based on LSTM or other deep learning networks to adjust the learning rate and training weights based on real-time feedback, ensuring the long-term stability and efficiency of the predictive model.

[0242] By predicting defect trends, the system can proactively identify potential fabric defects, enabling effective control and intervention to reduce unnecessary rework and scrap, maximizing production efficiency and product quality. This step not only increases the system's automation level but also provides a powerful technical foundation for the prevention and control of fabric defects.

[0243] Step 7: Dynamic Control Adjustment: After defect trend prediction is complete, the system generates dynamic control signals based on the prediction results. The control unit (PLC) fine-tunes multiple key parameters in the production process to prevent potential defects. The purpose of dynamic control adjustment is to adjust the production line process through real-time feedback, prevent defects from escalating, and ensure the quality and stability of the fabric during production. The following are the specific implementation details of this step:

[0244] Control Objectives and Strategies: The primary goal of dynamic control adjustments is to quickly and effectively adjust key parameters in the fabric production process based on predicted defect areas to prevent defects from developing into actual problems. The core of this control strategy is to respond to defect predictions in real time and, through precise adjustments, ensure that fabric quality is not affected. Defects are reduced by adjusting multiple execution modules on the production line (such as feed speed, cloth pressing force, and stitching paths). Each control adjustment must minimize the occurrence of fabric defects without affecting production efficiency. This requires not only precise adjustments but also the ability to ensure smooth production and operation.

[0245] Key Parameter Control: During the fabric production process, multiple parameters affect final product quality, including but not limited to feed speed, pressing force, seam path, and tension. The system fine-tunes these parameters based on predictions. If the prediction model indicates significant tension variations or deformation trends in a particular area of the fabric, the system automatically slows the feed speed. This helps slow the fabric during transport, allowing it more time to relax naturally and reducing breakage or deformation caused by excessive stretching. When adjusting feed speed, the system also considers the fabric's material and thickness, ensuring that even under high-speed production conditions, feed speed adjustments do not affect overall production efficiency. If it predicts that certain areas of the fabric may wrinkle or misalign seams due to uneven tension or an uneven surface, the system fine-tunes this by adjusting the pressure of the pressing rollers. Increasing the pressing force helps flatten the fabric and prevent folding or wrinkling. These adjustments are fine-tuned based on the fabric's material, elasticity, and thickness. For example, for thin, wrinkle-prone fabrics, the system will moderately increase pressure; for thicker, more elastic fabrics, it will appropriately reduce pressure to avoid excessive compression. If the system predicts potential misalignment in the fabric's sewing path or potential deformation during the sewing process, it will automatically fine-tune the sewing head's trajectory. Adjusting the sewing path ensures that the fabric's seam position consistently meets predetermined standards, avoiding quality issues caused by misaligned seams. When adjusting the sewing path, the system considers the fabric's actual conditions (such as tension and elasticity) to ensure that the adjusted path does not cause unnecessary interruptions or stops during the sewing process. Tension control is a crucial factor in fabric production; excessively high or low tension can affect fabric quality. By adjusting the tension adjustment module, the system can fine-tune fabric tension based on predicted defect areas, ensuring ideal tension at each stage and preventing wrinkles or deformation caused by uneven tension.

[0246] Real-time feedback mechanism: The system not only provides early warnings during the defect prediction stage, but also requires real-time feedback during the actual adjustment process to ensure the effectiveness of the adjustment measures. The feedback mechanism includes the following aspects:

[0247] Real-time Monitoring: Adjusted production parameters are immediately fed back to the system via sensors, continuously monitoring dynamic data such as fabric tension and deformation. This feedback is used by the system to verify the effectiveness of the adjustments and ensure that no new defects appear in the adjusted fabric. Through real-time data feedback, the system can quickly determine the success of control adjustments. If the predicted defects are not effectively corrected, the system will continue to optimize control signals and further adjust the production process.

