Method, device and equipment for automatically detecting optical loss of optical fiber connector

Abstract

The invention relates to the technical field of optical fiber communication, and discloses an optical fiber connector optical loss automatic detection method, device and equipment, and the device comprises a modulation signal dynamic adjustment module, an environment parameter adaptive calibration module and a deep learning prediction model. The modulation signal dynamic adjustment module enhances the separability of optical fiber signals through dynamic phase modulation, and the detection precision is improved; the environmental parameter self-adaptive calibration module compensates the influence of temperature, humidity and vibration on optical signals in real time, and the stability of the system is improved; the deep learning prediction model is based on multi-dimensional data fusion and weight dynamic optimization, high-precision prediction of the performance degradation trend of the optical fiber connector is achieved, closed-loop logic from data collection to real-time prediction is formed, experiments show that the detection precision of the method can reach 99.2%, the prediction accuracy is improved to 93% or above, the optical loss calculation error is reduced to 1%, and the method can be widely applied to the field of optical fiber connectors. And meanwhile, the response speed is increased to millisecond level. The method is suitable for monitoring and predicting the optical fiber performance in a complex environment, and the precision and the real-time processing capability are remarkably improved.

Description

Technical Field

[0001] The present invention relates to the technical field of optical fiber communication, and particularly to a method, device and equipment for automatically detecting the optical loss of an optical fiber connector, which are used to detect and predict the performance of the optical fiber connector in a complex environment. Background Art

[0002] With the wide application of optical fiber communication systems in high-bandwidth and low-latency application scenarios, the performance stability and service life of optical fiber connectors have become important factors to ensure the reliability of the system. In actual operation, the performance of optical fiber connectors is affected by various factors such as environmental temperature, humidity, vibration and signal load, and its performance may gradually degrade, resulting in increased optical loss, signal distortion, and even communication interruption. To address this issue, the prior art usually realizes performance monitoring and life prediction through the following several methods:

[0003] 1. Based on the method of regular manual detection: Most of the existing optical fiber connector detections adopt manual periodic testing, and measure the optical loss value through offline equipment. However, this method has problems of poor real-time performance and difficulty in capturing the dynamic changes of the connector performance. Especially in the case of high-frequency use or extreme environments, it may lead to detection lag or failure.

[0004] 2. Statistical analysis through simple data monitoring: Some systems introduce basic monitoring devices to evaluate the performance status of the connector by collecting optical signal intensity data. However, due to the lack of comprehensive compensation for environmental interference factors, the accuracy of the detection results is relatively low, and it is difficult to cope with the non-linear changes in complex environments.

[0005] 3. Using traditional mathematical models for life prediction: Some technologies in the industry use fixed mathematical models based on historical data fitting to predict the life of the connector performance. However, the fusion ability of such models for multi-dimensional parameters is limited, and they cannot be updated in real time and dynamically, making it difficult to accurately reflect the actual degradation trend of optical fiber connectors under complex working conditions, and the prediction error is relatively large.

[0006] To solve the above problems, some studies have tried to improve the prediction accuracy by introducing machine learning algorithms. However, the existing solutions often lack real-time calibration and dynamic adjustment of environmental parameters, resulting in difficult convergence in the model training process or insufficient adaptability, which further limits the practical application value of the prediction model. Therefore, the prior art still has the following deficiencies: insufficient real-time performance, difficulty in capturing the dynamic changes of the connector performance; significant influence of environmental interference, lack of effective calibration mechanism; insufficient multi-dimensional data fusion ability, and the accuracy and applicability of the prediction model need to be improved.

[0007] Therefore, how to achieve dynamic real-time detection of the performance of optical fiber connectors and perform high-precision life prediction in complex environments has become the technical problem to be solved by the present invention. Summary of the Invention

[0008] The technical problem solved by the present invention is to provide an automatic optical loss detection method, device and equipment for an optical fiber connector to solve the problems of poor real-time performance, insufficient environmental adaptability and low prediction accuracy proposed in the above background technology in view of the defects existing in the above-mentioned prior art.

[0009] To solve the above technical problems, the technical solutions adopted by the present invention are as follows:

[0010] An automatic optical loss detection method for an optical fiber connector includes the following steps:

[0011] Step 1, generation of multi-frequency composite modulation signals: Generate multiple groups of modulation signals through a multi-frequency composite modulation light source, and each group of modulation signals has different frequencies and phases. Specifically, it includes:

[0012] The modulation signal frequency f i is dynamically generated according to the following formula:

[0013] f i = f 0 + i·Δf

[0014] where f 0 is the base frequency, Δf is the frequency increment, and i is the modulation signal serial number;

[0015] The modulation signal phase φ i is dynamically set according to the following formula:

[0016] φ i = φ 0 + g(i)

[0017] where φ 0 is the initial phase, and g(i) is the phase adjustment function, which is determined according to the physical characteristics of the optical fiber connector and the detection scenario;

[0018] The modulation signal is injected into the optical fiber connector to be detected;

[0019] Step 2, acquisition and separation of optical reflection signals: The reflection signals at the interface of the optical fiber connector are collected in real time through an optical reflection receiving module, where:

[0020] The collected reflection signals include optical intensity, phase and delay characteristics;

[0021] The fast Fourier transform FFT is used to perform frequency-domain separation on the reflection signals to generate a reflection signal data set, and the data set includes the incident optical power P i (f 入射 ) and the reflected optical power P i (f 入射 (fi );

[0022] Step 3, Adaptive calibration of environmental parameters: Detect the temperature, humidity, and vibration information of the environment through the environmental sensing module, and dynamically adjust the optical power output and receiving sensitivity of the modulation signal according to the following calibration formula:

[0023] P 校准 = P 原始 ·(1 + α T ·ΔT + α H ·ΔH + α V ·ΔV)

[0024] Where:

[0025] P 校准 is the calibrated optical power, and P 原始 is the initial optical power;

