Glass fiber filament production broken filament early warning system and method based on machine vision

Abstract

The invention discloses a glass fiber filament production broken filament early warning system and method based on machine vision, and the system comprises an industrial camera module which is used for collecting a visual image in a glass fiber filament production process and obtaining the morphological characteristics of a filament body, and the morphological characteristics comprise the diameter of the filament body, surface scratches and a motion trail; the optical fiber sensor module is arranged below the glass fiber filaments and is used for detecting light transmittance fluctuation data of filament bodies; and the edge calculation unit is electrically connected with the industrial camera module and the optical fiber sensor module and is used for receiving visual image data transmitted by the industrial camera module in real time. Through machine vision, optical fiber light transmittance and a dual-mode detection framework, and in combination with dynamic weight adjustment logic, when dust shielding or light intensity fluctuation exists in a workshop, the visual image data can be displayed in real time; the system can automatically adjust the weight ratio of visual detection and light transmittance detection, and the visual information is compensated by using light transmittance data, so that the defect missed judgment caused by environmental interference on a single mode is avoided.

Description

Technical Field

[0001] This invention belongs to the field of machine vision and glass fiber filament production technology, and particularly relates to a machine vision-based glass fiber filament breakage early warning system and method. Background Technology

[0002] In the production of glass fiber filaments, filament breakage leads to raw material waste, production line downtime, and reduced time spent on subsequent filament drawing. Therefore, filament breakage early warning is a crucial step in ensuring production continuity and economic efficiency. Current mainstream filament breakage early warning solutions rely on industrial camera vision inspection. By acquiring filament images and analyzing morphological features such as diameter changes and surface scratches, they identify potential precursors to breakage. Some solutions supplement visual information by incorporating a single type of sensor (such as a tension sensor). For different batches of glass fiber filaments (e.g., filaments of different materials and diameters), current technologies typically require manual labeling of numerous samples and training of dedicated detection models to adapt to batch characteristics. In multi-filament parallel production scenarios, tracking and positioning rely heavily on the filament's positional coordinates to ensure basic monitoring of individual filaments and meet basic early warning requirements in conventional production environments. Existing technologies are ill-suited to the complex scenarios and flexible requirements of glass fiber filament production, exhibiting several limitations: In terms of interference resistance, pure visual inspection is susceptible to dust and light intensity fluctuations in the workshop, leading to decreased image clarity and reduced defect recognition accuracy. Single-sensor-assisted solutions cannot achieve data complementarity and struggle to cope with environmental changes. Regarding batch adaptation, reliance on numerous manually labeled samples and lengthy model training results in significant equipment downtime during new batch switching, hindering rapid response to the multi-batch, small-volume production demands of flexible manufacturing. Multi-filament tracking, relying solely on position coordinates, is prone to ID confusion when filaments intertwine or obstruct each other, leading to incorrect warning positioning and increased risk of unintended shutdowns. Warning mechanisms are often simple "single shutdown / prompt" response modes, unable to flexibly output processing strategies based on defect severity, resulting in either excessive downtime impacting production efficiency or missed defects leading to filament breakage losses. Furthermore, some solutions suffer from poor hardware compatibility, requiring customized equipment and making integration with existing production line control systems difficult, resulting in high costs and challenges in widespread application. Summary of the Invention

[0003] In order to overcome the above-mentioned defects of the prior art, the present invention provides a glass fiber filament breakage early warning system and method based on machine vision, which solves the problems of poor anti-interference, slow adaptation, mixed positioning, rigid response and low compatibility in the prior art for glass fiber filament breakage early warning.

[0004] To achieve the above objectives, the present invention provides the following technical solution: A machine vision-based glass fiber filament breakage early warning system includes: An industrial camera module is used to acquire visual images during the glass fiber filament production process and obtain the morphological characteristics of the filament, including filament diameter, surface scratches, and motion trajectory. The fiber optic sensor module is located below the glass fiber filament and is used to detect the transmittance fluctuation data of the filament. The edge computing unit is electrically connected to the industrial camera module and the fiber optic sensor module, respectively, and receives visual image data transmitted from the industrial camera module and transmittance fluctuation data transmitted from the fiber optic sensor module in real time. The edge computing unit has the following built-in features: The identification number generation module is used to assign a unique identification number to each fiber when a new batch of fiberglass filaments is put into production, and to associate and store the identification number with the exclusive characteristics of the batch of filaments. The exclusive characteristics include the light absorption coefficient, surface reflectivity threshold and light transmittance benchmark value corresponding to the filament material. The dual-modal detection module, with machine vision inspection at its core, combines transmittance data from fiber optic sensors to extract defect features of the filament. These defect features include diameter abrupt changes ≥0.5μm, surface scratches ≤1μm, and transmittance fluctuations exceeding batch-specific benchmark values. The module also adjusts the weighting of visual inspection and transmittance detection based on environmental parameters, including dust concentration and light intensity, with a dust concentration threshold of 0.5mg / m³. 3 The threshold for judging light intensity fluctuations is 10%. The batch adaptation module is used to learn and adapt to new batches of yarn through a small number of samples. It only needs to collect 50-100 visual images and transmittance data samples with identification numbers to complete the adaptation of defect judgment standards. During the production process, every 1000m of yarn is produced, 10 unlabeled samples are automatically collected to update the judgment standards. The early warning execution unit is electrically connected to the edge computing unit and is used to output graded early warning signals based on the defect level and the silk trajectory offset corresponding to the identification number output by the dual-modal detection module.

