Intelligent optical fiber welding auxiliary operation device, welding state pre-detection method and storage medium

Through intelligent fiber fusion auxiliary operation device and deep learning model, the problem of poor fusion quality of fibers in harsh environments is solved, efficient and professional fiber repair under low temperature and wind and sand conditions is achieved, and the recovery speed and signal of the communication network are improved. Transmission quality.

CN118915234APending Publication Date: 2024-11-08STATE GRID LIAONING SHENYANG ELECTRIC POWER SUPPLY COMPANY
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202410963148.9
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-07-18
Publication Date
2024-11-08

AI Technical Summary

Technical Problem

Optical fibers are susceptible to damage in harsh environments, especially in low temperatures and windy sand conditions, resulting in poor welding quality, affecting the recovery speed of the communication network and signal transmission.

Method used

An intelligent fiber fusion auxiliary operation device is designed, including a box, a high-definition camera, a STM32 controller and a heating tube. It combines a deep learning model to perform pre-detection of fiber fusion state, prevent pollution through a sealing structure, and maintain a constant temperature using a heating tube.

Benefits of technology

It improves the success rate and quality of fiber fusion splicing, ensures rapid and professional repair work in harsh environments, reduces damage and pollution during fusion splicing, and improves the recovery speed of communication networks.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118915234A_ABST
    Figure CN118915234A_ABST
Patent Text Reader

Abstract

The invention relates to the technical field of optical fiber fusion splicing devices, in particular to an intelligent optical fiber fusion splicing auxiliary operation device which comprises a box body, and a mounting seat is arranged at the bottom in the box body and used for mounting an optical fiber fusion splicer; a cover body is arranged on the upper end surface of the box body; two side surfaces of the box body are respectively provided with a through hole and an optical fiber inlet, and the optical fiber inlet and the through hole are respectively provided with a sealing structure; and a high-definition camera is installed for shooting and detecting the optical fiber welding state. Meanwhile, the invention further provides an optical fiber fusion state pre-detection method, a deep learning model for judging the optical fiber fusion position is trained and burnt to an stm chip, whether fusion at the position meets the requirement that the loss is within 0.02 dB or not is calculated, and optical fiber fusion operation is assisted through prompt tone. According to the intelligent optical fiber fusion auxiliary operation device provided by the invention, the optical fiber fusion splicer is put into the device, so that the effects of keeping warm, preventing sand and preventing wind are achieved, the intelligent optical fiber fusion auxiliary operation device can be suitable for intelligent optical fiber fusion work in various severe weather and environments, and the success rate of fusion is improved.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of optical fiber fusion splicing devices, and in particular to an intelligent optical fiber fusion splicing auxiliary operation device, a fusion splicing state pre-detection method and a storage medium. Background Art

[0002] Fiber optic transmission, with its significant advantages such as high bandwidth, low loss, strong anti-interference and high security, has become an indispensable long-distance, high-bandwidth data transmission solution in the modern communications field.

[0003] However, optical fiber is also at risk of damage in a complex and ever-changing environment. Environmental factors such as temperature and humidity changes, natural disasters, physical damage, and equipment failures may all cause damage to optical fiber. Therefore, when optical fiber is damaged, rapid and professional repair work is particularly important. In this process, the optical fiber fusion splicer plays a vital role. It uses advanced hot-melt technology to accurately fuse the broken optical fiber end faces together to ensure lossless transmission of optical signals. It cannot be ignored that optical fiber fusion work faces huge challenges under severe weather conditions such as cold and windy sand. Low temperature makes the optical fiber more brittle and more easily damaged during fusion; windy sand may contaminate the optical fiber end face and affect the quality of fusion. This undoubtedly poses a serious obstacle to the speed of repair to restore the communication network. In addition, the quality and qualification of fusion also seriously affect the transmission of signals. Summary of the invention

