A method and device for precise control of temperature field and stress relief during annealing of glass beads

Abstract

The application discloses a kind of temperature field precision control and stress relief method and device in glass bead embryo annealing process, belong to glass heat treatment technical field;Method includes: obtaining bead embryo parameter matching annealing curve;Bead embryo is loaded on rotatable profiling bearing mechanism, and is placed into the annealing furnace with multiple partition independent heating module;In the heating and holding stage, drive bearing mechanism to rotate and make bead embryo heated surface dynamic change, based on multiple point temperature feedback independently adjust the power of each heating zone to maintain temperature field uniform;In cooling stage, according to the diameter of bead embryo, dynamically allocate the cooling rate of upper and lower heating zones, form asymmetric cooling curve to compensate the influence of gravity;Stress detection results are fed back to optimize annealing curve;Device includes furnace body, multiple partition independent heating module, rotatable profiling bearing mechanism and multiple point temperature acquisition module;The application realizes the precise control of annealing temperature field and the effective elimination of internal stress of glass bead embryo.

Description

Technical Field

[0001] This invention relates to the field of glass heat treatment technology, specifically a method and apparatus for precise temperature field control and stress relief during the annealing process of glass beads. Background Technology

[0002] The degree to which residual stress is eliminated in glass beads, especially spherical glass components used in optical instruments, precision grinding, shot peening and other fields, directly determines the impact resistance, optical uniformity and service life of the product. Annealing is a key process to eliminate internal stress in glass. The core of this process is to precisely control the temperature field distribution and change rate during heating, holding and cooling so that the viscous flow inside the glass is sufficient to relax the internal stress caused by uneven thermal history.

[0003] However, due to the small size, high sphericity, and high packing density of glass beads, temperature field control during annealing is more complex than that of flat glass. Existing technologies suffer from the following main technical defects in practical applications: The methods for controlling temperature field uniformity are crude; traditional annealing furnaces often employ single-point heating, and operators rely solely on empirical adjustments based on single thermocouple readings, leading to inconsistent heating histories within the same batch of glass bead preforms and significant dispersion in stress relief effects. Furthermore, the stress asymmetry problem during cooling is prominent; existing processes mostly involve overall cooling during the cooling stage, lacking differentiated control. Because the glass beads are in a static, stacked state, the upper and lower surfaces... Inconsistent surface cooling rates can easily generate asymmetric residual stress fields, i.e., stress birefringence, which seriously affects product strength. There is a lack of dynamic temperature control strategies for the spherical geometry. The heat conduction of glass beads is distributed in three dimensions with spherical symmetry, which is fundamentally different from the one-dimensional heat conduction of flat glass. Existing technologies follow the control logic of flat glass and ignore the rolling heating requirements and gravity compensation requirements of the beads. The process adaptability is insufficient when producing multiple specifications. When glass beads of different diameters or materials are processed in the same furnace, traditional processes cannot dynamically adjust the temperature control curve and often adopt conservative strategies, resulting in overheating of small beads or incomplete stress relief of large beads.

[0004] Furthermore, existing control algorithms mostly rely on PID or fuzzy logic, which essentially perform lag adjustment based on instantaneous temperature feedback. They cannot integrate multi-point temperature data to dynamically identify the spatial distribution of the temperature field, nor can they predict the evolution stage of internal stress in the bead blank. This lag in control logic causes control commands to always lag behind actual operating conditions, making it difficult to meet the stringent requirements of high-end glass beads for the consistency of residual stress.

[0005] Therefore, there is an urgent need for a method and apparatus that can achieve precise control of the annealing temperature field and symmetrical stress elimination based on the geometric characteristics of glass bead blanks, in order to solve the technical problems of low temperature field control accuracy, incomplete stress elimination, and control logic lagging behind changes in operating conditions in the existing technology. Summary of the Invention

[0006] The purpose of this invention is to overcome the shortcomings of existing technologies, such as coarse temperature field uniformity control, asymmetrical stress during cooling, lack of dynamic temperature control strategies for spherical geometry, and insufficient process adaptability when producing multiple specifications. This invention provides a method and apparatus for precise temperature field control and stress relief during the annealing process of glass beads. Through the coordinated control of multi-zone independent heating and a rotatable conformal bearing mechanism, intelligent identification of temperature field states based on a time-series depth network, compensation for gravity effects using asymmetrical cooling curves, and a multi-objective optimization feedback closed loop, this invention achieves a leap from experience-based annealing to data-driven intelligent annealing, solving the technical problems of incomplete stress relief and asymmetrical stress distribution within the glass beads.

[0007] The objective of this invention can be achieved through the following technical solutions: This application provides a method for precise temperature field control and stress relief during the annealing process of glass beads, including the following steps: The glass bead preform is loaded onto a rotatable conformal support mechanism inside the annealing furnace, which is equipped with a multi-zone independent heating module, the multi-zone independent heating module including multiple independently temperature-controlled heating zones; Temperature data from multiple temperature measurement points inside the annealing furnace are collected synchronously by a sensor acquisition device, and combined with the material and geometric parameters of the glass bead blank to form a comprehensive working condition dataset. The temperature data is fused using an intelligent control algorithm based on temporal deep networks to identify the temperature field distribution state inside the furnace and the stress relief stage of the glass bead preform. Based on the temperature field distribution and stress relief stage identification results, the intelligent control module calls the preset process parameter database data to generate the optimal power output command for each heating zone and the rotation command for the conformal bearing mechanism. The execution module receives power output commands from each heating zone and rotation commands from the contoured bearing mechanism, drives the multi-zone independent heating module to adjust the power of each heating zone, and simultaneously drives the contoured bearing mechanism to rotate at a preset speed, so that the heated surface of the glass bead blank changes dynamically, and performs coordinated adjustment of precise temperature field control and stress relief. The feedback optimization module collects residual stress detection data, system energy consumption data, and furnace temperature fluctuation data in real time, compares them with preset target values, calculates deviation values, and iteratively corrects the power output commands of each heating zone and the rotation commands of the conformal bearing mechanism through optimization algorithms to form closed-loop control.

[0008] Furthermore, after forming the comprehensive operating condition dataset, it also includes: Preprocess the key variables in the comprehensive working condition dataset; The spatial distribution features, temporal gradient change features, and multi-point temperature difference features of the preprocessed data are extracted, and the extracted features are normalized to remove outliers. The extracted features are constructed into operating condition feature vectors and associated with annealing performance data in historical production records to form a structured operating condition feature dataset. The structured operating condition feature dataset is derived from the comprehensive operating condition dataset. The structured operating condition feature dataset is specifically used for decision-making and optimization of intelligent control algorithms.

[0009] Furthermore, the intelligent control algorithm based on temporal deep networks is used to fuse the temperature data and identify the temperature field distribution state inside the furnace and the stress relief stage of the glass bead preform, specifically including: A temporal deep network model based on an encoder-decoder architecture is constructed as the core model of the intelligent control algorithm; wherein, the encoder is used to fuse and extract features from the temporal sequence of the temperature data, and the decoder is used to output the identification results of the temperature field distribution state and stress relief stage; The encoder's data fusion and feature extraction include: First, the temperature time series data from multiple temperature measurement points, which have undergone timestamp alignment and normalization preprocessing, are spliced ​​together along the channel dimension to form a multi-channel time series feature tensor, thus achieving the initial aggregation of multi-source data. Then, a one-dimensional convolutional neural network layer is used to perform convolution operations on the multi-channel temporal feature tensor. By sliding the convolution kernel in the time dimension, the local temporal features of different channels are fused across channels to extract the spatial correlation features between each temperature measurement point and obtain a convolutional feature sequence that incorporates spatial information. Finally, the convolutional feature sequence is input into a bidirectional gated recurrent unit network. The forward GRU and backward GRU capture the historical dependencies and future trends of the time series data respectively. The features are deeply fused in the time dimension to explore the forward and backward dynamic dependencies of temperature changes in time and output a deep time series feature vector that integrates spatial correlation and time series dependency. The state recognition of the decoder includes: introducing an attention mechanism into the deep temporal feature vector output by the bidirectional gated recurrent unit network, calculating the weights of different time steps, and focusing on the most critical temporal segments for judging the current state; based on the weighted feature vector, simultaneously outputting the temperature field distribution state and the stress relief stage recognition results of the glass bead preform through a fully connected classification network.

