Gas blocking type vacuum pump high speed drive shaft end sealing method

Abstract

This invention relates to the field of pump body sealing technology, and more particularly to a method for sealing the high-speed drive shaft end of a gas-blocking vacuum pump. The technical solution is based on a sealing device comprising a pump chamber and a rotor. The rotor is driven by a drive motor via an input shaft and is located within the pump chamber. A bearing housing supports the input shaft. A main chamber tritium concentration monitoring gauge is located inside the pump chamber, and a side chamber tritium concentration monitoring gauge is located outside the bearing housing. A purge port connects to the shaft end sealing gap. A gas cylinder is connected to the purge port via a control valve. An AI controller is connected to the main chamber tritium concentration monitoring gauge, the side chamber tritium concentration monitoring gauge, the drive motor, the control valve, and each sensing element. This invention utilizes AI-driven closed-loop adaptive control to dynamically match operating conditions and adjust the blowing parameters, effectively blocking tritium diffusion and improving the accuracy and stability of leakage control in extreme environments.

Description

Technical Field

[0001] This invention relates to the field of pump body sealing technology, and in particular to a method for sealing the high-speed drive shaft end of a gas-blocking vacuum pump. Background Technology

[0002] Future controlled nuclear fusion will use the deuterium-tritium reaction as its core reaction pathway. The vacuum maintenance and media transport in tokamak devices and tritium circulation systems rely on key tritium-related equipment such as Roots pumps, screw vacuum pumps, and high-speed centrifugal pumps, often employing high-speed shaft direct drive or gear transmission. Therefore, the sealing performance of critical components such as the high-speed drive shaft, chamber connection surfaces, and gearbox-pump body mating surfaces directly determines the operational safety and reliability of the fusion device. The sealing technology of the high-speed drive shaft end, as the main channel for media leakage, is particularly crucial.

[0003] Tritium, as the core medium in fusion reactions, possesses extremely high permeability and chemical reactivity. It readily combines with hydrogen and oxygen in the environment to form tritized water. Leakage not only poses long-term radiation hazards to operators but also significantly increases the cost of detritium removal from equipment, seriously threatening the safe and stable operation of the fusion system. Furthermore, during fusion operation, tritium-related pumps must withstand extreme conditions such as prolonged exposure to strong radiation, ultra-wide extreme temperature fluctuations, and ultra-high precision sealing constraints. The high-speed shaft system operates over a wide range of rotational speeds, and gear transmission is accompanied by continuous periodic vibrations. These multiple complex conditions, coupled with tritium's high permeability, significantly increase the difficulty of achieving a seal at the high-speed drive shaft end.

[0004] Currently, the sealing technologies used in the pumps and chambers of fusion tritium systems are mainly conventional dynamic sealing structures such as mechanical seals, lip seals, and labyrinth seals, combined with general static sealing components such as O-rings and metal gaskets. These technologies lack customized designs for the extreme and unique operating conditions of tritium fusion, making it difficult to address sealing failures caused by multi-field coupling. Several core defects are prevalent: First, insufficient radiation resistance and compatibility with tritium-contaminated media; long-term exposure to strong radiation and highly corrosive tritium environments easily leads to aging and failure of seals, making long-term stable sealing impossible. Second, poor stability of high-speed dynamic seals; difficulty adapting to the wide range of speed changes and periodic vibrations of high-speed shaft systems; and a tendency for abnormal increases in sealing gaps, leading to tritium leakage. Third, weak collaborative control capabilities for multiple leakage channels; and an inability to accurately monitor and control critical leakage channels such as shaft end sealing gaps.

[0005] More importantly, existing sealing structures can only achieve passive sealing functions and lack the ability to perceive multi-dimensional parameters, predict leakage risks, and adaptively adjust, thus failing to respond in real time to dynamic changes under extreme operating conditions. Therefore, addressing the core requirement of sealing the high-speed drive shaft end in fusion tritium scenarios, this paper proposes a gas-blocking vacuum pump high-speed drive shaft end sealing method. Summary of the Invention

[0006] The purpose of this invention is to address the shortcomings of existing technologies, such as poor radiation resistance, unstable dynamic sealing, and lack of active control capability, by proposing a gas-blocking vacuum pump high-speed drive shaft end sealing method.