[0004] In order to solve the above technical problems, the present invention provides an intelligent optical fiber fusion splicing auxiliary operation device, comprising a box body 1, a mounting seat 3 is provided at the bottom of the box body 1, and an optical fiber fusion splicer 4 is installed; a cover body 5 is provided on the upper end surface of the box body 1; through holes 6 and optical fiber inlets 7 are respectively provided on both side surfaces of the box body 1, and sealing structures are provided at the inlets 7 and the through holes 6;

[0005] The optical fiber inlet 7 is a long hole, which is arranged along the height direction of the side of the box body 1, and the upper end is open; a row of sealing brushes 11 is arranged on both sides of the long hole, and the two rows of sealing brushes 11 are overlapped and staggered at the inlet 7;

[0006] The inclined surface 10 is embedded with a controller 8 structure including an STM32 and a high-definition camera 9 .

[0007] Furthermore, the front and upper end surfaces of the box body 1 are designed to be transparent.

[0008] Furthermore, a slope 10 is provided at the connection between the front face and the upper end face of the box body 1 .

[0009] Furthermore, the sealing structure provided at the through hole 6 is a corrugated elastic rubber member 12 .

[0010] Furthermore, a heating tube and a thermocouple are provided in the box body 1, and the thermocouple is connected to a controller, and the controller controls the heating tube to perform heating.

[0011] Furthermore, a high-definition camera 9 is installed on the side of the box 1 to detect the optical fiber fusion state.

[0012] The present invention also provides a method for pre-detecting the state of optical fiber fusion splicing, based on any one of the above-mentioned intelligent optical fiber fusion splicing auxiliary operation devices, comprising:

[0013] The historical fiber splicing images are used to train the vision-based deep learning model, which includes a deformable feature extraction network, a content conversion feature fusion network, and a detection layer.

[0014] Use a high-definition camera to take a picture of the fiber pre-fusion state P, and input it into the deformable feature extraction network to obtain the output feature p1;

[0015] Input the output feature P1 into the content conversion feature fusion network to obtain the fused feature P2;

[0016] The Focal-EIoU loss function is used to analyze the fusion feature P2 to determine whether the fiber pre-splicing status image P is a successful splicing class or a splicing failure class.

[0017] Furthermore, the deformable convolution feature extraction network Deformable Convolution Network (DCN) can change the convolution kernel size according to the feature shape, increase the degree of freedom of convolution kernel sampling, and autonomously learn the offset in the fiber fusion detection task; specifically, an initial offset is set to ΔP n , where P n It is an enumeration of all pixel positions, and the starting position P0 of each position on the feature map of the fiber fusion image is calculated. The fiber fusion state picture P taken by the high-definition camera is input into this network, and its input feature is set to x(P0+P n +ΔP n ), the calculation formula is:

[0018]

[0019] Among them, y(P0) is the output result of the starting position P0, w(P n ) is the weight function, R is the receptive field set in the entire image, x(P0+P n +ΔP n ) is the input feature, due to the offset ΔP n It is usually a decimal, so it is implemented by bilinear interpolation, and the calculation formula is

[0020]

[0021] Among them, x(p) is the input feature after interpolation, that is, the output feature P1 of the fiber fusion state image P, G(q,p) is the bilinear interpolation function, and x(q) is the original input feature.

[0022] Furthermore, a multi-scale feature fusion network based on contextual transformer (CoT) is adopted. The multi-scale feature fusion network introduces the CoT module, inputs the feature p1 into the above CoT module, further extracts the feature, and outputs the fused feature P2. Specifically, the module uses 3×3 convolution to contextually encode the input key and generate a static context representation. The encoded key is connected to the input query, and a dynamic multi-head attention matrix is ​​learned through two consecutive 1×1 convolutions. In the multi-head attention mechanism, the input sequence is first transformed into three matrices: query, Q, key, K, and value, V.

