An ultrathin hollow coil winding system with self-learning optimization function
By integrating automated winding equipment with deep learning models, the problems of relying on manual parameter adjustment and high-dimensional data processing in existing winding equipment have been solved, enabling high-precision and stable production of ultra-thin hollow coils, and improving production efficiency and product quality.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- TANAC AUTOMATION
- Filing Date
- 2026-01-28
- Publication Date
- 2026-06-02
AI Technical Summary
Existing winding equipment relies on manual parameter adjustment, resulting in poor quality stability, significant challenges in high-dimensional data processing, and a lack of self-learning and continuous optimization capabilities, making it difficult to meet the high-precision production requirements of ultra-thin hollow coils.
The main body of the winding machine integrates automatic feeding, winding, heating and shaping, and unloading. It combines hardware detection modules and software control systems, and uses a deep learning model that integrates attention mechanism and graph neural network to realize real-time acquisition, analysis and optimization of parameters. It processes noise through PCA/autoencoder dimensionality reduction, wavelet denoising and Kalman filtering technology, LSTM/TCN to model time dependence, and DQN/distributed PPO to accelerate training, so as to achieve adaptive adjustment.
It has achieved fully automated and precise production, improved production continuity and stability, reduced labor costs, improved product quality consistency and production efficiency, enhanced system adaptability and flexibility, and reduced scrap rate and overall production costs.
Smart Images
Figure CN122136169A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of coil winding technology, and specifically to an ultra-thin hollow coil winding system with self-learning optimization function. Background Technology
[0002] Ultra-thin hollow coils are core components of electronic devices, sensors, and other products. Their dimensional accuracy, winding tightness, and consistency of electrical properties (resistance, inductance, etc.) directly determine the operational stability and lifespan of the end product. However, existing winding equipment suffers from the following key technical deficiencies: Relying on manual parameter adjustment results in poor quality stability: Traditional winding equipment mostly adopts a fixed parameter operation mode, requiring operators to manually adjust key indicators such as rotation speed, die spacing, tension, and heating parameters based on production experience. It is impossible to perceive the dynamic changes in the physical and electrical characteristics of the coil in real time, resulting in significant differences in dimensional accuracy, winding tightness, and electrical performance between different batches or even within the same batch of products. This leads to insufficient quality stability and makes it difficult to meet the high-precision production requirements of ultra-thin hollow coils.
[0003] High-dimensional data processing presents significant challenges: During the winding process, there are complex nonlinear relationships between the coil's physical parameters (size, thickness, etc.), electrical parameters (resistance, inductance, etc.), and equipment operating parameters (rotation speed, tension, etc.), and these parameters exhibit strong time dependence with the production process. Furthermore, data acquisition is susceptible to environmental interference and noise, and uneven data collection exists in actual production. This leads to existing machine learning models facing problems such as high computational complexity, slow training convergence, and weak generalization ability, making it difficult to accurately capture parameter correlation patterns and achieve effective parameter optimization.
[0004] Lack of self-learning and continuous optimization capabilities: Existing winding equipment does not have the function of autonomous data analysis and model iterative updates. When the coil specifications change, the characteristics of raw materials fluctuate, or the production environment changes, the parameters need to be manually adjusted again. The adaptability is poor and the production efficiency is low, which further restricts the level of intelligence and product qualification rate of ultra-thin hollow coil production. Summary of the Invention
[0005] To address the aforementioned issues, this invention discloses an ultra-thin hollow coil winding system with self-learning optimization capabilities, comprising a winding machine body, a hardware detection module, a software control system, and an execution adjustment module. The winding machine integrates an automatic feeding unit, a winding unit, a heating and shaping unit, a wire cutting unit, and a unloading unit, realizing fully automated operation of the coil winding process; The hardware detection module is used to collect the physical parameters, electrical parameters, and winding machine operating parameters of the coil in real time. The software control system is connected to the hardware detection module by signal, and has a built-in deep learning model that integrates attention mechanism and graph neural network. It is configured with a data preprocessing unit, a time-dependent modeling unit, a data optimization acquisition unit and an accelerated training unit to analyze parameter mapping relationship and generate optimization adjustment instructions. The execution adjustment module is electrically connected to both the software control system and the main body of the winding machine. It is used to receive optimization adjustment commands and drive the corresponding unit of the main body of the winding machine to adjust the operating parameters, thereby achieving adaptive adjustment and continuous optimization.
