Glass fiber fly yarn breakage early warning and disposal method based on AI visual analysis

Abstract

The application relates to the technical field of AI vision detection, and discloses a glass fiber flying yarn breakage early warning and disposal method based on AI vision analysis, which comprises the following steps: collecting normal operation videos of a furnace table, performing frame-by-frame processing on the videos, forming a plurality of pictures, manually labeling pictures containing broken yarns, using the labeled pictures to train a model, performing real-time detection on the furnace table through the model, starting an audible and visual alarm of the furnace table where the broken yarns are located when the number of detected broken yarns reaches a threshold value, obtaining alarm information of the furnace table where the broken yarns are located, processing the alarm information through defined rules, generating a 16-bit instruction, sending the instruction to corresponding staff, reminding the staff to perform on-site detection and processing, and judging staff ID information through an optical character recognition model.

Description

Technical Field

[0001] This invention relates to the field of AI visual inspection technology, specifically to a method for early warning and handling of glass fiber filament breakage based on AI visual analysis. Background Technology

[0002] In the sprue workshop, all fiber-drawing operations are conducted manually, with manual rewinding performed only when fly-out fibers are detected. Workers work in shifts, constantly moving back and forth for inspections, resulting in high labor intensity and delays in detecting fly-out fibers, leading to production waste. Furthermore, the sprue temperature reaches as high as 1200℃, with high radiant heat. The glass fiber filaments are approximately 10μm in diameter and are transparent, making it very difficult to observe broken fibers under these red-hot conditions, and also causing eye strain and discomfort.

[0003] To adjust the thermal balance of the stencil and improve operational continuity, the auxiliary heating current of the stencil needs to be adjusted based on the filament status and flow rate. This requires workers to manually record and input information such as the stencil furnace position, filament time, and processing time to provide initial data for temperature adjustment. This process is physically demanding and prone to errors and omissions in data input.

[0004] This invention is based on the YOLO optimized target detection algorithm, collects a set of images featuring broken glass fiber strands, establishes a lightweight target detection method, and embeds this method into an edge AI computing device. Through a serial port module, alarm signals from multiple furnace positions are output in hexadecimal format to the signal receiving device to remind workers to perform the wire guiding operation. Combined with an audible and visual alarm, it forms a complete detection and alarm device. This device features high detection accuracy, fast speed, small size, and stable operation. Summary of the Invention

[0005] This invention provides a method for early warning and handling of glass fiber filament breakage based on AI visual analysis, which helps to solve the problems mentioned in the background art.

[0006] This invention provides the following technical solution: a method for early warning and handling of glass fiber filament breakage based on AI visual analysis, comprising:

[0007] Obtain the working area for each furnace;

[0008] Each work area is equipped with a wire drawing machine, a staff member, multiple cameras, and an audible and visual alarm.

[0009] Each staff member is wearing a signal receiving device;

[0010] The cameras include camera number one and camera number two;

[0011] One edge AI computing device is used for every three furnaces;

[0012] All edge AI computing devices are numbered sequentially, starting from 1.

[0013] Each furnace platform under each edge AI computing device is numbered in order from furnace platform number one to furnace platform number three.

[0014] By collecting video information on the normal operation of the wire drawing machine in each furnace, and training the detection model;

[0015] The detection model is optimized in the first step by using compression and excitation operations;

[0016] The detection model is then optimized in the second step using a loss function and algorithm.

[0017] The wire drawing machine's operation process is monitored in real time using a detection model via edge AI computing devices, and an alarm is triggered based on the detection results.

[0018] If an alarm is triggered, instructions will be generated based on the alarm information and defined rules.

[0019] Based on the alarm information, some wire drawing machines were inspected.

[0020] The system sends instructions to staff to conduct on-site inspections and determines whether the staff meet the requirements of the current work area.

[0021] Optionally, the step of collecting video information on the normal operation of the wire drawing machine in each furnace and training the detection model includes:

[0022] Video footage of the wire drawing machine operating normally in each furnace was collected as a sample.

[0023] The sample includes the status events of each wire drawing machine in the furnace during the wire drawing process;

[0024] Status events include normal states and abnormal states;

[0025] The samples are processed frame by frame to form multiple sample images;

[0026] Manually label sample images containing abnormal states;

[0027] The abnormal condition refers to wire breakage that occurs during the operation of the wire drawing machine.

[0028] Manually labeled sample images are used to train the model, and this model is denoted as the detection model.

