A method for intelligent adjustment of UV curing parameters in optical fiber coating equipment

Abstract

This invention relates to the field of data processing technology, and in particular to an intelligent adjustment method for UV curing parameters in optical fiber coating equipment. The method acquires an acceleration data sequence of the optical fiber in the current time period. Based on the difference between each data point in the acceleration data sequence and the data from a previous time point, the data in the acceleration data sequence are clustered to obtain the target cluster containing the tension acceleration at the current time. Based on the distance between any two data points in the target cluster, the number of data points in the target cluster, and the distance between the target cluster and each other cluster, a parameter adjustment coefficient is obtained. Based on the parameter adjustment coefficient, a preset initial length scale in the Gaussian process regression model is adaptively adjusted to obtain an adaptive Gaussian process regression model, which in turn yields the predicted tension acceleration for future time points. The UV curing parameters are then adjusted based on the predicted tension acceleration, improving the accuracy of UV curing parameter adjustment.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, and in particular to a method for intelligent adjustment of ultraviolet curing parameters for optical fiber coating equipment. Background Technology

[0002] After optical fiber is drawn into its final shape, it needs to be uniformly coated with a photocurable resin to enhance its mechanical strength and prevent corrosion from moisture and chemical media. The coated resin then undergoes ultraviolet (UV) irradiation for curing, forming a dense, smooth, and strongly adhesive protective layer to ensure the long-term stability and transmission performance of the optical fiber. The UV curing process is a crucial step in determining the quality and performance of the optical fiber coating. Its effectiveness is easily affected by factors such as the drawing speed. Therefore, it is necessary to intelligently adjust the UV curing parameters according to the fiber's drawing speed to achieve adaptive optimization of the curing process, which is of great significance for improving the quality of optical fiber coating and production stability.

[0003] Traditionally, Gaussian process regression (GPR) models are used to predict the future pulling speed of optical fibers based on historical pulling speeds, and the prediction uncertainty is output. This allows for feedforward adaptive adjustment of UV curing parameters, proactively addressing changes in fiber pulling speed and reducing UV curing lag and over-adjustment. In the GPR model, the length scale directly controls the smoothness or complexity of the fitted function. Traditionally, the length scale is fixed. However, if the scale is too large, the model may fail to capture the true details in the data, leading to underfitting. Conversely, if the scale is too small, the model may interpret the offset of each data point as a true characteristic of the function, striving to capture every data point, including local fluctuations and subtle changes. This results in a function with many peaks and troughs, leading to poor generalization and inaccurate predictions of the fiber pulling speed in the near future.

[0004] Therefore, how to adaptively obtain the length scale in the GPR model to improve the accuracy of predicting the pulling speed of optical fiber in the short term, and thus improve the accuracy of adjusting the UV curing parameters, has become an urgent problem to be solved. Summary of the Invention

[0005] In view of this, embodiments of the present invention provide an intelligent adjustment method for UV curing parameters in optical fiber coating equipment, to solve the problem of how to adaptively obtain the length scale in the GPR model to improve the accuracy of predicting the pulling speed of optical fiber in a short period of time, and thus improve the accuracy of adjusting UV curing parameters.

[0006] This invention provides a method for intelligent adjustment of UV curing parameters in optical fiber coating equipment, the method comprising the following steps:

[0007] During the fiber drawing and coating process, the pulling acceleration of the fiber at the current moment and a preset number of historical moments before the current moment are obtained to obtain an acceleration data sequence. Based on the difference between each data point in the acceleration data sequence and the data at the corresponding moment before it, the data in the acceleration data sequence are clustered to obtain at least two clusters.

[0008] The cluster containing the current traction acceleration is denoted as the target cluster. The data distribution concentration of the target cluster is obtained based on the distance between every two data points in the target cluster. The consistency of data changes between the target cluster and other clusters is obtained based on the number of data points in the target cluster and the distance between the target cluster and each other cluster.

[0009] Based on the consistency between the concentration of the data distribution and the variation of the data, parameter adjustment coefficients are obtained, which are used to adjust the preset initial length scale in the Gaussian process regression model;

[0010] Based on the parameter adjustment coefficient, the preset initial length scale in the Gaussian process regression model is adaptively adjusted to obtain the adaptive Gaussian process regression model. The acceleration data sequence is input into the adaptive Gaussian process regression model to obtain the predicted tension acceleration of the optical fiber at future moments. Based on the predicted tension acceleration, the UV curing parameters are intelligently adjusted.

[0011] Preferably, the step of clustering the data in the acceleration data sequence based on the difference between each data point in the acceleration data sequence and the data before its corresponding time point to obtain at least two clusters includes:

[0012] For any data in the acceleration data sequence, the time corresponding to the data is obtained and recorded as the target time. The historical traction acceleration of the optical fiber is obtained at each time within a preset historical period before the target time, thus obtaining the historical data sequence of the data.

[0013] Calculate the absolute value of the difference between any data point and any historical traction acceleration in the historical data sequence to obtain the acceleration difference value. Calculate the time interval between any data point and any historical traction acceleration at the corresponding moment. Use the negative of the time interval as the independent variable of an exponential function with the natural constant as the base to obtain the degree of mutual reference between any data point and any historical traction acceleration.

