Method for manufacturing optical fiber preform and method for predicting characteristics

Abstract

A method for manufacturing and predicting the characteristics of optical fiber preforms is disclosed, which accurately predicts the characteristics of optical fibers based on the measurement results of the preforms. The method for manufacturing optical fiber preforms includes: a manufacturing step of manufacturing optical fiber preforms; a measurement step of measuring the refractive index distribution of the preforms; and a judgment step of determining whether the preforms are good or bad based on the characteristic prediction results obtained by inputting the measurement results from the measurement step into a pre-created characteristic prediction model. The characteristic prediction model is a model obtained through machine learning using training data. The training data uses the refractive index distribution data of previously manufactured optical fiber preforms as explanatory variables and the measurement results of the characteristics of optical fibers obtained by drawing the preforms as target variables.

Description

Technical Field

[0001] This disclosure relates to a method for manufacturing and predicting the properties of a base material for optical fibers. Background Technology

[0002] Patent document 1 discloses a method for predicting the properties of an optical fiber obtained from a matrix material based on the refractive index of the matrix material.

[0003] Existing technical documents

[0004] Patent documents

[0005] Patent Document 1: Japanese Patent Application Publication No. 2003-12337 Summary of the Invention

[0006] The technical problem that the invention aims to solve

[0007] Optical fibers have established standards for various characteristics, such as cutoff wavelength; fibers whose characteristics do not meet these standards cannot be shipped. Although these characteristics vary depending on the manufacturing of the preform and drawing conditions, the process from upstream preform manufacturing to obtaining the optical fiber takes several weeks. Therefore, if manufacturing conditions are to be changed based on measured values ​​of the fiber's characteristics, it takes time for the changes to take effect. For this reason, it is desirable to predict the fiber's characteristics at the preform stage and adjust the manufacturing conditions based on the predictions. However, current methods struggle to predict fiber characteristics with high accuracy at the preform stage.

[0008] The purpose of this disclosure is to provide a method for manufacturing optical fiber preforms and a method for predicting their properties that can accurately predict the properties of optical fibers based on the measurement results of the preforms.

[0009] Solutions for solving technical problems

[0010] The method for manufacturing optical fiber preform disclosed herein includes: a manufacturing step of manufacturing optical fiber preform; a measurement step of measuring the refractive index distribution of the optical fiber preform; and a judgment step of judging whether the optical fiber preform is good or bad based on the characteristic prediction result obtained by inputting the measurement result in the measurement step into a pre-created characteristic prediction model. The characteristic prediction model is a model obtained by machine learning using training data. The training data uses the refractive index distribution data of previously manufactured optical fiber preforms as explanatory variables and the measurement results of the characteristics of the optical fiber obtained by drawing the optical fiber preform as target variables.

[0011] Invention Effects

[0012] According to this disclosure, the characteristics of optical fibers can be predicted with high accuracy based on the measurement results of the matrix material used for optical fibers. Attached Figure Description

[0013] Figure 1 This is a block diagram illustrating an example of a manufacturing system for a method of manufacturing a matrix material for optical fibers according to an embodiment of the present disclosure.

[0014] Figure 2 This is a flowchart illustrating a method for manufacturing a preform for optical fibers according to an embodiment of the present disclosure.

[0015] Figure 3 This is a diagram illustrating an example of the refractive index distribution of a concave structure.

[0016] Figure 4 This is a diagram showing an example of the refractive index distribution of a trench structure. Detailed Implementation

[0017] [Description of embodiments of this disclosure]

[0018] First, the implementation aspects of this disclosure will be listed for explanation. (1) The method for manufacturing optical fiber preform of this disclosure includes: a manufacturing step for manufacturing optical fiber preform; a measurement step for measuring the refractive index distribution of the optical fiber preform; and a judgment step for judging whether the optical fiber preform is good or bad based on the characteristic prediction result obtained by inputting the measurement result in the measurement step into a pre-created characteristic prediction model, wherein the characteristic prediction model is a model obtained by machine learning using training data, wherein the training data uses the refractive index distribution data of previously manufactured optical fiber preforms as explanatory variables and the measurement result of the characteristics of the optical fiber obtained by drawing the optical fiber preform as the target variable.

