Method and system for measuring insertion loss of prefabricated optical cable high-density optical fiber connector

Abstract

The invention discloses an insertion loss measurement method and system for a prefabricated optical cable high-density optical fiber connector, and relates to the technical field of optical cable loss prediction.The method comprises the steps that end face cleaning is conducted on a connector to be measured in the factory leaving stage, and a factory leaving baseline certificate is generated and stored; carrying out a wet accelerated aging test, and collecting corresponding wet insertion loss, OTDR, an end face image and a sensor time sequence as wet characteristics; calculating a theoretical coupling loss baseline by using a physical coupling model and taking optical fiber parameters as input; constructing an underwater optical cable insertion loss prediction model, and predicting a final connector loss value of a wet state additional loss residual error by taking the wet state characteristics and the image characteristics as input; establishing a life distribution model, and carrying a wetting judgment strategy to judge whether a wetting event occurs or not; and triggering the wet accelerated degradation sub-model. According to the method, the problem of multi-band deletion of samples in training data due to difficulty in frequent sampling or end face microscopic examination after seabed deployment is solved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of optical cable loss prediction, in particular to a method and system for measuring the insertion loss of a high-density fiber connector of a prefabricated optical cable. BACKGROUND

[0002] With the rapid development of the interconnection of submarine optical cables, ocean observation networks and near-shore data centers, prefabricated optical cables and high-density fiber connectors are widely used to shorten the on-site access time, reduce the construction risk and improve the system integration efficiency. In these application scenarios, accurate measurement and long-term monitoring of the core-by-core insertion loss become the core engineering problems to ensure the reliability of the link, develop the operation and maintenance strategy and evaluate the remaining service life. The existing mainstream core-by-core IL measurement methods mainly include light source and optical power meter, OTDR, end face microscopic imaging and IEC 61300-3-35 grading; these means are good in the laboratory or factory environment, but there are many limitations when directly transplanted to the underwater in-service scene such as submarine, including: It is difficult to frequently sample or do end face microscopic examination after submarine deployment, many degradation signs (end face moisture, micro-cracks, small area corrosion) are invisible or only exposed after failure, resulting in extremely scarce known failure time / cause in training data and sample multiple deletion.

[0003] Laboratory accelerated aging test can quickly observe the wet state degradation curve, but there are non-negligible differences between the acceleration conditions and the temperature gradient of the real sea conditions, the real pressure fluctuation, the biological attachment dynamics, etc., thus causing the problem of poor generalization ability.

[0004] The coupling of various external factors makes the degradation present a highly nonlinear and history-dependent behavior, and it is difficult for single-factor models or linear statistical methods to describe the interaction effects, such as humidity + temperature + salinity jointly accelerating the degradation of sealing materials, and pressure cycling can amplify the optical contact degradation caused by micro-looseness, etc.

[0005] Therefore, the present application provides a method and system for measuring the insertion loss of a high-density fiber connector of a prefabricated optical cable. SUMMARY

[0006] The purpose of the present application is to provide a method and system for measuring the insertion loss of a high-density fiber connector of a prefabricated optical cable to solve the existing problems raised in the background.

[0007] To achieve the above purpose, the present application provides the following technical solution: a method for measuring the insertion loss of a high-density fiber connector of a prefabricated optical cable, comprising the following steps: S1, cleaning the end face of the connector to be tested at the factory stage and collecting the end face image, core-by-core IL and OTDR baseline, generating and storing the factory baseline certificate; S2. Before leaving the factory, conduct wet accelerated aging tests and collect corresponding wet insertion loss, OTDR, end face images and sensor time series as wet characteristics; S3. Using a physical coupling model, the fiber parameters are used as input to calculate the theoretical coupling loss baseline. ; S4. Construct an underwater optical cable insertion loss prediction model, using the aforementioned wet-state features and image features as input, to predict the wet-state additional loss residual. and uncertainty ; Calculate the final connector loss value ; S5. Establish a lifetime distribution model based on historical losses. The sequence and the wet intrusion event information are used to calculate the remaining useful life distribution (RUL) of each core using degradation curves and survival analysis methods, and output the maintenance priority; at the same time, a wet intrusion judgment strategy is implemented to determine whether a wet intrusion event has occurred; if a wet intrusion event is detected, the wet accelerated degradation sub-model is triggered and the process returns to S4 to re-estimate the degradation rate of each core.