[0026] ΔT, ΔH, and ΔV are the deviation values of environmental temperature, humidity, and vibration respectively;

[0027] α T , α H , α V are calibration coefficients obtained by experimental fitting;

[0028] Step 4, Calculation of optical loss: Based on the reflection signal dataset, calculate the optical loss value of the fiber optic connector through the following formula:

[0029]

[0030] Where: P 入射 (f i ) and P 反射 (f i ) are the incident optical power and reflected optical power at frequency f i respectively;

[0031] Calculate the comprehensive optical loss value L 综合 through the following weighted average formula:

[0032]

[0033] Where, w i is the dynamic weight coefficient, determined by the least squares fitting method according to the signal-to-noise ratio and frequency range of the reflection signal, and N is the total number of modulation signals;

[0034] Step 5, Loss traceability and fault classification: Based on the optical loss value and reflection signal characteristics, realize the traceability of the optical loss source and the classification of fault types through a pre-trained deep learning model, where:

[0035] The input of the deep learning model includes the intensity feature, phase feature, and delay feature of the reflected signal;

[0036] The model is trained using 1000 sets of optical reflection signal data collected in the laboratory through a training method based on a convolutional neural network, and outputs a classification result including the location of the loss source and the type of fault;

[0037] Step 6, performance degradation trend prediction: Through a time series analysis algorithm, based on the collected historical optical loss data, predict the future performance degradation trend of the fiber optic connector, and output a performance prediction report including the predicted time point and the loss change trend;

[0038] Step 7, result output: Upload the comprehensive optical loss value, traceability diagnosis result, and performance prediction report of the fiber optic connector to the cloud platform through a data interface, and the cloud platform performs visual processing on the data and generates maintenance suggestions.

[0039] As a further solution of the present invention, the specific form of the phase adjustment function g(i) of the modulation signal is:

[0040] g(i) = A·sin(B·i) + C

[0041] Wherein, A is the phase adjustment amplitude, determined based on the maximum dynamic range of the fiber optic reflection signal; B is the frequency factor, depending on the sampling rate of the modulation signal; C is the phase offset, determined by the dielectric constant of the fiber optic connector interface.

[0042] As a further solution of the present invention, in the reflected signal dataset, the incident optical power P 入射 (f i ) and P 反射 (f i ) are obtained through the following steps:

[0043] Use a high-sensitivity photodetector to sample the reflected signal;

[0044] Pass the sampled signal through a digital filter to suppress environmental noise and low-frequency interference;

[0045] Apply the fast Fourier transform FFT to extract the frequency domain features of the reflected signal, and calculate the optical power value based on the frequency domain features.

[0046] As a further solution of the present invention, the adaptive calibration of the environmental parameters further includes the following steps:

[0047] Collect no less than 1000 sets of historical data for the temperature, humidity, and vibration of the detection environment respectively, and establish a multi-dimensional data correlation model;

[0048] Use the weighted linear regression method to fit the calibration coefficients α T 、αH and α V , where the weighting factor is jointly determined by the time decay function of historical data and the fluctuation amplitude of real-time data;

[0049] Adjust the compensation parameters of the optical loss calculation formula in real time according to the calibration coefficient.

[0050] As a further solution of the present invention, the weighting coefficient w of the optical loss calculation i The dynamic determination of includes the following steps:

[0051] For each frequency f i Calculate the signal-to-noise ratio SNR i , and the formula is:

[0052]

[0053] where σ i is the mean square deviation of the background noise corresponding to the frequency f i , and is obtained by averaging multiple samplings;

[0054] According to the signal-to-noise ratio SNR i and the frequency distribution characteristics, use an optimization method based on Bayesian estimation to dynamically allocate the weighting coefficient w i .

[0055] As a further solution of the present invention, the convolutional neural network structure of the deep learning model includes the following modules:

[0056] The first convolutional layer is used to extract the time-domain features of the reflected signal;

[0057] The second convolutional layer is used to extract the frequency-domain features of the reflected signal;

[0058] The third convolutional layer generates a feature map of a specific fault mode through multi-scale feature fusion;

[0059] A fully connected layer inputs the feature map into a classifier and outputs the loss classification of the fiber optic connector and the identification result of the fault source.

[0060] As a further solution of the present invention, the time series analysis algorithm for predicting the performance degradation trend is based on a long short-term memory (LSTM) neural network, and the training of this neural network includes the following steps:

[0061] The input features are historical optical loss values L 综合 , detection environment parameters, and fault categories;

[0062] Use the backpropagation algorithm with gradient clipping to optimize the network weights;

[0063] The network output is the optical loss value at a future time point and its change trend.

[0064] As a further solution of the present invention, the result output further includes:

[0065] Storing the comprehensive optical loss value and performance prediction result of the fiber optic connector into the cloud through a distributed database interface;

[0066] Generating an optical loss trend graph and a prediction curve using a real-time visualization engine based on WebGL;

[0067] Generating maintenance suggestions according to the prediction curve, including the recommended maintenance time window and tips for replacing key components.

[0068] As a further solution of the present invention, the method is applied to a fiber optic connector detection device, which includes:

[0069] A modulation signal generation module, configured to generate a multi-frequency composite modulation signal and inject it into the fiber optic connector port;

[0070] An optoelectronic detection module, configured to collect the reflected signal returned from the fiber optic connector;

[0071] A data processing module, configured to denoise, separate, and calculate the optical loss of the reflected signal;

[0072] An output module, configured to output the detection result and fault classification information through a display screen or a network interface.

[0073] As a further solution of the present invention, the method is applied to a performance monitoring device of a fiber optic communication system, which includes:

[0074] A data acquisition unit, configured to obtain the reflected signal from the fiber optic connector in real time;

[0075] A data analysis unit, configured to perform optical loss traceability and performance prediction based on a deep learning model;

[0076] A network interface unit, configured to upload the analysis result to a cloud platform and generate a remote maintenance report.