[0019] Previously, methods using the finite element method to predict the refractive index distribution of optical fiber preforms were known, but the prediction accuracy was low. To improve prediction accuracy, periodic calibration was required, which became a burden on operators. According to this embodiment, by using a machine learning model created based on data from previously manufactured optical fiber preforms, the characteristics of the optical fiber preforms can be predicted with high accuracy. Therefore, the accuracy of determining whether a material is good or bad is improved, thereby increasing manufacturing efficiency.

[0020] (2) In (1) above, the characteristic prediction result may also include data related to at least one characteristic selected from the group consisting of cutoff wavelength, mode field diameter, and zero-dispersion wavelength. By using the method of this embodiment, the cutoff wavelength, mode field diameter, or zero-dispersion wavelength can be predicted with high accuracy based on the refractive index distribution of the optical fiber substrate.

[0021] (3) In (2) above, the characteristic prediction result may also include the cutoff wavelength, and the characteristic prediction model may also include drawing tension as the explanatory variable, and the cutoff wavelength corrected based on drawing tension as the target variable. Since the cutoff wavelength varies according to drawing tension, the corrected cutoff wavelength can be predicted by correcting the measured value based on drawing tension and including drawing tension as the explanatory variable.

[0022] (4) In (3) above, the manufacturing conditions in the drawing process can also be adjusted based on the predicted characteristics of the optical fiber preform. By changing the drawing tension according to the predicted cutoff wavelength value, the cutoff wavelength of the manufactured optical fiber can be adjusted to a more appropriate range. Through such adjustments, the range of optical fiber preforms that can be used in the product can be expanded, and manufacturing efficiency can be improved.

[0023] (5) In any of (1) to (4) above, the characteristic prediction model may also include a predicted value as the explanatory variable, the predicted value being calculated using the finite element method based on at least one of the target variables and the refractive index distribution. By incorporating the predicted value based on the finite element method as the explanatory variable, the prediction accuracy is improved.

[0024] (6) In any of (1) to (5) above, the refractive index distribution may also be a recessed structure or a grooved structure. Previously, it was difficult to predict characteristics based on complex refractive index distributions such as recessed structures or grooved structures. According to this embodiment, by using a characteristic prediction model obtained through machine learning, even optical fiber preforms with complex refractive index distributions can have their characteristics predicted with high accuracy.

[0025] (7) In (6) above, the characteristic prediction model may also include the difference ΔN between the refractive index of the fiber core and the refractive index of the inner cladding in the refractive index distribution, and the amount of depression as the difference between the refractive index of the inner cladding and the refractive index of the outer cladding as explanatory variables. By adding ΔN and the amount of depression as explanatory variables, the prediction accuracy is improved.

[0026] (8) The optical fiber preform characteristic prediction method of this disclosure includes: a measurement step of measuring the refractive index distribution of the optical fiber preform; and a prediction step of inputting the measurement results from the measurement step into a pre-created characteristic prediction model to predict the characteristics of the optical fiber preform. The characteristic prediction model is a model obtained through machine learning using training data. The training data uses refractive index distribution data of previously manufactured optical fiber preforms as explanatory variables and the measurement results of the characteristics of the optical fiber obtained by drawing the optical fiber preform as the target variable. According to this embodiment, by using a machine learning model created based on data from previously manufactured optical fiber preforms, the characteristics of the optical fiber preform can be predicted with high accuracy.

[0027] [Details of the embodiments of this disclosure]

[0028] The following describes specific examples of the manufacturing method and characteristic prediction method for the optical fiber preform of this disclosure with reference to the accompanying drawings. It should be noted that the present invention is not limited to these examples, but is intended to include all modifications within the meaning and scope of the claims, as illustrated by the claims. At least some of the embodiments described below can also be arbitrarily combined.

[0029] Figure 1 This is a block diagram illustrating an example of a manufacturing system used to implement the method for manufacturing the optical fiber preform of this disclosure. Figure 1 As shown, the manufacturing system 100 includes a base material manufacturing device 1, a refractive index measuring device 2, and a characteristic prediction device 3. In Figure 1 The document also shows an optical fiber manufacturing apparatus 4 for drawing an optical fiber substrate manufactured by a substrate manufacturing apparatus 1 to produce an optical fiber, and a characteristic measuring apparatus 5 for measuring the characteristics of the optical fiber manufactured by the optical fiber manufacturing apparatus 4.

