An optical fiber dynamic fatigue parameter testing device and method

Abstract

The application provides a kind of optical fiber dynamic fatigue parameter testing device, method, method includes: using high-speed camera to shoot for training and test image set;Second, image recognition model selection can adopt YOLOV5s Similar AI target recognition model;Third, model training, ready pre-training model, utilize special data set to carry out model recognition and tracking ability training, require can real-time identification optical fiber image, once optical fiber breaks, can first time output fracture signal, reach requirement by testing can be used for system integration, otherwise need to carry out again training, until reach requirement.The application meets the industry for the rapid test evaluation demand of optical fiber fatigue characteristics.

Description

Technical Field

[0001] This invention belongs to the field of optical fiber dynamic fatigue parameter testing, and particularly relates to an optical fiber dynamic fatigue parameter testing device and method. Background Technology

[0002] Fiber optic sensing technology originated in 1977 and has developed rapidly alongside fiber optic communication technology. It is a crucial indicator of a nation's level of informatization. Fiber optic sensing technology has been widely used in both military and civilian sectors, including defense, aerospace, and automotive electronics, and has a broad market. Fiber optic sensing encompasses two functions: sensing and transmitting external signals (the measured quantity). Sensing (or being sensitive) refers to the change in physical parameters of the light wave propagating in the fiber optic cable, such as intensity (power), wavelength, frequency, phase, and polarization state, caused by external signals according to their changing patterns. Measuring these changes in optical parameters is equivalent to "sensing" the change in the external signal. This "sensing" is essentially the real-time modulation of the light wave propagating in the fiber optic cable by the external signal. Transmission refers to the fiber optic cable transmitting the light wave modulated by the external signal to a photodetector for detection. The external signal is extracted from the light wave and processed as needed, i.e., demodulation. Therefore, fiber optic sensing technology includes two aspects: modulation and demodulation. That is, modulation techniques (or loading techniques) on how external signals (measured) modulate the optical wave parameters in the optical fiber, and demodulation techniques (or detection techniques) on how to extract external signals (measured) from the modulated optical waves.

[0003] Whether for transmission or sensing, optical fiber is the primary carrier. In practical applications, the ability of optical fiber to resist corrosion under specific stress environments—that is, its resistance to chemical or environmental corrosion under long-term stress—is extremely important. Therefore, GB / T15972.33-2024, the specification for optical fiber test methods, Part 33: Measurement methods and test procedures for mechanical properties—stress corrosion susceptibility parameters—provides detailed requirements on how to test these parameters and mainly specifies five test methods for stress corrosion susceptibility parameters.

[0004] According to standard requirements, any mechanical testing of optical fibers should determine fracture stress and fatigue properties under simulated conditions as close as possible to actual applications. For many practical optical fiber sensing applications, sensitivity enhancement designs are employed to improve fiber sensing sensitivity, such as winding a certain length of optical fiber onto a cylinder, in which case the fiber needs to be in a long-term bending state. Therefore, the two-point bending method for measuring dynamic stress corrosion susceptibility parameters is very suitable for testing sensing optical fibers. Currently, the industry mainly follows GB / T15972.33-2024 for testing, and the two-point bending device in this standard is shown in the figure below.

[0005] This device measures the strain / stress required for an optical fiber to break at two points by measuring the spacing between the pressure plates at the point of breakage. In the experiment, the pressure plates broke the optical fiber at four different pressure levels: 1 μm / s, 10 μm / s, 100 μm / s, and 1000 μm / s. Simultaneously, an optical fiber breakage detection system recorded the breakage distance at the point of breakage, thereby enabling the testing of dynamic stress corrosion sensitivity parameters.

[0006] The standard specifies three methods for detecting fiber optic breakage: ① Using an acoustic emission detector or sensor to detect fiber breakage. The breakage signal is transmitted via a computer to stop the movement of the pressure plate, and the spacing between the pressure plates is displayed when the breakage occurs. ② Mounting a force (pressure) sensor on a fixed pressure plate and connecting it to a suitable signal processing device to measure the force applied to the fiber during the test. When the fiber breaks, the force drops to zero, thus providing a method for detecting fiber breakage. ③ Providing light to the fiber in the test and monitoring the light output signal. When the fiber breaks, the light transmission signal disappears.

[0007] Currently, the fiber optic two-point bending test system that is commercially available and widely used in the industry is the TP-2 from FIBERSIGMA. This system uses the acoustic detection method in method ① to detect the acoustic signal of fiber breakage.

[0008] In summary, the shortcomings of existing technologies are:

[0009] I. Existing fiber optic two-point bending test systems primarily employ an acoustic emission detector or sensor to detect fiber breakage. The breakage signal is transmitted via a computer to stop the movement of the pressure plate, and the spacing between the pressure plates is displayed at the time of breakage. This method is severely susceptible to ambient noise interference, making it prone to false alarms. That is, even when the fiber is not broken, the sensor detects relevant noise, triggering a false alarm and generating a false breakage signal. This problem is particularly severe during low-speed testing.

[0010] Second, existing fiber optic two-point bending test systems only record the detected acoustic signals when detecting multiple fiber breaks, without recording which channel of the fixture the fiber broke. The signals cannot be traced back to their source. Often, when multiple samples are tested simultaneously, the equipment records breakage signals beyond those of the samples, requiring retesting, which affects test accuracy and efficiency.

[0011] Third, the recommended methods ② and ③ in the standard have poor operability and are difficult to implement. For example, method ③ requires passing light through the optical fiber and monitoring the light output signal. The number of tests at each speed is at least 15, which requires a lot of fusion splicing work, is time-consuming, has poor operability, and will affect the stress state of the sample after fusion splicing. Improper handling may affect the test results. Summary of the Invention

[0012] The purpose of this invention is to propose a device and method for testing the dynamic fatigue parameters of optical fibers. Addressing the shortcomings of current acoustic signal detection technology used in the two-point bending fatigue characteristic testing of optical fibers, this invention proposes a system and method for testing the fatigue characteristics of optical fibers using intelligent image detection. A high-speed camera captures the state of the optical fiber sample in real time during the test. An algorithm identifies the fiber state in real time and records the image of the fiber breakage moment and the spacing of the pressure plates, effectively reducing misjudgments. When testing multiple samples simultaneously, each fiber can be accurately recorded, allowing for traceability and significantly improving testing efficiency and accuracy, thus meeting the industry's need for rapid testing and evaluation of optical fiber fatigue characteristics.

[0013] To achieve the above objectives, a method for testing dynamic fatigue parameters of optical fibers is provided in a first aspect of the present invention, comprising the following steps:

[0014] Step 1: Acquire fiber optic images and pressure plate spacing using a high-speed camera; wherein, the pressure plate spacing is the distance between the inner sides of the clamp;

[0015] Step two is selected from one of (I)-(IV):

[0016] (I) Select a target recognition model, train the target recognition model based on the optical fiber image and the pressure plate spacing, and output the pressure plate spacing and the frame index where the breakage occurs at this time;

[0017] (II) An anomaly detection model is adopted, with good quality optical fiber as the normal value of the model and optical fiber breakage as the abnormal value. If the anomaly detection model outputs an anomaly, the pressure plate spacing and the frame index of the breakage occurrence are output at this time.

[0018] (III) An anomaly detection model is adopted, with good quality optical fiber as the normal value of the model and optical fiber breakage as the abnormal value. If the anomaly detection model outputs an anomaly, the pressure plate spacing and the frame index of the breakage occurrence are output and input into the detection model. Good quality optical fiber and broken optical fiber are marked, and the broken optical fiber number of the broken optical fiber and the pressure plate spacing corresponding to the frame are output.