[0029] Optionally, the first step of optimizing the detection model using compression and excitation operations includes:

[0030] The specific formula for compression is as follows:

[0031]

[0032] In the above formula, X represents the feature map, c represents the number of channels in the feature map, H and W are the height and width of the feature map, respectively, Z represents the compressed feature vector, and i and j represent the indices of the feature map's height and width in the spatial dimension, respectively. c X represents the value of the c-th channel in the compressed feature vector. c (i,j) represents the value of the c-th channel in the original feature map at spatial location (i,j);

[0033] The specific formula for the incentive operation is as follows:

[0034] S = F ex (z,W)=σ(W2δ(z,W1z))

[0035] In the above formula, F ex This represents the activation operation, z represents the input vector, and W represents the dimension, including W1 and W2. 2， W1 and W2 represent the dimensions used to compress and expand the vector in the fully connected layer, δ represents the activation function of the first fully connected layer, and σ represents the activation function after the second fully connected layer.

[0036] Optionally, the second step of optimizing the detection model using a loss function and algorithm includes:

[0037] The EIOU_LOSS loss function is used for identification. The specific formula for the loss function is as follows:

[0038]

[0039] In the above formula, ELOU_LOSS represents the loss function, IOU represents the ratio of the intersection to the union of the predicted and ground truth boxes, ρ represents the Euclidean distance, c is the diagonal length of the minimum closure between the predicted and ground truth boxes, and C w and C h These are the width and height of the minimum closure between the predicted bounding box and the ground truth bounding box, respectively, where b represents the center point of the predicted bounding box. gt w represents the center point of the ground truth bounding box, and w represents the width of the predicted bounding box. gt h represents the width of the ground truth bounding box, and h represents the height of the predicted bounding box. gt Indicates the height of the actual bounding box;

[0040] The model was optimized using the DIOU_NMS algorithm. The specific formula for the DIOU_NMS algorithm is as follows:

[0041]

[0042] In the above formula, IOU represents the ratio of the intersection to the union of the predicted and ground truth boxes, ρ represents the Euclidean distance, c is the diagonal length of the minimum closure between the predicted and ground truth boxes, and b represents the center point of the predicted box. g t This represents the center point of the actual bounding box.

[0043] Optionally, the step of using an edge AI computing device to perform real-time detection of the wire drawing machine's operation process using a detection model, and determining whether to issue an alarm based on the detection results, includes:

[0044] Each furnace drawing machine was used to acquire real-time operation video of the edge AI computing device using camera number one, and recorded as the target video;

[0045] The target video is processed frame by frame to form multiple target images;

[0046] Power on all edge AI computing devices and determine if they are functioning correctly.

[0047] If the edge AI computing device malfunctions, an alarm will be triggered to notify professionals for repair.

[0048] If the edge AI computing device is functioning properly, then the detection model is used to detect the target image;

[0049] If an abnormal state is detected, the target image with the abnormal state will be added to the abnormal image set.

[0050] Set the threshold a for the number of abnormal states;

[0051] If the number of elements in the abnormal image set is less than 1, continue the detection.

[0052] If the number of elements in the abnormal image set is equal to 'a', then the audible and visual alarms corresponding to all elements in the abnormal image set will be activated.

[0053] Optionally, if an alarm is triggered, an instruction is generated based on the alarm information and defined rules, including:

[0054] Obtain the wire drawing machine corresponding to each element in the abnormal image set, denoted as the abnormal machine, and enter it into the abnormal machine set;

[0055] Retrieve alarm information for each element in the abnormal machine set;

[0056] The alarm information includes the device number of the edge AI computing device to which the abnormal machine belongs, the furnace number to which the abnormal machine belongs, the activation status of the abnormal machine's switch, and the activation status of the audible and visual alarm corresponding to the abnormal machine.

[0057] The alarm information of each abnormal machine is processed by defining rules, instructions are generated, and the instructions are entered into the instruction set.

[0058] The API service interface is used to push the elements in the instruction set to the signal receiving devices of the corresponding staff members.

[0059] Optionally, the step of processing alarm information for each abnormal machine by defining rules and generating instructions includes:

[0060] The specific rules are defined as follows:

[0061] All wire drawing machines under the edge AI computing device are represented by 3-bit binary numbers. The weights of the 3-bit binary numbers are set to 8, 4 and 2 respectively from left to right, and each binary number is set to 0 or 1 to represent the wire drawing machine number.

[0062] Use 5-bit binary numbers to represent the audible and visual alarm and the internal switch of the wire drawing machine;

[0063] The internal switches of the wire drawing machine include switch number one, switch number two, switch number three, and switch number four.

