Optical fiber automatic winding method, device and equipment and computer readable storage medium

Abstract

The invention discloses an optical fiber automatic winding method, device and equipment and a computer readable storage medium, and the method comprises the steps: carrying out the real-time analysis of collected image data in an optical fiber winding process through a trained visual model, so as to achieve the anomaly detection, and outputting an anomaly type; based on the obtained current winding working condition parameters and the abnormal types obtained through judgment, using a reinforcement learning strategy model to carry out operation processing so as to generate and give corresponding intervention control actions aiming at the fiber winding equipment, thereby constructing a closed-loop response mechanism from image abnormal recognition to control strategy output; the abnormal state in the optical fiber winding process can be sensed in time and controlled in a targeted mode, the problems that due to the fact that an existing winding method lacks real-time monitoring and correcting capacity, tiny abnormity cannot be sensed, and the optimal intervention opportunity is missed are solved, the production efficiency is improved, the yield is stabilized, and the production cost is reduced. And support is provided for large-scale consistent manufacturing of the optical fiber hydrophone.

Description

Technical Field

[0001] This application relates to the field of intelligent manufacturing, specifically to an automatic optical fiber winding method, apparatus, equipment, and computer-readable storage medium. Background Technology

[0002] Fiber optic hydrophones are underwater acoustic sensing devices based on the principles of interference or scattering effects, and are widely used in underwater listening, earthquake monitoring, oil and gas exploration, marine scientific surveys, and military reconnaissance. Compared with traditional electroacoustic hydrophones, fiber optic hydrophones have advantages such as high sensitivity, strong resistance to electromagnetic interference, strong environmental adaptability, and ease of networking, making them an important development direction for underwater sensing technology.

[0003] In related technologies, the close-packing winding process of optical fibers is one of the core key steps in the manufacturing of fiber optic hydrophones. Typically, slender optical fibers need to be wound uniformly and tightly onto flexible or solid sensing elements (such as elastic diaphragms, metal cylinders, or polymer structures) to construct an interference cavity or acoustic sensitive region. This process places extremely high demands on winding precision, tension consistency, and close-packing uniformity. The winding quality directly determines the hydrophone's performance indicators, such as sensitivity, frequency response range, linearity, and signal-to-noise ratio.

[0004] However, existing fiber optic hydrophone winding methods lack real-time monitoring and correction capabilities, making it impossible to detect minor anomalies during the winding process and miss the best intervention opportunity. This results in low production efficiency, large fluctuations in yield, and seriously affects the large-scale standardized manufacturing of fiber optic hydrophones. Summary of the Invention

[0005] This application provides an automatic optical fiber winding method, apparatus, equipment, and computer-readable storage medium, which can solve the technical problems of difficult identification of abnormalities, lag in control response, and unstable winding quality in traditional winding methods in related technologies.

[0006] In a first aspect, embodiments of this application provide an automatic optical fiber winding method, the automatic optical fiber winding method comprising: The system performs anomaly detection and discrimination on the acquired image data based on the trained visual model. If an anomaly is detected, the anomaly type is output. Based on the current winding operating parameters and the identified anomaly types, the reinforcement learning strategy model is used to provide intervention and control actions for the fiber winding equipment.

[0007] In conjunction with the first aspect, in one implementation, after providing the intervention control action for the fiber winding equipment in the reinforcement learning strategy model based on the current winding condition parameters and the determined anomaly type, the method further includes: The intervention and control actions of the fiber winding equipment are evaluated and rewarded through a manual feedback mechanism, and an interactive feedback dataset is accumulated. The reinforcement learning policy model is continuously trained and optimized based on the interactive feedback dataset.

[0008] In conjunction with the first aspect, in one implementation, the step of performing anomaly detection and discrimination on the acquired image data based on the trained visual model, and outputting the anomaly type if an anomaly is detected, further includes: The image data acquired during the fiber optic winding process is initially assessed. If there are obvious defects or blurriness, the image data is preprocessed.

[0009] In conjunction with the first aspect, in one implementation, the preprocessing includes image denoising, size normalization, grayscale enhancement, histogram equalization, or data augmentation.

[0010] In conjunction with the first aspect, in one implementation, before performing anomaly detection and discrimination on the acquired image data based on the trained visual model, and outputting the anomaly type if an anomaly is determined, the method further includes: Abnormal image data during the optical fiber winding process is collected, and the abnormal image data is classified into abnormal types to form an image dataset; The image dataset is imported into the neural network model for training, resulting in a trained neural network model.

