A precise AI control method to prevent atomization in reduction furnaces

Abstract

This invention discloses a precise AI control method for preventing atomization in reduction furnaces, relating to the field of machine maintenance. Addressing the problems of existing technologies such as detection lag, strong coupling of multiple variables, and limited monitoring dimensions, this method achieves precise control through three steps: first, furnace condition coordination and multi-dimensional data acquisition, collecting real-time DCS parameters, dynamic data on silicon rod growth, and historical data; second, AI atomization prediction and intelligent parameter correction, relying on a trained AI model to analyze data and output optimal correction values, with a control response time ≤2s; and third, scenario-specific precise control and multi-variable collaborative optimization, covering four scenarios including sudden excessive atomization suppression and proportion imbalance control. This invention achieves preventative atomization control, improves raw material conversion rate, reduces energy consumption and labor costs, and ensures efficient and stable operation of the reduction reaction.

Description

Technical Field

[0001] This invention relates to the field of machine maintenance, and more specifically to a precise AI control method for preventing atomization in a reduction furnace. Background Technology

[0002] During the reduction furnace production process, numerous controllable variables influence the silicon core growth outcome, including the flow rate of TCS, the flow rate of hydrogen, the molar ratio of TCS to hydrogen, furnace pressure, furnace temperature, the flow rates of various cooling waters, and the power adjustment current and voltage. Loss of control over any of these variables will directly affect the silicon core growth result. Excessively high furnace temperatures cause trichlorosilane to react in the gas phase, generating silicon powder. This silicon powder floats in the gas phase, causing air pollution.

[0003] The existing reduction furnace control technology has the following shortcomings:

[0004] First, the high-temperature extreme environment and detection lag: the surface of the silicon rod is 1050-1080℃, and the gas phase degree inside the furnace is high, making it impossible to directly set up points to measure atomization. Second, strong coupling of multiple variables and fuzzy critical boundaries: strong coupling of TCS, H2 flow rate, temperature, pressure, rod diameter / spacing, etc. Third, the monitoring dimensions are limited, lacking real-time detection of the surface temperature of silicon rods and the gas phase temperature inside the furnace. Problems such as delayed fault warnings and nozzle blockage can easily cause atomization.

[0005] To address the aforementioned issues, this invention provides an AI control technology for a reduction furnace atomization that enables "real-time monitoring, intelligent regulation, and collaborative adaptation," thereby improving the production stability and accuracy of the reduction furnace, preventing atomization caused by excessively high gas phase temperature inside the furnace, and ensuring efficient and stable operation of the reduction reaction. Summary of the Invention

[0006] This invention proposes a precise AI control method to prevent atomization in a reduction furnace, the method comprising the following steps: S1. Furnace condition coordination and multi-dimensional data acquisition; Collect comprehensive operational data of the reduction furnace, including but not limited to: Key parameters of DCS in real time mainly include furnace temperature, pressure, TCS flow rate, hydrogen flow rate, various cooling water flow rates, power adjustment current, voltage, and exhaust gas dust concentration. The dynamic data of silicon rod growth mainly includes silicon rod diameter, spacing, surface condition, and growth rate; Historical data, including historical atomization case data, past process parameter adjustment records, and corresponding operational results data.

[0007] The aforementioned data is used, on the one hand, to correct deviations in the mechanism model and compensate for interference factors that are difficult to quantify precisely in the mechanism model, such as uneven local flow field in the furnace, changes in silicon rod thermal radiation, and fluctuations in raw material ratio; on the other hand, it is used to capture the dynamic trend of atomization precursors in real time, providing data support for subsequent risk prediction.

[0008] S2.AI atomization prediction and intelligent parameter correction; Based on an AI atomization prediction model trained on a large amount of historical operating data, the multi-dimensional monitoring data collected by S1 is quickly analyzed to accurately identify problems such as atomization parameter deviation and critical risk points. The model outputs the optimal correction values ​​for medium pressure, flow rate, and raw material feed in real time, automatically triggering dynamic adjustment of the feed parameters of the reduction furnace. The control response time is ≤2s, and the furnace reaction state can be kept stable without manual intervention. This avoids the lag and error of traditional manual experience adjustment and prevents atomization of the reduction furnace reaction from the source.