[0077] Compared with the prior art, the beneficial effects of the present invention are:

[0078] 1. Through the combination of the dynamic adjustment function of the modulation signal, the adaptive calibration mechanism of environmental parameters, and the multi-dimensional deep learning model, a complete technical closed-loop from data acquisition, processing, calculation to output is formed. By dynamically adjusting the phase of the modulation signal, the separability of the fiber optic reflection signal is effectively enhanced, thereby improving the accuracy of fiber optic connector loss calculation; the adaptive calibration of environmental parameters further compensates for the interference of external conditions (such as temperature, humidity, and vibration) on signal stability during the detection process; and the deep learning model based on the multi-layer convolutional neural network can accurately identify the source of optical loss and predict the performance degradation trend. This synergistic effect not only significantly improves the accuracy and stability of detection, but also enables the system to dynamically adapt to complex detection environments, realizing real-time analysis and intelligent feedback.

[0079] 2. In practical applications, the present technical solution can adapt to complex fiber optic communication scenarios, such as industrial environments with high temperature, high humidity, or large vibration frequencies. In these complex environments, traditional detection methods often suffer a significant decline in detection accuracy due to environmental disturbances, while the applied technical solution realizes the optimal compensation of the detection environment for core parameters through the environmental parameter calibration model and the adaptive adjustment of the modulation signal. In industrial environment tests, through the weight dynamic allocation mechanism, the signal-to-noise ratio of the reflection signal is increased to more than 85%, and the optical loss calculation error is reduced from 5% to 1%, providing a reliable guarantee for the safety and stability of fiber optic communication.

[0080] 3. By combining the convolutional neural network with historical fiber optic performance data and mining the time-domain and frequency-domain signal features, a performance degradation prediction model applicable to complex scenarios is constructed. Compared with existing methods, this model can update the weight parameters in real time and dynamically adjust the prediction results according to the latest environmental data and historical trends. Experimental results show that based on this prediction model, the prediction accuracy of the loss change trend of fiber optic connectors has been improved from 78% of the original technology to 93%, providing decision support for the operation and maintenance team for advance planning and troubleshooting.

[0081] The additional aspects and advantages of the present invention will be partially given in the following description, partially become obvious from the following description, or be understood through the practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0082] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for use in the description of the embodiments or the prior art. Obviously, the following drawings are only some embodiments of the present invention, and those of ordinary skill in the art can also obtain other drawings without creative efforts based on these drawings.

[0083] Figure 1 It is a flowchart of the dynamic closed-loop health monitoring and life prediction of the present invention. Detailed implementation mode

[0084] The technical solutions in the embodiments of the present invention will be clearly and completely described below. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0085] Please refer to Figure 1 , in the embodiment of the present invention, an automatic optical loss detection method for an optical fiber connector includes the following steps:

[0086] Step 1, generation of multi-frequency composite modulation signals: Generate multiple groups of modulation signals through a multi-frequency composite modulation light source, and each group of modulation signals has different frequencies and phases. Specifically, it includes:

[0087] The modulation signal frequency f i is dynamically generated according to the following formula:

[0088] f i = f 0 + i·Δf

[0089] where f 0 is the base frequency, Δf is the frequency increment, and i is the modulation signal serial number;

[0090] The modulation signal phase φ i is dynamically set according to the following formula:

[0091] φ i = φ 0 + g(i)

[0092] where φ 0 is the initial phase, and g(i) is the phase adjustment function, which is determined according to the physical characteristics of the optical fiber connector and the detection scenario;

[0093] The modulation signal is injected into the optical fiber connector to be detected;

[0094] Step 2, acquisition and separation of optical reflection signals: The reflection signals at the interface of the optical fiber connector are collected in real time through an optical reflection receiving module, where:

[0095] The collected reflection signals include light intensity, phase, and delay characteristics;

[0096] The fast Fourier transform FFT is used to perform frequency-domain separation on the reflection signals to generate a reflection signal data set, and the data set includes the incident optical power P i (f 入射 ) and the reflected optical power P i (f 入射 ) corresponding to different frequencies fi )

[0097] Step 3, Adaptive calibration of environmental parameters: Detect the temperature, humidity, and vibration information of the environment through the environmental sensing module, and dynamically adjust the optical power output and receiving sensitivity of the modulation signal according to the following calibration formula:

[0098] P 校准 = P 原始 ·(1 + α T ·ΔT + α H ·ΔH + α V ·ΔV)

[0099] Where:

[0100] P 校准 is the calibrated optical power, and P 原始 is the initial optical power;

[0101] ΔT, ΔH, and ΔV are the deviation values of environmental temperature, humidity, and vibration respectively;

[0102] α T , α H , α V are calibration coefficients obtained by experimental fitting;

[0103] Step 4, Calculation of optical loss: Based on the reflected signal dataset, calculate the optical loss value of the fiber optic connector through the following formula:

[0104]

[0105] Where: P 入射 (f i ) and P 反射 (f i ) are the incident optical power and reflected optical power at frequency f i respectively;

[0106] Calculate the comprehensive optical loss value L 综合 through the following weighted average formula:

[0107]

[0108] Where, w i is the dynamic weight coefficient, determined by the least squares fitting method according to the signal-to-noise ratio and frequency range of the reflected signal, and N is the total number of modulation signals;

[0109] Step 5, Loss traceability and fault classification: Based on the optical loss value and the characteristics of the reflected signal, realize the traceability of the source of optical loss and the classification of fault types through a pre-trained deep learning model, where:

[0110] The input of the deep learning model includes the intensity feature, phase feature, and delay feature of the reflected signal;

[0111] The model is trained using 1000 sets of optical reflection signal data collected in the laboratory through a training method based on a convolutional neural network, and outputs a classification result including the location of the loss source and the type of fault;

[0112] Step 6, Performance degradation trend prediction: Through a time series analysis algorithm, based on the collected historical optical loss data, predict the future performance degradation trend of the fiber optic connector, and output a performance prediction report including the predicted time point and the loss change trend;

[0113] Step 7, Result output: Upload the comprehensive optical loss value, traceability diagnosis result, and performance prediction report of the fiber optic connector to the cloud platform through a data interface, and the cloud platform performs visual processing on the data and generates maintenance suggestions.

