Control method for efficient and intelligent discharge of industrial silicon
Through the combination of multimodal sensor network and machine learning model, precise control of industrial silicon is achieved, solving the problems of timing judgment deviations and safety hazards in the existing technology, and improving production efficiency and safety.
Patent Information
- Application Number
- CN202510513974.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-04-23
- Publication Date
- 2025-08-08
AI Technical Summary
The existing industrial silicon release control technology cannot obtain the silicon liquid flow rate, melt phase state and environmental parameters simultaneously, resulting in a deviation in the judgment of the timing of the release, and lack of dynamic data fusion, making it difficult to balance energy consumption, efficiency and silicon purity. The traditional actuator has large errors and poses safety hazards.
The multi-modal sensor network is used to collect data in real time, combine the CNN-LSTM hybrid model for data fusion analysis, generate dynamic optimization control instructions for the tilt angle, furnace discharge speed and cooling rate, and execute it through a closed-loop control system, integrating exception processing modules and digital twin technology to achieve automated control throughout the process.
Significantly improve the accuracy of the timing prediction, reduce impurity residues and energy waste, reduce the risk of silicon liquid splash, improve production safety and efficiency, and shorten the process debugging cycle.
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of industrial silicon smelting, and in particular to a control method for efficiently and intelligently discharging industrial silicon. Background Art
[0002] Industrial silicon is a critical building block for photovoltaics, semiconductors, and alloy manufacturing. The efficiency and quality of its smelting process directly impact product performance and production costs. In industrial silicon smelting, the discharge of liquid silicon from the furnace is a key step in determining ingot purity, energy consumption, and production safety. Its control precision and intelligent level are crucial for process optimization.
[0003] However, the current industrial silicon furnace control technology still has the following technical bottlenecks: (1) Only temperature or pressure sensors are used to monitor the furnace status, and the silicon liquid flow rate, molten phase and environmental parameters cannot be obtained synchronously, resulting in deviations in the judgment of the best furnace discharge time, which can easily cause impurity residue or energy waste. (2) Relying on fixed thresholds or empirical formulas to set parameters, there is a lack of dynamic integration of real-time data and historical smelting laws, making it difficult to adapt to complex changes in the furnace and resulting in large prediction errors. (3) Existing strategies focus on a single goal (such as the shortest time), ignoring the balance between energy consumption, efficiency and silicon purity. Uneven cooling can easily lead to cracks in silicon ingots or energy waste. (4) Traditional actuators (such as hydraulic furnace tilting systems) have large angle errors and a high risk of silicon liquid splashing; responses to abnormal working conditions rely on manual labor, the protection mechanism has a high false trigger rate, and safety hazards are prominent. Summary of the Invention
[0004] The present invention provides a full-process control method integrating intelligent perception, dynamic optimization and precise execution to improve the intelligence level of industrial production.
[0005] The technical solution adopted by the present invention is: a control method for efficient and intelligent discharge of industrial silicon, comprising the following steps:
[0006] Step 1: Real-time collection of silicon liquid temperature, flow rate, melting state data and external environmental parameters in the smelting furnace;
[0007] Step 2: Use machine learning models to integrate and analyze the collected data and predict the optimal release time;
[0008] Step 3: Generate furnace discharge parameter control instructions based on the dynamic optimization algorithm, including furnace tilting angle, furnace discharge speed and cooling rate;
[0009] Step 4: Execute instructions and provide feedback adjustments through a closed-loop control system to achieve full-process automated control.
[0010] As a further improvement of the present invention, the data acquisition adopts a multimodal sensor network, including an infrared thermal imager, an electromagnetic flow meter, a laser rangefinder and an acoustic emission sensor.
[0011] As a further improvement of the present invention, the machine learning model is a hybrid model of a convolutional neural network (CNN) and a long short-term memory network (LSTM), and the input data includes historical smelting data, real-time process parameters and environmental variables.
[0012] As a further improvement of the present invention, the dynamic optimization algorithm is a multi-objective optimization model based on reinforcement learning, and the optimization objectives include minimizing energy consumption, maximizing furnace efficiency and achieving silicon purity standards.