[0064] The five binary digits, from left to right, represent switch number one, switch number two, switch number three, switch number four, and the sound and light alarm. Each binary digit can be set to 0 or 1 to indicate the status of the alarm and the internal switch of the wire drawing machine, with 0 indicating off and 1 indicating on.

[0065] The weights of the 5-bit binary numbers are set from left to right as 1, 8, 4, 2 and 1 respectively;

[0066] Use the 3-bit binary number as the first part;

[0067] Use the 5-bit binary number as the second part;

[0068] Combine the parts in the order from the first part to the second part to form an 8-bit binary number;

[0069] The 8-bit binary number is converted into a hexadecimal number based on the weight of each bit. The hexadecimal number is the instruction.

[0070] Optionally, the step of sending instructions to notify staff to conduct on-site testing and determining whether the staff meet the requirements of the current work area includes:

[0071] Randomly obtain the working area of ​​the element in the abnormal machine set, and denote it as the target area. Obtain the employee number of the corresponding staff member based on the target area, and denote it as the target employee number.

[0072] The employee IDs of staff in the target area are acquired in real time by the second camera, recorded as pre-selected employee IDs, and identified and judged according to the optical character recognition model.

[0073] If the pre-selected work number does not match the target work number, the signal receiving device of the worker corresponding to the target area will be re-sent instructions.

[0074] If the pre-selected work number matches the target work number, the staff in the target area will be notified to conduct an on-site inspection, and the abnormal machine corresponding to the current target area in the abnormal machine set will be removed. The elements in the abnormal machine set will be reselected and judged.

[0075] If the on-site inspection results are consistent with the alarm information, then an anomaly should be handled.

[0076] If the on-site inspection results are inconsistent with the alarm information, the current abnormal machine will be removed from the abnormal machine set.

[0077] If the staff finishes the abnormal handling operation, the optical character recognition model in the current work area will be turned off.

[0078] Optionally, the step of performing inspection operations on some wire drawing machines based on alarm information includes:

[0079] Enter all wire drawing machines into the machine collection;

[0080] Remove all abnormal machine set elements from the machine set;

[0081] Staff members conduct inspections of the elements within the machine set.

[0082] If an abnormal condition is found during the inspection, the abnormal condition image is obtained through the No. 1 camera of the current wire drawing machine and recorded in the inspection set;

[0083] The inspection operation ends when all elements in the machine set have been traversed.

[0084] The elements in the inspection set are used to train the detection model.

[0085] The present invention has the following beneficial effects:

[0086] 1. This method for early warning and handling of glass fiber fly filament breakage based on AI visual analysis uses AI visual analysis technology to monitor and warn of glass fiber fly filament breakage in real time, improving production safety and efficiency. By configuring intelligent equipment and optimizing algorithms, it can promptly detect and handle fly filament breakage problems, reducing production interruptions and material waste.

[0087] 2. This AI-based visual analysis-based method for early warning and handling of fiber filament breakage in glass fibers can more accurately identify the operating status of the drawing machine by collecting and labeling sample images of abnormal states and using the samples to train the detection model, thereby improving the accuracy and reliability of the early warning system.

[0088] 3. This AI-based visual analysis-based method for early warning and handling of broken glass fiber filaments introduces a lightweight ShuffleNetV2 network into the backbone network to reduce the number of parameters and computational load. An SE attention mechanism is embedded in the feature fusion network, which includes suppression and excitation operations. By learning the importance weights of each channel, the feature map is adaptively recalibrated, enabling the network to focus more on information-rich channels and suppress less important ones. By assigning different weights to different channels, the discriminative ability of features is enhanced, allowing the network to focus more on key information when processing images and improving the accuracy of feature recognition.

[0089] 4. This AI-based visual analysis-based method for early warning and handling of broken glass fiber filaments, by selecting the EIOU_LOSS loss function and the DIOU_NMS algorithm, provides more refined processing of the overlapping area between the predicted and ground truth bounding boxes, exhibits stronger adaptability to target shapes, and offers better accuracy and applicability. The EIOU_LOSS loss function considers three parts: overlapping area, center distance, and width-height loss. The width-height loss minimizes the difference in width and height between the target box and the anchor box, resulting in faster model convergence. This reduces the model training cycle, accelerates the model training process, and provides more accurate positioning information, thereby improving the accuracy of target detection.