[0014] Obtain the degree of cross-reference between any data and each historical traction acceleration in the historical data sequence, accumulate all the degrees of cross-reference to obtain the accumulated value, calculate the proportion of the degree of cross-reference between any data and any historical traction acceleration in the accumulated value, and obtain the weight of the acceleration difference value between any data and any historical traction acceleration.

[0015] The acceleration difference value and its weight between any data and each historical traction acceleration in the historical data sequence are obtained respectively. All acceleration difference values ​​are weighted and summed to obtain the weighted difference feature value of any data.

[0016] The weighted difference feature value of each data point in the acceleration data sequence is obtained separately. All weighted difference feature values ​​are mapped to a scatter plot. The horizontal axis of the scatter plot represents the time corresponding to each data point in the acceleration data sequence, and the vertical axis represents the weighted difference feature value of each data point in the acceleration data sequence. The data points in the scatter plot are clustered to obtain at least two clusters.

[0017] Preferably, the step of obtaining the data distribution concentration of the target cluster based on the distance between every two data points in the target cluster includes:

[0018] The data in the target cluster are sorted according to their corresponding time to obtain a time series. The time interval between each two adjacent data points in the time series is obtained. The average time interval and coefficient of variation of all time intervals are calculated. The coefficient of variation is used as the independent variable of an exponential function with the natural constant as the base to obtain the data distribution uniformity in the target cluster.

[0019] The sampling interval of the data in the acceleration data sequence is obtained and denoted as the standard time interval. The ratio between the average time interval and the standard time interval is calculated to obtain the data temporal distribution concentration of the target cluster. The product between the data temporal distribution concentration and the data distribution uniformity is calculated. The negative of the product is used as the independent variable of an exponential function with the natural constant as the base to obtain the data distribution concentration of the target cluster.

[0020] Preferably, the step of determining the degree of consistency of data changes between the target cluster and other clusters based on the amount of data in the target cluster and the distance between the target cluster and each other cluster includes:

[0021] The ratio between the number of all data in the target cluster and the number of all data in the acceleration data sequence is used to obtain the size characteristic value of the target cluster.

[0022] Obtain the centroids of all clusters, calculate the distance between the centroid of the target cluster and the centroid of each other cluster, and use the negative of the average of all distances as the independent variable of an exponential function with the natural constant as the base to obtain the similarity between the target cluster and each other cluster.

[0023] The product of the scale feature value and the similarity is calculated to obtain the degree of consistency in data variation between the target cluster and other clusters.

[0024] Preferably, obtaining the parameter adjustment coefficient based on the consistency between the data distribution concentration and the data variation includes:

[0025] The parameter adjustment coefficient is obtained by multiplying the reciprocal of the data distribution concentration with the consistency of the data variation.

[0026] Preferably, the step of adaptively adjusting the preset initial length scale in the Gaussian process regression model according to the parameter adjustment coefficient to obtain the adaptive Gaussian process regression model includes:

[0027] Calculate the product between the parameter adjustment coefficient and the preset initial length scale to obtain the initial adjusted length scale, and use the initial adjusted length scale as the length scale in the Gaussian process regression model to obtain the initial adjusted Gaussian process regression model;

[0028] According to a preset ratio, a training set and a validation set are obtained from the acceleration data sequence. The training set is input into the initial adjusted Gaussian process regression model to obtain the predicted value of each data in the validation set. Based on each data in the validation set and its predicted value, the mean square error of the validation set is obtained.

[0029] Using the gradient descent optimization algorithm, the initial adjustment length scale in the initial adjusted Gaussian process regression model is adjusted to obtain a new adjusted Gaussian process regression model. The training set is input into the new adjusted Gaussian process regression model to obtain the mean squared error of the validation set. The new adjusted Gaussian process regression model is used as the initial adjusted Gaussian process regression model, and the initial adjustment length scale in the initial adjusted Gaussian process regression model is adjusted until a preset convergence condition is met to obtain the new adjusted Gaussian process regression model corresponding to the minimum mean squared error. The new adjusted Gaussian process regression model corresponding to the minimum mean squared error is used as the adaptive Gaussian process regression model.

[0030] The beneficial effects of the embodiments of the present invention compared with the prior art are as follows:

[0031] This invention, during the fiber drawing and coating process, acquires the tension acceleration of the optical fiber at the current moment and a preset number of historical moments prior to the current moment, obtaining an acceleration data sequence. Based on the difference between each data point in the acceleration data sequence and the data at its corresponding previous moment, the data in the acceleration data sequence are clustered to obtain at least two clusters. The cluster containing the tension acceleration at the current moment is designated as the target cluster. Based on the distance between every two data points in the target cluster, the data distribution concentration of the target cluster is obtained. Based on the number of data points in the target cluster and the distance between the target cluster and each other cluster, the consistency of data change between the target cluster and other clusters is obtained. Based on the data distribution concentration and the consistency of data change, a parameter adjustment coefficient is obtained, which is used to adjust the preset initial length scale in the Gaussian process regression model. Based on the parameter adjustment coefficient, the preset initial length scale in the Gaussian process regression model is adaptively adjusted to obtain an adaptive Gaussian process regression model. The acceleration data sequence is input into the adaptive Gaussian process regression model to obtain the predicted tension acceleration of the optical fiber at future moments. Based on the predicted tension acceleration, the UV curing parameters are intelligently adjusted. Specifically, based on the difference between each data point in the acceleration data sequence and the data before its corresponding time point, the data in the acceleration data sequence are clustered to obtain the target cluster where the pulling acceleration at the current time point belongs. Based on the distance between every two weighted difference feature values ​​in the target cluster, the number of data points in the target cluster, and the distance between the target cluster and each other cluster, it can be determined whether the change in the pulling speed at the current time point is a periodic change or a sudden change in the fiber during the drawing process. This allows for adaptive adjustment of the length scale in the GPR model, ensuring the accuracy of the prediction of the fiber acceleration at future time points, and ensuring that the subsequent curing parameter adjustment can maintain high precision and robustness under different production conditions, thereby significantly improving the accuracy and stability of UV lamp curing parameter adjustment. Attached Figure Description