[0023] For each attention head, a scaled dot product attention operation is performed using independent Q, K, and V matrices;

[0024] The formula for the scaled dot product attention operation is:

[0025]

[0026] where d K is the dimension of the K matrix, used to scale the dot product to stabilize the softmax function;

[0027] The learned attention matrix is ​​multiplied with the input value to achieve a dynamic context representation of the input; the fusion of static and dynamic context representations is used as the output of this module.

[0028] Furthermore, the predicted box b is used to select the position of the fused feature P2 and compared with the reference box of successful or failed fusion. The Focal-Efficient Intersection over Union (Focal-EIoU) loss function is used to select the position of the fused feature P2 given the predicted box b and the target reference box b. gt , the loss function EIoU loss is defined as follows:

[0029] L EIoU =L IoU +L dis +L ssp (4)

[0030] Among them, LLoss represents the loss, L dis is the distance loss, L ssp is a penalty term for the size of the predicted box shape.

[0031]

[0032] Where ρ represents the Euclidean distance, c is the diagonal length of the smallest enclosing rectangle covering the prediction box and the reference box; at any time, if one of the two variables, width w or height h, increases, the other will decrease;

[0033] L Focal-EIoU =IoU γ L EIoU (6)

[0034] Where IoU = \b gt ∩b / b gt ∪b\,γ is a parameter used to control the degree of suppression of outliers.

[0035] A computer storage medium, wherein the computer readable storage medium stores a computer program, and when the computer program is executed by a processor, any of the above methods is implemented.

[0036] The intelligent optical fiber fusion splicing auxiliary operation device proposed by the present invention places an optical fiber fusion splicer in the device to keep warm, prevent sand and wind, and is suitable for intelligent optical fiber fusion splicing work in various severe weather and environments, and is easy to operate; in addition, the optical fiber fusion splicing pre-detection method provided by the present invention, when the optical fiber is pre-fixed on the optical fiber fusion splicer, the pre-fusion fixation image of the optical fiber is collected by a camera and transmitted to a controller with a built-in STM32 chip, so as to determine whether the fusion position outputs a corresponding detection result, prompt the on-site operator, and promptly correct the position of the optical fiber fusion splicing, thereby improving the success rate of the fusion splicing. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] Figure 1 It is a schematic structural diagram of an embodiment of a box of the present invention;

[0038] Figure 2 It is a schematic diagram of the side structure of the box;

[0039] Figure 3 This is the front view of the box.

[0040] Among them, 1. box body; 2. optical fiber; 3. mounting seat; 4. optical fiber fusion splicer; 5. cover body; 6. through hole; 7. line inlet; 8. controller; 9. high-definition camera; 10. inclined plane; 11. sealing brush; 12. corrugated elastic rubber part. DETAILED DESCRIPTION

[0041] The principles and features of the present invention are described below in conjunction with the accompanying drawings. The examples given are only used to explain the present invention and are not used to limit the scope of the present invention.

[0042] refer to Figure 1-3 The present invention provides an intelligent optical fiber fusion splicing auxiliary operation device, including a box body 1, a mounting seat 3 is arranged at the bottom of the box body 1, and a fiber fusion splicer 4 is installed; a cover body 5 is arranged on the upper end surface of the box body 1; through holes 6 and optical fiber inlets 7 are respectively arranged on both side surfaces of the box body 1, and sealing structures are arranged at the inlets and through holes 6.

[0043] As an improvement to the solution, in order to facilitate observation of the fusion splicing of the optical fibers in the box 1, the front and upper end surfaces of the box 1 are designed to be transparent.

[0044] As an improvement of the solution, when the optical fiber is being fused, the box 1 is located in front and below the operator, and the line of sight enters the box 1 from the intersection of the front and the upper end surface of the box 1. In order to prevent the edge line of the box 1 at the intersection from interfering with the vision, an inclined surface 10 is provided at the connection between the front and the upper end surface of the box 1, and the line of sight enters the box 1 for observation through the inclined surface 10.