[0010] Preferably, the generation of optimal power output commands for each heating zone and rotation commands for the conformal bearing mechanism specifically includes: The intelligent control module calls the preset process parameter database, which stores the historical optimal process parameters under different combinations of temperature field distribution states and stress relief stages. Based on the temperature field distribution and stress relief stage identification results, a similarity matching algorithm is used to match the process parameters of similar working conditions in the database. Combined with the real-time equipment operation stability and target residual stress requirements, the parameters are adjusted through a multi-objective optimization algorithm to determine the preliminary power output values ​​of each heating zone and the rotation speed of the conformal bearing mechanism. The initial power output values ​​of each heating zone and the rotation speed of the conforming bearing mechanism are integrated with the stress concentration prevention strategy based on the temperature field uniformity requirement. The safety deviation range of the power output of each heating zone is set according to the temperature field distribution. If the initial power output value exceeds the corresponding safety deviation range, the initial power output value is limited to the boundary value, and the rotation speed of the conforming bearing mechanism is adjusted synchronously to generate a coordinated control command for temperature field control and stress relief.

[0011] Preferably, the driving multi-zone independent heating module adjusts the power of each heating zone to maintain a uniform temperature field, while simultaneously driving the contour-following bearing mechanism to rotate at a preset speed, performing coordinated adjustment of precise temperature field control and stress relief, specifically including: The multi-zone independent heating module in the execution module converts the optimal power output command into control signals for the power controllers of each heating zone, driving the upper heating zone, lower heating zone and side heating zone to operate at the set power, and monitoring the actual power and temperature feedback of each heating zone in real time. The contour-following bearing mechanism drives the servo motor according to the rotation command, controls the tray to rotate intermittently at a set speed, monitors the tray load and rotation stability data in real time, and performs coordinated adjustment of the heating and cooling processes.

[0012] Preferably, the feedback optimization module collects residual stress detection data, system energy consumption data, and furnace temperature fluctuation data in real time, compares them with preset target values, calculates deviation values, and iteratively corrects the power output commands of each heating zone and the rotation commands of the conformal bearing mechanism through an optimization algorithm to form a closed-loop control, specifically including: The residual stress data of the annealed glass bead preform is collected by an online stress detector, the system energy consumption data is collected by a power sensor, and the temperature fluctuation data of each temperature measuring point in the furnace is collected by a multi-point temperature acquisition module. The calculation unit compares the collected residual stress data, system energy consumption data, and multi-point temperature fluctuation data with the preset target value determined based on product standards and historical best production data to calculate the deviation value. The calculation process of the deviation value is as follows: the absolute deviation between the measured residual stress value and the target value, the absolute deviation between the measured system energy consumption value and the target value, and the root mean square error between the measured temperature value and the set value at each temperature measurement point are calculated as temperature fluctuation deviations. Then, the three deviation values ​​are weighted and fused according to preset weight coefficients to obtain a comprehensive deviation evaluation function. An optimization algorithm is used to iteratively optimize the power output command of each heating zone and the rotation command of the conformal bearing mechanism in the next control cycle with the goal of minimizing the comprehensive deviation evaluation function. The optimal command obtained by optimization is then updated to the process parameter database of the intelligent control module for subsequent batch temperature field control and stress relief decisions.

[0013] Specifically, an optimization algorithm is employed to iteratively optimize the power output commands for each heating zone and the rotation commands for the conformal bearing mechanism in the next control cycle, with the goal of minimizing the comprehensive deviation evaluation function. This includes: Initialize the particle swarm by constructing a multi-dimensional search space using the current power output commands of each heating zone and the rotation commands of the contour bearing mechanism, and initialize the position and velocity of each particle in the particle swarm. Then, iterative optimization is performed, mapping the position coordinates of each particle in the current iteration to a set of candidate power output values ​​of each heating zone and rotational speed of the conformal bearing mechanism. The fitness value of the particle is calculated by substituting the deviation value into the comprehensive deviation evaluation function constructed by the deviation value. Based on the particle's current velocity, individual optimal position and global optimal position, the position and velocity of the next generation of particles are calculated according to the velocity and position update formula of the particle swarm optimization algorithm. When the preset termination condition is reached, the iteration stops, and the power output value of each heating zone corresponding to the global optimal position and the rotation speed of the conformal bearing mechanism are taken as the optimal command output for this optimization.

[0014] Furthermore, a device for precise temperature field control and stress relief during the annealing process of glass beads to achieve the above method includes a horizontal cylindrical furnace body, a multi-zone independent heating module, a rotatable contour-following bearing mechanism, a multi-point temperature acquisition module, a control module, and a hot air circulation disturbance module. The multi-zone independent heating module is installed on the inner wall of the furnace body and is divided into an upper heating zone, a lower heating zone and a side heating zone along the circumference of the furnace. The upper heating zone is installed on the inner wall of the top of the furnace, the lower heating zone is installed on the inner wall of the bottom of the furnace, and the side heating zone is installed on the inner walls of both sides of the furnace. Each heating zone is connected to an independent power controller. The rotatable contour-following support mechanism is located at the center of the bottom of the furnace body and includes a tray and a servo motor. The servo motor is installed on the outside of the furnace body and is used to drive the tray to rotate intermittently at a preset speed. The output shaft of the servo motor extends into the furnace chamber through a sealed bearing and is connected to the bottom of the tray for transmission. The surface of the tray is evenly distributed with arc-shaped grooves for accommodating glass bead blanks. The radius of curvature of the arc-shaped grooves is greater than the radius of the glass bead blanks, so that the glass bead blanks and the arc-shaped grooves are in point contact or line contact. The multi-point temperature acquisition module is installed on the furnace body and includes multiple thermocouples arranged in a matrix at different heights in the top, middle and bottom of the furnace and at different depths in the center and edge, for synchronously acquiring temperature data from multiple temperature measurement points in the furnace. The hot air circulation disturbance module is installed inside the furnace body and includes two sets of micro hot air circulation fans that are evenly spaced along the length of the furnace body. Each set of fans includes two fans installed on the inner walls of both sides of the furnace chamber and a guide plate installed on the side of the fan facing the center of the furnace chamber. The guide plate is inclined towards the center of the furnace chamber. The control module is located outside the furnace body and integrated into a control cabinet. The control cabinet contains a database unit, a temperature acquisition unit, an algorithm control unit, a drive control unit, and a stress feedback unit, all connected via an internal bus. The database unit stores the basic annealing temperature curves and corresponding residual stress thresholds for glass beads of different materials and diameters. The temperature acquisition unit is connected to a multi-point temperature acquisition module to acquire temperature data at various points within the furnace in real time. The algorithm control unit is connected to the database unit and the temperature acquisition unit to calculate the output power of each heating zone based on the temperature data using a decoupled control algorithm, and to calculate the asymmetric cooling curve based on the glass bead diameter. The drive control unit is connected to the algorithm control unit and the servo motor to control the start, stop, and speed of the servo motor. The stress feedback unit receives residual stress detection results and feeds them back to the database unit to optimize the basic annealing temperature curve.

