A DEVICE AND METHOD FOR CAMERA-BASED FABRIC DEFECT DETECTION AND CONTROLLED STOPPING IN KNITTING MACHINES.

Abstract

The invention describes a system that detects fabric defects occurring during production on knitting machines by analyzing images obtained through a camera system and circular illumination structure in real-time on an embedded computing unit. It generates segmentation outputs using an encoder-decoder-based deep learning model on these images, verifies candidate defect areas with post-processing filters, and enables rapid intervention against these defects only during active production, depending on the machine's operating status monitored by a rotation sensor. In case of defect detection, the machine is stopped via a relay structure compatible with the machine's existing control infrastructure and through a controlled electrical connection via a resistor, thereby reducing production losses and protecting machine components. Simultaneously, it records and monitors data related to the detected defects, along with segmentation outputs and defect verification results.It is an integrated mechanism and method that makes it possible to improve analysis and production processes.

Description

CAMERA-BASED FABRIC DEFECT DETECTION AND CONTROL IN KNITTING MACHINES. STOPPING DEVICE AND METHOD Technical Area The invention uses camera-based technology to detect fabric defects that may occur during production on knitting machines. Detection using imaging, deep learning-based image analysis, and segmentation techniques. and based on these findings, a controlled shutdown of the machine. It relates to the mechanism and method. The invention, in particular, features a camera and lighting structure positioned toward the fabric pool, and a local edge. The encoder-decoder based image processing model that runs on the device, at the pixel level. defect segmentation, post-processing filters, intelligent activation and error detection based on machine operating status. This relates to a device and method that includes a controlled stopping mechanism in this situation. State of the Art In knitted fabric production, quality control has long relied heavily on visual monitoring by operators. It is carried out based on this, but errors are most often made after the fabric is produced, not during production. or it is noticed during periodic human inspections; this situation increases the failure rate, It makes the control process subjective and provides a real-time, standardized audit. This reveals the need for infrastructure. Current applications are dependent on operator attention. the control cannot be maintained with the same precision throughout the production process, and some automated solutions are only effective. It is known to encompass specific types of errors or specific machine configurations. Solutions in the current state of the art generally involve final visual inspection and knitting. Camera-based image acquisition systems added to the machine later allow for the imaging of a specific fabric. The software processes images taken from the region, gives an alarm when an error is detected, or some These applications are based on the principle of stopping the machine. However, these solutions these include camera placement, lighting architecture, image processing infrastructure, and machine intervention. There are significant differences in terms of form; especially adaptation to different types of knitting machines. Obtaining a consistent image from the area of ​​the fabric where defects are most clearly visible is only possible during active production. during the inspection and ensuring the machine is controlled in a manner compatible with the existing machine logic. Issues such as halting operations are not addressed together in every solution. 35 2 In this context, the “Apparatus” numbered WO2025022426A1 is considered as an example of existing technology. The patent application is titled "Method for Detection of Textile Defects in Circular Knitting Machine". One or more devices placed on the fixed part for detecting textile defects in a circular knitting machine. multiple cameras, one or more light sources mounted on a rotating section, and light direction working with its components; parameters in the acquired images are determined using pre-defined parameters. a system that compares, warns the user, and can stop the machine if necessary. It explains. In the background section of the application, it is stated that in previous solutions, the rotating part was very much in place. numerous cameras are exposed to vibration, maintenance difficulties, space requirements, dust and oil stains, and It is stated that this creates disadvantages such as the need for slip rings; in the new solution, the cameras are fixed. It is recommended that the dimming of the light sources and the rotating part of the light source be placed on the dimming side. However, the aforementioned The application included a strategic camera focused on the fabric pool where defects are most clearly visible. its placement, a special circular lighting structure around the camera, only when the machine is in active production. Smart activation based on the activated turn sensor and the machine's existing rope break logic. Similarly, controlled stopping is achieved through a resistor between the +24V and ground lines. Its structure is not clearly defined; therefore, the solution involves machine-compatible and controlled intervention. It does not fully meet the approach. Another example is WO2024251692A1, “Method for Detecting Defects in a Knitted Fabric “Manufactured by an Automatic Knitting Machine, Corresponding System and Computer Program” This is a patent application titled [Patent Title]. In this application, the tube produced by an automatic circular knitting machine is described. A digital camera placed in a fixed position on the upper inner part of the fabric captures the inner surface of the fabric. by obtaining images and comparing these images with data from an accurate reference fabric It is stated that the presence of defects is automatically determined. The application claims that the method is low-cost. It can be implemented without making major changes to the shape and machine, the camera is positioned on top of the tube fabric positioning it in the neutral voltage zone improves image quality and is fundamental. In practice, even the mere detection of a defect serves as a warning to the operator. It is stated that this is advantageous. However, the approach is primarily based on comparison with a reference image. it was stated that it was based on, that the camera was in a fixed position facing the inner surface, and that the operator It appears that the focus is on providing information, with a special camera placement focused on the fabric pool. Ring lighting around the camera, smart activation based on production status with tour sensor, local Modular image processing architecture on the end device and instead of abruptly shutting down the machine after an error. The controlled stopping mechanism, compatible with the machine's existing stop logic, is clearly evident. It is not included. Therefore, the application in question is multi-component and machine integrated. It does not meet the need for a structure based on solid foundations. 