[0006] Preferably, the hardware detection module includes a physical parameter detection component and an electrical parameter detection component. The physical parameter detection component includes a vision sensor and a pressure sensor. The vision sensor is used to detect the outer dimensions, inner dimensions, and thickness of the coil, and the pressure sensor is used to detect the winding tightness of the coil.
[0007] Preferably, the electrical parameter detection component uses a high-precision multimeter and an inductance tester. The high-precision multimeter is used to collect the resistance, current, and voltage parameters of the coil, and the inductance tester is used to collect the inductance parameters of the coil.
[0008] Preferably, the operating parameters of the winding machine include rotational speed, die spacing, tension, heating current intensity and energizing time, and the hardware detection module synchronously collects the above operating parameters and transmits them to the software control system.
[0009] Preferably, the data preprocessing unit uses PCA or an autoencoder to reduce the dimensionality of high-dimensional data, and combines wavelet denoising and Kalman filtering techniques to suppress data noise and improve the quality of the original data.
[0010] Preferably, the time-dependent modeling unit constructs a time series model through an LSTM network or a TCN network to accurately capture the time dependency between coil parameters and winding machine operating parameters.
[0011] Preferably, the data optimization acquisition unit adopts an adaptive sampling strategy to supplement scarce data samples in a targeted manner, and combines data augmentation technology to expand the diversity of the dataset, thereby solving the problem of uneven data acquisition in actual production.
[0012] Preferably, the accelerated training unit uses a parallel DQN algorithm or a distributed PPO algorithm to accelerate the model training process, while also using a priority experience replay mechanism to optimize policy learning efficiency and improve the model training convergence speed.
[0013] Preferably, the execution adjustment module includes a speed adjustment component, a spacing adjustment component, a tension adjustment component, and a heating adjustment component, which respectively adjust the rotational speed of the winding machine, the die spacing, the winding tension, and the heating current intensity and energizing time.
[0014] Preferably, the software control system further includes a data storage unit and a model training unit. The data storage unit is used to store various collected parameters and model training data, and the model training unit continuously iterates and updates the deep learning model based on the optimized dataset to achieve continuous improvement of the system's self-learning capability.
[0015] Compared with the prior art, the beneficial effects of the present invention are as follows: 1. Completely eliminate reliance on manual labor and achieve fully automated and precise production. The system integrates automatic feeding, winding, heating and shaping, unloading and other full-process operation units. With the real-time acquisition capability of multi-dimensional parameters of the hardware detection module and the automatic analysis and optimization function of the software control system, there is no need for manual adjustment of key parameters such as rotation speed and mold spacing. It completely avoids quality fluctuations caused by human operation errors and reliance on experience, greatly reduces labor costs, and realizes unmanned control from raw materials to finished products, significantly improving production continuity and stability.
[0016] 2. This system efficiently addresses the challenges of high-dimensional data processing, improving model training and optimization efficiency. It tackles industry pain points such as parameter nonlinearity, strong time dependence, high noise interference, and uneven data collection during the winding process. A preprocessing combination of PCA / autoencoder dimensionality reduction + wavelet denoising and Kalman filtering denoising simplifies model computational complexity and improves data quality. LSTM / TCN time-dependent modeling captures dynamic parameter correlations. Adaptive sampling and data augmentation address data shortcomings. Combined with a parallel DQN / distributed PPO+PER accelerated training scheme, the system improves model training convergence speed, effectively solving the problems of computational complexity and weak generalization ability in existing models, providing technical support for accurate parameter optimization.
[0017] 3. Core algorithm innovation empowers consistent and high-precision product quality. The innovative design of deep learning model integrating attention mechanism and graph neural network can automatically filter key parameters affecting coil quality (such as winding tension and heating current intensity), ignore redundant information, and improve the optimization focus. At the same time, the graph structure is used to accurately capture the interaction relationship between physical parameters, electrical parameters and operating parameters, which significantly improves the prediction accuracy of the model.