[0008] A further improvement of this invention is that the wet accelerated aging test and wet characteristic acquisition in step S2 include: S21. Perform N sets of wet accelerated aging tests on the same batch of sample cables according to the preset test matrix. The test matrix includes at least a combination of pressure, humidity immersion mode, temperature and salt spray intensity. Each set of tests is run on several repeated samples to obtain a statistically repeatable degradation sequence. S22. In each wet accelerated aging test group, continuously collect and store: sensor records that change over time, event-triggered core-by-core insertion loss IL(t) sequence, OTDR waveform snapshots and their event list, end-face image sequence, and test metadata; S23. Obtain the wet characteristic set from the wet accelerated aging test data through the following processing: calculate the humidity peak, the time of first threshold breach, the cumulative wet exposure duration, the pressure sensitivity coefficient, the proportion of wet spot area on the end face, the deterioration of end face structural similarity, the amplitude and width changes of OTDR events before and after wet conditions, and the wet IL increment.

[0009] A further improvement of this invention is that the establishment and calculation of the physical coupling model includes: S31, using the mode field diameter at both ends of the optical fiber , As the core quantity, the Gaussian mode overlap approximation is used to calculate the fundamental coupling loss term caused by mode field mismatch. ; S32. Calculate the additional loss term caused by centering based on centering error. The eccentricity square ratio and Gaussian radius are used as approximations to correct the eccentricity and reflect the weakening effect of eccentricity on coupling efficiency. S33. Consider the bending radius and use an empirical index to calculate the bending loss term; S34, Combine the basic coupling loss term and the additional loss term. After normalizing the bending loss term, a weighted sum is performed to obtain the theoretical coupling loss baseline. .

[0010] A further improvement of this invention is that the underwater optical cable insertion loss prediction model adopts a hybrid structure of physical prior and residual learning, with theoretical coupling loss baseline. Using the baseline input, a neural network residual module is constructed to predict the residual of wet-state additional loss. and uncertainty log variance Final output connector loss value IL estimation ; The neural network includes an end-face image branch, an OTDR waveform branch, a time-series sensor branch, and a metadata branch; this means that self-supervised pre-training is performed on unlabeled end-face images and OTDR data, and then fine-tuned on data containing wet accelerated aging tests and factory baseline labels; when the model predicts a residual variance greater than a set residual variance threshold on wet samples, the weight of wet samples in the model is proportional to the variance.

[0011] A further improvement of this invention is that the process of establishing the lifetime distribution model includes processing the time series of predicted connector loss values ​​for each core. Construct degradation curves and fit parameters ; Let represent the factory baseline, a and b represent natural degradation terms, and c and d represent wet-accelerated degradation terms. The parameters are obtained using Bayesian methods, and their confidence intervals are provided. This represents a wet intrusion event indicator function; simultaneously, a survival analysis model is trained in parallel, and a wet intrusion judgment strategy is used to determine whether a wet intrusion event has occurred. When the result is negative... The value is 0. When the result is yes, after confirming the wet accelerated degradation terms a and b through the wet accelerated degradation sub-model, the remaining service life distribution of each core is generated by combining the degradation curve and the output of the survival model.

[0012] A further improvement of this invention is that the survival analysis model is constructed based on the Cox proportional hazards model, which is expressed as: , Indicates instantaneous failure rate. Indicates baseline hazard. This indicates the relative scaling of the covariates with respect to the baseline hazard. Indicates the most recent connector loss value The concatenation vector of wet features is used to obtain the survival function. , which represents the probability that the device has not failed by time t.

[0013] A further improvement of this invention is that the wet intrusion judgment strategy includes multi-layer judgment logic: Threshold determination layer, when the humidity sensor recorded a value that meets the time requirement... When the humidity at any time t within the time period is greater than or equal to a set humidity threshold, it is judged as initial wet intrusion. The linked evidence layer outputs the wet intrusion probability through a weakly supervised merger while the threshold judgment layer remains valid. ,when Greater than or equal to the set threshold At that time, it was determined to be a wet intrusion event.