[0019] (IV) An anomaly detection model is adopted, with good quality optical fiber as the normal value and optical fiber breakage as the abnormal value. If the anomaly detection model outputs an anomaly, the clamping plate spacing and the frame index of the breakage occurrence are output and input into the detection model. Good quality optical fiber and broken optical fiber are marked, and the broken optical fiber number and the clamping plate spacing corresponding to the frame are output. Based on the broken optical fiber number and the clamping plate spacing corresponding to the frame, the optical fiber is segmented using the segmentation method with the broken sample number. The number of connected regions of the optical fiber is calculated. When the number of connected regions is less than 1, the clamping plate spacing and the frame index of the breakage occurrence are output.

[0020] Step 3: Once the number of valid samples for each speed setting reaches the agreed threshold, the dynamic fatigue parameters are fitted by combining the pressure plate spacing and the fracture occurrence frame index to generate the fiber dynamic fatigue parameters.

[0021] Furthermore, the spacing between the pressure plates is calculated as follows:

[0022] In a static state, record the distance to the inside of the fixture and the number of pixels inside the fixture in the image;

[0023] If the fixture moves, the Hough transform is used to identify the straight line inside the fixture in the acquired image, and the distance inside the fixture is automatically calculated.

[0024] If the distance inside the clamp is greater than a preset threshold and the optical fiber is unbroken, then only the algorithm is used to calculate the distance; if the distance inside the clamp is less than the preset threshold, then the target recognition model is used to identify the state of the optical fiber.

[0025] Furthermore, the execution step (I) is as follows:

[0026] The optical fiber image sequence was acquired by a high-speed camera, and the pressure plate spacing sequence was obtained corresponding to the distance inside the fixture.

[0027] The optical fiber image is input into the classification model, which outputs the probability of optical fiber breakage, ranging from 0 to 1.

[0028] Based on the probability of fiber breakage, the geometric-probability coupling score is calculated by combining the current and previous time plate spacing, and the frame index corresponding to the maximum score is found.

[0029] The frame index corresponding to the maximum score and the pressure plate spacing corresponding to the frame index corresponding to the maximum score are used as the dynamic fatigue parameters of the optical fiber.

[0030] Furthermore, the structure of the classification model is as follows: a pre-trained and frozen convolutional neural network with a ResNet-18 backbone, and a Squeeze-and-Excitation attention mechanism added to each residual module, so that the classification model pays attention to the slender region of the optical fiber when extracting features.

[0031] Wherein, the geometry-probability coupling fraction The calculation is as follows:

[0032]

[0033] in, It is a moment The probability of fracture; and These are the pressure plate spacing at the current and previous moments, respectively; the second term in the fraction represents the normalized shrinkage rate of the spacing between two adjacent frames; It is a fixed weighting factor.

[0034] Furthermore, the anomaly detection model employs a convolutional autoencoder structure; its encoding part consists of multiple layers of convolution, batch normalization, and nonlinear activation layers, used to progressively compress the features of the input image; the decoding part reconstructs an image of the same size as the input image through deconvolution and upsampling operations.

[0035] Then, the specific execution steps of (II) are as follows:

[0036] The optical fiber image sequence was acquired by a high-speed camera, and the pressure plate spacing sequence was obtained corresponding to the distance inside the fixture.

[0037] The fiber optic image is input into the anomaly detection model to obtain the corresponding reconstructed image;

[0038] Anomaly scores are calculated based on the corresponding reconstructed images, as well as the corresponding fiber optic images and pressure plate spacing.

[0039] When the anomaly score is at its maximum, the corresponding frame index is selected as the moment when the break occurs;

[0040] By combining the frame index corresponding to the maximum abnormal score and the corresponding pressure plate spacing, the dynamic fatigue parameters of the optical fiber are generated.

[0041] Furthermore, the abnormal score The calculation is as follows:

[0042] ;

[0043] in, and These represent the pixel positions of the image. Input values ​​and reconstructed values; It is a normalized weight map generated during the calibration stage, with larger values ​​taken in the bottom region where the fiber stress is concentrated; and These represent the spacing between two adjacent pressure plates; This is the coupling weight constant, used to balance the contributions of visual error and geometric shrinkage.

[0044] Furthermore, if the output of the target recognition model and / or anomaly detection model and / or detection model passes the test, the current target recognition model and / or anomaly detection model and / or detection model is used for system integration; if the output of the target recognition model and / or anomaly detection model and / or detection model fails the test, the current target recognition model and / or anomaly detection model and / or detection model is retrained until it meets the requirements for passing the test.

[0045] Furthermore, the detection model adopts a single-stage object detection network of the YOLO series during the training phase, with CSPDarknet as the backbone and FPN combined with PAN structure for the feature fusion part. The detection head outputs the position parameters and classification probability of the candidate boxes.

[0046] The specific steps of outputting the current clamping plate spacing and the frame index of the breakage occurrence, inputting them into the detection model, marking good optical fibers and broken optical fibers, and outputting the breakage fiber number of the broken optical fiber and the clamping plate spacing corresponding to that frame include:

[0047] The fracture occurrence frame index is input into the detection model, which outputs a set of candidate boxes. ,in Indicates the position of the candidate box. This indicates the confidence level at which the current candidate box is judged as broken;

[0048] Within each candidate box, the candidate box region is first binarized and thinned to obtain the fiber skeleton, and then the connectivity index of the region is calculated. , defined as the ratio of the maximum number of connected channel pixels to the total number of foreground pixels within the bounding box, and taking into account the shrinkage of the pressure plate spacing, the pressure plate spacing at this time is... Initial reference distance from the device Perform normalization, and use it as a geometric factor in the determination;

[0049] Based on the connectivity index Initial reference spacing and the confidence level of the current candidate box being judged as broken. Calculate the detection coupling score of the candidate boxes :

[0050] ;

[0051] in, and These are the weight constants, used to balance the contributions of geometric shrinkage and connectivity penalty;

[0052] Within the same frame, calculate the detection coupling score for all candidate boxes. The largest value is selected to determine the broken fiber number, and then output.

[0053] Further, the specific execution steps of segmenting the fiber based on the broken fiber number and the clamping plate spacing corresponding to the frame, using the broken sample number and the segmentation method, calculating the number of connected regions of the fiber, and when the number of connected regions is less than 1, inputting the clamping plate spacing and the frame index of the broken fiber are as follows:

[0054] Based on the broken fiber number and the pressure plate spacing corresponding to the frame, a binary mask is obtained using a pre-trained segmentation model.

[0055] Based on the binary mask, the number of connected regions is calculated using the connected component algorithm; if the number of connected regions = 1, then the pixel region of each optical fiber is completely connected.

[0056] Calculate the breakage determination value based on the number of connected regions:

[0057] ;

[0058] in, This is the fracture determination value. It is the number of connected components; To normalize the gap ratio, a binary mask is used. The skeleton is extracted and the pixel distance between the endpoints of the main fracture is measured, and then normalized to obtain the ratio with the diagonal length of the region. The bridging ratio is the proportion of new pixels added after performing a closing operation with a fixed radius on the mask. It is the spacing between the pressure plates of the broken frame. This is the initial baseline spacing. Four weights. It is fixed during deployment and determined through calibration experiments;

[0059] If the breakage determination value is greater than the preset normal frame level, then the optical fiber is determined to be broken at the current breakage occurrence frame index.