[0045] As an improvement of the solution, the optical fiber inlet 7 is a long hole, which is arranged along the height direction of the side of the box body 1 and has an opening at the upper end; a row of sealing brushes 11 is arranged on both sides of the long hole, and the two rows of sealing brushes 11 are overlapped and staggered at the inlet, so that the seal is denser, and the disconnected optical fiber penetrates into the box body 1 from the sealing brushes 11 in the long holes on both sides, and the sealing brushes 11 can effectively prevent sand and dust.

[0046] As an improvement of the solution, the sealing structure provided at the through hole 6 is a corrugated elastic rubber member 12, and the operator extends both arms into the box body 1 from the through hole 6 to operate.

[0047] As an improvement of the scheme, a heating tube and a thermocouple are provided in the box 1, and the thermocouple is connected to the controller, and the controller controls the heating tube to heat. This heating structure is generally used in relatively cold weather to prevent the excessive cold from affecting the performance of the fusion splicer. Therefore, a temperature control device is added to keep a constant temperature in the box 1 at all times, so that the machine works in a mild environment. After the fusion is completed, the cover 5 is opened and the fused optical fiber 2 moves upward along the upper end of the long hole out of the box 1.

[0048] As an improvement to the solution, the controller 8 and the camera 9 are installed on the inclined surface 10, and the placement position should avoid blocking the line of sight. A deep learning model for determining the fiber fusion position is placed in the controller. The camera takes the fiber end face position and transmits it to the controller. The controller determines whether the fiber position before fusion meets the requirement that the loss after fusion is less than 0.02dB, and assists the fiber fusion operation through a prompt sound. A battery is provided in the box 1 for power supply.

[0049] The present invention also provides a method for pre-detecting the state of optical fiber fusion splicing, based on any one of the above-mentioned intelligent optical fiber fusion splicing auxiliary operation devices, comprising the following steps:

[0050] Step 1: Use historical fiber splicing images to train a vision-based deep learning model, which includes a deformable feature extraction network, a content conversion feature fusion network, and a detection layer; burn the model to the controller to calculate whether the fiber end face position meets the requirements of post-splicing loss within 0.02dB.

[0051] Use a high-definition camera to take a picture of the fiber pre-fusion state P, and input it into the deformable feature extraction network to obtain the output feature p1;

[0052] Deformable Convolution Network (DCN) can change the size of the convolution kernel according to the feature shape, increase the degree of freedom of convolution kernel sampling, and autonomously learn the offset in the fiber fusion detection task; specifically, an initial offset is set to ΔP n , where P n It is an enumeration of all pixel positions, and the starting position P0 of each position on the feature map of the fiber fusion image is calculated. The fiber fusion state picture P taken by the high-definition camera is input into this network, and its input feature is set to x(P0+P n +ΔP n ), the calculation formula is:

[0053]

[0054] Among them, y(P0) is the output result of the starting position P0, w(P n ) is the weight function, R is the receptive field set in the entire image, x(P0+P n +ΔP n ) is the input feature, due to the offset ΔP n It is usually a decimal, so it is implemented by bilinear interpolation, and the calculation formula is

[0055]

[0056] Among them, x(p) is the input feature after interpolation, that is, the output feature P1 of the fiber fusion state image P, G(q,p) is the bilinear interpolation function, and x(q) is the original input feature.

[0057] The output feature P1 is input into the content transformation feature fusion network to obtain the fused feature P2; a multi-scale feature fusion network based on the content converter (Contextual Transformer, CoT) is adopted. In order to allocate more resources to more important tasks under limited computing power and achieve efficient and accurate task processing, the CoT module is introduced, which aims to enhance the ability of visual representation by making full use of the contextual information between input keys, thereby improving the performance of computer vision tasks.