[0090] 5. This method for early warning and handling of fiberglass filament breakage based on AI visual analysis uses edge AI computing devices for real-time detection and determines whether the equipment can operate normally, eliminating the possibility of production stoppages caused by equipment failure. By setting thresholds to define alarm timing, it can quickly respond to abnormal states of the drawing machine, promptly notify staff for handling, reduce production interruption time, and improve production continuity and product quality.

[0091] 6. This AI-based visual analysis-based method for early warning and handling of fiberglass filament breakage generates instructions based on alarm information and defined rules. These instructions are then sent to the corresponding signal receiving devices via API service interfaces, which in turn notify the staff. This ensures timely transmission and execution of instructions, improving emergency response speed and processing efficiency.

[0092] 7. The AI-based visual analysis-based method for early warning and handling of fiberglass filament breakage has clear and standardized instructions generated through defined rules, which are easy for staff to understand and execute. It also facilitates the automated processing and recording of anomalies, improving the accuracy of operations and the traceability of data.

[0093] 8. This method for early warning and handling of fiberglass filament breakage based on AI visual analysis obtains the employee ID of the staff member corresponding to the abnormal machine, and uses signal receiving equipment and optical character recognition model to confirm the identity of the current staff member, ensuring that the correct staff member handles the abnormal situation, thereby improving the pertinence and efficiency of the handling.

[0094] 9. This AI-based visual analysis-based method for early warning and handling of fiberglass filament breakage can further ensure the normal operation of all drawing machines by conducting inspections of non-abnormal machines, promptly identifying and addressing potential problems, and improving the stability and reliability of the entire production system. Simultaneously, abnormal conditions discovered during inspections can be used to further train the detection model, continuously optimizing the model's early warning performance.

[0095] 10. This method for early warning and handling of fiberglass filament breakage based on AI visual analysis reduces manual labor by using automated equipment for filament quality inspection, thereby reducing the difficulty and intensity of workers' work and improving work efficiency and quality. Attached Figure Description

[0096] Figure 1 This is a schematic diagram of the process of the present invention. Detailed Implementation

[0097] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0098] Example 1, see Figure 1 A method for early warning and handling of glass fiber filament breakage based on AI visual analysis, including:

[0099] Obtain the working area for each furnace;

[0100] Each work area is equipped with a wire drawing machine, a staff member, multiple cameras, and an audible and visual alarm.

[0101] Each staff member is wearing a signal receiving device;

[0102] The cameras include camera number one and camera number two;

[0103] One edge AI computing device is used for every three furnaces;

[0104] All edge AI computing devices are numbered sequentially, starting from 1.

[0105] Each furnace platform under each edge AI computing device is numbered in order from furnace platform number one to furnace platform number three.

[0106] By collecting video information on the normal operation of the wire drawing machine in each furnace, and training the detection model;

[0107] The detection model is optimized in the first step by using compression and excitation operations;

[0108] The detection model is then optimized in the second step using a loss function and algorithm.

[0109] The wire drawing machine's operation process is monitored in real time using a detection model via edge AI computing devices, and an alarm is triggered based on the detection results.

[0110] If an alarm is triggered, instructions will be generated based on the alarm information and defined rules.

[0111] Based on the alarm information, some wire drawing machines were inspected.

[0112] By sending instructions to staff to conduct on-site testing, and determining whether the staff meet the requirements of the current work area;

[0113] By using AI visual analysis technology to monitor and warn of fiberglass filament breakage in real time, production safety and efficiency are improved. Through the configuration of intelligent equipment and algorithm optimization, filament breakage problems can be detected and dealt with in a timely manner, reducing production interruptions and material waste.

[0114] The process of collecting video information on the normal operation of the wire drawing machine in each furnace and training a detection model includes:

[0115] Video footage of the wire drawing machine operating normally in each furnace was collected as a sample.

[0116] The sample includes the status events of each wire drawing machine in the furnace during the wire drawing process;

[0117] Status events include normal states and abnormal states;

[0118] The samples are processed frame by frame to form multiple sample images;

[0119] Manually label sample images containing abnormal states;

[0120] The abnormal condition refers to wire breakage that occurs during the operation of the wire drawing machine.

[0121] Manually labeled sample images are used to train the model, and this model is denoted as the detection model.

[0122] By collecting and labeling sample images of abnormal states, and using these samples to train a detection model, the operating status of the wire drawing machine can be identified more accurately, thereby improving the accuracy and reliability of the early warning system.