[0060] To achieve the above objectives, a second aspect of the present invention provides an optical fiber dynamic fatigue parameter testing device, comprising the following modules:

[0061] An image acquisition unit is used to acquire fiber optic images and pressure plate spacing via a high-speed camera; wherein, the pressure plate spacing is the distance between the inner sides of the clamp;

[0062] The image analysis unit is used to perform one of (I)-(IV):

[0063] (I) Select a target recognition model, train the target recognition model based on the optical fiber image and the pressure plate spacing, and output the pressure plate spacing and the frame index where the breakage occurs at this time;

[0064] (II) An anomaly detection model is adopted, with good quality optical fiber as the normal value of the model and optical fiber breakage as the abnormal value. If the anomaly detection model outputs an anomaly, the pressure plate spacing and the frame index of the breakage occurrence are output at this time.

[0065] (III) An anomaly detection model is adopted, with good quality optical fiber as the normal value of the model and optical fiber breakage as the abnormal value. If the anomaly detection model outputs an anomaly, the pressure plate spacing and the frame index of the breakage occurrence are output and input into the detection model. Good quality optical fiber and broken optical fiber are marked, and the broken optical fiber number of the broken optical fiber and the pressure plate spacing corresponding to the frame are output.

[0066] (IV) An anomaly detection model is adopted, with good quality optical fiber as the normal value and optical fiber breakage as the abnormal value. If the anomaly detection model outputs an anomaly, the clamping plate spacing and the frame index of the breakage occurrence are output and input into the detection model. Good quality optical fiber and broken optical fiber are marked, and the broken optical fiber number and the clamping plate spacing corresponding to the frame are output. Based on the broken optical fiber number and the clamping plate spacing corresponding to the frame, the optical fiber is segmented using the segmentation method with the broken sample number. The number of connected regions of the optical fiber is calculated. When the number of connected regions is less than 1, the clamping plate spacing and the frame index of the breakage occurrence are output.

[0067] The fiber dynamic fatigue parameter generation unit is used to generate fiber dynamic fatigue parameters by combining the pressure plate spacing and the fracture occurrence frame index with the agreed threshold after the number of effective samples at each speed level reaches the agreed threshold.

[0068] The beneficial technical effects of the present invention are at least as follows:

[0069] This invention uses a high-speed camera to capture the state of optical fiber samples in real time during testing. The algorithm identifies the state of the optical fiber in real time and records the image of the moment the fiber breaks and the spacing of the pressure plates, effectively reducing misjudgments. When multiple samples are tested simultaneously, each optical fiber can be accurately recorded, making it traceable and effectively improving testing efficiency and accuracy, thus meeting the industry's need for rapid testing and evaluation of optical fiber fatigue characteristics. Attached Figure Description

[0070] The present invention will be further described with reference to the accompanying drawings, but the embodiments in the drawings do not constitute any limitation on the present invention. For those skilled in the art, other drawings can be obtained based on the following drawings without creative effort.

[0071] Figure 1 This is a schematic diagram of the steps and structure of a method for testing dynamic fatigue parameters of optical fibers according to the present invention.

[0072] Figure 2 This is a schematic diagram of the pressure plate spacing of the present invention.

[0073] Figure 3 This is a schematic diagram of fiber breakage and good product in this invention.

[0074] Figure 4This is a schematic diagram of the optical fiber connection area of ​​the present invention.

[0075] Figure 5 This is a schematic diagram of the fiber dynamic fatigue parameter testing method according to an embodiment of the present invention. Detailed Implementation

[0076] Embodiments of the present invention are described in detail below. Examples of these embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and should not be construed as limiting the present invention.

[0077] Optical fiber: Optical fiber is a thin, long fiber made of high-purity glass (such as silicon dioxide) or plastic. It achieves efficient transmission of optical signals through the principle of total internal reflection.

[0078] Stress corrosion susceptibility parameters: The stress corrosion susceptibility parameters of optical fibers are a type of mechanical property of optical fibers. They are indicators that measure the corrosion resistance of optical fibers under specific stress environments and are mainly used to evaluate the resistance of optical fibers to chemical or environmental corrosion under long-term stress.

[0079] Dynamic fatigue parameters: These are stress corrosion susceptibility parameters obtained through dynamic fatigue testing, primarily including the two-point bending dynamic method and the axial tension dynamic method. The parameters obtained by different methods vary, and the testing method is generally selected based on the actual application environment.

[0080] In one or more embodiments, such as Figure 1 As shown, a method for testing the dynamic fatigue parameters of optical fibers is disclosed, the method comprising the following steps:

[0081] Step 1: Acquire fiber optic images and pressure plate spacing using a high-speed camera; wherein, the pressure plate spacing is the distance between the inner sides of the clamp;

[0082] Step two is selected from one of (I)-(IV):

[0083] (I) Select a target recognition model, train the target recognition model based on the optical fiber image and the pressure plate spacing, and output the pressure plate spacing and the frame index of the breakage occurrence at this time;

[0084] (II) An anomaly detection model is adopted, with good quality optical fiber as the normal value of the model and optical fiber breakage as the abnormal value. If the anomaly detection model outputs an anomaly, the pressure plate spacing and the frame index of the breakage occurrence are output at this time.

[0085] (III) An anomaly detection model is adopted. The good optical fibers are regarded as the normal values of the model, and the fiber breakage is regarded as the abnormal value. If the anomaly detection model outputs an anomaly, the platen spacing and the breakage occurrence frame index at this time are output, input into the detection model, the good optical fibers and the broken optical fibers are marked, the broken fiber number of the broken optical fiber is output, and the platen spacing corresponding to this frame is output.

[0086] (IV) An anomaly detection model is adopted. The good optical fibers are regarded as the normal values of the model, and the fiber breakage is regarded as the abnormal value. If the anomaly detection model outputs an anomaly, the platen spacing and the breakage occurrence frame index at this time are output, input into the detection model, the good optical fibers and the broken optical fibers are marked, the broken fiber number of the broken optical fiber is output, and the platen spacing corresponding to this frame is output; based on the broken fiber number of the broken optical fiber and the platen spacing corresponding to this frame, the fiber is segmented using the same segmentation method as the broken sample number, the number of connected regions of the fiber is calculated, and when the number of connected regions is less than 1, the platen spacing and the breakage occurrence frame index are output.

[0087] Step 3: After the effective sample numbers of each speed gear reach the agreed threshold, the platen spacing and the breakage occurrence frame index are combined to enter dynamic fatigue parameter fitting to generate the fiber dynamic fatigue parameters.

[0088] Specifically, the method is described as follows:

[0089] As Figure 2 shown, record the distance d inside the fixture and the number of pixels inside the fixture in the image in the static state. During the experiment, the optical fiber will move under the action of the platen. The Hough transform is used to identify the straight line inside the fixture for the collected image, and the distance inside the fixture is automatically calculated. When d > s, the optical fiber is not broken, and only the algorithm is used to calculate the distance. When d < s, other AI models are used to identify the state of the optical fiber at the same time.

[0090] Furthermore, the explanations are as follows:

[0091] Method 1: Use a classification model to classify the states of the optical fibers into broken and good. Through a large number of collected images for training, if a break is identified during the test, the d at this time is output. The specific scheme is as follows:

[0092] In the two-point bending experiment of the optical fiber, in order to accurately identify the breakage moment of the optical fiber, it is necessary to comprehensively analyze the images and geometric parameters collected during the experiment. The core of this step is to combine the image sequence obtained by the high-speed camera with the platen spacing calculated in real time, use the convolutional neural network classification model to output the breakage probability, and on this basis, confirm the breakage occurrence time point and the corresponding geometric quantity through the coupling decision strategy.