[0248] Closed-Loop Control: This process utilizes a closed-loop control system, meaning that feedback from each control adjustment is fed back into the system for secondary optimization. This closed-loop feedback allows the system to rapidly respond to any anomalies or changes, ensuring that any deviations in the fabric production process are corrected as quickly as possible. For example, if an adjustment causes a new problem in the fabric (such as a seam deviation), the system quickly adjusts the relevant control parameters to prevent further defects.

[0249] Multi-Strategy Joint Control: In some complex production environments, a single control strategy may not be sufficient to address all issues. In these situations, the system employs multi-strategy joint control to simultaneously adjust multiple parameters. For example, if tension changes and localized wrinkles occur during the sewing process, the system will coordinate feed speed, pressure, and sewing path to achieve optimal control. The system can simultaneously adjust multiple key control parameters to ensure that the fabric remains in optimal condition throughout the entire production process. Each control adjustment takes into account the fabric's characteristics and production conditions to ensure that other parameters are not affected.

[0250] Balancing Optimization Objectives with Production Efficiency: Dynamic control adjustments focus not only on improving fabric quality but also on the overall efficiency of the production line. During adjustments, the system optimizes control strategies based on production needs, enabling the production line to maintain a high and stable production pace while ensuring fabric quality. Each time a control parameter is adjusted, the system calculates its impact on production efficiency and strives to minimize the adjustment process to avoid downtime or interruptions caused by over-adjustments. Throughout the production process, the system continuously optimizes each adjustment step, achieving maximum quality improvement with minimal control action, ensuring a balanced balance between production efficiency and fabric quality.

[0251] Human Intervention and Manual Adjustment: While the system enables automatic control and fine-tuning, it also provides an interface for human intervention in complex or unpredictable situations. When a predicted defect is too complex or cannot be fully resolved through automatic adjustments, the system will issue a warning to the operator through a visual interface and provide appropriate adjustment suggestions. The operator can then make manual adjustments based on the prompts to further optimize the production process.

[0252] Through dynamic control adjustments, the system can precisely adjust multiple key production process parameters (such as feed speed, cloth pressing force, and sewing path) based on defect trend predictions, effectively preventing and correcting potential defects. This process ensures refined management of the fabric production process, improving product quality while continuously optimizing production efficiency.

[0253] Step 8: Data Feedback and Model Optimization: During the fabric production process, every defect detection, prediction, and fine-tuning step generates a massive amount of data. This data is not only real-time information from the production line, but is also crucial for subsequent model optimization, system adaptation, and improved prediction accuracy. The key to data feedback and model optimization is to feed real-time production data back to the training module, thereby enhancing the system's learning capabilities. This allows the model to continuously adjust and optimize based on new data, thereby improving the accuracy of subsequent defect predictions. The following are the specific implementation details of this step:

[0254] Real-time data feedback mechanism: Throughout the production process, the system continuously collects real-time data from various modules (such as industrial cameras, stepper motor controllers, sensors, and execution control units). This data includes fabric images, motion trajectories, stress distribution, tension changes, deformation levels, control adjustment signals, and results. All of this real-time data is transmitted via a network system to a central control unit, which then integrates and synchronizes it to ensure data integrity and time consistency. Especially during real-time control adjustments, the system transmits back control parameters, adjustment measures, and subsequent feedback information at each moment, forming a complete data chain.

[0255] Real-time Feedback and Model Updates: Each time the system detects a potential defect and makes fine-tuning adjustments, the operator or automated system records the results of the adjustments. This includes the fabric state before and after the adjustment, whether the anticipated defect has been resolved, and whether any new issues have arisen during the adjustment process. This data allows the system to understand the effectiveness and limitations of the current adjustment strategy. This real-time feedback is fed back to the deep learning training module and used to update and optimize the model. Through supervised learning, the system compares the actual corrected results with the predicted defect results and calculates the deviation. These deviations serve as training data to continuously optimize the weight parameters in the prediction model.