[0030] The preform manufacturing apparatus 1 is an apparatus for manufacturing preforms for optical fibers. In this disclosure, the manufacturing method for the preforms for optical fibers is not particularly limited; for example, it may employ vapor phase synthesis methods such as VAD, OVD, or MCVD. The preform manufacturing apparatus 1 is not limited to a single apparatus and may include multiple apparatuses. For example, the preform manufacturing apparatus 1 may also include an apparatus for manufacturing glass microparticle packs using the vapor phase synthesis method illustrated above, and a sintering furnace for sintering the glass microparticle packs to make them transparent. In the preform manufacturing apparatus 1, various manufacturing conditions are set for manufacturing preforms for optical fibers with desired properties. The manufacturing conditions of the preform manufacturing apparatus 1 can also be adjusted by a control device (not shown). The parameters specifically adjusted as manufacturing conditions may vary depending on the method used, but for example, the amount and doping ratio of raw materials, heating temperature, and heating time can be adjusted. The preforms for optical fibers manufactured by the preform manufacturing apparatus 1 have a concentric circular structure including a high-refractive-index core and a low-refractive-index cladding covering the core.

[0031] The refractive index measuring device 2 measures the refractive index of the optical fiber preform manufactured by the preform manufacturing apparatus 1. As a method for measuring the refractive index, known methods can be used, for example, the refractive index distribution along the diameter can be obtained by analyzing the transmitted light when a laser is irradiated perpendicular to the axis of the preform. The refractive index measuring device 2 may also be included in the preform manufacturing apparatus 1. For an optical fiber preform, the refractive index can be measured at multiple locations along the length direction. When the refractive index is measured at multiple locations, all refractive index distribution data, a portion of the refractive index distribution data, or the average of multiple refractive index distributions can be used in the characteristic prediction described below. The preferred location for measuring the refractive index is the central portion along the length direction of the preform; specifically, it is preferably the portion 50 mm or more inside the end of the region of the optical fiber preform from which it is drawn into the product.

[0032] The characteristic prediction device 3 predicts the characteristics of the optical fiber obtained by drawing the optical fiber preform from the refractive index distribution data measured by the refractive index measuring device 2. The characteristic prediction device 3 includes a storage unit 31, a data acquisition unit 32, a characteristic prediction unit 33, a determination unit 34, and an output unit 35. The characteristic prediction device 3 can also be connected to other devices via cables or networks (not shown).

[0033] The hardware configuration of the feature prediction device 3 is not particularly limited. Each process (function) in the embodiments described below is implemented by a processing circuit including one or more processors. This processing circuit may also be composed of an integrated circuit that combines one or more memories, various analog circuits, and various digital circuits in addition to the one or more processors. The one or more memories store programs (commands) that cause the one or more processors to execute the aforementioned processes. The one or more processors may execute the aforementioned processes according to the programs read from the one or more memories, or they may execute the aforementioned processes according to logic circuits pre-designed to execute the aforementioned processes. The processors may be various processors suitable for computer control, such as CPUs (Central Processing Units), GPUs (Graphics Processing Units), DSPs (Digital Signal Processors), FPGAs (Field Programmable Gate Arrays), and ASICs (Application Specific Integrated Circuits). It should be noted that physically separate processors may also cooperate to execute the aforementioned processes. For example, the processors of multiple physically separate computers can also cooperate with each other via networks such as LAN (Local Area Network), WAN (Wide Area Network), and the Internet to execute the aforementioned processes. The programs can also be installed in the memory from external server devices via the aforementioned network, or they can be circulated in a state stored on recording media such as CD-ROM (Compact Disc Read Only Memory), DVD-ROM (Digital Versatile Disk Read Only Memory), and semiconductor memory, and installed from the aforementioned recording media into the memory.