[0058] This module uses 3×3 convolution to contextually encode the input key and generate a static context representation. The encoded key is connected to the input query and a dynamic multi-head attention matrix is ​​learned through two consecutive 1×1 convolutions. In the multi-head attention mechanism, the input sequence is first transformed into three matrices: query, Q, key, K, and value, V.

[0059] For each attention head, a scaled dot product attention operation is performed using independent Q, K, and V matrices;

[0060] The formula for the scaled dot product attention operation is:

[0061]

[0062] where d K is the dimension of the K matrix, used to scale the dot product to stabilize the softmax function;

[0063] The learned attention matrix is ​​multiplied by the input value to realize the dynamic context representation of the input; the fusion of static and dynamic context representation is used as the output of this module. The feature p1 is input into the above CoT module, and the feature is further extracted to output the fused feature P2.

[0064] The Focal-EIoU loss function is used to analyze the fusion feature P2 to determine whether the fiber pre-splicing state image P is a successful splicing class or a splicing failure class. Specifically,

[0065] Use the prediction box b to select the position of the fused feature P2 and compare it with the reference box of successful or failed fusion; when the prediction box b does not overlap with the reference box, the intersection over union (IoU) loss will have a gradient vanishing problem, resulting in a slow convergence speed and inaccurate detector. In order to solve this problem, the IoU loss function is improved to the Focal-EIoU loss function. Given the prediction box b and the target reference box b gt , the loss function EIoU loss is defined as follows:

[0066] L EIoU =L IoU +L dis +L ssp (4)

[0067] Among them, LLoss represents the loss, L dis is the distance loss, L ssp is a penalty term for the size of the predicted box shape.

[0068]

[0069] Where ρ represents the Euclidean distance, c is the diagonal length of the smallest enclosing rectangle covering the prediction box and the reference box; at any time, if one of the two variables, width w or height h, increases, the other will decrease;

[0070] L Focal-EIoU =IoU γ L EIoU (6)

[0071] Where IoU = \b gt ∩b / b gt ∪b\,γ is a parameter used to control the degree of suppression of outliers.

[0072] Through the above-mentioned Focal-EIoU loss function, it is distinguished whether the fusion feature P2 belongs to the successful welding class or the failed welding class, so that it can be concluded that the optical fiber welding state image P belongs to the successful welding class or the failed welding class, and further the conclusion of the success or failure of the optical fiber welding can be drawn.

[0073] A computer storage medium, wherein the computer readable storage medium stores a computer program, and when the computer program is executed by a processor, any of the above methods is implemented.

[0074] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. An intelligent optical fiber fusion auxiliary operation device, Features: The box body (1) comprises a mounting seat (3) at the bottom of the box body (1) for mounting an optical fiber fusion splicer (4); a cover (5) is provided on the upper end surface of the box body (1); through holes (6) and optical fiber inlets (7) are respectively provided on both side surfaces of the box body (1), and sealing structures are provided at the inlets (7) and the through holes (6); The optical fiber inlet (7) is a long hole, arranged along the height direction of the side of the box body (1), and the upper end is open; a row of sealing brushes (11) is arranged on both sides of the long hole, and the two rows of sealing brushes (11) are overlapped and staggered at the inlet (7); A controller (8) structure containing STM32 and a high-definition camera (9) are embedded on the inclined surface (10).

2. An intelligent optical fiber fusion splicing auxiliary operation device as claimed in claim 1, Features: A slope (10) is provided at the connection between the front face and the upper end face of the box body (1).

3. An intelligent optical fiber fusion splicing auxiliary operation device as claimed in claim 1, Features: The sealing structure arranged at the through hole (6) is a corrugated elastic rubber piece (12).

4. The intelligent optical fiber fusion splicing auxiliary operation device according to claim 1, Features: A heating tube and a thermocouple are arranged in the box (1); the thermocouple is connected to a controller, and the controller controls the heating tube to perform heating.

5. The intelligent optical fiber fusion splicing auxiliary operation device according to claim 1, Features: A high-definition camera (9) is installed on the side of the box (1) to detect the optical fiber fusion state.