[0256] Deep Learning Model Training: After each defect prediction and fine-tuning adjustment, the system continuously transmits new data, specifically the original data at the time of the defect, perturbation detection data, and adjustment results, for online training. This process allows the system to update the existing model through incremental learning, without requiring a complete retraining. This approach enables the system to continuously adapt to new fabric types, new production environments, and new defect manifestations during production. As production progresses, more sample data is accumulated, including defect characteristics for different fabrics and the effects of adjustments under different production conditions. Based on this new sample data, the system continuously expands the training set and enhances the model's learning capabilities.

[0257] Adaptive Mechanism and Online Learning: Over time, fabric types and production environments may change. For example, fabric characteristics such as material, thickness, elasticity, and texture may vary, potentially affecting the appearance and occurrence of defects. To address this, the system incorporates adaptive learning capabilities, automatically adjusting model parameters and strategies based on the characteristics of different fabrics. When the system detects a change in fabric type, it automatically switches to a learning mode tailored to the new fabric characteristics, rapidly updating the inertia map and prediction model using a small amount of sample data to ensure adaptability and stability. The system gradually updates the deep learning model through online learning, enabling it to dynamically respond to changing production conditions and fabric characteristics. For example, by continuously accumulating sample data from different batches, design styles, and production runs, the system can adjust the prediction model in real time to improve prediction accuracy.

[0258] Model Optimization and Accuracy Improvement: During each data feed and training process, the system uses a loss function (such as mean squared error (MSE) or cross-entropy loss) to measure the difference between the predicted value and the true value. By optimizing the loss function, the system adjusts the weights and parameters in the network to make the prediction model more accurate. Optimization algorithms such as gradient descent or the Adam optimizer are used to reduce the error between the predicted and actual results. With each defect prediction and fine-tuning control, the system performs incremental optimization based on the new data. This means that each optimization process is based on the previous training, reducing training time and improving the model's responsiveness in actual production. Incremental learning also avoids overfitting and ensures the model's generalization capabilities under different production conditions.

[0259] Step 9. Graph Adaptation and Update: During the fabric production process, fabrics from different batches, types, or process parameters have significant differences in their physical properties (such as elasticity, thickness, texture, etc.). To ensure that the defect prediction system can operate stably and efficiently in a diverse fabric environment, this system rapidly builds and iteratively updates the inertia tensor graph to adapt to the process of different fabrics without completely retraining the model. The following is the specific method:

[0260] Fabric Property Identification and Classification: As each new batch of fabric arrives online, the system automatically collects key physical parameters of the fabric, such as stretch, resilience, thickness, surface texture, optical reflectivity (which influences image acquisition), and tension response characteristics (automatically extracted through force feedback during feeding). This data is used to automatically classify the fabric and provide a classification basis for atlas generation.

[0261] Rapid inertial map construction: The system calls the preset basic inertial map template under the category to which the fabric belongs. In the early stages of production (such as the first 10 to 30 seconds), the system collects the fabric's real-time optical flow data and stress tensor, and combines them with the template for rapid fitting. It generates an adaptive inertial map for the fabric under the current working conditions, which serves as a benchmark reference map for defect detection and disturbance identification.

[0262] Dynamic update mechanism of the map: During the production process, if the system detects that the dynamic behavior of the fabric deviates continuously from the current map (such as exceeding the set offset tolerance threshold), the map update logic is triggered: the current motion data is integrated with the historical tensor trajectory; the tensor features are updated through weighted averaging or sliding windows to reconstruct the "steady-state inertia map"; and online adaptive map evolution is achieved to avoid misjudgments and missed detections.

[0263] Lightweight atlas switching and version caching: The system maintains an atlas cache library for common fabric types; if the production line switches to a fabric that has been used in the past, the system can directly load its historical atlas version without remodeling; for new fabrics, only a small amount of data needs to be sampled to generate a customized atlas, greatly reducing initialization overhead.

[0264] Linked adaptation of process parameters: Graph updates simultaneously drive dynamic matching of process parameters. For example, the pressure of the pressing roller is automatically adjusted to adapt to more elastic fabrics; the sewing speed curve is adapted to the dynamic stability index in the graph; and the seam position is fine-tuned to correct for the inertial offset of different fabrics during the sewing process.