[0034] The storage unit 31 stores a program for performing the processing of predicting the characteristics of the optical fiber, and a model for predicting the characteristics (hereinafter referred to as the characteristic prediction model). The determination process S3 for judging whether the optical fiber preform is good or bad, which will be described below, is performed, for example, by the CPU executing the program stored in the storage unit 31. The characteristic prediction model is a model obtained by machine learning using training data, which uses the refractive index distribution data of previously manufactured optical fiber preforms as explanatory variables and the measurement results of the characteristics of the optical fiber obtained by drawing the optical fiber preform as the target variable. By inputting the refractive index distribution data of the optical fiber preform into the characteristic prediction model, the predicted results of the characteristics of the optical fiber obtained by drawing the optical fiber preform are output.

[0035] The fiber characteristics used as the target variable in the characteristic prediction model are not particularly limited, but can be, for example, at least one characteristic selected from the cutoff wavelength, mode field diameter, and zero-dispersion wavelength. The cutoff wavelength is the shortest wavelength that propagates in single mode within the fiber. The mode field diameter represents the radial spread of light propagating within the fiber. The zero-dispersion wavelength is the wavelength at which wavelength dispersion is zero.

[0036] The machine learning algorithms used to generate feature prediction models are not particularly limited as long as they are classified as teacher-learning algorithms. They can also be classification or regression methods such as decision trees, logistic regression analysis, and multiple regression analysis, or ensemble learning that combines multiple learners. Specific examples of algorithms include LightGBM (Light Gradient Boosting Machine), Ridge, XGBoost (Extreme Gradient Boosting), and RandomForest.

[0037] The number of datasets related to past optical fiber parent materials and optical fibers used in generating the characteristic prediction model can be either over 5,000 or over 20,000. If the number of datasets is over 5,000, the prediction accuracy will be improved.

[0038] The data acquisition unit 32 acquires data input from an external source to the characteristic prediction device 3. The characteristic prediction unit 33 predicts the characteristics of the optical fiber based on the data acquired by the data acquisition unit 32 and the characteristic prediction model. The determination unit 34 determines whether the optical fiber preform is good or bad based on the prediction result of the characteristic prediction unit 33. The output unit 35 outputs the determination result of the determination unit 34. The functions of the data acquisition unit 32, the characteristic prediction unit 33, the determination unit 34, and the output unit 35 will be described in detail in the following description of the manufacturing method of the optical fiber preform.

[0039] Next, a method for manufacturing a preform for optical fibers according to one embodiment of this disclosure will be described. Figure 2 This is a flowchart of the manufacturing method of the optical fiber preform according to this embodiment. Figure 2 As shown, the method for manufacturing optical fiber preforms according to this embodiment includes a manufacturing step S1, a measurement step S2, and a judgment step S3. Figure 2 The diagram also shows that output process S4 is the process following decision process S3.

[0040] First, in manufacturing step S1, a pre-material for optical fiber is manufactured using a pre-material manufacturing apparatus 1. Next, in measurement step S2, the refractive index of the pre-material for optical fiber manufactured by the pre-material manufacturing apparatus 1 is measured using a refractive index measuring apparatus 2, and refractive index distribution data is obtained. The refractive index distribution data obtained here refers to numerical data of the refractive index at multiple (e.g., 1200 points) locations along, for example, the diameter of the pre-material for optical fiber.

[0041] Next, in the judgment step S3, the measurement results from the measurement step S2 are input into the pre-created characteristic prediction model to predict the characteristics of the optical fiber obtained by drawing the optical fiber base material, and the quality of the optical fiber base material is determined based on the characteristic prediction results. Specifically, the refractive index distribution data obtained by the refractive index measuring device 2 in the measurement process S2 is first input into the data acquisition unit 32 of the characteristic prediction device 3. Figure 1 (arrow R1). Next, the characteristic prediction unit 33 inputs the refractive index distribution data input to the data acquisition unit 32 into the characteristic prediction model stored in the storage unit 31, and obtains the prediction result of the characteristics of the optical fiber obtained from the optical fiber preform. The determination unit 34 determines whether the optical fiber preform is good or bad based on the characteristic prediction result obtained by the characteristic prediction unit 33.

[0042] The method for determining whether the optical fiber preform in the determination unit 34 is good or bad is not particularly limited. For example, it can also be determined based on whether the predicted values ​​of each characteristic calculated by the characteristic prediction model are within the specified allowable range.