[0011] In conjunction with the first aspect, in one implementation, the anomaly types include fiber optic loosening, fiber optic overlap, fiber optic offset, and fiber optic breakage. The intervention and control actions include pausing fiber winding, retracting the winding equipment's shaft by a fixed number of turns, increasing the tension of the winding equipment, decreasing the tension of the winding equipment, and calling for manual intervention.

[0012] In conjunction with the first aspect, in one embodiment, the current winding condition parameters include the fiber cabling position, the spindle speed of the winding equipment, the fiber tension value, the current number of fiber winding layers, and the current fiber winding position.

[0013] Secondly, embodiments of this application provide an automatic optical fiber winding device, the automatic optical fiber winding device comprising: The visual model recognition module is used to detect and identify anomalies in the acquired image data based on the trained visual model. If an anomaly is detected, the anomaly type is output. The reinforcement learning strategy generation module is used to input the current winding condition parameters and the identified anomaly type into the reinforcement learning strategy model, and to give the intervention control actions of the fiber winding equipment.

[0014] Thirdly, embodiments of this application provide an automatic optical fiber winding device, which includes a processor, a memory, and an automatic optical fiber winding program stored in the memory and executable by the processor. When the automatic optical fiber winding program is executed by the processor, it implements the steps of the automatic optical fiber winding method as described in some of the above embodiments.

[0015] Fourthly, embodiments of this application provide a computer-readable storage medium storing an automatic optical fiber winding program, wherein when the automatic optical fiber winding program is executed by a processor, it implements the steps of the automatic optical fiber winding method as described in some of the above embodiments.

[0016] The beneficial effects of the technical solutions provided in this application include: By using a trained visual model to analyze the acquired image data of the fiber optic winding process in real time, anomaly detection and anomaly type output are achieved. Then, based on the acquired current winding condition parameters and the anomaly type identified above as input information, a reinforcement learning strategy model is used for computation to generate and provide corresponding intervention and control actions for the fiber winding equipment. This constructs a closed-loop response mechanism from image anomaly recognition to control strategy output. It enables timely perception and targeted control of abnormal states during fiber optic winding, solving the problem that existing winding methods lack real-time monitoring and correction capabilities, resulting in the inability to detect minor anomalies and missing the best intervention opportunity. This is beneficial for improving production efficiency, stabilizing yield, and supporting the large-scale standardized manufacturing of fiber optic hydrophones. Attached Figure Description

[0017] Figure 1 This is a schematic flowchart of an embodiment of the automatic optical fiber winding method of this application; Figure 2 This is a schematic flowchart of another embodiment of the automatic optical fiber winding method of this application; Figure 3 This is a flowchart illustrating yet another embodiment of the automatic optical fiber winding method of this application. Figure 4 This is a schematic diagram of the training process for the Transformer-based visual model in this application; Figure 5 This is a schematic diagram of the training process for the reinforcement learning control strategy model in this application; Figure 6 This is a structural block diagram of the automatic optical fiber winding device of this application; Figure 7 This is a schematic diagram of the hardware structure of the automatic optical fiber winding equipment involved in the embodiments of this application. Detailed Implementation

[0018] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present application, and not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present application.

[0019] It's important to understand that fiber optic hydrophones are underwater acoustic sensing devices based on interference principles or scattering effects, widely used in underwater listening, earthquake monitoring, oil and gas exploration, marine scientific surveys, and military reconnaissance. Compared to traditional electroacoustic hydrophones, fiber optic hydrophones offer advantages such as high sensitivity, strong resistance to electromagnetic interference, high environmental adaptability, and ease of networking, making them an important development direction for underwater sensing technology.

[0020] In the manufacturing process of fiber optic hydrophones, the close-packing winding process of optical fibers is one of the core key steps. Typically, slender optical fibers need to be wound uniformly and tightly onto flexible or solid sensing elements (such as elastic diaphragms, metal cylinders, or polymer structures) to construct an interference cavity or acoustic sensitive region. This process places extremely high demands on winding precision, tension consistency, and close-packing uniformity. The winding quality directly determines the hydrophone's performance indicators, such as sensitivity, frequency response range, linearity, and signal-to-noise ratio.