[0114] As a further solution of the present invention, the specific form of the phase adjustment function g(i) of the modulation signal is: g(i) = A·sin(B·i) + C; where A is the phase adjustment amplitude, determined based on the maximum dynamic range of the fiber optic reflection signal; B is the frequency factor, depending on the sampling rate of the modulation signal; C is the phase offset, determined by the dielectric constant of the fiber optic connector interface. Among them, in terms of the influence of the dielectric constant on phase adjustment, the following formula can be referred to, that is, the relationship between phase adjustment and dielectric constant:

[0115]

[0116] Among them: Δφ: Phase adjustment amount. λ: Wavelength of light, usually 1.31μm or 1.55μm.

[0117] Refractive index, refractive index of the medium, determined by the relative dielectric constant ε of the material r of the material. d: Distance that the optical signal propagates in the connector material.

[0118] As a further solution of the present invention, the dynamic determination of the weighting coefficient w i for optical loss calculation includes the following steps:

[0119] For each frequency f i calculate the signal-to-noise ratio SNR i , and the formula is:

[0120]

[0121] where σ i is the mean square deviation of the background noise corresponding to the frequency f i , obtained by averaging multiple samplings;

[0122] According to the signal-to-noise ratio SNR i and the frequency distribution characteristics, use an optimization method based on Bayesian estimation to dynamically allocate the weight coefficient w i .

[0123] As a further solution of the present invention, the convolutional neural network structure of the deep learning model includes the following modules:

[0124] The first convolutional layer is used to extract the time-domain features of the reflected signal;

[0125] The second convolutional layer is used to extract the frequency-domain features of the reflected signal;

[0126] The third convolutional layer generates a feature map of a specific fault mode through multi-scale feature fusion;

[0127] A fully connected layer inputs the feature map into a classifier and outputs the loss classification of the fiber optic connector and the identification result of the fault source.

[0128] As a further solution of the present invention, the time series analysis algorithm for predicting the performance degradation trend is based on a long short-term memory (LSTM) neural network, and the training of this neural network includes the following steps:

[0129] The input features are historical optical loss values L 综合 , detection environment parameters, and fault categories;

[0130] Use the backpropagation algorithm with gradient clipping to optimize the network weights;

[0131] The network output is the optical loss value and its change trend at future time points.

[0132] As a further solution of the present invention, the result output further includes:

[0133] Store the comprehensive optical loss value and performance prediction result of the fiber optic connector in the cloud through a distributed database interface;

[0134] Use a real-time visualization engine based on WebGL to generate an optical loss trend graph and a prediction curve;

[0135] Generate maintenance suggestions according to the prediction curve, including the recommended maintenance time window and tips for replacing key components.

[0136] As a further solution of the present invention, the method is applied to a fiber optic connector detection device, and this device includes:

[0137] A modulation signal generation module is used to generate a multi-frequency composite modulation signal and inject it into the fiber optic connector port;

[0138] An optoelectronic detection module for collecting reflected signals returned from an optical fiber connector;

[0139] A data processing module for denoising, separating, and calculating optical loss of the reflected signals;

[0140] An output module for outputting detection results and fault classification information through a display screen or a network interface.

[0141] As a further solution of the present invention, the method is applied to a performance monitoring device of an optical fiber communication system, and the device includes:

[0142] A data acquisition unit for obtaining reflected signals in real time from an optical fiber connector;

[0143] A data analysis unit for tracing the source of optical loss and predicting performance based on a deep learning model;

[0144] A network interface unit for uploading the analysis results to a cloud platform and generating a remote maintenance report.

[0145] Embodiment 1:

[0146] In order to verify the application effect of the optical fiber connector performance detection and life prediction system proposed by the present invention in a complex environment, this embodiment combines the actual application scenario in the data center environment to elaborate on the specific implementation manner of the present invention in detail.

[0147] A large number of optical fiber connectors are arranged inside the data center. These connectors are susceptible to the influence of environmental temperature, humidity, vibration, and continuous high-load data streams during long-term operation, resulting in optical signal attenuation and connector performance degradation. Due to poor real-time performance and data loss, the traditional manual periodic detection method is difficult to capture the dynamic performance changes of the connectors in a timely manner. To address this problem, Embodiment 1 uses the technical solution of the present invention to construct a system integrating data acquisition, real-time analysis, and life prediction. The system configuration and operation process are as follows:

[0148] 1. Data acquisition module: Install a micro sensor module on each optical fiber connector in the data center to collect real-time environmental parameters (such as temperature, humidity, vibration amplitude) and optical signal parameters (such as light intensity, optical loss). These data are uploaded to the cloud platform through a communication interface to form a real-time monitoring data stream.

[0149] 2. Data processing and preprocessing: The collected raw data is preprocessed through the algorithm of the present invention in the cloud platform, including denoising, outlier removal, and normalization processing of multi-dimensional parameters. For example, when an abnormal change occurs in the environmental parameters of a certain connector (such as a sudden increase in temperature by 5°C or a sudden increase in humidity by 10%), the system corrects the data through a real-time filtering algorithm to reduce the interference of abnormal fluctuations on the overall model.

[0150] 3. Performance State Analysis: Using the preprocessed data, the present invention quantifies the performance state of fiber optic connectors by introducing a multi-dimensional non-linear dynamic model. This model integrates optical signal intensity, environmental parameters, and historical trend data, and calculates the health index of each connector through iterative calculations. The result of the health index is represented in the range of 0 - 100, where 0 - 50 indicates that the performance of the connector has significantly deteriorated and requires immediate maintenance; 50 - 80 indicates a slow decline in performance; and 80 - 100 indicates the normal performance range.