[0014] A further improvement of this invention is that the wet-accelerated degradation sub-model uses the wet dynamics fitted in the wet-accelerated aging test as a priori, adopts a parameterized form, and the initial values ​​of wet parameters c and d are obtained from the fitting results of the wet-accelerated aging test, and are mapped to the equivalent values ​​in the field through hierarchical Bayesian wet-accelerated aging test field mapping and EAF transformation; the mapping process outputs the posterior distribution of c and d to reflect the uncertainty; when a wet intrusion event occurs, the sub-model receives high-frequency IL and OTDR sensor data uploaded from the edge and performs online Bayesian posterior updates to c and d according to an incremental window to quickly converge to the wet parameter estimate of the specific sample and update the RUL; the output of the wet-accelerated degradation sub-model includes wet amplitude estimation, rate estimation, parameter confidence interval, IL time series prediction of the wet stage, and the updated RUL distribution; when the wet amplitude estimate or the updated RUL reaches the maintenance threshold, a maintenance instruction is automatically generated and the active sampling priority is updated.

[0015] On the other hand, the present invention provides an insertion loss measurement system for prefabricated high-density fiber optic connectors, comprising: The baseline acquisition module is used to clean the end face of the connector under test and acquire end face images, core-by-core IL and OTDR baselines during the factory manufacturing process, and generate and store the factory baseline certificate. The wet-state feature acquisition module is used to conduct wet-state accelerated aging tests before leaving the factory and acquire corresponding wet-state insertion loss, OTDR, end face images and sensor time series as wet-state features; The physical coupling module is used to calculate the theoretical coupling loss baseline by taking fiber parameters as input. ; The underwater optical cable insertion loss prediction model construction module is used to predict the wet-state additional loss residual using the aforementioned wet-state features and image features as input. and uncertainty ; Calculate the final connector loss value ; The lifetime distribution model building module is used to build models based on historical losses. The sequence and the wet intrusion event information are used to calculate the remaining service life distribution (RUL) of each core using degradation curves and survival analysis methods, and output the maintenance priority; at the same time, a wet intrusion judgment strategy is implemented to determine whether a wet intrusion event has occurred; if a wet intrusion event is detected, the wet accelerated degradation sub-model is triggered and the process returns to the underwater optical cable insertion loss prediction model construction module to re-estimate the degradation rate of each core.

[0016] Compared with the prior art, the beneficial effects of the present invention are: 1. This invention first combines a physical coupling model with a neural network-based residual learning module to improve the generalization capability in real-sea environments and reduce the expensive recovery costs caused by false alarms and misjudgments; it solves the problem of difficulty in frequently recovering the end face for visual inspection when deployed on the seabed; and it estimates the core-by-core loss and provides confidence levels through remotely measurable multimodal data under underwater conditions, reducing the dependence on recovery verification. 2. Secondly, a complete wet-state curve, OTDR end-face evolution, and sensor response were obtained through a systematic wet-state accelerated aging matrix experiment at the factory. Wet-state degradation parameters were fitted and used as priors. Hierarchical Bayesian mapping was used to map the parameters learned under accelerated conditions to the equivalent values ​​in the field. Online posterior update was achieved to quickly integrate high-frequency field observations into parameter estimation, significantly reducing RUL uncertainty. This solved the problem that seabed wet intrusion is usually invisible and degradation may deteriorate rapidly in a short period of time. Rapid detection and online re-estimation can shorten response time and reduce the risk of sudden interruption. 3. Modeling of in-service samples containing right censoring or interval censoring is performed by combining parameterized degradation curves and survival analysis; hierarchical Bayesian modeling is used to jointly infer wet accelerated aging group information with a small number of field-recovered samples. In seabed scenarios where censored data is predominant, the survival model and Bayesian hierarchical structure can unbiasedly utilize surviving sample information and quantify uncertainty; the problem of model inaccuracy caused by small number of sea trial samples, domain shift and multi-stress coupling is solved. Attached Figure Description

[0017] Figure 1 This is a flowchart of the insertion loss measurement method for prefabricated high-density optical fiber connectors of the present invention. Figure 2 This is a schematic diagram of the insertion loss measurement system for the prefabricated high-density optical fiber connector of the present invention. Detailed Implementation

[0018] The technical solution of the present invention will be described in detail below with reference to the accompanying drawings and specific embodiments. It should be understood that the embodiments of the present invention and the specific features in the embodiments are detailed descriptions of the technical solution of the present invention, rather than limitations thereof. In the absence of conflict, the embodiments of the present invention and the technical features in the embodiments can be combined with each other.

[0019] The term "and / or" is merely a description of the relationship between related objects, indicating that there can be three relationships. For example, A and / or B can represent three cases: A exists alone, A and B exist simultaneously, and B exists alone.