35 Also, document number WO2022060340A1, “An Artificial Vision System for Circular Knitting Machines The patent application is titled "Performing Real-Time Inspection and Classification of Fabric Defects". 3 This serves as an example of the current technique. This application requires at least two cameras for circular knitting machines. at least two lighting units, at least two encoders, and a computer for transmitting and storing images. an artificial vision system containing software with image processing and defect classification algorithms He explains that the system can track both the inside and outside of the fabric, and detect any defects. It states that it can store location data and stop the machine when a fault is detected. The application specifically noted that some previous systems only scanned a single surface, providing information on defect location. It was stated that these did not produce or include synchronization and appropriate lighting elements. The aim is to eliminate this. However, in this application, the images are transferred to a computer. The narrative is based on processing, generally using LEDs for lighting and camera carriers. The cabin structure is explained, focusing on interior and exterior surface tracking; controlled stopping mechanism. It does not include real-time detection and classification. Therefore, although the aforementioned application includes real-time detection and classification, even if it provides a controlled shutdown approach that protects machine health and production continuity, It does not fully resolve the standardized integration structure focused on fabric pools. Consequently, solutions within the known state of the art have quality that is dependent on human control. Although it has shown significant development from manual control to camera-based automated control, Applicability to different types of knitting machines, in the area where fabric defects are most apparent. Standardized image acquisition, intelligent activation that prevents unnecessary inspections outside of production, data locally processed and controlled, which safely interrupts production without damaging the machine. It does not present the stopping elements together and in an integrated manner. Existing solutions... The deficiency has made it necessary to make improvements in the relevant technical field. Brief Description of the Invention The present invention meets the aforementioned requirements while eliminating all disadvantages. and some additional advantages are that fabric can be damaged during production in knitting machines. error detection using camera-based imaging, deep learning-based image analysis, and segmentation. detection using techniques and, based on these detections, controlled operation of the machine It is related to a mechanism and method for stopping it. Based on the current state of the art, the aim of the invention is to improve the production process in knitting machines. By detecting fabric defects in real time, these defects can be corrected without damaging the machine. The aim is to ensure a swift and controlled response. 35 The purpose of the invention is to photograph fabric by means of a camera structure positioned towards the fabric pool. The aim is to ensure that errors on the surface are displayed in the clearest and most stable way. 4 Another purpose of the invention is to illuminate fabric thanks to the circular lighting structure located around the camera. The aim is to increase the visibility of defects such as lines, holes, and threads on the surface. Another objective of the invention is to create a localized system optimized to operate in real-time on the end device. fast and delay-free analysis of images obtained thanks to image processing architecture is to ensure. Another objective of the invention is encoder-decoder based deep learning that runs on the end device. Thanks to its model, numerous different fabric defects can be detected automatically and with high accuracy. The goal is to ensure that it is done. Another objective of the invention is to utilize pixel-level data generated by a deep learning model. Thanks to the segmentation map, defect areas on the fabric surface can be precisely identified. The goal is to ensure its determination. Another objective of the invention is to enable the encoder and decoder components to identify defects specific to knitted fabrics. Thanks to its structure, it detects subtle structural distortions and low-contrast defect regions. The goal is to ensure reliable identification. Another objective of the invention is to enable segmentation outputs to be passed through post-processing filters. The goal is to reduce false positive diagnoses. Another aim of the invention is to analyze the geometric properties, dimensional characteristics, orientations, and characteristics of defects. natural fabric texture thanks to the final processing mechanism that evaluates the continuity characteristics. The goal is to ensure that variations are separated from defects. Another purpose of the invention is thanks to the intelligent activation mechanism that works integrated with the tour sensor. The system should only be activated when the machine is in active production mode. Another objective of the invention is a control structure that can become passive depending on the machine's operating status. This helps to reduce unnecessary processing load and false positives. Another purpose of the invention is to provide a controlled stop function that is activated after fault detection. The goal is to ensure the machine can be stopped safely without sudden and damaging shutdowns. 