[0005] Preferably, a machine vision-based method for early warning of fiber breakage in glass fiber production includes the following steps: S1: When a new batch of glass fiber filaments is put into production, visual images of the filaments are captured by an industrial camera module. The identification number generation module assigns a unique identification number to each filament and stores the identification number in association with the exclusive characteristics of the batch of filaments. The exclusive characteristics include light absorption coefficient, surface reflectivity threshold and light transmittance benchmark value. S2: During the production process, the industrial camera module acquires real-time visual images of the filament and extracts the morphological features of the filament, including diameter, surface scratches and motion trajectory. The fiber optic sensor module simultaneously acquires the light transmittance fluctuation data of the filament and transmits them to the edge computing unit. S3: The dual-modal detection module uses morphological features extracted by machine vision as its core, combined with transmittance fluctuation data, to determine whether there are defects in the filament. When the real-time dust concentration in the workshop is ≤0.5mg / m³ 3 Furthermore, when the real-time light intensity fluctuation is ≤10%, the visual inspection weight accounts for 70%-80%, the transmittance inspection weight accounts for 20%-30%, and the focus is on judging the diameter change and surface scratches through morphological features. When the real-time dust concentration in the workshop is >0.5mg / m³ 3 When the real-time light intensity fluctuation is greater than 10%, the transmittance detection weight accounts for 60%-70%, and the visual inspection weight accounts for 30%-40%. The focus is on judging defects by whether the transmittance fluctuation exceeds the batch-specific benchmark value. S4: When a new batch is launched, the batch adaptation module quickly adapts the defect judgment standard based on 50-100 visual images and transmittance data samples; during the production process, every 1000m of yarn is produced, 10 unlabeled samples are automatically collected to update the judgment standard and adapt to the characteristic changes within the yarn batch. S5: The early warning execution unit outputs a graded early warning based on the defect level output by the dual-modal detection module and the silk trajectory offset corresponding to the identification number. The defect levels include minor, general, and severe. A shutdown signal is triggered when the defect level is severe and the trajectory offset is greater than 0.5mm. When the defect level is general and the wire spacing is less than the safety spacing, the safety spacing is 2-3 times the standard diameter of the wire in this batch, and an operation terminal prompt signal is output. When the defect level is minor and the duration is ≥30s, a production parameter adjustment prompt signal is output.

[0006] Preferably, the industrial camera module has a resolution of ≥1920×1080, a frame rate of ≥25fps, and a lens focal length of 8-16mm to meet the morphological feature extraction requirements of filament diameters of 5-20μm.

[0007] Preferably, the detection wavelength of the fiber optic sensor module is 532nm, the transmittance detection accuracy is ≤0.1%, the response time is ≤1ms, and the transmittance benchmark value in the batch-specific characteristics is: 92%-95% for E glass fiber and 88%-91% for AR glass fiber.

[0008] Preferably, the batch-specific characteristics associated with the identification number generation module also include the toughness parameter of the filament. The toughness parameter is associated with the material type in the batch-specific characteristics, wherein the toughness parameter of AR glass fiber filament is ≤2.5GPa, and the toughness parameter of E glass fiber filament is ≥3GPa. When judging defects, the dual-modal detection module adjusts the defect threshold in combination with the toughness parameter: when the toughness parameter is ≤2.5GPa, the diameter mutation threshold is lowered to 0.3μm, and the transmittance fluctuation threshold is lowered by 10%-15%.

[0009] Preferably, in step S1, the associated storage information of the identification number also includes the initial oscillation frequency of the filament, which is associated with the material type of the batch of filaments, wherein the initial oscillation frequency of AR glass fiber filament is 1-2Hz and the initial oscillation frequency of E glass fiber filament is 2-3Hz; in step S5, when the difference between the actual oscillation frequency of the filament and the initial oscillation frequency is >0.5Hz, the early warning execution unit increases the judgment weight of the defect level.

[0010] Preferably, in step S3, when the dual-modal detection module determines a diameter abrupt change, it uses multi-frame image superposition and comparison. When a diameter change ≥ 0.5 μm is detected in three consecutive frames, it is determined to be a diameter abrupt change defect, thus avoiding misjudgment caused by noise in a single frame image.

[0011] Preferably, in step S4, the batch adaptation module uses the MAML algorithm to complete the rapid adaptation, with ≤15 iterations and an adaptation time of ≤30 minutes, without the need for manual labeling of sample defect types.

[0012] Preferably, in step S5, the production parameter adjustment prompt signal includes drawing speed adjustment and cooling air temperature adjustment. The drawing speed adjustment range is ±5%, and the cooling air temperature adjustment range is ±2℃. The specific adjustment parameters are determined according to the material type in the batch-specific characteristics: when the material is AR glass fiber, the drawing speed adjustment range is ≤3%; when the material is E glass fiber, the cooling air temperature adjustment range is ≤1.5℃.

[0013] Preferably, the edge computing unit also has a built-in trajectory prediction module. Based on the historical movement trajectory of the silk corresponding to the identification number, the Kalman filter is used to predict the trajectory in the next 0.5-1 seconds. When the distance between the predicted trajectory and the adjacent silk is less than the safety distance, the safety distance is 2-3 times the standard diameter of the batch of silk, and the warning signal of the early warning execution unit is triggered in advance.