[0151] 4. Application of the Lifetime Prediction Model: The present invention adopts a machine learning-based lifetime prediction model, which analyzes historical data and real-time health indices. The model dynamically adjusts the risk weight coefficient and performs non-linear fitting on multi-dimensional data inputs to achieve high-precision prediction of the remaining lifetime of the connector. Specifically, the model uses known environmental data and historical failure records to train an adaptive prediction model. For example, if a certain connector shows a faster performance degradation trend in a high-humidity environment during past operations, the model will automatically increase the weight of the humidity parameter in the prediction process to enhance the accuracy of the result.

[0152] Through the above steps, the system has successfully achieved real-time performance monitoring and lifetime prediction of fiber optic connectors in data centers. Specific experimental results show that: the quantitative analysis of the health index enables maintenance personnel to intuitively judge the performance state of the connector, avoiding the problem of discontinuous data in previous manual detection methods; the introduction of the dynamic risk weight coefficient significantly improves the accuracy of lifetime prediction, and the average prediction error of the system in various complex environments has been reduced from 15% of the traditional model to 5%; real-time data collection and multi-dimensional data fusion enable the system to operate stably in complex scenarios with high-frequency operation, further reducing the failure rate of fiber optic connectors.

[0153] For example, in an actual test of a data center, due to long-term exposure to a high-temperature and high-humidity environment, the health index of a certain fiber optic connector continuously decreased. The system predicted in advance that the remaining lifetime of the connector was insufficient and promptly alerted the maintenance personnel, avoiding a potential communication interruption failure. At the same time, through the analysis report of the system, it was found that there were defects in the humidity control in this area, resulting in an accelerated degradation of the connector's performance. This discovery prompted the data center operator to optimize the regional environmental control strategy, fundamentally reducing the occurrence of similar problems.

[0154] Example 2:

[0155] This embodiment takes the maintenance of access layer nodes in an optical fiber communication network as the actual application scenario. In this scenario, the optical fiber connectors of the access layer nodes are an important part of the communication network, and their performance directly affects the stability of the entire network. The traditional maintenance method mainly relies on periodic detection and passive fault response, making it difficult to detect performance degradation problems in a timely manner, resulting in a relatively high risk of network interruption. In this experiment, the access layer nodes of a certain urban communication network were used as the object, and 200 optical fiber connectors were selected. Among them, 100 were maintained using the traditional method (control group), and 100 were maintained using the detection and prediction system of the present invention (experimental group). The experimental period was 6 months, and the performance of the two groups of connectors in terms of performance monitoring accuracy, fault warning accuracy, and maintenance efficiency was recorded. The technical implementation and experimental steps are as follows:

[0156] 1. System deployment and initialization: Deploy the performance detection and life prediction system of the present invention on the 100 optical fiber connectors in the experimental group, including a micro-sensor module, a data acquisition module, and a cloud platform analysis system. The core parameters such as the optical signal intensity and optical loss of the optical fiber connectors are collected in real time through the sensor module, and a complete data stream is formed in combination with environmental parameters (such as temperature, humidity, and vibration) and uploaded to the cloud platform.

[0157] 2. The control group adopts the traditional periodic detection method, and conducts manual detection on the connectors once a month, recording the optical signal attenuation data and environmental conditions.

[0158] 3. Data processing and analysis: The data of the experimental group passes through the intelligent preprocessing module of the cloud platform to complete outlier removal, multi-dimensional normalization processing, and dynamic threshold judgment. For example, when the optical signal intensity of a certain connector drops by more than 5% and is accompanied by high humidity fluctuations, the system automatically marks it as potentially abnormal and generates a health index and life prediction result through a non-linear dynamic model. The control group only relies on manual detection to record the historical optical signal changes and lacks real-time processing and dynamic analysis capabilities.

[0159] 4. Fault warning and maintenance response: The life prediction model of the experimental group generates the remaining life prediction value in real time through multi-dimensional parameter fitting and dynamic weight adjustment. When the health index of a certain connector is lower than 50 or the remaining life is lower than 30 days, the system issues a warning and recommends that the maintenance personnel take targeted maintenance measures. The control group only responds after detecting performance anomalies during manual detection, usually lagging behind the actual fault occurrence time.

[0160] Through 6 months of continuous experiments, the following comparison results were obtained: The accuracy of the optical signal intensity and environmental parameters monitored by the experimental group's system remained above 99%, capable of capturing minute fluctuations in real time; due to the overly long time intervals in the manual detection of the control group, approximately 20% of the initial characteristics of performance degradation were missed; the experimental group's system issued accurate fault warnings for 20 connectors in advance, and performance critical value drops occurred in 18 of them within 1 month as verified; the control group only detected 5 abnormal connectors during the same period, and 4 of them were detected only after actual faults occurred; the experimental group achieved precise maintenance through real-time analysis results, reducing the average maintenance time for each connector to 15 minutes and reducing the total maintenance cost by approximately 30%. Due to relying on manual periodic detection, the average maintenance time of the control group was 35 minutes and the total cost was relatively high.

[0161] Through the data fusion and dynamic analysis model, the present invention can capture subtle performance changes and generate a health index in real time, greatly improving the monitoring accuracy and avoiding the omission of performance degradation caused by data breakpoints in traditional methods; experiments show that the life prediction model of the present invention uses multi-dimensional non-linear analysis and dynamic weight adjustment, significantly improving the accuracy of fault prediction, especially being outstanding in complex environments such as high humidity and vibration fluctuations; by transforming the fault from passive response to active predictive maintenance, the present invention effectively reduces unnecessary maintenance operations to improve resource utilization.