[0020] Example 1 Figure 1 A flowchart illustrating the insertion loss measurement method for prefabricated high-density fiber optic connectors disclosed in this embodiment is shown, with the following steps: S1. During the factory stage, the connector under test is cleaned and the end face image is collected. The IL and OTDR baselines are obtained one by one and the factory baseline certificate is generated and stored. This is used to establish a complete baseline data package for each prefabricated optical cable and its high-density connector under "dry factory conditions"; the specific steps include: In a temperature-controlled cleanroom (or semi-clean testing station), the ambient temperature and relative humidity are recorded and written into the certificate. S2. Before leaving the factory, conduct wet accelerated aging tests and collect corresponding wet insertion loss, OTDR, end face images and sensor time series as wet characteristics; The complete shape of the IL curve under wet conditions is obtained to fit c and d in the degradation model; OTDR features and end-face microscopic image features in the early stage of wet invasion are used to train the residual learning model; and the accelerated response of sensor data (humidity, temperature, pressure) is also obtained. The typical curve shape of wet invasion degradation is determined.

[0021] The wet accelerated aging test and wet characteristic acquisition include: S21. Perform N sets of wet accelerated aging tests on the same batch of sample cables according to the preset test matrix. The test matrix includes at least a combination of pressure, humidity immersion mode, temperature and salt spray intensity. Each set of tests is run on several repeated samples to obtain a statistically repeatable degradation sequence. S22. In each wet accelerated aging test group, continuously collect and store: sensor records that change over time, including humidity H(t), temperature T(t), pressure P(t), event-triggered core-by-core insertion loss IL(t) sequence, OTDR waveform snapshots and their event list, end face image sequence, and test metadata (test conditions, sample ID, reference pigtail ID, timestamp). S23. Obtain the wet characteristic set from the wet accelerated aging test data through the following processing: calculate the peak humidity, the time to first threshold breach, and the cumulative duration of wet exposure. For humidity threshold (For example, at 75% RH) an indicator function is used to statistically analyze the total time spent above the threshold, which serves as a quantitative indicator of wet exposure and a pressure sensitivity coefficient. The unit is This represents the linear approximation sensitivity of IL under a unit pressure change. A large PS indicates that the joint is extremely sensitive to pressure, which may require improvements in sealing or mechanical restraint, the proportion of wet patch area on the end face, and deterioration of end face structural similarity. The end face image features are obtained by semantic segmentation or difference image to obtain wet patch coverage, salt crystal distribution, particle statistics, and scratch expansion. At the same time, the SSIM of the end face image and the factory baseline image are calculated as a measure of end face structural changes. The amplitude and width changes of the OTDR event before and after wetting include the difference in event amplitude, width, and energy before / after wetting. Event counts and locations of newly occurring events, as well as wet IL increments, including direct calculation of IL increments during the wet steady-state or transient phases.

[0022] S3. Using a physical coupling model, the fiber parameters are used as input to calculate the theoretical coupling loss baseline. The establishment and calculation of the physical coupling model include: S31, using the mode field diameter at both ends of the optical fiber , As the core quantity, the Gaussian mode overlap approximation is used to calculate the fundamental coupling loss term caused by mode field mismatch. ; ; S32. Calculate the additional loss term caused by centering based on the centering error (eccentricity e). The eccentricity square ratio and Gaussian radius are used as approximations to correct the eccentricity and reflect the weakening effect of eccentricity on coupling efficiency. S33. Consider the bending radius and use an empirical index to calculate the bending loss term; S34, Combine the basic coupling loss term and the additional loss term. After normalizing the bending loss term, a weighted sum is performed to obtain the theoretical coupling loss baseline. .

[0023] The physical coupling model provides a reliable engineering prior, enabling the residual model to focus its learning objectives on wet / coupling anomalies that the physical model cannot explain, thereby enhancing interpretability and sample efficiency. , Preferred data should be obtained from manufacturer specifications or production inspection records; if not directly available on-site, it can be estimated through factory measurements, reference fiber optic data, or calibration experiments. Centering deviation *e* can be estimated from manufacturing tolerances, mechanical measurements, or image analysis. The bending radius is provided by cable design or installation records. The effect of centering deviation *e* on coupling efficiency is approximately calculated as follows: Proportional attenuation; bending loss can be described by empirical relationships, and the coefficients of these empirical relationships can be obtained from similar fiber standards or experimental calibration.