[0014] The technical effects and advantages of the machine vision-based glass fiber filament breakage early warning system and method of the present invention are as follows: 1. This invention utilizes a dual-modal detection architecture combining machine vision and fiber optic transmittance, along with dynamic weight adjustment logic, to effectively address the low accuracy of traditional pure visual early warning systems in complex environments such as dust and light fluctuations. When dust obstructs the view or light intensity fluctuates in the workshop, the system automatically adjusts the weight ratio of visual detection and transmittance detection, using transmittance data to compensate for visual information. This avoids missed or false detections of defects caused by environmental interference in a single modality, ensuring the stability and reliability of fiber breakage precursor recognition across the entire production process. 2. This invention relies on a batch adaptation module based on the few-shot learning (MAML algorithm). This solution eliminates the need for a large number of labeled samples and lengthy model training required by traditional technologies. Only a small number (50-100 images + transmittance data) of "visual images + data" are needed to complete the adaptation of a new batch of glass fiber filaments, significantly shortening the adaptation time. Simultaneously, during production, the system can automatically collect unlabeled samples to update defect judgment criteria, adapting to minor changes in filament characteristics within a batch. This greatly reduces equipment downtime during batch changeovers, improves production line flexibility, and minimizes production capacity losses due to batch adaptation. 3. This invention uses a multi-filament management logic of "identity number + characteristic binding". The system assigns a unique identity number to each filament and associates it with its exclusive characteristics. Even if multiple filaments are intertwined or obscured, the identity of the filament can be confirmed through characteristic verification (rather than relying solely on position), avoiding ID confusion. Combined with the trajectory prediction module's prediction of the filament's movement trajectory, the risk of frictional filament breakage caused by filaments being too close together can be identified in advance, significantly improving the positioning accuracy of filament breakage warning in multi-filament parallel production scenarios and reducing false shutdowns and missed warnings caused by multi-filament intertwining. 4. This invention employs a graded early warning mechanism based on "defect level + trajectory deviation," which outputs targeted early warning signals according to the severity of the precursors to wire breakage (minor, moderate, severe): severe defects trigger rapid shutdown to prevent the wire breakage from escalating; moderate defects prompt focused monitoring to balance production efficiency and risk control; and minor but persistent defects prompt parameter adjustments to prevent wire breakage at its source. Simultaneously, early identification of precursors to wire breakage (based on micro-defects and trajectory anomalies) avoids the waste of raw materials and time required for re-threading after wire breakage, reducing economic losses during production. 5. The core hardware of this invention (industrial camera, fiber optic sensor, edge computing unit) all adopt mature industrial-grade equipment, without the need for customized special components. It can be directly connected to the control system of existing glass fiber production lines, with low modification difficulty and controllable cost. At the same time, the system's adaptability to glass fiber filaments of different materials and diameters allows it to cover most large-scale glass fiber filament production scenarios, and has broad application and promotion value. 6. This invention is not a simple superposition of modules, but forms a complete technical closed loop of "identity number associated characteristics - dual-modal detection adaptation environment - batch learning dynamic optimization - graded early warning implementation": the identity number provides characteristic benchmarks for dual-modal detection, dual-modal data provides sample support for batch adaptation, and batch adaptation results feed back into the accuracy of defect judgment. The synergistic effect of each link fundamentally solves the problem of traditional early warning technology. Attached Figure Description

[0015] Figure 1 This is a system flowchart of a machine vision-based glass fiber filament breakage early warning system and method proposed in this invention; Figure 2 This is a flowchart of a machine vision-based glass fiber filament production breakage early warning system and method proposed in this invention. Detailed Implementation

[0016] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0017] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0018] refer to Figure 1-2 This invention proposes a machine vision-based glass fiber filament breakage early warning system and method. It is based on a large-scale glass fiber filament production line (8-32 filaments produced in parallel), with a workshop configuration of temperature 25±5℃ and humidity 40%-60%. The core hardware adopts mature industrial-grade equipment and can be directly connected to existing production control systems (such as PLC). The test indicators include: early warning accuracy (breakage precursor recognition rate), false alarm rate (false trigger rate in non-breakage scenarios), new batch adaptation time, and anti-interference ability (performance stability under dust / light fluctuations). Example 1 This embodiment provides a machine vision-based glass fiber filament breakage early warning system and method for single-batch, multi-filament production of E-glass fiber filaments (10μm in diameter) under normal environmental conditions. Specific implementation details include: Purpose of implementation: Verify the accuracy of the system in identifying the precursory signs of multifilament breakage in E-glass under normal conditions (low dust, stable light) and the effectiveness of graded early warning. System Architecture: Industrial camera module: Hikvision MV-CA020-10GM, resolution 1920×1080, frame rate 30fps, lens focal length 12mm, installed 50cm directly above the filament, equipped with white ring LED light source (illuminance 500 lux), the main function is to acquire visual images of the filament and extract the morphological features of the filament, including the filament diameter, surface scratches and motion trajectory. Fiber optic sensor module: Keyence FS-V21R is selected, with a detection wavelength of 532nm, a transmittance detection accuracy of 0.05%, and a response time of 0.5ms. It is installed 30cm directly below the filament and aligned with the center of a single filament, and is responsible for detecting the transmittance fluctuation data of the filament. Edge computing unit: Utilizing NVIDIA Jetson AGXXavier (running Ubuntu 20.04 and PyTorch 1.12 framework), it incorporates three core functional modules: First, an identification number generation module, which can assign unique IDs (ID1-ID8) to 8 filaments and associate them with batch-specific characteristics; second, a dual-modal detection module, which can allocate weights for visual inspection and transmittance inspection based on environmental parameters, while simultaneously extracting filament defect features; and third, a batch adaptation module, which can automatically collect unlabeled samples to update defect judgment thresholds.