[0021] However, existing fiber optic hydrophone winding methods lack real-time monitoring and correction capabilities, making it impossible to detect minor anomalies during the winding process and miss the best intervention opportunity. This results in low production efficiency, large fluctuations in yield, and seriously affects the large-scale standardized manufacturing of fiber optic hydrophones.

[0022] To make the objectives, technical solutions, and advantages of this application clearer, the embodiments of this application will be described in further detail below with reference to the accompanying drawings.

[0023] In a first aspect, embodiments of this application provide an automatic optical fiber winding method.

[0024] In one embodiment, reference is made to Figure 1 , Figure 1 This is a schematic flowchart of the first embodiment of the automatic optical fiber winding method of this application. Figure 1 As shown, the automatic optical fiber winding method includes: S100: Based on the trained visual model, perform anomaly detection and discrimination on the acquired image data. If an anomaly is detected, the anomaly type is output. S200: Based on the current winding condition parameters and the identified anomaly type, input reinforcement learning strategy model to give intervention control actions for the fiber winding equipment.

[0025] In this embodiment, a trained visual model is used to analyze the acquired image data of the fiber optic winding process in real time to detect anomalies and output the anomaly type. Then, based on the acquired current winding condition parameters and the anomaly type identified above as input information, a reinforcement learning strategy model is used for computation to generate and provide corresponding intervention and control actions for the fiber winding equipment. This establishes a closed-loop response mechanism from image anomaly recognition to control strategy output. This enables timely perception and targeted control of abnormal states during fiber optic winding, solving the problem that existing winding methods lack real-time monitoring and correction capabilities, resulting in the inability to detect minor anomalies and missing the best intervention opportunity. This is beneficial for improving production efficiency, stabilizing yield, and supporting the large-scale standardized manufacturing of fiber optic hydrophones.

[0026] Furthermore, in one embodiment, after S200, there is also S300, which includes the following steps: S300-1: The intervention control actions of the fiber winding equipment are evaluated and rewarded through a manual feedback mechanism to accumulate an interactive feedback dataset; S300-2: Continuously train and optimize reinforcement learning policy models based on interactive feedback datasets.

[0027] In this embodiment, after the reinforcement learning strategy model outputs the intervention control actions of the fiber winding equipment, a human feedback mechanism is used to evaluate the effectiveness of these actions, thereby accumulating an interactive feedback dataset containing the control actions and corresponding effect evaluations. Based on this dataset, the reinforcement learning strategy model is continuously trained and optimized, enabling the model to dynamically adjust the strategy according to the actual control effect, forming a closed-loop mechanism for strategy iterative optimization. By introducing reward and punishment signals from human feedback, the adaptive capability and control accuracy of the reinforcement learning strategy model to complex winding conditions are improved, ensuring that the intervention control actions generated by the model are more in line with actual production needs. This enhances the effectiveness and stability of abnormal correction during fiber winding, providing more reliable intelligent control support for the high-quality, large-scale manufacturing of fiber optic hydrophones.

[0028] Furthermore, in one embodiment, the human feedback mechanism constructs an immediate reward function in the reinforcement learning strategy model based on human scoring or production result evaluation.

[0029] In this embodiment, the effectiveness of the intervention control actions output by the reinforcement learning strategy model is evaluated through manual scoring or production result assessment. The evaluation result is transformed into a reward / penalty signal of the immediate reward function in the reinforcement learning strategy model, enabling the model to adjust the strategy parameters based on the signal. The immediate reward function constructed through manual feedback guides the reinforcement learning strategy model to iterate towards generating better control actions, improving the model's adaptability to complex winding conditions and the accuracy of control actions, ensuring the effectiveness of abnormal correction during the fiber optic winding process, and providing support for the high-quality manufacturing of fiber optic hydrophones.

[0030] Furthermore, in one embodiment, S100 further includes the following step: S100-1: Perform preliminary judgment on the image data acquired during the optical fiber winding process. If there are obvious defects or blurriness, perform image data preprocessing.

[0031] In this embodiment, before using the trained visual model to perform anomaly detection and discrimination on the acquired optical fiber winding process image data, the image data is first pre-judged. If there are obvious bad pixels or blurry images, preprocessing is performed to eliminate interference factors in the image that affect the accuracy of subsequent anomaly detection, ensuring that the image data input to the visual model has good quality. Through the preliminary screening and preprocessing of the image data, the quality of the image data input to the visual model is improved, providing a reliable data foundation for the visual model to achieve high-precision anomaly detection and discrimination. This ensures the accuracy of the subsequent control strategy generation based on the anomaly detection results, which is conducive to enhancing the reliability of anomaly identification in the optical fiber winding process and providing support for improving the winding quality of optical fiber hydrophones.