[0007] The technical solution of this invention is as follows: A method for sealing the high-speed drive shaft end of a gas-blocking vacuum pump, based on a sealing device. The sealing device includes a pump chamber and a rotor. The rotor is driven by a drive motor through an input shaft and is located inside the pump chamber. A bearing seat supports the input shaft. A main chamber tritium concentration monitoring gauge is located inside the pump chamber, and a side chamber tritium concentration monitoring gauge is located outside the bearing seat. A purge port is connected to the shaft end sealing gap. A gas cylinder is connected to the purge port through a control valve. An AI controller is connected to the main chamber tritium concentration monitoring gauge, the side chamber tritium concentration monitoring gauge, the drive motor, the control valve, and each sensing element.

[0008] The sealing method includes the following steps:

[0009] S1. Multi-dimensional sealing data acquisition: Utilizing radiation-resistant and corrosion-resistant special sensing elements deployed in key parts of the pump cavity, bearing housing, and shaft system, combined with the tritium concentration monitoring gauges in the main cavity and the side cavity, multi-dimensional data such as pump cavity pressure, tritium concentration on both sides of the shaft end sealing gap, input shaft rotation speed, and sealing gap size are collected simultaneously.

[0010] S2. Data preprocessing and feature extraction: The collected multi-dimensional data is transmitted to the AI ​​controller. First, the 3σ criterion is used to remove abnormal data. Then, the Min-Max normalization method is used to map the data of different dimensions to the interval [0, 1]. Subsequently, the improved LSTM with attention mechanism built into the AI ​​controller is used to extract time series features and combine them with the historical operation database to build a sealed state feature model.

[0011] S3. Leakage risk trend prediction: Compare real-time concentration and pressure with multi-level thresholds, capture long-term and short-term dependencies through LSTM, and use fuzzy comprehensive evaluation method to determine the risk level. Output early warning for medium risk and above.

[0012] S4. Dynamic blowing control command generation: The AI ​​controller constructs a multi-objective optimization function with the goal of achieving the best sealing effect and minimizing the consumption of purge gas. It then uses a genetic algorithm to iteratively find the optimal blowing pressure and flow command.

[0013] S5. Adaptive airflow blocking and recovery: The AI ​​controller drives the control valve to adjust the gas supply from the gas cylinder, injects purge gas into the sealed gap through the purge port, forms a directional backflow airflow barrier, pushes the tritium gas back to the controllable recovery area of ​​the pump chamber, and feeds back the control parameters in a closed loop.

[0014] Optionally, the normalization formula in S2 is:

[0015]

[0016] in, This is the original data. For normalized data, The minimum value of the data. The maximum value of the data;

[0017] The attention mechanism of the improved LSTM calculates the feature weights for each time step using the following formula:

[0018]

[0019] in, For the first Attention weight at any moment For LSTM hidden states, , where are weight matrices. The bias vector is used to enhance the extraction of early warning features of leakage through differential weights.

[0020] Optionally, the attention mechanism includes a multimodal correlation module that performs cross-modal correlations on pressure, tritium concentration, rotational speed, gap, and drive motor current and temperature, using the Pearson correlation coefficient.

[0021]

[0022] in, For modal correlation degree, For the set of sealing state parameters, This is a set of motor operating parameters; used to mine multi-parameter linkage characteristics.

[0023] The attention mechanism has a dynamic weight optimization module that adaptively adjusts the attention weights of each time step and dimension feature according to the high-speed / low-speed conditions of the input axis, so that the feature extraction matches the working conditions.

[0024] The attention mechanism also includes a feature evolution mining module, which combines historical database analysis to study the evolution of sealing features with the aging and wear of bearing housings and input shafts, and constructs a sealing state model containing long-term evolution features.

[0025] Optionally, in S4, the genetic algorithm uses a multi-objective optimization function as the optimization objective, iterates through selection, crossover, and mutation, and outputs the optimal blowing control command with rotational speed and sealing gap size as real-time constraints.

[0026] Construct a multi-objective optimization function:

[0027]

[0028] in, To optimize the objective function, These are the weighting coefficients, and , This refers to the amount of tritium gas leaked. The flow rate of the purging gas.

[0029] Optionally, the purge port is a tapered directional push-back structure with a cone angle of 15°–30°, and the airflow velocity meets the following requirements:

[0030]

[0031] in, To purge airflow velocity, To sweep away traffic, For the flow area of ​​the purge port, is the cone angle velocity gain coefficient; to enhance the efficiency of tritium gas being pushed back into the pump chamber.

