A method for controlling a rotary motor of a crystal growing furnace

Abstract

The application provides a crystal growth furnace rotating motor control method, and relates to the technical field of crystal growth furnace. The crystal growth furnace rotating motor control method successfully optimizes the thermal field design by deeply studying the interaction between the furnace body structure and the crystal / melt temperature distribution, as well as the influence of the thermal field structure on the growth interface, accurately simulates the temperature distribution in the furnace by using numerical calculation and analysis technology, ensures the temperature uniformity in the crystal growth process, thereby improving the quality of the crystal and reducing the energy consumption. In addition, an artificial neural network model is used to predict the growth conditions and optimize the related parameters, realizing fine control of the crystal growth process. Finally, through experimental verification and adjustment, the accuracy of the simulation is ensured, and the applicability and accuracy of the model are improved. The automated and optimized control process significantly improves the efficiency of scientific research and production, shortens the research and development cycle, and accelerates the listing of new products.

Description

Technical Field

[0001] This invention relates to the field of crystal growth furnace technology, specifically to a method for controlling a rotary motor in a crystal growth furnace. Background Technology

[0002] A crystal growth furnace is an electric furnace used under specific conditions to melt and purify materials and grow them into single or polycrystalline materials. It is mainly used for the crystal growth of semiconductor materials (such as silicon and germanium), laser materials (such as yttrium aluminum garnet), and high-purity metal materials (such as tungsten and molybdenum). A crystal growth furnace typically consists of a vacuum chamber, a vacuum and gas filling system, a heating system, lifting and rotating mechanisms, temperature and pressure measurement and control systems, and other auxiliary equipment. High precision is required for both control and transmission. Crystal growth furnaces are classified into single-crystal furnaces and polycrystalline furnaces based on the structure of the crystal being grown. Single-crystal furnaces are further divided into Czochralski single-crystal furnaces and zone melting single-crystal furnaces based on different purification methods, while polycrystalline furnaces commonly used are polycrystalline silicon ingot furnaces. The Czochralski single-crystal furnace, also known as the Czochralski method, is a commonly used crystal growth method suitable for the mass production of large-size perfect crystals. This method controls the rotation and pulling speed of the seed crystal, causing the molten material to crystallize along the seed crystal and grow into a rod-shaped single crystal. Zone melting is a method that can significantly increase the purity of single-crystal materials during crystallization. It involves moving the material rod or heating element to shift the molten zone, causing crystallization and ultimately forming a single-crystal rod. Crystal growth furnaces have a wide range of applications, including semiconductor manufacturing, optical device manufacturing, and inorganic material preparation. These devices are crucial for modern industry and technological development because they can produce high-quality single-crystal materials, which are indispensable foundations for many high-tech products and research fields.

[0003] Under current technological conditions, the growth process of silicon carbide crystals is typically carried out at high temperatures exceeding 2000°C. Due to the limitations of high temperatures and confined spaces, traditional temperature monitoring methods are difficult to apply, making it challenging to directly observe and monitor temperature field changes during the growth process. When growing silicon carbide crystals using the physical vapor transport (PVT) method, the gaseous components often crystallize and accumulate at the center of the temperature-sensing aperture in the lowest temperature zone of the growth chamber. This phenomenon prevents direct observation of the seed crystal cap temperature, and as the accumulation at the temperature-sensing aperture increases, the temperature field at the crystal growth interface also changes, potentially leading to the formation of defects such as polytype inclusions and stress. Summary of the Invention

[0004] (a) Technical problems to be solved

[0005] To address the shortcomings of existing technologies, this invention provides a method for controlling a rotary motor in a crystal growth furnace, which solves the problems of difficulty in directly observing and monitoring temperature field changes during the growth process, as well as the potential for changes in the temperature field at the crystal growth interface, which may lead to the formation of defects such as polymorphic inclusions and stress.

[0006] (II) Technical Solution

[0007] To achieve the above objectives, the present invention provides the following technical solution: a method for controlling a rotary motor in a crystal growth furnace, specifically comprising the following steps:

[0008] S1. Establishing geometric and physical models

[0009] Based on the actual structure of the crystal growth furnace, a geometric model is established, including the crucible, heater and insulation material. At the same time, a physical model is established, including heat conduction, fluid dynamics and mass transfer. The Boltzmann equation and the lattice Boltzmann method are used to simulate crystal growth.