[0043] In this embodiment, in the output process S4 following the determination process S3, the output unit 35 outputs the determination result of the determination unit 34. For example, the output unit 35 may display the determination result on an output device such as a display and notify the operator of the determination result. Alternatively, the output unit 35 may send the determination result or an instruction based on the determination result to the optical fiber manufacturing apparatus 4, and change the manufacturing conditions of the optical fiber manufacturing apparatus 4 based on the sent determination result or instruction. Figure 1 Arrow R2). Alternatively, the output unit 35 can be configured to send a determination result or an instruction based on the determination result to the base material manufacturing apparatus 1, and to change the manufacturing conditions of the base material manufacturing apparatus 1 based on the sent determination result or instruction. Figure 1 Arrow R3).

[0044] Previously, methods such as the finite element method (FEM) were used to predict the properties of optical fibers obtained by drawing the fiber-forming material from the base material, based on characteristics such as the refractive index of the base material. However, it is difficult to predict the properties of optical fibers with high accuracy at the base material stage using the finite element method.

[0045] According to this embodiment, the refractive index distribution data of the optical fiber substrate is input into a characteristic prediction model obtained through machine learning to predict the characteristics of the optical fiber and determine whether they are good or bad. By adopting this configuration, the characteristics of the optical fiber can be predicted with high accuracy at the substrate stage.

[0046] Alternatively, if the characteristic prediction results include the cutoff wavelength—in other words, if the target variable includes the cutoff wavelength—the characteristic prediction model also includes fiber drawing tension as an explanatory variable, and a cutoff wavelength corrected based on fiber drawing tension as the target variable. Since the cutoff wavelength of the optical fiber varies with the fiber drawing tension, the corrected cutoff wavelength can be predicted by correcting the measured value based on the fiber drawing tension and including fiber drawing tension as an explanatory variable. It should be noted that when fiber drawing tension is included as an explanatory variable, since fiber drawing has not yet occurred at the time of characteristic prediction, an assumed fiber drawing tension is input to predict each characteristic.

[0047] Alternatively, if the characteristic prediction results include the cutoff wavelength, the manufacturing conditions in the fiber drawing process can be further adjusted based on the characteristic prediction results of the fiber preform. By changing the drawing tension according to the predicted cutoff wavelength value, the cutoff wavelength of the manufactured fiber can be adjusted to a more appropriate range. For example, if the drawing tension is increased, the cutoff wavelength of the manufactured fiber becomes longer. Therefore, even if the predicted cutoff wavelength of the fiber preform is less than the target value, by increasing the tension during drawing of the fiber preform, the cutoff wavelength of the manufactured fiber can be made greater than the predicted value, resulting in an fiber with the desired characteristics. According to this configuration, even if the predicted cutoff wavelength does not meet the reference, it can be adjusted to meet the reference by changing manufacturing conditions such as drawing tension, thus expanding the range of fiber preforms that can be used in products and improving manufacturing efficiency.

[0048] The "adjustment of manufacturing conditions in the drawing process" mentioned above does not necessarily mean that drawing must be actually performed according to those manufacturing conditions. That is, adjusting the manufacturing conditions here means calculating appropriate manufacturing conditions for the drawing process based on characteristic prediction results and mapping the calculated manufacturing conditions to the optical fiber preform. Therefore, aspects such as marking the calculated manufacturing conditions on the optical fiber preform when it is shipped in preform form are also included in the method described in this embodiment.

[0049] Alternatively, the characteristic prediction model in this embodiment may also include predicted values ​​as explanatory variables. These predicted values ​​are calculated using the finite element method based on the refractive index distribution, taking into account at least one of the target variables. For example, when the target variables are the cutoff wavelength, mode field diameter, and zero-dispersion wavelength of the optical fiber, a predicted value for at least one selected from the cutoff wavelength, mode field diameter, and zero-dispersion wavelength is calculated using the finite element method based on the refractive index distribution, and a model is created that includes these predicted values ​​as explanatory variables. By incorporating predicted values ​​based on the finite element method as explanatory variables, the prediction accuracy is improved.