35 Another objective of the invention is to achieve stopping through a resistor between the +24V and ground lines. This connection allows knitting machines to operate in compatibility with the existing stop logic. is to ensure. Another objective of the invention is to create an intervention mechanism compatible with the existing machine control structure. This ensures the mechanical protection of machine components. Another purpose of the invention is its integration structure, which is compatible with different types of knitting machines. The goal is to ensure that the system can be used in a wide range of applications. Another advantage of the invention is that the camera system monitors the fabric pool thanks to its strategic placement. The goal is to ensure a monitoring process that is consistent with the operators' existing control practices. Another purpose of the invention is to provide operators with instant notifications in case of errors through the generated error reporting system. The goal is to ensure a rapid response. Another aim of the invention is to provide real-time monitoring of the production process through local and remote monitoring interfaces. The goal is to ensure that it can be monitored in a timely manner. Another purpose of the invention is to improve quality through error records and production data generated by the system. The goal is to ensure that control processes are traceable and reportable. Another objective of the invention is to analyze production efficiency through a sensor-based data derivation structure. The goal is to ensure that it can be done. Another aim of the invention is to enable different technologies to be developed thanks to a modular deep learning and image processing infrastructure. The goal is to ensure that image processing models are updatable and selectable. Another objective of the invention is to improve quality thanks to the integrated structure of the system integrated into the production process. the control is performed continuously and automatically without relying on manual processes is to ensure. The structural and characteristic features and all the advantages of the invention are described in detail below. This will make it clearer, therefore the evaluation should also include these detailed explanations. This should be done taking that into consideration. 35 Detailed Description of the Invention 6 This detailed explanation describes the fabric that may be damaged during production in knitting machines, which is the subject of the invention. error detection using camera-based imaging, deep learning-based image analysis, and segmentation. detection using techniques and, based on these detections, controlled operation of the machine The mechanism and method for stopping it are merely examples to help better understand the issue. This is explained in a way that does not create any limiting effects. The device and method described in the invention relate to the detection of fabric defects that occur during production on knitting machines. real-time detection and adjustment of the machine's current state based on the detected errors. An integrated quality system that enables controlled shutdowns in a manner compatible with the operational structure. It is a control solution. Within this scope, the system includes image acquisition, lighting, data processing, and machinery. Monitoring the machine's condition and intervening in the machine are integrated within a single structure. by creating a continuous, automated and standardized monitoring mechanism It is structured in a way that can adapt to different types of knitting machines. It is designed based on the principle of both hardware and software components working together. The image processing infrastructure within the system provides real-time images on the end device. an encoder-decoder based deep learning model optimized to work It uses pixel-level imaging of images obtained from the fabric surface. It is structured to generate segmentation outputs and identify defect regions. The subject is segmentation outputs, geometric properties of defects, dimensional characteristics, Natural fabric texture processed with post-processing filters that evaluate orientations and continuity characteristics. This will enable the differentiation of variations from defects and reduce false positives. It is analyzed in this way. The invention consists of a camera system, a circular lighting structure, an embedded computing unit, and a turn. The system consists of a sensor, a machine stop relay structure, and a connection infrastructure. The camera system, It is responsible for capturing images of the fabric surface, and the system allows for the clearest possible identification of defects on the fabric. It is positioned so that it can see the fabric pool, which is the area from which it can be observed. The camera system, located inside the mesh area of ​​the knitting machine, ensures the fabric has a taut and smooth surface. It is positioned to face the area created by the fabric curling, folding, or creasing. This area is where the fabric might curl, fold, or Because it is a transitional region where errors can be observed without deformation, optically detectable errors. It represents an area where it is most prominently displayed. Thanks to this positioning, the fabric This makes it easier to visualize defects such as lines, holes, broken threads, or similar imperfections on the surface. Image quality is improved. Thus, segmentation-based defect detection is performed on the images. The accuracy of the 35 analyses was increased, revealing defect areas on the fabric surface at the pixel level. This contributes to the determination. In addition, the camera system is positioned perpendicular to the continuous flow direction of the fabric or on a specific point. positioned to display images aligned at an angle, preventing movement. 