[0010] S2. Data Calculation and Result Analysis

[0011] The governing equations are discretized using the finite volume method, as shown in the following formula:

[0012]

[0013] Where f represents the physical quantities of velocity, enthalpy, and turbulence parameters, u is the velocity field, D is the diffusion coefficient, and S is the source term. Then, the electromagnetic field is solved using the complex function method to ensure the correctness of the model. Next, global data simulation is performed to analyze the temperature distribution in the crystal and the melt, the argon convection between the heat shield and the quartz crucible, and the thermal stress in the crystal.

[0014] S3. Optimize thermal field structure

[0015] By changing the crucible shape, heater position, and insulation cage partition block position parameters, the crystal growth process was optimized using computer numerical simulation. The effects of the optimized heat shield on the heater, the sidewall carbon felt's ability to prevent heat loss, and the flow guiding system's effect on reducing SiO deposition on the upper wall were analyzed.

[0016] S4. Process Parameter Optimization

[0017] Artificial neural networks and genetic algorithms were used to optimize process parameters such as pulling speed, crystal rotation speed, ambient gas, and crucible wall temperature to obtain a tight and flat solid-liquid interface. The optimal crystal rotation and crucible rotation under the best thermal field distribution were determined using response surface methodology, and gradient descent was used to update the network weights.

[0018] Δw=-ηj'(w)

[0019] Where η is the learning rate, j ’ (w) is the gradient of the loss with respect to the parameters;

[0020] S5. Machine Learning Model Building and Training

[0021] Based on CFD simulation data, an artificial neural network is used to establish a model of the crystal growth process. The neural network model is trained to predict the CFD simulation results of SiC crystal growth and to optimize the growth conditions.

[0022] S6. Rotary motor control

[0023] The controller controls the rotary motor to rotate the seed crystal using the lifting device, causing the crystal to rotate. The torque is calculated using the BLDC motor sinusoidal rectification commutation formula.

[0024] Shaft torque = K t [I R sin(θ)+I S sin(θ+120°)+I T sin(θ+240°)]

[0025] Where θ is the electrical angle of the rotation axis, and K t I is the torque constant of the motor. R ,I S ,I T As the phase current, during the constant diameter growth stage, the controller controls the rotary motor to rotate the seed crystal of the pulling device, and at the same time controls the heating device to heat the crucible.

[0026] S7. Data Feedback and Closed-Loop Control

[0027] Encoders are used to provide feedback on the positions of the crucible rod and seed crystal rod, enabling closed-loop control and display. Parameters are set and displayed via a touchscreen, the real-time status of the motor is monitored, and control parameters are adjusted based on feedback. Specifically, a PID control algorithm is used for closed-loop control of key parameters in the crystal growth process.

[0028]

[0029] Among them, K p It is the proportionality coefficient, K i It is the integral coefficient, K d These are the differential coefficients, and e(t) is the deviation;

[0030] S8. Experimental Verification and Adjustment

[0031] Crystal growth experiments were conducted to verify the accuracy of the numerical simulation, and the model parameters were adjusted based on the results to improve the reliability of the simulation.

[0032] Preferably, S1. Establishing a geometric and physical model specifically includes the following steps: geometric modeling, basic assumptions, establishment of the governing equations, setting of boundary conditions and initial conditions, mathematical modeling, mesh generation, equation discretization, selection of numerical solution methods, and coupling solution of physical quantities.

[0033] Preferably, S5. Machine learning model establishment and training specifically includes the following steps: problem definition, data preparation, feature selection, model selection, training and validation, parameter tuning, model training, dataset partitioning, model method selection, defining hyperparameters, loading data, initializing the network, defining loss and optimizer, training the network, model evaluation, model saving, and using the model.

[0034] (III) Beneficial Effects

[0035] This invention provides a method for controlling a rotary motor in a crystal growth furnace. It offers the following advantages:

[0036] This invention provides a rotary motor control method for a crystal growth furnace. Numerical simulation technology is used to study the relationship between the furnace structure and the crystal / melt temperature distribution, the evolution of interfaces at different growth stages, and the mechanism by which the thermal field structure affects the growth interface, aiming to optimize the thermal field design. Specifically, through numerical calculation and result analysis, the temperature distribution within the crystal growth furnace is accurately simulated and analyzed to ensure temperature uniformity during crystal growth. This process plays a crucial role in improving crystal quality. Optimizing the thermal field structure can improve thermal efficiency, reduce energy consumption, and provide a more stable and uniform growth environment, contributing to improved crystal uniformity and repeatability. By optimizing thermal field parameters and process parameters, crystal growth rate and efficiency can be increased, growth cycle shortened, and production costs reduced. Data feedback and closed-loop control can monitor and adjust the growth process in real time, ensuring process stability and reducing growth defects caused by operational errors or environmental changes. Precise rotary motor control and torque calculation help maintain the accurate position and orientation of the crystal during growth, reducing crystal stress and defects, and improving the overall quality of the crystal. Automated control reduces the need for human operation, lowers the possibility of operational errors, and improves production reliability. Precise numerical simulations and optimization algorithms ensure consistency in each growth process, improving the repeatability of production. Optimized thermal field structures and process parameters reduce energy consumption, lower production costs, and minimize environmental impact. Optimization through machine learning models allows the system to adapt to different growth conditions and requirements, enhancing its flexibility and adaptability. Artificial neural network models can predict crystal growth conditions, optimize growth parameters, and achieve finer control. Experimental verification and adjustments ensure the accuracy of simulation results, and model parameters can be further adjusted based on experimental results, improving the model's applicability and accuracy. Automation and optimization of the entire control process significantly improve research and production efficiency, shorten R&D cycles, and accelerate the time to market for new products. Attached Figure Description

[0037] Figure 1 This is a schematic diagram of the method flow of the present invention. Detailed Implementation

[0038] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0039] Example 1: Self-designed and temperature control performance experiment of a cadmium zinc telluride crystal growth furnace

[0040] Implementation steps:

[0041] A CdZnTe crystal growth furnace based on a moving heating method.

[0042] The furnace heating unit includes four specifications, divided into six temperature control sections, and is raised and lowered by a ball screw linear guide driven by an industrial computer-controlled servo motor.

[0043] The furnace cavity is equipped with heating tubes consisting of corundum ceramic tubes and high-temperature metal heat pipes. The heating unit is controlled by high-precision platinum-rhodium-platinum thermocouples, Eurotherm, transformers, and thyristors.

[0044] A fuzzy + PID control algorithm is used to adjust and control the temperature distribution of the heating furnace.

[0045] Stability and control performance experiments were conducted during the temperature heating process.

[0046] Experimental results:

[0047] The heating temperature inside the furnace cavity is continuously controlled for 200 hours, with the temperature fluctuation at the same location being ±0.005℃ and the heating temperature deviation being ≤±0.1℃.

[0048] The lengths of the upper and lower constant temperature zones of the furnace cavity are 400mm and 240mm, respectively.

[0049] The length of the temperature gradient zone in the middle of the furnace cavity is approximately 136 mm.

[0050] The length of the constant temperature zone at the bottom of the furnace cavity is 240mm.

[0051] At a heating temperature of approximately 1098°C, the temperature gradient is 0.92°C·mm⁻¹.

[0052] Experimental results show that the furnace meets the independent design and temperature control performance requirements of a CdZnTe crystal growth furnace.

[0053] Example 2: Automated diameter control Czochralski growth of large-size YVO crystals

[0054] Implementation steps:

[0055] Maintain an appropriate temperature field distribution in the space.

[0056] Gradually increase the medium-frequency induction power of the crystal furnace to heat and melt the polycrystalline material. After adding polycrystalline material three times and performing high-temperature melting, set a power reduction program to slightly reduce the heating temperature.

[0057] (001) oriented YVO4 rod-shaped seed crystals are used for seeding growth. The seed crystal fixed to the seed crystal rod is moved down to the surface of the melt. The seed crystal is first washed in a rotating lifting state, and then the appropriate temperature is adjusted to perform proper seed crystal welding.

[0058] Once the crystal diameter has slightly increased and the growth aperture has stabilized, the growth process switches to automatic diameter control under the program module adjustment.

[0059] The crystal rotation speed is maintained at 8–15 r / min, and the pulling speed is 0.8–2.0 mm / h.

[0060] Experimental results:

[0061] The above method enables automated diameter-controlled Czochralski growth of large-size YVO crystals, with a stable crystal growth process and precise diameter control.

[0062] Example 3: Application in the control system of SiC (silicon carbide) crystal growth furnace

[0063] Implementation steps:

[0064] During the SiC crystallization process, three stepper motors are used. Two stepper motors drive the lead screw to move the crucible rod and the seed crystal rod up and down, while the other motor drives the crucible to rotate.

[0065] Two LM3106A CPU modules were selected: one to control the stepper motors of the crucible rod and the seed crystal rod, and the other to control the motor that controls the rotation of the crucible.