[0015] Specifically, the furnace body is divided into a preheating zone, a heat preservation zone, a slow cooling zone, and a fast cooling zone along the axial direction, and each zone is independently equipped with the multi-zone independent heating module.

[0016] Furthermore, the device also includes an online stress detection module, which is located at the discharge port at the end of the rapid cooling zone of the furnace body. The online stress detection module is used to detect the residual stress of the glass bead preform after annealing in real time and transmit the detection results to the stress feedback unit.

[0017] The beneficial effects of this invention are as follows: This invention, through synergistic innovation of device and method, employs multi-zone independent heating and a rotatable contour-following support mechanism, combined with multi-point temperature acquisition and intelligent algorithms, to achieve uniform temperature distribution in the furnace and dynamic changes in the heated surface of the bead blanks, solving the problems of large temperature differences and uneven heating in traditional annealing. During the cooling stage, the cooling rate is dynamically allocated according to the bead blank diameter to form an asymmetric cooling curve, accurately compensating for the influence of gravity, eliminating the asymmetric residual stress field, and avoiding stress birefringence. Simultaneously, a time-series deep network is used to identify the temperature field state and stress elimination stage in real time, providing a decision-making basis for process optimization. Furthermore, an adaptive closed-loop control is formed by combining residual stress, energy consumption, and other data feedback through a particle swarm optimization algorithm. In addition, a pre-set process parameter database achieves precise adaptation for mixed production of multiple specifications within the same furnace through similarity matching and multi-objective optimization, significantly improving production efficiency and product yield. Attached Figure Description

[0018] To better understand and implement this application, the technical solution is described in detail below with reference to the accompanying drawings.

[0019] Figure 1 A schematic flowchart illustrating the method for precise control of the annealing temperature field of glass bead preforms provided in an embodiment of the present invention; Figure 2 is a schematic diagram of the workflow of the temporal deep network model provided in the embodiment of the present invention; Figure 3 This is a schematic diagram of the overall structure of the glass bead annealing apparatus provided in an embodiment of the present invention; In the diagram: 1-Horizontal cylindrical furnace body, 2-Feed inlet, 3-Automatic furnace door, 4-Discharge outlet, 5-Preheating zone, 6-Insulation zone, 7-Slow cooling zone, 8-Fast cooling zone, 9-Multi-zone independent heating module, 901-Upper heating zone, 902-Lower heating zone, 903-Side heating zone, 10-Rotating contour-following support mechanism, 101-Tray, 102-Servo motor, 11-Thermocouple, 12-Miniature hot air circulating fan, 121-Guide plate, 13-Control module. Detailed Implementation

[0020] To further illustrate the technical means and effects adopted by the present invention to achieve its intended purpose, exemplary embodiments will be described in detail below, examples of which are illustrated in the accompanying drawings. In the following description, when referring to the drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this application.

[0021] The terminology used in this application is for the purpose of describing particular embodiments only and is not intended to be limiting of the application. The singular forms “a,” “the,” and “the” used herein are also intended to include the plural forms unless the context clearly indicates otherwise. It should also be understood that the term “and / or” as used herein refers to and includes any and all possible combinations of one or more of the associated listed items.

[0022] The following detailed description of the specific implementation methods, features, and effects of the present invention, in conjunction with the accompanying drawings and preferred embodiments, is provided in detail.

[0023] Example 1 Please see Figures 1-2 This embodiment provides a method for precise temperature field control and stress relief during the annealing process of glass beads, including the following steps: S1. The glass bead preform is loaded onto a rotatable conformal support mechanism inside the annealing furnace. The annealing furnace is equipped with a multi-zone independent heating module 9, which includes multiple independently temperature-controlled heating zones. First, based on the material and geometric parameters of the glass bead blank to be processed, the corresponding basic annealing temperature curve is matched from the database unit of the control module 13 as the initial reference curve for this annealing process; the material parameters include the glass expansion coefficient, softening point temperature and viscosity-temperature characteristics, and the geometric parameters include the bead blank diameter and roundness tolerance. Subsequently, the annealing furnace door is opened, and the glass bead blanks are evenly placed on the tray 101 of the rotatable contour-following support mechanism. The surface of the tray 101 is evenly distributed with arc-shaped grooves. The radius of curvature of the arc-shaped grooves is greater than the radius of the glass bead blanks, so that the glass bead blanks and the arc-shaped grooves are in point contact or line contact. This ensures that the bead blanks are stably placed on the tray 101 while minimizing the interference of the contact area on heat conduction. Then, the furnace door is closed and the hot air circulation disturbance module is activated to create forced convection inside the furnace and preheat the furnace to the initial set temperature. At the same time, the temperature data at different heights at the top, middle and bottom of the furnace and at different depths at the center and edge are monitored in real time through the multi-point temperature acquisition module, providing an initial temperature field distribution benchmark for subsequent steps. This step, through the synergistic cooperation of the point-contact loading method of the rotatable contour-following bearing mechanism and the hot air circulation preheating, creates a uniform and stable initial temperature field environment for the glass bead preform annealing process, and lays the technological foundation for subsequent multi-zone independent heating and dynamic rotation control.

[0024] S2. Temperature data from multiple temperature measurement points inside the annealing furnace are collected synchronously through a sensor acquisition device, and combined with the material and geometric parameters of the glass bead blank to form a comprehensive working condition dataset. First, taking advantage of the initial temperature field environment established by S1, the multi-point temperature acquisition module is started. Multiple thermocouples 11 arranged in a matrix at different heights in the top, middle and bottom of the furnace and at different depths in the center and edge are used to synchronously acquire real-time temperature data of each temperature measurement point at a preset sampling frequency, forming the original temperature time sequence. Subsequently, the material and geometric parameters of the current batch of glass beads are retrieved from the database unit of the control module 13, including the glass expansion coefficient, softening point temperature, viscosity-temperature characteristics, bead diameter and roundness tolerance. These static parameters are aligned and correlated with the collected dynamic temperature data on the timestamp to form a comprehensive working condition dataset containing multi-source information. Then, after forming the comprehensive working condition dataset, the key variables in the comprehensive working condition dataset are preprocessed: first, the spatial distribution characteristics, temporal gradient change characteristics and multi-point temperature difference characteristics of the temperature data are extracted; then, the extracted features are normalized, and the minimum and maximum normalization methods are used to map each feature value to the [0,1] interval, while identifying and removing 0 values ​​or outliers that are significantly deviated from the normal range due to sensor anomalies, to ensure data quality; Subsequently, a working condition feature vector is constructed: the normalized spatial distribution features, temporal gradient features, multi-point difference features are spliced ​​and fused with the bead blank material parameters and geometric parameters to form a multi-dimensional feature vector representing the current working condition. Finally, the constructed operating condition feature vector is associated with the annealing performance data obtained from historical production records. The annealing performance data includes residual stress detection values, energy consumption data, and corresponding process parameter combinations after annealing of historical bead blanks of the same specification. Through a similarity matching algorithm, the current operating condition feature vector is compared with the operating condition features in the historical database to filter out performance data under similar operating conditions. This data, together with the current feature vector, generates a structured operating condition feature dataset. The structured operating condition feature dataset and the comprehensive operating condition dataset are derived from each other, with the former being used specifically for decision-making and optimization of subsequent intelligent control algorithms.

[0025] This step transforms the raw sensor data into a structured chemical condition feature dataset containing historical experience information through multi-source data fusion and feature engineering, realizing a digital representation of the annealing process state and providing high-quality data support for subsequent temperature field state identification and intelligent decision-making.