[0032] To more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0033] Figure 1 This is a flowchart of a method for intelligent adjustment of ultraviolet curing parameters for optical fiber coating equipment, provided in Embodiment 1 of the present invention. Detailed Implementation

[0034] Embodiments of this disclosure are described in detail below, with examples of these embodiments illustrated in the accompanying drawings. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain this disclosure, and should not be construed as limiting it.

[0035] It should be noted that the terms "first," "second," etc., used in this disclosure and the accompanying drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this disclosure described herein can be implemented in orders other than those illustrated or described herein. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this disclosure. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this disclosure.

[0036] To illustrate the technical solution of the present invention, specific embodiments are described below.

[0037] See Figure 1 This is a flowchart of a method for intelligent adjustment of ultraviolet curing parameters in an optical fiber coating equipment, as provided in Embodiment 1 of the present invention. Figure 1 As shown, the method may include:

[0038] Step S101: During the fiber drawing and coating process, the pulling acceleration of the fiber at the current moment and a preset number of historical moments before the current moment are obtained to obtain an acceleration data sequence. Based on the difference between each data in the acceleration data sequence and the data before its corresponding moment, the data in the acceleration data sequence are clustered to obtain at least two clusters.

[0039] The UV curing process is a key step that determines the quality and performance of optical fiber coating. Its effect is easily affected by factors such as the fiber drawing speed. Therefore, it is necessary to predict the future drawing speed based on the fiber's historical drawing speed in order to intelligently adjust the UV curing parameters and achieve adaptive optimization of the curing process.

[0040] First, during the fiber drawing and coating process, the fiber pulling speed is collected in real time by a high-precision encoder or laser rangefinder. Since the fiber drawing process is an extremely high-speed, precise and dynamic continuous process, the response time scale of its key control loop is in the millisecond range. The pulling speed is usually between 10m / s and 40m / s, so the sampling frequency is set to 1000Hz. There is no limit here, and the implementer can set it according to the specific scenario.

[0041] Then, the historical pulling speed is obtained to predict the pulling speed of the optical fiber in the future. Since the UV curing parameters need to be adjusted in real time during the optical fiber drawing and coating process, and the time accuracy requirement is high, a large amount of historical data is needed to predict the pulling speed of the optical fiber in the future to improve the prediction accuracy. Therefore, in this embodiment of the invention, the pulling speed of the current moment and the pulling speed of each sampling moment within 1 minute before the current moment are combined into a speed data sequence to obtain the speed data sequence in the current time period, that is, the preset number is 60,000, which is used to predict the pulling speed of the optical fiber in the future. There is no limitation here, and the implementer can set it according to the specific scenario.

[0042] Traditional methods utilize Gaussian process regression (GPR) models to predict fiber pulling speeds in the near future based on historical fiber pulling speeds and output prediction uncertainties. This allows for feedforward adaptive adjustment of UV curing parameters, proactively addressing changes in fiber pulling speeds and reducing UV curing lag and over-adjustment. The length scale in the GPR model directly controls the smoothness or complexity of the fitted function. Traditionally, the length scale is fixed. However, if the scale is too large, the model may fail to capture the true details in the data, leading to underfitting. Conversely, if the scale is too small, the model may interpret each data point's offset as a genuine characteristic of the function, striving to capture every data point, including local fluctuations and subtle changes. This results in a function with numerous peaks and troughs, leading to poor generalization and inaccurate predictions of fiber pulling speeds in the near future. Therefore, in this embodiment of the invention, the length scale is dynamically adjusted according to the data changes in the speed data sequence to ensure that the prediction results and subsequent curing parameter adjustments maintain high accuracy and robustness under different production conditions, thereby significantly improving the accuracy and stability of adjusting the UV lamp curing parameters.

[0043] Considering that velocity is a state variable while acceleration is a process variable, during the fiber drawing process, changes in acceleration may reflect potential disturbances earlier and more sensitively than changes in velocity. Therefore, in this embodiment of the invention, the pulling velocity in the velocity data sequence of the current time period is converted into pulling acceleration to obtain an acceleration data sequence. Based on the changes in the data in the acceleration data sequence, the length scale in the GPR model is adaptively adjusted to obtain an adaptive Gaussian process regression model, which is used to predict the pulling velocity of the fiber at future times. The conversion of velocity into acceleration is a prior art and will not be described in detail here.