[0066] Two two-stage reducers are selected between the stepper motor and the lead screw of the crucible rod and the seed crystal rod, with the first stage having a ratio of 80:1 and the second stage having a ratio of 100:1.

[0067] A 12:1 single-stage reducer is used for crucible rotation.

[0068] Hetech's touchscreen was selected for parameter setting and display, and two PLC slave stations were connected via RS485 port using the ModBus protocol.

[0069] Experimental results:

[0070] The stepper motor parameters for the crucible rod and seed crystal rod are set to 2000 Pulse / r, the lead screw is 5mm, and the reduction ratio is 1:8000 for slow speed and 1:80 for fast speed.

[0071] The speed range is 0.002 to 0.50 mm / hour for slow speeds and 1 to 60 mm / min for fast speeds.

[0072] The encoder resolution is 100ppr, which means 100 pulses correspond to 5 millimeters.

[0073] Experimental results show that the system can precisely control the rotation and stretching motions during the SiC crystal growth process, ensuring that the crystal texture meets the requirements.

[0074] In this embodiment of the invention, the specific method is as follows:

[0075] S1. Establishing geometric and physical models

[0076] Based on the actual structure of the crystal growth furnace, a geometric model is established, including the crucible, heater, and insulation material. Simultaneously, a physical model is established, including heat conduction, fluid dynamics, and mass transfer. The Boltzmann equation and lattice Boltzmann method are used to simulate crystal growth. The establishment of the geometric and physical models specifically includes the following steps:

[0077] Geometric modeling research:

[0078] For industrial crystal furnaces used in actual crystal growth, a corresponding geometric model is constructed. This model can be imported using computer-aided design (CAD) files or constructed independently in simulation software.

[0079] Explanation of the physical process:

[0080] This paper elaborates on the key physical mechanisms in the crystal growth process, taking the PVT process as an example. A medium-frequency induction coil or resistance heater generates heat, creating axial and radial temperature gradients inside the crucible. The raw material decomposes upon heating, and the resulting gaseous components are transported to the low-temperature seed crystal surface at the top of the crucible under the influence of the temperature gradient, where recrystallization occurs, forming a single crystal.

[0081] Basic assumptions:

[0082] In constructing the physical model, the silicon melt is treated as an incompressible Newtonian fluid with constant physical properties, satisfying the Boussinesq approximation; argon is treated as an ideal gas, using a low Mach number approximation, and conforming to the ideal gas law; the absorption of radiation energy by the gas is ignored, and all radiating surfaces are considered as gray body surfaces; during the melting process of silicon raw materials, its density remains consistent with that of the melt.

[0083] Construction of the governing equations:

[0084] A set of governing equations was constructed, which included calculations of the thermal field in the solid region, the thermal convection of the melt, and the radiation in the furnace cavity. In terms of the thermal convection of the melt, the governing equations included the momentum equation, the energy equation, and the continuity equation.

[0085] Boundary conditions and initial conditions settings:

[0086] To set the corresponding boundary and initial conditions for the governing equations, the boundary conditions of the melt surface and solidification interface involve temperature, velocity, and radiation conditions.

[0087] Mathematical modeling process:

[0088] Transforming a physical model into a mathematical model involves defining the governing equations and boundary conditions. This process involves expressing the physical process in mathematical terms and clarifying the solution domain and boundaries of the equations.

[0089] Mesh generation techniques:

[0090] A mesh for numerical computation is generated on the geometric model. The quality and quantity of the mesh directly affect the accuracy and computational efficiency of the numerical simulation.

[0091] Equation discretization:

[0092] The governing equations are discretized and transformed into a system of algebraic equations that can be solved on a computer.

[0093] Physical quantity coupling solution strategy:

[0094] The SIMPLE method algorithm is used to solve for the coupled physical quantities.

[0095] S2. Data Calculation and Result Analysis

[0096] The governing equations are discretized using the finite volume method, as shown in the following formula:

[0097]

[0098] Where f represents the physical quantities of velocity, enthalpy, and turbulence parameters, u is the velocity field, D is the diffusion coefficient, and S is the source term. Then, the electromagnetic field is solved using the complex function method to ensure the correctness of the model. Next, global data simulation is performed to analyze the temperature distribution in the crystal and the melt, the argon convection between the heat shield and the quartz crucible, and the thermal stress in the crystal.