[0032] Optionally, the sensing element is a tritium-resistant pressure sensor and a tritium gas concentration sensor;

[0033] The concentration difference is obtained from the tritium concentration monitoring gauges in the main cavity and the side cavity.

[0034]

[0035] in, The concentration difference between the two sides of the sealed section The concentration inside the pump chamber. The concentration is on the outside of the bearing housing.

[0036] Optionally, the pressure difference can be obtained from a pressure sensor located inside the pump chamber and outside the bearing housing.

[0037]

[0038] in, The pressure difference between the two sides of the sealed section The pressure inside the pump chamber, This refers to the pressure on the outside of the bearing housing;

[0039] Combined concentration gradient Jointly assess the integrity of the seal and the risk of leakage.

[0040] Optionally, the AI ​​controller has a built-in feedback adjustment module that collects the actual opening degree of the control valve and the actual airflow parameters of the purge port in real time, and calculates the control error using the following formula:

[0041]

[0042] in, To control the error, To determine the actual flow rate of the purging gas, For optimal purge flow rate command, This is the actual blowing pressure. This is the optimal blowing pressure command;

[0043] when When the error exceeds the preset threshold, the AI ​​controller corrects the valve opening in real time to ensure a stable sealing effect.

[0044] Optionally, the speed regulation of the drive motor is linked to the air blowing control command of the AI ​​controller. When the input shaft speed exceeds a preset high-speed threshold, the AI ​​controller automatically increases the blowing gas flow rate and pressure, thereby increasing the blowing gas flow rate. It is positively correlated with the input shaft speed, compensating for airflow disturbance in the sealing gap caused by high-speed rotation.

[0045] Compared with the prior art, this application includes at least one of the following beneficial technical effects:

[0046] This invention deeply integrates multi-parameter sensing with artificial intelligence monitoring. It adopts AI monitoring technology and combines it with a multi-parameter sensing module to achieve real-time online monitoring of the sealing status in all dimensions. This enables real-time monitoring of the sealing status in all dimensions, captures early signs of leakage, and solves the pain points of traditional detection being lagging and incomplete in coverage, providing accurate data support for leakage risk prediction.

[0047] This invention relies on an improved LSTM neural network and fuzzy comprehensive evaluation method to achieve early prediction of leakage risks. Combined with an attention mechanism to adapt to working conditions, it breaks through the limitations of traditional passive sealing, realizes early prediction of sealing failure risks and multi-objective optimization decision-making, and outputs precise control commands, thus breaking through the limitations of traditional passive sealing protection and ensuring the safety of fusion systems and personnel.

[0048] By using AI closed-loop adaptive control and genetic algorithm optimization, combined with a tapered purge port to form a directional airflow barrier, and linked motor speed to compensate for disturbances, the defects of unstable high-speed dynamic sealing and inability to adapt to a wide range of working conditions are solved, thereby improving sealing reliability.

[0049] In summary, this invention utilizes AI-driven closed-loop adaptive control to dynamically match operating conditions and adjust the blowing parameters, effectively blocking tritium diffusion, improving the accuracy and stability of leakage control in extreme environments, and enabling the system to autonomously iterate and adapt using AI technology, reducing manual intervention and maintenance costs. It provides intelligent protection for the long-term safe operation of tritium-containing high-speed shaft-driven pumps and has outstanding engineering application value. Attached Figure Description

[0050] Figure 1 This is a schematic diagram of the sealing device proposed in this invention.

[0051] Reference numerals: 1. Pump chamber; 2. Rotor; 3. Bearing housing; 4. Main chamber tritium concentration monitoring gauge; 5. Side chamber tritium concentration monitoring gauge; 6. Drive motor; 7. AI controller; 8. Gas cylinder; 9. Control valve; 10. Purge port; 11. Input shaft. Detailed Implementation

[0052] The following specific examples illustrate the implementation of the present invention. Those skilled in the art can easily understand other advantages and effects of the present invention from the content disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments, and various details in this specification can also be modified or changed based on different viewpoints and applications without departing from the spirit of the present invention. It should be noted that, unless otherwise specified, the following embodiments and features described therein can be combined with each other.