7 Blur and flickering effects are minimized. This allows the image to be transferred to the image processing model. This improves the consistency of the data and facilitates the identification of low-contrast defects. This arrangement, It is applicable to different types of knitting machines and is a standard regardless of the machine type. It enables the creation of a viewing point and facilitates system integration. The circular lighting structure is integrated around the camera system and is positioned by the camera. It ensures that the images are obtained under adequate lighting. The circular lighting structure. The camera system is positioned in a ring-like formation around the perimeter of this structure. Thanks to this, light is directed onto the fabric surface in a way that is aligned with the camera axis and is circumferentially symmetrical. By transmitting light, homogeneous illumination is achieved across the surface. Thanks to this structure, the fabric surface... Potential shadowing is reduced, and details on the surface are revealed more homogeneously with a light distribution. This ensures that surface variations, holes, lines, and other features, particularly those caused by the yarn direction, are made more prominent. The visibility of low-contrast defects, such as subtle deformities, is enhanced. Additionally, rounded shapes are also improved. The lighting structure is directional on the fabric surface, compared to point or linear light sources. It improves the contrast balance on the surface by minimizing glare (reflection) effects. This allows... The noise level of the images obtained by the camera system is reduced, and the image Processing algorithms and segmentation-based deep learning models yield more stable and accurate results. It is produced in this way. However, the round lighting structure is suitable for different fabric types and different By enabling the creation of similar optical conditions in surface textures, the system allows for different production methods. This contributes to consistent performance under these conditions. Thus, errors in the image processing stage are avoided. Detection accuracy is increased, and defect areas on the fabric surface are identified at the pixel level. It contributes to their separation. The embedded computing unit processes images captured by the camera system. is responsible. The embedded computing unit is located near the camera system, usually in the machine. It is located on the surface or inside the control panel. Thanks to this close positioning... Image data is processed directly locally without being transferred to an external system, and data transmission is avoided. This prevents potential delays. This unit instantly stores the received images in memory. It runs segmentation-based image processing algorithms and a deep learning model, and analyzes fabrics. It detects whether there are any errors. Thanks to the local processing of the images, the system, It gains real-time analysis capability and the time between fault detection and machine intervention is reduced. is minimized. For this purpose, real-time computing is implemented on the end device within the embedded computing unit. a deep learning model with approximately 11 million parameters optimized to work It is available. In addition, the embedded computing unit is capable of operating under a continuous data stream. 35 is optimized, enabling stable and uninterrupted operation even under high processing loads. The software infrastructure running within the embedded computing unit supports different image processing models. It is designed in a modular structure to allow for its use. The software in question... 8 its infrastructure consists of encoder and decoder components structured for processing image data. It may include a deep learning architecture consisting of an encoder component that extracts images of the fabric surface. While enabling the extraction of distinctive features, the decoder component uses the extracted features. It generates a pixel-level segmentation output of the fabric surface. Segmentation output: defective or flawless classification for each pixel in the image. It can be generated in the form of a binary segmentation map. Thus, the defect on the fabric surface The location and boundaries of the regions can be determined at the pixel level. Encoder and decoder. the characteristic structures of defects seen in components, especially on knitted fabric surfaces It can be structured for learning purposes. In this way, line defects, holes, and thread defects can be identified. Detection of distortions and low-contrast structural defects is facilitated by the deep learning model. The segmentation outputs generated are then passed through one or more post-processing filters. can be subjected to these. The aforementioned final processing filters are geometrically based on the detected candidate defect regions. by evaluating their features, dimensional characteristics, orientations, and continuity properties, it is incorrect. This can help reduce positive detections. Post-processing filters also have an inherent knitted structure. It can be structured in a way that distinguishes natural texture variations from actual defects. Thus, the system Overall accuracy performance can be improved while the false alarm rate is reduced. Thanks to this modular structure. Different algorithms can be selected depending on different fabric types, production conditions, or types of defects. It can be updated or modified during operation. However, the embedded computing unit is external. Because it is configured to operate independently of systems, it can work without a network connection. This ensures the system operates continuously even in production environments where it is limited or has other limitations. The rotation sensor is responsible for monitoring the operating status of the knitting machine and the machine's rotation. It generates a signal depending on its movement. The rotation sensor is located close to the rotating parts of the machine. It detects the machine's speed by being positioned there. This sensor sends pulses at a specific frequency depending on the machine's speed. It generates signals and, through these signals, determines the instantaneous operating speed and motion status of the machine. The data obtained from this sensor is evaluated by the embedded computing unit, and It is determined whether the machine is in active production. During this evaluation, a specific cycle is considered. Situations above the threshold are considered active production, and the system is then put into operation. In addition, data from the rotation sensor synchronizes the image acquisition process with the machine movement. This also allows for the images to be taken at the correct time intervals. Within the scope of the structure, a data communication line is established between the components. Accordingly, The mechanism is only activated when the machine is producing, and is not activated when the machine is stopped or out of production. It is disabled in such cases. This prevents unnecessary data processing and the use of system resources. Consumption and false positive error detections are minimized, and the system operates more stably and efficiently. 35 is provided. In addition, the image processing model and post-processing filters are only available during active production. By enabling this process, efficient use of processing resources can be achieved. 