[0099] S3. Optimize thermal field structure

[0100] By changing the crucible shape, heater position, and insulation cage partition block position parameters, the crystal growth process was optimized using computer numerical simulation. The effects of the optimized heat shield on the heater, the sidewall carbon felt's ability to prevent heat loss, and the flow guiding system's effect on reducing SiO deposition on the upper wall were analyzed.

[0101] S4. Process Parameter Optimization

[0102] Artificial neural networks and genetic algorithms were used to optimize process parameters such as pulling speed, crystal rotation speed, ambient gas, and crucible wall temperature to obtain a tight and flat solid-liquid interface. The optimal crystal rotation and crucible rotation under the best thermal field distribution were determined using response surface methodology, and gradient descent was used to update the network weights.

[0103] Δw=-ηj'(w)

[0104] Where η is the learning rate, and j'(w) is the gradient of the loss with respect to the parameters;

[0105] S5. Machine Learning Model Building and Training

[0106] Based on CFD simulation data, an artificial neural network is used to establish a model of the crystal growth process. The neural network model is trained to predict the CFD simulation results of SiC crystal growth and to optimize growth conditions. The establishment and training of the machine learning model specifically includes the following steps:

[0107] Problem definition:

[0108] Clearly define the problem to be solved, including identifying target variables, evaluation indicators, and expected results.

[0109] Data preparation:

[0110] Collect, clean, and preprocess data, including handling missing values, outliers, and noise, to ensure data quality and consistency.

[0111] Feature selection:

[0112] Identify and select the data columns that are most helpful to the prediction target, including feature engineering, such as creating interaction items, multinomial features, or applying dimensionality reduction techniques.

[0113] Model selection:

[0114] Depending on the nature of the problem, choose a suitable machine learning algorithm, including decision trees, random forests, support vector machines, and neural networks.

[0115] Training and validation:

[0116] The model is trained using a training dataset and its performance is evaluated using a validation set. Cross-validation is a common method to ensure that the model performs well not only on a fixed subset of data.

[0117] Parameter tuning:

[0118] Finding the optimal combination of hyperparameters through techniques such as grid search, random search, or Bayesian optimization can significantly improve model performance.

[0119] Model training:

[0120] The process of learning from data experience using established model methods also requires model evaluation to adjust the hyperparameters of the algorithm, and finally select the model with better performance.

[0121] Dataset partitioning: The dataset is divided into a training set and a test set, and the training set can be further subdivided into a training set and a validation set.

[0122] Model and method selection:

[0123] Choose the appropriate model structure based on the problem, such as a convolutional neural network (CNN).

[0124] Define hyperparameters:

[0125] Hyperparameters are configuration settings used to adjust the model training method, such as learning rate, batch size, and training period.

[0126] Loading data:

[0127] Use the torchvision.datasets module to download and load datasets, and use DataLoader to handle batch processing and shuffling.

[0128] Initialize the network:

[0129] Instantiate the neural network and move it to the device (GPU or CPU).

[0130] Define the loss and optimizer:

[0131] Choose a loss function (such as cross-entropy loss) and an optimizer (such as the Adam optimizer).

[0132] Training the network:

[0133] The dataset is iterated through multiple epochs, and the model weights are updated based on the loss.

[0134] Model evaluation:

[0135] Define a function to check the accuracy of the model on the training and test datasets.

[0136] Model saving:

[0137] Use libraries like joblib or pickle to save the model to a file for future use.

[0138] Model used:

[0139] Applying a model to new data for prediction includes loading the model and predicting new data;

[0140] S6. Rotary motor control

[0141] The controller controls the rotary motor to rotate the seed crystal using the lifting device, causing the crystal to rotate. The torque is calculated using the BLDC motor sinusoidal rectification commutation formula.

[0142] Shaft torque = K t [I R sin(θ)+I S sin(θ+120°)+I T sin(θ+240°)]

[0143] Where θ is the electrical angle of the rotation axis, and K t I is the torque constant of the motor. R ,IS ,I T As the phase current, during the constant diameter growth stage, the controller controls the rotary motor to rotate the seed crystal of the pulling device, and at the same time controls the heating device to heat the crucible.

[0144] S7. Data Feedback and Closed-Loop Control

[0145] Encoders are used to provide feedback on the positions of the crucible rod and seed crystal rod, enabling closed-loop control and display. Parameters are set and displayed via a touchscreen, the real-time status of the motor is monitored, and control parameters are adjusted based on feedback. Specifically, a PID control algorithm is used for closed-loop control of key parameters in the crystal growth process.

