Self-adaptive tension control and optimization method based on optical fiber winding process

Abstract

The invention provides a self-adaptive tension control and optimization method based on an optical fiber winding process, which is particularly applied to tension real-time monitoring and quality optimization in the optical fiber winding process, and comprises the following steps: firstly, acquiring optical fiber tension data in real time through a tension sensor, eliminating noise by using Kalman filtering, and ensuring the accuracy of the data; then, the tension change trend of the optical fiber is predicted based on historical and real-time data and an LSTM model, a prediction basis is provided for tension control, a control system adjusts the rotating speed of a motor and a tension adjusting device through a PID controller and an MPC optimization algorithm according to a prediction result and real-time tension data, it is ensured that the tension is stable and conforms to a set range, and after winding is completed, the tension of the optical fiber is adjusted. According to the method, the YOLOv5 deep learning model is combined with GPU acceleration for optical fiber defect detection, surface defects are recognized in real time, a feedback result is used for optimizing a control strategy, and the method effectively improves the stability and quality in the optical fiber production process and is suitable for a high-speed and high-precision optical fiber winding production line.

Description

Technical Field

[0001] The present invention relates to the technical field of optical fiber winding, and particularly to a method for realizing adaptive tension control and optimization during the optical fiber winding process. Background Art

[0002] During the optical fiber winding process, the control of the optical fiber tension has a decisive influence on the quality of the optical fiber ring, the performance of the fiber optic gyroscope, and the efficiency of the entire production process. Excessive tension will cause defects such as micro-cracks and fractures in the optical fiber, thus affecting the transmission performance of the optical fiber and even causing it to fail prematurely during use; while too small tension will cause uneven arrangement of the optical fiber during winding, resulting in loose layers of the optical fiber, thereby affecting the structural stability and optical performance of the optical fiber, and may even cause damage and performance degradation of the optical fiber. Therefore, accurately controlling the tension during the optical fiber winding process has always been an important research direction in optical fiber manufacturing and applications.

[0003] With the continuous improvement of the requirements for the quality of optical fibers in high-performance optical devices such as fiber optic gyroscopes, the difficulty of tension control during the optical fiber winding process has also increased accordingly. Especially in modern optical fiber winding equipment, the optical fiber winding speed is continuously increasing, and the traditional tension control methods can no longer meet the requirements of real-time and accuracy for high-speed production. Due to the influence of multiple factors such as the accuracy of mechanical equipment, environmental interference, and differences in optical fiber characteristics during the optical fiber winding process, traditional tension control systems often cannot operate stably under the requirements of high speed and high precision, resulting in large tension fluctuations, and then generating winding defects, affecting the performance and production efficiency of the optical fiber. Therefore, it is urgent to develop a new type of optical fiber tension detection and control method that can adapt to the requirements of high speed and high precision. Summary of the Invention

[0004] The purpose of the present invention is to provide an adaptive tension control and optimization method based on the optical fiber winding process, aiming to achieve high-quality production of optical fiber products by accurately controlling the tension fluctuations during the optical fiber winding process. This method can effectively avoid optical fiber damage or performance degradation caused by uneven tension, and improve the stability and production efficiency during the optical fiber winding process by optimizing the tension control strategy, and is particularly suitable for high-speed optical fiber winding and the production of high-precision optical fiber products.

[0005] The present invention will be described in detail below with reference to the accompanying drawings.

[0006] The present invention provides an adaptive tension control and optimization method based on the optical fiber winding process, and the specific steps are as follows:

[0007] Step 1: Real-time detection of optical fiber tension and filtering of sampled data. Use a high-precision sensor to collect the tension data during the optical fiber winding process in real time, and preprocess the collected data to eliminate noise interference through Kalman filtering to ensure the accuracy of the tension data;

[0008] Step 2: Fiber optic tension analysis and prediction based on a time series prediction model. Based on historical data and real-time tension data, the tension trend is predicted through the LSTM algorithm based on GPU, and the law of tension change of the fiber optic under different winding states is analyzed to provide a prediction basis for the next tension adjustment;

[0009] Step 3: PID parameter tuning for fiber optic tension control assisted by MPC. According to the prediction results and current tension data, the control system adjusts the parameters of the PID controller to dynamically adjust the fiber optic tension in real time. At the same time, the model predictive control (MPC) algorithm is applied to optimize the control response to ensure that the fiber optic tension is always maintained within the set safe range;