[0018] 4. Self-learning and continuous optimization enhance system adaptability and flexibility. The software control system has a built-in model training unit that continuously updates the deep learning model based on new data generated during production. This allows the system to autonomously learn the optimal parameter combinations for different coil specifications, raw material characteristics, and production environments. When coil specifications change, raw material prices fluctuate, or environmental conditions change, the system automatically adapts and generates optimized solutions without manual parameter readjustment, significantly shortening product changeover cycles and improving production flexibility and market responsiveness.
[0019] 5. Significantly improves production efficiency and reduces overall production costs. Through automated operations, manual intervention is reduced; self-optimizing algorithms improve product qualification rates; and model iteration shortens debugging time. The overall production efficiency of the system is improved compared to existing equipment. At the same time, precise parameter optimization can reduce raw material waste, reduce scrap rates and rework costs. Combined with full-process data storage and traceability functions, it facilitates production process control and problem investigation, bringing significant economic benefits and management efficiency to enterprises, and providing core guarantees for the quality upgrade of electronic equipment, sensors and other terminal products. Attached Figure Description
[0020] Figure 1 This is a block diagram of the overall system architecture of the present invention; Figure 2 This is a detailed block diagram of the hardware detection module of the present invention; Figure 3 This is a detailed block diagram of the software control system of the present invention; Figure 4 This is a detailed block diagram of the adjustment module of the present invention; Figure 5 This is a flowchart of the system workflow of the present invention. Detailed Implementation
[0021] Please refer to Figures 1 to 5 The purpose of this invention is to overcome the technical defects of existing winding equipment and provide an ultra-thin hollow coil winding system with self-learning optimization function. This system solves the problems of existing equipment relying on manual parameter adjustment, poor quality stability, low model training efficiency and insufficient generalization ability caused by high-dimensional data nonlinearity, time dependence, noise interference and uneven data acquisition. Through hardware and software co-design and core algorithm innovation, the system achieves fully automated and precise production and continuous optimization of ultra-thin hollow coils.
[0022] System components: The ultra-thin hollow coil winding system with self-learning optimization function of this invention includes a winding machine body, a hardware detection module, a software control system, and an execution adjustment module. The modules work together to realize the automation, intelligence, and self-optimization of coil winding. The specific structure is as follows: 1. Main body of the winding machine As the basic execution carrier for coil winding, it integrates an automatic feeding unit, a winding unit, a heating and shaping unit, a wire cutting unit, and a unloading unit. It can realize the fully automated operation from raw material feeding, coil winding, heating and shaping, wire cutting to finished product unloading, without the need for manual intervention in intermediate links, and provides a stable operating platform for parameter acquisition and optimization adjustment.
[0023] 2. Hardware detection module In conjunction with the main body of the winding machine, it is used to collect multi-dimensional parameters in real time and comprehensively during the production process, providing data support for model training and parameter optimization of the software control system. This includes physical parameter detection components and electrical parameter detection components, as detailed below: Physical parameter detection component: Composed of a vision sensor and a pressure sensor; the vision sensor is used to detect the outer dimensions (outer diameter, length, etc.), inner dimensions (inner diameter, etc.) and thickness of the coil in real time, with a detection accuracy of up to the micrometer level; the pressure sensor is used to detect the winding tightness of the coil, and obtains the tightness data of the coil after winding through contact pressure sensing.
[0024] Electrical parameter testing components: High-precision multimeter and inductance tester are used; the high-precision multimeter is used to accurately collect the resistance, current and voltage parameters of the coil, with a measurement error ≤ ±0.1%; the inductance tester is used to collect the inductance parameters of the coil to ensure the accuracy of electrical performance data.
[0025] In addition, the hardware detection module also collects the operating parameters of the winding machine simultaneously, including rotation speed, die spacing, tension, heating current intensity and power-on time, to achieve full-dimensional coverage of production process parameters.
[0026] 3. Software control system As the core control unit of the system, it is connected to the hardware detection module by signal. It has a built-in deep learning model that integrates attention mechanism and graph neural network, and is configured with data preprocessing unit, time-dependent modeling unit, data optimization and acquisition unit, accelerated training unit, data storage unit and model training unit. Its specific functions are as follows: Data preprocessing unit: To address the high-dimensionality and noise interference issues of the collected data, PCA (principal component analysis) or autoencoder is used to reduce the dimensionality of the high-dimensional data, simplifying the computational complexity of the model; at the same time, wavelet denoising and Kalman filtering techniques are combined to effectively suppress noise caused by environmental interference and improve the purity and reliability of the original data.