[0024] S4. Construct an underwater optical cable insertion loss prediction model based on a neural network-based residual learning model, using the wet-state features and image features as inputs to predict the wet-state additional loss residual. and uncertainty ; Calculate the final connector loss value ; The underwater optical cable insertion loss prediction model adopts a hybrid structure of physical prior and residual learning, with theoretical coupling loss baseline. Using the baseline input, a neural network residual module is constructed to predict the residual of wet-state additional loss. and uncertainty log variance Final output connector loss value IL estimation ; The neural network includes an end-face image branch based on a convolutional neural network, an OTDR waveform branch, a timing sensor branch, and a metadata branch (referencing pigtail ID, insertion / removal count, etc.). The heteroscedastic Gaussian negative log-likelihood is used as the loss function to jointly learn the mean and sample uncertainty. The loss form is as follows: ; Self-supervised pre-training, such as contrastive learning or mask reconstruction, is performed on large-scale unlabeled end-face images and OTDR data. Fine-tuning is then performed on data containing wet accelerated aging tests and factory baseline labels. The fine-tuning process includes training only the fusion layer and regression / uncertainty head; Heteroscedastic Gaussian NLL (main task) is used to quickly determine the fusion and output layers without disrupting the pre-trained representation. Subsequently, 1–3 high-level blocks (the latter half of each encoder) are unfrozen, and the head continues training. The output ALT wet sample weights and factory dry sample weights are set to 1; this is used to adapt the features to the discriminative power of the wet residuals. When there are enough labeled samples (≥10,000), an adapter / LoRA is inserted into each encoder, and only the adapter and head are trained; this significantly suppresses overfitting and catastrophic forgetting.

[0025] When the residual variance predicted by the model on wet samples exceeds a set threshold, the weight of wet samples in the model is proportional to the variance, making the model pay more attention to samples with unstable predictions.

[0026] The physical model explains most of the conventional coupling loss, while the residual network focuses on learning additional losses that the physical model cannot cover, such as wetness, contamination, microcracks, and poor contact, thereby reducing the number of training samples required and improving interpretability.

[0027] S5. Establish a lifetime distribution model based on historical losses. The sequence and the wet intrusion event information are used to calculate the remaining useful life distribution (RUL) of each core using degradation curves and survival analysis methods, and output the maintenance priority; at the same time, a wet intrusion judgment strategy is implemented to determine whether a wet intrusion event has occurred; if a wet intrusion event is detected, the wet accelerated degradation sub-model is triggered and the process returns to S4 to re-estimate the degradation rate of each core; the degradation rate of each core is used to describe how fast IL changes over time, that is, the time derivative of IL.

[0028] The lifetime distribution model divides degradation into two parts: natural evolution and wet-triggered degradation. The wet term is determined by the time of wet invasion. The parameters take effect initially and tend towards a saturation value c over time, with the rate controlled by d. c and d under acceleration conditions can be estimated using ALT, and ALT parameters can be mapped to equivalent values ​​in the field using hierarchical Bayesian or EAF methods. Bayesian fitting directly yields the posterior distribution of the parameters and naturally propagates the parameter uncertainty to the RUL distribution.

[0029] Many samples from active-duty maritime vessels are censored (still in service), and survival models can directly handle these data using partial likelihood or interval censoring without introducing bias. The use of time-varying covariates allows survival models to sensitively reflect recent degradation dynamics.

[0030] The process of establishing the lifetime distribution model includes processing each core based on the time series of predicted connector loss values. Construct degradation curves and fit parameters ; Let represent the factory baseline, a and b represent natural degradation terms, and c and d represent wet-accelerated degradation terms. The parameters are obtained using Bayesian methods, and their confidence intervals are provided. This represents the indicator function for the wet intrusion event, prior to the occurrence of wet intrusion. , Represented as 0, the entire wet term has no effect; after wet invasion occurs, , The value is represented as 1, indicating that the wet degradation term is activated; simultaneously, the survival analysis model is trained in parallel, and a wet invasion judgment strategy is used to determine whether a wet invasion event has occurred. When the result is no, If the result is 0, after confirming the wet-accelerated degradation terms c and d through the wet-accelerated degradation sub-model, the remaining service life distribution of each core is generated by combining the degradation curve and the output of the survival model.