[0093] First, the image sequence of the optical fiber Data was acquired using a high-speed camera. A high-speed camera with a frame rate of at least 1000fps was used in the experiment to ensure that one frame was acquired every millisecond, avoiding missed detection of momentary breaks. The camera was configured with a global shutter, a resolution of 1920×1080, and a 50mm fixed-focus lens, with the acquisition area focused on the bottom of the U-shaped region of the optical fiber. The camera was synchronized with the loading device via hardware triggering, ensuring that each frame was timestamped. For example, in one experiment, when the pressure plate approached at a speed of 10μm / s, 300 frames were generated over 300ms, each clearly showing the gradual bending of the optical fiber as the pressure plate approached.

[0094] Secondly, the pressure plate spacing sequence The data was obtained through image processing methods. Before the experiment, the fixtures were statically calibrated, and the correspondence between the known physical distances on the inner sides of the fixtures and the number of pixels in the image was measured. During the experiment, the edges of the fixtures in the image were identified using Hough line detection, the pixel distances between the fixtures were calculated, and then converted into physical distances according to the calibration coefficients. In this way, as the pressure plates approach each frame, a sequence of pressure plate spacings corresponding to the image timestamps is formed. For example, in the 300 frames of images mentioned above, the initial spacing is 1.50 mm, which gradually decreases to 1.20 mm, and the data is synchronously stored in the computer memory.

[0095] After obtaining the image and geometric data, each frame of the image... Input into classification model The model is a pre-trained convolutional neural network with frozen parameters. The backbone uses a ResNet-18 architecture, and a Squeeze-and-Excitation attention mechanism is added to each residual module, allowing the model to focus more on the slender regions of the optical fiber during feature extraction. The model input is a single-frame image scaled to 256×256 pixels, and the output is the probability of an optical fiber breakage. The range is from 0 to 1. For example, in the experiment, the model outputs probabilities mostly between 0.1 and 0.3 under normal conditions, but the probability rises rapidly to over 0.9 when a significant crack appears.

[0096] To avoid relying on one To address the resulting misjudgments, this step involved designing a geometry-probability coupled score. The formula is as follows:

[0097] ;

[0098] in, It is a moment The probability of fracture; and These are the current and previous time intervals, respectively; the second term in the fraction represents the normalized shrinkage rate of the interval between two adjacent frames; It is a fixed weighting factor, typically calibrated experimentally to a value between 0.2 and 0.5. Normalization ensures that this term and the probability are within the same numerical range, facilitating fusion. For example, in a certain experiment, mm, If mm, then the shrinkage rate is 0.0039; if at this time , Then the fraction If light reflection occurs, causing... However, if the geometry does not shrink, then the term is 0, the fraction is still 0.78, and it will not be misjudged as a break.

[0099] The calculated fraction sequence throughout the experiment It shows a steady upward trend until it spikes sharply at the moment of break. The frame index corresponding to the maximum score is found using the following formula. :

[0100] ;

[0101] This frame was determined to be the moment the break occurred. To further improve robustness, in practical applications... Three consecutive frames are taken for confirmation, requiring the center frame score to be a local maximum and the probability of breakage to be low. The value is greater than a threshold (e.g., 0.5), and the spacing is trending towards contraction. If the condition is not met, the process reverts to the frame corresponding to the second largest value and re-evaluates.

[0102] The final output includes the fracture occurrence frame index. The frame rate can be converted into a specific time point, for example, the 236th frame corresponds to 236ms; at the same time, the pressure plate spacing corresponding to that frame is also output. For example, in the above example, it is 1.020 mm. Both results serve as important inputs for subsequent fatigue parameter calculations.

[0103] In summary, this step utilizes images acquired by a high-speed camera and pressure plate spacing data, combined with a classification model and geometric shrinkage rate to construct a coupling score. This avoids misjudgments caused by relying solely on visual or geometric quantities, ensuring the synchronization of fracture time and physical quantities. The entire method demonstrated high accuracy and reproducibility in experiments, making it a crucial component of the fiber optic dynamic fatigue parameter testing system.

[0104] Method 2: An anomaly detection model is used. Good quality optical fiber is considered normal, while broken optical fiber is considered an anomaly. If the model outputs an anomaly, the value d at that point is output. The specific scheme is as follows:

[0105] In the two-point bending experiment of optical fiber, the basic idea of ​​using anomaly detection model to identify fracture is as follows: Figure 3-4 As shown, an autoencoder model is trained using a large number of image samples from high-quality optical fibers, enabling it to accurately reconstruct the appearance features of normal optical fibers. When an optical fiber breaks during the experiment, the model can no longer completely reconstruct the breakage area, resulting in a significant increase in the difference between the input image and the reconstructed image. This "reconstruction failure" anomaly is used to determine the breakage. Unlike the classification method in the previous step, this model does not require a large number of breakage samples during training, but only relies on high-quality data. This "learning only the normal" strategy is more suitable for transient and rare events such as optical fiber breakage.

[0106] During data acquisition, the high-speed camera continues to acquire fiber optic image sequences at a frequency of over 1000 frames per second. The camera is configured with a global shutter and a resolution of 1920×1080 pixels, with the lens fixedly focused on the bottom area of ​​the U-shaped fiber optic cable. Each frame of the image is accompanied by a hardware timestamp shared with the loading device to ensure synchronization with the geometry. Simultaneously, the image processing module performs straight-line detection on the fixture edges and calculates the corresponding pressure plate spacing based on static calibration coefficients. For example, in an experiment with a loading rate of 10 μm / s, the camera recorded the entire process of the initial spacing gradually decreasing from 1.50 mm to 1.20 mm, acquiring approximately 300 frames, each corresponding to a pressure plate spacing value.

[0107] Anomaly detection model A convolutional autoencoder structure is employed. Its encoding part consists of multiple convolutional layers, batch normalization, and nonlinear activation layers to progressively compress the features of the input image; the decoding part reconstructs an image of the same size as the input image through deconvolution and upsampling operations. During the training phase, the model uses only a large number of high-quality fiber images, and the optimization objective is to minimize the reconstruction error between the input and output images. To enhance the model's sensitivity to the bottom of the fiber's U-shape and potential breakage areas, a spatial weight map is introduced onto the training data. It assigns higher weights to areas where fiber stress is concentrated and lower weights to background areas, ensuring that the model learns features more accurately in key areas.

[0108] During the experiment, each frame of the image Input to The corresponding reconstructed image is obtained. Then, the weighted reconstruction error is calculated, and anomaly scores are constructed by combining geometric shrinkage information. :

[0109] ;

[0110] in, and These represent the pixel positions of the image. Input values ​​and reconstructed values; It is a normalized weight map generated during the calibration stage, with larger values ​​taken in the bottom region where the fiber stress is concentrated; and These represent the spacing between two adjacent pressure plates; This is a coupling weight constant used to balance the contributions of visual error and geometric shrinkage. This defined fraction can significantly improve the performance when a crack occurs, while remaining stable under ambient light reflection or slight occlusion. For example, in one experiment, the weighted reconstruction error of a normal frame was around 0.01, but when a crack appeared in frame 236, the error suddenly increased to 0.15. When shrinking from 1.024 mm to 1.020 mm, the normalized shrinkage term is approximately 0.0039. ,but It is significantly higher than the levels in the previous frames.