[0032] Furthermore, in one embodiment, the preprocessing includes image denoising, size normalization, grayscale enhancement, histogram equalization, or data augmentation.

[0033] In this embodiment, to address the potential issues in the acquired fiber optic winding process image data that may affect the recognition accuracy of subsequent visual models, such as noise, inconsistent dimensions, uneven grayscale distribution, and insufficient information, image data is preprocessed by image denoising to eliminate interference signals, size normalization to unify image specifications, grayscale enhancement and histogram equalization to optimize image contrast and detail recognition, and data augmentation to expand the diversity of effective samples. This preprocessing provides high-quality input data for the visual model. Through these preprocessing operations, the quality and consistency of the image data are improved, enhancing the accuracy and robustness of the visual model in identifying abnormal states during fiber optic winding. This lays a data foundation for high-precision execution of subsequent anomaly detection, thereby ensuring the reliability of real-time monitoring and correction of the fiber optic winding process and contributing to the stable improvement of fiber optic hydrophone winding quality.

[0034] Furthermore, in one embodiment, before S100, there is also S000, which includes the following steps: S000-1: Collect abnormal image data during the optical fiber winding process, classify the abnormal image data by abnormal type, and form an image dataset; S000-2: Import the image dataset into the neural network model for training to obtain a trained neural network model.

[0035] In this embodiment, image data containing various abnormal states during the optical fiber winding process are collected and classified according to the abnormality type to construct an image dataset. This dataset is then imported into a neural network model for training. The model learns the feature information of different abnormal types in the image data, forming the ability to identify abnormal states during optical fiber winding. This results in a well-trained neural network model that can be used for subsequent real-time anomaly detection. By constructing an image dataset containing classified abnormal types and training the neural network model, a model foundation is provided for high-precision anomaly detection and discrimination based on a visual model in the subsequent S100. This ensures that the visual model can effectively identify various abnormal states during the optical fiber winding process, laying the model support for real-time monitoring and correction of the optical fiber winding process, and helping to improve the stability of the optical fiber hydrophone winding quality.

[0036] Furthermore, in one embodiment, the abnormality types include fiber loosening, fiber overlap, fiber offset, and fiber breakage; the intervention control actions include pausing fiber winding, retracting the winding equipment's shaft by a fixed number of turns, increasing the tension of the winding equipment, decreasing the tension of the winding equipment, and calling for manual processing.

[0037] In this embodiment, the possible anomalies during fiber winding are defined as loose fiber winding, fiber overlap, fiber offset, and fiber breakage. Intervention and control actions are limited to pausing fiber winding, retracting the winding equipment's shaft by a fixed number of turns, increasing the tension of the winding equipment, decreasing the tension of the winding equipment, and calling for manual intervention. After identifying the specific anomaly types using a trained visual model, and combining this with the current winding conditions, the reinforcement learning strategy model outputs corresponding actions from the defined intervention and control actions, forming a precise control link for specific anomalies. By clearly defining the correspondence between anomaly types and intervention and control actions, the targeting of anomaly identification and the adaptability of control actions are improved. This ensures that the reinforcement learning strategy model can output effective intervention measures based on specific anomaly types, improving the accuracy and efficiency of anomaly correction during fiber winding and helping to ensure the consistency of fiber optic hydrophone winding quality.

[0038] Furthermore, in one embodiment, the current winding condition parameters include the fiber cabling position, the spindle speed of the winding equipment, the fiber tension value, the current number of fiber winding layers, and the current fiber winding position.

[0039] In this embodiment, the fiber optic cable position, the spindle speed of the winding equipment, the fiber tension value, the current number of fiber winding layers, and the current fiber winding position are used as current winding condition parameters. These parameters, along with the anomaly type identified by the visual model, are input into the reinforcement learning strategy model. This allows the model to combine specific working condition information and anomalies during the winding process to generate intervention and control actions adapted to the current state, thus constructing a strategy decision-making basis based on multi-dimensional working condition parameters and anomaly types. By incorporating the aforementioned specific winding condition parameters, the comprehensiveness and accuracy of the reinforcement learning strategy model's perception of the current winding state are improved, ensuring that the output intervention and control actions are highly matched with the actual winding conditions. This enhances the pertinence and effectiveness of anomaly correction, further ensuring the quality of fiber optic winding and supporting the high-quality manufacturing of fiber optic hydrophones.