[0019] Early warning execution unit: adopts Siemens S7-1200 PLC, connects to the workshop audible and visual alarm (supports yellow / red flashing) and the production line stop button, and its main function is to output graded early warning signals. Implementation steps: S1 (ID Generation): Before the new batch of E glass filaments goes online, the industrial camera continuously captures 10 clear images of the filaments (unobstructed and unblurred). The ID generation module of the edge computing unit clusters the filaments by "filament position + initial diameter" and assigns unique IDs (ID1-ID8) to the 8 filaments. At the same time, it reads the batch process file and binds "light absorption coefficient 0.8, surface reflectivity threshold 30 (grayscale value), light transmittance baseline value 93%, toughness parameter 3.2GPa" with the IDs and stores them to form an "ID-batch characteristics" mapping table. S2 (Real-time Data Acquisition): After the production line starts (drawing speed 12m / min), the industrial camera acquires 30 frames of images per second. Through the image preprocessing algorithm (Gaussian filtering for noise reduction + edge detection) of the edge computing unit, the morphological features of each filament are extracted in real time: the diameter is stable at 10±0.2μm, the surface scratches are ≤0.8μm, and the oscillation frequency of the motion trajectory is 2.5Hz. The fiber optic sensor synchronously acquires transmittance data and outputs one value every 10ms, with the fluctuation range controlled within ±0.5%. All data are transmitted to the edge computing unit in real time. S3 (Dual-modal detection): The workshop environmental monitoring sensor reports a real-time dust concentration of 0.3 mg / m³. 3 The edge computing unit's dual-modal detection module automatically allocated weights for the "real-time light intensity fluctuation of 5%": visual detection accounted for 75% (focusing on diameter abrupt changes and surface scratches), and transmittance detection accounted for 25% (assisting in defect verification). At the 15th minute of operation, the module detected that the diameters of three consecutive frames of ID3 filament images were 10.1μm, 10.5μm, and 10.6μm respectively (diameter abrupt change of 0.6μm), and the transmittance fluctuated by 0.8% synchronously (not exceeding the baseline value of 93%±1%), which was comprehensively judged as "general defects". S4 (Batch Adaptation): This batch consists of continuously produced E-glass filaments with no changes in batch characteristics, so no re-adaptation is required. For every 1000m of filament produced (approximately 83 minutes), the batch adaptation module of the edge computing unit automatically collects 10 unlabeled samples (including images of each ID filament and transmittance data), and updates the "diameter mutation judgment threshold" through self-supervised learning. Because the toughness parameter is stable (3.2GPa), the threshold remains unchanged at 0.5μm. S5 (Graded Early Warning): The edge computing unit transmits the "defect level (general) + trajectory offset (0.3mm)" of the ID3 filament to the early warning execution unit; the execution unit queries the "early warning rule table" and outputs the operation terminal prompt signal: the workshop audible and visual alarm flashes yellow, the central control screen displays "ID3 filament diameter is abnormal, it is recommended to pay close attention", and does not trigger a shutdown (to avoid excessive intervention). Implementation results: Technical specifications: Early warning accuracy rate 99.5% (accurately identified 100 signs of yarn breakage), false alarm rate 0.2% (only 2 cases were misjudged due to minor impurities on the yarn surface, which were corrected after manual intervention); Production benefits: By identifying ID3 filament defects 1.8 seconds in advance, the waste of raw materials due to re-threading after filament breakage is avoided (the loss of 500 yuan per filament breakage is about 500 yuan), reducing waste by 3 times per day and saving about 45,000 yuan per month. Example 2 This embodiment provides a machine vision-based glass fiber filament breakage early warning system and method for single-batch, multi-filament production of AR glass fiber filaments (8μm in diameter) in a dusty environment. Specific implementation details include: Purpose of implementation: The study verified that "under dust interference environment, the system can achieve accurate early warning of multi-filament AR glass through dynamic adjustment of dual-modal weights". System Architecture: Industrial camera module: Same as Example 1 (Hikvision MV-CA020-10GM), with the addition of a lens dust cover (to deal with dust) to ensure image clarity ≥90%, and the function remains to acquire visual images and extract morphological features. Fiber optic sensor module: Same as in Example 1 (Keyence FS-V21R), except that the transmittance detection accuracy is adjusted to 0.03% (because AR glass has lower transmittance, higher sensitivity is required), and its function is to detect transmittance fluctuation data. Edge computing unit: Same hardware as in Example 1 (NVIDIA Jetson AGXXavier), the dual-modal detection module adds a "dust concentration-weight mapping table", and the batch adaptation module pre-stores AR glass-specific characteristics to ensure rapid retrieval of detection rules. Early warning execution unit: Same as in Example 1 (Siemens S7-1200PLC), with the addition of a "stop signal delay trigger" function (300ms buffer to avoid accidental stop), which outputs graded early warning signals.

[0020] Implementation steps: S1 (ID Number Generation): When the AR glass filament is put into operation, the industrial camera captures 8 frames of images, and the edge computing unit assigns ID9-ID16 to the 8 filaments, associating them with exclusive characteristics: light absorption coefficient 1.2, surface reflectivity threshold 25 (grayscale value), transmittance baseline value 90%, and toughness parameter 2.3GPa. At the same time, the "transmittance fluctuation threshold is reduced by 15% (from ±1% to ±0.85%)" is written into the detection rules.