[0044] Under ideal conditions, the pulling speed of optical fiber during the drawing process is constant, i.e., the acceleration is zero. However, in actual production, the pulling speed may experience periodic fluctuations due to mechanical vibration, preform rotation, control loop oscillations, or sudden fluctuations due to unexpected events or rapid switching of system states, causing changes in acceleration. Since periodic fluctuations are highly structured and regular changes, their behavior is predictable. Therefore, if the current pulling speed is due to periodic fluctuations, a larger length scale should be chosen when using the GPR model to predict the optical fiber pulling speed in the near future to capture long-term trends. Conversely, if the current pulling speed is due to sudden fluctuations, a smaller length scale should be chosen to more sensitively respond to short-term changes, thereby improving the accuracy of predicting the optical fiber pulling speed in the near future and consequently improving the accuracy of adjusting UV curing parameters.

[0045] Therefore, after obtaining the acceleration data sequence, the weighted difference feature value of each data point in the acceleration data sequence is obtained based on the difference between each data point in the acceleration data sequence and the data point before its corresponding time. This value is then used to cluster the data in the acceleration data sequence to obtain at least two clusters. Furthermore, based on the distribution of the weighted difference feature values ​​in the clusters, the reason for the change in the traction speed at the current time is analyzed, so as to adaptively adjust the length scale in GPR.

[0046] Taking the a-th data point in the acceleration data sequence as an example, the specific method for obtaining the weighted difference feature value of the a-th data point is as follows:

[0047] For the a-th data in the acceleration data sequence, obtain the time corresponding to the a-th data and record it as the target time. Obtain the historical traction acceleration of the optical fiber at each time within 1 minute before the target time (there is no limit here, and the implementer can set it according to the specific scenario) to obtain the historical data sequence of the a-th data.

[0048] Calculate the absolute value of the difference between the a-th data point and any historical traction acceleration in the historical data sequence to obtain the acceleration difference value. Calculate the time interval between the a-th data point and any historical traction acceleration at the corresponding moment. Use the negative of the time interval as the independent variable of an exponential function with the natural constant as the base to obtain the degree of mutual reference between the a-th data point and any historical traction acceleration.

[0049] Obtain the degree of cross-reference between the a-th data point and each historical traction acceleration in the historical data sequence, accumulate all the degrees of cross-reference to obtain the accumulated value, calculate the proportion of the degree of cross-reference between the a-th data point and any historical traction acceleration in the accumulated value, and obtain the weight of the acceleration difference value between the a-th data point and any historical traction acceleration.

[0050] The acceleration difference value and its weight between the a-th data point and each historical traction acceleration in the historical data sequence are obtained respectively. The weighted sum of all acceleration difference values ​​is then performed to obtain the weighted difference feature value of the a-th data point.

[0051] In one embodiment, the formula for calculating the weighted difference eigenvalue of the a-th data point is:

[0052]

[0053] in, This represents the weighted difference eigenvalue of the a-th data point. This represents the number of all data points in the historical data sequence of the data point a. This represents the a-th data point in the acceleration data sequence. This represents the i-th historical traction acceleration in the historical data sequence of the a-th data point. The time interval between the a-th data point in the acceleration data sequence and the i-th historical traction acceleration in the historical data sequence of the a-th data point, and e represents the natural constant.

[0054] It should be noted that, Used to characterize the change in the a-th data point in the acceleration data sequence compared to historical time points. The larger the value, the faster the rate of change of the fiber drawing speed is within the historical time period corresponding to the a-th data point in the acceleration data sequence and the i-th historical traction acceleration in the historical data sequence of the a-th data point. The smaller the value, the smaller the time interval between the a-th data point in the acceleration data sequence and the i-th historical traction acceleration in its historical data sequence. and The greater the degree of cross-referencing between the two data points—that is, the more valuable the changes in the fiber's tension acceleration within the historical time period corresponding to the a-th data point in the acceleration data sequence and the i-th historical tension acceleration in the historical data sequence of the a-th data point—the better. The larger the value, the greater the degree of change in the data at the time corresponding to the a-th data point compared to historical times.

[0055] Furthermore, the weighted difference feature value of each data point in the acceleration data sequence is obtained, and all weighted difference feature values ​​are mapped onto a scatter plot. The horizontal axis of the scatter plot represents the time corresponding to each data point in the acceleration data sequence, and the vertical axis represents the weighted difference feature value of each data point in the acceleration data sequence. The HDBSCAN clustering algorithm is used to cluster the data points in the scatter plot, obtaining at least two clusters for further analysis of the data distribution within the clusters, thereby adaptively adjusting the length scale in the GPR. The HDBSCAN clustering algorithm is an existing technology and will not be described in detail here.

[0056] Step S102: The cluster in which the traction acceleration at the current moment is located is recorded as the target cluster. The data distribution concentration of the target cluster is obtained based on the distance between every two data points in the target cluster. The consistency of data change between the target cluster and other clusters is obtained based on the number of data points in the target cluster and the distance between the target cluster and each other cluster.