[0050] In this embodiment, the refractive index distribution of the optical fiber substrate may also be a recessed structure or a grooved structure. Figure 3 and Figure 4 This is a graph showing an example of refractive index distribution, with the horizontal axis representing the position along the diameter and the vertical axis representing the refractive index. Figure 3 The refractive index distribution of the optical fiber substrate with a recessed structure is shown. For example... Figure 3 As shown, in the recessed structure, the refractive index of the inner cladding A2, which is located in the cladding region adjacent to the fiber core A1, is lower than that of the outer cladding A3. Figure 4 The refractive index distribution of the matrix material for optical fibers with a trench structure is shown. For example... Figure 4 As shown, in the trench structure, in the cladding region, a low refractive index portion (groove) A4 is provided between the inner cladding A2 and the outer cladding A3.

[0051] Previously, it was difficult to predict the characteristics of optical fibers based on complex refractive index distributions such as concave or grooved structures. According to this embodiment, by using a characteristic prediction model obtained through machine learning, the characteristics of optical fibers can be predicted with high accuracy even for optical fiber base materials with complex refractive index distributions.

[0052] like Figure 3 As shown, in the recessed structure, the difference between the refractive index of the core A1 and the refractive index of the inner cladding A2 is called ΔN, and the difference between the refractive index of the inner cladding A2 and the refractive index of the outer cladding A3 is called the recess amount D. Figure 4 As shown, in the trench structure, the difference between the refractive index of the core A1 and the refractive index of the inner cladding A2 is called ΔN, and the difference between the refractive index of the inner cladding A2 and the refractive index of the outer cladding A3 is called the depression amount D. In this embodiment, the characteristic prediction model may also include the difference between the refractive index of the core and the refractive index of the inner cladding, ΔN, and the depression amount D, which is the difference between the refractive indices of the inner and outer claddings, as explanatory variables. By adding ΔN and the depression amount D as explanatory variables, the prediction accuracy is improved. ΔN and the depression amount D can be calculated based on the refractive index distribution data through processing by a processor, etc., or they can be used when the refractive index distribution is approximated as... Figure 3 and Figure 4 The values ​​are calculated based on the stepped distribution shown.

[0053] In this disclosure, the optical fiber can also be manufactured by drawing the optical fiber preform manufactured by the preform manufacturing apparatus 1 into fibers using the optical fiber manufacturing apparatus 4. After the characteristics of the optical fiber are measured by the characteristic measuring apparatus 5, the measurement results are input to the data acquisition unit 32 of the characteristic prediction apparatus 3. Figure 1 (Arrow R4). By combining the characteristic measurement results input from the characteristic measurement device 5 and the refractive index distribution data input from the refractive index measurement device 2, new training data can be used. Alternatively, the characteristic prediction device 3 can add data including the refractive index distribution data of the optical fiber preform obtained in this way and the characteristic measurement results of the optical fiber manufactured from the optical fiber preform to the dataset to create a new characteristic prediction model or to fine-tune an existing characteristic prediction model. By using the dataset of sequentially manufactured optical fibers to create or adjust the characteristic prediction model, the prediction accuracy of the characteristic prediction model can be improved.

[0054] Furthermore, this disclosure also relates to a method for predicting the characteristics of optical fiber preforms. The characteristic prediction method includes: a measurement step of measuring the refractive index distribution of the optical fiber preform; and a prediction step of inputting the measurement results from the measurement step into a pre-created characteristic prediction model to predict the characteristics of the optical fiber preform. The characteristic prediction model is a model obtained through machine learning using training data. The training data uses refractive index distribution data of previously manufactured optical fiber preforms as explanatory variables and the measurement results of the characteristics of optical fibers obtained by drawing the optical fiber preform as target variables. According to the method of this embodiment, the characteristics of optical fiber preforms can be predicted with high accuracy, and the prediction results can be used for quality assurance, verification, etc., of optical fiber preforms.