[0027] Time-dependent modeling unit: By constructing a time series model through LSTM (Long Short-Term Memory Network) or TCN (Temporal Convolutional Network), the model accurately captures the time dependence of coil parameters and winding machine operating parameters on the production process, thereby improving the model's adaptability to dynamic production scenarios.
[0028] Data optimization and acquisition unit: Adopts an adaptive sampling strategy to supplement the collection of scarce data samples in the production process; combined with data augmentation techniques (such as parameter perturbation, time series expansion, etc.) to expand the diversity of the dataset, solve the problem of uneven data collection in actual production, and provide a high-quality and comprehensive dataset for model training.
[0029] Accelerated Training Unit: Employing parallel DQN (Deep Q-Network) or distributed PPO (Proximal Policy Optimization) algorithms, the unit processes model training tasks in parallel, significantly improving training efficiency. Simultaneously, it incorporates a Priority Experience Replay (PER) mechanism, adjusting learning priorities based on the contribution of samples to model optimization, thereby accelerating model training convergence and shortening the optimization cycle.
[0030] Deep learning model: The core innovation lies in the integration of attention mechanism and graph neural network; the attention mechanism can automatically identify and focus on key sensor data (such as tension parameters that affect coil inductance, mold spacing parameters that affect dimensional accuracy, etc.), ignore secondary redundant information, reduce the computational load of the model and improve the optimization targeting; the graph neural network accurately captures the interaction relationship between physical parameters, electrical parameters and operating parameters by constructing a graph structure between parameters, further improving the model's generalization ability and prediction accuracy.
[0031] Data storage unit: Used to store various collected parameters (physical parameters, electrical parameters, operating parameters), model training data, and optimization and adjustment records, providing data support for model iteration and production process traceability.
[0032] Model training unit: Based on the optimized dataset, the deep learning model is continuously iterated and updated, enabling the model to learn new production rules and continuously improve the system's self-learning ability and optimization accuracy.
[0033] 4. Execute the adjustment module It is electrically connected to both the software control system and the main body of the winding machine, and is used to receive optimization and adjustment commands output by the software control system and drive the corresponding units of the main body of the winding machine to complete parameter adjustments, including speed adjustment components, spacing adjustment components, tension adjustment components, and heating adjustment components, as detailed below: Speed adjustment component: Adjusts the rotation speed of the winding machine to adapt to the winding density requirements of different coils; Spacing adjustment component: Adjusts the mold spacing to ensure the accuracy of the inner and outer dimensions of the coil; Tension adjustment component: Adjusts the tension during the winding process to prevent the coil from becoming loose or overstretched; Heating adjustment component: Adjusts the heating current intensity and energizing time to ensure the heating and shaping effect of the coil and improve structural stability.
[0034] Working principle: The workflow of the ultra-thin hollow coil winding system with self-learning optimization function of this invention is as follows: After the system is started, the main body of the winding machine automatically performs the entire process of feeding, winding, heating and shaping, cutting and unloading according to the preset initial parameters; Simultaneously, the hardware detection module uses components such as vision sensors, pressure sensors, high-precision multimeters, and inductance testers to collect the physical parameters, electrical parameters, and winding machine operating parameters of the coil in real time, and transmits the raw data to the software control system in real time. After receiving the raw data, the software control system first performs dimensionality reduction and noise suppression on the high-dimensional data through the data preprocessing unit to obtain clean and effective data; then, the time-dependent modeling unit constructs a time series model through the LSTM / TCN network to capture the dynamic correlation characteristics of parameters with the production process. To address the issue of uneven data collection, the data optimization collection unit supplements scarce samples through an adaptive sampling strategy and expands the dataset by combining data augmentation techniques, generating a high-quality optimized dataset. The accelerated training unit employs the parallel DQN / distributed PPO algorithm and PER mechanism to drive the deep learning model to train and converge quickly. The model uses an attention mechanism to filter key data and a graph neural network to capture the interaction between variables, accurately analyzes the mapping relationship between physical parameters, electrical parameters and operating parameters, and identifies the core operating parameters that affect the quality of the coil. The software control system generates targeted optimization and adjustment instructions based on the model analysis results and transmits them to the execution adjustment module; The adjustment module drives the corresponding units of the winding machine body to dynamically adjust the rotation speed, die spacing, tension, heating parameters, etc. through various adjustment components, so as to achieve adaptive optimization of production parameters; During the production process, the data storage unit continuously stores new production data and optimization records, and the model training unit continuously iterates and updates the deep learning model based on this data, so that the system's self-learning ability and optimization accuracy are continuously improved, ensuring that the product quality remains stable and meets the preset standards in long-term production.