[0021] S2 (Real-time Data Acquisition): Dust is generated in the workshop due to raw material feeding. The environmental monitoring sensor shows a real-time dust concentration of 0.8 mg / m³. 3 The images captured by the industrial camera showed a slight haze (grayscale value fluctuation ±8), and the extracted ID12 filament diameter was 8±0.3μm and the surface scratch was 1.0μm. The transmittance data collected by the fiber optic sensor fluctuated by ±1.2%, and all data were transmitted to the edge computing unit in real time. S3 (Dual-modal detection): The edge computing unit determines "dust concentration > 0.5 mg / m³". 3"Light intensity fluctuation 12% > 10%", the dual-modal detection module adjusted the weights: transmittance detection accounted for 65% and visual detection accounted for 35%; an ID12 filament transmittance fluctuation of 2.1% was detected (exceeding the adjustment threshold of 90% ± 0.85%), and a surface scratch of 1.0 μm was confirmed by the simultaneous visual image, and the overall judgment was "serious defect". S4 (Batch Adaptation): This batch is for the production of a single AR glass. 10 samples are automatically collected for every 1000m of filament produced. Due to the stable dust concentration, the detection threshold has not been adjusted. S5 (Leveled Early Warning): ID12 yarn trajectory offset of 0.6mm, the early warning execution unit triggers a stop signal, and the production line stops 300ms later to prevent yarn breakage from escalating. Implementation results: Technical specifications: 99.3% accuracy rate for early warning in dusty environments, 0.3% false alarm rate, and 300ms shutdown response time (compliant with industrial safety standards). Production benefits: Successfully avoided one AR glass wire breakage (AR glass raw material cost is 30% higher than E glass, and the loss of a single wire breakage is about 650 yuan), reduced the number of wire breakages in dusty environments by 2 per month, and saved about 39,000 yuan in costs. Example 3 This embodiment provides a machine vision-based glass fiber filament breakage early warning system and method for batch switching production from E glass (10μm) to AR glass (8μm). Specific implementation details include: Purpose of implementation: The system was validated to demonstrate that "the system can quickly adapt to different batches of glass filaments through learning from a small number of samples". System Architecture: Industrial camera module: Same as in Examples 1-2 (Hikvision MV-CA020-10GM), with the addition of a "batch switching image fast acquisition" mode (acquiring 50 frames of samples within 10 seconds), which is used to quickly obtain the visual samples required for batch adaptation. Fiber optic sensor module: Same as in Example 2 (Keyence FS-V21R, detection accuracy 0.03%), supports "one-click import of batch characteristics" (can directly read AR glass transmittance benchmark), and its function is to synchronously collect transmittance sample data. Edge computing unit: The hardware is consistent with that in Examples 1-2 (NVIDIA Jetson AGXXavier). The batch adaptation module has a built-in MAML algorithm (up to 15 iterations) and supports "automatic sample filtering" (removing fuzzy samples) to ensure adaptation efficiency and accuracy. Early warning execution unit: Same as in Examples 1-2 (Siemens S7-1200PLC), with the addition of "Batch Adaptation Complete Prompt" (green light always on), which is used to indicate the adaptation status. Implementation steps: S1 (Identity Number Generation): After the E glass production is completed, the characteristic data of the original ID1-ID8 are cleared; when the AR glass is put into operation, 8 frames of images are quickly acquired, ID17-ID24 are assigned, and basic characteristics (diameter 8μm, initial oscillation frequency 1.8Hz) are temporarily stored. S2 (Real-time Data Acquisition): Activate the "Batch Adaptation Sample Acquisition" mode. The industrial camera acquires 80 clear images (including different angles of each ID filament) within 10 seconds, and the fiber optic sensor simultaneously acquires 80 sets of transmittance data (90%±0.5%). All data is transmitted to the edge computing unit. S3 (Dual-modal detection): During the adaptation phase, real-time detection is not performed. Only the defect features in the sample data (such as the gray-scale gradient features of AR glass scratches) are analyzed to provide a basis for subsequent detection rule updates.

[0022] S4 (Batch Adaptation): The batch adaptation module of the edge computing unit calls the MAML algorithm, taking 80 samples (within the range of 50-100 samples) as input, and fine-tunes the "batch-specific feature layer": First, the model converges after 12 iterations; second, the detection rules are automatically updated, changing the transmittance benchmark value from 93% (E glass) to 90% (AR glass), and the diameter mutation threshold is lowered from 0.5μm to 0.3μm; third, the adaptation takes 25 minutes, and the green light of the warning execution unit is constantly lit, indicating that the adaptation is complete. S5 (Leveled Early Warning): After the adaptation is completed and production is started, the first AR glass filament (ID17) shows a sudden change in diameter of 0.3μm. The system judges it as a "general defect" and outputs a prompt to the operation terminal. There are no false alarms.