[0162] For example, in a network optimization project, a certain communication company tried to apply the system of the present invention to monitor the performance and predict the life of 1000 optical fiber connectors. In the 3rd month after deployment, the system detected that the health index of 12 of these connectors continued to drop below 40, predicting insufficient remaining life. The maintenance personnel carried out precise maintenance based on the system warning, successfully avoiding communication interruption accidents caused by connector failures. At the same time, through the analysis report generated by the system, the company found that the relatively high humidity in certain areas was the main fault inducement, thus optimizing the control strategy for the computer room environment and further improving the network operation stability.

[0163] Example 3:

[0164] This example demonstrates the detailed implementation of the data adaptive calibration and dynamic compensation formula. First, the calibration formula is optimized and boundary conditions are supplemented, that is, the dynamic calibration formula is used to adjust the optical power in real time according to the deviation of environmental parameters to enhance the robustness of the system. The calibration formula is as follows:

[0165] P 校准 = P 原始 ·(1 + W T ·α T ·ΔT + W H ·α H ·ΔH + W V ·αV ·ΔV)

[0166] Parameter source and boundary conditions: P 原始 : The optical power output of the modulation signal, provided by the light source module, with an initial range of 0.5 mW to 10 mW.

[0167] W T ,W H ,W V : Dynamic weights, respectively representing the relative impacts of temperature, humidity, and vibration on the optical power. The calculation is shown in Section 1.2.

[0168] α T ,α H ,α V : The calibration coefficients are obtained through experimental fitting and are applicable to the following environmental ranges respectively:

[0169] Temperature (ΔT): -20°C to 80°C, corresponding to α T = 0.003. Humidity (ΔH): 0% to 95%, corresponding to α H = 0.002. Vibration (ΔV: 0g to 2g (acceleration due to gravity)), corresponding to α V = 0.0015. If the parameters exceed the above ranges, the system will issue an alarm, and it is recommended to use a high-sensitivity compensation module. In a high-humidity environment (humidity 85% - 95%) for example, the calibration formula can reduce the calculation error of optical loss from 4% to 1%. In a strong vibration scenario (vibration amplitude > 1.5g), the deviation of the calibrated optical power does not exceed 0.5%.

[0170] And in the calculation and applicability of dynamic weights, the calculation method of dynamic weights is as follows:

[0171]

[0172] Special treatment: When |ΔT| + |ΔH| + |ΔV| = 0|, the weights are uniformly set to Boundary condition supplement: When the parameter change amplitude is extremely small (|ΔT|, |ΔH|, |ΔV| < 1), the influence of dynamic weights on the calibration formula can be ignored.

[0173] In terms of the refinement of dynamic modulation signal generation and acquisition, the dynamic frequency generation formula of the modulation signal is as follows:

[0174]

[0175] Parameter explanation and value range: f 0: Fundamental frequency, set to 1 MHz, to ensure the high bandwidth requirements of signal modulation. Δf: Frequency increment, set to 10 Hz, suitable for common optical fiber communication interfaces. If the signal reflection characteristics are complex (such as multimode optical fiber), it can be adjusted to 1 Hz to improve the frequency separation ability. β: Dynamic adjustment factor, experimentally fitted to 0.05. Signal-to-noise ratio, where σ i is the mean square deviation of the background noise, calculated by sampling average.

[0176] And in extreme condition adaptation, for example, when the SNR is extremely low (<10), Δf in the formula will automatically increase to 30 Hz to avoid signal separation failure. Experimental verification shows that the signal separation accuracy rate has increased from 85% to 98% after adjustment.

[0177] Furthermore, in terms of optimizing the acquisition process, wavelet threshold denoising can be used: filter the high-frequency noise in the reflected signal. The experimental results show that the calculation deviation of the signal power after denoising is reduced by 50%; and, hardware requirements: the sampling rate needs to reach 10 GHz to ensure high-precision capture of signal reflection.

[0178] The deep learning model combines CNN and LSTM. The detailed configuration is as follows: in terms of the convolutional layer, the first convolutional layer: kernel size 3×3, stride 1, used to extract the time-domain features of the optical signal; the second convolutional layer: kernel size 5×5, stride 2, extract the frequency-domain features and embed the environmental features at the same time; LSTM layer: input historical optical loss data, the number of neurons in the hidden layer is 50, used to predict the performance degradation trend; fully connected layer: integrate the convolutional and LSTM features and output the classification results (loss traceability and fault classification).

[0179] In terms of the adaptability of model training, such as the transfer learning method: pre-train in the laboratory environment and fine-tune using real-scene data. The parameters for fine-tuning include the weights of the convolutional layer and the time step of the LSTM; and the ratio of the fine-tuning set to the training set: 3:7, to ensure the generalization ability. In a high-humidity environment, the model classification accuracy rate reaches 96.5%, and the prediction error is controlled within 2.5%; in a scenario where the vibration amplitude exceeds 2 g, the model still maintains an accuracy rate of more than 95%.

[0180] For the hardware adaptability and environmental adaptability, such as the accuracy of the temperature sensor: ±0.1 °C, applicable range -40 °C to 85 °C; the accuracy of the humidity sensor: ±1%, applicable range 0% to 100%; the sensitivity of the vibration sensor: 0.1 g, sampling rate 10 Hz; in an extreme environment where the humidity > 95%, vibration > 2 g, and temperature > 80 °C, the system can improve the performance by adding anti-interference modules (such as vibration dampers, humidity isolation covers), which all belong to the extended implementation methods known to those of ordinary skill in the art.

[0181] Example 4:

[0182] To verify the adaptability of the present invention in various complex environments, in this embodiment, three typical scenarios are used as the experimental basis, namely the high-humidity environment of the data center, the industrial vibration environment, and the extreme temperature difference environment, to simulate typical complex situations that may be encountered in actual use:

[0183] High-humidity environment of the data center: The humidity range is set to 80%-95%, and the temperature is controlled at 25°C ± 2°C to simulate the usage scenario of high-density fiber optic connectors.