[0031] The survival analysis model is constructed based on the Cox proportional hazards model, which is expressed as follows: , Indicates instantaneous failure rate. Represents the baseline hazard, and represents the failure rate curve at that moment without relying on covariates (an unknown function, not specifically parameterized in the Cox model). This indicates the relative scaling of the covariate with respect to the baseline hazard; if the value of a covariate makes... As the value increases, the corresponding instantaneous failure rate increases exponentially. Indicates the most recent connector loss value The concatenation vector of wet features is used to obtain the survival function. , representing the probability that the device has not failed by time t; Cox hazard represents the rate of failure at the next instant, given that the user has already survived at time t. A higher covariate will amplify the failure rate through an exponential function, thus reducing the survival function S(t).

[0032] The wet intrusion detection strategy includes multiple layers of judgment logic: Threshold determination layer, when the humidity sensor recorded a value that meets the time requirement... When the humidity at any time t within the time period is greater than or equal to a set humidity threshold, it is judged as initial wet intrusion. The linked evidence layer outputs the wet intrusion probability through a weakly supervised merger while the threshold judgment layer remains valid. ,when Greater than or equal to the set threshold At that time, it was determined to be a wet intrusion event.

[0033] The wet-accelerated degradation sub-model uses the wet dynamics fitted in the wet-accelerated aging test as a priori, and adopts a parameterized form. The initial values ​​of wet parameters c and d are obtained from the fitting results of the wet-accelerated aging test, and are mapped to the equivalent values ​​in the field through hierarchical Bayesian wet-accelerated aging test field mapping and EAF transformation. The mapping process outputs the posterior distribution of c and d to reflect the uncertainty. When a wet intrusion event occurs, the sub-model receives high-frequency IL and OTDR sensor data uploaded from the edge and performs online Bayesian posterior updates to c and d according to an incremental window to quickly converge to the wet parameter estimate of the specific sample and update the RUL. The output of the wet-accelerated degradation sub-model includes wet amplitude estimation, rate estimation, parameter confidence interval, IL time series prediction for the wet stage, and updated RUL distribution. When the wet amplitude estimate or the updated RUL reaches the maintenance threshold, a maintenance instruction is automatically generated and the active sampling priority is updated.

[0034] The insertion loss measurement method for the prefabricated high-density fiber optic connector may also include an active sampling strategy, continuously collecting time-series and periodic core-by-core loss and OTDR measurements of cable humidity, temperature, and pressure during the in-service phase; for each connector, a candidate scoring function is calculated by weighting the model weights after the true label obtained by the underwater optical cable insertion loss prediction model's prediction variance, Monte Carlo or influence function approximation, and the service value of the impact of connector failure on the system (such as bandwidth loss, critical node impact). For in-service cables, sort candidates by scoring function and select the largest K, where K is determined by budget and operational window, e.g., 3–5 points per month. Issue ROV tasks: conduct wet docking tests or sample and recover end faces for microscopic inspection and complete IL / RL / OTDR scanning at candidate points on-site. Store the collected, high-precision data in a database for retraining or Bayesian posterior update of the underwater optical cable insertion loss prediction model. Feed the on-site data back to the model to reduce laboratory-to-real-sea discrepancies.

[0035] The threshold and weight settings can be set by default according to the present invention, or they can be set by those skilled in the art.

[0036] Example 2 Figure 2 This invention illustrates a framework diagram of an insertion loss measurement system for prefabricated high-density fiber optic connectors. Based on the same inventive concept as Embodiment 1, this invention provides an insertion loss measurement system for prefabricated high-density fiber optic connectors, comprising: The baseline acquisition module is used to clean the end face of the connector under test and acquire end face images, core-by-core IL and OTDR baselines during the factory manufacturing process, and generate and store the factory baseline certificate. The wet-state feature acquisition module is used to conduct wet-state accelerated aging tests before leaving the factory and acquire corresponding wet-state insertion loss, OTDR, end face images and sensor time series as wet-state features; The physical coupling module is used to calculate the theoretical coupling loss baseline by taking fiber parameters as input. ; The underwater optical cable insertion loss prediction model construction module is used to predict the wet-state additional loss residual using the aforementioned wet-state features and image features as input. and uncertainty ; Calculate the final connector loss value ; The lifetime distribution model building module is used to build models based on historical losses. The sequence and the wet intrusion event information are used to calculate the remaining service life distribution (RUL) of each core using degradation curves and survival analysis methods, and output the maintenance priority; at the same time, a wet intrusion judgment strategy is implemented to determine whether a wet intrusion event has occurred; if a wet intrusion event is detected, the wet accelerated degradation sub-model is triggered and the process returns to the underwater optical cable insertion loss prediction model construction module to re-estimate the degradation rate of each core.