[0026] S3. The temperature data is fused using an intelligent control algorithm based on a time-series deep network to identify the temperature field distribution state in the furnace and the stress relief stage of the glass bead blank. First, a temporal deep network model based on an encoder-decoder architecture is constructed as the intelligent control model. The specific construction process is as follows: The temporal deep network model adopts an encoder-decoder architecture. The encoder is responsible for feature extraction and fusion of the input temperature time series data, while the decoder is responsible for identifying the temperature field distribution state and stress relief stage based on the feature vector output by the encoder. The encoder and decoder are connected through an attention mechanism, enabling the decoder to focus on the most critical time series segment for judging the current state. The model adopts an end-to-end training method, using temperature time series data from historical production data as input and labeled temperature field distribution state and stress relief stage labels as supervision signals. The network parameters are optimized through a backpropagation algorithm. The data fusion and feature extraction process of the encoder specifically includes: First, real-time temperature time-series data is obtained from the structured chemical condition feature dataset generated in step S2. The temperature time-series data from multiple temperature measurement points, after timestamp alignment and normalization preprocessing, are concatenated along the channel dimension to form a multi-channel time-series feature tensor. ,in Indicates the length of the time series. This indicates the number of temperature measurement point channels, enabling preliminary aggregation of multi-source data. Then, a one-dimensional convolutional neural network layer is used to perform convolution operations on the multi-channel temporal feature tensor. By sliding the convolution kernel in the time dimension, cross-channel fusion of local temporal features from different channels is performed to extract spatial correlation features between temperature measurement points; specifically, setting... Each size is The convolution kernel has a stride of 1 and padding mode of "same". After the convolution operation, the convolutional feature sequence is obtained. ,in Indicates the number of convolution kernels. A convolutional feature sequence that incorporates spatial information; Finally, the convolutional feature sequence H is input into a bidirectional gated recurrent unit network. The forward GRU captures the historical dependencies of the temporal data in forward chronological order, while the backward GRU captures future trends in reverse chronological order, thus performing deep fusion of features in the temporal dimension. The outputs of the forward GRU and the backward GRU are concatenated at each time step to obtain a deep temporal feature vector. ,in Indicates the number of GRU hidden units. To create a deep temporal feature vector that integrates spatial correlation and temporal dependence; The state recognition process of the decoder specifically includes: the depth temporal feature vector output by the encoder. An attention mechanism is introduced, first calculating the attention weights at each time step. ,in For a trainable weight matrix, This is the output of the encoder at the last time step. For the first The bidirectional GRU outputs at each time step are used; then, the features at each time step are weighted and summed according to the attention weights to obtain the context vector. ; context vector With encoder global features By concatenating the features, a comprehensive feature vector is obtained. ; Combine feature vectors The input is a fully connected classification network, which contains two parallel fully connected output layers: one for outputting the temperature field distribution state and the other for outputting the stress relief stage. The temperature field distribution state output layer uses a softmax activation function, and the output dimension corresponds to a preset number of uniformity levels. The stress relief stage output layer uses a softmax activation function, and the output dimension corresponds to a preset number of stages. The category with the highest probability between the two output layers is taken as the final recognition result. Temperature time-series data from multiple batches are collected from the historical production records generated in step S2. Process experts annotate the data, labeling each time window with the corresponding temperature field distribution and stress relief stage, thus forming a training dataset. The cross-entropy loss function is used as the optimization objective, and the Adam optimizer is used to iteratively train the network parameters of the time-series deep network model until the model's recognition accuracy on the validation set converges. The trained model is then deployed to the intelligent control module 13 for real-time processing of the online temperature data. The preprocessed real-time temperature data is input into the trained temporal deep network model, and the decoder directly outputs the identification results of the temperature field distribution state and the stress relief stage. This step constructs a temporal deep network with an encoder and decoder architecture, utilizes one-dimensional convolution to extract local spatial features, bidirectional gated recurrent units to mine temporal dependencies, and attention mechanisms to focus on key temporal segments, thereby achieving real-time and accurate identification of the temperature field distribution state in the furnace and the stress relief stage of the glass bead preform. This provides a decision-making basis for the intelligent optimization of subsequent process parameters.

[0027] S4. Based on the temperature field distribution and stress relief stage identification results, the intelligent control module 13 calls the preset process parameter database data to generate the optimal power output command for each heating zone and the rotation command for the conformal bearing mechanism. First, the intelligent control module 13 receives the temperature field distribution state identification result and the stress relief stage identification result output from step S3. These two identification results will be used as key indicators of the current operating condition; The process of matching process parameters for similar operating conditions in the database using a similarity matching algorithm specifically includes: the intelligent control module 13 accessing the preset process parameter database. The database stores historical optimal process parameters under different combinations of temperature field distribution and stress relief stages, and each historical record includes an operating condition identifier. Corresponding process parameter combinations And the efficiency indicators of this production; the current identification results As a query condition, it is matched against the working condition identifiers of each historical record in the database, prioritizing matches of historical working conditions that are exactly the same; if no exact match is found, a similarity matching algorithm is used to calculate the similarity between the current working condition and each historical working condition: ;in This is a rank difference function; for the same rank, take... One level apart Two levels apart If the difference is more than three levels, take , and The weighting coefficients and ; Filter out the top with the highest similarity Each historical record will be used to select the corresponding combination of process parameters as a candidate parameter set. ; Combining real-time equipment operational stability with target residual stress requirements, a multi-objective optimization algorithm is used to adjust parameters and determine the initial power output values ​​for each heating zone and the rotational speed of the conformal bearing mechanism. This process specifically includes: using a candidate parameter set... As the initial solution space, a multi-objective optimization problem is constructed, with optimization objectives including: , , ;in The residual stress value is predicted based on similar historical data. To meet the target residual stress requirements of this production, The historical standard deviation of the heating zone power. To predict the temperature field distribution based on similar historical data, a non-dominated sorting genetic algorithm with an elitist strategy is employed for multi-objective optimization. The population is initialized with a set of candidate parameters and combinations of parameters generated by minor mutations. Iterative evolution is performed through selection, crossover, and mutation operations. In each generation, the three objective function values ​​for each individual in the population are calculated. Non-dominated sorting and crowding distance calculations are performed to select the optimal solution set on the Pareto front. From the Pareto optimal solution set, feasibility is screened based on real-time equipment operational stability constraints, selecting solutions that meet the constraints. Then, based on the priority of the target residual stress requirements, the solution with the smallest deviation from the target residual stress is selected as the initial power output value for each heating zone. Rotation speed of the conformal bearing mechanism ; The process of integrating stress concentration prevention strategies based on temperature field uniformity requirements specifically includes: obtaining the current temperature field distribution state identified in step S3. Based on the uniformity level, the safe deviation range of power output for each heating zone is set. When the uniformity level is excellent, the allowable deviation range is... When the grade is "Good", the allowable deviation range is: When the grade is medium, the allowable deviation range is When the grade is poor, the allowable deviation range is: The power output values ​​of each heating zone were initially determined. Compare the power value of each heating zone with the corresponding safety deviation range, and check whether the power value exceeds the allowable range; if the first Preliminary power output value of each heating zone Exceeding the safety deviation range Then, it is restricted to the nearest boundary value to obtain the corrected power value: Simultaneously, the rotational speed of the contouring bearing mechanism is adjusted synchronously according to the magnitude of the power correction. If the correction magnitude exceeds the set threshold... If the rotation speed is too low, the residence time of the bead blank in that temperature zone can be extended. The rotation speed adjustment formula is as follows: ;in The adjustment coefficient is used to adjust the power output value of each heating zone. With the adjusted rotational speed of the contour-following load-bearing mechanism This process integrates commands to generate coordinated control instructions that ensure precise temperature field control and stress relief. The complete control scheme is output to the execution module; This step uses a similarity matching algorithm to filter process parameters with similar operating conditions from a historical database, combines a multi-objective optimization algorithm to balance residual stress requirements with equipment operating stability, and then uses an anti-stress concentration strategy to perform safety constraints on power output and coordinated speed adjustment based on the temperature field distribution. This achieves intelligent decision-making and precise generation from identification results to coordinated control commands.