[0123] The first step of optimizing the detection model using compression and excitation operations includes:

[0124] The specific formula for compression is as follows:

[0125]

[0126] In the above formula, X represents the feature map, c represents the number of channels in the feature map, H and W are the height and width of the feature map, respectively, Z represents the compressed feature vector, and i and j represent the indices of the feature map's height and width in the spatial dimension, respectively. c X represents the value of the c-th channel in the compressed feature vector. c (i,j) represents the value of the c-th channel in the original feature map at spatial location (i,j);

[0127] The specific formula for the incentive operation is as follows:

[0128] S = F ex (z,W)=σ(W2δ(z,W1z))

[0129] In the above formula, F ex This represents the activation operation, z represents the input vector, and W represents the dimension, including W1 and W2. 2， W1 and W2 represent the dimensions used to compress and expand the vector in the fully connected layer, δ represents the activation function of the first fully connected layer, and σ represents the activation function after the second fully connected layer.

[0130] A lightweight ShuffleNetV2 network is introduced into the backbone network to reduce the number of parameters and computational cost. An SE attention mechanism is embedded in the feature fusion network. The SE attention mechanism includes suppression and activation operations. By learning the importance weight of each channel, the feature map is adaptively recalibrated, enabling the network to pay more attention to information-rich channels and suppress less important channels. By assigning different weights to different channels, the discriminative ability of features is enhanced, allowing the network to pay more attention to key information when processing images and improve the accuracy of feature recognition.

[0131] The second step of optimizing the detection model using a loss function and algorithm includes:

[0132] The EIOU_LOSS loss function is used for identification. The specific formula for the loss function is as follows:

[0133]

[0134] In the above formula, ELOU_LOSS represents the loss function, IOU represents the ratio of the intersection to the union of the predicted and ground truth boxes, ρ represents the Euclidean distance, c is the diagonal length of the minimum closure between the predicted and ground truth boxes, and C w and C h These are the width and height of the minimum closure between the predicted bounding box and the ground truth bounding box, respectively, where b represents the center point of the predicted bounding box. gt w represents the center point of the ground truth bounding box, and w represents the width of the predicted bounding box. gt h represents the width of the ground truth bounding box, and h represents the height of the predicted bounding box. gt Indicates the height of the actual bounding box;

[0135] The model was optimized using the DIOU_NMS algorithm. The specific formula for the DIOU_NMS algorithm is as follows:

[0136]

[0137] In the above formula, IOU represents the ratio of the intersection to the union of the predicted and ground truth boxes, ρ represents the Euclidean distance, c is the diagonal length of the minimum closure between the predicted and ground truth boxes, and b represents the center point of the predicted box. gt Indicates the center point of the true bounding box;

[0138] By selecting the EIOU_LOSS loss function and the DIOU_NMS algorithm, the processing of the overlapping area between the predicted bounding box and the ground truth bounding box is more refined, the adaptability to the target shape is stronger, and the accuracy and applicability are better. The EIOU_LOSS loss function considers three parts: overlapping area, center distance, and width and height loss. The width and height loss minimizes the difference in width and height between the target box and the anchor box, making the model converge faster. This can reduce the model training cycle, speed up the model training process, and provide more accurate localization information, thereby improving the accuracy of target detection.

[0139] The process of using an edge AI computing device to perform real-time detection of the wire drawing machine's operation using a detection model, and determining whether to issue an alarm based on the detection results, includes:

[0140] Each furnace drawing machine was used to acquire real-time operation video of the edge AI computing device using camera number one, and recorded as the target video;

[0141] The target video is processed frame by frame to form multiple target images;

[0142] Power on all edge AI computing devices and determine if they are functioning correctly.

[0143] If the edge AI computing device malfunctions, an alarm will be triggered to notify professionals for repair.

[0144] If the edge AI computing device is functioning properly, then the detection model is used to detect the target image;

[0145] If an abnormal state is detected, the target image with the abnormal state will be added to the abnormal image set.

[0146] Set the threshold a for the number of abnormal states;

[0147] The threshold 'a' for the number of abnormal states is a constant and can be adjusted according to production conditions to determine when to trigger an alarm.

[0148] If the number of elements in the abnormal image set is less than 1, continue the detection.

[0149] If the number of elements in the abnormal image set is equal to 'a', then the audible and visual alarms corresponding to all elements in the abnormal image set will be activated.

[0150] By using edge AI computing devices for real-time detection and determining whether the equipment can operate normally, the probability of production stoppages caused by equipment failures can be eliminated. By setting thresholds to define alarm timing, the abnormal status of the wire drawing machine can be responded to quickly, and staff can be notified in a timely manner to handle the situation, reducing production interruption time and improving production continuity and product quality.