[0010] Step 4: Adaptive adjustment of fiber optic tension feedback. According to the adjustment results of the tension control module, the execution feedback module adjusts the actual tension by adjusting the motor speed and the tension adjustment device. This process is carried out in real time to ensure the stability and uniformity of the fiber optic tension;

[0011] Step 5: Fiber optic quality inspection and optimization based on machine vision. After winding is completed, the quality of the fiber optic product is inspected through optical imaging and defect detection technology based on YOLOv5 to confirm the effect of tension control, and the tension control algorithm is further optimized through feedback data.

[0012] The present invention has the following advantages compared with the prior art:

[0013] The adaptive tension control and optimization method based on the fiber optic winding process has the following advantages:

[0014] 1. Improve the quality of fiber optic: By precisely controlling the tension during the fiber optic winding process, fiber optic damage caused by excessive or too small tension is avoided, and the strength, stability and optical performance of the fiber optic are improved.

[0015] 2. Enhance production efficiency: The present invention can detect and adjust tension fluctuations in real time during high-speed fiber optic winding, reduce defects and production pauses caused by uneven tension, and significantly improve production efficiency.

[0016] 3. High-speed data processing: The data recognition and training process is accelerated based on the Nvidia RTX1060 GPU, achieving a training speed far exceeding that of hardware platforms such as CPUs and FPGAs, further improving production efficiency.

[0017] 4. Adapt to complex production environments: The present invention can dynamically adjust control strategies under different environmental conditions to adapt to the variability during the fiber optic winding process, ensuring high-quality production of fiber optic products. Description of the Drawings

[0018] Figure 1 is the flow chart of an adaptive tension control and optimization method based on the optical fiber winding process of the present invention;

[0019] Figure 2 is the logical structure diagram of an adaptive tension control and optimization method based on the optical fiber winding process of the present invention;

[0020] Figure 3 is the schematic diagram of the optical fiber tension analysis and prediction based on the time series prediction model in the second step of the present invention;

[0021] Figure 4 is the schematic diagram of the PID parameter tuning and adaptive feedback control assisted by MPC of the present invention;

[0022] Figure 5 is the schematic diagram of the optical fiber quality detection and optimized annotation training set based on machine vision in the fifth step of the present invention;

[0023] Figure 6 is the optical fiber detection of the test example of the present invention Figure 1 ;

[0024] Figure 7 is the optical fiber detection of the test example of the present invention Figure 2 . Detailed implementation manners

[0025] The detailed implementation manners of the present invention realize efficient tension control during the optical fiber winding process by combining advanced adaptive tension control and optimization methods, ensuring the quality of optical fiber products. The following combines actual application scenarios to introduce each step in detail.

[0026] Embodiment

[0027] Hardware environment: In this embodiment, an NVIDIA RTX 1060 GPU is used as the main computing platform, providing powerful parallel computing capabilities to support the efficient execution of deep learning algorithms. In addition, the computer is configured with an Intel Core i7 processor, 16GB of memory, and sufficient hard disk space for large-scale data storage and real-time computing.

[0028] Software environment: Ubuntu 20.04 is used to support image processing and deep learning model training. The following development environments are used:

[0029] PyTorch 2.1.2: Used for the training and inference of deep learning models. PyTorch provides powerful tensor computing and automatic differentiation functions, suitable for the implementation of deep learning optimization algorithms in the present invention.

[0030] Torchvision 0.4.0: Used for image processing related functions, supporting operations such as data preprocessing and data augmentation to improve the accuracy of image detection.

[0031] CUDA 12.1.0: Used for GPU accelerated computing to improve the efficiency of deep learning model training and inference processes, significantly shortening the computing time.

[0032] NumPy 1.22.4: Used for numerical calculation and data processing, supporting mathematical calculation and optimization processing of optical fiber tension data.

[0033] OpenCV 4.5.2: Used for image acquisition and processing, providing clear and accurate visual support for image data during the optical fiber winding process.

[0034] Matplotlib 3.4.3: Used for plotting visual image data and tension change curves to help analyze the tension fluctuations and defects during the optical fiber winding process.