[0111] Throughout the experiment, the fraction sequence was calculated continuously. Normally, the level remains low and fluctuates slightly under normal conditions until a significant peak occurs at the moment of fracture. The fracture moment is determined by the following formula:

[0112] ;

[0113] That is, take the frame index where the score reaches the maximum value. As the moment when fracture occurs. To further improve robustness, during engineering implementation, [the following is considered]: Local verification is performed on the three consecutive frames. The frame must not only contain a local maximum, but also have a reconstruction error significantly higher than adjacent frames and a shrinking spacing trend. If these conditions are not met, the system reverts to the second-largest value and re-evaluates. This verification mechanism avoids misidentifying false peaks caused by noise or light spot reflections in a single frame as breaks in the image.

[0114] Finally, the output of this step is the fracture occurrence frame index. and the corresponding pressure plate spacing In the aforementioned experiment, frame 236 was determined to be the break point, with a frame index of 236 and a corresponding interval of... mm. These two results are stored together and used as the core input for subsequent dynamic fatigue parameter fitting.

[0115] In summary, this step, through the coupling of reconstruction errors in the anomaly detection model and the geometric shrinkage of the pressure plate, forms a fracture identification mechanism suitable for dynamic fatigue testing of optical fibers. It avoids the problem of relying on a large number of fracture samples, strengthens the focus on the U-shaped bottom region through a spatial weighted graph, and significantly improves anti-interference capability by combining geometric constraints, ensuring accurate synchronization of fracture time and physical quantities. This method demonstrates reproducibility and stability under multi-sample parallel testing, high-speed loading, and complex ambient light conditions, and is a key link in ensuring data reliability throughout the entire invention.

[0116] Method 3: Use the detection model as the AI ​​model to mark good optical fibers and broken optical fibers, as shown in the figure below. When a broken optical fiber is detected, output the value d at that time.

[0117] In the two-point bending experiment of optical fiber, the detection model method can simultaneously complete "fiber localization" and "breakage determination" in a single frame image, which is particularly suitable for scenarios involving parallel testing of multiple optical fibers. This step follows the breakage occurrence frame index output in the previous step. and the corresponding pressure plate spacing In the same frame image The above utilizes the detection model Each optical fiber is individually marked to determine which fiber broke. This combines "when it broke" with "which fiber broke," forming a traceable closed-loop determination that ensures the accuracy and repeatability of experimental data under parallel sample conditions.

[0118] The image data acquisition method remains consistent with the preceding steps of the system: a high-speed camera captures images of the experimental area at a sampling rate of over 1000 frames per second to ensure no fracture is missed. Each frame has a resolution of 1920×1080 pixels and includes a hardware timestamp, strictly synchronized with the pressure plate spacing data. In the fracture frames requiring judgment... The system directly reads the corresponding image from the cache. This frame of image completely covers all optical fibers. Meanwhile, geometric quantities... The results, already calculated in the previous steps, will continue to be used as physical constraints in this step.

[0119] Detection model During the training phase, a single-stage object detection network from the YOLO series was used, with CSPDarknet as the backbone and an FPN+PAN structure for feature fusion. The detection head outputs the position parameters of the candidate boxes. And classification probabilities. During training, a large number of manually labeled images of good and broken optical fibers are used. The model learns the location and category label of each candidate box. After training, the parameters are frozen. During the inference phase, the model can directly output the box positions and breakage confidence of multiple optical fibers by inputting an image. This structure ensures that even if there are multiple optical fibers in a single frame, the model can simultaneously detect and label their states.

[0120] In the broken frame running Output a set of candidate boxes ,in Indicates the position of the candidate box. This indicates the confidence level at which the candidate box is determined to be broken. However, confidence level alone can be affected by light reflection, camera noise, or candidate box bounce. Therefore, this step introduces geometric and morphological constraints specific to the two-point bending characteristics of the fiber. Specifically, within each candidate box, the image region is first binarized and thinned to obtain the fiber skeleton, and then the connectivity index of that region is calculated. This is defined as "the ratio of the maximum number of connected channel pixels to the total number of foreground pixels within the frame". If the fiber optic cable remains intact, Approximately 1; if the fiber optic cable is broken, A significant decrease. At the same time, considering the shrinkage of the pressure plate spacing, Initial reference distance from the device Normalize it and use it as a geometric factor in the determination.

[0121] Combining the above three quantities, this step defines the detection coupling score for candidate boxes. :

[0122] ;

[0123] in, It is the fracture confidence level output by the detection model; It is the spacing between the pressure plates of the broken frame; It is the reference spacing initially calibrated in the experiment; It is the connectivity index of the candidate boxes; and These are weighting constants used to balance the contributions of geometric shrinkage and connectivity penalty. The innovation of this score lies in the fact that it not only relies on the detection model's judgment but also introduces physical constraints (the degree of pressure plate shrinkage) and image morphological indicators (connectivity). The combination of these three significantly improves the stability of fracture localization. For example, in an experiment involving four optical fibers, the model output a fracture probability of 0.91 for the third fiber, while the other three were below 0.3; simultaneously, the pressure plate spacing in that frame... Compared to the baseline value, the density has shrunk by approximately 0.3. In the four candidate frames, the connectivity index of the third fiber is only 0.18, while the other three are all greater than 0.9. After substituting into the formula, the coupling score of the third fiber is significantly higher than that of the other three, thus accurately determining that the third fiber in frame 236 has broken.

[0124] Calculate all candidate boxes within the same frame. The broken fiber number can be determined by selecting the largest value. :

[0125] ;

[0126] If scores are tied, the candidate with the higher geometric factor is selected; if they are still tied, the candidate box located closer to the center of the sample row or column is selected. This strategy enables the unique identification of broken optical fibers even in complex scenarios.

[0127] The output of this step is: Broken fiber number. and the spacing between the pressure plates corresponding to that frame. These two outputs are related to the breakpoint index of the preceding step. Combined, this forms a complete tracing result of "time-object-geometric quantity". The entire process directly utilizes the data output from the previous step, completing the localization and determination within the same frame, ensuring reproducibility and highlighting the inventiveness of the solution: by combining the output of the detection model with physical geometric quantities and image connectivity, a determination formula specifically for the two-point bending experiment scenario of optical fiber is proposed. This formula can distinguish specific samples when multiple samples are measured simultaneously, and avoids the false alarm problem common in traditional detection models.

[0128] Method 4: Use the segmentation method to segment the optical fiber, calculate the number of connected regions in the optical fiber, and output d when the number of connected regions is less than 1.

[0129] In the two-point bending experiment of optical fiber, the segmentation model serves to provide pixel-level evidence of fracture, so that the fracture time has been output by methods 2 and 3. and fracture sample number Based on this, a final confirmation is made as to whether the sample has truly fractured, and the distance between the pressure plates at the moment of fracture is simultaneously output. This approach not only supplements the probability and bounding box results output by classification and detection models, but also provides intuitive geometric and connectivity evidence, forming a complete closed loop of time-object-geometry-pixel.