[0040] On the other hand, this application proposes an automatic fiber winding method based on visual model recognition and reinforcement learning feedback control, see [link to relevant documentation]. Figure 2 This includes the following steps: Step 101: Set up industrial cameras and light sources to periodically take overhead photos of the fiber winding area from a certain angle; Step 102: Perform image anomaly detection using a Transformer-based visual model; Step 103: If the image is normal, continue winding the fiber; if the image is abnormal, output adjustment control commands through the reinforcement learning policy model. Step 104: The fiber winding device executes the control command and continues winding.

[0041] The following example, a specific automatic fiber winding process based on visual model recognition and reinforcement learning feedback control, will be used to illustrate the implementation process of the embodiments of this application in detail.

[0042] like Figure 3 As shown, this is a flowchart of a preferred embodiment of the automatic fiber winding method based on visual model recognition and reinforcement learning feedback control in a certain area. The automatic fiber winding process in this embodiment includes: Step 201: Collect abnormal image data during the fiber winding process. Illuminate the fiber winding area with a light source to make the image clear and distinguishable. The industrial camera takes pictures of the image at a fixed cycle. Step 202: After the industrial control computer acquires the image, it will perform a preliminary judgment on the image to check whether the data is clean and whether there are obvious bad pixels or blur. If there is an unclean image, proceed to step 203 for data preprocessing. If the image is clean, proceed to step 204 for anomaly detection. Step 203: Preprocess the image, mainly including: image denoising, size normalization, grayscale enhancement or histogram equalization, and data augmentation. After processing, proceed to step 204. Step 204: Based on the Transformer visual model, perform anomaly recognition on the fiber optic winding image. This model takes image patches (emedding) as input, performs global modeling through a multi-layer multi-head self-attention mechanism, and finally outputs the category or anomaly probability of the image. Step 205: Determine whether there is a fiber winding abnormality based on the output of the visual model. If there is no abnormality, proceed to step 206 without intervention and continue winding the fiber. If there is an abnormality, proceed to step 207 and intervene and control the fiber using a reinforcement learning strategy. Step 206: The system maintains its current state, and the fiber winding continues without triggering any control commands; Step 207: Collect state information and output control commands through a reinforcement learning policy model; Step 208: The fiber winding device executes the control action commands output by the reinforcement learning strategy to repair abnormal winding.

[0043] Figure 4 The training process of the visual model is presented, and its construction method mainly includes three stages: image encoding, Transformer feature extraction, and classification output, as detailed below: First, input image Divided into quantities of Image patches, each patch being [size missing]. Linearly map each patch to Dimensional embedding space:

[0044] in, It is the learned embedding matrix.

[0045] Add position encoding Preserving spatial order information, the final input sequence with added positional encoding is:

[0046] Finally, the input sequence is formed. It is then fed into the Transformer backbone network.

[0047] Secondly, the Transformer backbone consists of multiple stacked encoder layers, each of which includes the following two key substructures: (1) Multi-head self-attention mechanism: In each layer, the input sequence is first linearly transformed to obtain the query. ,key ,value :

[0048] Single-head attention is:

[0049] Multi-head attention splicing mapping:

[0050] in, , , and These are learnable weight parameters.

[0051] (2) Feedforward Network: The feature vector at each location is nonlinearly mapped through two fully connected layers.

[0052] The entire encoder layer has a residual connection structure:

[0053]

[0054] After stacking The layer encoder module outputs a high-dimensional semantic feature representation. ; Finally, the features are classified and the average representation of all tokens in the sequence is taken and input into the classifier head to output the anomaly category prediction result:

[0055] The probability distribution of each anomaly type is output after passing through a linear layer and softmax:

[0056] in, For the anomaly category prediction results, This represents the number of abnormal types (e.g., normal, excessive fiber winding gap, fiber overlap, fiber breakage).

[0057] The standard cross-entropy loss function is used for supervised training.

[0058] in, One-hot encoding of the real label. These are the model's predicted values.

[0059] The AdamW optimizer is used for parameter updates during training.