[0057] After dividing the acceleration data sequence into at least two clusters in step S101, the cluster containing the weighted difference feature value of the traction acceleration at the current moment is recorded as the target cluster. Based on the data distribution in the target cluster, the reasons for the change in traction acceleration at the current moment are analyzed, and the length scale in GPR is adaptively adjusted: If the data in the target cluster is highly concentrated on the time axis, and the difference between the weighted difference feature value of the traction acceleration at the current moment and the weighted difference feature value of the traction acceleration at historical moments is large, it indicates that the reason for the change in traction acceleration at the current moment may be a sudden event or a rapid switch in system state, etc. In this case, a smaller length scale should be selected to respond more sensitively to short-term changes; conversely, if the data in the target cluster is relatively loosely distributed on the time axis, and the weighted difference feature value of the traction acceleration at the current moment is similar to the weighted difference feature value of the traction acceleration at historical moments, it indicates that the reason for the change in traction acceleration at the current moment may be a periodic fluctuation caused by mechanical vibration, precast rod rotation, control loop oscillation, etc. In this case, a larger length scale should be selected to capture long-term trends.

[0058] In this embodiment of the invention, the data distribution concentration of the target cluster is first obtained based on the distance between every two weighted difference feature values ​​in the target cluster, which reflects the distribution concentration of the data in the target cluster on the time axis. Then, based on the number of data in the target cluster and the distance between the target cluster and each other cluster, the consistency of data change between the target cluster and other clusters is obtained, which reflects the similarity between the weighted difference feature value of the traction acceleration at the previous time and the weighted difference feature value of the traction acceleration at historical time. Then, by combining the data distribution concentration and the consistency of data change, the length scale in the GPR model is adaptively adjusted to obtain an adaptive Gaussian process regression model, which is used to predict the traction speed of the optical fiber at future time.

[0059] Specifically, the data distribution concentration of the target cluster is obtained based on the distance between every two weighted difference feature values ​​in the target cluster.

[0060] The data in the target cluster are sorted according to their corresponding time to obtain a time series. The time interval between each two adjacent data points in the time series is obtained. The average time interval and coefficient of variation of all time intervals are calculated. The coefficient of variation is used as the independent variable of an exponential function with the natural constant as the base to obtain the data distribution uniformity in the target cluster.

[0061] The sampling interval of the data in the acceleration data sequence is obtained and denoted as the standard time interval. In this embodiment of the invention, the standard time interval is 1ms. The ratio between the average time interval and the standard time interval is calculated to obtain the data temporal distribution concentration of the target cluster. The product between the data temporal distribution concentration and the data distribution uniformity is calculated. The negative number of the product is used as the independent variable of the exponential function with the natural constant as the base to obtain the data distribution concentration of the target cluster.

[0062] In one embodiment, the formula for calculating the data distribution concentration of the target cluster is:

[0063]

[0064] Where G represents the data distribution concentration of the target cluster, Let represent the average time interval between any two adjacent data points in the time series, d represent the standard time interval, Y represent the coefficient of variation of the time interval between any two adjacent data points in the time series, and e represent the natural constant. This represents an exponential function with the natural constant e as the base.

[0065] It should be noted that, The smaller Y is, the smaller the time interval between two adjacent data points in the time series, and the higher the consistency of the time interval between two adjacent data points. Consequently, the larger G is, the greater the concentration of the weighted difference feature values ​​in the target cluster on the time axis, and the greater the possibility that the change in the pulling speed at the current moment is caused by sudden events or rapid switching of system states, which are sudden fluctuations.

[0066] Furthermore, based on the amount of data in the target cluster and the distance between the target cluster and each other cluster, the degree of consistency in data changes between the target cluster and other clusters is obtained, specifically:

[0067] The ratio between the number of all data in the target cluster and the number of all data in the acceleration data sequence is used to obtain the size characteristic value of the target cluster.

[0068] Obtain the centroids of all clusters, calculate the distance between the centroid of the target cluster and the centroid of each other cluster, and use the negative of the average of all distances as the independent variable of an exponential function with the natural constant as the base to obtain the similarity between the target cluster and each other cluster.

[0069] The product of the scale feature value and the similarity is calculated to obtain the degree of consistency in data variation between the target cluster and other clusters.

[0070] In one implementation, the formula for calculating the consistency of data changes between the target cluster and other clusters is as follows:

[0071]

[0072] Where D represents the degree of consistency in data variation between the target cluster and other clusters, n represents the total number of data points in the target cluster, N represents the total number of data points in the acceleration data sequence (i.e., the total number of data points in all clusters), and z represents the total number of clusters. This represents the distance between the centroid of the target cluster and the centroid of the r-th other cluster. This represents an exponential function with the natural constant as its base.

[0073] It should be noted that the larger n is, the higher the consistency between the weighted difference characteristic value of the pulling acceleration at the current moment and the weighted difference characteristic value of the pulling acceleration at the historical moment. Consequently, the larger D is, the more consistent the way the fiber pulling speed changes at the current moment is with the way the fiber pulling speed changes at the historical moment. That is, the more similar the reasons for the change in the pulling speed at the current moment are with the reasons for the change in the pulling speed at the historical moment. The greater the possibility that the change in the pulling acceleration at the current moment is caused by periodic fluctuations such as mechanical vibration, preform rotation, and control loop oscillation. The smaller the value, the smaller the difference between the target cluster and other clusters. The higher the similarity between the weighted difference feature values ​​in the target cluster and the weighted difference feature values ​​in other clusters, the larger D becomes. The higher the consistency between the current pulling speed and the historical pulling speed data, the greater the possibility that the change in the pulling speed at the current moment is caused by periodic fluctuations such as mechanical vibration, precast rod rotation, and control loop oscillation.