[0130] The experimental data source was consistent with the system's preceding steps. During the experiment, the high-speed camera acquired image sequences at a frame rate exceeding 1000 frames per second, with each frame being 1920×1080 pixels. Strict synchronization with the loading device was maintained via hardware triggering, ensuring that each frame corresponded to a specific pressure plate spacing value. Pressure plate spacing. It is obtained by detecting the straight line inside the fixture in the image and combining it with the statically calibrated pixel-physical quantity conversion coefficient. This is used in the keyframes for fracture determination. The system directly reads the corresponding frame image from the cache. And obtain the fracture sample number from the previous step. To confirm at the pixel level whether the sample is truly broken, this step uses a trained segmentation model. The model employs a U-Net architecture. The encoding layer includes multiple convolutional layers, batch normalization, and non-linear activation to extract multi-scale features. The decoding layer gradually restores spatial resolution through upsampling and convolution, finally outputting a probability map of the same size as the input. During training, manually labeled fiber optic mask images are used, with labels distinguishing between "fiber optic pixels" and "background pixels." After sufficient training and freezing, the model generates pixel-level segmentation results for the input image region during the inference phase.

[0131] In the judgment frame In the middle, the system first determines the number Find the corresponding rectangular region of interest in the "sample position and image coordinate mapping table" established during calibration. This area was entered into A binary mask is obtained. The number of connected components is then calculated within the mask using the connected component algorithm. Under normal circumstances, the pixel regions of each optical fiber are completely connected. If the optical fiber breaks, the connected region will split into two or more parts. Consequently, the number of connected regions increases. To avoid misjudgments caused by small noise or local shadows, this step designs an innovative "connectivity-gap-geometric coupling score," which adds two indicators, the gap ratio and the bridging ratio, to the number of connected regions. Combined with the shrinkage of the pressure plate, this forms a comprehensive judgment formula for the scenario of this invention.

[0132] ;

[0133] in, This is the fracture determination value. It is the number of connected components; To normalize the gap ratio, by... The skeleton is extracted and the pixel distance between the endpoints of the main fracture is measured, and then normalized to obtain the ratio with the diagonal length of the region. The bridging ratio is the proportion of new pixels added after a fixed-radius closing operation on the mask, reflecting whether the gap may be an adhesion that is "not completely broken". It is the spacing between the pressure plates of the broken frame. This is the initial baseline spacing. Four weights. Fixed during deployment, determined through calibration experiments. For example, in a frame if This indicates that the optical fiber is divided into two segments; skeleton analysis yields the notch ratio. Bridgeable ,at the same time Compared to the baseline spacing of 1.50mm, the shrinkage is approximately 0.32; when the weight is taken as... At that time, the coupling score of the frame A value greater than 0.8 is significantly higher than the level of normal frames (approximately 0.1–0.2), therefore, it is determined that this fiber optic cable is in a faulty frame. It has broken.

[0134] In actual implementation, the system... Thresholds obtained from statistics during the deployment phase Make comparisons. When At that time, confirm that the fracture has occurred and output the result. Otherwise, it is marked as "incomplete fracture" or "requires review". To further ensure stability, the system also archives this information in the log. and the segmentation result mask This facilitates later manual inspection and reproduction.

[0135] The final output is the spacing between the pressure plates of the broken frame. This method provides pixel-level segmentation evidence in conjunction with the fiber optic fracture morphology by introducing two specific indicators—notch ratio and bridging ratio—and combining them with the experimental geometry of pressure plate shrinkage. This ensures that the determination not only depends on the change in the number of pixels but also closely aligns with the physical process. Its role in fiber optic dynamic fatigue testing lies in providing visualized, pixel-level evidence for each fracture event, avoiding the uncertainty caused by single probabilities or the number of connections, and further guaranteeing the traceability of experimental results and the integrity of the invention.

[0136] Once the number of valid samples at each speed level reaches a predetermined threshold, the dynamic fatigue parameters are fitted using the pressure plate spacing and the fracture occurrence frame index to generate the fiber dynamic fatigue parameters. Specifically:

[0137] First, data acquisition and alignment no longer require any preprocessing: each record has been indexed in the frame where the break occurred using one of the aforementioned four methods. The output shows the spacing between the pressure plates in the same frame. (Image-geometric alignment complete), along with the loading velocity of the sample. (Obtained synchronously from speed settings and encoder counting via motion control software). Then, according to sample number and speed setting... The fragmented frame snapshots are dropped into pairs, and their paths are simultaneously written to the same record for audit review. Before entering the calculation, only one minimal consistency check is performed: if the fragmented frame snapshot shows "not completely broken" or... If an item exceeds the effective working range specified by the static calibration of the device, it will be marked as a verification and will not be included in this calculation; all other items will be included in the geometric-mechanical mapping.

[0138] The geometric-mechanical mapping uses a minimum variable explicit relationship for two-point bending: the fracture spacing... First, the equivalent bending radius of the U-shaped bottom is mapped, then the fiber core surface strain is obtained and linearly mapped to the fracture stress. To avoid introducing complex analytical derivations, this scheme fixes the "radius-spacing" scaling factor during the static calibration stage of the device (setting multiple spacings with standard gauge blocks, acquiring the bottom curve, and performing circle / ellipse fitting to obtain the scaling factor), thus simplifying the online calculation into a one-step explicit expression:

[0139] ;

[0140] here, The longitudinal elastic modulus of the optical fiber material (entered into the system parameter library before the test is started by stretching a single filament or using the supplier's consistent specifications). The outer diameter of the optical fiber (a fixed value obtained by microscopic measurement or online diameter measurement before sample loading, which is written into the parameters of this batch). The dimensionless proportionality coefficient obtained from the static calibration of the device (fitted from the "spacing-equivalent radius" calibration curve, fixed at the device level); For this sample in the fracture frame The spacing between the pressure plates (output in the same frame by one of the aforementioned four methods). The denominator in this formula... Normalization was performed (by) (absorbing geometric differences), therefore it can be directly compared with Substitution is straightforward and requires no additional unit conversion. For example: A batch of parameters is... , , (All three are pre-registered constants), in There is a record on the gear. Substituting the values, we can obtain... (On par with the other samples in the same grade). All Accordingly They are grouped according to speed level.

[0141] Once the number of valid samples for each speed setting reaches a predetermined threshold (e.g., no less than 15 samples per setting), dynamic fatigue parameter fitting begins. The fitting follows the generally accepted engineering practice of approximating linearity in logarithmic coordinates based on a power-law relationship: each recorded... As a sample point, a weighted least squares approach is used to reduce the impact of low-confidence points. The weights can be the product of "fragmentation frame determination confidence × mapping interval stability" (both derived from existing logs in the preceding steps, requiring no additional data collection). To ensure transparency and ease of implementation, the slope and parameters are calculated using the explicit formulas below; in practice, regression can be completed simply by performing a weighted summation on the point set for each grade:

[0142]

[0143] in, The regression weight for this sample point is obtained by multiplying the confidence and stability scores from the aforementioned log; the algorithm can be implemented using common weighted regression. , It is a weighted average; for – The regression slope; This represents the desired dynamic fatigue parameters. For example, the weighted regression analysis after summing the results at four speeds yields... ,but The system simultaneously saves the fitting residuals, weights, and corresponding fragmented frame snapshot paths for easy review.

[0144] To ensure a closed-loop process, the system outputs... At the same time, automatically generate the speed for each gear. The system provides a distribution summary and a list of sampled broken frames. If a certain sample level is insufficient or the weighted residual is significantly large, it will automatically provide a "supplementary test suggestion," which only requires supplementing that sample level and does not require repeating the already completed sample levels.