[0184] Industrial vibration environment: The test scenario is set in the high-frequency mechanical vibration area, with the vibration amplitude ranging from 0.5g to 2g and the frequency from 10Hz to 50Hz, to verify the stability of the system in a high-vibration environment.

[0185] Extreme temperature difference environment: The rapid temperature difference change from -20°C to 80°C is simulated through a temperature chamber to detect the performance of the system under extreme temperature conditions.

[0186] In each scenario, 100 fiber optic connectors of different models are monitored for a long time, and the following key data are collected:

[0187] Environmental parameters: The temperature, humidity, and vibration amplitude are collected in real time, and the change values per minute are recorded. Optical signal data: The light intensity, optical loss value, and signal-to-noise ratio are collected through the photoelectric detection module. Equipment operation status: The failure occurrence frequency and degradation trend of the fiber optic connector under different environmental conditions are recorded.

[0188] 2. Data preprocessing and dynamic adjustment method.

[0189] The collected experimental data is large in quantity and noisy, and effective preprocessing must be carried out to ensure the accuracy of the analysis. The following methods are adopted in this embodiment:

[0190] Outlier removal: The 3σ principle is used to remove outliers in the environmental sensor or optical signal acquisition to ensure the continuity and accuracy of the data.

[0191] Data normalization processing: Multidimensional data such as temperature, humidity, and vibration are normalized to eliminate the influence of different dimensions on model training.

[0192] Dynamic threshold adjustment: The judgment threshold for optical signal attenuation is updated in real time during the monitoring process. For example, in a high-humidity environment, the system dynamically lowers the threshold for judging optical loss to adapt to the change of signal characteristics.

[0193] And to improve the accuracy and applicability of the model, this embodiment has improved the model training from the following aspects:

[0194] Expansion of training data sources: Combine the experimentally collected data (300 groups) and introduce the publicly available actual optical fiber fault data sets in the industry (such as 1000 groups of historical data provided by a certain optical fiber manufacturing enterprise) to enrich the model training samples.

[0195] Model transfer learning: First, pre-train the model using the publicly available data set, and then fine-tune it with the experimentally collected data to ensure that the model has both generality and can adapt to specific scenarios. Data augmentation method: Generate virtual samples by perturbing the collected data (such as adding noise, changing the sampling frequency) to enhance the model's robustness to environmental changes. Performance evaluation metrics: During the training process, the core evaluation metrics are the optical loss prediction error, signal classification accuracy, and fault diagnosis sensitivity, and the hyperparameters of the model (such as the learning rate, number of network layers, etc.) are dynamically adjusted.

[0196] At the same time, in order to ensure the applicability of the system in different environments, the performance of the hardware module in this embodiment is further optimized and verified:

[0197] Optimization of environmental sensors: Use industrial-grade high-precision sensors (such as the temperature sensor accuracy is ±0.1°C, and the humidity sensor accuracy is ±1%) to enhance the system's adaptability to extreme environments. Enhancement of the optoelectronic detection module: Add a signal filtering module to reduce noise during the acquisition of optical signals in a high-vibration environment. Extreme condition test: In an extreme environment of -20°C to 80°C, humidity of 95%, and vibration amplitude of 2g, the system can operate stably for 72 hours, and the optical loss calculation error remains within 1%.

[0198] Through the above experiments and optimizations, the following results are obtained in this embodiment: When the humidity is as high as 95%, the optical loss calculation error is reduced from 4% to 1.5%, and the prediction accuracy is increased to 92.3%; under the conditions of vibration amplitude of 2g and frequency of 50Hz, the signal-to-noise ratio remains above 80%, and the system stability is not significantly affected; when the temperature change range is from -20°C to 80°C, the optical signal acquisition and analysis module operates normally, and there is no obvious data loss phenomenon.

[0199] For those skilled in the art, the present invention is not limited to the details of the above exemplary embodiments, and the present invention can be implemented in other specific forms without departing from the spirit or basic characteristics of the present invention. Therefore, in any regard, the embodiments should be regarded as exemplary and non-limiting, and the scope of the present invention is defined by the appended claims rather than the above description. Therefore, it is intended to embrace all changes falling within the meaning and scope of the equivalent elements of the claims within the present invention.

Claims

1. A method for automatically detecting optical loss of an optical fiber connector, characterized in that: The following steps are involved: Step 1, generation of multi-frequency composite modulation signals: generating multiple groups of modulation signals by a multi-frequency composite modulation light source, each group of modulation signals having different frequencies and phases, specifically including: Modulation signal frequency f i Generated dynamically according to the following formula: f i =f0+i·Δf Among them, f0 is the basic frequency, Δf is the frequency increment, and i is the modulation signal number; Modulation signal phase φ i Dynamically set according to the following formula: φ i =φ0+g(i) Among them, φ0 is the initial phase, g(i) is the phase adjustment function, which is determined according to the physical characteristics of the optical fiber connector and the detection scenario; The modulated signal is injected into the optical fiber connector to be detected; Step 2, collection and separation of optical reflection signals: The reflection signals at the optical fiber connector interface are collected in real time by an optical reflection receiving module, wherein: The collected reflection signal contains light intensity, phase and delay characteristics; The reflected signal is separated in the frequency domain by using the fast Fourier transform (FFT) to generate a reflected signal data set, which contains the reflected signal data corresponding to different frequencies f i The incident light power P 入射 (f i ) and reflected light power P 入射 (f i ); Step 3, adaptive calibration of environmental parameters: The temperature, humidity and vibration information of the detection environment are detected by the environmental sensing module, and the optical power output and receiving sensitivity of the modulated signal are dynamically adjusted according to the following calibration formula: P 校准 =P 原始 ·(1+a T ·ΔT+α H ·ΔH+α V ·ΔV) in: P 校准 is the calibrated optical power, P 原始 is the initial optical power; ΔT, ΔH, and ΔV are the deviation values ​​of ambient temperature, humidity, and vibration, respectively; α T , α H , α V is the calibration coefficient obtained from experimental fitting; Step 4, calculation of optical loss: Based on the reflection signal data set, the optical loss value of the optical fiber connector is calculated by the following formula: Where: P 入射 (f i ) and P 反射 (f i ) are the frequencies f i The incident light power and reflected light power under The comprehensive optical loss value L is calculated by the following weighted average formula: 综合 : Among them, w i is the dynamic weight coefficient, which is determined by the least square fitting method according to the signal-to-noise ratio and frequency range of the reflected signal, and N is the total number of modulated signals; Step 5, loss tracing and fault classification: Based on the optical loss value and reflection signal characteristics, the optical loss source is traced and the fault type is classified through a pre-trained deep learning model, where: The input of the deep learning model includes the intensity feature, phase feature, and delay feature of the reflected signal; The model is trained using 1,000 sets of light reflection signal data collected in the laboratory through a convolutional neural network training method, and outputs classification results including the location of the loss source and the type of fault. Step 6, performance degradation trend prediction: through the time series analysis algorithm, based on the collected historical optical loss data, predict the future performance degradation trend of the optical fiber connector, and output a performance prediction report containing the predicted time point and loss change trend; Step 7, result output: The comprehensive optical loss value, traceability diagnosis result and performance prediction report of the optical fiber connector are uploaded to the cloud platform through the data interface. The cloud platform visualizes the data and generates maintenance suggestions.