[0146]

[0147] Among them, K p It is the proportionality coefficient, K i It is the integral coefficient, K d These are the differential coefficients, and e(t) is the deviation;

[0148] S8. Experimental Verification and Adjustment

[0149] Crystal growth experiments were conducted to verify the accuracy of the numerical simulation, and the model parameters were adjusted based on the results to improve the reliability of the simulation.

[0150] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for controlling a rotary motor in a crystal growth furnace, characterized in that, Specifically, the following steps are included: S1. Establishing geometric and physical models Based on the actual structure of the crystal growth furnace, a geometric model is established, including the crucible, heater and insulation material. At the same time, a physical model is established, including heat conduction, fluid dynamics and mass transfer. The Boltzmann equation and the lattice Boltzmann method are used to simulate crystal growth. S2. Data Calculation and Result Analysis The governing equations are discretized using the finite volume method, as shown in the following formula: Where f represents the physical quantities of velocity, enthalpy, and turbulence parameters, u is the velocity field, D is the diffusion coefficient, and S is the source term. Then, the electromagnetic field is solved using the complex function method to ensure the correctness of the model. Next, global data simulation is performed to analyze the temperature distribution in the crystal and the melt, the argon convection between the heat shield and the quartz crucible, and the thermal stress in the crystal. S3. Optimize thermal field structure By changing the crucible shape, heater position, and insulation cage partition block position parameters, the crystal growth process was optimized using computer numerical simulation. The effects of the optimized heat shield on the heater, the sidewall carbon felt's ability to prevent heat loss, and the flow guiding system's effect on reducing SiO deposition on the upper wall were analyzed. S4. Process Parameter Optimization Artificial neural networks and genetic algorithms were used to optimize process parameters such as pulling speed, crystal rotation speed, ambient gas, and crucible wall temperature to obtain a tight and flat solid-liquid interface. The optimal crystal rotation and crucible rotation under the best thermal field distribution were determined using response surface methodology, and gradient descent was used to update the network weights. Δw=-ηj'(w) Where η is the learning rate, and j'(w) is the gradient of the loss with respect to the parameters; S5. Machine Learning Model Building and Training Based on CFD simulation data, an artificial neural network is used to establish a model of the crystal growth process. The neural network model is trained to predict the CFD simulation results of SiC crystal growth and to optimize the growth conditions. S6. Rotary motor control The controller controls the rotary motor to rotate the seed crystal using the lifting device, causing the crystal to rotate. The torque is calculated using the BLDC motor sinusoidal rectification commutation formula. Shaft torque = K t [I R sin(θ) + I S sin(θ + 120°) + I T sin(θ + 240°)] Where θ is the electrical angle of the rotation axis, and K t I is the torque constant of the motor. R ,I S ,I T As the phase current, during the constant diameter growth stage, the controller controls the rotary motor to rotate the seed crystal of the pulling device, and at the same time controls the heating device to heat the crucible. S7. Data Feedback and Closed-Loop Control Encoders are used to provide feedback on the positions of the crucible rod and seed crystal rod, enabling closed-loop control and display. Parameters are set and displayed via a touchscreen, the real-time status of the motor is monitored, and control parameters are adjusted based on feedback. Specifically, a PID control algorithm is used for closed-loop control of key parameters in the crystal growth process. Among them, K p It is the proportionality coefficient, K i It is the integral coefficient, K d These are the differential coefficients, and e(t) is the deviation; S8. Experimental Verification and Adjustment Crystal growth experiments were conducted to verify the accuracy of the numerical simulation, and the model parameters were adjusted based on the results to improve the reliability of the simulation.

2. The method for controlling a rotary motor in a crystal growth furnace according to claim 1, characterized in that: The S1. Establishing the geometric and physical model specifically includes the following steps: geometric modeling, basic assumptions, establishment of the governing equations, setting of boundary conditions and initial conditions, mathematical modeling, mesh generation, equation discretization, selection of numerical solution methods, and coupling solution of physical quantities.

3. The method for controlling a rotary motor in a crystal growth furnace according to claim 1, characterized in that: The S5. Machine Learning Model Building and Training specifically includes the following steps: problem definition, data preparation, feature selection, model selection, training and validation, parameter tuning, model training, dataset partitioning, model method selection, defining hyperparameters, loading data, initializing the network, defining loss and optimizer, training the network, model evaluation, model saving, and using the model.

Images