[0028] S5. The execution module receives the power output command of each heating zone and the rotation command of the conforming bearing mechanism, drives the multi-zone independent heating module 9 to adjust the power of each heating zone, and drives the conforming bearing mechanism to rotate at a preset speed, so that the heated surface of the glass bead blank changes dynamically, and performs coordinated adjustment of precise temperature field control and stress elimination. First, the execution module receives the coordinated control instructions generated in step S4. The instructions include the target power value for each heating zone and the target rotation speed of the contour-following bearing mechanism; The power adjustment process of the multi-zone independent heating module 9 specifically includes: the execution module sending the target power value of each heating zone to the corresponding power controller; after receiving the target power value, each power controller first reads the actual power feedback value of the current heating zone and calculates the power deviation; a PID control algorithm is used to perform closed-loop power adjustment, and the proportional, integral, and derivative coefficients of the controller are pre-tuned according to the thermal response characteristics of the heating zone; the controller output is converted into the firing angle of the thyristor or the duty cycle of the solid-state relay to adjust the power of the heating element so that the actual power gradually approaches the target power value; during the adjustment process, the power controller monitors the deviation between the actual power and the target power in real time, and when the absolute value of the deviation is less than the set threshold and remains stable for more than the set time, it is determined that the power adjustment of the heating zone is completed; at the same time, the actual power and temperature feedback data of each heating zone are uploaded to the control module 13 in real time for instruction optimization in the next cycle; The rotation drive process of the contour-following bearing mechanism specifically includes: the execution module sending the target rotation speed to the servo motor 102 driver; the servo motor 102 driver calculating the corresponding control pulse frequency based on the target rotation speed and using a vector control algorithm to drive the servo motor 102 to rotate at the set rotation speed; the servo motor 102 driving the tray 101 to rotate through the sealed bearing, causing the glass bead blanks loaded in the arc-shaped groove on the surface of the tray 101 to rotate together with the tray 101; during the rotation process, the encoder built into the servo motor 102 provides real-time feedback of the actual rotation speed to the driver, and the driver uses a PID algorithm to perform closed-loop adjustment of the rotation speed to ensure that the deviation between the actual rotation speed and the target rotation speed is controlled within ±1%; at the same time, the rotation mode is set to intermittent rotation, that is, it operates in a cycle mode of rotation, pause, rotation, and the ratio of rotation time to pause time in each rotation cycle is dynamically adjusted according to the diameter of the bead blank. The larger the diameter of the bead blank, the greater the proportion of pause time is appropriately increased to ensure that all parts of the bead blank are heated evenly; the servo motor 102 driver also monitors the load change of the tray 101 in real time, and when the load fluctuation is detected to exceed the set threshold, it automatically adjusts the output torque to maintain rotational stability; The coordinated adjustment process of heating and rotation specifically includes: during the heating and heat preservation stage, the execution module simultaneously drives the heating module and the supporting mechanism to work together in the manner described above; the heating module outputs power according to the target, causing the furnace temperature to gradually rise to the set value; the supporting mechanism rotates at the target speed, causing the heated surface of the glass beads to change dynamically, avoiding local overheating; during the rotation process, the beads are in point contact or line contact within the arc-shaped groove, slowly rolling as the tray 101 rotates, continuously changing their relative position with the heat source, achieving uniform heating in three-dimensional space; during the cooling stage, the execution module, according to S... The cooling curve command generated in the 4 steps dynamically adjusts the power output of each heating zone. The upper heating zone 901 and the lower heating zone 902 are adjusted according to different cooling rates to form asymmetrical cooling. At the same time, the supporting mechanism continues to rotate at the set speed to ensure that the temperature gradient of each part of the bead blank is uniform during the cooling process and to compensate for the influence of gravity. Throughout the heating and cooling process, the execution module continuously collects the actual power, actual temperature, actual rotation speed of tray 101 and load data of each heating zone and compares them with the target command. If any deviation or abnormality occurs, an alarm is immediately triggered and the system automatically switches to the safety protection mode. This step achieves precise adjustment of the heating zone power through PID closed-loop control algorithm, stable control of the bearing mechanism speed through servo motor 102 vector control and encoder feedback, and through the coordinated cooperation of heating and rotation, the heated surface of the glass bead preform changes dynamically during the heating stage and cools down uniformly during the cooling stage, thus achieving coordinated adjustment of precise temperature field control and stress relief.

[0029] S6. The residual stress detection data of the glass bead blank, system energy consumption data and furnace temperature fluctuation data are collected in real time through the feedback optimization module. The data are compared with the preset target value, the deviation value is calculated, and the power output command of each heating zone and the rotation command of the conformal bearing mechanism are iteratively corrected through the optimization algorithm to form a closed loop control.

[0030] To ensure the real-time control of the annealing process, the particle swarm optimization algorithm described in this embodiment adopts a two-layer architecture that separates offline optimization and online control: After each batch of annealing is completed, the control system uses the residual stress, energy consumption, and temperature fluctuation data collected for that batch to start the particle swarm optimization algorithm for offline iterative optimization during the preparation of the next batch of loading; the optimal instructions obtained by optimization are updated to the process parameter database for online control calls in subsequent batches; the online control layer directly calls the optimal parameters in the database and realizes real-time adjustment of heating power and rotation speed through PID controller, with a single control cycle of less than 100 milliseconds, which fully meets the real-time requirements of industrial annealing furnaces; First, after each batch of glass bead preforms has been annealed, the feedback optimization module is activated, and the measured residual stress of the annealed glass bead preforms is collected using an online stress detector. The measured system energy consumption during this annealing process was collected using a power sensor. The temperature time sequence data of each temperature measuring point in the furnace during the entire annealing process is collected by the multi-point temperature acquisition module, and the temperature fluctuation deviation between the measured temperature value and the set value of each temperature measuring point is calculated. The calculation process for the deviation value is as follows: calculate the residual stress deviation separately. ,in The residual stress target value is determined based on the product standard; system energy consumption deviation. ,in The target energy consumption per unit is determined based on historical best production data; temperature fluctuation deviation. ,in Number of temperature measurement points For the first The measured temperature at each temperature measuring point The set temperature at the temperature measurement point at the corresponding time; the three deviation values ​​are weighted and fused to obtain the comprehensive deviation evaluation function. ,in The weighting coefficients and The weighting coefficients are dynamically adjusted based on product quality requirements and production cost requirements; The particle swarm optimization algorithm is used to iteratively optimize the power output commands of each heating zone and the rotation commands of the contour-following bearing mechanism, specifically including: The process of initializing the position and velocity of each particle in a particle swarm includes: setting the particle swarm size to... The position vector of each particle This represents a set of candidate heating zone power output values ​​and the rotational speed of the contour-following load-bearing mechanism, where... The number of heating zones is determined by the actual power output value of each heating zone in the current control cycle and the rotational speed of the conformal bearing mechanism, which are used as a reference. Random disturbances are then applied in the vicinity of these values ​​to generate... The initial positions of the particles are determined to ensure uniform distribution of particles in the solution space; the velocity vector of each particle... Initialize to a small range of random values, with the velocity range set to 10%-20% of the position range; initialize the individual optimal position for each particle. It calculates the fitness value of each particle and selects the position of the particle with the lowest fitness value as the global optimal position. ; The process of mapping the position coordinates of each particle in the current iteration to a set of candidate power output values ​​for each heating zone and rotational speed of the conformal bearing mechanism, and substituting these values ​​into a comprehensive deviation evaluation function to calculate the fitness value of the particle, specifically includes: in the... In each iteration, each particle position vector This is analyzed as a specific combination of process parameters, namely the power values ​​of each heating zone. and the rotational speed of the load-bearing mechanism Input this set of process parameters into a process simulation model or a prediction model based on historical data to predict the residual stress values ​​that may be generated if this set of parameters is used for annealing production. System energy consumption value and temperature fluctuation value ; Calculate the fitness value of the particle based on the predicted value. A smaller fitness value indicates better process parameters; The process of calculating the position and velocity of the next generation of particles according to the velocity and position update formula of the particle swarm optimization algorithm specifically includes: based on the... Particle velocity in the next iteration Optimal position of individual particles and the global optimal position Calculate the first Speed ​​of iteration: in Inertial weights control the degree to which a particle maintains its current velocity; These are learning factors, which control the degree to which particles learn towards their individual optimality and global optimality, respectively. for A random number vector within the interval; calculate the first [number] based on the updated velocity. Position of the next iteration: After the update, check if the position exceeds the preset power and speed range. If it does, limit it to the boundary values ​​and set the speed in the corresponding direction to zero or reverse it; calculate the fitness value of the new position. Then update the individual's optimal position. Otherwise, maintain the original individual's optimal position; after all particles have been updated, find the particle with the smallest fitness value in the current iteration, if the fitness value is less than Then update the global optimal position. ; Iteration termination judgment and output: Repeat the above iteration process until the preset maximum number of iterations is reached. , or continuous The improvement in the global optimal fitness value is less than the set threshold. The iteration terminates when the time is right; the final global optimal position is obtained. The power output values ​​of each heating zone and the rotation speed of the contour bearing mechanism are used as the optimal command output for this optimization. The optimized instructions are updated to the process parameter database of the intelligent control module 13 for temperature field control and stress relief decision-making in the next control cycle, forming an adaptive closed-loop control. This step involves real-time acquisition of residual stress, system energy consumption, and temperature fluctuation data after annealing, calculation of a comprehensive deviation evaluation function, and iterative optimization of the power output of each heating zone and the rotational speed of the supporting mechanism using a particle swarm optimization algorithm. The optimized instructions are then updated to the database, achieving adaptive closed-loop control based on actual production feedback, which enables continuous optimization of process parameters as production data accumulates.