[0035] TensorFlow 2.x: If further optimization of the training process is required, TensorFlow can be used as an alternative for training and tuning deep learning models.

[0036] Optical fiber winding platform: Includes a take-up motor (model 57HSZ3N), a pay-off motor (model 57HSZ3N), a tension sensor (model JZHL-M1), a dancer wheel, a traversing motor (model SST42D), a ball screw, a screw slide table and other equipment. The take-up and pay-off motors jointly control the optical fiber winding process, and the tension sensor monitors the tension of the optical fiber in real time to ensure that the data is fed back to the control system in real time.

[0037] Optical imaging platform: Includes a camera (OV5640 camera produced by OmniVision) and its peripheral PCB circuit, a strip LED array with stepless brightness adjustment and a lifting bracket.

[0038] The following describes the specific implementation manners of the present invention with reference to the accompanying drawings.

[0039] As Figures 1 to 5 shown, an adaptive tension control and optimization method based on the optical fiber winding process includes the following steps:

[0040] Step 1: Real-time detection of optical fiber tension and filtering of sampled data.

[0041] Construction of the optical fiber winding platform:

[0042] According to the specific requirements of fiber winding, the fiber diameter is set to 125μm, the winding speed is 30 - 60r / min, and the tension control range is 5g ± 10%. This setting can ensure that the tension during fiber winding remains within the ideal range, avoiding fiber breakage caused by excessive tension or uneven winding caused by too low tension.

[0043] The take-up motor is supported by a bracket and connected to the take-up wheel using a coupling. The pay-off motor is the same. The tension sensor (model JZHL-M1) measures the variable-pressure fiber tension through a guide pulley and is equipped with a digital transmitter (model BSQ-DG) to convert analog data into RS232 digital signals for GPU processing. The wire arranging mechanism consists of a stepper motor (model SST42D), a ball screw, and a screw slide. For every 1° rotation of the ball screw, it linearly moves approximately 11μm, meeting the wire arranging accuracy requirements.

[0044] Selection and setup of sensors:

[0045] A high-precision tension sensor (model JZHL-M1) is installed on the fiber winding platform to measure the variable-pressure fiber tension through a guide pulley. This sensor can accurately measure the instantaneous tension value of the fiber during winding. To ensure the accuracy of the measured data, the sensor is also equipped with a digital transmitter (model BSQ-DG) to convert the analog signal into an RS232 digital signal and transmit it to the GPU processing platform in real time through an interface. The above structure is as Figure 2 shown.

[0046] Data filtering and preprocessing:

[0047] The GPU main processing platform uses the CUDA acceleration algorithm to process the tension data collected by the sensor in real time to ensure the accuracy of the data under noise interference. To ensure the accuracy of subsequent tension analysis and prediction, the Kalman filter method is selected here to process the data in real time, weakening the noise of the fiber tension data and optimizing the signal quality. The Kalman filter is a recursive algorithm that can estimate the state of a dynamic system in a noisy environment. Assuming that the change in fiber tension can be represented by a state space model, the Kalman filter optimizes the state estimate through historical data and current observations.

[0048] State update:

[0049]

[0050] Among them, is the current estimated state of the fiber tension, z k is the current sensor measurement value, and K k is the Kalman gain.

[0051] Kalman gain:

[0052]

[0053] Among them, P k-1 is the estimated error covariance at the previous moment, and R is the covariance of the micro-sensor noise.

[0054] If there are still some slight fluctuations or noise effects in the data after Kalman filtering, methods such as the moving average method can be used to smooth the data.

[0055] Step 2: Fiber optic tension analysis and prediction based on the time series prediction model;

[0056] Construction of a deep learning model based on historical data and real-time data:

[0057] To more accurately predict the change trend of the fiber optic tension, the system combines historical data and real-time collected tension data to construct a deep learning optimization algorithm model. The specific algorithm used is LSTM (Long short-term memory), which is particularly suitable for processing time series data and can capture the temporal features and long-term and short-term dependencies in the data. As a special type of RNN (Recurrent Neural Network), LSTM controls the flow of information by introducing input gates, forget gates, and output gates, and can selectively remember the tension data useful for the current prediction and suppress irrelevant noise. During the fiber winding process, the tension is affected by multiple factors such as the equipment state, material properties, and winding speed, and these factors have long-term dependencies at different time points. LSTM can effectively capture these complex temporal dependencies. The LSTM algorithm is deployed on Nvidia RTX 1060 for parallel computing, and the main algorithm is as follows:

[0058] In the input gate:

[0059] i t = σ(W i · [h t-1 , x t + b i ) (3)

[0060] Among them, i t is the activation value of the input gate, x t is the input data at that moment, which affects the tension of the fiber optic, h t-1 is the hidden state at the previous moment, and the tension output predicted at the previous moment.