[0053] Example

[0054] like Figure 1 As shown, this invention proposes a gas-blocking vacuum pump high-speed drive shaft end sealing method, based on a sealing device. The sealing device includes a pump chamber 1 and a rotor 2. The rotor 2 is driven by a drive motor 6 through an input shaft 11 and is located inside the pump chamber 1. A bearing seat 3 supports the input shaft 11. A main chamber tritium concentration monitoring gauge 4 is located inside the pump chamber 1, and a side chamber tritium concentration monitoring gauge 5 is located outside the bearing seat 3. A purge port 10 is connected to the shaft end sealing gap. A gas cylinder 8 is connected to the purge port 10 through a control valve 9. An AI controller 7 is connected to the main chamber tritium concentration monitoring gauge 4, the side chamber tritium concentration monitoring gauge 5, the drive motor 6, the control valve 9, and each sensing element. The speed adjustment of the drive motor 6 is linked to the purge control command of the AI ​​controller 7. When the speed of the input shaft 11 exceeds a preset high-speed threshold, the AI ​​controller 7 automatically increases the purge gas flow rate and pressure, thereby increasing the purge gas flow rate. It is positively correlated with the rotational speed of the input shaft 11, compensating for the airflow disturbance in the sealing gap caused by high-speed rotation.

[0055] The sealing method includes the following steps:

[0056] S1. Multi-dimensional sealing data acquisition: Utilizing radiation-resistant and corrosion-resistant special sensing elements deployed in the pump chamber 1, bearing housing 3, and key parts of the shaft system, the sensing elements are tritium-resistant pressure sensors and tritium gas concentration sensors; combined with the main chamber tritium concentration monitoring gauge 4 and the side chamber tritium concentration monitoring gauge 5, multi-dimensional data such as the pressure inside the pump chamber 1, the tritium gas concentration on both sides of the shaft end sealing gap, the rotation speed of the input shaft 11, and the sealing gap size are collected simultaneously;

[0057] The concentration difference is obtained from the tritium concentration monitoring gauge 4 in the main cavity and the tritium concentration monitoring gauge 5 in the side cavity:

[0058]

[0059] in, The concentration difference between the two sides of the sealed section The concentration inside pump chamber 1, The concentration is on the outer side of bearing housing 3;

[0060] The pressure difference is obtained from the pressure sensor inside the pump chamber 1 and outside the bearing housing 3:

[0061]

[0062] in, The pressure difference between the two sides of the sealed section The pressure inside pump chamber 1, The pressure on the outer side of bearing housing 3;

[0063] Combined concentration gradient Jointly assess the integrity of the seal and the risk of leakage.

[0064] S2. Data Preprocessing and Feature Extraction: The collected multi-dimensional data is transmitted to the AI ​​controller 7. First, outlier data is removed using the 3σ criterion. Then, the Min-Max normalization method is used to map data of different dimensions to the [0, 1] interval. The normalization formula is:

[0065]

[0066] in, This is the original data. For normalized data, The minimum value of the data. The maximum value of the data;

[0067] The AI ​​controller 7 has a built-in feedback adjustment module that collects the actual opening degree of the control valve 9 and the actual airflow parameters of the purge port 10 in real time, and calculates the control error using the following formula:

[0068]

[0069] in, To control the error, To determine the actual flow rate of the purging gas, For optimal purge flow rate command, This is the actual blowing pressure. This is the optimal blowing pressure command;

[0070] when When the error exceeds the preset threshold, the AI ​​controller 7 corrects the opening of the control valve 9 in real time to ensure a stable sealing effect.

[0071] Subsequently, the improved LSTM with attention mechanism built into the AI ​​controller 7 was used to extract time-series features, and a sealing state feature model was constructed by combining it with the historical operation database. The attention mechanism includes a multimodal correlation module, which performs cross-modal correlations on pressure, tritium concentration, rotational speed, gap, and drive motor 6 current and temperature, using the Pearson correlation coefficient.

[0072]

[0073] in, For modal correlation degree, For the set of sealing state parameters, This is a set of motor operating parameters; used to mine multi-parameter linkage characteristics.

[0074] The attention mechanism has a dynamic weight optimization module that adaptively adjusts the attention weights of each time step and dimension feature according to the high-speed / low-speed conditions of the input axis 11, so that the feature extraction matches the working conditions.