[0013] As a further improvement of the present invention, the tilting furnace angle control adopts an adaptive PID algorithm to adjust the servo motor torque in real time according to the silicon liquid flow rate, with an error accuracy of ≤0.5°.
[0014] As a further improvement of the present invention, the cooling rate is controlled by adjusting the spray pressure and coverage of the atomizing water cooling nozzle and predicting the cooling uniformity using a thermodynamic simulation model.
[0015] As a further improvement of the present invention, it also includes an abnormal operating condition processing module, which triggers emergency furnace tilting protection and synchronously starts a redundant cooling system when it detects that the pressure fluctuation in the furnace exceeds a threshold.
[0016] As a further improvement of the present invention, the closed-loop control system integrates digital twin technology, and realizes pre-verification of the control strategy by mapping the physical equipment status in real time through the virtual furnace body.
[0017] As a further improvement of the present invention, after the silicon ingot is taken out of the furnace, the composition of the silicon ingot is analyzed based on X-ray fluorescence spectroscopy (XRF), and the results are fed back to the model for self-learning iterative optimization.
[0018] An intelligent furnace discharge control system that implements the above method consists of a data acquisition layer, a decision-making layer, an execution layer, and a human-computer interaction layer. The data acquisition layer includes a multimodal sensor array and edge computing nodes; the decision-making layer includes an industrial server that deploys an optimization algorithm; the execution layer includes high-precision servo motors, variable-frequency water pumps, and intelligent valve groups; and the human-computer interaction layer includes an AR visualization terminal and a remote monitoring platform.
[0019] The beneficial effects of the present invention are as follows: (1) The present invention collects silicon liquid temperature, flow rate, melting state and environmental parameters in real time through a multimodal sensor network, and combines the CNN-LSTM hybrid model to dynamically analyze data characteristics, thereby significantly improving the prediction accuracy of the furnace discharge timing (deviation ≤ 3%) and avoiding impurity residue or energy waste caused by manual experience.
[0020] (2) The present invention adopts an adaptive PID algorithm to control the tilting angle of the furnace, and the servo motor torque matches the change of the silicon liquid flow rate in real time. The angle control error is ≤0.5°, which effectively suppresses the risk of silicon liquid splashing.
[0021] (3) The present invention uses XRF component analysis to provide real-time feedback on silicon ingot quality data, driving the self-learning iteration of the machine learning model, so that the furnace parameters can dynamically adapt to raw material fluctuations or equipment aging, shortening the process debugging cycle by 70%. DETAILED DESCRIPTION
[0022] In order to make the technical problems, technical solutions and beneficial effects to be solved by this application more clearly understood, this application is further described in detail below in conjunction with the embodiments. It should be understood that the embodiments described herein are only used to explain this application and are not intended to limit this application.
[0023] The present invention provides a control method for efficiently and intelligently discharging industrial silicon, comprising the following steps:
[0024] Step 1: Real-time collection of silicon liquid temperature, flow rate, melting state data and external environmental parameters in the smelting furnace;
[0025] Step 2: Use machine learning models to integrate and analyze the collected data and predict the optimal release time;
[0026] Step 3: Generate furnace discharge parameter control instructions based on the dynamic optimization algorithm, including furnace tilting angle, furnace discharge speed and cooling rate;
[0027] Step 4: Execute instructions and provide feedback adjustments through a closed-loop control system to achieve full-process automated control.
[0028] The data acquisition in the present invention adopts a multimodal sensor network, including an infrared thermal imager, an electromagnetic flow meter, a laser rangefinder and an acoustic emission sensor.
[0029] The machine learning model described in the present invention is a hybrid model of convolutional neural network (CNN) and long short-term memory network (LSTM), and the input data includes historical smelting data, real-time process parameters and environmental variables.
[0030] The dynamic optimization algorithm described in the present invention is a multi-objective optimization model based on reinforcement learning, and the optimization objectives include minimizing energy consumption, maximizing furnace efficiency and achieving silicon purity standards.
[0031] The tilting angle control in the present invention adopts an adaptive PID algorithm to adjust the servo motor torque in real time according to the silicon liquid flow rate, with an error accuracy of ≤0.5°.