[0031] Example 2 Please see Figure 3 This embodiment provides a device for precise temperature field control and stress relief during the annealing process of glass beads, which is applied to the method for precise temperature field control and stress relief during the annealing process of glass beads described in Embodiment 1. The device includes a horizontal cylindrical furnace body 1, a multi-zone independent heating module 9, a rotatable contour-following bearing mechanism 10, a multi-point temperature acquisition module, a control module 13, and a hot air circulation disturbance module. The horizontal cylindrical furnace body 1 is divided into a preheating zone 5, a heat preservation zone 6, a slow cooling zone 7, and a fast cooling zone 8 along the axial direction. Heat insulation partitions are installed between each zone to reduce thermal interference between adjacent zones. A feed inlet 2 and a discharge outlet 4 are respectively provided at both ends of the furnace body. An automatic furnace door 3 is installed at the feed inlet 2. The rotatable contour-following bearing mechanism 10 runs through the entire preheating zone 5, heat preservation zone 6, slow cooling zone 7, and fast cooling zone 8 along the axial direction of the furnace body to form a continuous material conveying channel. An online stress detection module is provided at the discharge outlet 4. The multi-zone independent heating module 9 is disposed on the inner wall of the furnace body. Within each axial zone, it is further divided into an upper heating zone 901, a lower heating zone 902, and a side heating zone 903 along the circumference of the furnace chamber, for independent and precise temperature control of different spatial locations within the furnace chamber. The upper heating zone 901 is disposed on the inner wall at the top of the furnace chamber, i.e., at the highest point of the furnace body's cross-section circumferential direction, and is continuously arranged along the entire axial length of the furnace body. It uses an infrared radiation heating plate and is mainly used to radiate heat to the upper surface of the bead blank. The lower heating zone 902 is disposed on the inner wall at the bottom of the furnace chamber, i.e., at the lowest point of the furnace body's cross-section circumferential direction, and is continuously arranged along the entire axial length of the furnace body. The heating elements are arranged continuously along the length of the furnace and use resistance wire heating elements, mainly used to conduct heat to the lower surface of the bead blank. The side heating zone 903 is set on the inner wall of the left and right sides of the furnace, that is, on the arc-shaped side wall between the upper heating zone 901 and the lower heating zone 902. It is arranged continuously along the entire axial length of the furnace body and uses silicon carbide rod heating elements, mainly used to radiate heat to the side of the bead blank and make up for the blind spots of the upper and lower heating. Each heating zone is connected to an independent power controller. The power controller adopts the silicon controlled rectifier power regulation method and receives the instructions of the control module 13 to precisely adjust the output power to realize the independent closed-loop control of each heating zone. The rotatable contour-following support mechanism 10 is located at the center of the bottom of the furnace body, extending through the entire axial length of the furnace body. It includes a tray 101 and a servo motor 102, used to support the glass bead preforms and drive their rotation, causing the heated surface of the preforms to change dynamically during annealing. The tray 101 extends along the furnace body axially through the entire preheating zone 5, heat preservation zone 6, slow cooling zone 7, and fast cooling zone 8, forming a continuous material support surface to achieve continuous conveying of the glass bead preforms in each process zone. The tray 101 is made of high-temperature resistant ceramic material, with evenly distributed arc-shaped grooves on its surface for accommodating the glass bead preforms. The radius of curvature of the arc-shaped grooves is greater than the radius of the glass bead preforms. This design ensures that the glass bead blank and the arc-shaped groove make point or line contact, minimizing the interference of the contact area on heat conduction while ensuring that the bead blank can roll freely during rotation. The servo motors 102 are installed on the outside of the furnace body below, with multiple motors spaced apart along the axial direction. Each servo motor 102 drives a section of the tray 101, enabling segmented independent control to adapt to the different rotation speed requirements of bead blanks of different diameters. The output shaft of the servo motor 102 extends into the furnace chamber through a sealed bearing and is connected to the bottom of the tray 101. The sealed bearing adopts a high-temperature resistant graphite sealing structure to prevent heat leakage from the furnace chamber and ensure the furnace chamber's airtightness. The multi-point temperature acquisition module is installed on the furnace body and includes multiple thermocouples 11 arranged in a matrix at different heights in the top, middle and bottom of the furnace and at different depths in the center and edges. These thermocouples are used to monitor the temperature distribution at various locations in the three-dimensional space within the furnace in real time. In each axial partition, at least nine temperature measuring points are arranged along the circumferential and radial directions to simultaneously collect temperature data from multiple temperature measuring points within the furnace, providing a data basis for temperature field uniformity assessment and control. All thermocouples 11 are connected to the temperature acquisition unit of the control module 13 to transmit the collected temperature signals to the control module 13 in real time. The hot air circulation disturbance module is installed inside the furnace body and includes multiple sets of micro hot air circulation fans 12 evenly spaced along the length of the furnace body. Each set of fans includes two fans installed on the inner walls of both sides of the furnace chamber and a guide plate 121 installed on the side of the fan facing the center of the furnace chamber, which is used to form forced convection in the furnace chamber and eliminate temperature stratification. The guide plate 121 is inclined towards the center of the furnace chamber to guide the hot air to converge towards the center of the furnace chamber, forming a uniform circulating flow field and improving the temperature field uniformity. The control module 13 is located outside the furnace body and integrated into a control cabinet. The control cabinet contains a database unit, a temperature acquisition unit, an algorithm control unit, a drive control unit, and a stress feedback unit. These units are connected via an internal bus for data sharing and collaborative control. The database unit stores basic annealing temperature curves, historical optimal process parameters, and corresponding residual stress thresholds for glass bead blanks of different materials and diameters, providing data support for process parameter matching and optimization. The temperature acquisition unit is connected to a multi-point temperature acquisition module to acquire temperature data from various points within the furnace in real time, and performs filtering, amplification, and analog-to-digital conversion processing to provide high-quality input data for the algorithm control unit. The algorithm control unit is connected to the database unit and the temperature acquisition unit, and incorporates the features described in Embodiment 1. A temporal deep network model and particle swarm optimization algorithm are used to identify the temperature field distribution and stress relief stage based on temperature data, calculate the output power of each heating zone, and calculate the asymmetric cooling curve based on the glass bead diameter, realizing intelligent conversion from data to decision. The drive control unit is connected to the algorithm control unit, the power controller of each heating zone, and each servo motor 102, and is used to convert the instructions generated by the algorithm control unit into specific control signals, control the output duty cycle of the power controller and the start, stop and speed of the servo motor 102, and realize precise drive of the heating module and the supporting mechanism. The stress feedback unit is connected to the online stress detection module at the discharge port 4, and is used to receive the residual stress detection results and feed them back to the database unit to optimize the basic annealing temperature curve and process parameter database, forming a closed-loop optimization. The online stress detection module is located at the discharge port 4 at the end of the rapid cooling zone 8 of the furnace body. It uses a polarized light stress detector to detect the residual stress of the glass bead preform in real time when it exits the furnace body, and transmits the detection results to the stress feedback unit to provide a basis for feedback optimization. This device achieves precise and uniform temperature distribution within the furnace through the synergistic effect of the multi-zone independent heating module 9 and the hot air circulation disturbance module. It achieves dynamic heating of the glass bead preform during the annealing process through the rotatable contour-following bearing mechanism 10. It achieves real-time perception and decision-making of the temperature field state through the multi-point temperature acquisition module and the intelligent control module 13. It achieves closed-loop optimization through the online stress detection module and the stress feedback unit. This provides complete hardware support for the method described in Example 1 and realizes fully automated control from temperature acquisition, state recognition, instruction generation to execution feedback.