[0023] Implementation results: Technical specifications: Batch adaptation time 25 minutes (traditional pure vision solution requires 24 hours), warning accuracy rate after adaptation 99.2%; Production benefits: Batch changeover downtime was reduced from 24 hours to 25 minutes, and equipment utilization increased by 97.9% (calculated based on 3 batch changes per month, an additional 23.5 hours of production, an increase of approximately 1.2 tons of glass fiber, and a benefit of approximately 24,000 yuan). Example 4 This embodiment provides a machine vision-based glass fiber filament breakage early warning system and method for the production of high-toughness E-glass (toughness 3.5 GPa) vs. low-toughness E-glass (toughness 2.8 GPa). Specific implementation details include: Purpose of implementation: The system was validated to show that "the defect judgment threshold is adjusted according to the toughness parameters of the filament, adapting to the detection needs of materials with different toughness." System Architecture: Industrial camera module: Same as Example 1 (Hikvision MV-CA020-10GM), with the addition of a "toughness parameter-threshold mapping" algorithm, which can automatically adjust the diameter detection accuracy according to toughness, and its function is to accurately extract the morphological features of filaments with different toughness. Fiber optic sensor module: Same as in Example 1 (Keyence FS-V21R), the transmittance threshold is adjusted synchronously with the toughness parameter, and its function is to match the transmittance detection requirements of filaments with different toughness. Edge computing unit: The hardware is consistent with that in Example 1 (NVIDIA Jetson AGXXavier). The identification number generation module adds a "toughness parameter input interface", and the dual-modal detection module supports "dynamic threshold switching" to ensure the accuracy of defect judgment for different toughness filaments. Early warning execution unit: Same as in Example 1 (Siemens S7-1200PLC), with the addition of a "production parameter adjustment suggestion output" function (connected to the wire drawing machine control system), which outputs targeted production adjustment prompts. Implementation steps: S1 (Identity Number Generation): Two batches of E-glass fibers are launched: one is a high-toughness batch (toughness 3.5GPa), assigned ID25-ID32, with a diameter mutation threshold of 0.5μm; the other is a low-toughness batch (toughness 2.8GPa), assigned ID33-ID40, with a diameter mutation threshold of 0.4μm. S2 (Real-time Data Acquisition): Two batches are produced separately, and diameter and light transmittance data are collected: the diameter fluctuation of high-toughness filament is ±0.2μm, and the diameter fluctuation of low-toughness filament is ±0.3μm. All data are transmitted to the edge computing unit in real time. S3 (Dual-modal detection): First, a sudden change in the diameter of the high-toughness ID28 filament by 0.4μm (<0.5μm) is judged as "no defect"; second, a sudden change in the diameter of the low-toughness ID35 filament by 0.4μm (≥0.4μm) with a simultaneous fluctuation in transmittance of 1.0% is judged as "general defect". S4 (Batch Adaptation): Both batches are E glass. Only the toughness correlation threshold is adjusted. No need to re-adapt the samples. S5 (Level-based early warning): If the ID35 filament defect persists for 32s (≥30s), the early warning execution unit outputs a "production parameter adjustment prompt signal": the drawing speed is reduced from 12m / min to 11.6m / min (adjustment range 3.3%), and the cooling air temperature is maintained at 25℃ (no air temperature adjustment is required for E glass); after adjustment, the ID35 filament diameter fluctuation returns to ±0.2μm. Implementation results: Technical specifications: 99.4% accuracy in identifying defects in filaments of different toughness; filament breakage rate decreased by 40% after parameter adjustment; Production benefits: The breakage rate of low-toughness E glass was reduced from 5‰ to 3‰, reducing the number of breakages by 4 per month and saving approximately 20,000 yuan in raw material costs. Example 5 This embodiment provides a machine vision-based glass fiber filament breakage early warning system and method for 32-filament parallel interlacing production (multi-filament scenario). Specific implementation details include: Purpose of implementation: The verification system demonstrates its ability to accurately locate IDs and predict trajectories in multi-filament intertwined scenarios, thus avoiding misjudgments and filament breakage caused by intertwining. System Architecture: Industrial camera module: Same as Example 1 (Hikvision MV-CA020-10GM), with the addition of a "multi-filament region segmentation" algorithm (which can simultaneously identify the position of 32 filaments), and the frame rate is increased to 35fps (to ensure real-time trajectory acquisition). Its function is to accurately acquire multi-filament visual images and trajectory data. Fiber optic sensor module: adopts 32 channels Keyence FS-V21R (1 channel for each filament) to synchronously detect the transmittance of each filament, avoiding sensor confusion during interlacing. Its function is to assist in multi-filament ID verification and transmittance detection. Edge computing unit: The hardware is consistent with that in Example 1 (NVIDIA Jetson AGXXavier), with the addition of a trajectory prediction module (based on Kalman filtering), which supports "parallel computing of 32-wire trajectories" and can predict trajectories in the future 0.5-1s. Its function is to predict multi-wire trajectories and accurately associate them with IDs. Early warning execution unit: Same as in Example 1 (Siemens S7-1200PLC), with the addition of a "multi-filament early warning priority sorting" function (processing filaments with high trajectory deviation risk first), which is to efficiently output early warning signals in multi-filament scenarios. Implementation steps: S1 (ID Number Generation): Assign ID41-ID72 to 32 E glass filaments, associate with a safety spacing of 20μm (10μm×2 times), and synchronously store the initial transmittance benchmark (93%) of each filament to ensure that the ID is uniquely bound to the filament characteristics. S2 (Real-time Data Acquisition): The industrial camera acquires 35 frames of images per second, and the trajectory prediction module records the coordinates of each filament in real time (accuracy ±0.1mm) and calculates the spacing between adjacent filaments (e.g., the spacing between ID69 and ID70 is 22μm); 32 fiber optic sensors synchronously acquire transmittance as the basis for ID verification (e.g., ID69 transmittance 92.8%, ID70 transmittance 93.1%), and all data is transmitted to the edge computing unit in real time.

[0024] S3 (Dual-modal detection): At the 30th minute of production, ID69 and ID70 show an intertwining trend. The trajectory prediction module predicts through Kalman filtering that the distance between the two filaments will decrease to 15μm (<20μm safe distance) within the next 0.8s. At the same time, the dual-modal detection module verifies the IDs through "transmittance characteristics" (to avoid ID confusion caused by intertwining) and confirms that they are ID69 and ID70. S4 (Batch Adaptation): This batch is produced using a single multi-filament process and does not require adaptation. S5 (Leveled Early Warning): The early warning execution unit outputs a prompt according to priority: "The gap between ID69 and ID70 will be insufficient. It is recommended to adjust the guide wheel." The worker completes the adjustment within 0.8s, and the gap between the two wires is restored to 22μm to avoid friction and wire breakage. Implementation results: Technical specifications: 98% breakage prevention rate in multi-filament interweaving scenarios, 99.6% ID positioning accuracy (no ID confusion); Production benefits: The 32-filament production line reduces interlacing breakage by 3 times per month, saves about 6 hours of re-drawing time, produces an additional 0.3 tons of glass fiber, and generates a profit of about 6,000 yuan. Comparative Example 1 This comparative example provides a traditional purely visual fiber breakage early warning solution, including the following: Solution configuration: Using only the industrial camera module of Example 1 (a batch adaptation / trajectory prediction module without fiber optic sensors and edge computing units), defect judgment relies on fixed grayscale thresholds (surface scratches > 1.2 μm, diameter abrupt changes > 0.8 μm), there is no ID association, and multi-wire positioning relies solely on position coordinates. Implementation steps and results: Under normal conditions: The diameter of the E glass filament suddenly changes by 0.6μm (<0.8μm), which is not detected and leads to filament breakage; In a dusty environment: AR glass filament images are blurry, surface scratches are 1.0μm (<1.2μm), unrecognized, and broken filaments; When switching between batches: switching from E glass to AR glass requires manual annotation of 5,000 samples and training the model takes 24 hours; When multiple filaments are intertwined: ID confusion occurs, and the defect of ID69 is mistakenly identified as ID70, leading to an erroneous shutdown.

[0025] Test results and shortcomings: In a normal environment (corresponding to the scenario in Example 1), the traditional solution achieves a warning accuracy of 92%, a false alarm rate of 2.5%, a new batch adaptation time of 24 hours, and a multi-filament interlacing positioning accuracy of 85%. In a dusty environment (corresponding to the scenario in Example 2), the warning accuracy drops to 82%, the false alarm rate rises to 5.8%, the new batch adaptation time remains 24 hours, and the multi-filament interlacing positioning accuracy is 78%. In a batch switching scenario (corresponding to the scenario in Example 3), the warning accuracy is only 80%, the false alarm rate is 6.3%, the new batch adaptation time remains 24 hours, and the multi-filament interlacing positioning accuracy is 80%. Overall, the traditional solution has significant shortcomings in anti-interference, batch adaptation efficiency, and multi-filament positioning accuracy. The core differences from this solution: Lacks dual-modal anti-interference capability: Accuracy in dusty environments is 17.3% lower than this solution; No batch adaptation function: the switching time is 57.6 times that of this solution (24h vs 25min); Multi-wire positioning without ID association: accuracy is 14.6% lower than this solution; No trajectory prediction: Cannot prevent interlacing and breakage, with a breakage rate 40% higher than this solution.