[0151] If an alarm is triggered, instructions are generated based on the alarm information and defined rules, including:

[0152] Obtain the wire drawing machine corresponding to each element in the abnormal image set, denoted as the abnormal machine, and enter it into the abnormal machine set;

[0153] Retrieve alarm information for each element in the abnormal machine set;

[0154] The alarm information includes the device number of the edge AI computing device to which the abnormal machine belongs, the furnace number to which the abnormal machine belongs, the activation status of the abnormal machine's switch, and the activation status of the audible and visual alarm corresponding to the abnormal machine.

[0155] The alarm information of each abnormal machine is processed by defining rules, instructions are generated, and the instructions are entered into the instruction set.

[0156] The elements within the instruction set are pushed to the signal receiving devices of the corresponding staff members through the API service interface;

[0157] Instructions are generated based on alarm information and defined rules. These instructions are then sent to the corresponding signal receiving devices via API service interfaces, and the signal receiving devices notify the staff. This ensures the timely delivery and execution of instructions, improving emergency response speed and processing efficiency.

[0158] The process of defining rules to process alarm information from each malfunctioning machine and generating instructions includes:

[0159] The specific rules are defined as follows:

[0160] All wire drawing machines under the edge AI computing device are represented by 3-bit binary numbers. The weights of the 3-bit binary numbers are set to 8, 4 and 2 respectively from left to right, and each binary number is set to 0 or 1 to represent the wire drawing machine number.

[0161] Use 5-bit binary numbers to represent the audible and visual alarm and the internal switch of the wire drawing machine;

[0162] The internal switches of the wire drawing machine include switch number one, switch number two, switch number three, and switch number four.

[0163] The five binary digits, from left to right, represent switch number one, switch number two, switch number three, switch number four, and the sound and light alarm. Each binary digit can be set to 0 or 1 to indicate the status of the alarm and the internal switch of the wire drawing machine, with 0 indicating off and 1 indicating on.

[0164] The weights of the 5-bit binary numbers are set from left to right as 1, 8, 4, 2 and 1 respectively;

[0165] Use the 3-bit binary number as the first part;

[0166] Use the 5-bit binary number as the second part;

[0167] Combine the parts in the order from the first part to the second part to form an 8-bit binary number;

[0168] The 8-bit binary number is converted into a hexadecimal number according to the weight of each binary digit. The hexadecimal number is the instruction.

[0169] Instructions generated by defining rules have a clear and standardized format, making them easy for staff to understand and execute. They also facilitate the automated handling and recording of anomalies, improving operational accuracy and data traceability.

[0170] The step of sending instructions to staff to conduct on-site testing and determining whether staff meet the requirements of the current work area includes:

[0171] Randomly obtain the working area of ​​the element in the abnormal machine set, and denote it as the target area. Obtain the employee number of the corresponding staff member based on the target area, and denote it as the target employee number.

[0172] The employee IDs of staff in the target area are acquired in real time by the second camera, recorded as pre-selected employee IDs, and identified and judged according to the optical character recognition model.

[0173] If the pre-selected work number does not match the target work number, the signal receiving device of the worker corresponding to the target area will be re-sent instructions.

[0174] If the pre-selected work number matches the target work number, the staff in the target area will be notified to conduct an on-site inspection, and the abnormal machine corresponding to the current target area in the abnormal machine set will be removed. The elements in the abnormal machine set will be reselected and judged.

[0175] If the on-site inspection results are consistent with the alarm information, then an anomaly should be handled.

[0176] If the on-site inspection results are inconsistent with the alarm information, the current abnormal machine will be removed from the abnormal machine set.

[0177] If the staff finishes the abnormal handling operation, the optical character recognition model in the current work area will be turned off;

[0178] By obtaining the employee ID of the staff member corresponding to the abnormal machine, the identity of the current staff member is confirmed using signal receiving equipment and optical character recognition model, ensuring that the correct staff member handles the abnormal situation and improving the targeting and efficiency of the handling.

[0179] The aforementioned inspection operation on some wire drawing machines based on alarm information includes:

[0180] Enter all wire drawing machines into the machine collection;

[0181] Remove all abnormal machine set elements from the machine set;

[0182] Staff members conduct inspections of the elements within the machine set.

[0183] If an abnormal condition is found during the inspection, the abnormal condition image is obtained through the No. 1 camera of the current wire drawing machine and recorded in the inspection set;

[0184] The inspection operation ends when all elements in the machine set have been traversed.