[0075] The attention mechanism also includes a feature evolution mining module, which combines historical database analysis to study the evolution of sealing features with the aging and wear of bearing housing 3 and input shaft 11, and constructs a sealing state model containing long-term evolution features.

[0076] S3. Leakage Risk Trend Prediction: Real-time concentration and pressure are compared with multi-level thresholds. LSTM is used to capture long-term and short-term dependencies, and a fuzzy comprehensive evaluation method is employed to determine the risk level. Warnings are issued for medium and above risks. The improved LSTM's attention mechanism calculates the time-step feature weights using the following formula:

[0077]

[0078] in, For the first Attention weight at any moment For LSTM hidden states, , where are weight matrices. The bias vector is used to enhance the extraction of early warning features of leakage through differential weights.

[0079] S4. Generation of dynamic air blowing control commands: The AI ​​controller 7 constructs a multi-objective optimization function with the goal of achieving the best sealing effect and minimizing the consumption of purge gas. Iterative optimization is performed through a genetic algorithm to obtain the optimal air blowing pressure and flow commands. The genetic algorithm uses the multi-objective optimization function as the optimization target, and iterative optimization is performed through selection, crossover, and mutation. With rotational speed and sealing gap size as real-time constraints, the optimal air blowing control command is output.

[0080] Construct a multi-objective optimization function:

[0081]

[0082] in, To optimize the objective function, These are the weighting coefficients, and , This refers to the amount of tritium gas leaked. The flow rate of the purging gas.

[0083] S5. Adaptive airflow blocking and recovery: AI controller 7 drives control valve 9 to adjust the gas supply of gas cylinder 8, injects purge gas into the sealing gap through purge port 10, forms a directional back-push airflow barrier, pushes tritium gas back to the controllable recovery area of ​​pump chamber 1, and feeds back the control parameters in a closed loop.

[0084] The purge port 10 is a tapered directional push-back structure with a cone angle of 15°–30°, and the airflow velocity meets the following requirements:

[0085]

[0086] in, To purge airflow velocity, To sweep away traffic, For the flow area of ​​the purge port, is the cone angle velocity gain coefficient; to enhance the efficiency of tritium gas being pushed back to pump chamber 1.

[0087] Specifically, during system operation, the drive motor 6 drives the rotor 2 to rotate at high speed within the pump chamber 1 via the input shaft 11. The pump chamber 1 is a tritium-contaminated working environment. The bearing housing 3 provides support and positioning for the input shaft 11, but the clearance between the shafts inevitably becomes a potential channel for tritium gas leakage. Tritium gas can easily diffuse from the pump chamber to the outer cavity along this clearance, thereby contaminating the drive motor 6 and affecting the safety of system operation.

[0088] To enable the prediction and control of leakage risks, tritium-resistant pressure sensors and tritium concentration sensors are simultaneously deployed in key parts of the pump chamber 1, bearing housing 3, and shaft system. These sensors collect multi-dimensional operating parameters in real time, such as pump chamber pressure, sealing gap size, and shaft speed. They work in conjunction with the main chamber tritium concentration monitoring gauge 4 and the side chamber tritium concentration monitoring gauge 5 to achieve real-time monitoring of the sealing status in all dimensions, capture early signs of leakage, and solve the pain points of traditional detection being lagging and incomplete in coverage. This provides accurate data support for the prediction of leakage risks. The main chamber tritium concentration monitoring gauge 4 and the side chamber tritium concentration monitoring gauge 5 simultaneously monitor and collect pressure and tritium concentration signals on both sides of the sealing section in real time, and transmit the collected multi-dimensional monitoring data to the AI ​​controller 7. The AI ​​Controller 7 is based on an improved LSTM neural network algorithm. It introduces an attention mechanism to achieve multimodal feature association and adaptive matching of operating conditions. It combines the aging evolution law of equipment to perform long-term trend analysis. At the same time, it uses the fuzzy comprehensive evaluation method to complete the risk level determination, realize the early prediction of leakage risk, break through the limitations of traditional passive sealing, and ensure the safety of fusion system and personnel. It compares, analyzes and extracts features from the real-time monitored concentration and pressure parameters with preset thresholds, and combines the historical operation database to predict leakage trends and identify the potential risk of tritium gas leakage to the atmospheric pressure side.