[0032] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any simple modifications, equivalent changes and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the scope of the present invention.

Claims

1. A method for precise temperature field control and stress relief during the annealing process of glass bead preforms, characterized in that, Includes the following steps: The glass bead preform is loaded onto a rotatable conformal support mechanism inside the annealing furnace. The annealing furnace is equipped with a multi-zone independent heating module (9), which includes multiple independently temperature-controlled heating zones. Temperature data from multiple temperature measurement points inside the annealing furnace are collected synchronously by a sensor acquisition device, and combined with the material and geometric parameters of the glass bead blank to form a comprehensive working condition dataset. The temperature data is fused using an intelligent control algorithm based on temporal deep networks to identify the temperature field distribution state inside the furnace and the stress relief stage of the glass bead preform. Based on the temperature field distribution and stress relief stage identification results, the intelligent control module (13) calls the preset process parameter database data to generate the optimal power output command for each heating zone and the rotation command for the conformal bearing mechanism. The execution module receives the power output command of each heating zone and the rotation command of the conformal bearing mechanism, drives the multi-zone independent heating module (9) to adjust the power of each heating zone, and drives the conformal bearing mechanism to rotate at a preset speed, so that the heated surface of the glass bead blank changes dynamically, and performs coordinated adjustment of precise temperature field control and stress elimination. The feedback optimization module collects residual stress detection data, system energy consumption data, and furnace temperature fluctuation data in real time, compares them with preset target values, calculates deviation values, and iteratively corrects the power output commands of each heating zone and the rotation commands of the conformal bearing mechanism through optimization algorithms to form closed-loop control.

2. The method for precise temperature field control and stress relief during the annealing process of glass bead preforms according to claim 1, characterized in that: After generating the comprehensive working condition dataset, the following is also included: Preprocess the key variables in the comprehensive working condition dataset; The spatial distribution features, temporal gradient change features, and multi-point temperature difference features of the preprocessed data are extracted, and the extracted features are normalized to remove outliers. The extracted features are constructed into operating condition feature vectors and associated with annealing performance data in historical production records to form a structured operating condition feature dataset. The structured operating condition feature dataset is derived from the comprehensive operating condition dataset. The structured operating condition feature dataset is specifically used for decision-making and optimization of intelligent control algorithms.

3. The method for precise temperature field control and stress relief during the annealing process of glass bead preforms according to claim 1, characterized in that: The intelligent control algorithm based on time-series deep networks is used to fuse the temperature data and identify the temperature field distribution state inside the furnace and the stress relief stage of the glass bead preform. Specifically, this includes: A temporal deep network model based on an encoder-decoder architecture is constructed as the core model of the intelligent control algorithm; wherein, the encoder is used to fuse and extract features from the temporal sequence of the temperature data, and the decoder is used to output the identification results of the temperature field distribution state and stress relief stage; The encoder's data fusion and feature extraction include: First, the temperature time series data from multiple temperature measurement points, which have undergone timestamp alignment and normalization preprocessing, are spliced ​​together along the channel dimension to form a multi-channel time series feature tensor, thus achieving the initial aggregation of multi-source data. Then, a one-dimensional convolutional neural network layer is used to perform convolution operations on the multi-channel temporal feature tensor. By sliding the convolution kernel in the time dimension, the local temporal features of different channels are fused across channels to extract the spatial correlation features between each temperature measurement point and obtain a convolutional feature sequence that incorporates spatial information. Finally, the convolutional feature sequence is input into a bidirectional gated recurrent unit network. The forward GRU and backward GRU capture the historical dependencies and future trends of the time series data respectively. The features are deeply fused in the time dimension to explore the forward and backward dynamic dependencies of temperature changes in time and output a deep time series feature vector that integrates spatial correlation and time series dependency. The state recognition of the decoder includes: introducing an attention mechanism into the deep temporal feature vector output by the bidirectional gated recurrent unit network, calculating the weights of different time steps, and focusing on the most critical temporal segments for judging the current state; based on the weighted feature vector, simultaneously outputting the temperature field distribution state and the stress relief stage recognition results of the glass bead preform through a fully connected classification network.

4. The method for precise temperature field control and stress relief during the annealing process of glass bead preforms according to claim 1, characterized in that: The generation of optimal power output commands for each heating zone and rotation commands for the conformal bearing mechanism specifically includes: The intelligent control module (13) calls the preset process parameter database, which stores the historical optimal process parameters under different temperature field distribution states and stress relief stage combinations; Based on the temperature field distribution and stress relief stage identification results, a similarity matching algorithm is used to match the process parameters of similar working conditions in the database. Combined with the real-time equipment operation stability and target residual stress requirements, the parameters are adjusted through a multi-objective optimization algorithm to determine the preliminary power output values ​​of each heating zone and the rotation speed of the conformal bearing mechanism. The initial power output values ​​of each heating zone and the rotation speed of the conforming bearing mechanism are integrated with the stress concentration prevention strategy based on the temperature field uniformity requirement. The safety deviation range of the power output of each heating zone is set according to the temperature field distribution. If the initial power output value exceeds the corresponding safety deviation range, the initial power output value is limited to the boundary value, and the rotation speed of the conforming bearing mechanism is adjusted synchronously to generate a coordinated control command for temperature field control and stress relief.