[0074] Thus, we obtained the data distribution concentration of the target cluster and the degree of consistency in data changes between the target cluster and other clusters.

[0075] Step S103: Based on the consistency between the data distribution concentration and the data change, obtain the parameter adjustment coefficient, which is used to adjust the preset initial length scale in the Gaussian process regression model.

[0076] After obtaining the data distribution concentration of the target cluster and the consistency of data changes between the target cluster and other clusters through step S102, the two are combined to more accurately evaluate the comprehensive characteristics of the target cluster in the time and quantity dimensions. This allows for the acquisition of parameter adjustment coefficients for adaptively adjusting the preset initial length scale in the Gaussian process regression model, so that the adaptively adjusted length scale can reflect both the sensitivity of local changes and the stability of the overall trend.

[0077] The specific method for obtaining the parameter adjustment coefficient is as follows:

[0078] The parameter adjustment coefficient is obtained by multiplying the reciprocal of the data distribution concentration with the consistency of the data variation.

[0079] In one embodiment, the formula for calculating the parameter adjustment coefficient is:

[0080]

[0081] Where Y represents the parameter adjustment coefficient, G represents the data distribution concentration, and D represents the consistency of data changes.

[0082] It should be noted that the smaller G is and the larger D is, the looser the data distribution in the target cluster is over time. The more consistent the way the fiber drawing speed changes at the current moment is with the way it changes in the past moment, the more likely the fiber drawing speed at the current moment has changed due to periodic fluctuations. In this case, a larger length scale should be chosen, i.e., the larger Y is, to capture long-term trends. Conversely, the larger G is and the smaller D is, the more concentrated the data in the target cluster is on the time axis. The more inconsistent the way the fiber drawing speed changes at the current moment is with the way it changes in the past moment, i.e. the fiber drawing speed at the current moment may have changed due to sudden events. In this case, a smaller length scale should be chosen, i.e., the smaller Y is, to respond more sensitively to short-term changes in the fiber drawing speed.

[0083] Step S104: Based on the parameter adjustment coefficient, the preset initial length scale in the Gaussian process regression model is adaptively adjusted to obtain the adaptive Gaussian process regression model. The acceleration data sequence is input into the adaptive Gaussian process regression model to obtain the predicted tension acceleration of the optical fiber at future moments. Based on the predicted tension acceleration, the UV curing parameters are intelligently adjusted.

[0084] Since the parameter adjustment coefficient obtained in step S103 is based on the data characteristics of the traction acceleration within the current time period (within 1 minute before the current moment), it is essentially a feature value of the traction acceleration within the current time period, used to reveal the inherent structural characteristics of the data. That is, the parameter adjustment coefficient itself is not a hyperparameter in the Gaussian process regression model, but rather provides guidance for adjusting the hyperparameter. The adjustment of the hyperparameter should be based on the proportional relationship reflected by the feature value, so it cannot be directly used to adjust the preset initial length scale in the Gaussian process regression model. Therefore, in this embodiment of the invention, after obtaining the parameter adjustment coefficient for adjusting the preset initial length scale in the Gaussian process regression model through step S103, the gradient descent optimization algorithm is used to find the "proportional relationship" of the parameter adjustment coefficient to adaptively adjust the preset initial length scale in the Gaussian process regression model, resulting in an adaptive Gaussian process regression model. Specifically:

[0085] Calculate the product between the parameter adjustment coefficient and the preset initial length scale to obtain the initial adjusted length scale. In the Gaussian process regression model, the length scale is generally initialized to 1.0, so the initial length scale is set to 1.0. There is no restriction here. The implementer can set it according to the specific scenario and use the initial adjusted length scale as the length scale in the Gaussian process regression model to obtain the initial adjusted Gaussian process regression model.

[0086] The last 40% of the acceleration data sequence is obtained. The first 20% of the data in the last 40% is used as the training set, and the last 20% of the data in the last 40% is used as the validation set. There is no restriction here. The implementer can set the ratio of the training set and the validation set according to the specific scenario. The training set is input into the initial adjusted Gaussian process regression model to obtain the predicted value of each data in the validation set. Based on each data in the validation set and its predicted value, the mean square error of the validation set is obtained.