[0089] When parameters are detected to exceed the threshold or abnormal fluctuations occur, the AI ​​controller 7 combines the shaft rotation speed and real-time data of the sealing gap, and uses a genetic algorithm to perform multi-objective optimization with the goal of achieving the best sealing effect and minimizing the consumption of purging gas. It outputs the optimal purging pressure and flow command, adjusts the control valve 9, and dynamically optimizes the gas supply of the gas cylinder 8 to ensure that the purging effect matches the system operating conditions.

[0090] Purge gas, adapted to real-time operating conditions, is injected into the shaft end sealing gap through purge port 10. Purge port 10 adopts a 15°–30° tapered directional push structure to form a high-speed directional airflow barrier. At the same time, AI controller 7 adjusts the purging parameters in real time according to the rotation speed of input shaft 11 to compensate for the airflow disturbance in the sealing gap caused by high-speed rotation. This solves the defects of unstable high-speed dynamic sealing and inability to adapt to wide operating conditions, improves sealing reliability, and directionally pushes the leaked tritium gas back to the controllable recovery area of ​​pump chamber 1, confining it inside the boundary of prohibited leakage. The entire system autonomously completes data acquisition, risk prediction, command output, and closed-loop feedback without manual intervention, significantly reducing purge gas consumption and overall operation and maintenance costs. It is highly adaptable to the long-term operation requirements of tritium-related pumps in controlled nuclear fusion and has outstanding engineering application value. It achieves dynamic closed-loop sealing in conjunction with shaft rotation conditions.

[0091] The above specific embodiments are merely several optional embodiments of the present invention. Based on the technical solutions of the present invention and the relevant teachings of the above embodiments, those skilled in the art can make various alternative improvements and combinations to the above specific embodiments.

Claims

1. A method for sealing the end of a high-speed drive shaft of a gas-blocking vacuum pump, characterized in that, Based on the sealing device, the sealing device includes a pump chamber (1) and a rotor (2). The rotor (2) is driven by a drive motor (6) through an input shaft (11) and is located in the pump chamber (1). The bearing seat (3) is used to support the input shaft (11). The main chamber tritium concentration monitoring gauge (4) is located inside the pump chamber (1), and the side chamber tritium concentration monitoring gauge (5) is located outside the bearing seat (3). The purge port (10) is connected to the shaft end sealing gap. The gas cylinder (8) is connected to the purge port (10) through a control valve (9). The AI ​​controller (7) is connected to the main chamber tritium concentration monitoring gauge (4), the side chamber tritium concentration monitoring gauge (5), the drive motor (6), the control valve (9), and each sensing element. The sealing method includes the following steps: S1. Multi-dimensional sealing data acquisition: Using radiation-resistant and corrosion-resistant special sensing elements deployed in the pump chamber (1), bearing housing (3) and key parts of the shaft system, combined with the main chamber tritium concentration monitoring gauge (4) and the side chamber tritium concentration monitoring gauge (5), multi-dimensional data such as the pressure in the pump chamber (1), the tritium concentration on both sides of the shaft end sealing gap, the rotation speed of the input shaft (11), and the sealing gap size are collected simultaneously. S2. Data preprocessing and feature extraction: The collected multi-dimensional data is transmitted to the AI ​​controller (7). First, the 3σ criterion is used to remove abnormal data. Then, the Min-Max normalization method is used to map the data of different dimensions to the interval [0, 1]. Subsequently, the improved LSTM with attention mechanism built into the AI ​​controller (7) is used to extract time series features and a sealed state feature model is constructed by combining the historical operation database. S3. Leakage risk trend prediction: Compare real-time concentration and pressure with multi-level thresholds, capture long-term and short-term dependencies through LSTM, and use fuzzy comprehensive evaluation method to determine the risk level. Output early warning for medium risk and above. S4. Generation of dynamic blowing control instructions: The AI ​​controller (7) constructs a multi-objective optimization function with the goal of achieving the best sealing effect and minimizing the consumption of purge gas. It then uses a genetic algorithm to iteratively find the optimal blowing pressure and flow instructions. S5. Adaptive airflow blocking and recovery: The AI ​​controller (7) drives the control valve (9) to adjust the gas supply of the gas cylinder (8), injects the purging gas into the sealing gap through the purging port (10), forms a directional backflow airflow barrier, pushes the tritium gas back to the controllable recovery area of ​​the pump chamber (1), and feeds back the control parameters in a closed loop.