5. The method for precise temperature field control and stress relief during the annealing process of glass bead preforms according to claim 1, characterized in that: The driving multi-zone independent heating module (9) adjusts the power of each heating zone to maintain a uniform temperature field, and simultaneously drives the contour-following bearing mechanism to rotate at a preset speed, performing coordinated adjustment of precise temperature field control and stress relief, specifically including: The multi-zone independent heating module (9) in the execution module converts the optimal power output command into the control signal of the power controller of each heating zone, drives the upper heating zone (901), lower heating zone (902) and side heating zone (903) to operate at the set power, and monitors the actual power and temperature feedback of each heating zone in real time. The contour-following bearing mechanism drives the servo motor (102) according to the rotation command, controls the tray (101) to rotate intermittently at the set speed, monitors the load and rotation stability data of the tray (101) in real time, and performs coordinated adjustment of the heating and cooling processes.

6. The method for precise temperature field control and stress relief during the annealing process of glass bead preforms according to claim 1, characterized in that: The feedback optimization module collects residual stress detection data, system energy consumption data, and furnace temperature fluctuation data in real time, compares them with preset target values, calculates deviation values, and uses an optimization algorithm to iteratively correct the power output commands of each heating zone and the rotation commands of the conformal bearing mechanism, forming a closed-loop control. Specifically, this includes: The residual stress data of the annealed glass bead preform is collected by an online stress detector, the system energy consumption data is collected by a power sensor, and the temperature fluctuation data of each temperature measuring point in the furnace is collected by a multi-point temperature acquisition module. The calculation unit compares the collected residual stress data, system energy consumption data, and multi-point temperature fluctuation data with the preset target value determined based on product standards and historical best production data to calculate the deviation value. The calculation process of the deviation value is as follows: the absolute deviation between the measured residual stress value and the target value, the absolute deviation between the measured system energy consumption value and the target value, and the root mean square error between the measured temperature value and the set value at each temperature measurement point are calculated as temperature fluctuation deviations. Then, the three deviation values ​​are weighted and fused according to preset weight coefficients to obtain a comprehensive deviation evaluation function. An optimization algorithm is adopted to minimize the comprehensive deviation evaluation function. The power output command of each heating zone and the rotation command of the conformal bearing mechanism in the next control cycle are iteratively optimized. The optimal command obtained by optimization is updated to the process parameter database of the intelligent control module (13) for subsequent batch temperature field control and stress relief decision-making.

7. The method for precise temperature field control and stress relief during the annealing process of glass bead preforms according to claim 6, characterized in that: An optimization algorithm is employed to iteratively optimize the power output commands for each heating zone and the rotation commands for the conformal bearing mechanism in the next control cycle, with the objective of minimizing the comprehensive deviation evaluation function. Specifically, this includes: Initialize the particle swarm by constructing a multi-dimensional search space using the current power output commands of each heating zone and the rotation commands of the contour bearing mechanism, and initialize the position and velocity of each particle in the particle swarm. Then, iterative optimization is performed, mapping the position coordinates of each particle in the current iteration to a set of candidate power output values ​​of each heating zone and rotational speed of the conformal bearing mechanism. The fitness value of the particle is calculated by substituting the deviation value into the comprehensive deviation evaluation function constructed by the deviation value. Based on the particle's current velocity, individual optimal position and global optimal position, the position and velocity of the next generation of particles are calculated according to the velocity and position update formula of the particle swarm optimization algorithm. When the preset termination condition is reached, the iteration stops, and the power output value of each heating zone corresponding to the global optimal position and the rotation speed of the conformal bearing mechanism are taken as the optimal command output for this optimization.

8. A device for precise temperature field control and stress relief during the annealing process of glass bead preforms to implement the method of any one of claims 1-7, characterized in that: It includes a horizontal cylindrical furnace body (1), a multi-zone independent heating module (9), a rotatable contour-following support mechanism (10), a multi-point temperature acquisition module, a control module (13), and a hot air circulation disturbance module; The furnace body is provided with a feed inlet (2) and a discharge outlet (4) at both ends, and an automatic furnace door (3) is installed at the feed inlet (2); the rotatable contour-following bearing mechanism (10) runs through the entire preheating zone (5), heat preservation zone (6), slow cooling zone (7) and fast cooling zone (8) along the furnace body axis to form a continuous material conveying channel; The multi-zone independent heating module (9) is installed on the inner wall of the furnace body and is divided into an upper heating zone (901), a lower heating zone (902) and a side heating zone (903) along the circumference of the furnace. The upper heating zone (901) is installed on the inner wall of the top of the furnace body, that is, at the highest point of the circumferential direction of the furnace body cross-section, and is continuously arranged along the entire axial length of the furnace body. The lower heating zone (902) is installed on the inner wall of the bottom of the furnace body, that is, at the lowest point of the circumferential direction of the furnace body cross-section, and is continuously arranged along the entire axial length of the furnace body. The side heating zone (903) is installed on the inner walls of the left and right sides of the furnace body, that is, on the arc-shaped side wall between the upper heating zone (901) and the lower heating zone (902), and is continuously arranged along the entire axial length of the furnace body. Each heating zone is connected to an independent power controller. The rotatable contour-following support mechanism (10) is located at the center of the bottom of the furnace body and includes a tray (101) and a servo motor (102). The servo motor (102) is installed on the outside of the furnace body and is used to drive the tray (101) to rotate intermittently at a preset speed. The output shaft of the servo motor (102) extends into the furnace chamber through a sealed bearing and is connected to the bottom of the tray (101) for transmission. The surface of the tray (101) is evenly distributed with arc-shaped grooves for accommodating glass bead blanks. The multi-point temperature acquisition module is installed on the furnace body and includes multiple thermocouples (11) arranged in a matrix at different heights in the top, middle and bottom of the furnace and at different depths in the center and edge, for synchronously acquiring temperature data of multiple temperature measurement points in the furnace. The hot air circulation disturbance module is installed inside the furnace body and includes two sets of micro hot air circulation fans (12) evenly spaced along the length of the furnace body. Each set of fans includes two fans installed on the inner walls of both sides of the furnace chamber, and a guide plate (121) installed on the side of the fan facing the center of the furnace chamber. The guide plate (121) is inclined towards the center of the furnace chamber. The control module (13) is located outside the furnace body and integrated into the control cabinet. The control cabinet contains a database unit, a temperature acquisition unit, an algorithm control unit, a drive control unit, and a stress feedback unit. The units are connected to each other via an internal bus. The database unit is used to store the basic annealing temperature curves and corresponding residual stress thresholds of glass beads of different materials and diameters. The temperature acquisition unit is connected to a multi-point temperature acquisition module to acquire temperature data of each point in the furnace in real time. The algorithm control unit is connected to the database unit and the temperature acquisition unit to calculate the output power of each heating zone based on the temperature data using a decoupling control algorithm, and to calculate the asymmetric cooling curve based on the diameter of the glass beads. The drive control unit is connected to the algorithm control unit and the servo motor (102) to control the start, stop, and speed of the servo motor (102). The stress feedback unit is used to receive the residual stress detection results and feed them back to the database unit to optimize the basic annealing temperature curve.

9. The device for precise temperature field control and stress relief during the annealing process of glass beads according to claim 8, characterized in that: The furnace body is divided into a preheating zone (5), a heat preservation zone (6), a slow cooling zone (7) and a fast cooling zone (8) along the axial direction. Each zone is independently equipped with the multi-zone independent heating module (9).

10. The device for precise temperature field control and stress relief during the annealing process of glass bead preforms according to claim 8, characterized in that: The device also includes an online stress detection module, which is located at the discharge port (4) at the end of the rapid cooling zone (8) of the furnace body. It is used to detect the residual stress of the glass bead preform after annealing in real time and transmit the detection results to the stress feedback unit.

Images