9 The machine stop relay system stops the machine if a defect is detected on the fabric. It is the component that enables the machine to be stopped. The machine stop relay structure is the component that enables the knitting machine to stop in its current state. It is electrically connected to the control system. This configuration is between the +24V line on the machine and the ground line. It stops the machine by creating a controlled connection between them through a specific resistance. By using the appropriate resistance value, current limiting is achieved, preventing sudden and sudden changes in the machine's control system. This prevents high current overload. This shutdown approach involves directly cutting off the power or Unlike sudden cut-off operations equivalent to an emergency stop button, the machine's own internal control It creates a mechanism that triggers the logic and provides a gradual stop. This allows the machine to... not a sudden and damaging stop, but a controlled stop in accordance with the existing operating principles. This is ensured. In addition, this structure allows the yarn breakage sensor, which is standard in knitting machines, to operate. It operates on a similar principle, enabling integration compatible with the existing machine infrastructure. It offers and can be implemented without any additional mechanical intervention in the system. Thanks to this method... sudden damage that can occur to sensitive mechanical components such as transmissions, gears, belts and needles Overloads and mechanical stresses are prevented. However, the machine stop relay structure prevents errors. Depending on the detection, it is controlled to be triggered only when necessary, thus avoiding unnecessary triggers. This prevents repeated shutdowns and ensures production continuity. defect detected by image processing model and confirmed by post-processing filters Machine intervention may be necessary depending on the region. The connectivity infrastructure enables all components in the system to communicate with each other. It includes electrical and data transmission lines. These lines are separate for data transmission and control signal transmission. It can be configured as follows. This configuration includes a camera system, an embedded computing unit, a tour sensor, and a machine. It enables the transmission of data and control signals between the stop relay structure. Also, the camera... deep learning model and post-processing filters of image data obtained by the system It also enables the data to be transferred to the embedded computing unit for processing. The connectivity infrastructure... Data structured to meet the high bandwidth requirements of image data. low-latency signal lines used for transmitting control signals together with the transmission lines This structure allows different types of data to be synchronized within the same system. This makes it possible to transmit image data, segmentation outputs, and defect verification. Simultaneous data flow can be created between the data, lap sensor signals, and control signals. In addition, the connectivity infrastructure is resistant to electromagnetic interference that may occur in industrial production environments. Data transmission reliability is ensured thanks to its shielded connectors and modular cabling structure. is increased. However, the connections between the system components will form a modular structure. This arrangement facilitates maintenance, troubleshooting, and component replacement procedures. 35 It also allows the system to communicate with external monitoring and control interfaces, and this This enables remote access, data logging, live monitoring, and system configuration processes. This makes it possible to implement it. Thanks to this infrastructure, the system can operate independently at the local level. a flexible communication architecture that can operate independently but also integrate with external systems when needed It provides segmentation outputs, defect records, and analysis results related to fault detection. It can be transferred to external monitoring systems. In the method described in the invention, images of the fabric pool are first captured by the camera system. During this process, the circular lighting structure actively works to provide the images with appropriate lighting. It enables the acquisition of images under specific conditions. The images obtained by the camera system are displayed at a specific time. Continuous monitoring of the fabric surface by taking measurements at intervals or in a continuous stream. The resulting images are transferred to the embedded computing unit, where they are stored in temporary memory. It is stored. Thanks to this temporary memory structure, comparisons can be made between consecutive image frames. This allows for the implementation of these measures and makes it possible to monitor the changes that occur over time. The embedded computing unit performs encoder-decoder-based deep learning on the acquired images. By running the model and segmentation algorithms, we can determine if there are any defects or anomalies on the fabric surface. It analyzes whether or not it exists. During this analysis, the pattern continuity in the image, the surface structure, and the expected characteristics are examined. Reference features are taken into account. Each pixel in the image is analyzed by the deep learning model. A binary segmentation map can be created that produces a classification of defective or flawless products. However, during the analysis process, specific regions on the image are examined first and then processed. The load is optimized and real-time operational performance is maintained. In the segmentation map After identifying candidate defect regions, one or more post-processing filters are applied. These can be evaluated. Additionally, changes in environmental lighting or natural features on the fabric surface can also be considered. In order to prevent false positives caused by variations, a threshold is set within the algorithm. Values ​​and comparison criteria can be adjusted dynamically. The aforementioned last process filters, candidate geometric properties, dimensional characteristics, orientations, and continuity of defect regions By evaluating their characteristics, we can help reduce misclassifications. In this way, both This makes it possible to detect low-contrast defects and ensures the system's stability and reliability. The process ensures the natural texture resulting from the knitted structure is preserved through final processing filters. By separating variations from defect regions, false positives are reduced and the system Its accuracy can be improved. Based on data received from the rotation sensor, the machine's operating status is continuously monitored. Situations where the machine speed is above a certain level are considered active production, and Only in this case is the image processing process kept active. This assessment is obtained from the tour sensor. This is performed by analyzing the pulse frequency over a specific time interval and instantaneously. To avoid being affected by fluctuations, the decision-making mechanism is based on a specific continuity criterion. It is activated. When the machine is paused, the system automatically becomes inactive and unnecessary functions are disabled. 