2. The gas-blocking vacuum pump high-speed drive shaft end sealing method according to claim 1, characterized in that, The normalization formula in S2 is: ， in, This is the original data. For normalized data, The minimum value of the data. The maximum value of the data; The attention mechanism of the improved LSTM calculates the feature weights for each time step using the following formula: ， in, For the first Attention weight at any moment For LSTM hidden states, , where are weight matrices. The bias vector is used to enhance the extraction of early warning features of leakage through differential weights.

3. The gas-blocking vacuum pump high-speed drive shaft end sealing method according to claim 2, characterized in that, The attention mechanism includes a multimodal correlation module that performs cross-modal correlations on pressure, tritium concentration, rotational speed, gap, and drive motor (6) current and temperature, using the Pearson correlation coefficient: ， in, For modal correlation degree, For the set of sealing state parameters, This is a set of motor operating parameters; used to mine multi-parameter linkage characteristics. The attention mechanism is equipped with a dynamic weight optimization module, which adaptively adjusts the attention weights of each time step and dimension feature according to the high-speed / low-speed working conditions of the input axis (11), so that the feature extraction matches the working conditions; The attention mechanism also includes a feature evolution mining module, which combines historical database analysis to study the evolution of sealing features with the aging and wear of bearing housing (3) and input shaft (11), and constructs a sealing state model containing long-term evolution features.

4. The gas-blocking vacuum pump high-speed drive shaft end sealing method according to claim 1, characterized in that, The genetic algorithm in S4 uses a multi-objective optimization function as the search objective, and iteratively optimizes through selection, crossover, and mutation, and outputs the optimal blowing control command with rotation speed and sealing gap size as real-time constraints. Construct a multi-objective optimization function: ， in, To optimize the objective function, These are the weighting coefficients, and , This refers to the amount of tritium gas leaked. The flow rate of the purging gas.

5. The gas-blocking vacuum pump high-speed drive shaft end sealing method according to claim 1, characterized in that, The purge port (10) is a tapered directional push-back structure with a cone angle of 15°–30°, and the airflow velocity satisfies: ， in, To purge airflow velocity, To sweep away the traffic, For the flow area of ​​the purge port, is the cone angle velocity gain coefficient; to enhance the efficiency of tritium gas being pushed back to the pump chamber (1).

6. The gas-blocking vacuum pump high-speed drive shaft end sealing method according to claim 1, characterized in that, The sensing elements are a tritium-resistant pressure sensor and a tritium gas concentration sensor; The concentration difference is obtained from the main cavity tritium concentration monitoring gauge (4) and the side cavity tritium concentration monitoring gauge (5): ， in, The concentration difference between the two sides of the sealed section The concentration inside the pump chamber (1) The concentration is on the outside of the bearing housing (3).

7. A method for sealing the high-speed drive shaft end of a gas-blocking vacuum pump according to claim 6, characterized in that, The pressure difference is obtained from the pressure sensor inside the pump chamber (1) and outside the bearing housing (3): ， in, The pressure difference between the two sides of the sealed section The pressure inside the pump chamber (1), For the pressure on the outside of the bearing housing (3); Combined concentration gradient Jointly assess the integrity of the seal and the risk of leakage.

8. The gas-blocking vacuum pump high-speed drive shaft end sealing method according to claim 1, characterized in that, The AI ​​controller (7) has a built-in feedback adjustment module that collects the actual opening degree of the control valve (9) and the actual airflow parameters of the purge port (10) in real time, and calculates the control error using the following formula: ， in, To control the error, To determine the actual flow rate of the purging gas, To obtain the optimal purge flow command, This refers to the actual blowing pressure. This is the optimal blowing pressure command; when When the error exceeds the preset error threshold, the AI ​​controller (7) corrects the opening of the control valve (9) in real time to ensure a stable sealing effect.

9. A method for sealing the high-speed drive shaft end of a gas-blocking vacuum pump according to claim 5, characterized in that, The speed regulation of the drive motor (6) is linked with the air blowing control command of the AI ​​controller (7). When the speed of the input shaft (11) exceeds the preset high-speed threshold, the AI ​​controller (7) automatically increases the flow rate and pressure of the blowing gas, thereby increasing the air blowing speed. It is positively correlated with the rotational speed of the input shaft (11) to compensate for the airflow disturbance in the sealing gap caused by high-speed rotation.

Images