[0037] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0038] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0039] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0040] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0041] The embodiments of the present invention have been described above with reference to the accompanying drawings. However, the present invention is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms under the guidance of the present invention without departing from the spirit and scope of the claims. All of these forms are within the protection scope of the present invention.

Claims

1. A method for measuring the insertion loss of a prefabricated high-density fiber optic connector, characterized in that: Includes the following steps: S1. During the factory manufacturing process, the connector under test is cleaned at the end face and the end face image is collected. The IL and OTDR baselines are obtained core by core, and the factory baseline certificate is generated and stored. S2. Before leaving the factory, conduct wet accelerated aging tests and collect corresponding wet insertion loss, OTDR, end face images and sensor time series as wet characteristics; S3. Using a physical coupling model, the fiber parameters are used as input to calculate the theoretical coupling loss baseline. ; S4. Construct an underwater optical cable insertion loss prediction model, using the aforementioned wet-state features and image features as input, to predict the wet-state additional loss residual. and uncertainty ; Calculate the final connector loss value ; S5. Establish a lifetime distribution model based on historical losses. The sequence and the aforementioned wet intrusion event information are used to calculate the remaining useful life distribution (RUL) of each core using degradation curves and survival analysis methods, and output the maintenance priority; at the same time, a wet intrusion judgment strategy is implemented to determine whether a wet intrusion event has occurred; If a wet intrusion event is detected, the wet accelerated degradation sub-model is triggered and the S4 is returned to re-estimate the degradation rate per core.

2. The method for measuring the insertion loss of a prefabricated high-density fiber optic connector according to claim 1, characterized in that: Step S2, which involves wet accelerated aging testing and wet characteristic acquisition, includes: S21. Perform N sets of wet accelerated aging tests on the same batch of sample cables according to the preset test matrix. The test matrix includes at least a combination of pressure, humidity immersion mode, temperature and salt spray intensity. Each set of tests is run on several repeated samples to obtain a statistically repeatable degradation sequence. S22. In each wet accelerated aging test group, continuously collect and store: sensor records that change over time, event-triggered core-by-core insertion loss IL(t) sequence, OTDR waveform snapshots and their event list, end-face image sequence, and test metadata; S23. Obtain the wet characteristic set from the wet accelerated aging test data through the following processing: calculate the humidity peak, the time of first threshold breach, the cumulative wet exposure duration, the pressure sensitivity coefficient, the proportion of wet spot area on the end face, the deterioration of end face structural similarity, the amplitude and width changes of OTDR events before and after wet conditions, and the wet IL increment.

3. The method for measuring the insertion loss of a prefabricated high-density fiber optic connector according to claim 2, characterized in that: The establishment and calculation of the physical coupling model include: S31, using the mode field diameter at both ends of the optical fiber , As the core quantity, the Gaussian mode overlap approximation is used to calculate the fundamental coupling loss term caused by mode field mismatch. ; S32. Calculate the additional loss term caused by centering based on centering error. The eccentricity square ratio and Gaussian radius are used as approximations to correct the eccentricity and reflect the weakening effect of eccentricity on coupling efficiency. S33. Consider the bending radius and use an empirical index to calculate the bending loss term; S34, Combine the basic coupling loss term and the additional loss term. After normalizing the bending loss term, a weighted sum is performed to obtain the theoretical coupling loss baseline. .

4. The method for measuring the insertion loss of a prefabricated high-density fiber optic connector according to claim 3, characterized in that: The underwater optical cable insertion loss prediction model adopts a hybrid structure of physical prior and residual learning, with theoretical coupling loss baseline. Using the baseline input, a neural network residual module is constructed to predict the residual of wet-state additional loss. and uncertainty log variance Final output connector loss value IL estimation ; The neural network includes an end-face image branch, an OTDR waveform branch, a time-series sensor branch, and a metadata branch; this means that self-supervised pre-training is performed on unlabeled end-face images and OTDR data, and then fine-tuned on data containing wet accelerated aging tests and factory baseline labels; when the model predicts a residual variance greater than a set residual variance threshold on wet samples, the weight of wet samples in the model is proportional to the variance.