[0060] in, These are the current model parameters. For the next updated model parameters, It's the learning rate. It is a first-order moment estimate. It is a second-order moment estimate. It is a constant term, and to prevent the denominator from being zero, it is often set to zero. .

[0061] The training process employs cosine annealing to update the learning rate, gradually decaying it from its initial value. This helps improve convergence quality and avoid oscillations.

[0062] in, yes Learning rate at any given moment It is the initial learning rate. It is the minimum learning rate. It is the total training time.

[0063] After training, the model is exported in ONNX or TensorRT format and deployed to edge devices to achieve real-time image detection and anomaly recognition during the fiber winding process. The output is used by the control system to determine whether to call reinforcement learning strategies for correction.

[0064] Figure 5 The training process of the reinforcement learning policy model is presented. First, the state space and action space are set, and the current winding condition parameters (such as fiber cabling position, spindle speed, fiber tension value, current winding layer number, current winding position, etc.) are collected and combined with the visual recognition results to form a state vector.

[0065] in, Represents the position of the ribbon cable. Represents spindle speed, Represents tension value, Represents the current number of winding layers, Represents the previous action, The code representing the anomaly type in visual recognition; The reinforcement learning policy model outputs action numbers: ,exist At any given time, 0 represents continuing winding, 1 represents pausing winding, 2 represents the shaft retracting a fixed number of turns, 3 represents increasing tension, 4 represents decreasing tension, and 5 represents calling for manual intervention.

[0066] Secondly, a reinforcement learning reward function is defined to use the policy to learn reasonable action preferences. The reward mainly covers three aspects:

[0067] in, The reward for anomaly correction is intended to encourage the strategy to correct visually identified winding anomalies in a timely and effective manner. This represents an efficiency incentive, the purpose of which is to encourage strategies to maintain the winding rhythm as much as possible while ensuring quality, thereby improving production efficiency. This indicates a penalty, which is a behavior that may lead to more serious consequences, such as: the anomaly being exacerbated instead of being fixed, incorrect actions requiring human intervention, or the controller issuing conflicting or invalid commands. , , These are pre-defined weighting factors.

[0068] Finally, using classic deep reinforcement learning algorithms, such as DQN, DDPG, PPO, A3C, and SAC, an iterative training strategy based on interactive data is employed to provide action instructions for the winding device under abnormal conditions, thereby correcting the winding abnormalities.

[0069] As can be seen, the embodiments of this application have the following beneficial effects: Firstly, the embodiments of this application integrate a visual model based on the Transformer architecture, which can achieve high-precision identification of multiple types of anomalies (such as loose winding, overlap, offset, fiber breakage, etc.) during the automatic fiber winding process. Compared with traditional image detection methods, it has stronger generalization ability and anomaly perception ability, and greatly improves the controllability and consistency of winding quality.

[0070] Secondly, the embodiments of this application propose a control strategy based on reinforcement learning feedback mechanism, which can automatically output adjustment actions, such as pause, tension adjustment or retraction operation, according to the visual recognition results. It has autonomous correction and dynamic response capabilities, effectively reducing manual intervention and improving the level of automation and intelligence.

[0071] Third, the entire method has the ability to continuously iterate and optimize. After actual deployment, it can continue to learn and improve the performance of the strategy. It is suitable for complex and ever-changing production environments, and is especially suitable for close-packed winding scenarios in the manufacturing process of high-precision fiber optic devices such as fiber optic hydrophones. It has good industrial application prospects and promotion value.

[0072] Secondly, embodiments of this application also provide an automatic optical fiber winding device.

[0073] In one embodiment, reference is made to Figure 6 , Figure 6 This is a functional module diagram of an embodiment of the automatic fiber optic winding device of this application. Figure 6 As shown, the automatic fiber winding device includes: a visual model recognition module, which is used to detect and judge anomalies in the acquired image data based on a trained visual model, and output the anomaly type if an anomaly is judged; and a reinforcement learning strategy generation module, which is used to input the current winding condition parameters and the judged anomaly type into the reinforcement learning strategy model to give the intervention control action of the fiber winding equipment.