[0087] Using the gradient descent optimization algorithm, the initial adjustment length scale in the initial adjusted Gaussian process regression model is adjusted to obtain a new adjusted Gaussian process regression model. The training set is input into the new adjusted Gaussian process regression model to obtain the mean squared error of the validation set. This mean squared error is used as the objective function in the gradient descent optimization algorithm. There are no restrictions here; implementers can set other objective functions according to specific scenarios, such as mean absolute error, Huber loss, etc. The new adjusted Gaussian process regression model is used as the initial adjusted Gaussian process regression model. The initial adjustment length scale in the initial adjusted Gaussian process regression model is adjusted until a preset convergence condition is met, obtaining the new adjusted Gaussian process regression model corresponding to the minimum mean squared error. Common convergence conditions include: the gradient is close to zero, for example, the gradient value is less than... The change in the objective function value is less than a preset threshold, for example, the change in the objective function value is less than... To reach the maximum number of iterations, for example, 50, in this embodiment of the invention, a preset convergence condition is set as the change in the objective function value being less than a preset threshold, where the preset threshold is... There are no restrictions here; implementers can set convergence conditions according to specific scenarios, using the newly adjusted Gaussian process regression model corresponding to the minimum mean square error as the adaptive Gaussian process regression model. The gradient descent optimization algorithm is existing technology and will not be elaborated upon here.

[0088] After obtaining the adaptive Gaussian process regression model, the acceleration data sequence is input into the adaptive Gaussian process regression model to obtain the predicted traction acceleration of the optical fiber at future moments.

[0089] Based on the current pulling speed and predicted pulling acceleration of the optical fiber, the future pulling speed of the optical fiber at future moments is calculated using kinematic principles. Then, the future pulling speed is compared with the optimal curing speed range set for the UV curing process. If the future pulling speed is higher than the maximum value in the optimal curing speed range, it means that the residence time of the coating on the optical fiber in the UV curing zone is shortened and the curing energy is insufficient, requiring an increase in UV lamp power. If the future pulling speed is lower than the minimum value in the optimal curing speed range, it indicates that the residence time is longer and the curing energy may be excessive, requiring an appropriate reduction in UV lamp power to avoid over-curing or thermal damage. The formula for calculating the UV lamp power adjustment can be used as follows: The required UV lamp power is obtained, and the calculated UV lamp adjustment power is sent to the UV curing control module in real time to achieve dynamic adjustment of the UV lamp output power. This ensures that the coating can always achieve a stable, sufficient, and uniform curing effect under different production conditions and speed fluctuations. The intelligent adjustment of UV curing parameters based on predicted tensile acceleration is existing technology and will not be elaborated here.

[0090] In summary, this invention, during the fiber drawing and coating process, acquires the tension acceleration of the optical fiber at the current moment and a preset number of historical moments prior to the current moment, obtaining an acceleration data sequence. Based on the difference between each data point in the acceleration data sequence and the data at its corresponding previous moment, the data in the acceleration data sequence are clustered to obtain at least two clusters. The cluster containing the tension acceleration at the current moment is designated as the target cluster. Based on the distance between every two data points in the target cluster, the data distribution concentration of the target cluster is obtained. Based on the number of data points in the target cluster and the distance between the target cluster and each other cluster, the consistency of data change between the target cluster and other clusters is obtained. Based on the data distribution concentration and the consistency of data change, a parameter adjustment coefficient is obtained, which is used to adjust the preset initial length scale in the Gaussian process regression model. Based on the parameter adjustment coefficient, the preset initial length scale in the Gaussian process regression model is adaptively adjusted to obtain an adaptive Gaussian process regression model. The acceleration data sequence is input into the adaptive Gaussian process regression model to obtain the predicted tension acceleration of the optical fiber at future moments. Based on the predicted tension acceleration, the UV curing parameters are intelligently adjusted. Specifically, based on the difference between each data point in the acceleration data sequence and the data before its corresponding time point, the data in the acceleration data sequence are clustered to obtain the target cluster where the pulling acceleration at the current time point belongs. Based on the distance between every two weighted difference feature values ​​in the target cluster, the number of data points in the target cluster, and the distance between the target cluster and each other cluster, it can be determined whether the change in the pulling speed at the current time point is a periodic change or a sudden change in the fiber during the drawing process. This allows for adaptive adjustment of the length scale in the GPR model, ensuring the accuracy of the prediction of the fiber acceleration at future time points, and ensuring that the subsequent curing parameter adjustment can maintain high precision and robustness under different production conditions, thereby significantly improving the accuracy and stability of UV lamp curing parameter adjustment.

[0091] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be included within the protection scope of the present invention.