[0009] S3. Precise control and multi-variable collaborative optimization for different scenarios; Based on different production conditions and process stages, targeted control strategies are implemented, which mainly include the following four application scenarios: S3.1 Suppression of sudden over-atomization regulation By using a digital twin model to predict atomization trends in real time, a dual cooling regulation system combining chemical and physical methods is initiated. Chemical regulation includes increasing hydrogen flow rate, rapidly diluting TCS concentration, suppressing reaction rate, and curbing atomization intensification within tens of seconds; Physical regulation includes AI synchronously assessing energy input requirements. If coordinated regulation is needed, the voltage is reduced to decrease energy input, thus providing dual protection against atomization suppression.

[0010] S3.2 Imbalance in proportions and regulation of low sedimentation rate When the silicon powder concentration is consistently below the lower limit, silicon rod growth is slow, and the surface is too bright (high temperature but insufficient reactants): Reduce the hydrogen flow rate in stages (1-2% each time) and slowly increase the effective TCS concentration to stimulate the reaction. AI monitors the recovery of silicon powder concentration and changes in furnace temperature in real time, dynamically fine-tuning the adjustment range to prevent overshooting of parameters and the risk of atomization.

[0011] S3.3 Adaptive Control of Process Stages Based on the characteristics of different stages in the entire production cycle of the reduction furnace, a differentiated hydrogen flow control strategy was formulated: During the furnace start-up stage, given the characteristics of the silicon core being thin and having a small surface area, a "high hydrogen flow rate" strategy is adopted to create a highly diluted reaction environment and strictly suppress atomization, laying a dense and pure silicon core foundation for subsequent growth. During the rapid deposition period, the silicon rod is already thick and has a large surface area. On the premise of ensuring no atomization occurs, gradually optimize and reduce the hydrogen / silicon ratio (relatively reduce hydrogen) to improve the raw material utilization rate and deposition speed. During the finishing stage, to prevent excessive internal stress in the silicon rod and deterioration of the surface quality, moderately increase hydrogen to end the reaction smoothly.

[0012] S3.4 Multivariable collaborative optimization control In response to parameter contradictions under complex working conditions (such as a decrease in deposition rate and high atomization), the AI performs multivariable linkage control without separately adjusting a single parameter: Simultaneously execute three adjustments within 120 seconds - reduce the hydrogen flow rate in line A by 1.5% (enhance the reaction power), slightly increase the corresponding silicon core voltage by 0.2V (compensate for the heat loss caused by the change in gas composition and maintain energy balance), and slightly reduce the total TCS flow rate by 1% (prevent excessive reaction); Through the collaborative linkage of multiple parameters, construct a dynamically optimal chemical reaction environment, taking into account both production efficiency and atomization suppression effect.

[0013] Compared with the prior art, the beneficial effects of the present invention include: (1) Improve the raw material utilization rate and reduce the loss cost. The AI can dynamically adjust the atomization parameters according to the raw material characteristics and the furnace working conditions, avoiding waste of raw materials caused by uneven reactions. For example, in a polysilicon reduction furnace, the AI can regulate the raw material ratio in stages, eliminating the loss caused by excessive generation of amorphous silicon powder in the later stage and increasing the effective conversion rate.

[0014] (2) Reduce energy consumption expenditure and cut down the energy cost. The polysilicon reduction furnace can reduce the ineffective energy consumption caused by abnormal atomization through the collaborative control of atomization and the furnace temperature and pressure by the AI, further compressing the energy expenditure.

[0015] (3) Reduce labor and maintenance costs and improve operation efficiency. Traditional atomization control relies on technicians to manually adjust parameters based on experience. The AI can achieve full-process automated closed-loop control, saving time for technicians every day and greatly reducing the labor burden.