[0055] Example

[0056] Based on a dataset of approximately 20,000 data points, including refractive index distribution data of previously manufactured optical fiber preforms and the properties of optical fibers obtained by drawing these preforms, a characteristic prediction model was created using machine learning algorithms. LightGBM, Ridge, XGBoost, and RandomForest were used as the machine learning algorithms. The explanatory variables of the characteristic prediction model were the refractive index distribution data of the optical fiber preforms (one-dimensional refractive index data along approximately 1200 points along the diameter), and the target variables were the cutoff wavelength, mode field diameter (MFD), and zero-dispersion wavelength. Using the obtained characteristic prediction model, predictions were made for each characteristic as the target variable based on the refractive index distribution data of 2000 evaluation datasets, and the root mean square error (RMSE) was calculated based on the predicted and measured values. The results are shown in Table 1.

[0057] For comparison, the finite element method (FEM) was used to predict characteristics for the same evaluation dataset, and the RMSE of each objective variable was calculated. The results are shown in Table 1. Table 1 shows the relative values ​​of the RMSE of each characteristic prediction model when the RMSE in the finite element method case is set to 1.

[0058] [Table 1]

[0059] As shown in Table 1, in any case where a feature prediction model generated by machine learning was used, the RMSE was smaller than that of the finite element method, and the prediction accuracy was improved.

[0060] Explanation of reference numerals in the attached figures

[0061] 100: Manufacturing system; 1: Base material manufacturing device; 2: Refractive index measuring device; 3: Characteristic prediction device; 31: Storage unit; 32: Data acquisition unit; 33: Characteristic prediction unit; 34: Judgment unit; 35: Output unit; 4: Optical fiber manufacturing device; 5: Characteristic measuring device; A1: Fiber core; A2: Inner cladding; A3: Outer cladding; A4: Low refractive index portion (groove).

Claims

1. A method for manufacturing a matrix material for optical fibers, comprising: Manufacturing process, manufacturing the base material for optical fibers; The measurement process includes measuring the refractive index distribution of the parent material used in the optical fiber; and The judgment process, based on the characteristic prediction results obtained by inputting the measurement results from the measurement process into a pre-created characteristic prediction model, determines whether the optical fiber base material is of good or bad quality. The characteristic prediction model is a model obtained by machine learning using training data. The training data uses the refractive index distribution data of the optical fiber base material manufactured in the past as the explanatory variable and the measurement results of the characteristics of the optical fiber obtained by drawing the optical fiber base material as the target variable.

2. The method for manufacturing the optical fiber preform according to claim 1, wherein, The characteristic prediction results include data related to at least one characteristic selected from the group consisting of cutoff wavelength, mode field diameter, and zero-dispersion wavelength.

3. The method for manufacturing the optical fiber preform according to claim 2, wherein, The characteristic prediction results include the cutoff wavelength. The characteristic prediction model also includes drawing tension as an explanatory variable. The characteristic prediction model also includes the cutoff wavelength, which is corrected based on the drawing tension, as the target variable.

4. The method for manufacturing the optical fiber preform according to claim 3, wherein, The manufacturing conditions in the fiber drawing process are adjusted based on the predicted characteristics of the optical fiber preform.

5. The method for manufacturing the optical fiber preform according to any one of claims 1 to 4, wherein, The characteristic prediction model also includes predicted values ​​as explanatory variables, which are calculated using the finite element method based on the refractive index distribution with respect to at least one of the target variables.

6. The method for manufacturing the optical fiber preform according to any one of claims 1 to 4, wherein, The refractive index distribution is a concave structure or a groove structure.

7. The method for manufacturing the optical fiber preform according to claim 6, wherein, The characteristic prediction model also includes the difference ΔN between the refractive index of the fiber core and the refractive index of the inner cladding in the refractive index distribution, and the amount of depression, which is the difference between the refractive index of the inner cladding and the refractive index of the outer cladding, as explanatory variables.

8. A method for predicting the characteristics of a matrix material for optical fibers, comprising: The measurement process involves determining the refractive index distribution of the matrix material used in optical fibers; and The prediction process involves inputting the measurement results from the measurement process into a pre-created characteristic prediction model to predict the characteristics of the optical fiber's base material. The characteristic prediction model is a model obtained by machine learning using training data. The training data uses the refractive index distribution data of the optical fiber base material manufactured in the past as the explanatory variable and the measurement results of the characteristics of the optical fiber obtained by drawing the optical fiber base material as the target variable.

Images