Claims

1. A method for intelligent adjustment of UV curing parameters in optical fiber coating equipment, characterized in that, The intelligent adjustment method for ultraviolet curing parameters in optical fiber coating equipment includes: During the fiber drawing and coating process, the pulling acceleration of the fiber at the current moment and a preset number of historical moments before the current moment are obtained to obtain an acceleration data sequence. Based on the difference between each data point in the acceleration data sequence and the data at the corresponding moment before it, the data in the acceleration data sequence are clustered to obtain at least two clusters. The cluster containing the current traction acceleration is denoted as the target cluster. The data distribution concentration of the target cluster is obtained based on the distance between every two data points in the target cluster. The consistency of data changes between the target cluster and other clusters is obtained based on the number of data points in the target cluster and the distance between the target cluster and each other cluster. Based on the consistency between the concentration of the data distribution and the variation of the data, parameter adjustment coefficients are obtained, which are used to adjust the preset initial length scale in the Gaussian process regression model; Based on the parameter adjustment coefficient, the preset initial length scale in the Gaussian process regression model is adaptively adjusted to obtain the adaptive Gaussian process regression model. The acceleration data sequence is input into the adaptive Gaussian process regression model to obtain the predicted tension acceleration of the optical fiber at future moments. Based on the predicted tension acceleration, the UV curing parameters are intelligently adjusted. The step of adaptively adjusting the preset initial length scale in the Gaussian process regression model according to the parameter adjustment coefficient to obtain the adaptive Gaussian process regression model includes: Calculate the product between the parameter adjustment coefficient and the preset initial length scale to obtain the initial adjusted length scale, and use the initial adjusted length scale as the length scale in the Gaussian process regression model to obtain the initial adjusted Gaussian process regression model; According to a preset ratio, a training set and a validation set are obtained from the acceleration data sequence. The training set is input into the initial adjusted Gaussian process regression model to obtain the predicted value of each data in the validation set. Based on each data in the validation set and its predicted value, the mean square error of the validation set is obtained. Using the gradient descent optimization algorithm, the initial adjustment length scale in the initial adjusted Gaussian process regression model is adjusted to obtain a new adjusted Gaussian process regression model. The training set is input into the new adjusted Gaussian process regression model to obtain the mean squared error of the validation set. The new adjusted Gaussian process regression model is used as the initial adjusted Gaussian process regression model, and the initial adjustment length scale in the initial adjusted Gaussian process regression model is adjusted until a preset convergence condition is met to obtain the new adjusted Gaussian process regression model corresponding to the minimum mean squared error. The new adjusted Gaussian process regression model corresponding to the minimum mean squared error is used as the adaptive Gaussian process regression model.

2. The intelligent adjustment method for ultraviolet curing parameters of an optical fiber coating equipment according to claim 1, characterized in that, The acceleration data sequence is clustered based on the difference between each data point and the data points preceding its corresponding time point, resulting in at least two clusters, including: For any data in the acceleration data sequence, the time corresponding to the data is obtained and recorded as the target time. The historical traction acceleration of the optical fiber is obtained at each time within a preset historical period before the target time, thus obtaining the historical data sequence of the data. Calculate the absolute value of the difference between any data point and any historical traction acceleration in the historical data sequence to obtain the acceleration difference value. Calculate the time interval between any data point and any historical traction acceleration at the corresponding moment. Use the negative of the time interval as the independent variable of an exponential function with the natural constant as the base to obtain the degree of mutual reference between any data point and any historical traction acceleration. Obtain the degree of cross-reference between any data and each historical traction acceleration in the historical data sequence, accumulate all the degrees of cross-reference to obtain the accumulated value, calculate the proportion of the degree of cross-reference between any data and any historical traction acceleration in the accumulated value, and obtain the weight of the acceleration difference value between any data and any historical traction acceleration. The acceleration difference value and its weight between any data and each historical traction acceleration in the historical data sequence are obtained respectively. All acceleration difference values ​​are weighted and summed to obtain the weighted difference feature value of any data. The weighted difference feature value of each data point in the acceleration data sequence is obtained separately. All weighted difference feature values ​​are mapped to a scatter plot. The horizontal axis of the scatter plot represents the time corresponding to each data point in the acceleration data sequence, and the vertical axis represents the weighted difference feature value of each data point in the acceleration data sequence. The data points in the scatter plot are clustered to obtain at least two clusters.

3. The intelligent adjustment method for ultraviolet curing parameters of an optical fiber coating equipment according to claim 1, characterized in that, The step of obtaining the data distribution concentration of the target cluster based on the distance between every two data points in the target cluster includes: The data in the target cluster are sorted according to their corresponding time to obtain a time series. The time interval between each two adjacent data points in the time series is obtained. The average time interval and coefficient of variation of all time intervals are calculated. The coefficient of variation is used as the independent variable of an exponential function with the natural constant as the base to obtain the data distribution uniformity in the target cluster. The sampling interval of the data in the acceleration data sequence is obtained and denoted as the standard time interval. The ratio between the average time interval and the standard time interval is calculated to obtain the data temporal distribution concentration of the target cluster. The product between the data temporal distribution concentration and the data distribution uniformity is calculated. The negative of the product is used as the independent variable of an exponential function with the natural constant as the base to obtain the data distribution concentration of the target cluster.

4. The intelligent adjustment method for ultraviolet curing parameters of an optical fiber coating equipment according to claim 1, characterized in that, The process of determining the degree of consistency in data changes between the target cluster and other clusters based on the amount of data in the target cluster and the distance between the target cluster and each other cluster includes: The ratio between the number of all data in the target cluster and the number of all data in the acceleration data sequence is used to obtain the size characteristic value of the target cluster. Obtain the centroids of all clusters, calculate the distance between the centroid of the target cluster and the centroid of each other cluster, and use the negative of the average of all distances as the independent variable of an exponential function with the natural constant as the base to obtain the similarity between the target cluster and each other cluster. The product of the scale feature value and the similarity is calculated to obtain the degree of consistency in data variation between the target cluster and other clusters.

5. The intelligent adjustment method for ultraviolet curing parameters of an optical fiber coating equipment according to claim 1, characterized in that, The step of obtaining the parameter adjustment coefficient based on the consistency between the data distribution concentration and the data variation includes: The parameter adjustment coefficient is obtained by multiplying the reciprocal of the data distribution concentration with the consistency of the data variation.

Images