[0016] (4) From macro to micro. It goes beyond the macro means of simply controlling temperature (voltage), directly intervenes in the micro world of chemical reaction kinetics, and the control is more fundamental.

[0017] (5) From lag to foresight. Based on the prediction model, the AI can adjust hydrogen in advance when the atomization trend deviates slightly, rather than making large adjustments when it becomes serious, achieving preventive control.

[0018] (6) From single-point to global chemical equilibrium. The AI understands that the change in hydrogen will affect a series of side reactions and gas-phase compositions, and its adjustment is to maintain a dynamic and optimal chemical environment. Attached Figure Description

[0019] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0020] Figure 1 This is an overall flowchart of the method of the present invention.

[0021] Figure 2 This is a diagram of the intelligent parameter correction interface for the method of the present invention. Detailed Implementation

[0022] Example 1 The overall flowchart of the precise AI control method for preventing atomization in a reduction furnace described in this invention is as follows: Figure 1 As shown, the specific steps are as follows: S1. Furnace condition coordination and multi-dimensional data acquisition; Collect comprehensive operational data of the reduction furnace, including but not limited to: furnace temperature, pressure, TCS flow rate, hydrogen flow rate, various cooling water flow rates, power adjustment current, voltage, exhaust gas dust concentration; silicon rod diameter, spacing, surface condition, growth rate; historical atomization case data, past process parameter adjustment records, and corresponding operational results data.

[0023] The aforementioned data is used, on the one hand, to correct deviations in the mechanism model and compensate for interference factors that are difficult to quantify precisely in the mechanism model, such as uneven local flow field in the furnace, changes in silicon rod thermal radiation, and fluctuations in raw material ratio; on the other hand, it is used to capture the dynamic trend of atomization precursors in real time, providing data support for subsequent risk prediction.

[0024] S2.AI atomization prediction and intelligent parameter correction; Based on an AI atomization prediction model trained on a large amount of historical operating data, the multi-dimensional monitoring data collected by S1 is quickly analyzed to accurately identify problems such as atomization parameter deviation and critical risk points. The model outputs the optimal correction values ​​for medium pressure, flow rate, and raw material feed in real time, automatically triggering dynamic adjustment of the feed parameters of the reduction furnace. The control response time is ≤2s, and the furnace reaction state can be kept stable without manual intervention. This avoids the lag and error of traditional manual experience adjustment and prevents atomization of the reduction furnace reaction from the source.

[0025] S3. Precise control and multi-variable collaborative optimization for different scenarios; Based on different production conditions and process stages, targeted control strategies are implemented, which mainly include the following four application scenarios: S3.1 Suppression of sudden over-atomization regulation By using a digital twin model to predict atomization trends in real time, a dual cooling regulation system combining chemical and physical methods is initiated. Chemical regulation includes increasing the hydrogen flow rate, rapidly diluting the TCS concentration, depressing the reaction rate, and curbing the exacerbation of atomization within dozens of seconds; Physical regulation includes AI synchronously evaluating the energy input requirements. If collaborative regulation is needed, the voltage is reduced to decrease the energy input, providing dual guarantees for the atomization inhibition effect.

[0026] S3.2 Regulation of ratio imbalance and low deposition rate When it is monitored that the silicon powder concentration continuously remains below the lower limit, the silicon rod grows slowly and the surface is overly bright (high temperature but insufficient reactants): Reduce the hydrogen flow rate in stages (with an adjustment range of 1 - 2% each time), and slowly increase the effective concentration of TCS to stimulate the reaction; AI monitors the recovery of the silicon powder concentration and the change of the furnace temperature in real time, dynamically fine-tuning the adjustment range to prevent overshoot of parameter adjustment and resulting in atomization risks.