2. A method, device and equipment for automatically detecting optical loss of an optical fiber connector according to claim 1, characterized in that: The specific form of the phase adjustment function g(i) of the modulation signal is: g(i)=A·sin(B·i)+C Among them, A is the phase adjustment amplitude, which is measured based on the maximum dynamic range of the fiber reflection signal; B is the frequency factor, which depends on the sampling rate of the modulation signal; C is the phase offset, which is determined by the dielectric constant of the fiber connector interface.

3. The method, device and equipment for automatically detecting optical loss of an optical fiber connector according to claim 1, characterized in that: In the reflection signal data set, the incident light power P 入射 (f i ) and P 反射 (f i ) includes the following steps: The reflected signal is sampled using a highly sensitive photodetector; The sampled signal is passed through a digital filter to suppress environmental noise and low-frequency interference; Fast Fourier transform (FFT) is used to extract the frequency domain characteristics of the reflected signal, and the optical power value is calculated based on the frequency domain characteristics.

4. A method, device and apparatus for automatically detecting optical loss of an optical fiber connector according to claim 1, characterized in that: The adaptive calibration of the environmental parameters further comprises the following steps: Collect no less than 1,000 sets of historical data on the temperature, humidity and vibration of the testing environment, and establish a multidimensional data association model; Fitting the calibration coefficient α using the weighted linear regression method T , α H , α V , where the weight factor is determined by the time decay function of historical data and the fluctuation amplitude of real-time data; The compensation parameters of the optical loss calculation formula are adjusted in real time according to the calibration coefficient.

5. The method, device and equipment for automatically detecting optical loss of an optical fiber connector according to claim 1, characterized in that: The weighting coefficient w for the optical loss calculation is i The dynamic determination of includes the following steps: For each frequency f i Calculate the signal-to-noise ratio SNR i , the formula is: Among them, σ i is the frequency f i The corresponding background noise mean square error is obtained by averaging multiple samples; According to the signal-to-noise ratio SNR i and frequency distribution characteristics, using the Bayesian estimation-based optimization method to dynamically assign weight coefficients w i .

6. A method, device and equipment for automatically detecting optical loss of an optical fiber connector according to claim 1, characterized in that: The convolutional neural network structure of the deep learning model includes the following modules: The first convolutional layer is used to extract the time domain features of the reflected signal; The second convolutional layer is used to extract the frequency domain features of the reflected signal; The third convolutional layer generates feature maps of specific fault modes by fusing multi-scale features; A fully connected layer inputs the feature map into the classifier and outputs the loss classification and fault source identification results of the optical fiber connector.

7. The method, device and equipment for automatically detecting optical loss of an optical fiber connector according to claim 1, characterized in that: The time series analysis algorithm for predicting performance degradation trends is based on a long short-term memory (LSTM) neural network, and the training of the neural network includes the following steps: The input feature is the historical optical loss value L 综合 , detection environment parameters and fault categories; Optimize network weights using back-propagation with gradient clipping; The network output is the optical loss value at a future time point and its changing trend.

8. The method, device and equipment for automatically detecting optical loss of an optical fiber connector according to claim 1, characterized in that: The result output further includes: The comprehensive optical loss value and performance prediction results of the optical fiber connector are stored in the cloud through a distributed database interface; Use WebGL-based real-time visualization engine to generate optical loss trend graphs and prediction curves; Generate maintenance recommendations based on prediction curves, including recommended maintenance time windows and key component replacement reminders.

9. The method, device and equipment for automatically detecting optical loss of an optical fiber connector according to claim 1, characterized in that: The method is applied to an optical fiber connector detection device, which comprises: A modulation signal generation module, used to generate a multi-frequency composite modulation signal and inject it into the optical fiber connector port; A photoelectric detection module is used to collect the reflected signal returned from the optical fiber connector; A data processing module, used for denoising, separation and optical loss calculation of reflected signals; The output module outputs the detection results and fault classification information through the display screen or network interface.

10. The method, device and equipment for automatically detecting optical loss of an optical fiber connector according to claim 1, characterized in that: The method is applied to a performance monitoring device of an optical fiber communication system, the device comprising: A data acquisition unit, used for acquiring a reflection signal from an optical fiber connector in real time; Data analysis unit, used for optical loss tracing and performance prediction based on deep learning models; The network interface unit is used to upload the analysis results to the cloud platform and generate remote maintenance reports.

Images