35 analyses are prevented. Additionally, false triggers occur during stop-start transitions of the machine. To prevent this, the system uses a delayed activation and deactivation logic. By running it in this way, more stable control is achieved. This allows for the efficient use of system resources. 11 This ensures accuracy and prevents false positives. Furthermore, the deep learning model and post-processing filters... By ensuring it is only run during active production, the processing load is reduced and real-time operation is achieved. Work performance is preserved. However, thanks to this structure, the image processing process is only Analysis efficiency is increased by conducting the analysis during time periods containing meaningful production data, and the resulting The reliability of the results is increased. This situation makes the production of a segmentation-based defect detection model more reliable. It helps maintain the accuracy of defect detection by preventing it from being affected by external images. If an error is detected on the fabric as a result of image processing, embedded computation is used. A control signal is generated by the unit. Prior to the generation of this signal, fault detection takes place. To improve accuracy, verification can be performed on multiple consecutive image frames. and prevents temporary or noise-induced false detections. Furthermore, the deep learning model... The segmentation outputs generated are passed through post-processing filters to identify candidate defects. Verification of the regions can be ensured. The generated control signal is delivered without delay or in advance. The machine stop signal is transmitted to the relay structure with a defined, very short delay time. Machine stop relay Its structure receives this signal and establishes a controlled connection between the +24V line and the ground line via a resistor. This triggers and stops the machine. This stopping process is triggered by the machine's existing rope break sensor. It works on a similar principle, thus allowing production to be interrupted without damaging the mechanical components. This also prevents the system from being triggered again after the control signal is transmitted. This is done by applying a specific waiting period to avoid unnecessary repetitive stop commands. This is prevented. For this purpose, a time delay circuit is included within the machine stop relay structure or A software delay mechanism can be used. This structure allows for quick intervention in case of errors. This also ensures that the system operates in a stable and reliable manner. Segmentation outputs and by evaluating the validation results obtained from the final processing filters together Stoppages caused by false alarms can be reduced. As soon as an error is detected, the system also creates an error log, and these logs are stored locally or remotely. The generated logs, image data from the moment of the failure, and processed analysis data are transferred to monitoring interfaces. outputs, segmentation outputs, defect validation results, defect type information, occurrence time, and It is configured to include data such as the relevant machine status. Users can access these interfaces through them. It can track error types, their occurrence times, and live images. It can also identify detected defects. They can view segmentation results and verification information for their respective regions. Similarly Over time, the system records the timestamp and machine ID information for each error event. It provides historical traceability and reporting capabilities. This allows operators to respond quickly. This ensures that production processes become more traceable. Additionally, the data collected by the system... 35 data points are stored for long-term analysis to identify error occurrence trends and recurring errors. This allows for the identification of problem sources and the optimization of production parameters. It recognizes this data. This data may also include segmentation outputs and defect verification records. 12 However, by using this data in preventive maintenance processes, machine failures can be detected in advance. It contributes to prediction. It also helps in analyzing the temporal distributions of different defect types. It allows. Thanks to this structure, the invented mechanism and method improve the quality control process in knitting machines. By making it continuous, automated and safe, we both reduce production losses and improve the quality of machinery and equipment. It maintains product quality. Thanks to the segmentation-based deep learning model, the fabric It becomes possible to identify defects on the surface at the pixel level. Furthermore, the real Timely error detection and controlled intervention capabilities prevent errors from spreading throughout the production process. By preventing waste, the waste rate is reduced and production continuity is increased. With post-processing filters. System reliability is increased by reducing the false alarm rate through the verification of fault areas. It is being enhanced. However, the integration structure works in harmony with the existing machine infrastructure. Thanks to this, the system can be implemented without requiring any additional mechanical changes, and is beneficial for businesses. It offers a practical solution. At the same time, the data obtained by the system can be analyzed. Thanks to this, quality control processes can be managed in a data-driven manner, and production It becomes possible to continuously improve its performance. Segmentation outputs and More detailed assessment of defect trends using defect verification data. This ensures both the protection of machine components and a longer lifespan. In the long term, a more stable and sustainable production infrastructure is being created.