[0074] In this embodiment, the visual model recognition module uses a trained visual model to perform real-time anomaly detection and discrimination on the collected image data of the optical fiber winding process. When an anomaly is detected, the specific anomaly type is output. At the same time, the reinforcement learning strategy generation module receives the current winding condition parameters and the above-mentioned anomaly type as input, and after processing with the help of the reinforcement learning strategy model, it gives the intervention control action of the winding equipment. The two modules work together to build a functional link from image anomaly recognition to control action output. This realizes the automatic recognition of abnormal states in the optical fiber winding process and the generation of targeted control actions, which solves the problem of the inability to monitor and accurately correct deviations in real time due to the lack of such collaborative functions in existing devices. This is conducive to improving the automation and intelligence level of optical fiber winding and provides device support for the high-quality and large-scale manufacturing of optical fiber hydrophones.

[0075] Furthermore, in one embodiment, the automatic fiber winding device further includes an image acquisition module, which includes a light source, a high-speed camera, and a power supply, for acquiring images of the fiber winding process.

[0076] Furthermore, in one embodiment, the automatic fiber winding device also includes a state perception and data processing module, which is located in the industrial control computer, and is used to collect current winding condition information and provide it to the reinforcement learning strategy generation module.

[0077] Furthermore, in one embodiment, the automatic optical fiber winding device further includes a control execution module, which includes a rotating shaft, a wire laying mechanism, tension adjustment and other control execution devices for performing control actions.

[0078] Thirdly, embodiments of this application provide an automatic optical fiber winding device, which can be a personal computer (PC), laptop computer, server, or other device with data processing capabilities.

[0079] Reference Figure 7 , Figure 7This is a schematic diagram of the hardware structure of the automatic optical fiber winding device involved in the embodiments of this application. In the embodiments of this application, the automatic optical fiber winding device may include a processor, a memory, a communication interface, and a communication bus.

[0080] The communication bus can be of any type and is used to interconnect the processor, memory, and communication interface.

[0081] Communication interfaces include input / output (I / O) interfaces, physical interfaces, and logical interfaces used for interconnecting devices within the automated fiber optic winding equipment, as well as interfaces used for interconnecting the automated fiber optic winding equipment with other devices (such as other computing devices or user equipment). Physical interfaces can be Ethernet interfaces, fiber optic interfaces, ATM interfaces, etc.; user equipment can be displays, keyboards, etc.

[0082] Memory can be various types of storage media, such as random access memory (RAM), read-only memory (ROM), non-volatile RAM (NVRAM), flash memory, optical storage, hard disk, programmable ROM (PROM), erasable PROM (EPROM), electrically erasable PROM (EEPROM), etc.

[0083] The processor can be a general-purpose processor, which can call the automatic fiber winding program stored in the memory and execute the automatic fiber winding method provided in the embodiments of this application. For example, the general-purpose processor can be a central processing unit (CPU). The method executed when the automatic fiber winding program is called can be referred to in the various embodiments of the automatic fiber winding method of this application, and will not be repeated here.

[0084] Those skilled in the art will understand that Figure 7 The hardware structure shown does not constitute a limitation of this application and may include more or fewer components than shown, or combine certain components, or have different component arrangements.

[0085] Fourthly, embodiments of this application also provide a readable storage medium.

[0086] The present application has a readable storage medium storing an automatic optical fiber winding program, wherein when the automatic optical fiber winding program is executed by a processor, it implements the steps of the automatic optical fiber winding method described above.

[0087] The method implemented when the automatic optical fiber winding program is executed can be referred to in various embodiments of the automatic optical fiber winding method of this application, and will not be repeated here.

[0088] It should be noted that the sequence numbers of the embodiments in this application are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments.

[0089] The terms "comprising" and "having," and any variations thereof, in the specification, claims, and accompanying drawings of this application are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or apparatus that includes a series of steps or units is not limited to the listed steps or units, but may optionally include steps or units not listed, or may optionally include other steps or units inherent to such process, method, product, or apparatus. The terms "first," "second," and "third," etc., are used to distinguish different objects, etc., and do not indicate a sequence, nor do they limit "first," "second," and "third" to different types.

[0090] In the description of the embodiments of this application, terms such as "exemplary," "for example," or "for instance" are used to indicate examples, illustrations, or explanations. Any embodiment or design described as "exemplary," "for example," or "for instance" in the embodiments of this application should not be construed as being more preferred or advantageous than other embodiments or designs. Specifically, the use of terms such as "exemplary," "for example," or "for instance" is intended to present the relevant concepts in a concrete manner.