[0027] S3.3 Adaptive regulation in the process stage According to the characteristics of different stages in the full production cycle of the reduction furnace, formulate a differential hydrogen flow rate regulation strategy: In the furnace starting stage, aiming at the characteristics of thin silicon cores and small surface areas, adopt the "higher hydrogen flow rate" strategy to construct a highly diluted reaction environment, strictly inhibit atomization, and lay a dense and pure silicon nucleus foundation for subsequent growth; During the rapid deposition period, the silicon rod is already thick and has a large surface area. On the premise of ensuring no atomization occurs, gradually optimize and reduce the hydrogen / silicon ratio (relatively reduce hydrogen) to improve the raw material utilization rate and deposition speed; In the finishing stage, to prevent excessive internal stress in the silicon rod and deterioration of the surface quality, moderately increase hydrogen to end the reaction smoothly.

[0028] S3.4 Multi-variable collaborative optimization regulation Aiming at parameter contradictions under complex working conditions (such as a decrease in deposition rate and high atomization), AI performs multi-variable linkage regulation and does not adjust a single parameter alone. Through multi-parameter collaborative linkage, construct a dynamically optimal chemical reaction environment, taking into account both production efficiency and atomization inhibition effect.

[0029] Example 2 This example is applied to a 60-pair rod polysilicon reduction furnace for producing electronic-grade polysilicon raw materials. The goal is to solve the atomization problem caused by multi-parameter coupling and detection lag during the production process, while improving the raw material conversion rate and production stability. The production raw materials are industrial-grade trichlorosilane and high-purity hydrogen. The normal working temperature range of the reduction furnace is 1050 - 1080 °C, the working pressure is 0.3 - 0.5 MPa, and the designed silicon rod growth cycle is 120 hours.

[0030] S1. System initialization and model loading Start the reduction furnace and its supporting equipment, including the DCS control system, AI control server, various sensors and actuators, and perform equipment self-checks to ensure that all equipment is operating normally, sensor data acquisition is stable, and actuators are responsive.

[0031] The pre-trained AI atomization prediction model (trained based on 1,000 batches of historical production data, covering atomization cases under different raw material ratios, temperature and pressure conditions) and the digital twin model of the reduction furnace (based on SolidWorks modeling and calibrated with Fluent fluid simulation data, which can map the flow field and temperature field distribution inside the furnace in real time) are loaded into the AI ​​control server.

[0032] Set the basic process parameters: initial TCS flow rate 30kg / h, initial total hydrogen flow rate 80Nm³ / h (40Nm³ / h for channel A and 40Nm³ / h for channel B), hydrogen-silicon molar ratio 10:1, initial furnace pressure 0.4MPa, initial power adjustment voltage 1495V (corresponding to the silicon core preheating requirement during the furnace start-up stage), atomization warning threshold (silicon powder concentration in the exhaust gas ≥8mg / m³), and silicon powder concentration lower limit threshold (≤3mg / m³).

[0033] S2. Real-time acquisition of multi-dimensional data The following data is collected by the DCS control system at a frequency of 10Hz and transmitted to the AI ​​control server in real time: Temperature data: silicon rod surface temperature, furnace gas phase temperature; Pressure data: Real-time pressure inside the furnace; Flow data: TCS real-time flow, A / B line hydrogen real-time flow, cooling water real-time flow; Atomization characterization data: silica powder concentration in exhaust gas; Silicon rod growth data: real-time diameter and growth rate of silicon rod; Electrical parameter data: real-time power adjustment voltage and current. Simultaneously, the server accesses historical atomization case data (including parameter change curves from 50 typical atomization failures) to correct deviations in the mechanism model and compensate for interference factors such as uneven local flow fields within the furnace and changes in silicon rod thermal radiation.

[0034] S3. AI Atomization Prediction and Intelligent Parameter Correction The AI ​​atomization prediction model quickly analyzes the real-time data collected by S2, and combines historical data with simulation results from the digital twin model to identify parameter deviations and atomization risk trends.

[0035] For different production conditions, the system dynamically outputs optimal correction parameters, which are sent to the DCS system via a communication module to drive the actuators to complete parameter adjustments. The control response time is ≤2 seconds, and the entire process requires no manual intervention. The parameter adjustment page is as follows: Figure 2 As shown.