Claims
1. A self-learning optimization system for winding ultrathin hollow coils, characterized in that, It includes the main body of the winding machine, a hardware detection module, a software control system, and an execution adjustment module; The winding machine integrates an automatic feeding unit, a winding unit, a heating and shaping unit, a wire cutting unit, and a unloading unit, realizing fully automated operation of the coil winding process; The hardware detection module is used to collect the physical parameters, electrical parameters, and winding machine operating parameters of the coil in real time. The software control system is connected to the hardware detection module by signal, and has a built-in deep learning model that integrates attention mechanism and graph neural network. It is configured with a data preprocessing unit, a time-dependent modeling unit, a data optimization acquisition unit and an accelerated training unit to analyze parameter mapping relationship and generate optimization adjustment instructions. The execution adjustment module is electrically connected to both the software control system and the main body of the winding machine. It is used to receive optimization adjustment commands and drive the corresponding unit of the main body of the winding machine to adjust the operating parameters, thereby achieving adaptive adjustment and continuous optimization.
2. The ultra-thin hollow coil winding system with self-learning optimization function according to claim 1, characterized in that, The hardware detection module includes a physical parameter detection component and an electrical parameter detection component. The physical parameter detection component includes a vision sensor and a pressure sensor. The vision sensor is used to detect the outer dimensions, inner dimensions, and thickness of the coil, and the pressure sensor is used to detect the winding tightness of the coil.
3. The ultra-thin hollow coil winding system with self-learning optimization function according to claim 1, characterized in that, The electrical parameter detection component uses a high-precision multimeter and an inductance tester. The high-precision multimeter is used to collect the resistance, current and voltage parameters of the coil, and the inductance tester is used to collect the inductance parameters of the coil.
4. The ultra-thin hollow coil winding system with self-learning optimization function according to claim 1, characterized in that, The operating parameters of the winding machine include rotational speed, die spacing, tension, heating current intensity and energizing time. The hardware detection module synchronously collects the above operating parameters and transmits them to the software control system.
5. The ultra-thin hollow coil winding system with self-learning optimization function according to claim 1, characterized in that, The data preprocessing unit uses PCA or an autoencoder to reduce the dimensionality of high-dimensional data, and combines wavelet denoising and Kalman filtering techniques to suppress data noise and improve the quality of the original data.
6. The ultrathin hollow coil winding system with self-learning optimization function according to claim 1, characterized in that, The time-dependent modeling unit constructs a time series model through an LSTM network or a TCN network to accurately capture the time dependency between coil parameters and winding machine operating parameters.
7. The ultra-thin hollow coil winding system with self-learning optimization function according to claim 1, characterized in that, The data optimization and acquisition unit employs an adaptive sampling strategy to supplement scarce data samples and combines it with data augmentation technology to expand the diversity of the dataset, thereby solving the problem of uneven data collection in actual production.
8. The ultrathin hollow coil winding system with self-learning optimization function according to claim 1, characterized in that, The accelerated training unit uses the parallel DQN algorithm or the distributed PPO algorithm to accelerate the model training process, and at the same time, it is equipped with a priority experience replay mechanism to optimize the policy learning efficiency and improve the model training convergence speed.
9. The ultrathin hollow coil winding system with self-learning optimization function according to claim 1, characterized in that, The execution adjustment module includes a speed adjustment component, a spacing adjustment component, a tension adjustment component, and a heating adjustment component, which respectively adjust the rotational speed of the winding machine, the die spacing, the winding tension, and the heating current intensity and energizing time.
10. The ultrathin hollow coil winding system with self-learning optimization function according to claim 1, characterized in that, The software control system also includes a data storage unit and a model training unit. The data storage unit is used to store various collected parameters and model training data. The model training unit continuously iterates and updates the deep learning model based on the optimized dataset to continuously improve the system's self-learning capability.