[0091] In the description of the embodiments of this application, unless otherwise stated, " / " means "or". For example, A / B can mean A or B. The "and / or" in the text is merely a description of the relationship between related objects, indicating that there can be three relationships. For example, A and / or B can mean: A exists alone, A and B exist simultaneously, and B exists alone. In addition, in the description of the embodiments of this application, "multiple" means two or more.

[0092] In some processes described in the embodiments of this application, multiple operations or steps are included in a specific order. However, it should be understood that these operations or steps may not be executed in the order they appear in the embodiments of this application, or they may be executed in parallel. The sequence number of the operation is only used to distinguish different operations, and the sequence number itself does not represent any execution order. In addition, these processes may include more or fewer operations, and these operations or steps may be executed sequentially or in parallel, and these operations or steps may be combined.

[0093] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) as described above, and includes several instructions to cause a terminal device to execute the methods described in the various embodiments of this application.

[0094] The above are merely preferred embodiments of this application and do not limit the patent scope of this application. Any equivalent structural or procedural transformations made using the content of this application's specification and drawings, or direct or indirect applications in other related technical fields, are similarly included within the patent protection scope of this application.

Claims

1. An automatic optical fiber winding method, characterized in that, The automatic optical fiber winding method includes: The system performs anomaly detection and discrimination on the acquired image data based on the trained visual model. If an anomaly is detected, the anomaly type is output. Based on the current winding operating parameters and the identified anomaly types, the reinforcement learning strategy model is used to provide intervention and control actions for the fiber winding equipment.

2. The automatic optical fiber winding method as described in claim 1, characterized in that, In the reinforcement learning strategy model based on the current winding condition parameters and the identified anomaly type, after giving the intervention control action of the fiber winding equipment, it also includes: The intervention and control actions of the fiber winding equipment are evaluated and rewarded through a manual feedback mechanism, and an interactive feedback dataset is accumulated. The reinforcement learning policy model is continuously trained and optimized based on the interactive feedback dataset.

3. The automatic optical fiber winding method as described in claim 1, characterized in that, The method of performing anomaly detection and discrimination on the acquired image data based on the trained visual model, and outputting the anomaly type if an anomaly is detected, also includes: The image data acquired during the fiber optic winding process is initially assessed. If there are obvious defects or blurriness, the image data is preprocessed.

4. The automatic optical fiber winding method as described in claim 3, characterized in that, The preprocessing includes image denoising, size normalization, grayscale enhancement, histogram equalization, or data augmentation.

5. The automatic optical fiber winding method as described in claim 1, characterized in that, Before performing anomaly detection and discrimination on the acquired image data based on the trained visual model, and outputting the anomaly type if an anomaly is detected, the following steps are also included: Abnormal image data during the optical fiber winding process is collected, and the abnormal image data is classified into abnormal types to form an image dataset; The image dataset is imported into the neural network model for training, resulting in a trained neural network model.

6. The automatic optical fiber winding method as described in claim 1, characterized in that, The anomaly types include loose fiber winding, fiber overlap, fiber offset, and fiber breakage; The intervention and control actions include pausing fiber winding, retracting the winding equipment's shaft by a fixed number of turns, increasing the tension of the winding equipment, decreasing the tension of the winding equipment, and calling for manual intervention.

7. The automatic optical fiber winding method as described in claim 1, characterized in that, The current winding conditions parameters include the fiber cabling position, the spindle speed of the winding equipment, the fiber tension value, the current number of fiber winding layers, and the current fiber winding position.

8. An automatic optical fiber winding device, characterized in that, The automatic optical fiber winding device includes: The visual model recognition module is used to detect and identify anomalies in the acquired image data based on a trained visual model. If an anomaly is detected, the anomaly type is output. The reinforcement learning strategy generation module is used to input the current winding condition parameters and the identified anomaly type into the reinforcement learning strategy model, and to give the intervention control actions of the fiber winding equipment.

9. An automatic optical fiber winding device, characterized in that, The automatic fiber winding device includes a processor, a memory, and an automatic fiber winding program stored in the memory and executable by the processor, wherein when the automatic fiber winding program is executed by the processor, it implements the steps of the automatic fiber winding method as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores an automatic optical fiber winding program, wherein when the automatic optical fiber winding program is executed by a processor, it implements the steps of the automatic optical fiber winding method as described in any one of claims 1 to 7.

Images