Claims

1. Camera-based detection of fabric defects that may occur during production on knitting machines. Detection using imaging and analysis techniques and based on these findings It is a mechanism for stopping the machine in a controlled manner, and its feature is; - the fabric is directed into the fabric pool within the cage area of ​​the knitting machine and the fabric aligned with the flow direction, perpendicular to the flow direction or at a specific angle, to follow the flow direction of the surface. a camera system positioned in this way, - The light is projected onto the fabric surface in alignment with the camera system axis and circumferentially symmetrical. a circular structure integrated around the camera system to transmit the information in this way. lighting structure, - locally processes the images captured by the camera system, into the camera system. located nearby, on the machine or inside the control panel, Defect detection on fabric surface using an encoder-decoder based deep learning model It will produce segmentation outputs to identify regions at the pixel level. an embedded computing unit structured in this way, - a rotation sensor that generates a signal based on the rotational movement of the knitting machine, - In case of fault detection, an electrical fault is detected in the knitting machine's existing control system. an intervening machine stop relay structure, - a link that enables the transmission of data and control signals between the aforementioned components It includes infrastructure.

2. The device conforms to Claim 1 and its feature is that the embedded computing unit is suitable for different fabric types. Depending on the manufacturing conditions or types of errors, data is stored in the memory unit and processed by the processor. the selection of at least one of the multiple image processing algorithms being run, or with a modular software infrastructure that allows for modification of the fabric surface configured to generate pixel-level segmentation outputs from images It involves a deep learning architecture consisting of encoder and decoder components.

3. The device conforms to Claim 1 and its feature is that the rotation sensor monitors the rotating part of the knitting machine. located in a position mechanically related to its elements and corresponding to the machine revolution. It is configured to generate electrical signals with a pulse frequency.

4. The device conforms to Claim 1 and its characteristic is that it connects the embedded computing unit to the tour sensor. transferring pulse signals generated by the rotation sensor to the embedded computing unit. It includes an electrical data communication line that provides this.

5. The device conforms to Claim 1 and its feature is that the machine stop relay structure of the knitting machine It will create a controlled connection between the 35 +24V line and the ground line through a specific resistor. This means the knitting machine is electrically connected to the existing control system. 14 6. The device conforming to either of claims 1 or 5 has the following feature: machine stop relay. Its structure is the electrical connection between the knitting machine's +24V line and the ground line. It is structured in a way that will create 7. The device conforming to either of claims 1 or 6 has the following feature: machine stop relay. its structure is a control circuit that limits re-triggering following a trigger signal. It includes.

8. The system complies with Claim 1, and its features include: the connection infrastructure, image data, and a data transmission line for transmitting segmentation outputs and control signals It includes a separate signal transmission line for transmission.

9. The device complies with Claim 1, and its characteristic feature is that the connection infrastructure is for industrial production. connection elements protected against electromagnetic interference that may occur in their environments and includes a modular cabling structure.