[0265] Operation interface prompts and manual correction suggestions: If the pattern evolution is abnormal (such as excessive fluctuations or abnormal update frequency), the system will issue a warning on the interface, prompting the operator to check the fabric tension, temperature and humidity or equipment status; it supports manual intervention to adjust the pattern parameters or lock the current pattern, which is suitable for small batches of special process fabrics.

[0266] This system uses single-channel, high-dimensional modeling based on industrial camera image data, eliminating the need for fabric contact sensors. By integrating inertial flow spectrum fitting, disturbance trend learning, and dynamic process feedback into a closed-loop control chain, it enables proactive prediction, precise fine-tuning, and adaptive scheduling in continuous, high-speed fabric production. This system has broad application value in the production of diverse styles, multiple fabrics, and high-speed garments.

[0267] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. An automatic underwear production line, characterized in that: Including the following settings: An automatic cloth roll feeding system (1) is used for continuously conveying cloth piece 1 (front piece) and cloth piece 2 (crotch bottom); An automatic cutting bed (2) for cutting the first and second pieces of cloth; The cloth piece grabbing mechanism (3) comprises two sets of manipulators, which respectively grab the cloth piece 1 and the cloth piece 2 and stack them as required; Sewing module group 1 (4) and sewing module group 2 (5) are used to sew the overlapping cloth pieces together to form a crotch cloth piece unit under the positioning and clamping of the jig; A leg rubber band sewing control unit (7) comprises a rubber band feeding mechanism (7a) and a rubber band sewing mechanism (7b) for sewing the leg rubber band to the leg edge of the crotch cloth unit; A waist rubber band sewing module (8) is used to sew the waist rubber band to the waist opening of the crotch cloth unit through a waist feeding mechanism (8a); A cloth piece folding and sewing module (9) comprises a folding manipulator (9a) and a sewing device (9b), which is used to fold the cloth piece with the rubber band sewn on it in half and seal the edges to form a trouser shape; An ultrasonic welding module (10) is used to weld and fix the edges of the folded trouser pieces; An edge sewing module (11) is used for sewing the excess cloth edge after welding; The various workstations are connected in series via a synchronous conveyor line (12), and the rhythm is coordinated uniformly by a control system.

2. The automatic underwear production line according to claim 1, characterized in that: The two sets of manipulators in the cloth piece grabbing mechanism (3) are both provided with adsorption ends, which can respectively grab the cloth piece one and the cloth piece two and release them onto the fixture synchronously.

3. The automatic underwear production line according to claim 1, characterized in that: The sewing module group 1 (4) and the sewing module group 2 (5) are respectively arranged at parallel workstations, and the cloth pieces are driven by the jig to enter the two sewing mechanisms for sewing synchronously.

4. The automatic underwear production line according to claim 1, characterized in that: The rubber band feeding mechanism (7a) can synchronously feed the rubber band according to the beat of the cloth piece, and the rubber band sewing mechanism (7b) sews the rubber band and cuts off the excess part of the edge.

5. The automatic underwear production line according to claim 1, characterized in that: The folding manipulator (9a) in the cloth piece folding and sewing module (9) has a multi-joint structure and cooperates with the folding clamp to fold the trouser piece along the central axis, and the sewing device (9b) performs edge seam processing.

6. The automatic underwear production line according to claim 1, characterized in that: The ultrasonic welding module (10) performs auxiliary welding.

7. The automatic underwear production line according to claim 1, characterized in that: The hemming sewing module (11) comprises an identification module and a cutting head hemming mechanism, and performs precise hemming sewing according to the edge contour.

8. The automatic underwear production line according to any one of claims 1 to 7, characterized in that: The control system is a centralized PLC controller that can realize dynamic scheduling and alarm joint control of the entire process modules of cloth grabbing, rubber band feeding, folding and welding, and beat synchronization.

Images