[0036] S4. Precise control and execution based on different scenarios (1) Start-up stage (0-10 hours) Operating characteristics: The silicon core is thin (8-15mm in diameter) and has a small surface area. It is prone to atomization due to excessively vigorous reaction, so a dense silicon core foundation is required.

[0037] Control strategy: AI maintains a "high hydrogen flow rate", stabilizing the total hydrogen flow rate at 85 Nm³ / h (42.5 Nm³ / h for A path and 42.5 Nm³ / h for B path), increasing the hydrogen-silicon molar ratio to 12:1, and highly diluting the TCS concentration; at the same time, the power regulation voltage is controlled to gradually increase from 1495V to 1865V (increasing by 37V per hour) to avoid local overheating of the silicon core.

[0038] Monitoring results: The concentration of silicon powder in the exhaust gas remained stable at 2-3 mg / m³, with no atomization phenomenon. The diameter of the silicon rod increased uniformly to 15 mm, and the silicon core structure was dense.

[0039] (2) Rapid deposition period (10-100 hours) Operating conditions: The diameter of the silicon rod has reached 15-80mm, the surface area has increased, the deposition rate needs to be increased, and the risk of fogging needs to be avoided.

[0040] Control Strategy: The AI ​​gradually optimizes the hydrogen / silicon ratio, reducing the total hydrogen flow rate by 1 Nm³ / h every 5 hours (synchronous adjustment of A / B circuits), gradually decreasing the hydrogen-silicon molar ratio from 12:1 to 8:1; simultaneously, the TCS flow rate is gradually increased from 30 kg / h to 45 kg / h, and the power regulation voltage is gradually increased from 1865V to 2119V (corresponding to the reference voltage curve). During the control period, the silicon powder concentration in the exhaust gas is monitored in real time. When the concentration approaches 5 mg / m³, the reduction of hydrogen flow rate is paused, and the current parameters are maintained for 1 hour before further adjustment to prevent overreaction.

[0041] Monitoring results: The growth rate of silicon rods was stable at 0.8-1.0 mm / h, the raw material utilization rate was improved, the concentration of silicon powder in the tail gas was always controlled at 3-6 mg / m³, no atomization occurred, and the diameter of silicon rods increased uniformly to 80 mm.

[0042] (3) Sudden over-atomization suppression (80th hour of production) Sudden change in operating conditions: Due to fluctuations in the purity of the raw material TCS (instantly dropping to 99.8%), the reaction rate inside the furnace suddenly increased, and the concentration of silicon powder in the tail gas rose from 5 mg / m³ to 12 mg / m³ within 30 seconds, triggering an atomization warning.

[0043] Regulation Strategy: The digital twin model predicts that fogging will continue to worsen, and AI immediately initiates a dual cooling approach combining chemical and physical methods. The chemical control involves simultaneously increasing the hydrogen flow rate of both A and B paths by 5 Nm³ / h (total flow rate increased from 60 Nm³ / h to 70 Nm³ / h) to rapidly dilute the TCS concentration and suppress the reaction rate. The physical control involves AI assessing excessive energy input and reducing the power adjustment voltage from 2034V to 2000V to decrease energy input.

[0044] The effect of the adjustment was that the concentration of silicon powder in the exhaust gas dropped to 8 mg / m³ after 30 seconds and to 5 mg / m³ after 60 seconds, and the atomization trend was suppressed. Subsequently, the hydrogen flow rate was maintained at 70 Nm³ / h and the voltage at 2000 V for 2 hours. After the TCS purity recovered, the original process parameters were gradually restored.

[0045] (4) Adjustment of imbalanced ratio and low deposition rate (90th hour of production) Operating characteristics: The silicon powder concentration was continuously lower than 3 mg / m³ (dropped to 2 mg / m³), the silicon rod growth rate dropped to 0.6 mm / h, and the silicon rod surface was too bright (the surface temperature rose to 1075℃). It was determined that the hydrogen ratio was too high and the effective concentration of TCS was insufficient.