[0061] In the forget gate:

[0062] f t = σ(W f · [h t-1 , xt +b f ) (4)

[0063] Among them, f t represents the activation value of the forget gate, which controls whether the tension information at the previous moment is retained.

[0064] Memory cell update:

[0065]

[0066] Among them, C t is the memory cell state at the current moment, C t-1 is the memory state at the previous moment, is the candidate memory update value at the current moment, f t and i t control the proportion of information flowing in and out.

[0067] Output gate:

[0068] o t = σ(W o ·[h t-1 , x t +b o ) (6)

[0069] Among them, o t determines the influence degree of the memory cell on the output at the current moment.

[0070] Final output:

[0071] h t = o t ·tanh(C t ) (7)

[0072] Among them, h t is the output state at the current moment, which is used to generate the predicted value or as the input for the next moment.

[0073] Through the training of a large amount of historical data, the LSTM model can effectively identify the change rules of the optical fiber tension under different operating conditions, ensuring that accurate tension trend information can be provided during prediction.

[0074] Optical fiber tension prediction based on LSTM:

[0075] The input of the LSTM model includes the tension data at the current moment and the tension sequence within a certain past time range. Specifically, the LSTM model accepts the following inputs: the tension value at the current moment, which is collected by a real-time sensor; the historical tension data, including the tension values at the past N moments, forming time series data. The LSTM predicts the changing trend of the fiber tension within the next T moments based on these inputs, and the predicted values are usually the tension values at the next few time steps. The predicted tension values will be used as a reference for the next tension adjustment and provide a basis for the decision-making of the control system.

[0076] Step 3: Fiber tension control based on PID parameter tuning assisted by MPC.

[0077] MPC (Model Predictive Control) uses a dynamic system model to optimize the control strategy by predicting the future tension changing trend. During the fiber winding process, MPC calculates the control input based on the prediction results of the future tension by the LSTM model. The core of the MPC controller is an optimization problem, that is, according to the current system state and the tension prediction for a period of time in the future, to find a set of control signals that can optimize the system performance. MPC calculates the optimal control strategy by solving a constrained optimization problem. Set the objective optimization function and constraint conditions, where the purpose of the objective optimization function is to make the fiber tension as close as possible to the target value, and the constraint conditions ensure that the tension is within the specified safe range. At each moment, the MPC controller solves the above optimization problem to obtain a series of control inputs. The actual control strategy will select the control input at the current moment and then recalculate the future control signals. After each execution of the controller, it will obtain the actual fiber tension value and new sensor data, and feedback this information to the MPC controller. Based on the feedback data, the controller will recalculate the control strategy to cope with the changes in the system state.

[0078] With the assistance of the MPC controller, the PID controller is used to finely tune the control signal. In the fiber winding system, the role of the PID controller is to calculate a control signal based on the error between the real-time tension and the target tension, and fine-tune the behavior of the system through an adaptive feedback mechanism to further accurately control the fiber tension. The control signal of the PID controller can be expressed as:

[0079]

[0080] where, e(t) = T target -T actual (t) is the tension error at the current moment, K p , K i , K d are the proportional, integral, and differential gain coefficients respectively, T target is the target tension, Tactual (t) is the current actual tension. The PID parameters are adjusted by the Ziegler-Nichols method.

[0081] Step Four: Adaptive adjustment of optical fiber tension feedback.

[0082] After performing PID and MPC control, the system will obtain feedback data in real time and compare the difference between the actual tension and the target tension. If the actual tension exceeds the preset safety range, the feedback mechanism will adjust the speed of the motor or the parameters of other regulating devices to quickly return to the target tension range. According to the feedback signal, the control system will adjust the speed of the take-up motor or the pay-off motor in a timely manner. By appropriately increasing or decreasing the motor speed, the tension is precisely controlled. In addition, the tension regulating device (such as a tension sensor and an electric regulating mechanism) will fine-tune the tension according to the control signal. The feedback signal ensures that this adjustment process is fast and stable, thus ensuring uniform tension of the optical fiber and achieving adaptive tension control.