[0032] The cooling rate described in the present invention is controlled by adjusting the spray pressure and coverage of the atomizing water cooling nozzle and predicting the cooling uniformity using a thermodynamic simulation model.
[0033] The present invention also includes an abnormal operating condition processing module, which triggers emergency furnace tilting protection and synchronously starts a redundant cooling system when it detects that the pressure fluctuation in the furnace exceeds a threshold.
[0034] The closed-loop control system described in the present invention integrates digital twin technology, and realizes pre-verification of the control strategy by mapping the physical equipment status in real time through the virtual furnace body.
[0035] In the present invention, after the silicon ingot is taken out of the furnace, the composition of the silicon ingot is analyzed based on X-ray fluorescence spectroscopy (XRF), and the results are fed back to the model for self-learning iterative optimization.
[0036] An intelligent furnace discharge control system that implements the above method consists of a data acquisition layer, a decision-making layer, an execution layer, and a human-computer interaction layer. The data acquisition layer includes a multimodal sensor array and edge computing nodes; the decision-making layer includes an industrial server that deploys an optimization algorithm; the execution layer includes high-precision servo motors, variable-frequency water pumps, and intelligent valve groups; and the human-computer interaction layer includes an AR visualization terminal and a remote monitoring platform.
[0037] Example:
[0038] In a certain industrial silicon smelting plant, the control method of the present invention is applied to carry out intelligent control of the entire process of silicon liquid discharge, and the specific implementation is as follows.
[0039] (1) Multimodal data acquisition
[0040] An infrared thermal imager (resolution 0.1°C) is installed on the side wall of the smelting furnace, an electromagnetic flowmeter (range 0-50L / s, accuracy ±0.5%) is embedded in the furnace bottom, a laser rangefinder (sampling frequency 100Hz) is configured at the furnace mouth to monitor the silicon liquid level, and an acoustic emission sensor (frequency band 20kHz-1MHz) is arranged outside the furnace body to capture abnormal signals of the melting state. At the same time, the ambient temperature and humidity (range -10°C to 50°C) and dust concentration (detection limit 0.1mg / m 3 ).
[0041] (2) Model prediction and decision-making
[0042] Real-time data (including historical 100 heat smelting data) is fed into a CNN-LSTM hybrid model (CNN layer count: 3, LSTM unit count: 128). The model outputs a prediction for the optimal furnace discharge timing. The discharge command is triggered when the silicon liquid temperature reaches 1680±10°C, the flow rate stabilizes at 12L / s, and the melt phase is uniform.
[0043] (3) Dynamic parameter optimization
[0044] Based on the reinforcement learning algorithm (Q-learning framework, reward function weights: energy consumption 0.4, efficiency 0.3, purity 0.3), the control parameters are generated: the tilting angle is set to 35.5°, the discharge speed is 8 L / s, and the cooling rate is 15°C / s.
[0045] (4) Precise execution and feedback
[0046] A servo motor (rated torque 200 N·m, response time ≤ 50 ms) adjusts the furnace tilt angle using an adaptive PID algorithm (proportional coefficient Kp = 2.5, integral time Ti = 0.1 s), with actual angle fluctuation ≤ 0.3°. Atomizing water-cooling nozzles (adjustable pressure 0.5-2 MPa) adjust the spray coverage area (distributed in a 1.2 m diameter ring) based on thermodynamic simulation results, achieving cooling uniformity deviation of <5%.
[0047] (V) Exception handling and self-optimization
[0048] When the furnace pressure suddenly increased to 15kPa (threshold 12kPa), the system initiated redundant cooling (opening two additional sets of nozzles) within 3 seconds and tilted the furnace back to a safe angle of 10°. After exiting the furnace, XRF testing of the silicon ingots revealed a purity of 99.92%. This data was fed back to the model to update the parameter library, reducing the prediction error for the next heat to 2.1%.
[0049] It can be seen from the above embodiments that the present invention can effectively cope with complex working conditions such as fluctuations in raw material composition through multi-source data fusion, digital twin pre-verification and model self-learning mechanism, while significantly reducing energy consumption and manual intervention intensity while ensuring silicon purity.