[0046] Control strategy: The AI ​​reduces the hydrogen flow rate in stages. In the first stage (0-30 minutes), the total hydrogen flow rate is reduced by 1% (from 65 Nm³ / h to 64.35 Nm³ / h), and in the second stage (30-60 minutes), it is reduced by another 1% (to 63.7 Nm³ / h). At the same time, the TCS flow rate is kept constant at 45 kg / h and the power adjustment voltage is kept constant at 2080 V. The silicon powder concentration and temperature changes are monitored in real time.

[0047] Control effect: After 60 minutes, the silicon powder concentration rose back to 3.5 mg / m³, the silicon rod growth rate recovered to 0.8 mm / h, and the surface temperature dropped to 1060℃. There was no risk of atomization caused by over-adjustment.

[0048] (5) Multivariate collaborative optimization (100th hour of production) Conflicting operating conditions: The deposition rate drops to 0.7 mm / h, while the concentration of silica powder in the exhaust gas rises to 7 mg / m³ (close to the warning threshold). Adjusting a single parameter cannot simultaneously "increase the deposition rate" and "suppress atomization".

[0049] Control strategy: AI executes multi-variable linkage control, simultaneously completing three adjustments within 120 seconds: 1. Hydrogen flow rate in route A is reduced by 1.5% (from 32 Nm³ / h to 31.52 Nm³ / h). 2. The corresponding silicon core power regulation voltage increases slightly by 0.2V (from 2080V to 2080.2V). 3. Total TCS flow rate decreased slightly by 1% (from 45 kg / h to 44.55 kg / h).

[0050] Reduce hydrogen to increase reaction kinetics, slightly increase voltage to compensate for heat loss, and slightly decrease TCS to prevent overreaction.

[0051] After adjustment, the silicon powder concentration stabilized at 5 mg / m³, and the deposition rate rebounded to 0.8 mm / h, achieving a dynamic balance under contradictory operating conditions.

[0052] (6) Final stage (100-120 hours) Operating conditions: The diameter of the silicon rod has reached the target value (100mm), and it is necessary to avoid excessive internal stress and deterioration of surface quality.

[0053] Control strategy: The AI ​​will moderately increase the hydrogen flow rate, increasing the total hydrogen flow rate by 2 Nm³ / h every 5 hours, gradually increasing it from 60 Nm³ / h to 75 Nm³ / h, restoring the hydrogen-silicon molar ratio to 10:1; at the same time, the TCS flow rate will be gradually reduced from 45 kg / h to 30 kg / h, and the power regulation voltage will be gradually reduced from 2080V to 1544V.

[0054] Monitoring results: The silicon rod surface was uniform, without cracks or burrs, the internal stress test was qualified, the silicon powder concentration in the exhaust gas was stable at 3-4 mg / m³, and the reaction ended peacefully.

Claims

1. A precise AI control method for preventing atomization in a reduction furnace, characterized in that, Includes the following steps: S1. Furnace condition coordination and multi-dimensional data acquisition: Collects full-dimensional operation data of the reduction furnace, covering real-time key parameters of DCS, dynamic data of silicon rod growth, and historical data. S2.AI atomization prediction and intelligent parameter correction: Based on an AI atomization prediction model trained with a large amount of historical operating data, it analyzes the monitoring data, outputs the optimal correction value, and automatically triggers parameter adjustment; S3. Precise control and multi-variable collaborative optimization for different scenarios: Targeted control strategies are implemented based on different production conditions and process stages. Preventive control of atomization is achieved through full-process automated closed-loop control.

2. The precise AI control method for preventing atomization in a reduction furnace according to claim 1, characterized in that, The key DCS real-time parameters in step S1 include furnace temperature, pressure, TCS flow rate, hydrogen flow rate, various cooling water flow rates, power adjustment current, voltage, and exhaust gas dust concentration. The dynamic data of silicon rod growth includes silicon rod diameter, spacing, surface condition, and growth rate; Historical data includes historical atomization case data, past process parameter adjustment records, and corresponding operational results data.