[0083] Step Five: Optical fiber quality inspection and optimization based on machine vision.

[0084] Use a high-resolution camera to collect optical fiber images under a bar-shaped LED array to ensure uniform and clear light. Process the images through OpenCV for denoising, enhancing contrast, and cropping and scaling to ensure the prominence of the optical fiber in the images. Manually annotate optical fiber defects (such as cracks and scratches) on the labelimg software, and label the data as categories and positions. Use GPU acceleration to train the YOLOv5 model, perform data augmentation, optimize the network structure, and improve the detection accuracy. Use the trained YOLOv5 model to detect optical fiber defects in the images in real time and output the defect categories and positions. The recognition process is as follows:

[0085] Prediction result = YOLOv5(I) = {type, x, y, w, h} (9)

[0086] where I is the input image, x and y are the center coordinates of the defect box, and w and h are the width and height of the defect box.

[0087] Filter defects according to the output box and confidence level to determine whether the optical fiber is qualified. Feed back the YOLOv5 detection results to the production system, analyze common defects, and adjust production parameters (such as tension, speed, etc.). Feed the detection results back into the model for incremental training and optimization to improve the detection accuracy. Feed the detection results back into the production process, and combine with the PID and MPC control algorithms to optimize the optical fiber winding parameters. If certain defects are detected frequently, the system can adjust parameters such as tension and speed in the optical fiber winding process through the feedback data to avoid the recurrence of the same defects.

[0088] Such as Figures 6 to 7As shown, test case: Optical fiber detection:

[0089] When YOLOv5 detects axial arrangement non-uniformity defects (confidence level ≥ 0.75) on the surface of the optical fiber, the system automatically associates the following multi-source data:

[0090] Tension control historical curve (sampling frequency 1kHz)

[0091] Motor speed fluctuation record (resolution ±0.1r / min)

[0092] Environmental temperature and humidity sensor data (accuracy ±0.5℃)

[0093] Among them, the optimization method is:

[0094] Defect-parameter correlation analysis: Calculate the correlation between the defect occurrence rate and the standard deviation of tension and environmental temperature and humidity through the Pearson correlation coefficient, and screen out strongly correlated parameters with |r|>0.7. For example: when the environmental humidity > 65%, the occurrence probability of bubble defects increases by 2.8 times; when the standard deviation of tension σ > 0.25g, the growth slope k of microcrack defects = 0.12 / hour;

[0095] Dynamic parameter compensation: According to the correlation analysis results, implement a hierarchical control strategy: First-level compensation: Adjust the integral time constant Ti of the PID controller (±20%); Second-level compensation: Modify the prediction time domain length of the MPC (10→15 steps); Third-level compensation: Activate the standby drying unit to maintain the humidity ≤ 60%;

[0096] Technical effect: In the production verification for 6 consecutive months, this optimization method achieved: the axial arrangement defect rate decreased from 1.2% to 0.35%; the tension control accuracy increased by 22% (σ = 0.18g → 0.14g); the overall equipment efficiency (OEE) increased by 17.5 percentage points.

[0097] It should be clear that the embodiments involved in the present invention have diverse implementation manners. It can be independently completed relying on computer hardware, or can be achieved through an organic combination of hardware and software, or can be implemented by computer instructions stored in a non-transitory computer-readable memory. In the implementation process, the method can use standard programming techniques. Specifically, a computer program is constructed through a non-transitory computer-readable storage medium configured with a computer program. The storage medium configured in this way can prompt the computer to operate in a specific and predefined manner, and this operation process is realized according to the method described in the specific embodiment and the attached drawings. Each program can be written in a high-level procedure or object-oriented programming language when communicating with the computer system. Of course, if the actual need dictates, the program can also be implemented using assembly language or machine language. In any case, the language used can be a compiled or interpreted language. In addition, for a specific purpose, the program is capable of running on a programmed application-specific integrated circuit to fully exert its functions.