[0185] The elements in the inspection set are used to train the detection model;

[0186] By conducting routine inspections of non-abnormal machines, the normal operation of all wire drawing machines can be further ensured, potential problems can be identified and addressed in a timely manner, and the stability and reliability of the entire production system can be improved. Simultaneously, abnormal conditions discovered during the inspection process can be used to further train the detection model and continuously optimize the performance of the model's early warning system.

[0187] By using automated equipment for filament quality inspection, the amount of manual labor is reduced, the difficulty and intensity of workers' work are lowered, and work efficiency and quality are improved.

[0188] 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, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, 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 process, method, article, or apparatus.

[0189] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the technical principles of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A multi-furnace station glass fiber fly filament breakage early warning and disposal method based on AI visual analysis, characterized by: This includes: acquiring the working area of ​​each furnace; configuring each working area with one wire drawing machine, one signal receiving device, multiple cameras, and one audible and visual alarm; assigning one worker to each signal receiving device; including two cameras (Camera 1 and Camera 2); and using one edge AI computing device for every three furnaces. All edge AI computing devices are numbered sequentially, starting from 1; the furnaces under each edge AI computing device are numbered sequentially from furnace number one to furnace number three. By collecting video information of the normal operation of the wire drawing machine in each furnace and training a detection model, the detection model is first optimized by compression and excitation operations; the detection model is then optimized in the second step by loss function and algorithm; the operation process of the wire drawing machine is monitored in real time by the detection model through edge AI computing device, and an alarm is triggered based on the detection results. If an alarm is triggered, instructions are generated based on the alarm information and defined rules, including: Obtain the wire drawing machine corresponding to each element in the abnormal image set, denoted as the abnormal machine, and enter it into the abnormal machine set; obtain the alarm information for each element in the abnormal machine set; the alarm information includes the device number of the edge AI computing device to which the abnormal machine belongs, the furnace number to which the abnormal machine belongs, the start status of the abnormal machine switch, and the start status of the audible and visual alarm corresponding to the abnormal machine. Alarm information from each malfunctioning machine is processed by defining rules, generating instructions, and then entered into an instruction set. The specific rules are defined as follows: All wire drawing machines under the edge AI computing device are represented using 3-bit binary numbers. The weights of these 3-bit binary numbers are set to 8, 4, and 2 sequentially from left to right, with each bit representing the machine number by setting 0 or 1. The sound and light alarm and the internal switches of the wire drawing machines are represented using 5-bit binary numbers. The internal switches of the wire drawing machines include switch number 1, switch number 2, switch number 3, switch number 4, and switch number 4. These 5-bit binary numbers, from left to right, represent switch number 1, switch number 2, switch number 3, switch number 4, and the sound and light alarm, with each bit representing the machine number by setting 0 or 1. Alternatively, 1 indicates the switch status of the alarm and wire drawing machine, 0 indicates off, and 1 indicates on; the weights of the 5-bit binary numbers are set sequentially from left to right as 1, 8, 4, 2, and 1; the 3-bit binary number is used as the first part; the 5-bit binary number is used as the second part; the two parts are combined in the order of the first part to the second part to form an 8-bit binary number; the 8-bit binary number is converted into a hexadecimal number according to the weight of each bit, and the hexadecimal number is the instruction; the elements in the instruction set are pushed to the signal receiving devices of the corresponding staff members through the API service interface; Based on alarm information, inspections were conducted on some wire drawing machines; instructions were sent to staff to conduct on-site inspections and to determine whether the staff met the requirements of the current work area, specifically including: The system randomly selects the work area of ​​each element within the abnormal machine set and marks it as the target area. It then obtains the employee ID of the corresponding worker within the target area and marks it as the target employee ID. The system also uses a second camera to capture the employee IDs of workers within the target area in real time and marks them as pre-selected employee IDs. These are then identified and judged using an optical character recognition model. If the pre-selected employee ID does not match the target employee ID, the system re-sends instructions to the signal receiving device of the worker corresponding to the target area. If the pre-selected employee ID matches the target employee ID, the system notifies the workers in the target area to conduct an on-site inspection, removes the abnormal machine corresponding to the current target area from the abnormal machine set, and re-selects and judges elements within the abnormal machine set. If the on-site inspection results are consistent with the alarm information, anomaly handling is performed; if the on-site inspection results are inconsistent with the alarm information, the current abnormal machine is removed from the abnormal machine set; if the staff member ends the anomaly handling operation, the optical character recognition model of the current work area is turned off; by obtaining the staff member's employee number corresponding to the abnormal machine, the identity of the current staff member is confirmed using the signal receiving equipment and the optical character recognition model.