10. Camera-based detection of fabric defects that may occur during production on knitting machines. Detection using imaging and analysis techniques and based on these findings It is a method for stopping the machine in a controlled manner, and its characteristic feature is; - with the help of a circular lighting structure, the fabric pool is illuminated under suitable lighting conditions. Obtaining images of the fabric surface using a directed camera system, - transferring the acquired images to the embedded computing unit, - the aforementioned images are processed by the embedded computing unit using an encoder-decoder based process. segmentation of the fabric surface is analyzed using a deep learning model. generating the outputs and checking whether there are any defects or anomalies on the fabric surface. determination, - According to the data obtained from the rotation sensor, the knitting machine is in active production. the determination that it is not, - the generated segmentation outputs are processed using one or more post-processing filters evaluation and verification of candidate defect areas, - In case of error detection, the knitting machine is stopped via the relay system. stopping it in a controlled manner, - This includes the steps involved in creating an error log and transferring it to the monitoring interfaces.

11. This method complies with Claim 10 and its characteristic is that images are transmitted via a camera system at specific times. This involves taking the data at intervals or in a continuous flow as a process step.

12. This method complies with Claim 11 and its characteristic is that the acquired images are embedded in the computing unit. storing it in its temporary memory and comparing consecutive image frames. It includes the steps involved in the process. 35 13. This method complies with Claim 12 and its characteristic is that the acquired image data is processed using an encoder-decoder. Each pixel in the image is analyzed using a deep learning model based on the model. Generating segmentation outputs based on defective or flawless classification. It includes the process step.

14. This method complies with claim 13 and its characteristic is that it provides segmentation outputs on the image. Prior examination of identified candidate defect areas and specific image regions This involves optimizing the processing load by including a processing step.

15. This method complies with Claim 10 and is characterized by its adaptability to changes in ambient lighting or fabric. Threshold to reduce false detections caused by natural variations on the surface Dynamically adjusting the values ​​and comparison criteria is the process step. It includes.

16. This method complies with claim 15 and its characteristic is that the pulse frequency obtained from the rotation sensor is specific. by analyzing it over a period of time, it can be determined whether the knitting machine is in active production. It involves the process step of determining whether or not it exists.

17. This method complies with claim 16 and its characteristic is that the speed of the knitting machine is above a certain threshold value. In these cases, the image processing process must be kept active and the knitting machine must be running. deactivating the image processing process when it is stopped or out of production. It includes the process step.

18. This method complies with Claim 10 and its characteristic is that it is used during the transitions when the knitting machine stops and starts. Delayed activation of the image processing process to prevent false triggers. and involves the process step of running it with delayed deactivation logic.

19. This method complies with Claim 10 and is characterized by the use of multiple methods to improve the accuracy of error detection. Validation is performed on successive image frames and in segmentation outputs. The process involves verifying the identified candidate defect regions using post-processing filters. It includes the step.

20. This is a method compliant with Claim 19, characterized by its ability to recover embedded calculations if an error is detected. a control signal is generated by the unit and that control signal is transmitted to the machine The signal to stop is transmitted to the relay structure either without delay or with a predefined short delay. It includes the process step.

21. This method complies with Claim 20 and its characteristic feature is that it stops knitting via a machine stop relay structure. The electrical connection between the machine's +24V line and the ground line is made through a specific resistor. This involves a process step that stops the knitting machine by creating a false alarm.

22. This method complies with claim 21 and its characteristic feature is that it eliminates unnecessary processes after the transmission of the control signal. A specific waiting period is applied to the process to prevent repeated stop commands. It includes the step.

23. This method complies with Claim 10 and its characteristic feature is; information on the type of error and the time of its occurrence at the time of error detection. 35 timestamps, machine identification information, image data, and processed analysis outputs, a fault log consisting of segmentation outputs and fault validation results It includes the creation process step. 16 24. This method complies with claim 23 and its characteristic is that the generated error log can be monitored locally or remotely. transferring this information to the interfaces and providing the user with live view, error type and error time information. presentation, as well as display of segmentation outputs and defect verification information. It includes the process step.

25. This method complies with Claim 24 and is characterized by its use in long-term analysis of error logs. storage and identification of error occurrence trends based on those records, Identifying recurring problem sources, optimizing production parameters. and the process step of analyzing the segmentation outputs and defect verification results. It includes.

26. This method complies with Claim 25 and its characteristic is that it analyzes past errors on stored error logs. analysis of the data and segmentation outputs and defect verification results. This involves analyzing historical data related to the process.