[0145] like Figure 5As shown, this invention uses a designed intelligent image recognition algorithm to calculate the spacing between the pressure plates when the optical fiber breaks, thereby measuring the strain / stress value required when the optical fiber breaks with a two-point bending geometry. The optical fiber dynamic fatigue parameter testing device mainly consists of a computer, a motion control module, an image acquisition module, a stepper motor, a high-speed camera, pressure plates, and fixtures. The computer serves as the control and data processing hub. Through the motion control module, it drives the stepper motor to move the movable pressure plates at a certain speed. Simultaneously, the image acquisition module transmits the optical fiber test images from the high-speed camera in real time. The intelligent image algorithm in the computer identifies and analyzes the optical fiber state in real time, recording the image at the moment of fiber breakage and the pressure plate spacing. The stepper motor stops when all the optical fiber samples have been broken and the test is complete. The high-speed camera's recording range needs to cover all optical fibers in the fixture, with the focal point at the bottom of the U-shaped area of ​​the optical fiber, and the shooting frequency should be higher than 1000 frames / second. The image acquisition module can achieve high-speed image acquisition, with an acquisition rate higher than 1000 frames / second.

[0146] Image intelligent recognition algorithms need to possess millisecond-level high-speed automatic recognition and tracking capabilities for fiber optic targets. Targeted training is required to enable tracking and inference capabilities. The specific process is as follows: First, a dedicated dataset is constructed for model training and testing. This method primarily focuses on recognizing fiber optic images, thus requiring high-speed cameras to capture image sets for training and testing. Second, the image recognition model is selected. AI target recognition models such as YOLOv5s can be used, as the target objects are relatively simple. However, in experiments, the fiber optic cable will move under the pressure of the pressure plate, and the bending angle will change. Therefore, the model needs to possess millisecond-level high-speed target recognition and tracking capabilities. Third, model training is performed. A pre-trained model is prepared, and the model's recognition and tracking capabilities are trained using the dedicated dataset. The algorithm should be able to recognize fiber optic images in real time and output a breakage signal immediately upon fiber optic breakage. Once the requirements are met during testing, the algorithm can be used for system integration; otherwise, retraining is required until the requirements are met.

[0147] A second aspect of the present invention provides an optical fiber dynamic fatigue parameter testing device, comprising the following modules:

[0148] An image acquisition unit is used to acquire fiber optic images and pressure plate spacing via a high-speed camera; wherein, the pressure plate spacing is the distance between the inner sides of the clamp;

[0149] The image analysis unit is used to perform one of (I)-(IV):

[0150] (I) Select a target recognition model, train the target recognition model based on the optical fiber image and the pressure plate spacing, and output the pressure plate spacing and the frame index of the breakage occurrence at this time;

[0151] (II) An anomaly detection model is adopted, with good quality optical fiber as the normal value of the model and optical fiber breakage as the abnormal value. If the anomaly detection model outputs an anomaly, the pressure plate spacing and the frame index of the breakage occurrence are output at this time.

[0152] (III) An anomaly detection model is adopted, with good quality optical fiber as the normal value of the model and optical fiber breakage as the abnormal value. If the anomaly detection model outputs an anomaly, the pressure plate spacing and the frame index of the breakage occurrence are output and input into the detection model. Good quality optical fiber and broken optical fiber are marked, and the broken optical fiber number of the broken optical fiber and the pressure plate spacing corresponding to the frame are output.

[0153] (IV) An anomaly detection model is adopted, with good quality optical fiber as the normal value and optical fiber breakage as the abnormal value. If the anomaly detection model outputs an anomaly, the clamping plate spacing and the frame index of the breakage occurrence are output and input into the detection model. Good quality optical fiber and broken optical fiber are marked, and the broken optical fiber number and the clamping plate spacing corresponding to the frame are output. Based on the broken optical fiber number and the clamping plate spacing corresponding to the frame, the optical fiber is segmented using the segmentation method with the broken sample number. The number of connected regions of the optical fiber is calculated. When the number of connected regions is less than 1, the clamping plate spacing and the frame index of the breakage occurrence are output.

[0154] The fiber dynamic fatigue parameter generation unit is used to generate fiber dynamic fatigue parameters by combining the pressure plate spacing and the fracture occurrence frame index with the agreed threshold after the number of effective samples at each speed level reaches the agreed threshold.

[0155] This invention can be a method, apparatus, system, and / or computer program product. The computer program product may include a computer-readable storage medium having computer-readable program instructions loaded thereon for performing various aspects of the invention.

[0156] In the above embodiments, the descriptions of each embodiment have different focuses. For parts that are not described in detail or recorded in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0157] Although embodiments of the invention have been shown and described, those skilled in the art will understand that various changes, modifications, substitutions and variations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the claims and their equivalents.

[0158] The various illustrative logic blocks, modules, and circuits described in conjunction with the embodiments disclosed herein can be implemented or performed using a general-purpose processor, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA) or other programmable logic device, discrete gate or transistor logic, discrete hardware components, or any combination thereof designed to perform the functions described herein. The general-purpose processor may be a microprocessor, but in alternatives, it may be any conventional processor, controller, microcontroller, or state machine. The processor may also be implemented as a combination of computing devices, such as a combination of a DSP and a microprocessor, multiple microprocessors, one or more microprocessors cooperating with a DSP core, or any other such configuration.

[0159] The steps of the methods or algorithms described in conjunction with the embodiments disclosed herein may be embodied directly in hardware, in a software module executed by a processor, or in a combination of both. The software module may reside in RAM memory, flash memory, ROM memory, EPROM memory, EEPROM memory, registers, hard disk, removable disk, CD-ROM, or any other form of storage medium known in the art. An exemplary storage medium is coupled to a processor such that the processor can read and write information to / from the storage medium. In an alternative, the storage medium may be integrated into the processor. The processor and storage medium may reside in an ASIC. The ASIC may reside in a user terminal. In an alternative, the processor and storage medium may reside as discrete components in the user terminal.

[0160] In one or more exemplary embodiments, the described functionality may be implemented in hardware, software, firmware, or any combination thereof. If implemented in software as a computer program product, the functionality may be stored or transmitted as one or more instructions or code on or through a computer-readable medium. A computer-readable medium includes both computer storage media and communication media, encompassing any medium that facilitates the transfer of a computer program from one location to another. A storage medium may be any available medium accessible to a computer. By way of example and not limitation, such a computer-readable medium may include RAM, ROM, EEPROM, CD-ROM or other optical disc storage, disk storage or other magnetic storage devices, or any other medium that can be used to carry or store desired program code in the form of instructions or data structures and is accessible to a computer. Any connection is also legitimately referred to as a computer-readable medium. For example, if the software is transmitted from a website, server, or other remote source using coaxial cable, fiber optic cable, twisted pair, digital subscriber line (DSL), or wireless technologies such as infrared, radio, and microwave, then the coaxial cable, fiber optic cable, twisted pair, DSL, or wireless technologies such as infrared, radio, and microwave are included in the definition of a medium. As used in this article, disk and disc include compact discs (CDs), laser discs, optical discs, digital multi-purpose discs (DVDs), floppy disks, and Blu-ray discs. Disks typically reproduce data magnetically, while discs reproduce data optically using lasers. Combinations of these should also be included within the scope of computer-readable media.