5. The method for measuring the insertion loss of a prefabricated high-density fiber optic connector according to claim 4, characterized in that: The process of establishing the lifetime distribution model includes processing each core based on the time series of predicted connector loss values. Construct degradation curves and fit parameters ; Let represent the factory baseline, a and b represent natural degradation terms, and c and d represent wet-accelerated degradation terms. The parameters are obtained using Bayesian methods, and their confidence intervals are provided. This represents a wet intrusion event indicator function; simultaneously, a survival analysis model is trained in parallel, and a wet intrusion judgment strategy is used to determine whether a wet intrusion event has occurred. When the result is negative... The value is 0. When the result is yes, after confirming the wet accelerated degradation terms a and b through the wet accelerated degradation sub-model, the remaining service life distribution of each core is generated by combining the degradation curve and the output of the survival model.

6. The method for measuring the insertion loss of a prefabricated high-density fiber optic connector according to claim 5, characterized in that: The survival analysis model is constructed based on the Cox proportional hazards model, which is expressed as follows: , Indicates instantaneous failure rate. Indicates baseline hazard. This indicates the relative scaling of the covariates with respect to the baseline hazard. Indicates the most recent connector loss value The concatenation vector of wet features is used to obtain the survival function. , which represents the probability that the device has not failed by time t.

7. The method for measuring the insertion loss of a prefabricated high-density fiber optic connector according to claim 6, characterized in that: The wet intrusion detection strategy includes multiple layers of judgment logic: Threshold determination layer, when the humidity sensor recorded a value that meets the time requirement... When the humidity at any time t within the time period is greater than or equal to a set humidity threshold, it is judged as initial wet intrusion. The linked evidence layer outputs the wet intrusion probability through a weakly supervised merger while the threshold judgment layer remains valid. ,when Greater than or equal to the set threshold At that time, it was determined to be a wet intrusion event.

8. The method for measuring the insertion loss of a prefabricated high-density fiber optic connector according to claim 7, characterized in that: The wet-accelerated degradation sub-model uses the wet dynamics fitted in the wet-accelerated aging test as a priori, and adopts a parameterized form. The initial values ​​of wet parameters c and d are obtained from the fitting results of the wet-accelerated aging test, and are mapped to the equivalent values ​​in the field through hierarchical Bayesian wet-accelerated aging test field mapping and EAF transformation. The mapping process outputs the posterior distribution of c and d to reflect the uncertainty. When a wet intrusion event occurs, the sub-model receives high-frequency IL and OTDR sensor data uploaded from the edge and performs online Bayesian posterior updates to c and d according to an incremental window to quickly converge to the wet parameter estimate of the specific sample and update the RUL. The output of the wet-accelerated degradation sub-model includes wet amplitude estimation, rate estimation, parameter confidence interval, IL time series prediction for the wet stage, and updated RUL distribution. When the wet amplitude estimate or the updated RUL reaches the maintenance threshold, a maintenance instruction is automatically generated and the active sampling priority is updated.

9. An insertion loss measurement system for prefabricated high-density fiber optic connectors, used to perform insertion loss measurement of prefabricated high-density fiber optic connectors as described in any one of claims 1-8, characterized in that: include: The baseline acquisition module is used to clean the end face of the connector under test and acquire end face images, core-by-core IL and OTDR baselines during the factory manufacturing process, and generate and store the factory baseline certificate. The wet-state feature acquisition module is used to conduct wet-state accelerated aging tests before leaving the factory and acquire corresponding wet-state insertion loss, OTDR, end face images and sensor time series as wet-state features; The physical coupling module is used to calculate the theoretical coupling loss baseline by taking fiber parameters as input. ; The underwater optical cable insertion loss prediction model construction module is used to predict the wet-state additional loss residual using the aforementioned wet-state features and image features as input. and uncertainty ; Calculate the final connector loss value ; The lifetime distribution model building module is used to build models based on historical losses. The sequence and the aforementioned wet intrusion event information are used to calculate the remaining useful life distribution (RUL) of each core using degradation curves and survival analysis methods, and output the maintenance priority; at the same time, a wet intrusion judgment strategy is implemented to determine whether a wet intrusion event has occurred; If a wet intrusion event is detected, the wet accelerated degradation sub-model is triggered and the underwater optical cable insertion loss prediction model building module is returned to re-estimate the degradation rate per core.

Images