[0026] Comparing Examples 1-5 with Comparative Example 1, the following is a summary of the comparison between Examples 1-5 (this solution) and Comparative Example 1 (traditional pure vision solution) in terms of applicable scenarios, technical architecture, key performance, and production benefits: In terms of applicable scenarios, this solution covers all scenarios in glass fiber filament production: Example 1 is for the normal environment of E glass multifilaments (low dust, stable light), and Example 2 is for the dust environment of AR glass multifilaments (dust concentration 0.8mg / m³). 3 Example 3 solves the batch switching between E glass and AR glass, Example 4 adapts to E glass with different toughness (3.5GPa / 2.8GPa), and Example 5 adapts to 32-filament parallel interlacing scenarios; while the comparative example can only be used in a simple and single environment, and cannot work stably in dust, batch switching, and multi-filament interlacing scenarios. For example, it fails directly due to image blurring in a dusty environment.

[0027] In terms of technical architecture, this solution is a complete architecture of "industrial camera + fiber optic sensor + edge computing unit (including ID number generation, dual-modal detection, batch adaptation module) + early warning execution unit", which supports dual-modal collaboration, few-sample adaptation and multi-filament trajectory prediction. In contrast, the scale only relies on industrial camera, has no dual-modal anti-interference capability (no fiber optic sensor), no batch adaptation function (no corresponding module), no multi-filament ID positioning and trajectory prediction, and the defect judgment relies only on fixed grayscale threshold, which is a single technical dimension.

[0028] In terms of key performance, this solution is comprehensively superior: For early warning accuracy, Examples 1-5 all achieve 99.2%-99.5%, compared to only 92% in normal environments, 82% in dusty environments, and only 80% in batch switching scenarios; the false alarm rate is as low as 0.2%-0.3%, compared to 2.5%-6.3% in the comparison examples; the batch adaptation time is only 25 minutes in this solution, compared to 24 hours in the comparison examples (57.6 times faster); the multi-filament positioning accuracy reaches 99.6%, compared to a maximum of only 85% in the comparison examples (as low as 78% in interleaved scenarios), and there is also an ID confusion problem.

[0029] In terms of production revenue, this solution has significant economic value: Example 1 saves RMB 45,000 in waste costs due to broken wires per month; Example 2 saves RMB 39,000 per month; Example 3 increases revenue by RMB 24,000 due to the adaptation to produce more glass wires quickly; Example 4 saves RMB 20,000 per month; and Example 5 saves RMB 6,000 per month. In contrast, the comparative example suffers from more than RMB 120,000 in additional raw material costs per month due to frequent wire breakage (such as failure to detect a sudden change in diameter of 0.6μm under normal conditions) and prolonged downtime due to adaptation, and the equipment utilization rate is 15%-20% lower than that of this solution.

[0030] In summary, this solution addresses the pain points of traditional solutions, such as weak anti-interference, slow adaptation, and poor positioning, through multi-module collaboration and full-scenario adaptation, combining technological advancement with practicality in production.

[0031] The above embodiments can be implemented in whole or in part by software, hardware, firmware or other arbitrary combinations. When implemented by software, the above embodiments can be implemented in whole or in part in the form of a computer program product.

[0032] Those skilled in the art will recognize that the modules and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0033] In addition, the functional modules in the various embodiments of this application can be integrated into one processing module, or each module can exist physically separately, or two or more modules can be integrated into one module.

[0034] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of protection of the claims.

[0035] In conclusion, 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, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A machine vision-based glass fiber filament breakage early warning system, characterized in that, include: An industrial camera module is used to acquire visual images during the glass fiber filament production process and obtain the morphological characteristics of the filament, including filament diameter, surface scratches, and motion trajectory. The fiber optic sensor module is located below the glass fiber filament and is used to detect the transmittance fluctuation data of the filament. The edge computing unit is electrically connected to the industrial camera module and the fiber optic sensor module, respectively, and receives visual image data transmitted from the industrial camera module and transmittance fluctuation data transmitted from the fiber optic sensor module in real time. The edge computing unit has the following built-in features: The identification number generation module is used to assign a unique identification number to each fiber when a new batch of fiberglass filaments is put into production, and to associate and store the identification number with the exclusive characteristics of the batch of filaments. The exclusive characteristics include the light absorption coefficient, surface reflectivity threshold and light transmittance benchmark value corresponding to the filament material. The dual-modal detection module, with machine vision inspection at its core, combines transmittance data from fiber optic sensors to extract defect features of the filament. These defect features include diameter abrupt changes ≥0.5μm, surface scratches ≤1μm, and transmittance fluctuations exceeding batch-specific benchmark values. The module also adjusts the weighting of visual inspection and transmittance detection based on environmental parameters, including dust concentration and light intensity, with a dust concentration threshold of 0.5mg / m³. 3 The threshold for judging light intensity fluctuations is 10%. The batch adaptation module is used to learn and adapt to new batches of yarn through a small number of samples. It only needs to collect 50-100 visual images and transmittance data samples with identification numbers to complete the adaptation of defect judgment standards. During the production process, every 1000m of yarn is produced, 10 unlabeled samples are automatically collected to update the judgment standards. The early warning execution unit is electrically connected to the edge computing unit and is used to output graded early warning signals based on the defect level and the silk trajectory offset corresponding to the identification number output by the dual-modal detection module.