[0098] Furthermore, the method can be effectively implemented on various types of computing platforms, as long as these computing platforms are operably connected to a suitable environment. Its coverage is extensive, including but not limited to personal computers, minicomputers, mainframes, workstations, network or distributed computing environments, separate or integrated computer platforms, and platforms that can communicate with charged particle tools or other imaging devices, etc. Various aspects of the present invention can be implemented in the form of machine-readable code stored on a non-transitory storage medium or device. Whether these storage media or devices are removable or integrated into the interior of the computing platform, such as hard disks, optical read and / or write storage media, RAM, ROM, etc., can be read by a programmable computer. When the storage medium or device is read by the computer, it can be used to configure the computer and make it execute the process described in detail herein. In addition, the machine-readable code or part of its content can be transmitted through wired or wireless networks. When such transmission media contain instructions or programs that combine with a microprocessor or other data processors to implement the steps described above, the invention described herein covers these and other different types of non-transitory computer-readable storage media. When programmed according to the methods and techniques described in the present invention, the scope of the present invention also includes the computer itself, which means that the entire computer system belongs to the protection scope of the present invention during the process of running the relevant program to achieve the technical effects of the invention.

[0099] In summary, the above-described content is only a preferred embodiment of the present invention. However, the present invention is not limited to the embodiments mentioned above. As long as other implementation means can achieve the same technical effects as the present invention, they should all be included within the protection scope of the present invention. Within the defined protection scope of the present invention, various different modifications and variations are allowed in its technical solutions and / or implementation manners. This reflects the flexibility and adaptability of the present invention in the process of technology application and development, aiming to adapt to the ever-changing technical requirements and application scenarios, and at the same time leaving a broad space for subsequent technical improvement and innovation.

Claims

1. An adaptive tension control and optimization method based on the optical fiber winding process, characterized in that It includes the following steps: Step 1: Real-time detection of optical fiber tension and filtering of sampled data; Step 2: Analysis and prediction of optical fiber tension based on a time series prediction model; Step 3: PID parameter tuning for optical fiber tension control assisted by MPC; Step 4: Adaptive adjustment of optical fiber tension feedback; Step 5: Quality detection and optimization of optical fiber based on machine vision.

2. The adaptive tension control and optimization method based on the optical fiber winding process according to claim 1, characterized in that For the said Step 1: The method for real-time detection of optical fiber tension and filtering of sampled data is as follows: High-precision sensors are used to collect the tension data during the optical fiber winding process in real time, and the collected data is preprocessed. The Kalman filter is used to eliminate noise interference to ensure the accuracy of the tension data.

3. An adaptive tension control and optimization method based on the optical fiber winding process according to claim 1, characterized in that, For the said Step 2: The method for real-time detection of optical fiber tension and filtering of sampled data is as follows: Based on historical data and real-time tension data, the LSTM algorithm based on GPU is used to predict the tension trend, analyze the tension change law of the optical fiber under different winding states, and provide a prediction basis for the next tension adjustment.

4. An adaptive tension control and optimization method based on the optical fiber winding process according to claim 1, characterized in that For the said Step 3: The method for PID parameter tuning for optical fiber tension control assisted by MPC is as follows: According to the prediction result of Step 2 and the current tension data, the control system adjusts the parameters of the PID controller, dynamically adjusts the optical fiber tension in real time, and at the same time applies the model predictive control (MPC) algorithm to optimize the control response to ensure that the optical fiber tension is always maintained within the set safe range.

5. An adaptive tension control and optimization method based on the optical fiber winding process according to claim 1, characterized in that, For the said Step 4, the method for adaptive adjustment of optical fiber tension feedback is as follows: According to the adjustment result of the tension control module in Step 3, the execution feedback module adjusts the motor speed and the tension adjustment device to perform actual tension adjustment. This process is carried out in real time to ensure the stability and uniformity of the optical fiber tension.

6. An adaptive tension control and optimization method based on the optical fiber winding process according to claim 1, characterized in that, For the said Step 5, the method for quality detection and optimization of optical fiber based on machine vision is as follows: After the winding described in Step 4 is completed, the quality of the optical fiber product is detected through optical imaging and the YOLOv5 defect detection technology, the effect of the tension control is confirmed, and the tension control algorithm is further optimized through the feedback data.

Images