6. A method for pre-detecting the state of optical fiber fusion splicing, based on the intelligent optical fiber fusion splicing auxiliary operation device according to any one of claims 1 to 5 above, It is characterized in that include, The historical fiber splicing images are used to train the vision-based deep learning model, which includes a deformable feature extraction network, a content conversion feature fusion network, and a detection layer. Use a high-definition camera to take a picture of the fiber pre-splicing state P, and input it into the deformable feature extraction network to obtain the output feature p1; Input the output feature P1 into the content conversion feature fusion network to obtain the fused feature P2; The Focal-EIoU loss function is used to analyze the fusion feature P2 to determine whether the fiber pre-splicing status image P is a successful splicing class or a splicing failure class.

7. A method for pre-detecting optical fiber fusion state as claimed in claim 6, It is characterized in that The deformable convolution feature extraction network DCN can change the convolution kernel size according to the feature shape, increase the degree of freedom of convolution kernel sampling, and autonomously learn the offset in the fiber fusion detection task; specifically, an initial offset is set to ΔP n , where P n It is an enumeration of all pixel positions, and calculates the starting position P of each position on the feature map of the fiber fusion image. 0 , input the fiber splicing status picture P taken by the high-definition camera into this network, and set its input feature as x(P 0 +P n +ΔP n ), the calculation formula is: Among them, y(P 0 ) is the starting position P 0 The output result, w(P n ) is the weight function, R is the receptive field set in the entire image, x(P 0 +P n +ΔP n ) is the input feature, due to the offset ΔP n It is usually a decimal, so it is implemented by bilinear interpolation, and the calculation formula is: Among them, x(p) is the input feature after interpolation, that is, the output feature P1 of the fiber fusion state image P, G(q,p) is the bilinear interpolation function, and x(q) is the original input feature.

8. A method for pre-detecting optical fiber fusion state as claimed in claim 6, It is characterized in that A multi-scale feature fusion network based on content converter is adopted. The multi-scale feature fusion network introduces CoT module, inputs feature p1 into the above CoT module, further extracts features, and outputs fused feature P2. Specifically, the module uses 3×3 convolution to contextually encode the input key and generate a static context representation. The encoded key is connected to the input query, and a dynamic multi-head attention matrix is ​​learned through two consecutive 1×1 convolutions. In the multi-head attention mechanism, the input sequence is first transformed into three matrices of query (Query, Q), key (Key, K) and value (Value, V) through different linear transformations. For each attention head, a scaled dot product attention operation is performed using independent Q, K, and V matrices; The formula for the scaled dot product attention operation is: where d K is the dimension of the K matrix, used to scale the dot product to stabilize the softmax function; The learned attention matrix is ​​multiplied with the input value to achieve a dynamic context representation of the input; the fusion of static and dynamic context representations is used as the output of this module.

9. A method for pre-detecting optical fiber fusion state as claimed in claim 6, It is characterized in that Use the prediction box b to select the position of the fusion feature P2 and compare it with the reference box of successful or failed welding; Using the focal-effective intersection-over-union loss function, given the predicted box b and the target reference box b gt , the loss function EIoU loss is defined as follows: L EIoU =L IoU +L dis +L ssp (4) Among them, L(Loss) represents the loss, L dis is the distance loss, L ssp is the penalty term for the shape and size of the prediction box; Where ρ represents the Euclidean distance, c is the diagonal length of the smallest enclosing rectangle covering the prediction box and the reference box; at any time, if one of the two variables, width w or height h, increases, the other will decrease; L Focal-EIoU =IoU γ L EIoU (6) Among them, IoU = \b gt ∩b / b gt ∪b\,γ is a parameter used to control the degree of suppression of outliers.

10. A computer storage medium, It is characterized in that The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the method according to any one of claims 6 to 9 is implemented.