2. A machine vision-based method for early warning of fiber breakage in glass fiber filament production, characterized in that, Includes the following steps: S1: When a new batch of glass fiber filaments is put into production, visual images of the filaments are captured by an industrial camera module. The identification number generation module assigns a unique identification number to each filament and stores the identification number in association with the exclusive characteristics of the batch of filaments. The exclusive characteristics include light absorption coefficient, surface reflectivity threshold and light transmittance benchmark value. S2: During the production process, the industrial camera module acquires real-time visual images of the filament and extracts the morphological features of the filament, including diameter, surface scratches and motion trajectory. The fiber optic sensor module simultaneously acquires the light transmittance fluctuation data of the filament and transmits them to the edge computing unit. S3: The dual-modal detection module uses morphological features extracted by machine vision as its core, combined with transmittance fluctuation data, to determine whether there are defects in the filament. When the real-time dust concentration in the workshop is ≤0.5mg / m³ 3 Furthermore, when the real-time light intensity fluctuation is ≤10%, the visual inspection weight accounts for 70%-80%, the transmittance inspection weight accounts for 20%-30%, and the focus is on judging the diameter change and surface scratches through morphological features. When the real-time dust concentration in the workshop is >0.5mg / m³ 3 When the real-time light intensity fluctuation is greater than 10%, the transmittance detection weight accounts for 60%-70%, and the visual inspection weight accounts for 30%-40%. The focus is on judging defects by whether the transmittance fluctuation exceeds the batch-specific benchmark value. S4: When a new batch is launched, the batch adaptation module quickly adapts the defect judgment standard based on 50-100 visual images and transmittance data samples; during the production process, every 1000m of yarn is produced, 10 unlabeled samples are automatically collected to update the judgment standard and adapt to the characteristic changes within the yarn batch. S5: The early warning execution unit outputs a graded early warning based on the defect level output by the dual-modal detection module and the silk trajectory offset corresponding to the identification number. The defect levels include minor, general, and severe. A shutdown signal is triggered when the defect level is severe and the trajectory offset is greater than 0.5mm. When the defect level is general and the wire spacing is less than the safety spacing, the safety spacing is 2-3 times the standard diameter of the wire in this batch, and an operation terminal prompt signal is output. When the defect level is minor and the duration is ≥30s, a production parameter adjustment prompt signal is output.

3. The machine vision-based glass fiber filament breakage early warning system as described in claim 1, characterized in that, The industrial camera module has a resolution of ≥1920×1080, a frame rate of ≥25fps, and a lens focal length of 8-16mm to meet the requirements for extracting morphological features of filaments with a diameter of 5-20μm.

4. The machine vision-based glass fiber filament breakage early warning system as described in claim 1, characterized in that, The optical fiber sensor module has a detection wavelength of 532nm, a transmittance detection accuracy of ≤0.1%, and a response time of ≤1ms. The transmittance benchmark values ​​in the batch-specific characteristics are: 92%-95% for E glass fiber and 88%-91% for AR glass fiber.

5. The machine vision-based glass fiber filament breakage early warning system as described in claim 1, characterized in that, The batch-specific characteristics associated with the identification number generation module also include the toughness parameter of the filament. The toughness parameter is associated with the material type in the batch-specific characteristics, wherein the toughness parameter of AR glass fiber filament is ≤2.5GPa, and the toughness parameter of E glass fiber filament is ≥3GPa. When judging defects, the dual-modal detection module adjusts the defect threshold in combination with the toughness parameter: when the toughness parameter is ≤2.5GPa, the diameter mutation threshold is lowered to 0.3μm, and the transmittance fluctuation threshold is lowered by 10%-15%.

6. The method for early warning of fiber breakage in glass fiber production based on machine vision as described in claim 2, characterized in that, In step S1, the associated storage information of the identification number also includes the initial oscillation frequency of the filament. The initial oscillation frequency is associated with the material type of the filament in this batch. The initial oscillation frequency of AR glass fiber filament is 1-2Hz, and the initial oscillation frequency of E glass fiber filament is 2-3Hz. In step S5, when the difference between the actual oscillation frequency of the filament and the initial oscillation frequency is greater than 0.5Hz, the early warning execution unit increases the judgment weight of the defect level.

7. The method for early warning of fiber breakage in glass fiber production based on machine vision as described in claim 2, characterized in that, In step S3, when the dual-modal detection module determines a sudden change in diameter, it uses multi-frame image superposition and comparison. When a diameter change of ≥0.5μm is detected in three consecutive frames, it is determined to be a diameter sudden change defect, thus avoiding misjudgment caused by noise in a single frame image.

8. The method for early warning of fiber breakage in glass fiber production based on machine vision as described in claim 2, characterized in that, In step S4, the batch adaptation module uses the MAML algorithm to complete the fast adaptation, with ≤15 iterations and an adaptation time of ≤30 minutes, without the need for manual labeling of sample defect types.

9. The method for early warning of fiber breakage in glass fiber production based on machine vision as described in claim 2, characterized in that, In step S5, the production parameter adjustment prompt signal includes drawing speed adjustment and cooling air temperature adjustment. The drawing speed adjustment range is ±5%, and the cooling air temperature adjustment range is ±2℃. The specific adjustment parameters are determined according to the material type in the batch-specific characteristics: when the material is AR glass fiber, the drawing speed adjustment range is ≤3%; when the material is E glass fiber, the cooling air temperature adjustment range is ≤1.5℃.

10. The machine vision-based glass fiber filament breakage early warning system as described in claim 1, characterized in that, The edge computing unit also has a built-in trajectory prediction module. Based on the historical movement trajectory of the silk corresponding to the identification number, it uses Kalman filtering to predict the trajectory in the next 0.5-1 seconds. When the distance between the predicted trajectory and the adjacent silk is less than the safety distance, which is 2-3 times the standard diameter of the batch of silk, the warning execution unit will be triggered in advance.

Images