[0161] The prior description of this disclosure is provided to enable any person skilled in the art to make or use this disclosure. Various modifications to this disclosure will be apparent to those skilled in the art, and the general principles defined herein may be applied to other variations without departing from the spirit or scope of this disclosure. Therefore, this disclosure is not intended to be limited to the examples and designs described herein, but should be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for testing dynamic fatigue parameters of optical fibers, characterized in that, The method includes: Step 1: Acquire fiber optic images and pressure plate spacing using a high-speed camera; wherein, the pressure plate spacing is the distance between the inner sides of the clamp; Step 2: Employ an anomaly detection model, using good fiber as the normal value and fiber breakage as the anomaly value. If the anomaly detection model outputs an anomaly, output the clamping plate spacing and the breakage occurrence frame index at that time, input them into the detection model, mark the good fiber and the broken fiber, and output the broken fiber number of the broken fiber and the clamping plate spacing corresponding to that frame. Based on the broken fiber number and the clamping plate spacing corresponding to that frame, segment the fiber corresponding to the broken sample number, calculate the number of connected regions of the fiber, and when the number of connected regions is less than 1, output the clamping plate spacing and the breakage occurrence frame index. Step 3: Once the number of valid samples for each speed setting reaches the agreed threshold, the dynamic fatigue parameters are fitted by combining the pressure plate spacing and the fracture occurrence frame index to generate the fiber dynamic fatigue parameters. The detection model uses a single-stage object detection network from the YOLO series during the training phase, with CSPDarknet as the backbone and FPN combined with PAN structure for the feature fusion part. The detection head outputs the position parameters and classification probabilities of the candidate boxes. The specific steps of outputting the current clamping plate spacing and the frame index of the breakage occurrence, inputting them into the detection model, marking good optical fibers and broken optical fibers, and outputting the breakage fiber number of the broken optical fiber and the clamping plate spacing corresponding to that frame include: The fracture occurrence frame index is input into the detection model, which outputs a set of candidate boxes. ,in Indicates the position of the candidate box. This indicates the confidence level at which the current candidate box is judged as broken; Within each candidate box, the candidate box region is first binarized and thinned to obtain the fiber skeleton, and then the connectivity index of the region is calculated. , defined as the ratio of the maximum number of connected channel pixels to the total number of foreground pixels within the bounding box, and taking into account the shrinkage of the pressure plate spacing, the pressure plate spacing at this time is... Initial reference distance from the device Perform normalization, and use it as a geometric factor in the determination; Based on the connectivity index Initial reference spacing and the confidence level of the current candidate box being judged as broken. Calculate the detection coupling score of the candidate boxes : ； in, and These are the weight constants, used to balance the contributions of geometric shrinkage and connectivity penalty; Within the same frame, calculate the detection coupling score for all candidate boxes. The largest value is selected to determine the broken fiber number, and then output. The specific steps for segmenting the fiber corresponding to the fractured sample number based on the fractured fiber number and the pressure plate spacing corresponding to the frame, and outputting the pressure plate spacing and fracture occurrence frame index when the number of connected regions is less than 1, are as follows: Based on the broken fiber number and the pressure plate spacing corresponding to the frame, a binary mask is obtained using a pre-trained segmentation model. Based on the binary mask, the number of connected regions is calculated using the connected component algorithm; if the number of connected regions = 1, then the pixel region of each optical fiber is completely connected. Calculate the breakage determination value based on the number of connected regions: ； in, This is the fracture determination value. It is the number of connected components; To normalize the gap ratio, a binary mask is used. The skeleton is extracted and the pixel distance between the endpoints of the main fracture is measured, and then normalized to obtain the ratio with the diagonal length of the region. The bridging ratio is the proportion of new pixels added after performing a closing operation with a fixed radius on the mask. It is the spacing between the pressure plates of the broken frame. It is the initial baseline spacing; four weights It is fixed during deployment and determined through calibration experiments; If the breakage determination value is greater than the preset normal frame level, then the optical fiber is determined to be broken at the current breakage occurrence frame index.

2. The method for testing dynamic fatigue parameters of optical fibers according to claim 1, characterized in that, The spacing between the pressure plates is calculated as follows: In a static state, record the distance to the inside of the fixture and the number of pixels inside the fixture in the image; If the fixture moves, the Hough transform is used to identify the straight line inside the fixture in the acquired image, and the distance inside the fixture is automatically calculated. If the distance inside the clamp is greater than a preset threshold and the optical fiber is unbroken, then only the algorithm is used to calculate the distance; if the distance inside the clamp is less than the preset threshold, then the target recognition model is used to identify the state of the optical fiber.

3. The method for testing dynamic fatigue parameters of optical fibers according to claim 1, characterized in that, The anomaly detection model adopts a convolutional autoencoder structure; its encoding part consists of multiple convolutional layers, batch normalization, and nonlinear activation layers, which are used to progressively compress the features of the input image. The decoding part reconstructs an image of the same size as the input through deconvolution and upsampling operations; The anomaly detection model uses good fiber as the normal value and fiber breakage as the abnormal value. If the anomaly detection model outputs an anomaly, the current clamping plate spacing and the frame index of the breakage occurrence are output and input into the detection model. Good fiber and broken fiber are marked, and the broken fiber number of the broken fiber and the clamping plate spacing corresponding to that frame are output. The specific execution steps are as follows: The optical fiber image sequence was acquired by a high-speed camera, and the pressure plate spacing sequence was obtained corresponding to the distance inside the fixture. The fiber optic image is input into the anomaly detection model to obtain the corresponding reconstructed image; Anomaly scores are calculated based on the corresponding reconstructed images, as well as the corresponding fiber optic images and pressure plate spacing. When the anomaly score is at its maximum, the corresponding frame index is selected as the moment when the break occurs; By combining the frame index corresponding to the maximum abnormal score and the corresponding pressure plate spacing, the dynamic fatigue parameters of the optical fiber are generated.

4. The method for testing dynamic fatigue parameters of optical fibers according to claim 3, characterized in that, The abnormal score The calculation is as follows: ； in, and These represent the pixel positions of the image. Input values ​​and reconstructed values; It is a normalized weight map generated during the calibration stage, with larger values ​​taken in the bottom region where the fiber stress is concentrated; and These represent the spacing between two adjacent pressure plates; This is the coupling weight constant, used to balance the contributions of visual error and geometric shrinkage.

5. The method for testing dynamic fatigue parameters of optical fibers according to claim 1, characterized in that, If the anomaly detection model and / or the output of the detection model pass the test, the current anomaly detection model and / or the detection model will be used for system integration; if the anomaly detection model and / or the output of the detection model fails the test, the current anomaly detection model and / or the detection model will be retrained until it meets the requirements for passing the test.

6. An apparatus for performing the optical fiber dynamic fatigue parameter testing method as described in claim 1, characterized in that, The device includes: An image acquisition unit is used to acquire fiber optic images and pressure plate spacing via a high-speed camera; wherein, the pressure plate spacing is the distance between the inner sides of the clamp; The image analysis unit employs an anomaly detection model, treating good optical fibers as normal values ​​and fiber breaks as anomalies. If the anomaly detection model outputs an anomaly, it outputs the current clamping plate spacing and the frame index of the breakage occurrence, inputs these values ​​into the detection model, marks good and broken optical fibers, and outputs the breakage fiber number of the broken optical fiber and the clamping plate spacing corresponding to that frame. Based on the breakage fiber number and the clamping plate spacing corresponding to that frame, the unit segments the optical fiber corresponding to the broken sample number, calculates the number of connected regions in the optical fiber, and outputs the clamping plate spacing and the frame index of the breakage occurrence when the number of connected regions is less than 1. The fiber dynamic fatigue parameter generation unit is used to generate fiber dynamic fatigue parameters by combining the pressure plate spacing and the fracture occurrence frame index with the agreed threshold after the number of effective samples at each speed level reaches the agreed threshold.

Images