3. The precise AI control method for preventing atomization in a reduction furnace according to claim 1, characterized in that, The data collected in step S1 is used, on the one hand, to correct the deviation of the mechanism model and compensate for interference factors that are difficult to quantify accurately in the mechanism model, such as uneven local flow field in the furnace, changes in silicon rod thermal radiation, and fluctuations in raw material ratio. On the other hand, it captures the dynamic trends of precursors to atomization in real time, providing data support for risk prediction.

4. The precise AI control method for preventing atomization in a reduction furnace according to claim 1, characterized in that, The scenario-specific precise control in step S3 includes control to suppress sudden excessive atomization, control to address imbalances and low deposition rates, adaptive control at different process stages, and multi-variable synergistic optimization control.

5. The precise AI control method for preventing atomization in a reduction furnace according to claim 4, characterized in that, The method for suppressing sudden excessive atomization involves predicting the atomization development trend in real time using a digital twin model and initiating dual chemical and physical cooling regulation. The chemical regulation includes increasing the hydrogen flow rate, rapidly diluting the TCS concentration, suppressing the reaction rate, and curbing the atomization intensification within tens of seconds. The physical regulation includes AI synchronously assessing energy input requirements. If coordinated regulation is needed, the voltage is reduced to decrease energy input, thus providing dual protection for the atomization suppression effect.

6. The precise AI control method for preventing atomization in a reduction furnace according to claim 4, characterized in that, The imbalance in the ratio and the low deposition rate are controlled by reducing the hydrogen flow rate in stages and slowly increasing the effective concentration of TCS to stimulate the reaction when the silicon powder concentration is continuously below the lower limit, the silicon rod growth is slow and the surface is too bright. AI monitors the recovery of silicon powder concentration and changes in furnace temperature in real time, dynamically fine-tuning the adjustment range to prevent overshooting of parameters and the risk of atomization.

7. The precise AI control method for preventing atomization in a reduction furnace according to claim 4, characterized in that, The aforementioned adaptive control of the process stages involves formulating differentiated hydrogen flow control strategies based on the characteristics of different stages throughout the entire production cycle of the reduction furnace. During the furnace start-up stage, given the characteristics of the silicon core being thin and having a small surface area, a high hydrogen flow rate strategy is adopted to create a highly diluted reaction environment and strictly suppress atomization, laying a dense and pure silicon core foundation for subsequent growth. During the rapid deposition period, the silicon rod is already thick and has a large surface area. Under the premise of ensuring that atomization does not occur, the hydrogen / silicon ratio is gradually reduced to improve the raw material utilization rate and deposition rate. To prevent excessive internal stress and surface quality deterioration of the silicon rod during the final stage, hydrogen gas is added to ensure a smooth and gradual end to the reaction.

8. The precise AI control method for preventing atomization in a reduction furnace according to claim 4, characterized in that, The multivariate collaborative optimization control addresses parameter contradictions under complex operating conditions. The AI ​​simultaneously performs multi-parameter linkage adjustments within 120 seconds, without adjusting a single parameter individually, thus constructing a dynamically optimal chemical reaction environment.

9. The precise AI control method for preventing atomization in a reduction furnace according to claim 1, characterized in that, The AI ​​atomization prediction in step S2 also includes system initialization and model loading steps, starting the reduction furnace and supporting equipment and performing self-checks, loading the pre-trained AI atomization prediction model and the digital twin model of the reduction furnace, and setting basic process parameters, atomization warning threshold and silicon powder concentration lower limit threshold.

10. The precise AI control method for preventing atomization in a reduction furnace according to claim 1, characterized in that, The AI ​​atomization prediction model in step S2 is trained based on more than 1,000 batches of historical production data, covering atomization cases under different raw material ratios, temperature and pressure conditions.

Images