2. The method for early warning and handling of glass fiber breakage in multiple furnace positions based on AI visual analysis according to claim 1, characterized in that: The process of collecting video information on the normal operation of the wire drawing machine in each furnace and training a detection model includes: Video information of the normal operation of the wire drawing machine in each furnace is collected as a sample; the sample contains the state events of each wire drawing machine in the furnace during the wire drawing process; the state events include normal state and abnormal state; the sample is processed frame by frame to form multiple sample images; the sample images containing abnormal states are manually labeled; the abnormal state is the wire breakage phenomenon that occurs during the operation of the wire drawing machine; the manually labeled sample images are used to train the model, and the model is recorded as the detection model.

3. The method for early warning and handling of glass fiber breakage in multiple furnace positions based on AI visual analysis according to claim 1, characterized in that: The first step of optimizing the detection model using compression and excitation operations includes: The specific formula for compression is as follows: ; In the above formula, X represents the feature map, c represents the number of channels in the feature map, H and W are the height and width of the feature map, respectively, Z represents the compressed feature vector, and i and j represent the indices of the feature map's height and width in the spatial dimension, respectively. This represents the value of the c-th channel in the compressed feature vector. This represents the value of the c-th channel in the original feature map at spatial location (i,j); The specific formula for the incentive operation is as follows: ; In the above formula, Let z represent the activation operation, z represent the input vector, and W represent the dimension, including... and and This represents the dimension used for compressing and expanding vectors in the fully connected layer. This represents the activation function of the first fully connected layer. This represents the activation function following the second fully connected layer.

4. The method for early warning and handling of glass fiber breakage in multiple furnace positions based on AI visual analysis according to claim 1, characterized in that: The second step of optimizing the detection model using a loss function and algorithm includes: using the loss function EIOU_LOSS for identification, the specific formula of which is as follows: Where ELOU_LOSS represents the loss function, and IOU represents the ratio of the intersection to the union of the predicted and ground truth bounding boxes. Let represent the Euclidean distance, and c be the diagonal length of the minimum closure between the predicted and ground truth boxes. and and represent the width and height of the minimum closure between the predicted bounding box and the ground truth bounding box, respectively, where b represents the center point of the predicted bounding box. represents the center point of the ground truth bounding box, and w represents the width of the predicted bounding box. h represents the width of the ground truth bounding box, and h represents the height of the predicted bounding box. Indicates the height of the actual bounding box; The model was optimized using the DIOU_NMS algorithm. The specific formula for the DIOU_NMS algorithm is as follows: In the above formula, IOU represents the ratio of the intersection to the union of the predicted bounding box and the ground truth bounding box. Let represent the Euclidean distance, c be the diagonal length of the minimum closure between the predicted and ground truth boxes, and b be the center point of the predicted box. This represents the center point of the actual bounding box.

5. The method for early warning and handling of glass fiber breakage in multiple furnace positions based on AI visual analysis according to claim 1, characterized in that: The process of using an edge AI computing device to perform real-time detection of the wire drawing machine's operation using a detection model, and determining whether to issue an alarm based on the detection results, includes: Each furnace drawing machine operates in real time using a single camera under the edge AI computing device, and is recorded as the target video. The target video is then processed frame by frame to form multiple target images. Start all edge AI computing devices and determine if they can operate normally; if the edge AI computing devices cannot operate normally, issue an alarm and notify professionals for repair; if the edge AI computing devices can operate normally, use the detection model to detect the target image. If an abnormal state is detected, the target image with the abnormal state is added to the abnormal image set; a threshold 'a' is set for the number of abnormal states; if the number of elements in the abnormal image set is less than 'a', the detection continues. If the number of elements in the abnormal image set is equal to 'a', then the audible and visual alarms corresponding to all elements in the abnormal image set will be activated.

6. The method for early warning and handling of glass fiber breakage in multiple furnace positions based on AI visual analysis according to claim 1, characterized in that: The step of performing inspection operations on some wire drawing machines based on alarm information includes: recording all wire drawing machines into a machine set; removing all abnormal machine set elements from the machine set; having staff perform inspection operations on the elements in the machine set; if an abnormal state is found during the inspection, acquiring an image of the abnormal state through the first camera of the current wire drawing machine and recording it into the inspection set; ending the inspection operation when all elements in the machine set have been traversed; and training the detection model with the elements in the inspection set.

Images