[0050] In summary, the control method for efficient and intelligent discharge of industrial silicon according to the present invention exhibits high flexibility and safety when faced with abnormal working conditions. The built-in exception handling module of the system can quickly respond to emergencies such as abnormal pressure in the furnace, and effectively avoid potential safety hazards through the synchronous start-up of emergency furnace tilting protection and redundant cooling systems. In addition, the system can also perform component analysis on silicon ingots after discharge based on X-ray fluorescence spectroscopy (XRF) technology, feed back real-time quality data to the machine learning model, and drive the model to perform self-learning iterative optimization. This process not only improves the dynamic adaptability of discharge parameters, enabling the system to more accurately respond to challenges brought about by raw material fluctuations or equipment aging, but also significantly shortens the process debugging cycle and improves production efficiency.
[0051] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.
Claims
1. A control method for efficient and intelligent discharge of industrial silicon, characterized in that: The following steps are involved: Step 1: Real-time collection of silicon liquid temperature, flow rate, melting state data and external environmental parameters in the smelting furnace; Step 2: Use machine learning models to integrate and analyze the collected data and predict the optimal release time; Step 3: Generate furnace discharge parameter control instructions based on the dynamic optimization algorithm, including furnace tilting angle, furnace discharge speed and cooling rate; Step 4: Execute instructions and provide feedback adjustments through a closed-loop control system to achieve full-process automated control.
2. The control method for efficient and intelligent discharge of industrial silicon according to claim 1 is characterized in that: The data collection adopts a multimodal sensor network, including an infrared thermal imager, an electromagnetic flow meter, a laser rangefinder and an acoustic emission sensor.
3. The control method for efficient and intelligent discharge of industrial silicon according to claim 1 is characterized in that: The machine learning model is a hybrid model of convolutional neural network (CNN) and long short-term memory network (LSTM), and the input data includes historical smelting data, real-time process parameters and environmental variables.
4. The method for controlling efficient and intelligent discharge of industrial silicon according to claim 1, characterized in that: The dynamic optimization algorithm is a multi-objective optimization model based on reinforcement learning, and the optimization objectives include minimizing energy consumption, maximizing furnace efficiency and achieving silicon purity standards.
5. The control method for efficient and intelligent discharge of industrial silicon according to claim 1 is characterized in that: The tilting furnace angle control adopts an adaptive PID algorithm to adjust the servo motor torque in real time according to the silicon liquid flow rate, with an error accuracy of ≤0.5°.
6. The method for controlling efficient and intelligent discharge of industrial silicon according to claim 1, characterized in that: The cooling rate is controlled by adjusting the spray pressure and coverage of the atomizing water cooling nozzle and predicting the cooling uniformity using a thermodynamic simulation model.
7. The method for controlling efficient and intelligent discharge of industrial silicon according to claim 1, characterized in that: It also includes an abnormal operating condition processing module, which triggers emergency furnace tilting protection and synchronously starts the redundant cooling system when it detects that the pressure fluctuation in the furnace exceeds the threshold.
8. The method for controlling efficient and intelligent discharge of industrial silicon according to claim 1, characterized in that: The closed-loop control system integrates digital twin technology, which maps the physical equipment status in real time through the virtual furnace body to achieve pre-verification of the control strategy.
9. The method for controlling efficient and intelligent discharge of industrial silicon according to claim 1, characterized in that: After the silicon ingot is taken out of the furnace, its composition is analyzed based on X-ray fluorescence spectroscopy (XRF), and the results are fed back to the model for self-learning iterative optimization.
10. An intelligent tapping control system for implementing the method according to any one of claims 1 to 9, characterized in that: It consists of data acquisition layer, decision-making layer, execution layer, and human-computer interaction layer. The data acquisition layer includes multimodal sensor arrays and edge computing nodes; the decision-making layer includes industrial servers that deploy optimization algorithms; the execution layer includes high-precision servo motors, variable-frequency water pumps, and intelligent valve groups; and the human-computer interaction layer includes AR visualization terminals and remote monitoring platforms.